# WHAT DETERMINES SOCIAL BEHAVIOR? INVESTIGATING THE ROLE OF EMOTIONS, SELF-CENTERED MOTIVES, AND SOCIAL NORMS

EDITED BY: Corrado Corradi-Dell'Acqua, Leonie Koban, Susanne Leiberg, Patrik Vuilleumier and Ernst Fehr PUBLISHED IN: Frontiers in Human Neuroscience and Frontiers in Psychology

#### *Frontiers Copyright Statement*

*© Copyright 2007-2016 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

> *The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88919-964-8 DOI 10.3389/978-2-88919-964-8

# About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

# Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

## Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

# What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# **WHAT DETERMINES SOCIAL BEHAVIOR? INVESTIGATING THE ROLE OF EMOTIONS, SELF-CENTERED MOTIVES, AND SOCIAL NORMS**

Topic Editors:

**Corrado Corradi-Dell'Acqua**, University of Geneva, Switzerland **Leonie Koban,** University of Colorado, USA **Susanne Leiberg,** University of Zurich, Switzerland **Patrik Vuilleumier,** University of Geneva, Switzerland **Ernst Fehr,** University of Zurich, Switzerland

Human behavior and decision making is subject to social and motivational influences such as emotions, norms and self/other regarding preferences. The identification of the neural and psychological mechanisms underlying these factors is a central issue in psychology, behavioral economics and social neuroscience, with important clinical, social, and even political implications. However, despite a continuously growing interest from the scientific community, the processes underlying these factors, as well as their ontogenetic and phylogenetic development, have so far remained elusive. In this Research Topic we collect articles that provide challenging insights and stimulate a fruitful controversy on the question of "what determines social behavior."

Indeed, over the last decades, research has shown that introducing a social context to otherwise abstract tasks has diverse effects on social behavior. On the one hand, it may induce individuals to act irrationally, for instance to refuse money, but on the other hand it improves individuals' reasoning, in that formerly difficult abstract problems can be easily solved. These lines of research led to distinct (although not necessarily mutually exclusive) models for socially-driven behavioral changes. For instance, a popular theoretical framework interprets human behavior as a result of a conflict between cognition and emotion, with the cognitive system promoting self-interested choices, and the emotional system (triggered by the social context) operating against them. Other theories favor social norms and deontic heuristics in biasing human reasoning and encouraging choices that are sometimes in conflict with one's interest. Few studies attempted to disentangle between these (as well as other) models. As a consequence, although insightful results arise from specific domains/tasks, a comprehensive theoretical framework is still missing.

Furthermore, studies employing neuroimaging techniques have begun to shed some light on the neural substrates involved in social behavior, implicating consistently (although not exclusively) portions of the limbic system, the insular and the prefrontal cortex. In this context, a challenge for present research lies not only in further mapping the brain structures implicated in social behavior, or in describing in detail the functional interaction between these structures, but in showing how the implicated networks relate to different theoretical models.

This is Research Topic hosted by members of the Swiss National Center of Competence in Research "Affective Sciences – Emotions in Individual Behaviour and Social Processes". We collected contributions from the international community which extended the current knowledge about the psychological and neural structures underlying social behavior and decision making. In particular, we encouraged submissions from investigators arising from different domains (psychology, behavioral economics, affective sciences, etc.) implementing different techniques (behavior, electrophysiology, neuroimaging, brain stimulations) on different populations (neurotypical adults, children, brain damaged or psychiatric patients, etc.). Animal studies are also included, as the data reported are of high comparative value. Finally, we also welcomed submissions of meta-analytical articles, mini-reviews and perspective papers which offer provocative and insightful interpretations of the recent literature in the field.

**Citation:** Corradi-Dell'Acqua, C., Koban, L., Leiberg, S., Vuilleumier, P., Fehr, E., eds. (2016). What Determines Social Behavior? Investigating the Role of Emotions, Self-Centered Motives, and Social Norms. Lausanne: Frontiers Media. doi: 10.3389/978-2-88919-964-8

# Table of Contents

# **Section 1: Introduction**

*07 Editorial: What Determines Social Behavior? Investigating the Role of Emotions, Self-Centered Motives, and Social Norms*

Corrado Corradi-Dell'Acqua, Leonie Koban, Susanne Leiberg and Patrik Vuilleumier

*12 Social behavior in the "Age of Empathy"?—A social scientist's perspective on current trends in the behavioral sciences* Svenja Matusall

# **Section 2: Personal Determinants Sub-section 2.1: Genetic polymorphisms**


# **Sub-section 2.2: Individual traits**

*48 The social regulation of threat-related attentional disengagement in highly anxious individuals*

Erin L. Maresh, Lane Beckes and James A. Coan


Ralf Veit, Lilian Konicar, Jens G. Klinzing, Beatrix Barth, Özge Yilmaz and Niels Birbaumer

*91 What can we learn about emotion by studying psychopathy?* Abigail A. Marsh

# **Sub-section 2.3: Neurodevelopmental and Psychiatric Disorders**


Anna M. Järvinen and Ursula Bellugi

*124 Integrating intention and context: assessing social cognition in adults with Asperger syndrome* Sandra Baez, Alexia Rattazzi, María L. Gonzalez-Gadea, Teresa Torralva,

Nora Silvana Vigliecca, Jean Decety, Facundo Manes and Agustin Ibanez


Tiziana Zalla and Marco Sperduti

# **Sub-section 2.4: Emotion induction and Regulation**


Eliran Halali, Yoella Bereby-Meyer and Axel Ockenfels


Mascha van 't Wout, Sara Faught and David Menino

# **Section 3: Environmental Determinants Sub-section 3.1: Social Values and Norms**

*222 Economic and evolutionary hypotheses for cross-population variation in parochialism* Daniel J. Hruschka and Joseph Henrich

*232 Neurocognitive mechanisms underlying value-based decision-making: from core values to economic value* Tobias Brosch and David Sander

*240 The role of interest in the transmission of social values* Fabrice Clément and Daniel Dukes


# **Sub-section 3.2: Social Context**


Dario Bombari, Marianne Schmid Mast, Tobias Brosch and David Sander

*283 Understanding and accounting for relational context is critical for social neuroscience*

Elizabeth Clark-Polner and Margaret S. Clark

*297 The ethology of empathy: a taxonomy of real-world targets of need and their effect on observers*

Stephanie D. Preston, Alicia J. Hofelich and R. Brent Stansfield


# **Section 4: Social Behavior in the Medial Prefrontal Cortex**


Pyungwon Kang, Jongbin Lee, Sunhae Sul and Hackjin Kim

*377 Investment and repayment in a trust game after ventromedial prefrontal damage*

Giovanna Moretto, Manuela Sellitto and Giuseppe di Pellegrino

*387 Neural correlates of the behavioral-autonomic interaction response to potentially threatening stimuli*

Tom F. D. Farrow, Naomi K. Johnson, Michael D. Hunter, Anthony T. Barker, Iain D. Wilkinson and Peter W. R. Woodruff

# Editorial: What Determines Social Behavior? Investigating the Role of Emotions, Self-Centered Motives, and Social Norms

Corrado Corradi-Dell'Acqua1, 2, <sup>3</sup> \*, Leonie Koban1, 2, 4, Susanne Leiberg<sup>5</sup> and Patrik Vuilleumier 1, 2

<sup>1</sup> Swiss Centre for Affective Sciences, University of Geneva, Geneva, Switzerland, <sup>2</sup> Laboratory for Neurology and Imaging of Cognition, Department of Neurosciences and Clinic of Neurology, University Medical Center, Geneva, Switzerland, <sup>3</sup> Département de Psychologie, Faculté de Psychologie et des Sciences de l'Éducation, Université de Genève, Geneva, Switzerland, <sup>4</sup> Department of Psychology and Neuroscience, University of Colorado, Boulder, CO, USA, <sup>5</sup> Department of Economics, University of Zürich, Zürich, Switzerland

Keywords: social behavior, ultimatum game, emotions, decision making, contextual appraisal, medial prefrontal cortex (mPFC), autism spectrum disorders (ASD), oxytocin receptor gene

**The Editorial on the Research Topic**

### **What Determines Social Behavior? Investigating the Role of Emotions, Self-Centered Motives, and Social Norms**

In the last decade, a growing research effort in behavioral sciences, especially psychology and neuroscience, has been invested in the study of the cognitive, biological, and evolutionary foundations of social behavior. Differently from the case of sociology, which studies social behavior also at the group level in terms of organizations and structures, psychology and neuroscience often define "social" as a feature of the individual brain that allows an efficient interaction with conspecifics, and thus constitutes a possible evolutionary advantage (Matusall). In this view, an extremely wide range of mental and neural processes can be classified as "social," from the coding of relevant sensory stimuli about conspecifics (facial expressions, gestures, vocalizations, etc.), to the selection and planning of behavioral responses in complex interpersonal settings (economic transactions, negotiations, etc.). Despite such heterogeneity, there is a converging interest in the scientific community toward the identification of neural and psychological mechanisms that underlie all the many facets of social behavior, and their comparison across species and cultures.

This Research Topic was initiated by researchers from the Swiss National Center of Competence in Research "Affective Sciences—Emotions in Individual Behaviour and Social Processes," a multidisciplinary institution devoted to the study of affect-related processes across various disciplines (from psychology and neuroscience through to history, philosophy, art, and economy). In keeping with this spirit, this Research Topic comprehends 38 contributions from an interdisciplinary community each addressing specific psychological and neural phenomena that can be defined as "social." In particular, we collected both theoretical and empirical contributions, concerning animals, human individuals (neurotypical adults and children, but also individuals with neurological, psychiatric and developmental disorders) as well as human groups, engaged in either laboratory-controlled settings or real-life situations. Although the theoretical models and the applied research techniques (psychophysical, physiological, neuroimaging, genetic) are very diverse, they converge with a global framework suggesting that the determinants of social

Edited and reviewed by: Hauke R. Heekeren, Freie Universität Berlin, Germany

> \*Correspondence: Corrado Corradi-Dell'Acqua corrado.corradi@unige.ch

Received: 27 October 2015 Accepted: 20 June 2016 Published: 07 July 2016

#### Citation:

Corradi-Dell'Acqua C, Koban L, Leiberg S and Vuilleumier P (2016) Editorial: What Determines Social Behavior? Investigating the Role of Emotions, Self-Centered Motives, and Social Norms. Front. Hum. Neurosci. 10:342. doi: 10.3389/fnhum.2016.00342 behavior can be described across two independent dimensions: (1) a personal-to-environmental dimension, and (2) a transientto-stable dimension. These contributions thus represent an important cornerstone for building an interdisciplinary and comprehensive model of how individuals deal with the complexity of their social environment.

# PERSONAL-TO-ENVIRONMENTAL DIMENSION

For the purpose of this editorial, we can schematically describe social interactions as cases in which an individual is engaged in a given social environment. Importantly, the individual and the environment exert reciprocal influence on one another, as individual changes could cause, and be caused by, changes in the outside world. Within this context, we can define a behavior of interest any change of the individual's state over time (overt response, brain modulation, etc.), which in turn can be related to two main explanatory variables: a representation of the current state of the individual (to know how a person will change one needs to know how this person is) and a representation of the current state of the environment (to know how a person will change one needs to know what surrounds him/her). Thus, the personal-to-environmental dimension distinguishes between those determinants of social behavior that are attributable to idiosyncratic features of the individual from those that are related to specificities of the environment with which the individual is interacting. Such simplified model fits well our Research Topic, as the various contributions highlight the role of many factors that, despite their diversity, can be readily classified as personal or environmental.

Among the personal factors, the role played by genetic polymorphisms is well-described in the present Research Topic through the use of knock-out mice and endophenotype approaches in humans. In all these cases, the implicated genes are known to affect major functions of hormonal and neurotransmitter systems within brain networks important for social cognition. For instance, mice lacking the β2 subunit of neuronal nicotinic receptors of acetylcholine exhibit impaired behavior (relative to wild type mice) when competing with conspecifics for rewards (Chabout et al.) Furthermore, following a rich body of literature documenting how intranasal administration of oxytocin affects human social behavior (see Ebner et al; Haas et al.; Järvinen and Bellugi, as reviews), several contributions address the role played by the oxytocin gene receptor (OXTR). Taking a developmental perspective, Ebner et al. show how OXTR polymorphisms differently affect young and older adults' responses in medial prefrontal cortex (MPFC) to facial emotional expressions. Haas et al. suggest how OXTR polymorphisms might explain variations in individual cooperative behavior by affecting the structure and function of key brain areas for social behavior such the amygdala, the superior temporal sulcus, and the anterior cingulate cortex. It is possible that brain regions with high density of oxytocin receptors (such as the amygdala) affect social behavior through their regulatory role on the autonomic nervous system, an hypothesis put forward by Järvinen and Bellugi to account for social dysfunctional behavior in Williams Syndrome, in addition to more classic effects on cognition or learning. Finally, Hruschka and Henrich point out that genetic polymorphism might even explain some cultural differences, as suggested by the controversial evidence that collectivistic (as opposed to individualistic) societies might most frequently exhibit allelic variation of serotonin-transporter-linked polymorphic region (Chiao and Blizinsky, 2010; Eisenberg and Hayes, 2011).

A few studies also highlighted the role played in social behavior by individual traits: these are habitual patterns of behavior, thoughts and emotions that are relatively stable over time. Although of unclear etiology, inter-individual trait variability has been often used in the literature as a powerful factor that explains behavioral differences in the neurotypical population. This is the case of several studies from the present Research Topic, who report for instance that individual empathic traits can influence the decoding of emotional facial expressions Huelle et al., or monetary decisions on behalf of unknown people (O'Connell et al.). Furthermore, (Maresh et al.) find that the neural response to electrical shocks (and the degree to which this is affected by social proximity) is modulated by individual anxiety trait, a measure of idiosyncratic sensitivity to stressors. Finally, this Research Topic includes multiple studies on individuals exhibiting traits diagnostic of psychopathy, a developmental syndrome characterized by low levels of empathy, guilt, and remorse, but increased aggressive and antisocial behavior (Marsh). In particular, individuals with high psychopathic scores exhibit altered neural and behavioral responses in many experimental manipulations related to fear conditioning (Veit et al.), fear empathy (Marsh), or moral cognition (Tassy et al.). The case of psychopathy highlights the close tie between individual traits and the presence of disorders, which can be considered in some cases as extreme variants of normative behavioral patterns (Hare and Neumann, 2005; Walton et al., 2008). Consistently, several studies report atypical social behavior in individuals with psychiatric diagnoses or neurodevelopmental syndromes. For instance, individuals with schizophrenia and bipolar disorders show impairments in tasks involving the inference of others' thoughts and emotions (Caletti et al.). In a similar vein, individuals with Autism Spectrum Disorder or Asperger Syndrome display atypical behavior in several tasks (see Zalla and Sperduti, for review) ranging from visual processing of emotional facial expressions (Corradi-Dell'Acqua et al). to the inference of others' states, empathy, and moral cognition (Baez et al.).

Among environmental factors, several studies in the present Research Topic highlight the role played by social norms. These can be understood as representations of community's desires and expectations about end states that guide our evaluation of events and the selection of behavioral responses (see Brosch and Sander, for more details on norms and values). In particular, Hruschka and Henrich point out that socioeconomic rules (related to religion or market) can explain the degree to which populations are eager to exhibit in-group biases. Furthermore, Clément and Dukes discuss how one's interest toward events in the environment might be biased by their normative significance, i.e., by the degree to which these are relevant for social norms and for the self-concept in the community. Additional contributions suggest how people's behavior during situations involving division of goods can be understood prevalently in terms of fairness norms or equality heuristics, according to which people are eager to sanction unequal divisions even at their own expenses (Civai). For instance, Shaw and Olson show that children from 6 to 8 years of age will correct (or at least attempt to minimize) unequal distributions of tokens between two unknown kids. In adults, two articles suggest a major role of fairness heuristics in the well-known Ultimatum Game task (Civai; Guney and Newell): in both cases the authors argue that individuals (responders) refuse money which is freely offered to them when part of an unequal division, regardless of their ongoing emotional response (Civai) or of the alleged intentions of person (the proposer) who is making the offer (Guney and Newell).

# STABLE-TO-TRANSIENT DIMENSION

Most of the studies reviewed in the previous section describe factors that, despite their difference, can be classified as stable, i.e., they are held to exert a long-lasting effect on individual social behavior. These can be understood as general behavioral determinants, which transcend specific situations. Although important, stable determinants have only an approximate predictive power, as a large variability of individual social behavior can be explained in terms of transient factors related to the specificities of the interpersonal situation. For instance, as individual social behavior can be partly explained by idiosyncratic features of the individual, they can as well be affected by factors that temporally alter the individual's state and how he/she interacts with the social environment.

Several studies document that people's social behavior can be affected by manipulating their preexisting emotional state, for instance by showing them arousing stimuli, exposing them to stressful vs. rewarding conditions, or engaging them in emotion regulation strategies. As for the case of genetic polymorphisms, these preexisting emotional states can alter the mental and brain processes critical for individual social behavior, thus showing how affective and social functioning might rely on partially overlapping systems. For instance, Eskine presents compelling evidence that people's moral coding might be grounded in the same processes underlying gustatory disgust (see also Eskine et al., 2011, 2012). Likewise, in line with a rich body of literature showing how empathetic reactions to others' pain and disgust recruit similar neural structures as those involved in first-hand experiences of pain and disgust (Corradi-Dell'Acqua et al., 2011, 2016; Bernhardt and Singer, 2012, but see Krishnan et al., 2016), Marsh argues that dysfunctions in fear experience might lead to a reduced capacity to recognize fear in others (see also Adolphs et al., 1994).

Several contributions examine the role of preexisting emotional states in decision-making using behavioral economics paradigms. The theoretical framework underlying most of these studies posits that individual decisions result from the interaction of at least two different brain systems (Dual-System model—see Halali et al.): the cognitive/deliberate system (slow, controlled, cognitively-demanding, and instantiated mainly in prefrontal cortex) and the affective system (fast, automatic, cognitively non-demanding, and instantiated predominantly in limbic regions). As these two systems might promote conflicting courses of actions, transient emotional induction can be used as a mean to strengthen the affective contribution to a decision, as shown by Eimontaite et al. who find that inducing anger in people makes them less cooperative in social decision-making tasks such the Trust Game and the Prisoner Dilemma. Using a complementary approach, some studies engaged participants in emotion regulations strategies, by asking them to up- or down-regulate their emotional responses. Such regulation was found to have a significant impact on subsequent behavior (Grecucci et al.; van't Wout et al.) and brain responses (Grecucci et al.) in tasks such the Ultimatum and Dictator Game.

# CONTEXTUAL AND SOCIAL APPRAISAL

Accounts such as the Dual-System Model have been criticized for their dichotomous separation between cognition and emotion, which appears oversimplistic and not supported by empirical evidence (e.g., Moll et al., 2008; Shackman et al., 2011; Koban and Pourtois, 2014; Phelps et al., 2014). Alternative theoretical frameworks suggest instead that emotion is not a unitary construct opposed to cognition, and that distinct affective/motivational components may impact behavior in different (and in some cases opposite) ways (Moll et al., 2008; Phelps et al., 2014). In particular, appraisal theories of emotions (e.g., the Component Process Model by Scherer, 1984, 2009) propose that affective experience is critically determined by a series of cognitive evaluations (appraisal checks) of the environment in terms of events' novelty, valence, impact on one's goals, and how they can be dealt with. For instance, sadness is based on the awareness of the presence of a salient negative event (e.g., the occurrence of a terminal disease), undermining personal goals (it will end one's life), against which no course of action seems effective. The same event can instead induce an emotional response of higher arousal (such as anger or rage), if associated with the belief that a solution (a treatment) is available. In this perspective, the Component Process Model is not merely a theory of emotions, but can be seen as a comprehensive framework in which cognitive evaluation of the environment, affective reactions, and preparation of a behavioral response are integrated into a unique system.

For the purpose of this editorial, the appraisal checks proposed by the Component Process Model (Scherer, 1984, 2009) are good candidate processes for explaining how the social environment should not be considered as a stable construct exerting longlasting effects on individual behavior, but also as the result of multiple contextual or transitory factors that, when combined together, make each inter-personal situation unique. In accord with this view, several contributions to this Research Topic suggest that individual affective and behavioral responses might be determined by evaluations of the social context, some of which correspond to the same appraisal checks described in the Component Process Model. For instance, Maresh et al. show that, in anxious individuals, neural responses to threatening electrical stimuli are modulated by whether participants are alone or close to a person that could be a stranger or a friend. Furthermore, Clark-Polner and Clark review how interpersonal behavior (e.g., reaction to others' emotions, providing and receiving social support) are affected by the context of the relationship. Similarly, Baez et al. suggest that the social proficiency of individuals with Asperger Syndrome could improve when the contextual information from social settings is made explicit. Finally, Alexopoulos et al. had participants playing as responders in a modified Ultimatum Game task, and find that the neural activity in MPFC to unfair offers is affected by whether they could retaliate against the proposer (which reflects a change in coping potential).

Due to the dynamic properties of interpersonal relationships and interactions, simple appraisal checks such the assessment of novelty, valence, coping potential, etc. are often not sufficient to tackle the complexities of social situations. Among the many contextual/transitory properties of the environment that need to be appraised, there is also the presence of other human beings, each with their own mental states and cognitive appraisals. Let's imagine, for instance, the case in which an individual is observing a friend, in the attempt to infer his/her emotional states. It is reasonable that, to do so, the individual might model the behavior of the observed friend in relation of the most likely determinants, including his/her contextual appraisal. In particular, the individual can assess if the friend is sad, by checking whether he/she believes to be terminally ill and that a treatment might not be available (see also Corradi-Dell'Acqua et al., 2014). This is an example of social appraisal, in which each individual represents contextual aspects of the social environment also in terms of how other bystanders evaluate the same environment from their point of view (see Manstead and Fischer, 2001; Clément and Dukes). Social appraisal refers to individuals' metacognitive abilities, and has close ties with concepts such as mentalizing, theory-of-mind, and perspective taking. Importantly, the role played by social appraisal has been highlighted in this Research Topic by articles focusing on impression formation (Kuzmanovic et al.), interpersonal relationships (Bombari et al.) and monetary transactions (Halali et al.; Tomasino et al.). In particular, the behavioral and neural responses of individuals (responders) to unfairness in the Ultimatum Game can be affected by whether the monetary transaction is framed by the proposer in terms of offer ("I give") or acquisition ("I take"; Sarlo et al., 2013; Tomasino et al..) Furthermore, Halali et al. suggest that, when playing as proposers in the Ultimatum and Dictator Game tasks, participants most automatic choices are driven by considerations about whether the responder can retaliate against a potential unfair treatment.

Social appraisal can be differentiated from other kinds of contextual evaluations at the neural level. In particular, in line with existing models on the organization of MPFC (Lieberman, 2007; Forbes and Grafman, 2010; Corradi-Dell'Acqua et al., 2015), Bzdok et al. use meta-analytical evidence to propose a segregation between a dorsal portion, involved in topdown, controlled, metacognitive abilities, and a ventral portion involved in bottom-up, automatic evaluative-related processes. This segregation is also supported by Kang et al. who show how the dorsal MPFC is implicated in accurately estimating other people's preferences, whereas the ventral MPFC is recruited when using the Self as a proxy for the estimation. Furthermore, Grossmann reports that, already at the age of 5 months, dorsal MPFC might be implicated in triadic interactions, in which infants establish eye contact with others, in order direct their attention to specific objects/events in the external environment (see also Grossmann and Johnson, 2010). It should be stressed, however, that this segregation between dorsal and ventral regions is at odds with other studies from our Research Topic: on the one side, Farrow et al. implicate the dorsal (but not ventral) MPFC in the processing and evaluation of threatening words, picture and sounds; on the other hand, ventral (but not dorsal) MPFC is associated with processes related to social appraisal, such as the differential treatment of human and computer opponents in monetary transactions (Moretto et al.), or the conformity to the decision of in-group peers in a perceptual estimation task (Stallen et al.).

# CONCLUSIONS

In the last decades, psychologist and neuroscientists invested a considerable amount of research to investigate the ability to act "socially," which is considered an evolutionary advantage of many species (Matusall). The present Research Topic is a collection of a large number (38) of original contributions from an interdisciplinary community which together highlight that determinants of individual social behavior should be best understood along at least two different dimensions. This general perspective represents the backbone for a comprehensive and articulated model of how people and their brains interact with each other in social contexts. However, despite its appeal, it remains unclear how the model put forward in this editorial relates to particular paradigms with high ecological value, where it is more difficult to neatly disentangle the relative contribution of personal/environmental or stable/transient determinants. This is for instance the case of Preston et al. who investigated hospitalized terminal patients, measuring the emotional reactions elicited in observers and whether they were related to the frequency with which aid was delivered. In this perspective, a great challenge for future research in social psychology and neuroscience will indeed be to develop more accurate predictive models of social behavior and to make them applicable to ecologically valid settings.

# AUTHOR CONTRIBUTIONS

All authors listed, have made substantial, direct and intellectual contribution to the work, and approved it for publication.

# REFERENCES


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Corradi-Dell'Acqua, Koban, Leiberg and Vuilleumier. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# *Svenja Matusall\**

*MINDLab and Interacting Minds Centre, Aarhus University, Aarhus, Denmark*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Anelis Kaiser, University of Freiburg, Germany Gregory Hollin, University of Nottingham, UK Rebekka Klein, Martin-Luther-University of Halle-Wittenberg, Germany*

#### *\*Correspondence:*

*Svenja Matusall, MINDLab and Interacting Minds Centre, Aarhus University, Jens Chr, Skous Vej 4, Building 1483, 3rd Floor, 8000 Aarhus, Denmark e-mail: s.matusall@hum.au.dk*

Recently, several behavioral sciences became increasingly interested in investigating biological and evolutionary foundations of (human) social behavior. In this light, prosocial behavior is seen as a core element of human nature. A central role within this perspective plays the "social brain" that is not only able to communicate with the environment but rather to interact directly with other brains via neuronal mind reading capacities such as empathy. From the perspective of a sociologist, this paper investigates what "social" means in contemporary behavioral and particularly brain sciences. It will be discussed what "social" means in the light of social neuroscience and a glance into the history of social psychology and the brain sciences will show that two thought traditions come together in social neuroscience, combining an individualistic and an evolutionary notion of the "social." The paper concludes by situating current research on prosocial behavior in broader social discourses about sociality and society, suggesting that to naturalize prosocial aspects in human life is a current trend in today's behavioral sciences and beyond.

**Keywords: social neuroscience, prosocial behavior, history of neuroscience, epistemology, science studies**

## **INTRODUCTION**

Recently, several behavioral sciences, for instance neuroeconomics (e.g., Fehr and Fischbacher, 2003), primatology (e.g., De Waal, 2009) and social neuroscience (e.g., Frith and Frith, 2010), became increasingly interested in investigating biological and evolutionary foundations of (human) social behavior. Scholars from these fields argue that the biology of humans is itself much more prosocial than previously thought. Prosocial behavior is a core element of human nature. It is rooted in each individual, has evolved during the course of evolution, is located in the brain, its genes, functions, hormones and neurotransmitters and is embedded in an environment. A central concept of this new perspective on human nature is the "social brain" (Brothers, 1990) that is not only able to communicate with the environment but rather to interact directly with other brains via neuronal mind reading capacities such as empathy (see Young, 2012a).

Taking social neuroscience as an example, this paper explores the notion of "social" in contemporary behavioral sciences and how a new concept of human nature emerges. At the core of this new concept is the notion that default human behavior is prosocial. The paper sets out to investigate what "social" means in social neuroscience. (1), the research field is introduced before a glance in the history of the social sciences shows that "social" is by no means an unambiguous term (2). The historical roots of the social brain are explored (3) and the paper concludes (4) by situating current research on social behavior in broader discourses about sociality and society, suggesting that the trend to look for prosocial aspects in human life, culture and society also takes place in other spheres of society.

## **WHAT IS SOCIAL NEUROSCIENCE?**

Social neuroscience is much more diverse than this brief perspective paper could picture and hence this paper's aim can only be to outline general trends within the field. The term "social neuroscience" was first coined by social psychologists Gary Berntson and John Cacioppo in 1992 (Cacioppo and Berntson, 1992). They propose a cooperation between social psychology and neuroscience in order to avoid the pitfalls of reductionism by adding multiple perspectives to given problems. But it took another decade before a field with research groups, professorships, university courses, textbooks, conferences, societies, and journals emerged that calls itself social neuroscience (Matusall et al., 2011). In this process, a second important impetus came from a paper by Ochsner and Lieberman (2001), who should also be named among the founding figures of the field.

Many of social neuroscience's topics of interest fall into the realm of classic social psychology, for instance the study attitudes, prejudices and stereotypes (Matusall, 2012). Interestingly, however, is the field's new focus on emotion, empathy and altruism (cf. Decety and Ickes, 2009; Singer and Lamm, 2009). Recently, prosocial behavior moved into the center of attention, not only in social neuroscience but also in other behavioral sciences such as primatology and anthropology (cf. De Waal, 2009; Tomasello, 2009).

#### **WHAT DOES SOCIAL MEAN IN SOCIAL NEUROSCIENCES?**

In social neuroscience, prosocial behavior is sought in genes, brains and evolutionary past. "Social" is simultaneously understood as a capacity of the organism's brain to cope with the environment and as an evolutionary advantage of the species. This perspective on the social differs fundamentally from sociology's perspective, where the social can be anything from the sum of individual actions to power relations or social structures. The list of phenomena having been defined to be social in the course of the history of the social sciences is rather long and diverse as (Greenwood, 1997, p. 3) points out by giving a random collection of those phenomena: "states, families, armies, religious organizations, literary societies, mobs, street brawls, people chatting on a street corner, the Roman Catholic Church, the Renaissance, insect communication, dominance hierarchies among primates, language, financial instruments, and traffic flow in a city." Thus, "social" is by no means an unambiguous term and for understanding social neuroscience's notion of "social," it is crucial to look into the history of experimental social psychology, which is one of social neuroscience's intellectual parent disciplines. Looking at the questions social neuroscientists tackle in their research, it soon becomes evident that they focus on the way social stimuli are perceived and processed in the brain—no matter whether they study empathy, attitudes toward out-group members or voters' behavior. This individual-centered approach may be self-evident for social neuroscientists, yet it is a historically contingent approach as will be shown in the next section.

## **GENEALOGY OF A CONCEPT**

The individualistic perspective on the social has a long tradition in experimental social psychology: since its emergence in the 1920's, this discipline has understood itself as a branch of individual psychology (Allport, 1924), investigating whether and how the perception and processing of social stimuli differed from the perception and processing of non-social stimuli. In order to apply experimental methods to such questions, social psychologists had to frame their objects of investigation as statistically measurable. In this process, the social was redefined as a quality of countable entities. This perspective differed from theories in 19th century social psychology that connected the social with morality and religion, respectively with institutionalized power (Danziger, 1997). Moreover, the individualist notion of the social had a crucial role in defining and defending the individualistic American Way of Life against collectivist notions of society and the individual (Rose, 1998). The political background of its emergence seems all but forgotten by those employing this notion of social today as a variable investigated by experimental methods. Most social neuroscientists are trained in social psychology and most positions are located in psychology departments. Their research questions and their argumentation stand in the tradition of experimental social psychology. By relocating the "social" in the individual's brain and neurobiology, social neuroscientists are in line with their predecessors in treating it as an individual capacity.

#### **PERSPECTIVES FROM SOCIOLOGY**

Looking with the eyes of a sociologist, investigating problems in small pieces, such as brain activation, entails the risk of losing the perspective on the broader picture and taking the small piece for the whole problem (Star, 1983). The experimental design of "social" in social neuroscience research requires rendering research in a quantitative fashion. <sup>1</sup> This does not necessarily imply a reduction of complexity in the stimuli presented but in the questions asked. If complex issues such as voters' emotional reactions to election outcomes or empathy with members of an "out-group" are measured by quantitative tools, it has to be assumed that complex phenomenona can be split up into several problems and thus are not more than the sum of their parts. This approach differs fundamentally from hermeneutic approaches towards complex phenomena, which are more interested in meaning than in mechanisms and which are dominant in humanities and non-quantifying social sciences.

To some extent, social neuroscientists seem to be aware of this and pay credit to the problem of complexity by drawing on the notion of levels (Cacioppo and Berntson, 1992; Ochsner and Lieberman, 2001). Cacioppo and Berntson (1992) maintain that although the brain is an essential component of all social beings, brain, behavior and society are each too complex to be reduced to one another. Hence, social neuroscience aims to combine data generated on different levels to reach a better comprehension of social behavior. Yet, knowledge from other disciplines can only be integrated if compatible with the standards of quantifying sciences and qualitative knowledge is difficult to incorporate in such paradigms.

# **HISTORY OF THE SOCIAL BRAIN<sup>2</sup>**

Not only in social psychology, also in the brain sciences, questions about the "social" have a long tradition. The relationship between the brain and the social has been an issue of hot debate ever since the emergence of modern brain science in the late 18th century. In these debates, the pendulum has been swinging happily back and forth between seeing either nature or nurture as responsible for human behavior. Early 19th century's phrenologists, for instance, defined a cerebral faculty for each human property and thus saw a clear causal direction from brain to behavior, while psychiatrists in the second half of the 19th century made harmful social conditions responsible for psychiatric disorders and thus reversed causal directions (Hagner, 2007). Theories of evolution were central to 19th and early 20th century's concepts of the brain and the social. These theories were associated with a hierarchical organization of brain areas: the younger, more evolved parts such as intellectual capacities or morality controlled older parts such as drives and emotions (e.g., Jackson, 1884).

Not least as a reaction to the role medicine and biological sciences played in Nazi ideology, after the Second World War research in the West was dominated by behaviorism, cybernetics and cognitive science (Hagner, 2007). During that time questions about human interactions did not play a role in mainstream neuroscience and psychology. This began to change slowly in the 1980's and with even more force in the 1990's

<sup>1</sup>On the potential dangers of the "mereological fallacy", see Bennett and Hacker, 2003 and also Krüger, 2010.

<sup>2</sup>For more detailed historical analyses of discourses on the social brain and its relationship to society, see the recent work by anthropologist Allan Young: Young, 2011, 2012a,b. For a philosophical perspective on prosociality in neuroeconomics, and particularly a critical examination of the notion of altruistic punishment, see e.g., Klein, 2012.

when the social brain returned to the debate in three independent theories about the relationship between brain and social: the *social brain hypothesis*, the *somatic marker hypothesis* and the *mirror neuron theory*, which will be discussed in next section.

## **THE SOCIAL BRAIN SINCE THE 1990's**

The social brain hypothesis suggests that the size of the neocortex and the group size of mammals living in social groups correlate (Brothers, 1990; Dunbar, 1998). The bigger the group, the more complex the social situations which the brain has to process. Certain cognitive skills evolved to cope with social complexity. Consequently, the way we act in social interactions is determined by evolutionary heritage. The social brain hypothesis does not explicitly discuss the impact of history, culture, society, or life experiences on social cognition abilities in an individual or a group. Only in an evolutionary time frame these factors may have an impact on how future generations may engage with each other (Matusall, 2012). Nor does it answer the "hen and egg" question of whether the complex social groups or the cerebral capacities for processing them was first; or whether both evolved together. What it does is providing an evolutionary explanation for both, human sociality and the species' big brains.

The second theory, the somatic marker hypothesis was introduced by neuropsychiatrist Antonio Damasio and it suggests that positive experiences are connected with positive memories leaving a positive somatic marker, i.e., an incentive for deciding in favor of similar actions in future decision-making processes while negative experiences are connected with negative memories leaving negative a somatic marker, i.e., an alarm bell, leading to deciding against similar actions in future decision-making processes. These markers are acquired during socialization not only through experienced events but also by incorporating norms and rules and can change throughout life if new experiences occur (Damasio et al., 1991). This means a crucial shift in thinking about the social and the brain, which is later taken up by social neurosciences and related disciplines (Cacioppo and Berntson, 2005; Glimcher et al., 2009; Ariely and Berns, 2010). The somatic marker hypothesis couples biology with cultural and social environments. Somatic markers and thus the ability to act socially is part of the biological make-up with which humans are born, yet the way this sociality takes shape depends on the particular beliefs and values of the society one is born into (Damasio, 1994).

Around the same time when Damasio developed his somatic marker hypothesis, in Italy a team of neuroscientists reported to have found a neural basis of the capacity of primates to engage with others (di Pellegrino et al., 1992). It followed an everincreasing interest in these neurons, which were soon named mirror neurons, and their hypothesized function included a growing number of areas of social life (e.g., Gallese, 2003). This theory did not only seem to explain human social behavior, development and learning but also how we participate, for example, in another person's joy and distress automatically, by biological default. Yet, after the first excitement faded away, mirror neurons became contested (see for instance Hickok, 2008; Gallese et al., 2011) and it is too early to decide whether the mirror neuron theory will become canonical knowledge in the attempt of how mind and brain work. Like other such theories such as the concept of brain plasticity, mirror neuron theory enjoys a broad popularity outside the scientific community—perhaps not least because it provides a biology based on prosociality. The idea of biologically automatic responses to other people's behavior and even emotions is alluring, since it seems to argue in favor of a prosocial default of human nature. Even though feeling does not automatically lead to acting, being able to empathize may lay a foundation for prosocial action.

These three theories and their focus on social aspects of the human condition differ from preceding notions of human nature in one fundamental respect: Homo sapiens are understood as a social and empathic species rather than an individualistic one. Contrary to older models, it is now suggested that it comes quite naturally to humans to act prosocially. Evidence for the prosocial nature of humankind is found in humans' evolutionary history and the neurobiological and hormonal substrate of the brain. By looking at social behavior from this perspective, it appears that cooperation and altruism are beneficial. Working together, so the argument goes, made life easier and increased the chances of survival of the group's offspring (see e.g., Brothers, 1990 and Dunbar, 1998).

# **FUTURE PERSPECTIVES**

Evolutionary reasoning about prosociality can be summarized as follows: since Homo sapiens are a social species, organized in communities, individuals, who are able to decipher social stimuli and to act in prosocial ways had better chances of reproduction and hence, social brains evolved.<sup>3</sup> This evolutionary heritage equips contemporary humans with the tools for coping with the complexity of social organizations and to engage in social relationships. Not everyone acts prosocially all the time, but every healthy person bears in themselves the potential to do so and has the option to act on that potential. This perspective on sociality means a shift in the conceptual framework of what it is the norm and what needs explanation. While protagonists of this new version of human nature do not deny that aggression is as much part of human nature as is empathy, it now becomes marked as the other, the trait which needs to be explained and this also provides a new perspective on pathologies such as psychopathy or autism, which are now defined by their lack of empathy (e.g., Baron-Cohen, 2011; Blair, 2011). But not only pathologies, even everyday behavior such as envy is interpreted in terms of empathy, respectively the lack thereof (e.g., Shamay-Tsoory, 2009). This does not mean that antisocial behavior is no longer a part of this paradigm. Yet, it becomes the other, the non-normal, which needs to be explained.

In social neuroscience, the individualistic notion of social rooting in American social psychology and the more collectivist notion of the social rooting in anthropology come together and thus in this framework, social relations are intelligibly investigated within the individual. The focus is not on structures, institutions,

<sup>3</sup>The relationship between prosociality, cooperation, and altruism is complex and by no means uncontested in evolutionary psychology and other behavioral sciences. For overviews over the debate see e.g., Henrich and Henrich, 2006; Boyd and Richerson, 2009.

power relations, all things that can potentially be changed, but on the social as a biological category—nature—that cannot be changed. Sociality becomes a naturalized, innate quality and thus every "normal" individual is capable of behaving prosocially. At a time when responsibility for social cohesion is de-centralized, the neural capacity for prosociality is found.

#### **NEOSOCIALITY?**

Social neuroscience's notion of social relates to a new notion of what human beings are and how they normally act, in short a new version of a biologically based human nature. In this narrative, sociality is the driving force behind human evolution.

The notion of "social" employed in social neuroscience research is located in the individual brain, its ability to decode a certain kind of stimuli and to interact with others. It is a noteworthy historical concomitance that the investigation of social interactions via social structures or collective processes is replaced by the investigation of processes that take place within individuals at the same time when, in a broader societal setting, collectivist solutions have been replaced by more individual solution (e.g., in welfare, see for instance Sennett, 2006; Lessenich, 2008). Rabinow (1999) described this development as the transformation towards a "biosociality"—social structures become less important while identities are more and more based on individual (i.e., genetic) attributes than on social or group attributes. Investigating the social via communal genetic make-up or individuals' brains is rather different from studying the external conditions for a social structure.

## **REFERENCES**


In this approach, prosocial behavior becomes something innate and thus every normal individual is capable of behaving prosocially.

#### **CONCLUSION**

Social neuroscience is an interdisciplinary endeavor aiming to investigate sociality. Taking its methods from social psychology and cognitive neuroscience and its explanatory frame from evolutionary anthropology, it defines the social as both a feature of Homo sapiens' environment and an inherent human capacity to cope and survive. Doing so, it contributes to a new, prosocial notion of human nature. The lens through which social behavior is studied, has changed.

Yet, at the moment, both its focus on quantitative methods and reservations from many arts and social sciences exclude qualitatively operating social science from participating in this endeavor. A methodological and epistemological openness on both sides would be desirable because this could really increase knowledge about social conditions of human nature. Examples for such openness and collaborations can for instance be found in projects on "neurofeminism" (Bluhm et al., 2012; Dussauge and Kaiser, 2012; Einstein, 2012; Matusall, in press). These projects experiment with collaborations bridging the gap between qualitative and quantitative disciplines.

### **ACKNOWLEDGMENTS**

The research was in part funded by ESF grant number 2423, SNF grant number 100011-116725/1 and MINDLab.


puzzle of human cooperation. *Cogn. Syst. Res.* 7, 220–245. doi: 10.1016/j.cogsys.2005.11.010


Young, A. (2012b). The social brain and the myth of empathy. *Sci. Context* 25, 401–424.

**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 18 February 2013; accepted: 15 May 2013; published online: 31 May 2013.*

*Citation: Matusall S (2013) Social behavior in the "Age of Empathy"?—A social scientist's perspective on current trends in the behavioral sciences. Front. Hum. Neurosci. 7:236. doi: 10.3389/ fnhum.2013.00236*

*Copyright © 2013 Matusall. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Making choice between competing rewards in uncertain vs. safe social environment: role of neuronal nicotinic receptors of acetylcholine

# *Jonathan Chabout , Arnaud Cressant , Xian Hu , Jean-Marc Edeline and Sylvie Granon\**

*Centre de Neuroscience Paris Sud, Centre National de la Recherche Scientifique UMR 8195, Université Paris Sud 11, Orsay, France*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Paul E. Smaldino, Johns Hopkins University, USA Robert Liu, Emory University, USA*

#### *\*Correspondence:*

*Sylvie Granon, Centre de Neuroscience Paris Sud, Centre National de la Recherche Scientifique UMR 8195, Université Paris Sud 11, Bâtiment 446, 15 Bd Clémenceau, Orsay 91405, France e-mail: sylvie.granon@u-psud.fr*

In social environments, choosing between multiple rewards is modulated by the uncertainty of the situation. Here, we compared how mice interact with a conspecific and how they use acoustic communication during this interaction in a three chambers task (no social threat was possible) and a Social Interaction Task, SIT (uncertain situation as two mice interact freely). We further manipulated the motivational state of the mice to see how they rank natural rewards such as social contact, food, and novelty seeking. We previously showed that beta2-subunit containing nicotinic receptors −β2∗nAChRs- are required for establishing reward ranking between social interaction, novelty exploration, and food consumption in social situations with high uncertainty. Knockout mice for <sup>β</sup>2∗nAChRs <sup>−</sup>β2−*/*−mice- exhibit profound impairment in making social flexible choices, as compared to control -WT- mice. Our current data shows that being confronted with a conspecific in a socially safe environment as compared to a more uncertain environment, drastically reduced communication between the two mice, and changed their way to deal with a social conspecific. Furthermore, we demonstrated for the first time, that β2−*/*<sup>−</sup> mice had the same motivational ranking than WT mice when placed in a socially safe environment. Therefore, β2∗nAChRs are not necessary for integrating social information or social rewards *per se*, but are important for making choices, only in a socially uncertain environment. This seems particularly important in the context of Social Neuroscience, as numerous animal models are used to provide novel insights and to test promising novel treatments of human pathologies affecting social and communication processes, among which Autistic spectrum disorders and schizophrenia.

**Keywords: decision-making, social interaction, mice, beta2 nAChRs, ultrasonic vocalization, uncertainty**

### **INTRODUCTION**

Choosing among different rewards relies on multiple processes such as gaining knowledge about existing rewards and their respective value, integrating our own motivational state for each of them, as well as our individual goals. The ability to establish a rank between different rewards is thus a complex process that allows cognitive flexibility, goal focusing, and appropriate decision-making (Chambers et al., 2007; Körding, 2007; Badre, 2012; Smaldino and Richerson, 2012). In addition, the decision making process is complicated by different kinds of dilemmas, such as the one reflected by the exploitation/exploration process (Sutton and Barto, 1998), with striatal dopaminergic mechanisms being strongly linked to the automaticity of exploitation (Everitt et al., 2008; Maia, 2009). Exploration, by contrast, varies following the uncertainty of the different outcomes in competition and the prefrontal cortex plays a pivotal role in tracking uncertainty levels (Daw and Doya, 2006; Daw et al., 2006; Strauss et al., 2011).

In some environments, such as social ones, choosing between concurrent rewards is highly modulated by the uncertainty of the situation. Indeed, if social contacts constitute a reward for social mammals (Panksepp and Lahvis, 2007; Trezza et al., 2011), they may also trigger unknown reactions from social partners, thus making social environment uncertain and potentially risky.

We previously showed that animals lacking beta2 subunit of neuronal nicotinic receptors (β2−*/*−mice) showed impaired behavioral flexibility and difficulty to switch from one reward to another, whether the switch was between social interaction and food consumption, food retrieval and novelty exploration, or novelty exploration and social contact (Granon et al., 2003; Serreau et al., 2011). Particularly, in a social interaction task (SIT) designed to emphasize free social interaction, with potential risk of aggressiveness by an unknown conspecific (Cambon et al., 2010), we showed that β2−*/*−mice exhibited higher level of dominance and lower level of flexibility, in relation with their prefrontal hyper-monoaminergia (Coura et al., 2013). In addition, using a dedicated software to pinpoint social decisions by the probabilistic analysis of more than 20 social sequences within the normal social repertoire (De Chaumont et al., 2012), we showed that depleting the noradrenergic prefrontal innervation in normal mice shrinks the decision tree in this task, with lesioned mice making more rigid and non-adaptive decisions leading to aggressiveness (Coura et al., 2013). A deeper analysis of β2−*/*<sup>−</sup> mice' behavior (De Chaumont et al., 2012) was performed by the off line dissection of their behavioral repertoire during the SIT. We identified one peculiar dual riskprone posture, called "back-to-back," that requires the progressive development of tolerance from both mice. Indeed, when in this "back-to-back" posture, both mice of the dyad tolerate to be outside of the field of view of the other mouse. We showed that this specific posture emerged progressively while social contact frequency decreased. We thus postulated that this posture that does not exist at first when animals just met, reflects the tolerance they develop for a novel adult male conspecific.

As β2−*/*<sup>−</sup> mice did not integrate their partner's behavioral choices -stop, escape, approach- for adapting their own choices, this risk-prone posture virtually never emerged in β2−*/*<sup>−</sup> mice, leading to the continuous reinforcement of a unique motivation (i.e., social contact), instead of a switch between novelty exploration and social reinforcement. It is noticeable that the β2−*/*−mice flexible defect in the SIT was overcome by reexpression of the beta2-containing nAChRs into the prefrontal cortex -PFC- of β2−*/*−mice, thus showing the need for functional cholinergic transmission within the PFC for such integrative processes (Avale et al., 2011).

As we showed that the "pro-social" behavior of β2−*/*<sup>−</sup> mice was neither due to an impulsive phenotype nor to a biased evaluation of food or social reward values (Serreau et al., 2011), we wondered, here, whether β2−*/*<sup>−</sup> mice exhibited difficulties in dealing with competing rewards when they can make free choice, in a safe environment. Indeed, in previous work (Serreau et al., 2011), we put in a same novel arena a novel conspecific and attractive food. We saw that β2−*/*<sup>−</sup> mice disengaged less easily from a reinforcing behavior than WT mice, if reinforcements were in conflict with one another. Also, if WT mice frequently switched from one motivation to the other, the frequency of these transitions were biased in β2−*/*<sup>−</sup> mice in favor of social motivation. We particularly observed that β2−*/*<sup>−</sup> mice were more ready to discard a food reward if the social conspecific approached them (Serreau et al., 2011). It was therefore unkown whether β2−*/*<sup>−</sup> mice were more attracted by the social partner because social rewards were more interesting to them, or if they replied more strongly to a social partner that they may perceive as a putative threat. The latter point could be linked to their major increase in dominance behaviors (Coura et al., 2013), and their proneness to exhibit rigid follow behaviors (De Chaumont et al., 2012).

In the current study, we thus defined two types of environments: a "socially safe" one, represented by the 3-chamber apparatus, in which the test animal did not make real physical contact with the social partner, although it was able to see, smell and hear it. Therefore, there was no physical threat, and the choices made by the test mouse were more likely to rely on its own internal state and motivation. The second type of social environment was a large and novel cage in which a dyad of mice interacted freely. The risk of physical threat and dominance existed, although real aggressiveness was extremely rare in the C57BL/6 strain, in this particular protocol (Coura et al., 2013). We defined this situation as "socially uncertain." It is noticeable that both environments were novel and that the putative stress induced by novelty was diminished by prior exploration.

Here, we compared the ability of WT and β2−*/*<sup>−</sup> mice in comparing different natural rewards two by two -social, food or novelty exploration- in a "safe" environment, the three chambers task (Crawley, 2007; Chadman et al., 2008; Silverman et al., 2010a,b). The particularity of the task is that the test mouse is free to explore each reward, without any threat resulting from another male mouse's direct contact. In addition, we wondered whether mice emit ultrasonic vocalizations—USVs—when they were in contact with non-social rewards (such as food). Indeed, it is known that mice emit USVs in both social or non-social contexts (Panksepp and Lahvis, 2007; Jamain et al., 2008; Scattoni et al., 2008, 2010; Chabout et al., 2012). Our recent work showed that the number of emitted USVs correlates with the duration of social contact, and were strongly modulated by motivational/emotional states (Chabout et al., 2012). Acoustics parameters, like peak frequency, duration and number of calls, were dependent of the behavioral context, with high frequency USVs uttered in social (positive or attractive status) context while low frequency USVs were uttered in restrain (negative status) context. Therefore, the integrated analysis of behavioral and communication data may provide novel insight as to the emotional states of mice when confronted to competing rewards in a safe environment.

The aim of this study consisted in providing answers to three main questions:


## **MATERIALS AND METHODS ETHICS STATEMENT AND ANIMALS**

The animals were treated according to the ethical standards defined by the Center National de la Recherche Scientifique for animal health and care in strict compliance with the EEC recommendations (*n*◦86/609). All efforts were made to minimize animal discomfort and to reduce the number of animals used. We tested 25 β2−*/*<sup>−</sup> and 40 C57BL/6J -hereby called WT- male mice, all reared and purchased from Charles Rivers Laboratories France (L'Arbresle Cedex, France). β2−*/*<sup>−</sup> mice were originally generated from a 129/Sv ES cell line as described previously (Picciotto et al., 1995) and backcrossed onto the C57BL/6J strain for 20 generations. Because littermates are not available in the breeding facilities and as the number of backcrosses was high, we used C57BL/6J mice as controls.

They were 11–12 weeks old at their arrival and remained housed in a standard rearing facility in collective cages (4, 5 animals per cage) during one week before any experiment. Room ventilation, temperature and humidity were controlled with a 12/12 light-dark cycle (light on at 8:00 am). They received *ad libitum* water (throughout all experiments) and standard chow (quantity depending on the experiment).

For sessions of "three chambers task" experimental mice were placed in individual cages three weeks before the experiment while animals used as social stimuli remained in collective cages.

For the SIT mice were thereafter placed in individual cages 3 weeks before the experiment while visitor animals remained in collective cages. Visitor mice were all male C57BL/6J mice while experimental mice were either β2−*/*<sup>−</sup> or C57BL/6J mice.

#### **BEHAVIORAL PROCEDURES**

The succession of behavioral procedures is depicted on **Figure 1**.

#### *Three chambers tasks (3Ch)*

Two sessions of 3Ch (respectively, without and with food deprivation) were performed with 1 month interval with the same animals, and by keeping the same group of individuals. The apparatus was a rectangular box (64 × 42 cm) made from translucent Plexiglas. It was divided in three compartments of equal surface by Plexiglas walls. Light was set at 100 Lux and a numeric camera (Hercules®) was placed above the cage allowing to record mouse displacements. We used three different rewards according to the different groups. As a social reward, a naïve C57BL/6J male mouse was placed under a cup. Cups were Plexiglas cylinders with multiple holes to allow breathing, acoustic communication and nose-pokes from both mice. A glass of water was placed on top of the cup to prevent displacement and the test mouse from climbing. Food rewards were sucrose pellets (14 mg, Bio-Serv®), and a cup similar to the one described above was used as a novel object. For the two sessions of 3Ch all mice were habituated to consume sucrose pellets 3 days before the experimental days.

This apparatus allowed us to test mice's preference between two rewards we put in competition. Thus, we used three independent groups, Social *vs.* Food (8 WT, 8 β2−*/*<sup>−</sup> mice), Social *vs.* Object (8 WT, 9 β2−*/*<sup>−</sup> mice), Food *vs.* Object (8 WT, 8 β2−*/*<sup>−</sup> mice). Each group of mice was exposed to only two rewards at a time.

• For the first 3Ch session mice were fed *ad libitum*, with test mice isolated three weeks before the experiment. Tested mice were habituated 10 min to the central room of the apparatus and 10 min to the entire empty apparatus. Social reward mice were habituated to be under the cup for 15 min 3 times per days for 2 days prior experiment.

• The second 3Ch session (after food deprivation) was performed 1 month after the first one. The same groups of animals were maintained during the two sessions. Mice from Social *vs.* Food and Food *vs.* Object groups were deprived of food 2 weeks before the experiment. For deprivation standard chow was given so as to adjust and maintain at 85% of their free feeding weight. Only mice from the Social *vs.* Object condition were not deprived. The latter group allows the control of the repetition effect by comparing the results between the first and the second session of 3Ch in the same mice. Since the mice were tested twice in the same apparatus, we wanted to check that habituation to the maze would not impact the results. Test mice were habituated 10 min to the central room of the apparatus and 10 min to the entire empty apparatus. Social stimuli mice were re-exposed to the cup for 15 min the day before the experiment.

For both 3Ch sessions, after habituation phases, the two rewards were placed in opposite rooms, at the opposite of the microphones side (see **Figure 1B**). Location (left or right) of each reward was alternated across subject. The test phase lasted 10 min during which video and USVs were recorded. At the end of the

experiment, both test and social stimuli mice were replaced in their respective home cage. Between each trial the apparatus was cleaned first with 60% ethanol then with distilled water.

## *Social interaction task (SIT)*

The day of the experiment each animal was allowed to visit alone the novel environment for 30 min consisting of a transparent Plexiglas cage containing fresh bedding (50 × 30 × 30 cm) placed in an unfamiliar quiet room. The experimental cage was situated on a table, under a numeric video camera (Hercules®) connected to a computer (recording at 33 frames per s). Light was set at 100 Lux by undirected bulbs. After 30min habituation of the test mouse, a "visitor" mouse was gently introduced into the cage. "Visitors" were male mice unknown from the test mouse, of the same age from the C57BL/6J strain. "Visitors" had always been maintained in social cages. Each dyad was used only once.

### *Parameters of the tasks*

In the 3Ch experiment, we scored the time spent and the number of entrances in each reward room, as well as the number and time of contact with each reward. We considered an entrance when the animal placed the two forepaws in one room, and contact with the reward when the animal was less than 1 cm away from the reward. We scored USVs when the animal was in the reward chamber. Therefore, as the time spent in each chamber may vary, we expressed the USVs as the number of calls divided by the time spent in the reward chambers.

In the SIT experiment, we scored manually the duration and number of social contacts and analyzed the behavioral sequences between the two conspecifics for 8 min. Likewise, we scored USVs during the 8 min experiment.

## *Control measures*

*Olfaction tests.* Olfactory tests were devoted to test if mice of both genotypes were able to detect smells (i.e., small volatile molecules carried by the air). These odors are detected by neurons of the main olfactory epithelium. We therefore checked olfactory discrimination between water, orange flavor, and urine of male mice. By contrast, pheromones are detected by a specialized and distinct olfactory system, the vomeronasal organ (Dulac, 2000). As pheromones are present in high concentration in litter, we also subjected mice to a second olfactory test and compared their behavior when confronted to a clean *vs.* a used litter. Both olfactory experiments conducted in 24 WT and 25 β2 KO mice were tested in a transparent cage of Plexiglas (50 × 30× 30 cm). Their procedures are described below.

*Experiment 1: Comparison between three olfactory stimuli.* This first olfactory experiment was used to test the ability of both groups of mice to discriminate volatile odors. The experiment consisted of 30 min habituation to the cage. During this habituation period, an empty tube was taped to one largest side wall of the cage. The tube consisted of a Pasteur pipette (of which the tip was broken off) with a piece of filter paper (2 × 2 cm) rolled into it. It was taped onto the wall of the cage with a distance of 9 cm between the tip of the pipette and the bottom of the cage. This habituation period was followed by three times 2 min exposure to water, orange and urine odors, successively inserted in the tube. The orange consisted of a 1% solution of natural orange flavor in water. The urine sample was collected from groups of social C57BL6/J male mice. After collection the urine was kept in 0.5 ml Eppendorf tubes and frozen until use. A 20μl drop of the odor sample was added to the filter before the exposure.

We measured the time spent sniffing the tip of the pipette thanks to *off line* video analyses. The sniffing area was defined by a 2 cm diameter circle around the tip.

*Experiment 2: Fresh litter vs. used litter.* Litter taken from social cages -used litter- which may contain some volatile compounds but that contains mostly non-volatile ones (pheromones), was used to test mice's sensitivity to non-volatile odors components. Two Petri dishes (diameter 10 cm) were taped on the cage's floor. The floor was divided into eight equal square pieces by a piece of paper placed under the cage. Petri dishes contained fresh or used litter (mixture from four different cages of six same genotype male mice). Right or left position of each dish was randomized between each trial. The tested mouse was placed at the center of the cage and freely explored the environment during 15 min. The experimental cage was situated on a table, under a numeric video camera (Hercules®) connected to a computer (recording at 33 frames per s). Light was set at 100 Lux by undirected bulbs.

The time spent digging in each dish, number of exploration moves into the litter (front paws in the cup), number of rearing and grooming were measured and analyzed *off line* on the videos.

*Auditory tests.* Thresholds for the averaged Auditory Brainstem Response (ABR) were used as an electrophysiological measure of auditory sensitivity (Willott and Erway, 1998; Willott, 2006). These measures were made at the end, after all the behavioral procedures above. For this, calibrated stimuli were delivered using speaker equipment manufactured by DELTAMED. A maximum sound pressure level (SPL re: 20 WPa) of 80 dB was employed for all stimuli. Mice were anesthetized with mixed Xylazine (10 mg/Kg) and Ketamine (150 mg/Kg). Sub-dermal needle electrodes were inserted at the vertex (active), ventrolaterally to the left ear (reference) and in a paw muscle (ground). Mice were tested with tone pips (100 μs rise/fall; 10 ms duration; 1, 2, 4, 5, 8, 12, 16, 24, and 32 kHz). ABR thresholds were obtained for each frequency by reducing the SPL at 10 dB steps and finally at 5 dB steps up and down to identify the lowest level at which an ABR could be recognized. All records were computerized by software; Centor USB, DELTAMED.

#### *Ultrasonic vocalization recording*

In all experiments, except olfaction tests, a condenser ultrasound microphone Polaroid/CMPA was placed above the experimental chamber, high enough so that the receiving angle of the microphone covered the whole area of the test cage. For the 3Ch condition, one microphone was placed above each side chamber with an angle allowing full chamber coverage but avoiding any recording from the opposite chamber (**Figure 1**). Microphones were connected to an ultrasound recording interface Ultrasound Gate 416 H, which was itself plugged into a personal computer equipped with the recording software Avisoft Recorder USG (Sampling frequency: 250 kHz; FFT-length: 1024 points; 16-bits). All recording hardware and software were from Avisoft Bioacoustics® (Berlin, Germany).

#### *Acoustic variables*

For all behavioral conditions USVs were analyzed off line with SASLab Pro (Avisoft Bioacoustic®, Berlin, Germany). Spectrograms were generated for each detected call (Sampling frequency: 250 kHz; FFT-length: 1024 points; 16-bit; Blackman window; overlap: 87.5%; time resolution: 0.512 ms; frequency resolution: 244 Hz). For SIT condition audio recordings were disturbed by the background noise originating from the animals moving and/or digging in the fresh bedding. We nevertheless kept the bedding because social interactions may have been affected by its absence and we wanted to match as closely as possible to our classical experimental conditions (Granon et al., 2003). However, this prevented an automatic analysis of acoustic data.

We recorded the total number of calls emitted by each pair of mice, and manually measured different variables related to peak frequency [*Pf*start (peak frequency at the beginning of the call), *Pf*end (peak frequency at the end of the call), *Pf*min (minimum peak frequency), *Pf*max (maximum peak frequency)] for each call allowing us to calculate the *Pf*mean as *Pf*mean = *(Pf*min + *Pf*max*)*/2.

#### *Synchronization of audio and video files*

We performed a "clap" with our fingers in the field of the camera to time-matched video and audio files. In the audio files, we cut the information before this sound and in the video files we selected the exact frame of this event and started from this point. This manual synchronization allowed us to analyze which USVs were emitted during contact and non-contact events for SIT condition, and which USVs were emitted when test mouse was actually present in the related reward room in the 3Ch.

#### **STATISTICAL ANALYSES**

Statistical analyses were made with Statview® software. ANOVA repeated measures were used to compare the reward factors two by two (Social vs. Food, Social vs. Object, and Food vs. Object). Repeated measures ANOVA were used to compare subject performances. *Post-hoc* analyses were performed using Wilcoxon signed-rank (for dependent variables) or Mann-Whitney (for independent variables) non parametric tests only when appropriate. Correlation data were analyzed with a Spearman correlation test between behavioral measures and number of calls. The significance threshold was set at *p <* 0*.*05. For all *post-hoc* paired comparisons a Bonferroni correction was applied (α = α*/k*; where α is significance threshold and *k* the number of comparisons).

## **RESULTS**

#### **REWARD RANKING OF β2−***/***<sup>−</sup> MICE IN SAFE ENVIRONMENT**

We first analyzed contact time in all the non-deprived conditions (Social vs. Food, Social vs. Object, Food vs. Object). For all conditions, there was a major reward effect [S vs. F: *F(*1*,* <sup>15</sup>*)* = 42*.*38, *P <* 0*.*0001; S vs. O: *F(*1*,* <sup>14</sup>*)* = 32*.*85, *P <* 0*.*0001; F vs. O: *F(*1*,* <sup>14</sup>*)* = 11*.*18, *P* = 0*.*0048] and no genotype effect [S vs. F: *F(*1*,* <sup>15</sup>*)* = 0*.*75, *P* = NS; S vs. O: *F(*1*,* <sup>14</sup>*)* = 1*.*65, *P* = NS; F vs. O: *F(*1*,* <sup>14</sup>*)* = 0*.*325, *P* = NS]. There was an interaction only in the Food vs. Object condition [interaction genotype × condition: *F(*1*,* <sup>14</sup>*)* = 7*.*7, *P* = 0*.*01]. A more detailed comparison between rewards revealed that WT and β2−*/*<sup>−</sup> mice stayed longer in contact with the social reward, then with the Food [**Figure 2A** and **Table 1**, WT: S vs. *<sup>F</sup>*: *<sup>z</sup>* = −2*.*38, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>*.*017; <sup>β</sup>2−*/*−: S vs. F: *z* = −2*.*54, *P* = 0*.*011], but that they also prefer the Social as compared to a novel object (**Figure 2A** and **Table 1**, WT: S vs. O: *<sup>z</sup>* = −2*.*38, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>*.*017; <sup>β</sup>2−*/*−: S vs. O: *<sup>z</sup>* = −2*.*38, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>*.*017). However, β2−*/*<sup>−</sup> mice spent similar time in contact with the food and the novel object in the Food vs. Object condition, while WTs spent more time in contact with the novel object (**Figure 2A** and **Table 1**, WT: F vs. O: *<sup>z</sup>* = −2*.*52, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>*.*011; <sup>β</sup>2−*/*−: F vs. O: *z* = −0*.*42, *P* = 0*.*67). We noticed that the number of entrance in each compartment (data not show) was similar in both genotypes and for all the conditions. Therefore, even if β2−*/*<sup>−</sup> mice are hyperactive (Granon et al., 2003), this cannot explain the difference between the two genotypes concerning the time spent in contact with each reward.

When mice were in contact with rewards they emitted USVs. The number of calls was dependent of the time spent in each compartment (if they spent more time in the food compartment

**FIGURE 2 | Reward ranking of β2−***/***<sup>−</sup> and WT mice in safe environment. (A)** Time spent in contact with each reward. **(B)** Number of ultrasonic vocalizations— USVs—emitted in contact with each reward divided per the time spent in the room. Data are presented for WT and β2−*/*<sup>−</sup> mice as mean ± SE. <sup>∗</sup>*p <* 0*.*005; for Mann-Whitney paired comparisons.

**Table 1 | Summary of the reward ranking according to the two exposures of the three chambers tasks, before food deprivation and after food deprivation, for WT and β2−***/***<sup>−</sup> mice.**


the probability to emit calls was higher). To circumvent this bias, we calculated the ratio of the number of USVs divided by the time spent in a given compartment. Results showed that when mice were not food deprived, such as in Social vs. Food and Social vs. Object conditions, there was a reward effect [S vs. F: *F(*1*,* <sup>15</sup>*)* = 9*.*94, *P* = 0*.*006; S vs. O: *F(*1*,* <sup>15</sup>*)* = 25*.*59, *p* = 0*.*0002], but no genotype effect [S vs. F: *F(*1*,* <sup>15</sup>*)* = 1*.*70, *p* = NS; S vs. O: *F(*1*,* <sup>15</sup>*)* = 0*.*019, *p* = NS] and no interaction reward × genotype (S vs. F: *p* = NS, S vs. O: *p* = NS). We showed that WT and β2−*/*<sup>−</sup> mice always emitted more USVs in contact with social rewards than in contact with object rewards, while only β2−*/*<sup>−</sup> emitted more USVs in contact with social rewards in the Social vs. Food condition (**Figure 2B** and **Table 1**, WT: S vs. *F*: *z* = – 1.68, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*09; S vs. O: *<sup>z</sup>* <sup>=</sup> –2.1, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*03; <sup>β</sup>2−*/*−: S vs. F: *<sup>z</sup>* <sup>=</sup> –2.19, *p* = 0*.*02; S vs. O: *z* = –2.38, *p* = 0*.*01). We showed that there was no difference between USVs uttered in Food vs. Object condition for both genotypes [Reward effect: *F(*1*,* <sup>15</sup>*)* = 0*.*019, *p* = NS; genotype effect: *F(*1*,* <sup>15</sup>*)* = 1*.*72, *p* = NS].

Furthermore, we analyzed the peak frequency mean, Pf mean, of these calls in each compartment. Interestingly, WT and β2−*/*<sup>−</sup> showed no differences (not shown). However, the Pf mean was lower in contact with social reward for both genotypes (WT: 51.9 <sup>±</sup> 1.6 kHz; <sup>β</sup>2−*/*−: 48.7 <sup>±</sup> 1.4 kHz) than Pf mean in food reward (WT: 61.4 <sup>±</sup> 2.8 kHz; <sup>β</sup>2−*/*−: 64 <sup>±</sup> 4.3 kHz) as well as in contact with the object reward (WT: 61.1 <sup>±</sup> 2.6 kHz; <sup>β</sup>2−*/*−: 60.2 <sup>±</sup> 2.8 kHz). This result led us to think that the social mouse placed under the cup, although habituated to the procedure, contributed to the low frequency calls.

#### **ADAPTIVE BEHAVIOR OF β2−***/***<sup>−</sup> MICE WHEN MOTIVATIONAL STATE CHANGES**

In the second sessions of 3Ch task, all animals were food deprived except for the Social vs. Object group. As animals didn't need to be deprived (no food involved), this condition allowed us to control the repetition effect between the first and the second 3Ch exposures. There was no repetition effect between the first and the second Social vs. Object experiment for the time in contact with rewards [Social: repetition effect: *F(*1*,* <sup>14</sup>*)* = 1*.*364, NS; genotype effect: *F(*1*,* <sup>14</sup>*)* = 0*.*194, NS, Object: repetition effect: *F(*1*,* <sup>14</sup>*)* = 1*.*859, NS; genotype effect: *F(*1*,* <sup>14</sup>*)* = 3*.*30, NS].

In all conditions (Social vs. Food, Food vs. Object), when the motivational state changed after food deprivation, there was no difference between WT and β2−*/*<sup>−</sup> mice [genotype effect: S vs. F: *F(*1*,* <sup>14</sup>*)* = 2*.*31, NS; F vs. O: *F(*1*,* <sup>14</sup>*)* = 0*.*043, NS], but there was a reward effect [S vs. F: *F(*1*,* <sup>15</sup>*)* = 289*.*61, *p <* 0*.*0001; F vs. O: *<sup>F</sup>(*1*,* <sup>14</sup>*)* <sup>=</sup> <sup>287</sup>*.*65, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*0001]. As expected, WT and <sup>β</sup>2−*/*<sup>−</sup> mice spent most of their time in contact with the Food reward as compared with the Social reward (**Figure 3A** and **Table 1**, WT:*<sup>z</sup>* = −2*.*36, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*01; <sup>β</sup>2−*/*−: *<sup>z</sup>* = −2*.*66, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*007), or with Object rewards (**Figure 3A** and **Table 1**, WT: *z* = -2.52, *p* = 0*.*011; <sup>β</sup>2−*/*−: *<sup>z</sup>* = −2*.*38, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*01). In addition, we showed that both WT and β2−*/*<sup>−</sup> mice spent more time in contact with social than object rewards during the second session of three chamber task [**Figure 3B** right panel, genotype effect: *F(*1*,*14*)* = 0*.*74, *p* = NS, Reward effect: *F(*1*,*14*)* = 32*.*84, *p <* 0*.*0001; WT: *z* = −2*.*52, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*011, <sup>β</sup>2−*/*−: *<sup>z</sup>* = −2*.*38, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*017].

As previously, we showed that there was no difference in the number of entries in each compartment (data not shown). These results showed that even if mice spent more time in contact with the food rewards, they did not neglect the other rewards.

Regarding the emission of calls, only the Social vs. Food (**Figure 3C**) and the Social vs. Object conditions (**Figure 3D**, right panel) showed a reward effect when animals were food deprived [S vs. F: *F(*1*,* <sup>14</sup>*)* = 26*.*81, *p* = 0*.*0001, S vs. O: *F(*1*,* <sup>14</sup>*)* = 22*.*17, *p* = 0*.*0003] but no genotype effect [*F(*1*,* <sup>14</sup>*)* = 2*.*21, NS] or interaction reward × genotype [S vs. F: *F(*1*,* <sup>14</sup>*)* = 1*.*91, NS, S vs. O: *F(*1*,* <sup>14</sup>*)* = 0*.*53, NS]. The amount of calls emitted in the Food vs. object condition was similar for both genotypes for both rewards, even if WT mice showed a marginally significant trend to emit more USVs when in contact with the novel object than with food (F vs. O: WT: *z* = −1*.*85, *p* = 0*.*06). Actually in WT and β2−*/*<sup>−</sup> mice, the ratio USVs/Time in contact was always higher with social rewards than with food rewards (S vs. F: WT: *<sup>z</sup>* = −2*.*36, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*01; <sup>β</sup>2−*/*−: *<sup>z</sup>* = −2*.*66, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*007; S vs. O: WT: *<sup>z</sup>* = −2*.*19, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*02; <sup>β</sup>2−*/*−: *<sup>z</sup>* = −2*.*54, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*01).

We observed that food deprivation altered some USV features, like peak frequency and duration of calls. In both WT and β2−*/*<sup>−</sup> mice there was a significant reduction between non-deprived and deprived conditions in the mean peak frequency for both Social vs. Food [data not shown, Social: condition effect: *F(*1*,* <sup>15</sup>*)* = 17*.*47, *p* = 0*.*0009, genotype effect: *F(*1*,* <sup>14</sup>*)* = 3*.*40, NS, interaction condition × genotype: *F(*1*,* <sup>15</sup>*)* = 0*.*017, NS; Food: condition effect: *F(*1*,* <sup>13</sup>*)* = 9*.*05, *p* = 0*.*01, genotype effect: *F(*1*,* <sup>13</sup>*)* = 4*.*12, *p* = NS, interaction condition × genotype: *F(*1*,* <sup>13</sup>*)* = 0*.*28, NS], and Food vs. Object conditions [data not shown, Food: condition effect: *F(*1*,* <sup>14</sup>*)* = 5*.*33, *p* = 0*.*03, genotype effect: *F(*1*,* <sup>14</sup>*)* = 2*.*49, *p* = NS, interaction condition × genotype: *F(*1*,* <sup>14</sup>*)* = 0*.*41, NS; Object: condition effect: *F(*1*,* <sup>13</sup>*)* = 1*.*28, *p* = 0*.*27, genotype effect: *F(*1*,* <sup>13</sup>*)* = 0*.*44, NS, interaction condition × genotype: *F(*1*,* <sup>13</sup>*)* = 1*.*08, NS]. Indeed, during Social vs. food, WT mice showed a significant reduction of Pf mean (15.79%) when in

contact with the social reward (*U* = 7, *p* = 0*.*02), but not with the food reward (*<sup>U</sup>* <sup>=</sup> 15, NS). In addition, <sup>β</sup>2−*/*<sup>−</sup> mice showed a significant reduction of Pf mean (16.33%) when in contact with the social (*U* = 14, *p* = 0*.*01) and with the food rewards (16.67%; *U* = 14, *p* = 0*.*0.3). In the Food vs. Object condition, both WT and β2−*/*<sup>−</sup> mice showed a significant decrease in Pf mean when in contact with the Food reward (respectively 12.84% *U* = 11, *p* = 0*.*02, 19.96% *U* = 9, *p* = 0*.*01). Neither WT nor β2−*/*<sup>−</sup> mice showed such a decrease when in contact with the object reward (WT: *<sup>U</sup>* <sup>=</sup> 27, NS; <sup>β</sup>2−*/*−: *<sup>U</sup>* <sup>=</sup> 12, NS).

#### **IMPORTANCE OF SOCIAL FEEDBACK**

When mice were tested in SIT 1 month after the last session of 3Ch, we observed the typical phenotype of β2−*/*<sup>−</sup> mice (Granon et al., 2003; Avale et al., 2011; Serreau et al., 2011; De Chaumont et al., 2012). β2−*/*<sup>−</sup> mice spent more time in contact with the conspecifics than WT mice (**Figure 4A**,*U* = 24.5, *p <* 0*.*0001) and showed increased follow behaviors (**Figure 4A**, *U* = 28, *p <* 0*.*0001).

We wanted to know why we observed little, if any, difference in the 3Ch task while in SIT, WT and β2−*/*<sup>−</sup> mice behave very differently. Thus, we directly compared WT and β2−*/*<sup>−</sup> mice behavior and USVs in these two social conditions that trigger different "levels" of social reward. Indeed, the SIT provides full social contact (physical contact, movements of both mice, visual, olfactory and auditory feedbacks) while the three chamber task provides only limited amount of social information (visual, olfactory and auditory, with contact limited to nose pokes). We showed that the level of social information impacted the time spent in contact with social reward [**Figure 4B**, condition effect: *<sup>F</sup>(*1*,* <sup>31</sup>*)* <sup>=</sup> <sup>111</sup>*.*2, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*0001]. Indeed, both WT and <sup>β</sup>2−*/*<sup>−</sup> mice spent more time in social contact during the SIT than during the 3Ch task (WT: *<sup>z</sup>* = −2*.*79, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*0052; <sup>β</sup>2−*/*−: *<sup>z</sup>* = −3*.*62, *p* = 0*.*0003).

Like observed in our previous experiments, β2−*/*<sup>−</sup> mice spent significantly more time in contact with the conspecific than WT mice in SIT (*U* = 24.5, *p <* 0*.*0001), but not in the 3Ch (*U* = 100, NS), as illustrated in **Figure 4**.

We also analyzed USVs emitted during both SIT and 3Ch conditions (**Figure 4C**). We showed that there was no difference between WT and β2−*/*<sup>−</sup> mice but call rate (number of USVs per min) varied between conditions [genotype effect, *F(*1*,* <sup>31</sup>*)* = 0*.*007, *p* = 0*.*93, NS; condition effect, *F(*1*,* <sup>31</sup>*)* = 24*.*62, *p <* 0*.*0001, interaction genotype × condition, *F(*1*,* <sup>31</sup>*)* = 0*.*023, NS]. WT and β2−*/*<sup>−</sup> mice emitted drastically more USVs during the SIT (WT: 61.9 <sup>±</sup> 10.76, <sup>β</sup>2−*/*−: 64.77 <sup>±</sup> 22.13 USVs per min) than during the 3Ch task (WT: 3*.*11 ± <sup>0</sup>*.*35, <sup>β</sup>2−*/*−: 2*.*<sup>28</sup> <sup>±</sup> <sup>0</sup>*.*25 USVs per min). Furthermore, in the SIT, there was a positive and significant correlation between the time in social contact and the number of USVs emitted for WT (**Figure 5**), (*r*s = 0.821, *n* = 16, *p* = 0*.*0015), but not for <sup>β</sup>2−*/*<sup>−</sup> mice (*r*<sup>s</sup> <sup>=</sup> 0.434, *<sup>n</sup>* <sup>=</sup> 17, NS). There was no such correlation in the 3Ch task for any genotype (WT: *<sup>r</sup>*<sup>s</sup> <sup>=</sup> 0.099, *<sup>n</sup>* <sup>=</sup> 16, NS; <sup>β</sup>2−*/*−: *<sup>r</sup>*<sup>s</sup> <sup>=</sup> 0.314, *<sup>n</sup>* <sup>=</sup> 17, NS).

# **CONTROL MEASURES**

## *Olfactory tests*

*Experiment 1.* β2−*/*<sup>−</sup> mice spent significantly less time sniffing odors, whatever it was than WT mice [genotype effect: *F(*1*,* <sup>47</sup>*)* = 21*.*32, *P <* 0*.*0001]. For both genotypes no differences were detected between global time spent sniffing water, orange or urine [condition effect: *F(*2*,* <sup>47</sup>*)* = 1*.*49, *P* = NS; interaction genotype × condition: *F(*2*,* <sup>94</sup>*)* = 0*.*647, *P* = NS, data not shown]. However, when comparing the last exposure to water and the first exposure to orange, both WT and β2−*/*<sup>−</sup> mice reacted to the change (parametric *t*-test *p* = 0*.*056 and *p* = 0*.*028, respectively). When comparing the last exposure to orange and the first exposure to urine, only WT mice reacted to the change (parametric *t*-test *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*046), while <sup>β</sup>2−*/*<sup>−</sup> mice did not show significant difference

**FIGURE 4 | Effect of social feedback. (A)** Percentage of time spent in contact to, and in following the conspecific during the social interaction task (SIT). **(B)** Percentage of time spent in contact with the conspecific in the social interaction task (SIT) and in the 3Ch task (3Ch). **(C)** Number of ultrasonic vocalizations—USVs—emitted in each condition. Datas are presented for WT and <sup>β</sup>2−*/*<sup>−</sup> mice as mean <sup>±</sup> SE. ∗∗∗*<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*0001 for Mann-Whitney paired comparisons.

(*p* = 0*.*1), likely because of a large inter-individual variability which could be associated with their hyperactive phenotype.

*Experiment 2.* The second olfactory test was to check the interest of β2−*/*<sup>−</sup> mice in social odors such as pheromones. We compared their behavior when exposed to litter taken from social cages vs. clean litter. There was no difference between WT and <sup>β</sup>2−*/*<sup>−</sup> [genotype effect: *<sup>F</sup>(*1*,* <sup>47</sup>*)* <sup>=</sup> <sup>0</sup>*.*042, *<sup>p</sup>* <sup>=</sup> NS], and both genotypes spent drastically more time in contact with social than with clean litter [data not shown, condition effect: *F(*1*,* <sup>47</sup>*)* = 156*.*03, *P <* 0*.*0001; interaction genotype × condition: *F(*1*,* <sup>47</sup>*)* = 2*.*05, *<sup>p</sup>* <sup>=</sup> NS]. These control experiments showed that <sup>β</sup>2−*/*<sup>−</sup> mice exhibit similar interest for social olfactory cues as WT animals.

To control for putative auditory defects in β2−*/*<sup>−</sup> mice, we analyzed their ABR. Results showed no difference between WT and <sup>β</sup>2−*/*<sup>−</sup> mice [*F(*1*,* <sup>28</sup>*)* <sup>=</sup> <sup>0</sup>*.*139, NS] and both genotypes exhibited auditory thresholds that were function of the tone frequency [*F(*8*,* <sup>28</sup>*)* = 115*.*13, *p <* 0*.*0001] and that were similar to previously published ABR thresholds (Buran et al., 2010). Therefore, the lack of correlation between the number of USVs and the time spent in social contact during the SIT in β2−*/*<sup>−</sup> mice is unlikely to be due to auditory problems (**Figure 6**).

# **DISCUSSION**

The aim of the present study was to determine whether, and if yes how, mice rank natural rewards like food, exploration and social contact. In addition, we wondered whether being in the 3Ch task, i.e. a task in which the test mouse can make choices without interference from another mouse, would impact on this rank and how social information were integrated to choose between rewards. We further asked whether β2nAChRs, known to be necessary for showing adapted social interaction, would be involved in such reward ranking.

We focused here on three main natural rewards in rodents: novelty exploration, interaction with an unkown conspecific, and food consumption. By contrast with our previous studies, we used a three chamber task (3Ch task) to assess the

**FIGURE 6 | Auditory brainstem responses (ABR) elicited by WT and β2−***/***<sup>−</sup> mice.** Graph showed the auditory brainstem responses for WT and β2−*/*<sup>−</sup> mice for different pure frequency sounds (logarithmical scale) at different thresholds (dB).

rank spontaneously established by mice between natural rewards in a safe environment. Indeed, in the 3Ch task these rewards competed two by two, and mice can freely move from one reward to another without any interference from a conspecific. We then modified food motivation by food deprivation, and assessed the ability of mice to adapt to their motivational state. In both situations, we compared behaviors and ultrasonic vocalizations.

Previous works showed that when these three rewards competed in socially unpredictable environment like SIT, β2−*/*<sup>−</sup> mice showed impaired organization in their choices toward the rewards, namely, they exhibited difficulty in switching between the different rewards (Granon et al., 2003; Serreau et al., 2011). In such environment, the tested mouse faced an unknown conspecific -the visitor mouse- which moved freely and showed reciprocal and non-aggressive social contact. We further showed that β2−*/*<sup>−</sup> mice exhibited decision-making defects and lacked behavioral flexibility, whether the rewards were of social nature (De Chaumont et al., 2012), or not (Granon et al., 2003). However, in the SIT context, the visitor mouse strongly interacted with the test mouse, thus potentially affecting its decisions. β2−*/*<sup>−</sup> mice also exhibited a high level of dominance toward the visitor mouse (Coura et al., 2013) and were less likely to allow the visitor mouse approaching (Serreau et al., 2011; De Chaumont et al., 2012). To circumvent this issue, we used here a safe and predictable environment, the three chambers environment (Crawley, 2000; Moy et al., 2004; Nadler et al., 2004; Silverman et al., 2010a). In this task, the test mouse (either a WT or a β2−*/*<sup>−</sup> mouse) was the only decision-maker, as the stimulus mouse was kept under a cup during "Social" reward sessions. This test therefore allowed us to establish the natural preference exhibited by the tested mice. In this context, our current results show that the rank established between rewards was adaptive for both genotypes: it changed similarly in WT and β2−*/*<sup>−</sup> mice when the motivation level of mice changed, i.e., when animals were food deprived. As compared to a socially more unpredictable environment (Serreau et al., 2011), the 3Ch experiment revealed that the establishment of a rank between competing motivations was strongly modulated by the social risk level of the task, or by the putative interference from an unkown adult male mouse. Indeed, we show here that non food-deprived WT mice ranked their motivations in a specific order, from the most preferred reward to the less preferred one: social *>* novel object *>* food. This result confirmed that novelty exploration is one of the preferred natural rewards in mice (Avale et al., 2011; Bourgeois et al., 2012). Whether this was reinforced by the paucity of the laboratory rearing in standard cages, i.e., containing no items, remains to be investigated (Van Praag et al., 2000; Kulesskaya et al., 2011). It is noticeable that scoring the number of entrance in a specific compartment was not sufficient as this measure did not allow discrimination between the different rewards, contrary to the scoring of duration of contact with each reward.

Our results revealed that non-food deprived β2−*/*<sup>−</sup> mice, like WT mice, chose the social reward in the first place. However, they spent equal time in contact with the novel object and the food. As we previously showed that β2−*/*<sup>−</sup> mice are not more -or less- sensitive to food reward than WT (Serreau et al., 2011), the current data may suggest that for β2−*/*<sup>−</sup> mice, food can be considered as an interesting novel object, when the food motivation is low.

The number of USVs emitted was significantly higher when mice faced the social reward than when they faced any of the two other rewards. This was true for both WT and β2−*/*<sup>−</sup> mice. If this measure obviously increased when having two mice instead of one, it also confirmed the stimulating effect of the social context on USV's emission (Vignal et al., 2005; Arriaga et al., 2012; Chabout et al., 2012). It is interesting to note that having two mice instead of one did not simply multiply by two the number of USVs. Indeed, comparing the number of USVs in the 3Ch (social condition) and in the SIT, two experimental conditions in which a dyad of mice was recorded, clearly showed that the number of animals was not a critical factor. By contrast, the type of social contact they can have was likely to be a major factor. Indeed, we showed here that in the 3Ch task, the distribution of USVs was statistically not correlated to the time spent in contact with the rewards. This was not the case in the SIT during which both mice exchanged not only olfactory, auditory and visual information but also could touch and react to each other. Our ABR and olfactory control experiments showed that it is unlikely that differences between WT and β2−*/*<sup>−</sup> mice were due to difference in integrating olfactory or auditory information, although it must be noticed that ABR did not measure auditory responses for the highest USVs.

The behavioral results also showed that in the 3Ch environment, both genotypes reacted similarly to the food deprivation and re-organized their reward ranking when their motivation for food changed: they both decreased the time spent in contact with social reward or with the novel object, and increased drastically the time spent in contact with the food, as expected. This showed that both groups were similarly sensitive to food deprivation and adapted their reward preference to their motivation level. However, the number of USVs emitted in contact of each reward was similar in food deprived and non-deprived conditions. Both β2−*/*<sup>−</sup> and WT mice emitted more USVs in the social compartment. However, both groups showed similar alteration in the mean frequency of emitted USVs (about 5–10% lower) after food deprivation. It has been shown that when rats (Brudzynski, 2007) or mice (Chabout et al., 2012) were subjected to a negative emotional context, such as foot shock of restrain stress exposure, the frequency of their USVs was lower than when they were exposed to a positive or rewarding stimulus. Our current results therefore suggest that food deprivation induced a slightly negative emotional state that was reflected by the frequency of the USVs emitted. However, USVs thus emitted did not discriminate between the different rewards, suggesting that the lower emotional state induced by food deprivation was not counterbalanced by other types of rewards.

Notably, β2−*/*<sup>−</sup> mice exposed to the 3Ch task showed no difference with WT mice concerning their behavior and USVs. However, when these animals were subsequently subjected to the SIT, they exhibited a drastic behavioral impairment: they specifically showed increased social contact duration and follow behavior. These results, that are similar to our previously data obtained in the SIT in animals that were not exposed to other tasks before (Granon et al., 2003; Avale et al., 2008; De Chaumont et al., 2012; Coura et al., 2013), highlight two important points. First, although β2−*/*<sup>−</sup> mice exhibited normal motivation for social reward when tested in the 3Ch, they showed altered social interaction when the social environment was unpredictable. Second, our results emphasized the importance of social feedback. We showed that in the 3Ch task, the number of USVs was drastically and significantly reduced, as compared to that emitted during the SIT. Furthermore, we showed that in WT mice, this number was not correlated to the duration of social contact, although this was the case in the SIT. A large part of USV emission in the SIT was therefore likely to be associated with the social feedback received by the dyad. Another alternative hypothesis, not exclusive with the first one, is that USVs accompanied the attentional load generated by the task. This load would be higher in unpredictable tasks and so would be the number of USVs.

The lack of correlation between the duration of social contact and USVs in β2−*/*<sup>−</sup> mice subjected to the SIT could mean that these mice were not sensitive to the social reward. However, results obtained in the 3Ch task indicate that this is unlikely. Rather, the lack of correlation may indicate that β2−*/*<sup>−</sup> mice did not integrate USVs in social behavior.

Our current data revealed that the 3Ch task and the SIT are both very complementary in the study of mice social interaction. The 3Ch task is very useful to ensure that animals exhibit normal preference for a social conspecific, as compared to other types of rewards. It can also be used to show that animals exhibit normal social approach or interest. However, the fact that very few USVs were obtained in this task limits its use. The SIT, by contrast, and because it allows social feedback from both conspecifics, can be used for studying behavioral social patterns and strategies as well as how acoustic communication is integrated in these patterns. Placing animals in such environment, although it remains quite different from a naturalistic context, allows to study how mice face risk and, potentially threat, from another unknown individual, or develop dominance. Using the SIT, we previously showed that re-expressing the beta2-containing nAChRs into the prefrontal cortex of β2−*/*<sup>−</sup> mice was sufficient to restore normal pattern of social interaction (Avale et al., 2011). Whether this behavioral restoration would be associated with restoration of USV-social contact duration correlation and whether this correlation is necessary for the restoration of social flexibility remains to be unraveled.

What determines social behaviors remains unclear. However, the fact that they were conserved during evolution process and shared by most animal species suggests that there are great benefits to them. Here, although we did not compare the three rewards at the same time, we demonstrated that mice establish a rank among competing natural rewards. Social reward was the preferred one, if mice have been socially deprived for a few weeks. We also provide clear evidence that β2-containing nAChRs are not involved in motivational ranking *per se*, as β2−*/*<sup>−</sup> mice showed normal reward ranking in a safe social situation. These receptors were also not involved in the monitoring of internal states as β2−*/*<sup>−</sup> mice adapted, like WT mice to food or social deprivation. We also highlight that social feedback and acoustic communication are related. It remains unclear, however, if social feedback impacts communicational abilities or, in the contrary, if alteration of USV features impact social behaviors.

## **AUTHOR CONTRIBUTION**

Jonathan Chabout and Arnaud Cressant analyzed the data, performed the statistical analyses. Jonathan Chabout carried out the behavioral experiments and participated in the writing process. Xian Hu and Arnaud Cressant carried out some behavioral experiments. Jean-Marc Edeline and Sylvie Granon contributed to interpretation and writing process. Sylvie Granon designed, contributed, and coordinated the experimental work.

### **REFERENCES**


*Sci. U.S.A.* 109, 19878–19879. doi: 10.1073/pnas.1216902109.


30, 7587–7597. doi: 10.1523/ JNEUROSCI.0389-10.2010


mice. *Autism Res.* 1, 147–158. doi: 10.1002/aur.22


to the symptoms of autism. *Brain Pathol.* 17, 448–459. doi: 10.1111/j.1750-3639.2007.00096.x


system do? *Science* 318, 606–610. doi: 10.1126/science.1142998


in the BTBR T+tf/J mouse model of autism. *PLoS ONE* 3:e3067. doi: 10.1371/journal.pone.0003067


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 14 April 2013; accepted: 27 July 2013; published online: 27 August 2013. Citation: Chabout J, Cressant A, Hu X, Edeline J-M and Granon S (2013) Making choice between competing rewards in uncertain vs. safe social environment: role of neuronal nicotinic receptors of acetylcholine. Front. Hum. Neurosci. 7:468. doi: 10.3389/fnhum. 2013.00468*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Chabout, Cressant, Hu, Edeline and Granon. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Oxytocin and socioemotional aging: Current knowledge and future trends

#### *Natalie C. Ebner <sup>1</sup> \*, Gabriela M. Maura1, Kai MacDonald2, Lars Westberg3 and Håkan Fischer 4,5*

*<sup>1</sup> Department of Psychology, University of Florida, Gainesville, FL, USA*


*<sup>4</sup> Department of Psychology, Stockholm University, Stockholm, Sweden*

*<sup>5</sup> Aging Research Center, Karolinska Institute, Stockholm, Sweden*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

*Reviewed by:*

*Gregory R. Samanez-Larkin, Yale University, USA Brian W. Haas, University of Georgia, USA*

#### *\*Correspondence:*

*Natalie C. Ebner, Department of Psychology, University of Florida, Psychology Building, 945 Center Drive, PO Box 112250, Gainesville, FL 32611, USA e-mail: natalie.ebner@ufl.edu*

The oxytocin (OT) system is involved in various aspects of social cognition and prosocial behavior. Specifically, OT has been examined in the context of social memory, emotion recognition, cooperation, trust, empathy, and bonding, and—though evidence is somewhat mixed-intranasal OT appears to benefit aspects of socioemotional functioning. However, most of the extant data on aging and OT is from animal research and human OT research has focused largely on young adults. As such, though we know that various socioemotional capacities change with age, we know little about whether age-related changes in the OT system may underlie age-related differences in socioemotional functioning. In this review, we take a genetic-neuro-behavioral approach and evaluate current evidence on age-related changes in the OT system as well as the putative effects of these alterations on age-related socioemotional functioning. Looking forward, we identify informational gaps and propose an *Age-Related Genetic, Neurobiological, Sociobehavioral Model of Oxytocin* (*AGeNeS-OT model*) which may structure and inform investigations into aging-related genetic, neural, and sociocognitive processes related to OT. As an exemplar of the use of the model, we report exploratory data suggesting differences in socioemotional processing associated with genetic variation in the oxytocin receptor gene (*OXTR*) in samples of young and older adults. Information gained from this arena has translational potential in depression, social stress, and anxiety-all of which have high relevance in aging—and may contribute to reducing social isolation and improving well-being of individuals across the lifespan.

#### **Keywords: oxytocin, aging, socioemotional functioning, amygdala, anterior cingulate**

Social and emotional processes and their associated genetic and neurobiological mechanisms in aging are still incompletely understood (Nielsen and Mather, 2011). In this paper we propose to combine neuroendocrine, genetic, and sociobehavioral approaches to examine the role of the oxytocin (OT) system in the context of socioemotional aging. Aspects of the OT system warranting investigation include: (1) changes in endogenous and dynamic OT levels; (2) changes in systems which directly impact OT function (i.e., gonadal hormones); (3) genetic variation in aspects of the OT system, including the gene for oxytocin (*OXT*), its receptor (*OXTR*), and the related CD38 system; (4) changes in OT-rich neural regions; (5) the effect of exogenous OT. There is increasing evidence that OT plays a significant role in many of the socioemotional capacities that undergo age-related changes. However, to date, very little is known about the role of OT in human aging (Huffmeijer et al., 2012). Thus, it will be crucial for future research to clarify links between age-related changes in the aforementioned aspects of the OT system and changes in neural processing and subsequent alterations in experience as well as behavior in socioemotional domains in older compared to young adults.

To foreshadow, this focused review conceptually integrates two lines of research. First, we summarize evidence for age-associated changes in socioemotional capacities (Isaacowitz et al., 2007; Ruffman et al., 2008; Scheibe and Carstensen, 2010). Second, we review evidence for the involvement of OT in socioemotional functioning (Bartz et al., 2011; Meyer-Lindenberg et al., 2011; Van IJzendoorn and Bakermans-Kranenburg, 2012). Synthesizing these two lines of work, we present an *Age-Related Genetic, Neurobiological, Sociobehavioral Model of Oxytocin* (*AGeNeS-OT model*) which may stimulate questions and organize investigations into the role of OT in socioemotional aging. As an example of the use of the *AGeNeS–OT* model, we report preliminary data suggesting neural and behavioral differences in socioemotional processing associated with genetic variations in *OXTR* in samples of young and older adults. We conclude by suggesting future directions for research implied by the model. Ultimately, these investigations will increase our understanding of the role of OT in aging and will have the potential for generating new interventions to improve health and well-being.

# **SOCIOEMOTIONAL FUNCTIONING AND AGING**

From life's beginning, humans are confronted with critical, survival-enhancing socioemotional stimuli related to self and others. To maintain successful social interactions and avoid the negative consequences of social isolation (Baumeister and Leary, 1995; Norman et al., 2011), we learn to quickly and accurately process, respond to, and remember social cues (Baron-Cohen et al., 2000; Grady and Keightley, 2002; Adolphs, 2003). Socioemotional functioning may become particularly relevant in old age when-due to the experience of increasing physical ailment, dependency, and age-related social losses-the experience of social isolation often increases with negative effects on physical and mental health (Cornwell and Waite, 2009).

The extant literature suggests a mixed picture of age-related changes in socioemotional capabilities: Some capacities (e.g., emotion regulation, emotional problem solving) improve with age, whereas other skills (e.g., recognition of emotions in others) decline (cf. Scheibe and Carstensen, 2010). In particular, across various studies, older compared to young adults show increased emotion regulation capacity (Carstensen, 2006; Blanchard-Fields et al., 2007; Riediger et al., 2009; Scheibe and Blanchard-Fields, 2009; Voelkle et al., 2013) and greater confidence in this ability (Lawton et al., 1992; Gross and Levenson, 1997; Kessler and Staudinger, 2009). The majority of older adults are well-adjusted emotionally and report relatively high levels of affective wellbeing and emotional stability as documented in cross-sectional (Carstensen et al., 2000) as well as longitudinal (Carstensen et al., 2011) studies (see also Charles et al., 2001; Teachman, 2006). In addition, older compared to young adults are at least equally (and often more) effective in their ability to regulate their emotional experiences, autonomic arousal, and outward display of negative emotions in language and faces when instructed to do so (Kunzmann et al., 2005; Magai et al., 2006; Phillips et al., 2008), and show improved socioemotional problem solving capacity (Blanchard-Fields et al., 2007).

At the same time, older adults often show increased difficulties in accurate recognition of social and emotional cues (for reviews see Isaacowitz et al., 2007; Ruffman et al., 2008; see also Ebner and Johnson, 2010; see **Figure 1A**). Recent functional magnetic resonance imaging (fMRI) data suggests that these difficulties are associated with greater activity in dorsomedial prefrontal cortex (dmPFC) in older compared to young adults during facial emotion reading, particularly for angry expressions (Williams et al., 2006; Keightley et al., 2007; Ebner et al., 2012; see **Figure 1C**). This association comports with previous evidence that dmPFC is involved in complex processing and cognitive and emotional control (Amodio and Frith, 2006). Another age-related change in socioemotional functioning is that older compared to young adults demonstrate more interpersonal trust (List, 2004; Castle et al., 2012). This change may be due to the difficulty older adults often have in "reading" the emotions of others, as suggested by recent findings that older compared to young adults are less proficient at detecting lies, mediated by deficits in emotion recognition (Ruffman et al., 2012). With respect to changes in memory, there is evidence that the majority of older adults experience declines in remembering critical socioemotional cues, including names (Crook et al., 1993; Verhaeghen and Salthouse, 1997) and faces (Bartlett et al., 1989; Grady et al., 1995; Ebner and Johnson, 2009; see **Figure 1B**). Finally, in terms of social motivation, there is robust evidence that older adults are more avoidance-oriented and less approach-oriented than young adults (Ebner et al., 2006; Freund, 2006; Nikitin et al. in revision).

Importantly, the mechanisms underlying these age-related changes in socioemotional functioning are not well-understood yet. One potential explanation is differences in visual processing (Isaacowitz et al., 2006; Ebner et al., 2011), perhaps as a function of age-related changes in motivation (Mather and Carstensen, 2005; Carstensen, 2006; Samanez-Larkin and Carstensen, 2011). In particular, there is evidence that older compared to young adults spend more time looking at positive than negative information (Isaacowitz et al., 2006) and, when processing faces, spend less time viewing the eye region and more time viewing the mouth

(Firestone et al., 2007). This age-differential visual processing pattern may be important given that the eye vs. mouth regions of a face carry different socioemotional information (Calder et al., 2000; Ebner et al., 2011).

A complementary, mechanistic explanation for age-related changes in socioemotional function may be changes in brain structure or function in regions associated with socioemotional processing such as amygdala, PFC, insula, or fusiform gyrus (Keightley et al., 2007; Grady, 2008; Cacioppo et al., 2009; Ebner et al., 2012; see Ruffman et al., 2008; Samanez-Larkin and Carstensen, 2011; St. Jaques et al., 2013, for overviews). For instance, there is well-documented, age-related structural decline in regions such as the lateral PFC (lPFC), insula, and striatum (Raz, 2005; Raz et al., 2005). Regarding functional changes, one common finding is an age-related decrease in amygdala activation during the perception of emotional stimuli (especially negative stimuli) accompanied by an age-related increase in activity in a number of lPFC and mPFC regions (Iidaka et al., 2002; Gunning-Dixon et al., 2003; Fischer et al., 2005, 2010; Tessitore et al., 2005; but see Mather and Carstensen, 2005; Wright et al., 2006; Ebner et al., 2013).

Crucially, however, extant literature suggests that age-related differences in socioemotional processing cannot be explained solely by age-related visuoperceptual and/or neurocognitive changes (Samanez-Larkin and Carstensen, 2011). In addition, it may be that changes in socioemotional function are also linked with age-related alterations in neuroendocrine function. In particular, the neuropeptide OT appears as a particularly promising candidate, given increasing evidence of its role in socioemotional domains (Insel and Fernald, 2004; Donaldson and Young, 2008; Bartz et al., 2011; Meyer-Lindenberg et al., 2011; Norman et al., 2011). However, to date, we know very little about age-related changes in the OT system, particularly in the context of socioemotional aging (Huffmeijer et al., 2012).

# **OXYTOCIN AND SOCIOEMOTIONAL FUNCTIONING**

OT is a nine amino acid peptide, with peripheral and central functions (Gimpl and Fahrenholz, 2001). It is synthesized in magnocellular neurosecretory cells of paraventricular nuclei (PVN) and supraoptic nuclei (SON) of the hypothalamus and released through the posterior pituitary gland into the periphery (Insel, 2010). OT is also released into the brain by magnocellular dendrites (Leng and Ludwig, 2006) and by OT-releasing neurons projecting to specific brain regions such as the amygdala, hippocampus, and striatum (Kimura et al., 1992; Landgraf and Neumann, 2004; Knobloch et al., 2012). Human and animal studies combined suggest that the function of the OT system is reflected at a variety of physiological and anatomical levels, including: (1) peripheral hormone levels (i.e., plasma and saliva); (2) central hormone levels [i.e., in cerebrospinal fluid (CSF)]; (3) histological levels (i.e., presence and size of OT cells); (4) receptor levels (in OT receptor binding in defined brain regions); (5) genetic levels, or the level of "neuropeptidergic individuality" (MacDonald, 2012); i.e., polymorphisms related to *OXT* or *OXTR*, or genes related to OT release (i.e., *CD38*; Sauer et al., 2012, 2013).

In particular, accumulating evidence suggests that OT may serve as a key effector in socioemotional functioning such as emotion recognition, memory for faces, interpersonal trust, and bonding as briefly summarized next (see Bartz et al., 2011; Meyer-Lindenberg et al., 2011; Norman et al., 2011; Zink and Meyer-Lindenberg, 2012, for comprehensive overviews).

After the discovery that certain neuropeptides could be delivered intranasally to the human brain (Born et al., 2002), a number of experimental studies using intranasal OT revealed intriguing effects on diverse aspects of socioemotional functioning. For example, research in healthy adults suggests that OT impairs performance in verbal memory tasks (Ferrier et al., 1980; Heinrichs et al., 2004; but see Feifel et al., 2012), while enhancing recognition of social (i.e., faces) but not non-social stimuli (Rimmele et al., 2009; see also Heinrichs et al., 2004), especially for neutral and angry compared to happy faces (Savaskan et al., 2008). Furthermore, intranasal administration of OT increases overall gaze time toward faces (Guastella et al., 2008; Andari et al., 2010; Averbeck, 2010; Gamer et al., 2010) and increases emotion recognition, specifically of happy and fearful faces (and under certain conditions angry faces; see Shahrestani et al., 2013, for a recent review).

In addition, recent studies have shown that intranasal OT increases facial trustworthiness and attractiveness ratings (Theodoridou et al., 2009) as well as interpersonal trust and the willingness to take social risks (Kosfeld et al., 2005; Baumgartner et al., 2008; Phan et al., 2010). These effects of OT on trust seem to be particularly pronounced in positive social interactions (Zak et al., 2005; Mikolajczak et al., 2010) and with respect to ingroup vs. out-group members (Van IJzendoorn and Bakermans-Kranenburg, 2012). Moreover, these effects seem moderated by interindividual differences (Rockliff et al., 2011; but see Guastella et al., 2013), including genetic polymorphisms associated with OT function (Riedl and Javor, 2011; see also Rodrigues et al., 2009; MacDonald, 2012, for reviews).

Besides these effects on facial processing and trust, intranasal OT has been shown to influence social approach behavior, attachment, bonding, and social rejection with associated health benefits (Ditzen et al., 2009; Gouin et al., 2010; Scheele et al., 2012; Schneiderman et al., 2012; Fekete et al., 2013). For example, intranasal OT increased positive relative to negative behaviors during a laboratory couple conflict and reduced post-conflict cortisol levels (Ditzen et al., 2009). This potential stress reducingeffect of OT has been further documented by evidence that participants with increased plasma OT healed faster and had a greater number of positive interactions with partners during a 24-h hospital stay (Gouin et al., 2010; see also Kéri and Kiss, 2011; Kiss et al., 2011; see Taylor et al., 2006, for a discussion of OT's role during relaxation vs. stress; see also Feldman et al., 2011).

An ever-expanding body of neuroimaging data suggests that OT's effects on socioemotional functioning are due to its attenuation of the neural circuitry for anxiety and aversion and its activation of social reward neural networks (cf. Yoshida et al., 2009; Zink and Meyer-Lindenberg, 2012). In particular, a number of studies have provided evidence that the amygdala might be a key structure for the mediation of the social-cognitive effects of OT (Kirsch et al., 2005; Domes et al., 2007a; Petrovic et al., 2008; Singer et al., 2008; Labuschagne et al., 2010; Riem et al., 2011; Zink and Meyer-Lindenberg, 2012; cf. Huffmeijer et al., 2012; but see Domes et al., 2010). For example, OT attenuates amygdala response to fear-inducing stimuli (Kirsch et al., 2005). Baumgartner et al. (2008; see also Kosfeld et al., 2005; Mikolajczak et al., 2010) provide evidence that OT reduced betrayal aversion to breaches of trust via a reduction in bilateral amygdala activation and midbrain regions and greater ventral striatum and orbitofrontal cortex (OFC) activity. Furthermore, there are suggestions of specific modulatory influences of OT on subregions within the amygdala during processing of socioemotional information (Gamer et al., 2010; see also Huber et al., 2005; Viviani et al., 2011; Knobloch et al., 2012). These central effects, importantly, occur in interaction with a network of other neurochemicals including estrogen, dopamine, and serotonin (Riedl and Javor, 2011).

Thus there are suggestions in the literature that OT increases approach-related behaviors, while decreasing withdrawal-related behaviors (Kemp and Guastella, 2010). At the same time, however, there is evidence suggesting that OT may play a somewhat more complex role in social behavior than simply directing approach vs. avoidance behavior and/or attentional biases to positive and negative information, respectively. Rather, OT may increase social engagement, salience of social agents, and social value of processed information, largely independent of valence (Shamay-Tsoory et al., 2009; Tops, 2009; Shamay-Tsoory, 2010; Bartz et al., 2011). In line with this suggestion, brain regions such as the ventral tegmentum, PFC, nucleus accumbens, and insula associated with the social-reward neural network have shown sensitive to OT (Balleine et al., 2007; Riem et al., 2011; Wittfoth-Schardt et al., 2012; Groppe et al., 2013; Scheele et al., 2013).

The central effects of OT are mediated by its G-proteincoupled receptor, located on a variety of tissues including the brain, heart, kidney, and uterus (Loup et al., 1991; Gimpl and Fahrenholz, 2001). Polymorphisms of the gene encoding the OT receptor, *OXTR*, have been shown to contribute to individual differences in various social phenotypes (cf. Gimpl and Fahrenholz, 2001; Meyer-Lindenberg et al., 2011; Ebstein et al., 2012; Zink and Meyer-Lindenberg, 2012; Kumsta et al., 2013; Westberg and Walum, 2013). For example, *OXTR* single nucleotide polymorphisms (SNPs) have been associated with lower positive affect (Lucht et al., 2009), lower levels of responsiveness of mothers to their toddlers (Bakermans-Kranenburg and van IJzendoorn, 2008), lower empathy scores and increased stress reactivity (Rodrigues et al., 2009), non-verbal displays of prosociality (Kogan et al., 2011), and pair-bonding (Walum et al., 2012). *OXTR* SNPs have also been studied in relation with autism spectrum disorder (ASD; see Ebstein et al., 2012, for a review), with evidence that they contribute to risk for some phenotypes observed in ASD (Egawa et al., 2012; but see Tansey et al., 2010).

Taken together, this review highlights the importance of simultaneously considering behavioral, neural, and genetic perspectives when examining OT's role in socioemotional functioning, as will be discussed in more detail below (see **Figure 2**). In addition, it raises five important caveats and informational gaps.

First, some of the effects associated with OT are inconsistent and come from small, homogeneous samples, creating a

need for replication of key findings in larger, more representative samples.

Second, many of OT's effects seem to vary by individual difference variables such as the level of social proficiency (Bartz et al., 2011; but Guastella et al., 2013).

Third, there is increasing evidence suggesting that the effects of OT are dependent on context (Domes et al., 2007b) and influenced by early life experiences (see MacDonald, 2012, for a review). For example, women (Heim et al., 2008) and men (Meinlschmidt and Heim, 2007) who were abused or neglected as children showed altered OT system sensitivity as adults (e.g., decreased CSF level of OT; see also Winslow et al., 2003; Fries et al., 2005; but see Anderson, 2006; cf. MacDonald, 2012, for a review).

Fourth, due to both theoretical safety concerns using OT in women as well as the complexity introduced by OT's sex-specific effects, a large majority of studies conducted so far refer to men exclusively, even though there are growing indications that some of OT's effects may differ by sex (Savaskan et al., 2008; Guastella et al., 2009; Domes et al., 2010; Marsh et al., 2010; cf. MacDonald, 2012). This sex-specific pattern raises the possibility that the effects of OT on social cognition may be differentially regulated by gonadal steroids (estrogen and testosterone) or other sex-specific biological factors (Choleris et al., 2009; Gabor et al., 2012; see also Van Anders et al., 2011; see also Weisman and Feldman, 2013).

A fifth shortcoming in the current human literature on oxytocin—critical in the present context—is that current studies have almost exclusively been conducted with young adults. Given the aforementioned evidence of age-group differences in socioemotional functioning (Scheibe and Carstensen, 2010; Samanez-Larkin and Carstensen, 2011), a comprehensive examination of aging-related aspects of the OT system (including genetic, neurobiological, and behavioral aspects) is warranted (Huffmeijer et al., 2012).

### **OXYTOCIN AND AGING**

Despite a significant need for research addressing the growing older segment of the population, research on OT and aging is scarce and inconclusive. To date, the few studies that have addressed age-related differences in the OT system almost exclusively refer to non-human species with limited applicability to humans (Quinn, 2005). Also, studies conducted to date are characterized by large methodological differences in terms of species examined, OT parameters measured, brain regions targeted, etc., which makes a direct comparison difficult and a meta-analytic approach not feasible. Most importantly, a theoretical framework for generating hypotheses regarding age-related differences in the OT system (including changes in endogenous OT physiology, function, and differential response to exogenous OT) is entirely lacking (cf. Huffmeijer et al., 2012).

**Table 1** provides a summary of the current studies on OT and aging. Whereas some studies suggest no noticeable effects of aging on the OT system (Fliers et al., 1985; Zbuzek et al., 1988; Wierda et al., 1991; Arletti et al., 1995), other studies report agerelated change (Fliers and Swaab, 1983; Melis et al., 1992, 1995; Arsenijevic et al., 1995; Parker et al., 2010). Notably, some of the studies reporting comparability of the OT system across older and


*Y, Young subjects; MA, Middle-aged subjects; O, Older subjects; M, Male; F, Female; OT, Oxytocin; PVN, Paraventricular nuclei of hypothalamus; SON, Supraoptic nuclei (SON) of hypothalamus; AVP, Arginine vasopressin; NOS, Nitric oxide synthase; CSF, Cerebrospinal fluid.*

young subjects refer to peripheral OT levels (Fliers and Swaab, 1983; Zbuzek et al., 1988; Melis et al., 1992), whereas several of the studies documenting age-related change relate to central OT levels (Fliers and Swaab, 1983; Melis et al., 1992; Arsenijevic et al., 1995; Parker et al., 2010). Thus, it is possible that aging may change OT transmission in the CNS but not in the neurohypophyseal (peripheral) system (Melis et al., 1999). A summary of the evidence reported in **Table 1** would be that current evidence does not allow yet a firm conclusion of the existence or direction of age-related changes in the OT system, leaving the question open to empirical examination.

To our knowledge, only one very recent study explicitly examined the effects of intranasal OT in a group of older men (mean age of 80 years) focusing on OT's effects on social engagement and physical health (Barraza et al., 2013). Results from this double-blind, placebo-controlled 10-day clinical trial suggested improvement in dispositional gratitude in older adults in the OT compared to the placebo group. In addition, the OT group had a slower decline in physical functioning and decreased self-reported fatigue than the placebo group. No changes in mood, cardiovascular states, or social activity and engagement patterns were observed across the study interval. Importantly, this study did not include a comparison group of young adults and did not extensively explore OT's effects on other aspects of socioemotional functioning. Thus, it is critical to follow up on these first promising findings regarding OT and aging and to conduct systematic examinations of age differences in baseline levels of OT. In addition, a comprehensive evaluation of both single-dose as well as longer-term administration of intranasal OT and its effect on socioemotional functioning in young and older men and women is warranted. Finally, these studies should take into account genetic variations related to OT.

## **OXYTOCIN AND SOCIOEMOTIONAL AGING: AGE-RELATED GENETIC, NEUROBIOLOGICAL, SOCIOBEHAVIORAL MODEL OF OXYTOCIN**

Based on the following rationale, we propose an *OT X Age*interaction (see **Figure 3C**) as the guiding working hypothesis for future research on the role of OT in socioemotional aging: As mentioned above, there is early evidence that the beneficial effects of OT in socioemotional domains (see **Figure 3A**) vary by individual factors (Bartz et al., 2011). Notably, "preexisting social impairment" seems to play a role, in that more socially impaired individuals benefit more from OT than less socially impaired individuals (Bartz et al., 2010; Guastella et al., 2010; but see Bakermans-Kranenburg and van IJzendoorn, 2013). Also there may be a "ceiling effect," a point beyond which OT cannot further improve social abilities (Bartz et al., 2011). As laid out above, older adults experience deficits in various socioemotional capacities (Scheibe and Carstensen, 2010; see **Figure 3B**), rendering them more socially impaired than young adults in some regards. Therefore, it may well be that OT is particularly beneficial in older compared to young adults (see **Figure 3C**).

However, an alternative hypothesis exists: As reported above, even though some aspects of socioemotional functioning (i.e., emotion recognition and memory for emotional information) decline with age, other aspects increase or remain stable. That is,

Oxytocin X Age interaction effect. Schematic representation of guiding working hypotheses. YA, Young adults, OA, Older adults; OT, Oxytocin, P, Placebo.

given broad evidence for a positivity effect and for healthy socioemotional functioning in old age (Carstensen, 2006; Carstensen et al., 2011), as well as some evidence for increased trustworthiness in old age (Castle et al., 2012), on average, older adults can be described as highly positive, trustworthy, and prosocial. These characteristics may be adaptive in some contexts (e.g., social interactions within close relationships) but maladaptive in others (e.g., putting aging adults at greater susceptibility to fraud). This reasoning, combined with the current lack of proof that aging is associated with declines in the OT system and mixed evidence regarding OT's effect on cognitive performance (Heinrichs et al., 2004; Feifel et al., 2012), suggest the possibility that under certain circumstances OT may have harmful effects in older adults. Given that OT is currently being investigated in clinical populations such as schizophrenia (cf. MacDonald and Feifel, 2012, 2013), comprising samples of people who are late middle-aged, a thorough investigation of age-related aspects of the OT systemincluding beneficial or detrimental effects on outcome measures in socioemotional as well as cognitive domains-will be crucial.

As summarized above, the OT system is represented at genetic, neural, and behavioral levels (Meyer-Lindenberg et al., 2011). Furthermore, each of these levels and their functional interactions are influenced by the aging process. We therefore propose for future research in the domain of OT and socioemotional aging to adopt an *Age-Related Genetic, Neurobiological, Sociobehavioral Model of Oxytocin* (*AGeNeS-OT model*; **Figure 2**). In particular, this model suggests that a comprehensive examination of the central OT system should consider interactions between OT-related genes (*OXT, OXTR, CD38;* Meyer-Lindenberg et al., 2011; Sauer et al., 2013), the brain (e.g., amygdala, frontal cortex, brainstem, ventral tegmental area; Pedersen et al., 1994; Kirsch et al., 2005; Balleine et al., 2007; Baumgartner et al., 2008; Gamer et al., 2010), and behavior (e.g., social memory, emotion identification, approach/avoidance biases; Rimmele et al., 2009; Domes et al., 2010; Lischke et al., 2011), by combining genetics, functional and structural brain imaging, and sociobehavioral measures. Crucially, the model proposes that interactions between neuroendocrine and sociobehavioral factors need to be considered from a developmental perspective, taking age variations into account. Along these lines, the model offers a theoretical framework to address vital research questions: (1) Are OT-related genotypes associated with composition and quality of social networks in the elderly? How do brain structures involved in social processing such as mPFC and OFC, temporoparietal junction, or amygdala mediate these relationships? (2) Is older adults' increased social avoidance compared to approach motivation represented in neural processing differences in brain networks involving PFC and amygdala? To what extent do these associations interact with OT-related genotypes? (3) Are detrimental effects that early abuse has on morbidity and mortality in the elderly moderated by OT-related genotypes or OT levels? How is this relationship structurally and functionally represented in the brain? (4) Are effects of social relationships on cognitive functioning in the elderly mediated by the OT system (either OT-levels or OT-related genotypes)? Do structural changes in brain regions such as the hippocampus underlie this relationship?

In the attempt to provide a concrete empirical application of the *AGeNeS-OT Model*, we here present a preliminary report of an experiment in which we examined associations between *OXTR* polymorphisms, brain activity and behavioral response during reading of facial emotions in young and older adults. This exploratory, secondary data analysis was based on our group's previous finding of increased activation in ventromedial PFC (vmPFC) during emotion identification of happy compared to angry faces and increased dorsomedial PFC (dmPFC) activity to angry compared to happy faces (Ebner et al., 2012; see also Keightley et al., 2007) in both young and older adults. In the present set of analyses, we examined the extent to which these processing differences in mPFC would be further qualified when considering *OXTR* polymorphisms in both of the age groups. In particular, we examined (1) the extent to which *OXTR* polymorphisms were associated with differences in young and older adults' brain activity in bilateral mPFC (Haxby et al., 2000, 2002; Pessoa and Adolphs, 2010; Ebner et al., 2012) during a facial emotion reading task; and (2) the extent to which *OXTR* polymorphisms were associated with young and older adults' ability to read facial emotions.

Young [*n* = 25, 12 females, *M* = 25*.*1 years (*SD* = 3*.*6, range = 20–31)] and older [*n* = 29, 17 females, *M* = 68*.*3 years (*SD* = 2*.*8, range = 65–74)] healthy participants underwent fMRI on a 3T Siemens Magnetom TrioTim scanner, while identifying happy, neutral, and angry facial emotions (see Ebner et al., 2012, for details on participants, study design, and image acquisition). Participants were subsequently genotyped by KBioscience (http://www*.*kbioscience*.*co*.*uk) using KASPar methodology for 14 *OXTR* single nucleotide polymorphisms (SNPs in order from the 3 to the 5 end: rs7632287, rs6770632, rs1042778, rs237887, rs2268493, rs2254298, rs53576, rs237897, rs4686302, rs4564970, rs2301261, rs2268498, rs2270465, rs75775), previously shown to be associated with social behavior (Apicella et al., 2010; Meyer-Lindenberg et al., 2011; Ebstein et al., 2012; Walum et al., 2012; Westberg and Walum, 2013).

Data from this event-related fMRI study was analyzed using Statistical Parametric Mapping (SPM5; Wellcome Department of Imaging Neuroscience) and pre-processing and data analysis was conducted as reported in Ebner et al. (2012). The following *T*-contrasts were specified across young and older adults, based on our previous findings (Ebner et al., 2012): (1) *Happy Faces > Angry Faces*, (2) *Angry Faces > Happy Faces*. We focused on select regions of interest (ROIs: bilateral medial frontal gyrus and anterior cingulate gyrus) in which we had previously seen processing differences for happy vs. angry faces, at a threshold of *p <* 0*.*05, FDR corrected. For each region of activation identified by these two contrasts, peak voxel beta values were extracted for each participant to produce a single value for each condition of interest. These values are depicted in the bar graphs of **Figure 4**. In the fashion of follow-up *F*- and *t*-tests (*p <* 0*.*05), for each of the 14 *OXTR* SNPs that were genotyped, we examined differences in brain activation between polymorphisms across the total sample as well as separately for young and older adults. The most consistent associations found in these analyses were in relation to *OXTR* rs237887 (cf. Lerer et al., 2008; Israel et al., 2009; Liu et al., 2010; Lori et al., 2012; but see Apicella et al., 2010).

*OXTR* rs237887 AA carriers (*n* = 10 young participants; *n* = 10 older participants) and GA/GG carriers (*n* = 15 young participants; *n* = 19 older participants) were comparable in terms of chronological age, level of education, cognitive status (e.g., Mini Mental State Examination; Folstein et al., 1975; 2-Back Digits Task; Kirchner, 1958; Verbal Fluency Task; Lezak, 1995), and affective variables (Geriatric Depression Scale; Brink et al., 1982; Gottfries, 1997; State-Trait Anxiety Inventory; Spielberger et al., 1970).

For the contrast *Happy Faces > Angry Faces*, we found greater BOLD response to happy compared to angry faces in bilateral anterior cingulate cortex (ACC; MNI: *x* = 3, *y* = 45, *z* = 0 and *x* = −3, *y* = 51, *z* = 0) and bilateral mPFC (MNI: *x* = 3, *y* = 60, *z* = −3 and *x* = −3, *y* = 57, *z* = −3). **Figure 4A** shows brain activity in left ACC (MNI: *x* = −3, *y* = 51, *z* = 0) for this contrast. To then examine associations between *OXTR* rs237887 polymorphisms and brain activity during facial emotion identification of happy vs. angry faces in young and older adults, we conducted follow-up univariate ANOVA collapsed across young and older participants on extracted beta values at the peak voxel of activation. Left ACC activity was greater for AA carriers than GA/GG carriers [*F(*1*,* <sup>51</sup>*)* <sup>=</sup> <sup>6</sup>*.*51, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*014, <sup>η</sup><sup>2</sup> *p* = 0*.*11; see **Figure 4B**]. More interestingly, however, this effect was more pronounced in older than young adults, as tested in univariate ANOVAs conducted separately within young and older participants [Young participants: *<sup>F</sup>(*1*,* <sup>23</sup>*)* <sup>=</sup> <sup>2</sup>*.*38, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*136, <sup>η</sup><sup>2</sup> *p* = <sup>0</sup>*.*09; Older participants: *<sup>F</sup>(*1*,* <sup>26</sup>*)* <sup>=</sup> <sup>3</sup>*.*09, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*035, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*16; see **Figure 4C**]. A comparable pattern of results was found for right ACC [*F(*1*,* <sup>51</sup>*)* <sup>=</sup> <sup>6</sup>*.*34, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*015, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*11]. In addition, the results for left [*F(*1*,* <sup>51</sup>*)* <sup>=</sup> <sup>3</sup>*.*24, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*078, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*06] and

#### **Box 1 | Questions for future research.**

Error bars represent standard errors of the between-group differences.

1. Is aging accompanied by increases or decreases in central and peripheral release of OT? 2. Does the dynamic activity of the OT system change with age and, if so, how and why?

3. Do age-related differences in OT system dynamics underlie age-related differences in socioemotional functioning? If so, how do these changes frame our understanding of the age-associated changes in important social skills (i.e., reading facial emotions, face memory, approach, and avoidance behavior)?

4. How do OT-related individual genetic (and epigenetic) differences interact with neural and behavior age-related changes in socioemotional domains?

5. Does the OT system mediate some of the effects of adverse early experience on health and well-being? How does this play out in old age?

6. Does the OT system mediate some of the salutary psychological and health effects of ongoing social relationships (both intimate and larger social networks)? To what extent do age-related changes in social relationships influence these effects?

7. How do sex differences in OT system dynamics play out in the context of aging? For example, what is the role of age-related changes in estrogen and testosterone?

8. Does the OT system have a role in age-related changes in cognition and memory?

9. Might OT be an effective treatment for conditions like social anxiety or depression in the elderly? Would such treatment improve quality of life?

10. Might older adults be at increased risk of OT-related side effects (i.e., hyponatremia) with chronic dosing?

right [*F(*1*,* <sup>51</sup>*)* <sup>=</sup> <sup>1</sup>*.*29, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*261, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*03] mPFC pointed in the same direction but were not significant.

ACC is a brain region associated with affective processing (Bush et al., 2000; Amodio and Frith, 2006; Ebner et al., 2012), suggesting that AA compared to GA/GG carriers may process happy compared to angry faces more affectively. This interpretation was further supported by the finding that AA-genotype carriers of *OXTR* rs237887 (*M* = 1111 ms, *SD* = 171 ms) were faster at labeling happy expressions than individuals carrying a G-allele [*M* = 1212 ms, *SD* = 173 ms; *F(*1*,* <sup>50</sup>*)* = 4*.*26, *p* = 0*.*044, η2 *<sup>p</sup>* = 0*.*08], with comparable effects in young and older participants. No comparable effect was found for accuracy in emotion expression identification. However, interestingly, greater recruitment of right ACC in individuals carrying a G-allele was positively correlated (*r* = 0*.*35; *p* = 0*.*049) with accuracy in reading happy faces but uncorrelated in AA-genotype carriers (*r* = 0*.*05; *p* = 0*.*838). This suggests that GA/GG carriers, as the group who needed more time on the task, benefitted from recruiting ACC during the facial emotion reading task. This positive brainbehavior correlation in GA/GG carriers was comparable in young and older participants (Fisher's *z* = −0*.*42; *p* = 0*.*337).

For the contrast *Angry Faces > Happy Faces,* we found greater BOLD response to angry compared to happy faces in left mPFC (MNI: *x* = −6, *y* = 15, *z* = 51). In a follow-up univariate ANOVA collapsed across young and older participants on extracted beta values at the peak voxel of activation, activity in left mPFC did not vary by *OXTR* rs237887 polymorphism (*p >* 0*.*05).

To our knowledge this is the first study that considers young and older participants in a genetic-neuro-behavioral examination of facial emotion processing, as suggested in the *AGeNeS-OT model*. Though this secondary data analysis was largely exploratory and replication in a larger independent sample of young and older adults is warranted, our study provides some first indication of a role of *OXTR* rs237887 in reading positive compared to negative facial expressions, with some variation as a function of the age of the participant. Intriguingly, *OXTR* rs237887 has previously been associated with susceptibility for ASD (Liu et al., 2010), prosocial behavior (Israel et al., 2009, but see Apicella et al., 2010), and face recognition (Lori et al., 2012). We found improved processing of happy compared to angry faces for AA carriers compared to GA/GG carriers, as reflected in their faster response time in reading happy faces and their increased recruitment of ACC during emotion reading of happy compared to angry faces. Examining young and older participants separately, this increased activation of ACC in AA compared to GA/GG carriers was more pronounced in older than young participants. This is very interesting given broad evidence of preferential processing of positive over negative information in older compared to young adults (Mather and Carstensen, 2005). In addition, our findings suggest that GA/GG carriers' ability to correctly identify happy faces improved when recruiting ACC during the task.

## **FUTURE TRENDS IN RESEARCH ON OXYTOCIN AND SOCIOEMOTIONAL AGING**

Taken together, this research review indicates that a targeted investigation of age-related changes in the OT system—especially one that considers genetic, neural, and behavioral processes has the potential to substantively increase our understanding of socioemotional change in aging. We believe that our *AGeNeS-OT model* will be a fruitful conceptual basis in that it raises a set of vital research questions necessary to refine our understanding of OT-related dynamics in aging in socioemotional contexts (see **Box 1**). In addition, future research along those

# **REFERENCES**


*PLoS ONE* 5:e11153. doi: 10.1371/journal.pone.0011153


lines has great potential to inform both pharmacological and psychosocial interventions targeting social and emotional dysfunction in the elderly. In particular, there is an increasing body of research suggesting a significant role of OT in the context of various disorders characterized by socioemotional dysfunction such as social-bonding deficits or related to social anxiety and stress (Zetzsche et al., 1996; Heinrichs et al., 2003; Taylor et al., 2006; see MacDonald and Feifel, 2012, for an overview), deficits with great relevance in an aging context. Thus, future research toward implementation of pharmacological neuropeptide treatments with the potential to decrease emotional and social stress, anxiety, and depression (Arletti and Bertolini, 1987; Carter and Altemus, 1997) will be important. These interventions may consequently promote positive social interaction and willingness to engage in more frequently rewarding social risks (Heinrichs et al., 2003; Kosfeld et al., 2005), improving health and life quality up until late in life.

# **ACKNOWLEDGMENTS**

This research was supported in part by the NIH/NCATS Clinical and Translational Science Award to the University of Florida UL1 TR000064 (pilot award to Natalie C. Ebner) and the Swedish Research Council (2008-2356) and the Konung Gustaf V:s och Drottning Victorias Frimurarstiftelse (Håkan Fischer). Some of Kai MacDonald's work was supported by the Goodenough Neuroscience Research Fund. The authors wish to thank Drs. David Feifel and Ronald Cohen for constructive discussions regarding various aspects of this manuscript.


Kolevzon, A., et al. (2010). Oxytocin selectively improves empathic accuracy*. Psychol. Sci.* 21, 1426–1428. doi: 10.1177/0956797610383439


L. (2002). Sniffing neuropeptides: a transnasal approach to the human brain. *Nat. Neurosci.* 5, 514–516. doi: 10.1038/nn0602-849


women and men. Development and validation. *Behav. Res. Methods.* 42, 351–362. doi: 10.3758/BRM.42.1.351


synchrony: considering stress and affiliation components of human bonding*. Dev. Sci.* 14, 752–761. doi: 10.1111/j.1467-7687.2010.01021.x


social recognition. *Behav. Neurosci.* 126, 97–109. doi: 10.1037/a00 26464


627–628. doi: 10.1016/j.psyneuen. 2013.03.005


*Cogn. Affect. Neurosci.* 2, 292–302. doi: 10.1093/scan/nsm024


cortisol responses to social stress in humans*. Biol. Psychol.* 93, 304–307. doi: 10.1016/j.biopsycho. 2013.02.018


disorder (ASD) in the Japanese population*. J. Hum. Genet.* 55, 137–141. doi: 10.1038/jhg.2009.140


*Trends Cogn. Sci.* 9, 496–502. doi: 10.1016/j.tics.2005.08.005


postpartum onset of rat maternal behavior in the ventral tegmental and medial preoptic areas. *Behav. Neurosci.* 108, 1163. doi: 10.1037/0735–7044.108.6.1163


amygdala, insula, and inferior frontal gyrus responses to infant crying: a randomized controlled trial. *Biol. Psychiatry* 70, 291–297. doi: 10.1016/j.biopsych. 2011.02.006


368–374. doi: 10.1016/j.psyneuen. 2007.12.004


*Manual for the State-Trait Anxiety Inventory*. Palo Alto, CA: Consulting Psychologist Press.


face recognition, trust to ingroup, and trust to out-group*. Psychoneuroendocrinology* 37, 438–443. doi: 10.1016/j.psyneuen. 2011.07.008


social buffering in rhesus monkeys*. Neuropsychopharmacology* 28, 910–918. doi: 10.1038/sj.npp. 1300128


that oxytocin exerts anxiolytic effects via oxytocin receptor expressed in serotonergic neurons in mice*. J. Neurosci.* 29, 2259–2271. doi: 10.1523/JNEUROSCI.5593- 08.2009


Fischer 344 and Sprague-Dawley rats*. Neuroendocrinology* 48, 619–626. doi: 10.1159/000125072


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 28 May 2013; accepted: 01 August 2013; published online: 28 August 2013.*

*Citation: Ebner NC, Maura GM, MacDonald K, Westberg L and Fischer H (2013) Oxytocin and socioemotional aging: Current knowledge and future trends. Front. Hum. Neurosci. 7:487. doi: 10.3389/fnhum.2013.00487*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Ebner, Maura, MacDonald, Westberg and Fischer. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Navigating the complex path between the oxytocin receptor gene (*OXTR*) and cooperation: an endophenotype approach

# *Brian W. Haas1,2\*, Ian W. Anderson1 and Jessica M. Smith2*

*<sup>1</sup> Department of Psychology, University of Georgia, Athens, GA, USA*

*<sup>2</sup> Interdisciplinary Neuroscience Graduate Program, University of Georgia, Athens, GA, USA*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Maria Barth, Tufts University, USA Hidenori Yamasue, University of Tokyo, Japan*

#### *\*Correspondence:*

*Brian W. Haas, Department of Psychology, University of Georgia, 125 Baldwin Street, Athens, 30602 GA, USA e-mail: bhaas@uga.edu*

Although cooperation represents a core facet of human social behavior there exists considerable variability across people in terms of the tendency to cooperate. One factor that may contribute to individual differences in cooperation is a key gene within the oxytocin (OT) system, the OT reception gene (*OXTR*). In this article, we aim to bridge the gap between the *OXTR* gene and cooperation by using an endophenotype approach. We present evidence that the association between the *OXTR* gene and cooperation may in part be due to how the *OXTR* gene affects brain systems involved in emotion recognition, empathy/theory of mind, social communication and social reward seeking. There is evidence that the *OXTR* gene is associated with the functional anatomy of the amygdala, visual cortex (VC), anterior cingulate and superior temporal gyrus (STG). However, it is currently unknown how the *OXTR* gene may be linked to the functional anatomy of other relevant brain regions that include the fusiform gyrus (FG), superior temporal sulcus (STS), ventromedial prefrontal cortex (VMPFC), temporoparietal junction (TPJ) and nucleus accumbens (NAcc). We conclude by highlighting potential future research directions that may elucidate the path between *OXTR* and complex behaviors such as cooperation.

**Keywords:** *OXTR***, genetics, oxytocin, social-cognition, cooperation**

#### **INTRODUCTION**

Human beings have a unique affinity towards cooperating with one another to accomplish goals. The act of cooperation involves a common effort within a group for the collective benefit rather than seeking to accomplish goals solely for oneself. Cooperative people are characterized as socially tolerant, empathic, helpful, and compassionate (Cloninger et al., 1993). Although the tendency to cooperate is a common attribute across many people and cultures, there also exists considerable variability between people in cooperative motivation and behavior (Wischniewski et al., 2009). In this article, we highlight recent progress towards identifying genetic factors that contribute to individual differences in cooperation. Within this framework, we focus our attention on the oxytocin (OT) system and brain regions that subserve the tendency to cooperate.

Cooperation has been empirically investigated in several different ways. One common way that cooperation has been studied within social psychology and economic research is to use the Prisoner's Dilemma task. The prisoner's dilemma task provides the opportunity to characterize the value one attributes to their own outcome versus the value one attributes to the outcome of the entire group. Results of this research demonstrate that humans often act more cooperatively than a strict self-interest strategy would predict (Komorita and Parks, 1999).

Although many humans tend to be driven towards cooperation, there also exists considerable variability among people in how cooperative they tend to be (Wischniewski et al., 2009). For example, individual differences in personality traits, such as with agreeableness and conscientiousness, are associated with the tendency to cooperate (Witt et al., 2002; Volk et al., 2011). Another factor that may influence cooperation across humans is genetics (Cesarini et al., 2008). Many of the genes implicated in cooperation are known to affect the function of hormonal and neurotransmitter systems within a network of brain regions important for social-cognition.

One hormone associated with individual differences in cooperation and with the function of brain regions involved in socialcognition is OT (Ebstein et al., 2012; Yamasue, 2013). OT is a neuropeptide primarily synthesized in the hypothalamus and has broad effects on OT receptors throughout the central nervous system (Gimpl and Fahrenholz, 2001). One way to investigate the effect of OT on social behavior is to manipulate OT levels via intranasal administration. Following OT administration people display greater amounts of cooperative behavior as compared to individuals receiving placebo (Kosfeld et al., 2005; Declerck et al., 2013). These findings have motivated the search for genes within the OT system that are linked to individual differences in cooperation.

One gene within the OT system that has been linked to prosocial behaviors, such as cooperation, is the OT receptor gene (*OXTR*). *OXTR* is a gene located on chromosome 3p25 that codes for OT receptors (Kimura et al., 1992). There are several single nucleotide polymorphisms (SNPs) of *OXTR*, each of which codes for particular attributes of the OT receptor. Studying how behavioral or biological metrics vary according to *OXTR* SNPs provides insight as to the function of the *OXTR* gene. In this article, we focus on findings comparing social cognitive metrics, as related to cooperation, based on *OXTR* SNPs.

Behavioral genetic studies associating polymorphisms of the *OXTR* gene with laboratory measures of cooperation have revealed a mixed group of results. Israel et al. (2009) and Tabak et al. (2013) demonstrated an association between *OXTR* and lab measures of cooperation including the Dictator Game and the Social Values Orientation and Prisoner's Dilemma tasks. On the other hand, Apicella et al. (2010) reported no association between *OXTR* and cooperation during the Dictator and Trust Game. The inconsistency across these studies may indicate that *OXTR* is associated with some, but not all, of the underlying components of cooperative behavior in humans.

One effective method used to investigate the association between genes and complex behaviors, such as cooperation, is to use an endophenotype approach. An endophenotype represents an intermediate level between gene expression and a complex behavior or disease state (Gottesman and Gould, 2003). For example, there may be a weak or moderate association between a gene and the onset of an anxiety or mood disorder based on how the gene codes for the organization of neurons that react to psychosocial stress. In this example, the neuronal response to psychosocial stress is considered an endophenotype that exists between the gene and disease (Hamer, 2002; Gottesman and Gould, 2003). Considering endophenotypes holds promise in terms of elucidating the genetic etiology of cognition, social behavior and psychopathology.

# **A COMPONENT VIEW OF COOPERATION AND THE COOPERATIVE BRAIN**

The ability to cooperate effectively with others relies on a set of underlying social-cognitive constructs (Brosnan et al., 2010). Successful cooperation requires social cognitive constructs that include (but not limited to) to (i) acknowledging and recognizing the emotional states of others within a group (emotion recognition) (Elfenbein et al., 2007; Krumhuber et al., 2007); (ii) accurately interpreting the intentions of others (empathy/theory of mind) (Sally and Hill, 2006; Paal and Bereczkei, 2007); (iii) communicating effectively with others (social communication) (Miller et al., 2002); and (iv) seeking out and valuing social interaction (social reward seeking). Thus, effective cooperation may be characterized by the availability of specific socialcognitive resources. Considering each of these social cognitive constructs independently may elucidate how genes within the OT system influence cooperative motivation and behavior in humans.

Neuroimaging research demonstrates that specific brain networks carry out many of the social-cognitive constructs underlying cooperation. The ability to accurately acknowledge and recognize the emotional states of others (emotion recognition) is carried out through the ventral processing stream and relies on visual cortices, the fusiform gyrus (FG), superior temporal sulcus (STS) and areas within the prefrontal cortex (**Figure 1A**). The visual centers of the brain are within an emotional attention circuit (Rudrauf et al., 2008). When faced with information of high emotional saliency, the amygdala and visual cortex (VC) function to increase local attention recourses (Morris et al., 1998). The FG contains the fusiform face area, which is a highly specialized region for distinguishing between different types of faces (Kanwisher et al., 1997). The amygdala functions to signal and tag information that is highly emotionally salient (Aggleton, 2000). The STS is involved in processing social and emotional signals conveyed via body or biological motions (Thompson and Parasuraman, 2012). Lastly, the ventromedial and dorsolateral prefrontal cortex (DLPFC) categorizes and evaluates the salience of emotional stimuli (Mitchell and Greening, 2012; Roy et al., 2012). Combined, this brain network subserves the ability to evaluate and categorize the emotional states of others.

Being able to accurately interpret the intentions of others (empathy and theory of mind) relies on a brain network that includes areas within the frontal, temporal and parietal cortices (**Figure 1B**). The dorsomedial prefrontal cortex (DMPFC) and dorsal anterior cingulate cortex (ACC) generate appropriate emotional responses to other people's mental states (i.e., emotional empathy) (Fan et al., 2010). While the temporoparietal junction (TPJ) is involved in accurately interpreting the mental states and intentions of others (i.e., cognitive empathy or theory of mind) (Decety and Lamm, 2007). Together, these brain structures subserve the ability to accurately understand and respond to the emotions and intentions of other people.

Social communication (the ability to perceive, transmit and understand information between people) relies on brain regions that include several areas within the temporal and frontal lobes (**Figure 1C**). The superior temporal gyrus (STG) processes verbal and non-verbal social cues (Hickok and Poeppel, 2007; Hein and Knight, 2008). The superior temporal gyrus contains the primary auditory cortex, which functions to decode vocal communication signals (Hickok and Poeppel, 2007). Lastly the inferior frontal gyrus (IFG), and the premotor cortex (PMC) are both important regions involved in speech production (Price, 2012). Together, the integrity of these brain structures is critical for effective social communication to occur.

Being driven towards social interaction and the subjective sense of reward in response to social interaction may be associated with the tendency to cooperate. The subjective motivation towards social interaction is subserved by a brain network involved in salience and reward processing and includes the dopaminergic system and several regions within the ventral striatum and frontal lobe (Haber and Knutson, 2009; **Figure 1D**). Within the striatum, the nucleus accumbens (NAcc) is a critical area involved in reward processing and is conceptualized as the brain's "pleasure center." Within the frontal lobe, the ventromedial prefrontal cortex (VMPFC) and ACC function to signal and anticipate potential rewards (Rushworth et al., 2011). Thus, the brain's reward circuitry may contribute to how social interactions are experienced and valued during conditions that involve cooperation.

**FIGURE 1 | Schematic representation of brain networks subserving cooperation for each social-cognitive component: (A) Emotion Recognition, (B) Empathy and Theory of Mind, (C) Social Communication, (D) Social Reward.** Rounded boxes (with bold or grey outlines) signify brain regions implicated in social-cognitive constructs subserving cooperation. Rounded boxes with bold outlines signify brain

# *OXTR* **AND BRAIN MECHANISMS UNDERLYING COOPERATION**

In this section, we will consider the association between the *OXTR* gene and specific social-cognitive constructs and brain networks that subserve cooperation. There is some evidence that the *OXTR* gene is associated with emotion recognition. Lucht et al. (2012) and Rodrigues et al. (2009) reported that performance on the "Reading the Mind in the Eyes Test" varied according to *OXTR* polymorphisms. Additional, indirect support for the association between OT genes and emotion recognition comes from evidence that OT administration improves people's ability to recognize emotions (Domes et al., 2007; Bartz et al., 2010; Guastella et al., 2010).

regions shown to be structurally or functionally different according to the *OXTR* gene (SNPs or methylation) in humans. DLPFC, Dorsolateral Prefrontal Cortex; VC, Visual Cortex; STS, Superior Temporal Sulcus; FG, Fusiform Gyrus; VMPFC, Ventromedial Prefrontal Cortex; Amy, Amygdala; TPJ, Temporoparietal Junction; ACC, Anterior Cingulate Cortex; PMC, Premotor Cortex; IFG, Inferior Frontal Gyrus; STG, Superior Temporal Gyrus; NAcc, Nucleus Accumbens.

In terms of the brain, magnetic resonance imaging studies (MRI) studies show that the *OXTR* gene is associated with the structure and function of a subset of brain regions involved in emotion recognition (**Figure 1A**). O'Connell et al. (2012) demonstrated that *OXTR* polymorphisms are associated with VC (cuneus and inferior occipital gyrus) reactivity to fearful faces. In terms of the amygdala, both Furman et al. (2011) and Inoue et al. (2010) showed an association between the *OXTR* gene and amygdala volume and Tost et al. (2010) showed that the *OXTR* gene is associated with amygdala activity during an emotional face-matching task. For the FG, O'Connell et al. (2012) directly tested for activation differences according to *OXTR* variants within this region, but failed to identify any significant differences. A recent study investigated the association between methylation<sup>1</sup> of the *OXTR* gene and brain function during biological motion processing (Jack et al., 2012). The results indicated that increased methylation (typically associated with decreased expression) of *OXTR* is associated with greater activation within the superior temporal gyrus (STG). Although biological motion is more often linked with the function of the STS, this study provides preliminary evidence that the *OXTR* gene is associated with neural reactivity during the recognition of social information. Lastly, it is currently unknown if the *OXTR* gene is associated with the structure or function of the dorsolateral or VMPFC in humans.

There is evidence that the *OXTR* gene may be linked to individual differences in empathy and theory of mind. Behavioral studies show that *OXTR* is associated with self reported empathy (Rodrigues et al., 2009). Furthermore, Wu et al. (2012) demonstrated that distinct polymorphisms of the *OXTR* gene are associated with emotional and cognitive (i.e., theory of mind) empathy. In terms of the brain, there is evidence that the *OXTR* gene is associated with the structure and function of brain regions involved in emotional empathy (**Figure 1B**). Specifically, Furman et al. (2011) and Tost et al. (2010, 2011) demonstrated volumetric differences of the dorsal ACC according to *OXTR* polymorphisms. In terms of function, Tost et al.(2011) showed that the *OXTR* gene is associated with dorsal ACC activity during emotional face processing. However, it is currently unknown if *OXTR* is associated with the structure or function of key brain regions involved in cognitive empathy/theory or mind, such as the temporal parietal junction.

The *OXTR* gene may be linked to the ability to socially communicate. Behavioral research shows that *OXTR* is associated with the ability to comprehend information during vocal communication (Tops et al., 2011) and with the severity of communication deficits in autism (Jacob et al., 2007; Lerer et al., 2007; Campbell et al., 2011). There is limited evidence that *OXTR* is linked to brain mechanisms underlying social communication (**Figure 1C**). However, indirect support comes from research on the effect of OT administration on brain reactivity to vocal social signals (Riem et al., 2011, 2012). Riem et al. (2011) showed that parents that receive OT administration exhibit greater left inferior frontal gyrus reactivity to sounds of their child crying as compared to parents receiving placebo. In addition, *OXTR* methylation status is associated with left STG activation (Jack et al., 2012). However, the association between *OXTR* methylation and STG activity was found in response to a biological motion processing task. Therefore, it is currently unknown if the *OXTR* gene is associated with the functional anatomy of the STG when socially communicating. In addition, it is currently unknown how *OXTR* polymorphisms may affect the structure or function of other regions involved in social communication that include the inferior frontal gyrus and premotor area.

The *OXTR* gene may be associated with individual differences in social reward processing. Behavioral studies indicate that the *OXTR* gene is associated with trait reward sensitivity (Tost et al., 2010) and social motivation deficits in autism (Campbell et al., 2011). In terms of the brain, there is some evidence that the *OXTR* gene is associated with the structure and function of a subset of regions involved in reward processing (**Figure 1D**). Polymorphisms of *OXTR* are associated with ACC reactivity during emotion processing (Tost et al., 2011). In addition *OXTR* is associated with volumetric differences of the ACC (Tost et al., 2010; Furman et al., 2011). Indirect support of the association between *OXTR* and the neural basis of reward processing comes from research on the link between *OXTR* and dopamine transmission. Love et al. (2012), used positron emission tomography and showed that *OXTR* is associated with dopamine levels within the striatum in females. Lastly, research on animals demonstrates that the nucleus accumbens is densely populated with OT receptors (Ross et al., 2009). It is currently unknown however, how the *OXTR* gene affects the functional anatomy of the Nacc and VMPFC in humans.

A review of the research to date demonstrates that the *OXTR* gene may influence the functional anatomy of a subset of the brain regions implicated in cooperation. For emotion recognition, the *OXTR* gene is associated with the structure (Inoue et al., 2010; Furman et al., 2011) and function (Tost et al., 2010) of the amygdala and the function of the VC (O'Connell et al., 2012). This indicates that the *OXTR* gene may be associated with attention to emotionally salient stimuli, but not necessarily with face processing (fusiform gryus) or higher order categorization and evaluation of emotional stimuli (VMPFC and DLPFC). For empathy and theory of mind, the *OXTR* gene is associated with the structure (Tost et al., 2010, 2011; Furman et al., 2011) and function (Tost et al., 2011) of the ACC, though there is currently no evidence that *OXTR* is associated with the structure or function of the TPJ. These findings suggest that the *OXTR* gene may influence brain regions involved in emotional but not cognitive empathy. In terms of social communication, there is one study showing an association between *OXTR* methylation and the function of the STG (Jack et al., 2012). Therefore, it is currently not known how *OXTR* polymorphisms are linked to the functional anatomy within brain important for social communication. Lastly, for social reward processing, there is evidence that *OXTR* is associated with the structure (Tost et al., 2010, 2011; Furman et al., 2011) and function (Tost et al., 2011) of the ACC. In spite of many animal studies demonstrating that the NAcc is densely populated with OT receptors (Hammock and Young, 2006), it is currently unknown how the *OXTR* gene may affect the functional anatomy of the NAcc in humans.

# **POTENTIAL FUTURE RESEARCH DIRECTIONS AND CONCLUSION**

There is currently a lack of evidence that the *OXTR* gene is associated with brain function during tasks that explicitly involve cooperation. One potential strategy to elucidate the association between *OXTR* and the brain basis of cooperation is to utilize a version of the prisoner's dilemma task that can be used within a brain imaging environment (Rilling et al., 2012) and compare patterns of brain reactivity according to *OXTR* polymorphisms.

In addition, a potential strategy to explore how the *OXTR* gene may be associated with face processing is to use tasks and analysis procedures specifically designed to quantify the spatial extent of the fusiform face area (Weiner and Grill-Spector, 2012).

<sup>1</sup>Methylation is an epigenetic process that affects the expression of genes.

Support for the hypothesis that the *OXTR* gene is associated with emotional but not cognitive empathy may be obtained by using fMRI tasks designed to explicitly compare types of empathic processing (Sebastian et al., 2012). For social communication, MRI studies show that emotional prosody relies on a specific network of brain regions (Ethofer et al., 2006; Wiethoff et al., 2008). Emotional prosody tasks may be a promising tool to explore the association between *OXTR* and brain networks subserving social communication. Lastly, an effective way to investigate the association between *OXTR* and social reward processing may be to utilize tasks explicitly designed to assess social versus nonsocial (monetary) reward processing (Gossen et al., 2013). Based on findings that social relevance boosts the influence of OT on cooperative behavior (Declerck et al., 2010, 2013), the *OXTR* gene may have a greater impact on brain function during social reward processing as compared to monetary reward processing.

In this review, we have focused on a network of key brain regions involved in cooperation. In conclusion, there is limited evidence that the *OXTR* gene is directly linked to the functional anatomy of the brain network implicated in cooperation. The use of endophenotypes is a promising strategy that may help to elucidate this complex gene, brain and social-cognitive association.

#### **REFERENCES**


in negotiation. *J. Nonverbal. Behav.* 31, 205–223. doi: 10.1007/s10919-007- 0033-7


ship to vineland adaptive behavior scales and cognition. *Mol. psychiatry* 13, 980– 988. doi: 10.1038/sj.mp.4002087

Love, T. M., Enoch, M.-A., Hodgkinson, C. A., Peciña, M., Mickey, B., Koeppe, R. A., et al. (2012). Oxytocin gene polymorphisms influence human dopaminergic function in a sex-dependent manner. *Biol. Psychiatry* 72, 198–206. doi: 10. 1016/j.biopsych.2012.01.033

Lucht, M. J., Barnow, S., Sonnenfeld, C., Ulrich, I., Grabe, H. J., Schroeder, W., et al. (2012). Associations between the oxytocin receptor gene (OXTR) and "mindreading" in humans-an exploratory study. *Nord. J. Psychiatry* 67, 15–21. doi: 10. 3109/08039488.2012.700731


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 04 November 2013; published online: 28 November 2013.*

*Citation: Haas BW, Anderson IW and Smith JM (2013) Navigating the complex path between the oxytocin receptor gene (OXTR) and cooperation: an endophenotype approach. Front. Hum. Neurosci. 7:801. doi: 10.3389/fnhum.2013.00801*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Haas, Anderson and Smith. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The social regulation of threat-related attentional disengagement in highly anxious individuals

# *Erin L. Maresh , Lane Beckes and James A. Coan\**

*Department of Psychology, University of Virginia, Charlottesville, VA, USA*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Stuart W. G. Derbyshire, National University of Singapore, Singapore Marta Andreatta, University of Wuerzburg, Germany*

#### *\*Correspondence:*

*James A. Coan, Department of Psychology, University of Virginia, 102 Gilmer Hall, PO Box 400400, Charlottesville, VA 22904, USA e-mail: jcoan@virginia.edu*

Social support may normalize stress reactivity among highly anxious individuals, yet little research has examined anxious reactions in social contexts. We examined the role of both state and trait anxiety in the link between social support and the neural response to threat. We employed an fMRI paradigm in which participants faced the threat of electric shock under three conditions: alone, holding a stranger's hand, and holding a friend's hand. We found significant interactions between trait anxiety and threat condition in regions including the hypothalamus, putamen, precentral gyrus, and precuneus. Analyses revealed that highly trait anxious individuals were less active in each of these brain regions while alone in the scanner—a pattern that suggests the attentional disengagement associated with the perception of high intensity threats. These findings support past research suggesting that individuals high in anxiety tend to have elevated neural responses to mild or moderate threats but paradoxically lower responses to high intensity threats, suggesting a curvilinear relationship between anxiety and threat responding. We hypothesized that for highly anxious individuals, shock cues would be perceived as highly threatening while alone in the scanner, possibly due to attentional disengagement, but this perception would be mitigated if they were holding someone's hand. The disengagement seen in highly anxious people under conditions of high perceived threat may thus be alleviated by social proximity. These results suggest a role for social support in regulating emotional responses in anxious individuals, which may aid in treatment outcomes.

**Keywords: social regulation of emotion, trait anxiety, attentional disengagement, fMRI**

# **INTRODUCTION**

A large body of research suggests that social proximity and interaction confer benefits ranging from buffering stress to extending life (House et al., 1988; DeVries et al., 2003). These benefits may be linked to the way supportive social contact can attenuate threat responding in the brain (Coan et al., 2006b, 2013). Recent work suggests that social support may be especially important for people high in trait anxiety, as anxiety is characterized by increased reactivity to stressors (Bolger and Zuckerman, 1995; Conner et al., 2012). Still, many questions remain about how anxious people respond to perceived threats in a supportive social context. Our goal was to examine how the presence or absence of perceived social resources alters threat-related processing in the brains of highly anxious adults.

In general, high trait anxiety corresponds with increased responsiveness to stressors. This is observed in self reported anxiety (Bolger and Schilling, 1991), autonomic reactivity (Gonzalez-Bono et al., 2002), and hormonal output (Schlotz et al., 2006). Neuroimaging has also revealed increased stress-related activity in the central nervous system (Etkin et al., 2004). For example, when anticipating a shock, individuals high in trait anxiety show exaggerated activity in the dorsal anterior cingulate cortex (dACC), somatosensory cortex, motor cortex, and hippocampus, areas

related to vigilance, motor preparedness, and approach/avoidance conflict (Straube et al., 2009).

On the other hand, anxiety-related traits have also been associated with *decreased* responsivity to stress. For example, Jezova et al. (2004), found that participants with high trait anxiety had lower secretions of epinephrine, norepinephrine, and prolactin during a stressful public speaking task. Similarly, lower cortisol levels upon awakening were found in participants higher in trait anxiety (Walker et al., 2011). It has been suggested that the excessive, chronic activation of the stress response that anxious people experience may eventually lead to reduced responsiveness of the hypothalamic-pituitary-adrenal (HPA) axis (McEwen, 2007).

Recent work using functional magnetic resonance imaging (fMRI) has led some to postulate that anxiety has a curvilinear relationship with threat responding (Straube et al., 2009; Drabant et al., 2011). Straube et al. observed that while strong threats yielded a positive correlation between anxiety and activity in certain brain regions, this correlation was conspicuously negative in the ventral anterior cingulate cortex (vACC), a region within the cingulate associated with the modulation of physiological arousal (Allman et al., 2001). Similarly, modulating the intensity of anticipatory anxiety during shock threat led to monotonically linear increases in activity from safety to strong threat except in participants high in neuroticism (Drabant et al., 2011), a personality trait that is strongly related to anxiety (Luteijn and Bouman, 1988). Highly neurotic participants showed a relative decrease in neural activity in the insula and the dorsolateral prefrontal cortex (DLPFC) when shifting from moderate to strong threats. The authors theorize that this decreased activity signals a switch to an avoidant processing style in highly anxious individuals when the threat becomes severe. Furthermore, these findings suggest that anxiety alters an individual's *perception* of threat, which may function in the same way as altering the threat itself.

One method of modulating the perceived intensity of a threat is through social support. Indeed, recent work reveals that handholding mitigates the neural threat response, particularly when the hand-holder is a familiar relationship partner (Coan et al., 2006b; Conner et al., 2012). Less is known about how anxiety influences the extent and direction of this relationship. We do know that strong social ties buffer against the development of anxiety disorders (Plaisier et al., 2007), but little is known about how anxiety manifests in the context of social support.

The current study was designed to explore how the provision of social support may interact with anxiety in the neural response to threat. We measured brain activity using fMRI in a sample of participants who underwent a threat-of-shock paradigm under three conditions: while alone, while holding a stranger's hand, and while holding a friend's hand (cf. Coan et al., 2006b). Additionally, we examined how both trait and state anxiety moderated the neural threat response under these conditions. Due to differing reports in the literature, we proposed two competing hypotheses: (1) According to what we term the *potentiation* model, the relationship between anxiety and threat response is simply linear. That is, people with higher anxiety will show potentiated activity in brain areas related to stress and anticipatory anxiety (Coan et al., 2006a) in response to a threatening stimulus. The potentiation model predicts that under conditions of handholding, individuals with higher anxiety will show a reduction in threat-related brain activity such that they more closely resemble less anxious participants, particularly when holding a friend's hand (vs. a stranger's; cf. Conner et al., 2012). (2) By contrast, the *vigilance/disengagement* model suggests the relationship between trait anxiety and threat response is curvilinear, with moderate threats leading to increased vigilance in threat-related brain regions and strong threats leading to a strategy of disengagement from threat stimuli—and a concomitantly diminished neural threat response. Moreover, the vigilance/disengagement model predicts that people with higher trait anxiety will show *decreased* brain activity in areas related to anticipatory anxiety when anticipating the stimulus alone, because the intensity of the threat cue will be perceived as greater in the absence of handholding. Note that the potentiation and vigilance/disengagement models result in precisely opposite predictions. The potentiation model predicts higher trait anxiety will correspond with higher threat reactivity while alone and lower threat reactivity with social support. The vigilance/disengagement model predicts higher trait anxiety will correspond with lower threat reactivity while alone and higher threat reactivity with social support. Importantly, both models assume that social support decreases the perceived intensity of a threat cue (cf. Cohen and Wills, 1985; Coan, 2008).

# **MATERIALS AND METHODS**

## **PARTICIPANTS**

Twenty-seven participants and their opposite-gendered friends were recruited via flyers or drawn from a larger longitudinal study on adolescent social development (McElhaney et al., 2006; Chango et al., 2012). These participants are also part of a larger group in which we are studying the effects of handholding across different types of relationships (marriage, cohabitating, dating, and friends). Because this is a heterosexual sample in which the participants brought in opposite-gendered romantic partners, to maintain consistency, we requested opposite-gendered friends as well. We further requested each participant bring in a friend for whom they have not had romantic feelings. Respondents were excluded if they had current or past history of psychopathology, were pregnant, or exhibited risk for incident in the fMRI environment. Of the twenty-seven participants and their friends, two dyads were removed from final analyses for being outliers according to Mahalanobis distances. The final sample of 25 participants consisted of 13 males and 12 females, ages 23–26. Ten participants identified themselves as African-American and fifteen as White on a demographics questionnaire. Each member of the pair gave informed consent and was paid \$160 for his or her participation.

## **PROCEDURE**

Participants were screened via telephone and scheduled for a visit to the laboratory. During the screening, participants were informed they would receive a mild electric shock that is designed to be uncomfortable but not painful. On the scheduled day, the participant came in with his or her friend and both completed a battery of questionnaires assessing personality, attachment style, relationship measures, etc. For this study, we looked at results from the State-Trait Anxiety Inventory (STAI; Spielberger, 1983) for each participant. The STAI measures both state and trait anxiety, each using a 20-item questionnaire with a 4-point Likert scale, yielding scores ranging from 20 to 80.

## **SHOCK PARADIGM**

Two Ag-AgCl shock electrodes were placed on the participant's right or left ankle (counterbalanced across participants). The participant entered the fMRI scanner and anatomical scans were collected. Following this, the participant underwent the handholding paradigm. Participants viewed stimuli projected onto a screen at the back of the magnet's bore via a mirror placed on the head coil, and a button box was provided for the participant to respond to stimuli. Scanning was done under three conditions (Alone, Stranger, and Friend), the order of which was counterbalanced across participants. In the Alone condition, the participant underwent the experiment alone in the scanner. In the Stranger condition, the participant underwent the experiment while holding the handing of an anonymous experimenter of the opposite gender whom the participant did not meet until the end of the experiment. In the Friend condition, the participant held the hand of the opposite-gendered friend they had brought with them. Before each condition, the participant was informed whether he or she would be holding a stranger's hand, a friend's hand, or would be alone. The handholding partner sat on a stool next to the participant, with both participant and hand holder hands resting on the bed of the scanner, allowing each person to comfortably hold hands for the duration of the task.

During each condition, the participant observed twelve threat (a red "X" on a black background) and twelve safety (a blue "O" on a black background) cues in a random order for a total of twenty-four trials (**Figure 1**). The participant was informed that the threat cue indicates he or she has a 17% chance of being shocked (i.e., two of the twelve threat cues result in a shock), and the safety cue indicates he or she is safe from shock for that trial. To increase anticipatory anxiety in our participants, we did not apply the shock before the experimental procedure and instead used a uniform shock generated by a physiological stimulator (Coulbourn Instruments, Allentown, PA) that lasted for 20 ms at 4 mA. This current was selected to provide a shock that is uncomfortable but not painful.

Each trial began with a 1-s threat or safety cue followed by an anticipation period that varied among 4, 6, 8, or 10 s, during which the participant focused on a fixation cross. A small dot indicated the end of the anticipation period, during which the shock was delivered on 17% of the threat trials. A blank screen was then presented for a 4-, 6-, 8-, or 10-s resting period, separating each trial. At the end of each condition, the participant used the button box to rate his or her subjective feelings of unpleasantness (valence) and agitation (arousal) on the 9-point pictorial Self-Assessment Manikin (SAM) scales (Bradley and Lang, 1994). We did not observe any significant effects of handholding, or indeed of state or trait anxiety, on subjective reports of valence and arousal (all *p*'s *>*.18).

### **IMAGE ACQUISITION**

Images were acquired using a Siemens 3.0 Tesla MAGNETOM Trio high-speed magnetic imaging device with a CP transmit/receive head coil and integrated mirror. One hundred seventy-six high-resolution T1-magnetization-prepared rapidacquisition gradient echo slices were collected to determine the localization of function (1-mm slices, *TR* = 1900 ms, *TE* = 2*.*53 ms, flip angle = 9◦, FOV = 250 mm, voxel size = 1 × 1 × 1 mm). Two hundred sixteen functional T2∗-weighted Echo Planar images (EPIs) sensitive to BOLD contrast were collected per block, in volumes of twenty-eight 3.5-mm transversal echoplanar slices (1-mm slice gap) covering the whole brain (1-mm slice gap, *TR* = 2000 ms, *TE* = 40 ms, flip angle = 90◦, FOV = 192 mm, matrix = 64 × 64, voxel size = 3 × 3 × 3*.*5 mm).

Data was preprocessed using FMRIB's Software Library (FSL) software *(Version 5.98; www.fmrib.ox.ac.uk/fsl)*. Motion was corrected using FMRIB's Linear Image Registration Tool, an intramodal correction algorithm tool (MCFLIRT; Jenkinson et al., 2002). In a separate step, we performed slice scan-time correction and a high-pass filtering cutoff point of 100 s, removing signals that were irrelevant to the stimuli. We used BET (Smith, 2002) brain extraction, which eliminated unwanted, nonbrain material voxels in the fMRI data, and conducted spatial smoothing with a 5-mm full width at half minimum Gaussian kernel. Images were registered to the Montreal Neurological Institute (MNI) standard space by FLIRT (Jenkinson et al., 2002). Threat trials where participants actually received shocks were excluded from analysis due to possible movement artifacts.

## **FUNCTIONAL REGIONS AND DATA ANALYSIS**

Data analysis was conducted using FEAT (fMRI Expert Analysis Tool) Version 5.98 in the FSL package. For first level analysis, in order to compare the neural response to threat of shock, threat minus safety maps were created by subtracting the response to the safety cue from the response to the threat cue for each handholding condition. We chose to model the difference between the threat cue and the safety cue rather than between the threat cue and the resting period due to the ambiguity inherent in experimentally uncontrolled periods of rest (cf. Coan et al., 2006a). Moreover, only threat trials in which a shock did not occur were included for analysis in order to reduce undesirable movement artifact. For second level analysis, these data were collapsed across all three functional runs, one for each handholding condition, for each individual participant using a fixed effects model. The threat minus safe contrast from the first level was carried into the third level, where between-subjects analysis was done separately for each handholding condition, as well as on contrasts of handholding conditions (Alone minus Stranger, Alone minus Friend, Stranger minus Friend). This was accomplished using a mixed effects model with state and trait anxiety entered as covariates. All clusters were whole brain-corrected and met clusterwise thresholding of *z >* 2*.*3 and a corrected cluster significance level of *p <* 0*.*05. Anatomical labels for brain regions were identified using the Harvard-Oxford cortical and subcortical atlases. To more closely examine interactions between state and trait anxiety and handholding conditions, we extracted mean percent signal change from the hypothalamus, putamen, and multiple sites within the precuneus. All coordinates are reported in Montreal Neurological Institute (MNI) space.

#### **RESULTS**

#### **STATE AND TRAIT ANXIETY INVENTORY RESULTS**

Prior to entering the fMRI scanner, all participants completed both portions of the STAI (Spielberger, 1983). In our sample, participants scored *m* = 34*.*76, *SD* = 9*.*66 (range = 20–59) on the State portion and *m* = 32*.*6, *SD* = 9*.*08 (range = 20–50) on the Trait portion. To check for multicollinearity, we examined the correlation between state and trait anxiety. Because state and trait anxiety showed a moderate correlation in our sample (*r* = 0*.*42, *p* = 0*.*02), we tested each of them as separate predictors.

#### **MAIN EFFECTS OF THREAT AND HANDHOLDING**

We found main effects of threat cues minus safety cues in several areas previously found both by us and others to be active during threat anticipation (e.g., Ploghaus et al., 1999; Coan et al., 2006b). Some of these areas included the anterior cingulate cortex (ACC), posterior cingulate cortex (PCC), orbitofrontal cortex (OFC), frontal pole, angular gyrus, precentral gyrus, supramarginal gyrus, occipital cortex, caudate, putamen, pallidum, and thalamus. Main effects of handholding condition on the neural reactivity to threat in this sample have been reported elsewhere (Coan et al., 2013) and are therefore discussed only briefly here. As anticipated based on Coan et al. (2006b), threatrelated (threat minus safety) activity was significantly lower in the Friend condition than in the Alone condition in the ACC, the left superior frontal gyrus, and the left supplementary motor cortex. Interestingly, threat-related activity was lower in the Stranger condition compared to the Friend condition in the left putamen.

#### **TRAIT ANXIETY BY HANDHOLDING CONDITIONS**

We first identified regions of neural activity during threat that correlated with trait anxiety in each independent handholding condition (Alone, Stranger, and Friend). As described in detail below, significant negative correlations with trait anxiety were found in the Alone and Friend conditions (**Table 1**). No significant correlations between trait anxiety and neural activity were found in the Stranger condition.

#### **ALONE CONDITION**

Trait anxiety significantly negatively correlated with brain activity in the Alone condition in five main clusters (**Table 1**, **Figures 2A–E**). The first cluster reached peak activity in the left precuneus and extended to the PCC, right temporo-occipital inferior temporal gyrus, right temporal occipital fusiform gyrus, right lingual gyrus, left cerebellum, left superior lateral occipital cortex, left insula, left inferior frontal gyrus pars opercularis, left

#### **Table 1 | Main effects of trait/state anxiety levels, by condition.**

angular gyrus, left central opercular cortex, and precentral gyrus (**Figure 2A**). The second cluster peaked in the precentral gyrus and extended to the PCC, ACC, supplementary motor cortex, precentral gyrus, left postcentral gyrus, left middle frontal gyrus, left anterior supramarginal gyrus (**Figure 2B**). The third cluster peaked in the right lateral occipital cortex and extended to the right supracalcarine cortex, right lateral occipital cortex, and right temporo-occipital middle temporal gyrus. (**Figure 2C**) The fourth cluster peaked in the left lingual gyrus and extended to the left posterior parahippocampal gyrus, left posterior temporal fusiform cortex, left occipital fusiform gyrus, left temporal occipital fusiform cortex, and cerebellum (**Figure 2D**). The fifth cluster also peaked in the left precuneus and extended into the left superior lateral occipital cortex and left superior parietal lobule (**Figure 2E**).

**FIGURE 2 | Clusters of activity significantly correlated with trait anxiety in the Alone condition.** Blue areas indicate a negative correlation with trait anxiety. **(A)** Cluster 1, peak activation in left precuneus. **(B)** Cluster 2, peak activation in precentral gyrus. **(C)** Cluster 3, peak activation in right lateral occipital cortex. **(D)** Cluster 4, peak activation in left lingual gyrus. **(E)** Cluster 5, peak activation in left precuneus.


**+***, Indicates a positive correlation;* **−***, Indicates a negative correlation.*

### **FRIEND CONDITION**

Trait anxiety negatively correlated with brain activity in the Friend condition in two main clusters, with the first peaking in the left superior parietal lobule and extending to the PCC, left precuneus, and left superior lateral occipital cortex (**Figure 3A**), and the second peaking in the left frontal medial cortex and extending to the paracingulate gyrus and left frontal pole (**Figure 3B**).

### **STATE ANXIETY BY HANDHOLDING CONDITIONS**

Next, we identified regions of neural activity during threat that correlated with state anxiety in each independent handholding condition. Only the Friend condition yielded significant correlations (**Table 1**).

### **FRIEND CONDITION**

Significant positive correlations with state anxiety were found in the Friend condition in three main clusters (**Figures 3C–E**). The first cluster peaked in the PCC and extended to the precuneus (**Figure 3C**). The second cluster peaked in the frontal pole and extended to the paracingulate (**Figure 3D**). The third cluster peaked in the left superior lateral occipital cortex and extended to the left angular gyrus, left inferior lateral occipital cortex, and the temporo-occipital middle temporal gyrus (**Figure 3E**).

### **INTERACTIONS BETWEEN HANDHOLDING AND TRAIT ANXIETY**

Previous research has shown that handholding by and physical proximity to close relational partners tends to attenuate threatrelated neural activity (Coan et al., 2006b; Conner et al., 2012). To investigate whether levels of trait and state anxiety moderated this relationship, we employed additional contrasts (Alone minus Stranger, Alone minus Friend, Friend minus Stranger) to compare the association between anxiety and neural activity across handholding conditions. All three contrasts (Alone minus Stranger, Alone minus Friend, Friend minus Stranger) showed negative

**FIGURE 3 | Clusters of activity significantly correlated with trait or state anxiety in the Friend condition.** Blue areas indicate a negative correlation with trait anxiety; green areas indicate a positive correlation with state anxiety. **(A)** Cluster 1, peak activation in left superior parietal lobule. **(B)** Cluster 2, peak activation in left frontal medial cortex. **(C)** Cluster 3, peak activation in posterior cingulate cortex. **(D)** Cluster 4, peak activation in frontal pole. **(E)** Cluster 5, peak activation in left superior lateral occipital cortex.

correlations between trait anxiety and threat-related brain activity (**Table 2**).

# **ALONE MINUS STRANGER**

Subtracting brain activity correlated with trait anxiety in the Stranger condition from that in the Alone condition provides an index of brain areas that contain correlations with trait anxiety that are significantly stronger in the Alone compared to Stranger condition. The Alone minus Stranger contrast yielded one cluster of neural activity that was significantly and negatively correlated with trait anxiety. This cluster peaked in the left precentral gyrus and extended to the postcentral gyrus and precuneus. This indicated that for those with lower levels of trait anxiety, threat-related brain activity was higher in the Alone compared to Stranger condition, or, conversely, for those with higher levels of trait anxiety, brain activity during the threat cues contrasted with the safety cues was decreased in the Alone relative to the Stranger condition.

### **ALONE MINUS FRIEND**

Trait anxiety significantly and negatively correlated with the Alone minus Friend contrast in two clusters, with one cluster peaking in the hypothalamus and extending to the substantia nigra, right pallidum, thalamus, insula, and putamen (**Figure 4A**), and another cluster peaking in the left putamen and extending to the left thalamus, insula, and pallidum (**Figure 4B**). These effects indicated that for those higher in trait anxiety, neural threat activation was decreased in the Alone condition relative to the Friend condition.

## **FRIEND MINUS STRANGER**

The contrast Friend minus Stranger resulted in one cluster of activity significantly and negatively correlated with trait anxiety. This cluster peaked in the left precuneus and extended to the left postcentral gyrus, indicating that for those with higher trait anxiety, neural threat activity was decreased in the Friend condition relative to the Stranger condition (**Figure 5A**).

## **INTERACTIONS BETWEEN HANDHOLDING AND STATE ANXIETY**

State anxiety showed a significant negative correlation with threat-related brain activity in the Alone minus Friend contrast and a positive correlation in the Friend minus Stranger contrast (**Table 2**).

## **ALONE MINUS FRIEND**

One cluster in the Alone minus Friend contrast negatively correlated with state anxiety. This cluster peaked in the precuneus and extended to the PCC. This indicates that for those with higher state anxiety, neural activity was decreased in the Alone condition relative to the Friend condition (**Figure 4C**).

## **FRIEND MINUS STRANGER**

One cluster significantly positively correlated with state anxiety, peaking in the precuneus and extending into the PCC and lingual gyrus. In other words, for those higher in state anxiety, threatrelated brain activity was higher in the Friend relative to the Stranger condition (**Figure 5B**).


**Table 2 | Interactions between handholding conditions and trait/state anxiety levels.**

**+***, Indicates a positive correlation;* **−***, Indicates a negative correlation.*

#### **INDIVIDUAL THREAT AND SAFETY CUES WITH STATE AND TRAIT ANXIETY**

Because it is conceptually difficult to interpret correlations with fMRI contrast images, we considered the possibility that anxiety-related differences across handholding conditions were related to altered neural activity during the safety cues rather than during the threat cues. To explore this, we modeled the safety and threat cues independently with state and trait anxiety for each handholding condition. We saw no significant relationships between state or trait anxiety and neural activity during either the threat or safety periods in any condition, suggesting that people with higher state or trait anxiety did not have significantly different baseline or threat activity. We saw one exception: state anxiety was positively associated with neural activity during safety cues in the Stranger condition. This activity peaked in the left cuneal cortex and extended to the right cuneal cortex, the bilateral occipital pole, lateral occipital cortex, lingual gyrus, and cerebellum, and the right intracalcarine cortex.

## **DISCUSSION**

Previously, we found that supportive social contact delivered via handholding reduced threat-related neural activity in the ACC, left superior frontal gyrus, and left supplementary motor cortex (Coan et al., 2013). Using the same sample, we examined how anxiety levels might interact with the presence or absence of supportive social contact to predict neural responses in the presence of a potential threat. Although trait and state anxiety were moderately correlated, their associations with active threats—both while alone and in a social context—were quite different. On the one hand, although trait anxiety was unrelated to the threat-safe contrast during supportive handholding, the same contrast was *negatively* associated with trait anxiety when participants were alone. This pattern was observed throughout the brain, implicating processes as diverse as self-focus, emotion, and working memory (e.g., precuneus, PCC, portions of the default mode network, cf. Maddock et al., 2003; Cavanna and Trimble, 2006; Zhao et al., 2007); motor preparation and coordination (e.g., precentral gyrus, supplementary motor cortex, and cerebellum, cf. Liotti et al., 2000; Critchley et al., 2004); and even visual attention (e.g., lateral occipital cortex and lingual gyrus, cf. Hopfinger et al., 2000; Murray and Wojciulik, 2004). While holding a friend's hand, trait anxiety corresponded with decreased brain activity mainly in the superior parietal lobule and frontal medial cortex, whereas higher state anxiety corresponded with *increased* brain activity in areas such as the PCC, frontal pole, and lateral occipital cortex. When holding a stranger's hand, neither trait nor state anxiety showed any association with brain activity. Close examination of these results suggested that trait anxiety indeed corresponded with smaller differences between threat and safety cues when a participant was alone in the scanner, relative to holding a stranger's or friend's hand.

These findings are consistent with the vigilance/disengagement model—that a curvilinear association between anxiety and neural output exists, such that moderate threats induce increased neural threat activity indicative of increased arousal and orientation to the threat, whereas strong threats induce decreased neural activity, signaling a disengagement or avoidance of the stimulus. Based on these and earlier results, we propose that social support can alter the perception threat—as well as the brain's multifaceted response to that threat—especially when the support is provided by a familiar friend. Moreover, the seemingly paradoxical impact of support on individuals high in trait anxiety may suggest some important clinical implications, for how anxiety is both understood and treated.

Previously, anxiety and related traits such as neuroticism have been characterized by increased reactivity to stress (Bolger and Schilling, 1991; Mroczek and Almeida, 2004). Accordingly, many studies employing neuroimaging have observed increased activity in threat-related brain regions in anxious individuals when anticipating an aversive stimulus (Canli et al., 2001; Simpson et al., 2001; Simmons et al., 2006; Haas et al., 2007). However, this finding has not been universal—in line with our findings, some have reported *decreased* neural activity in more anxious individuals (Kumari et al., 2007; Straube et al., 2009; Drabant et al., 2011).

We speculate that one key variable in resolving these discrepant findings may be how intensely the participant perceives the aversive stimulus during the anticipatory period. In general, studies employ an unchanging threat (e.g., a fixed level of shock) throughout the experiment. To examine the effect of varying the threat level on brain activity, Straube et al. (2009) employed a threat-of-shock paradigm in which participants underwent fMRI scanning while viewing cues indicating they might receive either no shock, mild shock, moderate shock, or strong shock, as subjectively rated by the participant prior to the scan. Participants retroactively reported their levels of state anxiety while anticipating each threat level. During moderate threat, positive correlations between anxiety and activity in the ventromedial prefrontal cortex (VMPFC) and vACC were found. Yet, during strong threat, these correlations became negative, while activity in the dorsal ACC, somatosensory cortex, motor cortex, and hippocampus showed positive correlations with anxiety.

While the Straube et al. (2009) study provides evidence of a curvilinear relationship that may help explain our findings, it is important to note that they assessed state anxiety *after* the fMRI scan and individually for each level of threat, whereas we looked at general measures of state and trait anxiety administered prior to the shock task. A more recent study measured levels of trait neuroticism prior to employing an fMRI paradigm in which level of shock was varied (Drabant et al., 2011). Results of this study showed a negative correlation between neuroticism and activity in the inferior frontal gyrus and insula in strong shock trials compared to moderate shock trials. The authors suggested that people high in neuroticism may switch to an avoidant processing strategy in the face of high threat, as would be predicted by the vigilance/disengagement model. Highly threatening stimuli seem to result in lower levels of threat-related brain activity in people with greater anxiety-related traits.

The vigilance/disengagement model we propose is consistent with an inverted U-shaped model of arousal outlined by Wilken et al. (2000)in which increasing arousal input (e.g., greater threat) also increases arousal output (e.g., physiological arousal) up to a point, at which output begins to decrease. They suggested that highly trait anxious individuals might be more aroused (and vigilant) at their "baseline," such that severe stressors place them beyond the peak of the inverted U. Along these lines, we suggest first that the more anxious people in our sample may have perceived the threat of shock as a strongly threatening stimulus, leading to the regional neural deactivations we observed. Second, as we have previously documented the buffering effect of handholding on the brain's response to threat cues (Coan et al., 2006b, 2013), the administration of supportive handholding may have lowered the perception of threat in the highly anxious people to less intense levels (i.e., closer to the peak of the inverted U), resulting in increased neural activity. This model of the moderation of threat perception by social context is illustrated in **Figure 6**.

An important question to consider is whether our threat-ofshock paradigm could potentially be perceived as highly threatening. One limitation of our study is that we did not directly vary the level of shock, nor did we measure the subjective level of anxiety induced by the shock. However, we have several reasons to believe the nature of our shock paradigm is capable of inducing high levels of anticipatory anxiety. First, our threat cues did not indicate absolute certainty of shock; rather, we told the participants that the cues indicate a 17% chance of being shocked (and, indeed, we did shock them following 17% of the cues). This unpredictability may increase levels of negative affect and anxiety, as others have observed (Carlsson et al., 2006). Second, in contrast to other studies (Straube et al., 2009; Drabant et al., 2011), we did not shock the participants prior to entering the fMRI scanner; in other words, participants did not have a pre-formed expectation of the intensity of the shock until receiving a shock during the experiment. These factors, in combination with the tendency of people with

high trait anxiety to interpret stimuli as more threatening than those with low trait anxiety (Mogg et al., 2000), suggest that our more anxious participants viewed the shock as a strong threat.

An alternative explanation for our findings is that high trait anxiety serves as a buffering factor to physiological arousal under times of high stress. Highly trait anxious people show blunted secretion of stress hormones such as cortisol, adrenocorticotropic hormone, epinephrine, norepinephrine, and prolactin during a social stress task (Jezova et al., 2004) and lower electrodermal responses during cognitive and affective stressors (Wilken et al., 2000). A previous study found that individuals with higher levels of neuroticism show less discomfort and smaller autonomic nervous system reactivity to a high intensity stressor (LeBlanc et al., 2004). Interestingly, in the same sample, more neurotic individuals reported greater discomfort to a mild or moderate stressor compared to less neurotic individuals, a behavioral finding that further suggests the vigilance/disengagement model (LeBlanc et al., 2003).

While our study focused on levels of anxiety in a subclinical sample, decreases in brain activity have been observed in individuals with a variety anxiety disorders in response to negative or threatening stimuli. For example, PTSD patients show less activity in the thalamus, parahippocampal gyrus, and parietal areas compared to controls when recalling negative emotional states (Lanius et al., 2003). A study using magnetoencephalography found early increased frontal activity in response to aversive pictures in PTSD patients relative to controls, followed by deactivations in parieto-occipital areas (Adenauer et al., 2010). People with generalized anxiety disorder (GAD) show increased early cortical activity followed by reduced reactivity, relative to healthy controls (Weinberg and Hajcak, 2011). These and similar findings have been posited to be a "vigilance-avoidance" pattern, in which rapid assessment of a threat is followed by attentional disengagement from the stimulus once it has been deemed dangerous (Mogg et al., 2004). This attentional disengagement, while it may decrease anxiety in the moment, maintains the anxiety disorder in the long run, as it prevents an individual from habituating to the feared stimuli. Our fMRI findings may have captured attentional disengagement, similar to that seen in anxiety disorders, in our more anxious participants. That social support moderated anxious responding during threat has important implications for the etiology and maintenance of anxiety disorders. It may be that anxiety-prone individuals are particularly vulnerable to experiencing a threat as highly threatening in the absence of social support, yet the lowered arousal resulting from disengagement may paradoxically reinforce the avoidance of social contact. Further research should assess this possibility.

In conclusion, we examined how anxiety relates to the neural response to threat under conditions of social support. We

## **REFERENCES**


reactivity to emotional stimuli. *Behav. Neurosci.* 115, 33–42. doi: 10.1037/0735-7044.115.1.33


demonstrated that, when alone, participants with higher trait anxiety exhibited attenuated neural activity in several brain areas in response to a physically threatening stimulus, which we suggest is related to attentional disengagement. Upon receipt of social support via holding another person's hand, this effect largely disappeared or was reversed—brain activity in highly trait anxious people was indistinguishable from or slightly greater than that in less trait anxious people. These findings support a vigilance/disengagement model in which a curvilinear relationship between anxiety and threat results in decreased neural output past a certain threshold of threat intensity. That the provision of social support eliminated this effect suggests a role for supportive others in the treatment and prevention of anxiety disorders.

## **ACKNOWLEDGMENTS**

The authors acknowledge the moral and/or intellectual support of Marlen Z. Gonzalez, Casey Brown, Karen Hasselmo, Alexander Tatum, Zoe Englander, and Bethany A. Teachman. They also thank Joseph P. Allen for access to the KLIFF sample used in this study.

This work was supported by a National Institute of Mental Health grant, Award Number R01MH080725, awarded to James A. Coan.

response to threat. *Psychol. Sci*. 17, 1032–1039. doi: 10.1111/j.1467- 9280.2006.01832.x


threat anticipation vary as a function of threat intensity and neuroticism. *Neuroimage* 55, 401–410. doi: 10.1016/j.neuroimage.2010.11.040


and neuroticism in questionnaires. *Eur. J. Pers.* 2, 113–120. doi: 10.1002/per.2410020206


*Soc. Sci. Med.* 64, 401–410. doi: 10.1016/j.socscimed.2006.09.008


*Individ. Dif.* 51, 123–127. doi: 10.1016/j.paid.2011.03.026


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 16 April 2013; accepted: 11 August 2013; published online: 30 August 2013.*

*Citation: Maresh EL, Beckes L and Coan JA (2013) The social regulation of threat-related attentional disengagement in highly anxious individuals. Front. Hum. Neurosci. 7:515. doi: 10.3389/ fnhum.2013.00515*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Maresh, Beckes and Coan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The role of empathy in choosing rewards from another's perspective

# *Garret O'Connell , Anastasia Christakou , Anthony T. Haffey and Bhismadev Chakrabarti\**

*School of Psychology and Clinical Language Sciences, Centre for Integrative Neuroscience and Neurodynamics, University of Reading, Reading, UK*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Cintia R. Lucci, École Normale Supérieure, France Claudia Civai, University of Minnesota, USA Carla Sharp, University of Houston, USA*

#### *\*Correspondence:*

*Bhismadev Chakrabarti, School of Psychology and Clinical Language Sciences, Centre for Integrative Neuroscience and Neurodynamics, University of Reading, Reading RG6 6AL, UK. e-mail: b.chakrabarti@reading.ac.uk* As social animals, we regularly act in the interest of others by making decisions on their behalf. These decisions can take the form of choices between smaller short-term rewards and larger long-term rewards, and can be effectively indexed by temporal discounting (TD). In a TD paradigm, a reward loses subjective value with increasing delay presumably because it becomes more difficult to simulate how much the recipient (e.g., future self) will value it. If this is the case, then the value of delayed rewards should be discounted even more steeply when we are choosing for someone whose feelings we do not readily simulate, such as socially distant strangers. Second, the ability to simulate shows individual differences and is indexed by trait empathy. We hypothesized that individuals high in trait empathy will more readily simulate, and hence discount less steeply for distant others, compared to those who are low on trait empathy. To test these predictions, we asked 63 participants from the general population to perform a TD task from the perspectives of close and distant others, as well as their own. People were found to discount less steeply for themselves, and the steepness of TD increased with increasing distance from self. Additionally, individuals who scored high in trait empathy were found to discount less steeply for distant others compared to those who scored low. These findings confirm the role of empathy in determining how we choose rewards for others.

**Keywords: empathy, reward, temporal discounting, social distance, simulation**

## **INTRODUCTION**

As social beings, we do not just make decisions for ourselves, but regularly have to make decisions on behalf of others. We invest great effort into judging what someone else would like when buying gifts, or making plans for them. Consider the case of a husband trying to decide what his wife would prefer: a fancy dinner out this evening or a weekend trip away in 2 weeks' time. How do we make such decisions? A body of literature on how we make choices for ourselves shows that a key role in these decisions is played by our emotional state (Damasio et al., 1991). If our own emotions are crucial to making choices for ourselves, it follows that we need a good understanding of another person's emotions and mental states in order to make choices on their behalf. Empathy is a trait that quantifies this capacity to understand others' emotions and mental states and respond appropriately to them (for a review, see Chakrabarti et al., 2006). In the current paper, we examine the role of empathy in making choices on another's behalf in an intertemporal context.

One of the most commonly encountered choices are those between short-term and long-term rewards. Such intertemporal preferences are indexed by temporal discounting (TD). In a typical TD paradigm, a series of choices between smaller immediate and larger delayed monetary amounts are presented. The commonly observed response pattern is that with increasing delay, the more immediate though lesser rewards are preferred over larger, later rewards. The rate at which rewards are subjectively devalued slows down as delay increases, resulting in a steep-to-flat "discounting curve," suggesting that rewards are devalued with time more rapidly over shorter delays than longer delays (Ainslie, 1975). This discounting function has been associated with intelligence (Mischel and Metzner, 1962; Kirby et al., 2005; Shamosh et al., 2008), impulsivity (Bickel et al., 1999; de Wit et al., 2007; Christakou et al., 2011), and consequential life outcomes such as health, wealth and social-functioning (Mischel et al., 1989; Moffitt et al., 2011). While predictors of how individuals discount when they have to make choices about themselves have been well investigated, little research has focused on the discounting functions for others.

It has been suggested that we devalue delayed rewards because we empathise less with the feelings of their recipient (i.e., future selves) (Loewenstein, 1996). A key process underlying empathy is that of simulation, i.e., the ability to put ourselves in the shoes of others (Gordon, 1992; Shanton and Goldman, 2010). Simulation provides a potential mechanism to understand how another person feels by imagining how we ourselves would feel in their situation, and has been proposed to underlie theory of mind (Shanton and Goldman, 2010). This mechanism applies equally to ourselves, i.e., we put ourselves in the shoes of our future selves, to predict how we will feel in the future. Recent functional neuroimaging studies provide indirect evidence for simulation, by showing involvement of the ventromedial prefrontal cortex in making value-based decisions for self as well as for others (Nicolle et al., 2012; Suzuki et al., 2012; Janowski et al., 2013). Neural and other indices of simulation (e.g., vicarious pain responses) are greater if the person is a socially close one (i.e., familiar or liked) than if the person is socially distant (Singer et al., 2006; Xu et al., 2009; Cheng et al., 2010). Arguably, simulation is easiest if the person to simulate is one's own self (minimum social distance). Social distance could thus be viewed as a proxy measure for ease of simulation.

The above effect has direct implications for TD for self and others. For the self, it suggests that increasing the delay to reward reduces empathy for the recipient by increasing their social distance. This was supported by a set of studies by Bartels and Rips (2010) showing that social distance (operationalized by the authors as "psychological connectedness") with the future self was directly proportional to the rate of discounting. For the other, it predicts that people will tend to discount less steeply for themselves and close others (who are easy to simulate), compared to distant others (who are difficult to simulate) (see **Figure 1**).

The ability to put one's self in another's place and simulate their feelings is indexed by trait empathy. If TD changes as a function of simulation, it is expected that highly empathic people (who simulate easily) will discount less steeply when making choices on behalf of others. As a corollary, people low in empathy will find it difficult to simulate distant others, and hence will discount more steeply when making choices on their behalf. To test these predictions, we examined TD from the perspective of others at different social distances. It is important to note that this is not equivalent to *social discounting*, in which rewards for others are discounted between close and distant others with no delay (Jones and Rachlin, 2006).

Specifically, we predicted that:


# **MATERIALS AND METHODS**

### **PARTICIPANTS**

76 participants (38 female; age: *M* = 24*.*7 years, *SD* = 1*.*52), drawn largely from the university student population, consented to participate and received £6 for their time. An exclusion criterion was being a non-native English speaker. This study was approved by the University of Reading Research Ethics Committee.

## **TRAIT EMPATHY MEASURES**

Participants completed online versions of the Empathy Quotient (EQ; Baron-Cohen and Wheelwright, 2004) and the Interpersonal Reactivity Index (IRI; Davis, 1980). The personal distress subscale of the IRI was omitted as it was not directly relevant to this study.

## **SOCIAL DISTANCE PROCEDURE**

A social distance procedure was used to identify close and distant others. This task measures perceptions of others across dimensions of familiarity, similarity and kinship (Liviatan et al., 2008; Osinski, 2009 ´ ). Participants were first instructed to list persons they know in descending order of familiarity between 1 and 100 at selected positions [as described in (Jones and Rachlin, 2006)]. Persons identified at the 4th and 43rd positions were used as close and distant others respectively, based on the observation that these points covered the maximum rate of change of the social discounting curve in a previous report (Jones and Rachlin, 2006). There were no restrictions on the category of relationship that could be used for these positions (e.g., spouse, sibling, friend), nor was this data collected.

## **TEMPORAL DISCOUNTING TASK**

For this task, the following instructions were given by the experimenter:

"This task involves a series of choices between smaller amounts of money now or larger amounts of money later. However, you will also be asked to perform the task as if you were someone you know. Try and put yourself in their shoes and imagine how they would respond."

Participants were told to try and not to factor in particulars about their or the others' financial situation, only to select the preferred option. Because of the current study's focus on empathy, participants were instructed to make the decision *from the perspective* of the other, rather than *for the benefit* of other, which is more akin to altruism. To avoid self-bias (i.e., participants resorting to responding with their own preferences without considering how others might differ), decisions for self-took place after decisions for others had been completed, as suggested by Faro and Rottenstreich (2006).

The order of blocks (one each) for close and distant conditions was counterbalanced across participants. The task was run using E-Prime version 2.0. The person's perspective from whom the task was to be performed was shown before every trial and under the options during selection. Task options were between a variable immediate amount (*<* £100) or £100 at one of a randomly ordered sequence of 6 delays (weeks: 1, 3; months: 2, 5, 9, 18). Both immediate and delayed amounts were presented together and the participant selected the left or right amount with a keystroke (see **Figure 2**). The sides on which the immediate and delayed rewards were presented were counterbalanced across participants. The double-limits algorithm was used to estimate the variable amount (Johnson and Bickel, 2002) and indifference points (i.e., immediate amount value at which choices of £100 at a given delay were equally likely) used to map TD curves. Rate of TD was estimated using the area-under-the-curve (AuC) of the plot of indifference points against time (i.e., *the higher the AuC, the lower the steepness/rate of discounting*) (Myerson et al., 2001). To reduce floor/ceiling effects, participants who discounted less than 5% or more than 95% after a delay of 5 months in more than one condition were removed.

# **RESULTS**

After screening participants for the exclusion criteria (8 due to TD criteria, 5 due to non-native speaker criteria), 63 participants (34 females, 29 males; age: *M* = 23*.*8 years, *SD* = 1*.*38) remained. Due to the direction of predicted effects, results of planned *post-hoc* comparisons are reported at the 1-tailed level. IRI questionnaire data were lost for two participants due to a technical fault.

To test that individuals exhibited TD, a one-way repeated measures ANOVA was performed with delay as a within-subjects factor and indifference points in the self-condition as a dependent variable. A significant effect of delay was observed, *F(*5*,* <sup>310</sup>*)* = 180*.*56, *p <* 0*.*001, η<sup>2</sup> *<sup>p</sup>* = 0*.*477.

To test the effect of social distance on TD, a one-way repeated measures ANOVA was performed with social condition (self/close/distant other) as a within-subjects factor and rate of TD as a dependent variable. A significant effect of social distance was observed, *<sup>F</sup>(*2*,* <sup>124</sup>*)* <sup>=</sup> <sup>5</sup>*.*12, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*007, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*076. Planned contrasts (Bonferroni corrected) showed a significant difference between self (*M* = 958, *SD* = 58*.*92) and distant other (*M* = 731, *SD* = 62*.*39), *t(*62*)* = 3*.*07, *p* = 0*.*003, *d* = 0*.*39. A marginally significant difference was seen for TD between self and close other (*M* = 850, *SE* = 62*.*18), *t(*62*)* =

1*.*91, *p* = 0*.*06, *d* = 0*.*24, but not significant between TD for close and distant others [*t(*62*)* = 1*.*48, *p* = 0*.*143, *d* = 0*.*24] (**Figure 3**). Pearson's correlations were performed to examine the association between the steepness of discounting and trait empathy in specific task conditions (see **Table 1**). Specifically, TD for distant others was negatively associated with empathy scores (i.e., individuals high in empathy discounted less steeply for distant others, *r*EQ-distant*(*61*)* = 0*.*220, *p* = 0*.*042, *r*IRI-distant*(*59*)* = 0*.*310, *p* = 0*.*008). No association was found for self [*r*EQ-self*(*61*)* = 0*.*027, *p* = 0*.*416, *r*IRI-self*(*59*)* = 0*.*050, *p* = 0*.*352], or close conditions [*r*EQ-close*(*61*)* = 0*.*162, *p* = 0*.*103, *r*IRI-close*(*59*)* = 0*.*134, *p* = 0*.*152].

# **DISCUSSION**

Making decisions on behalf of others is a common part of everyday life, and yet how we make these choices remains largely unknown. In this experiment, we tested the hypothesis that this process is influenced by social distance and trait empathy. Choice behavior was operationalized using a TD paradigm, where participants were asked to choose between a series of immediate and distant rewards on behalf of close and distant others, as well as themselves. We found that (1) people discount less steeply for themselves compared to others, and that the steepness of discounting for others increases with social distance,

**FIGURE 3 | Temporal discounting curves for social conditions.** Error-bars at 95% CI and adjusted for within-subjects variance (significantly different at: ∗0*.*05 level; ∗∗0*.*01 level).

**Table 1 | Correlations between temporal discounting in each social condition and trait empathy measures (significantly correlated at the: ∗0***.***05 level; ∗∗0***.***01 level).**


**are presented vary within blocks).**

(2) compared to people who score low on trait empathy, highly empathic people discount less steeply for distant others.

TD was steeper as the social distance of reward recipients increased across conditions of self, close and distant others. Participants were explicitly instructed to simulate the reward recipients in the task (i.e., "put yourself in the shoes of the recipient"). The ease of simulation varies as a function of social distance, i.e., simulation is easiest when one is making choices about oneself, and participants were found to discount least steeply in this condition. People who are similar to one's self, or who are socially close, are easier to simulate compared to socially distant others. Consistent with this, rewards for self-similar persons are found to be higher in subjective value and show higher activation of reward-related brain areas, when compared to those for self-dissimilar persons (Mobbs et al., 2009). Finally, simulation is most difficult when making choices on behalf of a distant other. As expected, discounting for distant others produced the steepest slope. These parallel findings show that people tend to choose more immediate compared to delayed rewards as psychological connectedness with the reward recipient reduces (Bartels and Rips, 2010).

In this experiment, participants always performed TD for self-last to avoid reported self-bias effects (i.e., the increased tendency to use one's own preferences as a default in choosing for others after having made the same choices for self). It is possible that such biasing effects may work both ways, such that choices for oneself made after choices for others are biased toward others' predicted preferences. However, our results are concordant with those observed by Beisswanger et al. (2003), who used a between-group design to avoid order effects, and reported that choices for others were more impulsive compared to choices for self. Given that impulsive choices are associated with steeper TD (Alessi and Petry, 2003), this supports the present finding that intertemporal choices are more impulsive for others than for self.

Our results contrast with a recent report by Ziegler and Tunney (2012), which shows that choices for others in a TD paradigm become less impulsive as social distance increases. Critically, participants in their task were instructed to select the option that another *should* select. This frames the choice in a way that biases participants toward self-control (i.e., choosing the reward that is best for the recipient, which may not be the reward the recipient would choose on his/her own). This key difference in the frame of operation for the TD task can potentially explain the divergent results.

Simulation is a key empathic mechanism for internally representing the emotions and motivations of others (Keysers and Gazzola, 2007). Accordingly, we expected that the individuals who are high in trait empathy would be able to simulate distant others more easily, and hence discount less steeply when making choices on their behalf. This hypothesis was supported as trait empathy (using two separate trait measures) was inversely related to the steepness of TD for distant others. This finding replicates a previous report showing that people who score higher in trait empathy make more self-similar choices for others (Faro and Rottenstreich, 2006). Additionally, the results suggest a role for empathy in making intertemporal choices, which was elegantly predicted by Loewenstein almost two decades ago (Loewenstein, 1996). This result is also consistent with previous work that suggests a link between TD and other social behaviors. The steepness of TD is negatively correlated with altruistic tendencies (Harris and Madden, 2002; Yi et al., 2005). Steeper TD has also been reported in persons with social anxiety (Rounds et al., 2007), a trait marked by the reduced motivation to affiliate with others (Mallott et al., 2009).

In our study, no significant association was observed between TD for self/close other and trait empathy. This would be expected if the value of delayed rewards for recipients is indexed by individual differences in how easily we can simulate them. This null finding replicates a previous report in which a positive association between trait empathy and less-impulsive choices was noted for others, but not for self (Faro and Rottenstreich, 2006). We speculate that there are two possible reasons why this null finding was observed. First, both trait measures of empathy (IRI and EQ) ask questions about hypothetical unknown/distant others, which increases the sensitivity of these measures over larger social distances. Secondly, the self and close other conditions might be susceptible to ceiling effects in simulation; effects that compress individual differences and make their association with trait empathy difficult to observe. A possible method to overcome this limitation could be to use longer delays in the TD task, reducing the ease of simulation for future selves as done by Bartels and Rips (2010). A caveat of the current study is that the observed relationship between empathy and TD for others may not generalize to the entire lifespan, since the age range of the current study is fairly narrow. Future research should examine this relationship in other age groups, particularly adolescence, when immature discounting is observed (Christakou et al., 2011; Sharp et al., 2011). A second direction for future work is to test the hypothesized role of simulation in intertemporal choices for others using objective indices of simulation measured by psychophysiological and neuroimaging techniques. Current experiments in our lab are testing this.

In this experiment, we show how social distance influences choice of future rewards for self and others, by showing that people discount least steeply for themselves, and most steeply for distant others. We interpret this using a simulation based account of empathy that suggests that socially distant people are most difficult to simulate. Crucially, we find that trait empathy influences how we choose rewards for others; highly empathic people discount less steeply for distant others. Future research should examine these processes in psychopathological populations with deficits in both reward and empathy processes, such as people with addiction (Gizewski et al., 2012), attention disorders (Marton et al., 2008) and those diagnosed with Autism Spectrum Conditions (Schmitz et al., 2008; Dichter et al., 2010; Scott-Van Zeeland et al., 2010).

#### **ACKNOWLEDGMENTS**

Garret O'Connell is supported by a University of Reading doctoral studentship. Anastasia Christakou is supported by the Human Frontier Science Program. Anthony T. Haffey and Bhismadev Chakrabarti are supported by the Medical Research Council UK.

## **REFERENCES**


with delay discounting in middleaged adults. *Person. Individ. Differ.* 42, 111–121.


in disrupting affiliative behavior. *Depress. Anxiety* 26, 438–446.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 15 November 2012; accepted: 18 April 2013; published online: 23 May 2013.*

*Citation: O'Connell G, Christakou A, Haffey AT and Chakrabarti B (2013) The role of empathy in choosing rewards from another's perspective. Front. Hum. Neurosci. 7:174. doi: 10.3389/fnhum. 2013.00174*

*Copyright © 2013 O'Connell, Christakou, Haffey and Chakrabarti. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Unsupervised learning of facial emotion decoding skills

# *Jan O. Huelle1,2 †, Benjamin Sack1 †, Katja Broer 1, Irina Komlewa1 and Silke Anders1\**

<sup>1</sup> Social and Affective Neuroscience, Department of Neurology, Universität zu Lübeck, Lübeck, Germany <sup>2</sup> School of Ophthalmology, South West Peninsula Postgraduate Medical Education, Plymouth, UK

#### *Edited by:*

Leonie Koban, University of Colorado Boulder, USA

#### *Reviewed by:*

Brian Thomas Leitzke, University of Wisconsin–Madison, USA Katja Schlegel, University of Geneva, Switzerland

#### *\*Correspondence:*

Silke Anders, Social and Affective Neuroscience, Department of Neurology, Universität zu Lübeck, Ratzeburger Allee 160, 23538 Lübeck, Germany e-mail: silke.anders@ neuro.uni-luebeck.de

†Jan O. Huelle and Benjamin Sack contributed equally to this work.

Research on the mechanisms underlying human facial emotion recognition has long focussed on genetically determined neural algorithms and often neglected the question of how these algorithms might be tuned by social learning. Here we show that facial emotion decoding skills can be significantly and sustainably improved by practice without an external teaching signal. Participants saw video clips of dynamic facial expressions of five different women and were asked to decide which of four possible emotions (anger, disgust, fear, and sadness) was shown in each clip. Although no external information about the correctness of the participant's response or the sender's true affective state was provided, participants showed a significant increase of facial emotion recognition accuracy both within and across two training sessions two days to several weeks apart. We discuss several similarities and differences between the unsupervised improvement of facial decoding skills observed in the current study, unsupervised perceptual learning of simple visual stimuli described in previous studies and practice effects often observed in cognitive tasks.

**Keywords: dynamic facial expressions, emotional facial expressions, unsupervised learning, perceptual learning, social learning, cross-cultural learning, empathy**

# **INTRODUCTION**

Dating from Darwin's notion that "the different races of man express their emotions [...] with remarkable uniformity" (Darwin, 1872) facial expressions of emotion have long been viewed as a hard-wired product of evolution that is universally understood across human cultures and, to some extent, even mammalian species. Although most researchers now agree that human emotional facial expressions can vary considerably across social groups and cultures (for a meta-analysis see Elfenbein and Ambady, 2002), few studies have aimed to systematically investigate how encoding and decoding of facial expressions is shaped by social learning. Furthermore, the majority of studies that did investigate learning of facial emotion recognition aimed to develop training programs that might improve the participants' social or inter-cultural skills and therefore mixed different types of training (e.g., McAlpine et al., 1992; Stewart and Singh, 1995; Bölte et al., 2002; Silver et al., 2004; Solomon et al., 2004; Wölwer et al., 2005; Matsumoto and Hwang, 2011).

Theoretical work has suggested that associative learning during infancy might play an important role in the acquisition of facial decoding skills. The reasoning is that because infants are often exposed to similar emotional contexts as their mothers, the sight of their mother's facial expression in a given context becomes gradually associated with the infant's own emotional state in that context through Hebbian learning. Such associative learning, it is argued, can take place even if the infant's and the mother's emotional state are different because the mothers often mirror the infant's emotional state (Keysers and Perrett, 2004; Keysers and Gazzola, 2006). It has further been proposed that once these links have been established, contextual cues might be sufficient to fine-tune associations between observed facial expressions and emotional meaning. Indeed, the few studies that have systematically investigated learning of facial emotional recognition provide evidence that facial decoding skills can be sharpened both in adults (Elfenbein, 2006) and children (Beck and Feldman, 1989) if appropriate information about the affective content is provided on a trial-by-trial basis.

While such information might often be available during normal infant development, it will often be absent in adult life. Consider, for example, an individual observing the expressive emotional behavior of members of a different social group or culture. For this individual the emotion giving rise to the emotional display might be as obscure as the behavior itself. Thus, if cross peer-group and cross-cultural learning of facial emotional expressions can take place across the life span as suggested by the works by Elfenbein and others (Elfenbein and Ambady, 2002; Elfenbein, 2006), then some form of learning that does not rely on an external teaching signal might be effective in this learning.

The neural processes and mechanisms underlying unsupervised improvement of stimulus perception have extensively been studied in vision research. These studies provide consistent evidence that repeated exposure to simple visual stimuli such as tilted lines can lead to enhanced stimulus detection, discrimination or categorization in the complete absence of an external teaching signal (e.g., Poggio et al., 1992; Crist et al., 1997). A well-known example for this is the texture discrimination task in which participants learn to judge the orientation of a simple target stimulus (a number of aligned lines) among a number of distracter lines (Karni and Sagi, 1991). Interestingly, two recent studies that aimed to show that training with appropriate feedback can improve emotion recognition skills provided evidence that emotion recognition learning does not only take place if participants receive appropriate feedback, but might also occur in the complete absence of feedback (Blanch-Hartigan, 2012; Hurley, 2012).

Here, we provide further evidence that mere practice without an external teaching signal can improve facial emotion decoding skills in adults. In addition, we explore whether interpersonal traits can explain interindividual differences in learning. During two training sessions several days to weeks apart, participants saw video clips of dynamic facial expressions of five different women and were asked to decide which of four possible emotions (anger, disgust, fear, and sadness) was shown in each video. Although no information about the correctness of the participant's response or the woman's true affective state was provided, participants showed a significant increase of facial emotion recognition accuracy both within and between training sessions. This effect was modulated by stimulus duration and interpersonal traits. We discuss several similarities and differences between the unsupervised learning of facial decoding skills observed in the current study, unsupervised perceptual learning of simple visual and auditory stimuli described in previous studies and practice effects often observed in cognitive tasks.

### **MATERIALS AND METHODS**

#### **ETHICS STATEMENT**

Participants gave their informed consent before participation according to the guidelines of the American Psychological Association (http://www.apa.org/ethics) and the study was approved by the Ethics Committee of the Universität zu Lübeck. All data were analyzed anonymously.

#### **PARTICIPANTS**

Forty female participants were recruited from the Universität zu Lübeck, Germany. All participants were German-speaking Caucasians and none of the participants reported current or previous neurological or psychiatric illnesses. To investigate possible effects of the duration of the consolidation interval between the first and the second training session on learning, half of the participants had their second training sessions 2 days after the first training session (*2-days consolidation interval*), the other half 40–80 days (mean 59 days) after the first training session (*2-months consolidation interval*). Two participants were not available for the second training session; data of these participants were excluded from the analysis. The final sample consisted of 38 participants (20 with a 2-days consolidation interval, 18 with the 2-months consolidation interval) with an average age of 22.2 years (range 19–28 years).

#### **ASSESSMENT OF INTERPERSONAL TRAITS**

To examine possible relations between interpersonal traits and improvement of facial decoding skills participants completed the German 16-item version of the *Interpersonal Reactivity Index* (IRI, Davis et al., 2003), the *Saarbrücker Persönlichkeitsfragebogen* (SPF, http://psydok.sulb.uni-saarland.de/volltexte/2009/2363/) after the first training session. The IRI assesses the participant's interpersonal traits on four different subscales: spontaneous attempts to adopt the perspectives of other people (perspective-taking), tendency to identify with characters in movies, novels, plays, and other fictional situations (fantasy scale), feelings of warmth, compassion, and concern for others (empathic concern) and feelings

of anxiety and discomfort when observing another's negative experience (personal distress).

## **STIMULI**

In order to investigate subtle changes of ecologically valid facial emotion decoding skills we sought to use a stimulus set in which (i) senders expressed their true emotional state (rather than just showing a given prototypical facial expression) and (ii) senders communicated their true emotional state to a socially significant person (rather than just looking into a camera). Thus, we used video clips recorded in a previous fMRI (functional magnetic resonance imaging) study in which participants (senders) were asked to imagine and submerge themselves into a cued emotional situation and to facially express their feeling to their romantic partner who they believed was observing them online via a video camera (Anders et al., 2011). Analysis of the data from that study showed that observers were not only able to identify the sender's emotional state above chance at the behavioral level, but that showing and observing a given emotion evoked emotion-specific patterns of brain activity that were highly similar in the sender's and the observer's brain (Anders et al., 2011). For the current study, we selected videos clips of anger, disgust, fear, and sadness, each expressed by five different female Caucasian senders. These clips were selected from eight videos (two per emotion) recorded from each sender, whereby each video comprised four 20 s periods of a given emotion, separated by 20 s neutral periods. Only negative emotions were selected to avoid ceiling effects introduced by joy (which is usually very easily recognized among the negative emotions).

In order to permit the investigation of possible effects of stimulus duration on learning, videos were cut into clips of five different lengths (2 s, 4 s, 6 s, 8 s, and 10 s), each beginning with the onset of an emotional period. The final set of 100 different video clips contained one sample of each sender-by-emotion-by-length combination. These video clips were shuffled and divided into five subsets of twenty video clips, with the restriction that each subset contained one sample of each length-by-emotion combination and one sample of each sender-by-emotion combination. Subsets were presented in a counterbalanced order across participants, and a different order was used for the first and second training session of each participant. Analysis of hit rates for the five subsets during the first training session revealed no significant difference between stimulus subsets (one-way ANOVA with factor stimulus set, *F*[4,148] = 1.4, *p* = 0.23), indicating that facial expressions were evenly distributed across stimulus sets with regard to emotion recognition difficulty.

## **PROCEDURE**

Participants were tested in two training sessions, either 2 days or 40–80 days (mean 59 days) apart (see above). Video presentation during each training session was divided into five blocks, each containing one subset of video clips. Video clips were presented on a 15-- TFT laptop screen approximately 500 mm in front of the participant's face. Each video clip was preceded bya1s fixation cross on a dark background. Immediately after the video clip, a response screen appeared with four small boxes, each labeled

with one emotion ( "anger", "disgust", "fear", "sadness"), indicating the participant to convey her decision by button press. Four keys on the keyboard (*D*, *G*, *J*, *L*), each labeled with one emotion, were used as response buttons (whereby the order of the labeled boxes on the screen corresponded to the order of the response buttons on the keyboard). As soon as the participant had entered her response (maximal response interval of 5 s), the response screen was replaced with a dark screen for a fixed intertrial interval of 3 s. Importantly, the assignment of response buttons was counterbalanced across participants and a different assignment was used for the first and second training session for each participant. A response was defined as correct if the response button pressed by the participant corresponded to the emotion cued to the sender and as incorrect otherwise. A missing response was counted as an incorrect response. The presentation of a complete subset of video clips took a maximum of 20 × 15 s = 5 min, depending on the participant's response time. After each of these blocks, a short break was inserted (< 3 min), resulting in a maximum duration of 5 × 8 min = 40 min for each training session (**Figure 1**).

To familiarize participants with the experimental setting, each training session was preceded by three practice trials with video clips of a sender that was not used in the main experiment. Stimulus presentation and response logging were implemented with Presentation software (Neurobehavioral Systems Inc.,Albany, CA, USA).

#### **DATA ANALYSIS**

First, emotion recognition data were reduced by computing average hit rates and response times for each block and participant. Second, to obtain an estimate of initial performance and blockto-block increase (hit rates) or decrease (response times) of performance during each training session for each participant, a straight line with slope *bj* and constant *cj* was fitted through block averages, separately for each training session, using the least square criterion such that

$$y\_{ji} = \begin{array}{c} b\_{\text{j}}x\_{ji} + \ c\_{\text{j}} + \ e\_{ji}, \text{with } i = 1, 2, \dots, 5 \text{ and } j = 1, 2 \end{array}$$

where *yji* is the estimated hit rate in block *i* of training session *j*, *xji* is the mean-corrected number of block *i* of training session *j*, and *eji* is the error in block *i* of training session *j*.

In our main analysis, we then tested (i) whether learning slopes (*b*1,*b*2) were larger (hit rates) or smaller (response times) than zero (indicating learning within training sessions) and (ii) whether there was a significant increase (hit rates) or decrease (response times) of estimated performance from the first block of the first training session to the first block of the second training session (*y*2,1 − *y*1,1) (indicating consolidation across training sessions). To test for consolidation across training sessions, we used estimated hit rates/response times during the first block of each training session (*y*1,1 and *y*2,1) rather than average performance during each

session because they represent unbiased estimates of performance at the *beginning* of each training session.

For hit rates, we performed three additional analyses. First, to examine whether stimulus duration had an effect on learning, we tested for differences in initial performance (*y*1,1), learning slopes (*b*1, *b*2), and consolidation (*y*2,1 − *y*1,1) between short and long video clips. For this analysis, the parameters *b* and *y* were computed as described above, but this time separately for short videos (2–4 s) and long videos (8–10 s).

Second, to test for possible relations between interpersonal traits and (learning of) facial decoding skills, we correlated each participant's initial performance (*y*1,1) and average learning slopes (*b*<sup>1</sup> + *b*2) with her scores on the four IRI subscales (fantasy, empathic concern, perspective taking, personal distress).

Finally, we asked whether learning differed across emotions. Because of the limited number of trials per emotion, data were averaged across the five blocks of each training session for this analysis. Because hit rates for single categories can be affected by response biases, we computed average unbiased hit rates *huj*,*<sup>e</sup>* (Wagner, 1993), *huj*,*<sup>e</sup>* = (# of hits × # of hits)/(# of responses × # of stimuli) for each emotion and training session, where *huj*,*<sup>e</sup>* is the unbiased hit rate for emotion *e* in training session *j*. Differences between emotions were assessed by a four-by-two ANOVA with factors emotion and training session.

Student's *t*-test was used to test for differences unless otherwise indicated. In cases where we had a one-sided hypothesis, statistical tests were performed one-tailed, in all other cases two-tailed.

#### **RESULTS**

#### **MAIN ANALYSIS**

Behavioral data are summarized in **Table 1**. Participants showed a significant block-to-block increase of hit rates during both training sessions [*training session 1*, *T*(37) = 1.7, *p* = 0.046, *training session 2*, *T*(37) = 2.9, *p* = 0.033, **Figure 2A**], and there was no significant difference in learning slopes between training sessions [*training session 1 minus training session 2*, *T*(37) = –0.4, *p* > 0.50 (two-tailed)]. Learning slopes did not differ between the two groups [*2-days interval minus 2-months interval*, *training session 1*, *T*(36) = 0.1, *p* > 0.50 (two-tailed); *training session 2*, *T*(36) = 0.0, *p* > 0.50 (two-tailed)], and there was no interaction between consolidation interval and training session [*T*(36) = 0.1, *p* > 0.50 (two-tailed)]. This indicates that significant learning took place within training sessions, independent of the interval between training sessions.

Importantly, there was also a significant increase in hit rates from the first block of the first training session to the first block of the second training session [*T*(37) = 2.6, *p* = 0.007, **Figure 2B**]. Again there was no significant difference between groups [*2-days interval minus 2-months interval*, *T*(36) = –1.2, *p* > 0.10]. This indicates that increased emotion recognition accuracy consolidated across training sessions, independent of the consolidation interval between training sessions.

A similar pattern was observed for response times. There was a significant block-to-block decrease of response times during both training sessions [*training session 1*, *T*(37) = –3.7, *p* < 0.001; *training session 2*, *T*(37) = –2.0; *p* = 0.017], although this decrease was significantly stronger during the first than during the second training session [*training session 1 minus training session 2*, *T*(37) = –2.1, *p* = 0.021]. Learning slopes did not differ between groups in the first training session [*two-days interval minus longer interval*, *T*(36) = 0.3, *p* > 0.50 (two-tailed)], although in the second training session participants with a 2-days consolidation interval showed a stronger decrease of response times than participants in with a 2-months consolidation interval [*2-days interval*

**Table 1 | Mean hit rates, response times, and unbiased hit rates for all stimuli.**


Numbers in brackets indicate SEM (N = 38).

Participants showed a significant linear increase in emotion recognition accuracy within both training sessions (bold red lines in **(A)**. Emotion recognition accuracy also increased significantly from the first block of the first training session to the first block of the second training session, indicating consolidation of emotion recognition accuracy across training sessions (dashed red lines in **(A)**, bar charts in **(B)**. Filled circles in **(A)** represent block averages across all participants. Learning and consolidation

two training sessions (open circles and narrow dashed black line in **(A)**, only shown for the second training session) and participants who had a longer interval between the two training sessions (open circles and wide dashed line in **(A)**, only shown for the second training session). Numbers on the y-axis indicate percentage of correctly recognized facial expressions (note that chance level is 25 percent). Error bars indicate SEM. Asterisks indicate significant effects (p < 0.05).

*minus 2-months interval*, *T*(36) = –2.3, *p* = 0.027 (two-tailed)]; this interaction between consolidation interval and training session did not reach statistical significance [*T*(36) = –1.6, *p* > 0.10 (two-tailed)].

Response times decreased significantly from the first block of the first training session to the first block of the second training session [*T*(37) = –2.2, *p* = 0.017] and there was no significant difference between groups [*2-days interval minus 2-months interval*, *T*(36) = 0.4, *p* > 0.30]. Together, these data indicate that response times decreased both within and across training sessions, independent of the consolidation interval between training sessions.

#### **LONG vs. SHORT STIMULUS DURATION**

As expected, there was a trendfor long videos (8–10 s) to be initially recognized less accurately than short videos (2–4 s) [*long minus short videos*, *T*(37) = 1.3, *p* = 0.10]. This difference increased during the first training sessions and remained nearly stable during the second training session: while long videos showed a significant block-to-block increase of hit rates during the first and the second training session [*training session 1*, *T*(37) = 3.0, *p* = 0.002; *training session 2*, *T*(37) = 3.1, *p* = 0.002], short videos showed a significant block-to-block increase of hit rates only in the second training session [*training session 1*, *T*(37) = –0.3, *p* > 0.50; *training session 2*, *T*(37) = 2.2, *p* = 0.017, **Figure 3A**]. The difference between learning slopes for long and short videos in the first, but not in the second, training session was statistically significant [*long minus short videos*, *training session 1*, *T*(37) = 2.1, *p* = 0.021, *training session 2*, *T*(37) = 0.2, *p* > 0.50], with an interaction just below statistical significance [*stimulus duration x training session*, *T*(37) = 1.6, *p* = 0.059]. A similar trend was

observed when estimated hit rates during the first blocks of the first and second training sessions were compared [*long minus short videos*, *T*(37) = 1.6, *p* = 0.059, **Figure 3B**]. Together, these data show that initial performance was more accurate for long than for short videos, and that emotion recognition accuracy improved faster for long than for short videos.

#### **INTERPERSONAL TRAITS**

Participants' IRI scores deviated less than one SD from the norm of their German age reference group (Christoph Paulus, Normentabellen des SPF, Universität des Saarlandes, 2011) on all four subscales (perspective taking, mean = 3.5, SD = 0.6, norm 3.7; fantasy, mean = 3.5, SD = 0.8, norm 3.6; empathic concern, mean = 3.6, SD = 0.7, norm 3.6; personal distress, mean = 2.6, SD = 0.8, norm 2.8).

Overall, correlations between interpersonal traits and initial performance or learning were weak. However, we observed a significant positive correlation between empathic concern and initial hit rates (*y*1,1) for long videos [*r* = 0.27, *T*(36) = 1.7, *p* = 0.050 (uncorrected)] and between empathic concern and learning slopes for short videos [*r* = 0.36, *T*(36) = 2.3, *p* = 0.014 (uncorrected)]. Thus, empathic concern predicted both initial performance for long videos and improvement in emotion recognition accuracy for short videos.

#### **SINGLE EMOTIONS**

Average unbiased hit rates (Wagner, 1993) showed a significant increase from the first training session to the second training session for each and every emotion [anger, *T*(37) = 2.6, *p* = 0.007; disgust, *T*(37) = 2.9, *p* = 0.003; fear, *T*(37) = 3.2, *p* = 0.001; sadness, *T*(37) = 2.5, *p* = 0.009], and this improvement of

emotion recognition accuracy was similar across all emotions [four-by-two ANOVA with factors emotion and training session, emotion × training session interaction, *F*(3,11) = 0.8, *p* > 0.50, **Figure 4**].

# **DISCUSSION**

We observed a significant improvement of facial emotion decoding skills in healthy adults in a forced-choice emotion recognition paradigm without any external feedback. Participants' emotion recognition accuracy increased significantly both within and between two training sessions two days to several weeks apart. Although the study population and stimulus sample in the current study were limited to female Caucasian senders and observers, the current study extends previous evidence that facial emotion decoding skills can be significantly and sustainably improved by learning mechanisms that do not rely on external teaching signals.

The neural processes and mechanisms underlying unsupervised learning have extensively been studied in vision research, but improved performance after practice without feedback has also been observed in more cognitive tasks. Below, we discuss similarities and differences between the unsupervised improvement of facial decoding skills observed in the current study, unsupervised perceptual learning of simple stimuli, and other forms of unsupervised learning.

#### **COMPLEX VERSUS SIMPLE STIMULI**

Improvement of perceptual skills after repeated stimulus exposure without external feedback has been most intensively studied in the visual domain (e.g., Karni and Sagi, 1991; Poggio et al., 1992; Crist et al., 1997; more recently Özgen and Davies, 2002), but has also been observed in the auditory (e.g., Goudbeek et al., 2009) and olfactory (e.g., Li et al., 2006) modality. In these studies, participants were typically asked to detect, discriminate or categorize simple visual, auditory or olfactory stimuli. The decision boundary could either be explicitly given (such as "upright" for discrimination of tilted lines) or implicitly defined by the structure of the stimulus set (e.g., for a stimulus set consisting of lines whose tilt angles cluster around 45◦ and –45◦ tilt angle, respectively, "upright" can be derived as decision boundary from the structure of the stimulus set). In the first case, stimulus exposure results in enhanced perceptual discrimination along the relevant physical dimension (*perceptual discrimination learning*), particularly around the decision boundary. In the second case, stimulus exposure leads to learning of previously unknown categories (*perceptual category learning*), which in turn can result in perceptual discrimination learning. Both processes could in principle have contributed to the improvement of facial emotion decoding skills observed in the current study. However, the learning problem in the current study differed from that in studies using simple visual or auditory stimuli in at least two important factors: First, the physical feature space spanned by the facial emotional expressions used in the current study comprised far more dimensions than the space spanned by the simple stimuli used in previous studies. Second, participants in the current study had extensive prior (perceptual and semantic) knowledge about the categorical structure underlying the stimulus space.

#### **PHYSICAL FEATURE SPACE AND PRIOR KNOWLEDGE**

Recent studies show that humans easily learn new stimulus categories without feedback if these categories are defined by a single physical dimension (such as tilt angle), but are surprisingly inept in learning perceptual categories without an external teaching signal if learning requires the integration of two or more perceptual dimensions (such as tilt angle *and* length (*information integration learning*); Ashby et al., 1999; Goudbeek et al., 2009). This suggests that prior category knowledge might play an important role in facial emotion recognition learning.

In further support of this, a study on chimpanzee facial emotion recognition found that human observers perceived prototypical chimpanzee (*Pan troglodytes*) facial expressions categorically if they had previously learned (nonsense) verbal labels for each category (Fugate et al., 2010), while extensive perceptual experience with non-human primate facial expressions alone did not result in categorical perception (it should be noted though that participants in that study were also counted as having perceptual expertise if they had prior experience with a primate species other than chimpanzees). Another study on visual category learning found that semantic category knowledge can help to direct attention to relevant stimulus dimensions (Kim and Rehder, 2011).

In addition to semantic category knowledge, innate or learned perceptual knowledge might play an important role in facial emotion recognition learning. Specifically, innate or acquired neural algorithms that favor processing along biologically relevant higher-order perceptual dimensions (e.g., anger–disgust, anger–fear, anger–sadness, disgust–fear, disgust–sadness, fear– sadness) rather than physical dimensions (e.g., form and relative spacing of lips, brows, and eyes) could substantially reduce the dimensionality of the relevant perceptual space and thereby facilitate unsupervised learning. Empirical support for the assumption that such algorithms indeed develop early in life comes from the observation that infants, but not adults, readily learn multidimensional speech–sound categories by mere exposure (Maye et al., 2002; Goudbeek et al., 2009). In the current study, learning was facilitated both by empathic abilities and initial performance.

One important task for future studies will be to examine the effects of prior (learned or innate) semantic or perceptual knowledge on unsupervised learning of facial emotion decoding skills. This is particular interesting as observers will likely have less prior knowledge about the emotional behavior of senders who have a different social, cultural or ethnic background than the observer.

#### **SPECIFIC VERSUS GENERALIZED LEARNING**

Early studies on perceptual learning using simple physical stimuli in the visual domain found that training effects were remarkably specific to the particular stimuli used for training (e.g., an increased ability to discriminate distances between vertical lines did not generalize across line orientation or even visual location, Poggio et al., 1992; Crist et al., 1997). This has been taken as evidence that perceptual learning can take place very early in the visual processing stream (Gilbert, 2001). Thus the question arises whether the improvement of facial decoding skills observed in the current study is limited to the particular sample of individuals

**FIGURE 3 | Improvement of facial emotion recognition accuracy (long vs. short videos).** Participants showed a significant linear increase in emotion recognition accuracy within both training sessions for long videos but not for short videos; for short videos a significant increase in emotion recognition accuracy was observed only in the second training session (dark/bright bold red lines in **A**). In line with this, emotion recognition accuracy increased significantly from the first block of the first training

session to the first block of the second training session for long, but not for short videos (dark/bright dashed red lines in **A**, dark/bright bar charts in **B**). Dark/bright gray filled circles in **A** represent block averages for long/short videos across all participants. Numbers on the y-axis indicate percentage of correctly recognized facial expressions (note that chance level is 25%). Error bars indicate SEM. Asterisks indicate significant effects (p < 0.05).

seen during training or whether it generalizes beyond individual senders and maybe even sensory modalities.

Interestingly, there is accumulating evidence from neuroimaging studies that improved perceptual performance can be related to neural changes at different cortical levels, possibly depending on the particular perceptual task (Schoups et al., 2001; Schwartz et al., 2002; Furmanski, 2004; Little and Thulborn, 2005; Sigman et al., 2005; Li et al., 2006; Op de Beeck et al., 2006; Jiang et al., 2007; Law and Gold, 2008; van der Linden et al., 2008; Yotsumoto et al., 2008; Li et al., 2009; Wong et al., 2009; Zhang et al., 2010; Kahnt et al., 2011; Folstein et al., 2012; Myers and Swan, 2012), and that neural changes in higher cortical areas are associated with less specific learning effects (for review, see Sasaki et al., 2010). Extrapolating this evidence to the current study one might propose that if improved facial emotion decoding skills are related to neural plasticity in higher visual areas [e.g., occipito-temporal areas that support facial emotion recognition independent of facial identity (Fox et al., 2009)], then these learning effects should generalize beyond individual senders. Even more interestingly, one might ask whether learning effects can also generalize across sensory modalities. For example, it would be highly interesting to see whether perceivers who become more accurate at discriminating between facial emotional expression of different categories would at the same time become more accurate at discriminating vocal emotional expressions of the same categories (see Shams et al., 2011 for a related account). This would point towards increased discrimination accuracy at a neural level that receives input from different sensory modalities. Further combined behavioral and neuroimaging studies are needed to address these questions.

#### **ACTIVE DECISION MAKING AND STIMULUS SALIENCE**

Another factor that might have an important effect on unsupervised learning of facial decoding skills is explicit decision-making versus passive observation. One of the first reports of perceptual learning is the observation that passive exposure to visual stimuli can increase visual discrimination in rats (Gibson and Walk, 1956). In most perceptual learning studies in humans, participants were required to actively make a decision, but there are also a few studies that report perceptual learning after mere stimulus exposure in humans (e.g., Skrandies and Fahle, 1994). Although these findings suggest that explicit decision making is not essential for perceptual learning to occur, active decision making could still act as an enhancing factor. In a recent review, Sasaki et al. (2010) underline the role of signal strength in perceptual learning, and there is evidence that if participants are required to make a decision in the absence of external feedback an internal error signal is created that can serve as reinforcement signal and thereby facilitate learning (Daniel and Pollmann, 2012). Similarly, emotional salience might act as an internal signal amplifier and thereby facilitate learning in real life. Empirical evidence for this comes from a series of studies of physically abused children that showed that abused children recognize angry facial expressions more rapidly than controls (Pollak et al., 2009). Furthermore, compared to healthy controls, abused children's category boundaries for angry expressions were shifted towards fearful and sad facial expressions (Pollak and Kistler, 2002). Although these studies do not allow to completely separate effects of emotional salience from effects of frequent exposure they provide some evidence that emotional salience might play a role in learning of facial emotion recognition. Behavioral studies that closely model real life situations are needed to investigate the role explicit decision making, salience, and related factors in more detail.

#### **OTHER FORMS OF UNSUPERVISED LEARNING**

In a study on auditory perceptual learning, Hawkey et al. (2004) distinguished between *perceptual learning* (which refers to performance changes, "brought about through practice or experience, that improve an organism's ability to respond to its environment", p. 1055) and *procedural learning* (which refers to "improvement in performance on a task that results from learning the responds demands of the task", p. 1055). In the current study, *procedural learning* would refer to any improvement in performance that is not specific for facial emotional expressions (or, in fact, for any expressive emotional behavior, see below) but for features of the particular experimental set-up used in the current study, e.g., selecting and pressing the appropriate response button on a keyboard. Another possible factor that might confound results in studies that require participants to repeatedly classify stimuli into a number of predefined categories is that over the course of the experiment participants might acquire knowledge about a particular stimulus set (e.g., the frequency distribution of stimuli of a particular class) which could help them to develop response strategies that increase performance in the absence true stimulus-related learning (see e.g., Scherer and Scherer, 2011).

In the current study, we partly controlled for procedural learning by switching response buttons across training sessions. A more stringent control that should certainly be implemented in future studies would be to test the participants' facial decoding skills after training on a completely different experimental set-up (e.g., by showing the participants static images rather than videos and asking them to respond orally rather than via a computer).

Improved performance after practice without feedback has also frequently been observed in more cognitive tasks, for example when participants are tested on cognitive abilities (e.g., Hausknecht et al., 2002, 2007). A number of factors have been discussed to explain increased performance in such tasks, the most relevant for the current observation perhaps being reduced anxiety and increased motivation. Although these factors are probably more important in settings where participants know or have the impression that they being assessed for their personal abilities, future studies on facial decoding skills should include additional affective and motivational state questionnaires to control for these factors.

# **CONCLUSION**

In sum, the current study extends previous evidence that facial emotion decoding skills can be significantly and sustainably improved by learning mechanisms that do not rely on an external teaching signal. Such mechanisms might provide a basis for dynamic, life-long tuning of facial emotion decoding skills in humans. Importantly, the particular pattern of improvement of facial decoding skills observed in the current study, i.e., dependency of learning on stimulus duration and empathy-related personally traits, are difficult to explain by any confounding factors. Nevertheless, the results of the current study call for further systematic behavioral and neuroimaging studies that investigate (i) the degree to which unsupervised learning of facial emotion decoding skills relies on prior semantic and perceptual knowledge (ii) the degree to which improved emotion recognition generalizes across senders and sensory modalities, (iii) possible modulating effects of explicit decision making and stimulus salience and (iv) control more stringently for confounding effects. Such studies will, hopefully, (i) allow to develop efficient training programs to improve cross-cultural emotion decoding skills and (ii) draw the attention of the neuroscience community to the role of neural plasticity in human social behavior.

## **AUTHOR CONTRIBUTIONS**

Silke Anders, Jan O. Huelle, and Benjamin Sack conceived the experiment; Benjamin Sack and Katja Broer acquired data; Silke Anders, Irina Komlewa, and Benjamin Sack analyzed the data; Silke Anders wrote the manuscript; all authors edited the manuscript.

#### **ACKNOWLEDGMENTS**

This work was partly supported by the Bundesministerium für Bildung und Forschung (Federal Ministry of Education and Research, Grant 01GQ1105) and the Deutsche Forschungsgemeinschaft (German Research Association, Grant SFB654/2–2009 and Grant AN 755/2-1).

#### **REFERENCES**

Anders, S., Heinzle, J., Weiskopf, N., Ethofer, T., and Haynes, J.-D. (2011). Flow of affective information between communicating brains. *Neuroimage* 54, 439–446. doi: 10.1016/j.neuroimage.2010.07.004


and pervasive developmental disorder NOS. *J. Autism Dev. Disord.* 34, 649–668. doi: 10.1007/s10803-004-5286-y


Zhang, J., Meeson, A., Welchman, A. E., and Kourtzi, Z. (2010). Learning alters the tuning of functional magnetic resonance imaging patterns for visual forms. *J. Neurosci*. 30, 14127–14133. doi: 10.1523/JNEUROSCI.2204-10.2010

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 September 2013; paper pending published: 29 September 2013; accepted: 30 January 2014; published online: 27 February 2014.*

*Citation: Huelle JO, Sack B, Broer K, Komlewa I and Anders S (2014) Unsupervised learning of facial emotion decoding skills. Front. Hum. Neurosci. 8:77. doi: 10.3389/fnhum.2014.00077*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2014 Huelle, Sack, Broer, Komlewa and Anders. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited andthatthe original publication inthis journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# High levels of psychopathic traits alters moral choice but not moral judgment

# *Sébastien Tassy1,2\*, Christine Deruelle1, Julien Mancini 3,4, Samuel Leistedt 5,6 and Bruno Wicker <sup>1</sup>*

*<sup>1</sup> Institut de Neurosciences de la Timone, CNRS UMR 7289 and Aix-Marseille Université, Marseille, France*

*<sup>2</sup> Department of Psychiatry, Assistance Publique - Hôpitaux de Marseille, Sainte Marguerite University Hospital, Marseille, France*

*<sup>3</sup> School of Medicine, Aix Marseille Université, UMR912 SESSTIM (Economics and Social Health and Medical Information Processing), Marseille, France*

*<sup>4</sup> Public Health and Medical Information Department (SSPIM), Assistance Publique - Hôpitaux de Marseille, La Timone University Hospital, Marseille, France*

*<sup>5</sup> Forensic Psychiatry Unit, Department of Psychiatry, Erasme Academic Hospital, Université Libre de Bruxelles, Bruxelles, Belgium*

*<sup>6</sup> Medical Anthropology, SPF Justice, Brussels, Belgium*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Jack Van Honk, Utrecht University, Netherlands Kendall J. Eskine, Loyola University New Orleans, USA Paul Conway, The University of Western Ontario, Canada*

#### *\*Correspondence:*

*Sébastien Tassy, Pôle Universtaire de Psychiatrie, Sce Pr Azorin, Hôpital Sainte Marguerite, CHU de Marseille, 270 Bd de Sainte Marguerite, 13274 Marseille cedex 9, France e-mail: dr.sebastientassy@ gmail.com*

Psychopathy is a personality disorder frequently associated with immoral behaviors. Previous behavioral studies on the influence of psychopathy on moral decision have yielded contradictory results, possibly because they focused either on judgment (abstract evaluation) or on choice of hypothetical action, two processes that may rely on different mechanisms. In this study, we explored the influence of the level of psychopathic traits on judgment and choice of hypothetical action during moral dilemma evaluation. A population of 102 students completed a questionnaire with ten moral dilemmas and nine non-moral dilemmas. The task included questions targeting both judgment ("Is it acceptable to . . . in order to . . . ?") and choice of hypothetical action ("Would you . . . in order to . . . ?"). The level of psychopathic traits of each participant was evaluated with the Levenson Self-Report Psychopathy (LSRP) scale. Logistic regression fitted with the generalized estimating equations method analyses were conducted using responses to the judgment and choice tasks as the dependent variables and psychopathy scores as predictor. Results show that a high level of psychopathic traits, and more specifically those related to affective deficit, predicted a greater proportion of utilitarian responses for the choice but not for the judgment question. There was no first-order interaction between the level of psychopathic traits and other potential predictors. The relation between a high level of psychopathic traits and increased utilitarianism in choice of action but not in moral judgment may explain the contradictory results of previous studies where these two processes were not contrasted. It also gives further support to the hypothesis that choice of action endorsement and abstract judgment during moral dilemma evaluation are partially distinct neural and psychological processes. We propose that this distinction should be better taken into account in the evaluation of psychopathic behaviors.

#### **Keywords: moral, psychopathy, decision making, judgment, choice, emotion**

# **INTRODUCTION**

Psychopathy is a personality disorder characterized by emotional dysfunction, callousness, manipulativeness, reduced guilt, remorse and empathy, egocentricity, and antisocial behavior including impulsivity and poor behavioral control. Moreover, psychopaths frequently engage in morally inappropriate behavior, including taking advantage of others, lying, cheating, and abandoning relationships (Cleckley, 1941; Hare, 1999). Although psychopathy increases the probability of immoral behavior, experimental studies exploring its influence on decision making during moral dilemma evaluation have yielded contradictory results. Some studies report that psychopathy does not influence the decision (Blair, 1995; Glenn et al., 2009; Tassy et al., 2009; Cima et al., 2010) while others report that it is associated with higher probability of responses that favor the sacrifice of one individual for the greater welfare of many (i.e., utilitarian bias) (Glenn et al., 2010; Bartels and Pizarro, 2011; Koenigs et al., 2012). This latter result is consistent with the idea suggesting that emotion is a key element leading to non-utilitarian moral judgments and hence that individuals with low emotional responsiveness, such as those high in psychopathy, are expected to make more utilitarian judgment (Eslinger and Damasio, 1985; Greene et al., 2004; Koenigs et al., 2007). It is also consistent with the proposal that psychopathy tends to reduce the empathy for the victim, leading to greater concern for the mathematically rational ends than the emotionally aversive dimension (Greene et al., 2004, 2008; Crockett et al., 2010). Based on clinical experience, several authors have also reported that psychopaths are individuals with normal—or even higher—intelligence and a normal ability to judge, but whose actual behaviors remain particularly immoral (Cleckley, 1941; Hare, 1999; Glenn et al., 2010). This discrepancy between the intact ability to judge and an altered behavior suggests that these processes are at least partially independent, as proposed in the case of patients with ventromedial prefrontal brain lesions who exhibit inappropriate social behaviors but a preserved judgment ability (Eslinger and Damasio, 1985). Somewhat counter-intuitively, moral choice of action as reflected in actual behavior could thus be independent of moral judgment. Recent behavioral studies support such discrepancy by reporting experimental evidence for a divergence between judgment and choice of action during moral evaluation (Kurzban et al., 2012; Tassy et al., 2013). Moreover, moral choice of action and moral judgment could rely on partially distinct neural processes. A support to this hypothesis comes from results of a recent study showing that neural disruption before moral dilemma evaluation alters the judgment (objective evaluation) without modifying the subsequent choice of action (Tassy et al., 2012).

Psychopathy has traditionally been conceptualized in forensic samples. It describes a subset of individuals with Antisocial Personality Disorder who exhibit distinct personality features. However, recent taxometric studies suggest that psychopathy is a dimensional construct rather than a qualitatively distinct category of behavior and should be considered as an extreme variant of normal personality (Levenson et al., 1995; Hare and Neumann, 2005; Walton et al., 2008). The level of psychopathic traits would thus exist on a continuum in the general population and individual differences can be reliably assessed via self-report measures (Lilienfeld and Andrews, 1996; Edens et al., 2006). In the present study we seek to determine if a high level of psychopathic traits, in particular those related to primary psychopathy characterized by affective deficits (Karpman, 1946) and ventromedial prefrontal cortex (VMPFc) hypoactivation (Lotze et al., 2007), may be associated to an utilitarian preference in the specific case of moral *choice* during moral dilemma evaluation.

## **MATERIALS AND METHODS PSYCHOPATHY SCALE**

The Levenson Self-Report Psychopathy (LSRP) scale is the only global measure of psychopathic traits in the general population validated in French (Levenson et al., 1995; Chabrol and Leichsenring, 2006; Campbell et al., 2009). It consists of selfadministered questionnaires with 26 items in a 1–4 Likert-type agree/disagree rating scale. This scale is further subdivided into two subscales.

The LSRP1 is a subset of 16 items from the complete questionnaire constructed to determine the degree to which participants report interpersonal-affective characteristics that are associated with factor I of the Psychopathy Checklist-Revised (PCL-R) and that are the hallmark of primary Psychopathy based on Cleckley's and Hare's conceptualizations of the disorder (1, 2). The 10 remaining items compose the LSRP2, which measures the traits related to the social deviance associated with the factor II of the Hare Psychopathy Checklist.

#### **DILEMMAS**

We presented 10 moral and 9 non-moral dilemmas validated and used in a previous rTMS experiment (Tassy et al., 2012). Most of the dilemmas were directly inspired from the battery developed by Greene et al. (2001), translated and adapted to take into account cultural specificities. All were "Sacrificial" moral dilemmas, offering the opportunity to save many people from death (or serious physical consequence) at the cost of one person's life (or serious physical consequences) (Glenn et al., 2010) (e.g., Deadly fumes are rising up in the portion of a hospital where 53 patients are located. Thanks to the ventilation system you can divert the fumes to a room where one patient is sleeping. Is it acceptable to divert the fumes to the room where one patient is sleeping to prevent asphyxiation of 53 patients? Do you divert the fumes to the room where one patient is sleeping to prevent asphyxiation of 53 patients?). Each question was worded so that a positive response favored the survival of the highest number of people (utilitarian response). The non-moral dilemmas required decision making in simple contextual situations with no moral connotation whatsoever. "Appropriate" responses implied the maximization of beneficial overall consequences (e.g., You have to be at a very important meeting at 14 h. You can get there by car or subway. With the subway you will arrive just in time for your meeting. With the car you travel in a more enjoyable way but you will arrive late. Is it acceptable to use the subway instead of your car to be on time at the meeting? Do you use the subway instead of your car to be on time at the meeting?). Because psychopathy is associated to selfishness (Hare, 1999; Mokros et al., 2008), we tested if increasing the personal consequences of the decision would interact with psychopathy traits (Thomas et al., 2011). To do so, in five moral dilemma, the potential victim was supposed to be a family member of the subject (e.g., A train with no brakes is running toward 12 workers. You can divert the train by operating a switch, but it will then go on another track where your *cousin* is working.).

Two questions followed the text of each dilemma: one targeting judgment ("Is it acceptable to . . . in order to . . . ?") and one targeting choice of action1 ("Would you . . . in order to . . . ?"). To control for any order effect, two types of questionnaires were created: one where the judgment question preceded choice ("Order A" questionnaire), and one where the choice question preceded judgment ("Order B" questionnaire).

#### **POPULATION**

One hundred and two French university students participated to the study (91 females, 22.6 ± 2.3 years old). After receiving oral information about the nature of the experiment, participants completed two anonymous paper questionnaires, one with the moral and non-moral dilemmas and one with the Levenson items. Questionnaires were freely available at the end of a course. Students were free to bring them back later, anonymously, at a dedicated place. They were informed that by accepting to bring anonymously the questionnaires back, they gave their informed consent to participate. Half of the subjects completed a questionnaire "Order A" in which the judgment question preceded the choice question. The other half of each population completed a questionnaire "Order B" in which the choice question preceded the judgment question. The two groups had similar age, gender and LSRP psychopathy score (cf. **Table 1**).

<sup>1</sup>Our study is not immune to a usual critique that can be raised in this kind of setting involving moral dilemmas: what we consider here as an action is obviously what the participants think their action could be if they were to make the decision in real life. For ethical reasons, using questionnaires is as far as we can go given the life and death nature of the dilemmas.


**Table 1 | Participant characteristics based on the type of questionnaire (judgment/choice vs. choice/judgment).**

*Both groups are identical for studied variables (NS, no significant statistical difference; Mn, mean; SD, standard deviation; LSRP, Levenson Self-Report Psychopathy scale).*

The whole population showed an average total LSRP score of 49.19 (±7.98), an average LSRP1 score of 29.06 (±5.43), and an average LSRP2 score of 20.13 (±3.42). The two factors of psychopathy were significantly correlated (*r* = 0.613, *p* < 0.001).

It is important to emphasize that in the present study we examine psychopathy as a personality trait that varies within the normal population.

We studied the influence of the level of psychopathic traits on the probability of utilitarian responses to the judgment and choice questions.

#### **STATISTICS**

For non-moral dilemmas, "appropriate" and "inappropriate" responses were coded 1 and 0, respectively. For moral dilemmas, response to each question was coded 1 if it favored maximizing the good of more people at the expense of very few identified individuals ("utilitarian" response; e.g., sacrificing one person's life to save five), and 0 for the reverse situation.

All statistical analyses were performed using SPSS 17.0 (SPSS Inc., Chicago, IL). For univariate comparisons, we used Student's *t*-test for the means and a Chi-squared test for the percentages.

Multiple logistic regression analyses fitted with the generalized estimating equations method to account for the within-subject correlation (Koenigs et al., 2007) were conducted using each response to a dilemma as the dependent variable and entering total psychopathy score, sex, affective proximity of the victim, and the type of questionnaire as predictors. Sex was entered as control variable in all analyses because it has been reported to significantly influence moral decision making (Fumagalli et al., 2010). An additional regression was conducted in which both factors of psychopathy (LSRP1 and LSRP2) were simultaneously entered as predictors in place of the total psychopathy score. When the level of psychopathic traits had a significant effect, first order interaction between this level and other potential predictors were systematically studied.

## **RESULTS**

## **INFLUENCE OF STUDIED VARIABLES, ON NON-MORAL DILEMMA EVALUATION**

No variable significantly predicted responses to non-moral dilemmas either for the judgment or the choice question. Psychopathy traits thus do not influence response in easy non-moral decision making situations.

**Table 2 | Regression analyses demonstrating associations between utilitarian responses to moral dilemma and predictors.**


*Note: The beta values are from multiple regression models predicting utilitarian response to moral dilemma from total psychopathy score (per 10-point increase), sex, affective proximity of the victim, and the type of questionnaire. Numbers indicate standardized beta (*β*). \*Beta values for the LSRP1 and LSRP2 are from multiple regression models using sex, affective proximity, type of questionnaire, and both psychopathy factors (per 10-point increase) as predictors. Bold style is for the variables of interest.*

## **INFLUENCE OF STUDIED VARIABLES ON MORAL DILEMMA EVALUATION**

In the case of moral dilemmas, a high level of psychopathy traits, male sex, and affective distance with the victim significantly predicted utilitarian response to the choice question. For the judgment question, only the male sex significantly predicted utilitarian response, but neither the level of psychopathy traits nor the affective distance with the victim.

Order of the question (judgment before choice or vice-versa) did not influence the response for the judgment and for the choice (cf. **Table 2**).

## **INFLUENCE OF BOTH FACTORS OF PSYCHOPATHY ON DILEMMA EVALUATION**

Only higher LSRP1 (affective and interpersonal dimension) score significantly predicted a bias toward utilitarian response to the choice question (cf. **Table 2**).

## **INFLUENCE OF INTERACTION BETWEEN PSYCHOPATHY TRAITS AND OTHER SIGNIFICANT PREDICTORS**

We did not find any interaction with total psychopathy traits ( <sup>∗</sup>sex β = 0.18; *p* = 0.532; <sup>∗</sup>affective proximity β = −0.14; *p* = 0.370) or LRSP1 (∗sex β = 0.22; *p* = 0.742; <sup>∗</sup>affective proximity β = −0.16; *p* = 0.388) that significantly predicted utilitarian responses to the choice question. Increasing affective proximity of the victim (i.e., stronger personal consequences) did not interact with psychopathy traits' influence on moral choice responses.

# **DISCUSSION**

A high level of psychopathy traits does not predict utilitarian judgment during moral dilemmas evaluation, but it predicts utilitarian response in the case of choice. This effect is observed more specifically for the interpersonal-affective characteristics of psychopathy as measured by the LSRP1. This suggests that while the evaluative moral judgment of individuals with a high level of psychopathic traits (HP) remains identical to the judgment of individuals with normal/low level of psychopathic traits, they are more likely to make an effective choice decision that would inflict suffering or death to an individual for the greater welfare of more people. A high level of psychopathic traits thus influences the choice of hypothetical action endorsement embedded in a moral dilemma, but not moral judgment. Consistent with what is known in the case of patients with VMPFc lesions who exhibit emotional deficits and endorse utilitarian responses to moral dilemmas (Koenigs et al., 2007), we found that, in non-clinical individuals, scoring higher on a general measure of psychopathic traits and a measure of psychopathic traits targeting shallow affects and VMPFc hypofunctioning (Lotze et al., 2007) predicts utilitarian action endorsement preferences.

This may help explain the discrepant results of previous studies on moral dilemma evaluation in psychopathic individuals. Some studies indeed claimed that the ability to evaluate moral dilemmas is preserved in psychopathy (Cima et al., 2010), while others claimed that this ability is altered (Koenigs et al., 2012). Several factors are potentially responsible for this variability. It has been proposed that differences in the population from which the subjects were drawn may explain the discrepancy between these studies as most studies have sampled directly from a psychiatric population or a population of criminal offenders. Individuals diagnosed with psychopathy may be highly motivated to report in a manner that they believe will make them seem like an "average" individual because, among other reasons, they may be concerned that their responses may have consequences for their treatment or incarceration (Bartels and Pizarro, 2011). Another potential factor is the scale used to measure psychopathy (Koenigs et al., 2011). However, in the present study we used a scale that has been used in previous studies on moral decision making (Glenn et al., 2010) and shows a good concordance with the PCL-R (Brinkley et al., 2001) and SRPIII scales (Williams et al., 2003) used in other studies on moral evaluation in psychopathy (Cima et al., 2010; Bartels and Pizarro, 2011).

On the basis of our results, we propose that some discrepancies in the results of previous studies could originate from the use of moral dilemmas that differ on other dimensions, including the wording or structure of the evaluation question (O'Neill and Petrinovich, 1998; O'Hara et al., 2010), which may modify responses because they do not target the same evaluative psychological processes. Indeed, if one goes into details of the experimental procedures, responses were unaltered only when the question was worded as an evaluative judgment ("*Is it moral for you to . . .* ") (Cima et al., 2010). By contrast, psychopaths showed an altered response when the question was worded as a behavioral choice ("*Would you . . . in order to . . .*?") (Bartels and Pizarro, 2011; Koenigs et al., 2012). At the cerebral level right dorso-lateral prefrontal cortex (DLPFc) disruption alters moral judgment but not choice, which suggest that this structure is required to process allocentric integration of contextual information during moral judgment (Frith and de Vignemont, 2005; Tassy et al., 2012). On the contrary, a high level of psychopathy traits characterized by VMPFc dysfunction (Blair, 2007; Lotze et al., 2007; Koenigs, 2012), alters moral choice but not moral judgment. This suggests that moral choice of action mostly involves VMPFc. In the same line, Glenn et al. (2009) found that higher psychopathy scores are associated with reduced activity in VMPFc during moral choice of action ("*Would you . . . in order to*"). By contrast, as previously hypothesized (Tassy et al., 2009). The same study reported that higher psychopathy scores are also associated with increased activity in the rDLPFc during moral dilemma resolution. Individuals with a high level of psychopathic traits, because they lack some emotional reactions (Blair, 2007), may thus rely on allocentric judgment of the situation to make a choice decision.

As expected from results of previous psychological studies, affective proximity of the potential victims influences responses toward less utilitarianism (O'Neill and Petrinovich, 1998; Thomas et al., 2011). This is true for both judgment and choice, but seems to be stronger for the choice [β choice = 1.11, 95% confidence interval (0.87; 1.35) > β judgment = 0.01 (−0.14; 0.30)], which is coherent with recent studies (Kurzban et al., 2012; Tassy et al., 2013). A potential explanation could be that implication of a kin has strong personal consequences (Thomas et al., 2011), and personal consequences influence more action choice than abstract judgment (Sood and Forehand, 2005). We didn't find any interaction between level of psychopathy traits' influence and affective proximity. The response bias toward more utilitarianism of individuals with higher level of psychopathic traits is not influence by the affective proximity of the victim. It thus suggests that the response bias toward more utilitarianism of individuals with higher level of psychopathic traits is not influence by stronger personal consequences. This may appear opposite to selfishness theoretically associated with psychopathy (Campbell et al., 2009). A reason should be that psychopathy results from a strong default of emotional reaction to other distress. The distress of others emotionally arouses individuals with a low level of psychopathy traits. When the other is affectively proximate it potentiates this reaction. It results in decreased utilitarianism in responses (to reduce other's distress). Individuals with high level of psychopathic traits lack emotional reaction to the distress of others (Lotze et al., 2007), thus this reaction cannot be potentiated.

As emphasized by Sood and Forehand (Sood and Forehand, 2005), compared to judgment, choice elicits a greater degree of self-referent processing. Choice differs from judgment in agency because it implies projecting oneself into a situation of direct interaction using an egocentric frame of reference with potential self-relevant consequences. It thus generates strong emotional reactions and is also largely influenced by self-interest rational maximization. By contrast, judgment relies on an evaluation of the situation from a more allocentric perspective as defined by Frith and de Vignemont (Frith and de Vignemont, 2005). It is thus less influenced by self-interest maximization. Responses during moral dilemma evaluation indeed differ depending on whether evaluators are agents in the scenario rather than observers (Nadelhoffer and Feltz, 2008) and participants' intuition about their own or other's moral transgression activate distinct brain regions (Berthoz et al., 2006). It may explain the variation of the degree of utilitarianism in various dilemma responses where the dilemma induces an abstract judgment (reaction to moral violation by another person) or a choice of action (i.e., from a first person perspective) (Monin et al., 2007). This could also explain why some studies reported that people acknowledge moral norms and make appropriate moral judgment but fail to act in accordance with them, illustrating a capacity for "moral hypocrisy" (Batson et al., 1997).

Such a discrepancy was already noted in the field of developmental psychology (Blasi, 1980). In line with such a view, the results from the present study give further original experimental support to the notion that choice and judgment during moral dilemma evaluation are partially distinct psychological processes

## **REFERENCES**


self-report psychopathy scale measure the same constructs as Hare's psychopathy checklistrevised? *Pers. Individ. Dif.* 31, 1021–1038. doi: 10.1016/S0191- 8869(00)00178-1


(Tassy et al., 2012). In itself, this may be sufficient to raise a methodological concern in the study of moral cognition, and propose that choice of action and evaluative judgment should be conceptualized and tested separately. It would help understanding exactly which moral processes are affected in psychopathy, and could be used as a tool to test potential therapeutic effect (Krueger et al., 2012). More generally, it would improve in-depth understanding of moral motivation and cognition.

Overall, our empirical data nourish the debate on the role of our emotions and feelings about particular actions and outcomes as a source of our moral judgment and moral behavior (Moll et al., 2005, 2007; Shenhav and Greene, 2010) by revealing different patterns of moral evaluation in HP compared to Control normal population individuals, at least for a particular type of decision. Besides this, it brings a potential methodological answer to the variability observed in previous studies exploring moral evaluation in psychopathy. Overall, it highlights the fact that moral cognition consists of many levels of complexity that need to be better understood and taken into account.

# **FUNDING**

No current external funding sources for this study.

Psychopathic, not psychopath: taxometric evidence for the dimensional structure of psychopathy. *J. Abnorm. Psychol*. 115, 131–144. doi: 10.1037/0021-843X.115.1.131


D. (2004). The neural bases of cognitive conflict and control in moral judgment. *Neuron* 44, 389–400. doi: org/10.1016/j.neuron.2004.09.027


*Behav.* 19, 349–367. doi: 10.1016/ S1090-5138(98)00030-0


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 28 February 2013; accepted: 13 May 2013; published online: 04 June 2013.*

*Citation: Tassy S, Deruelle C, Mancini J, Leistedt S and Wicker B (2013) High levels of psychopathic traits alters moral choice but not moral judgment. Front. Hum. Neurosci. 7:229. doi: 10.3389/ fnhum.2013.00229*

*Copyright © 2013 Tassy, Deruelle, Mancini, Leistedt and Wicker. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Deficient fear conditioning in psychopathy as a function of interpersonal and affective disturbances

#### *Ralf Veit <sup>1</sup> \*†, Lilian Konicar 1,2,3 †, Jens G. Klinzing3, Beatrix Barth1, Özge Yilmaz 1,3 and Niels Birbaumer 1,4*

*<sup>1</sup> Institute of Medical Psychology and Behavioral Neurobiology, University of Tübingen, Tübingen, Germany*

*<sup>2</sup> International Centre for Ethics in the Sciences and Humanities, Research Training Group Bioethics, University of Tübingen, Tübingen, Germany*

*<sup>3</sup> Graduate School of Neural and Behavioural Sciences, International Max Planck Research School, University of Tübingen, Tübingen, Germany*

*<sup>4</sup> IRCCS Ospedale San Camillo, Venezia-Lido, Italy*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Arielle R. Baskin-Sommers, Harvard Medical School/McLean Hospital, USA Stefan Schulreich, Freie Universität Berlin, Germany*

#### *\*Correspondence:*

*Ralf Veit, Institute of Medical Psychology and Behavioral Neurobiology, University of Tübingen, Otfried-Müller-str. 47, 72076 Tübingen, Germany e-mail: ralf.veit@uni-tuebingen.de †These authors have contributed equally to this work.*

The diminished fear reactivity is one of the most valid physiological findings in psychopathy research. In a fear conditioning paradigm, with faces as conditioned stimulus (CS) and electric shock as unconditioned stimulus (US), we investigated a sample of 14 high psychopathic violent offenders. Event related potentials, skin conductance responses (SCR) as well as subjective ratings of the CSs were collected. This study assessed to which extent the different facets of the psychopathy construct contribute to the fear conditioning deficits observed in psychopaths. Participants with high scores on the affective facet subscale of the Psychopathy Checklist-Revised (PCL-R) showed weaker conditioned fear responses and lower N100 amplitudes compared to low scorers. In contrast, high scorers on the affective facet rated the CS+ (paired) more negatively than low scorers regarding the CS− (unpaired). Regarding the P300, high scores on the interpersonal facet were associated with increased amplitudes to the CS+ compared to the CS−, while the opposed pattern was found for the antisocial facet. Both, the initial and terminal contingent negative variation indicated a divergent pattern: participants with pronounced interpersonal deficits, showed increased cortical negativity to the CS+ compared to the CS−, whereas a reversed CS+/CS− differentiation was found in offenders scoring high on the antisocial facet. The present study revealed that deficient fear conditioning in psychopathy was most pronounced in offenders with high scores on the affective facet. Event related potentials suggest that participants with distinct interpersonal deficits showed increased information processing, whereas the antisocial facet was linked to decreased attention and interest to the CS+. These data indicate that an approach to the facets of psychopathy can help to resolve ambiguous findings in psychopathy research and enables a more precise and useful description of this disorder.

#### **Keywords: fear conditioning, electrophysiology, skin conductance, psychopathy, emotional-cognitive interaction**

# **INTRODUCTION**

Psychopathy is a personality disorder characterized by impaired affective processing and the persistent violation of the rights of others. The Psychopathy Checklist Revised (PCL-R) is a rating scale, widely used in research that gathers the multi faceted clinical construct of psychopathy. Factor analysis revealed four distinct, but intercorrelated facets (Hare, 2003). High scores on the first facet "interpersonal" describe a conning, manipulative, superficially charming person. The second facet captures "affective" deficits such as limited emotionality and lack of empathy. Both facets describe the core psychopathy feature already depicted by Cleckley (1941/1982) and covered by the original factor 1 proposed by Hare (1991). The other original factor 2 can also be subdivided into two facets: the "lifestyle" facet describes an impulsive, irresponsible and sensation seeking person, while the "antisocial behavior" facet encompasses mainly rule taking behavior and conflicts with the criminal law. There is an emerging consensus that factor 1 and 2 of the PCL-R represent two distinct components of psychopathy with different behavioral manifestations (Patrick and Bernat, 2009). It has been shown that both factors correlate in the opposed direction with measures of negative emotionality (Hicks and Patrick, 2006), supporting the multilayered construct of psychopathy. Moreover, different behavioral features of psychopathy may have a separate etiology (Patrick et al., 2007). For example, Molto et al. (2007) reported that response preservation deficits in psychopaths are linked to the antisocial feature but not to the interpersonal/affective characteristic of psychopathy. In the same vein, externalizing characteristics of psychopathy like antisocial behavior and substance abuse are said to be closely linked to factor 2 of the PCL-R (Patrick et al., 2005). Therefore, it is of special interest to reconsider well known psychopathy findings and their relation to subordinate PCL-R factors and related facets. Hereby it is possible to disentangle the etiological basis of certain psychopathy features.

One of the key findings in psychopathy is a reduced emotional responsiveness demonstrated by the diminished startle reflex potentiation during negative pictures (Lang et al., 1993; Patrick, 1994; Levenston et al., 2000; Vaidyanathan et al., 2009a). However, this finding depends strongly on the familiarity (Baskin-Sommers et al., 2013) and complexity (Sadeh and Verona, 2012) of the visual stimuli. Moreover, recent studies highlight the importance of attention modulation in instructed fear related tasks (Newman et al., 2010), which would undermine the thesis of a general impaired fear reactivity in psychopathy. Concerning the different components of psychopathy, the study of Patrick et al. (1993) revealed, that subjects scoring high on the affective component and low on antisocial behavior showed the strongest fearpotentiated startle deficit. Focusing on a continuous-compared to a discrete-analyses, a linear relationship between a reduced startle potentiation during aversive pictures and increasing scores in the interpersonal and affective facets of psychopathy was found (Sadeh and Verona, 2012).

Another important peripheral physiological finding in psychopathy research is the lack of electrodermal anticipatory fear responses to stimuli associated with punishment. This phenomenon was found in a countdown procedure (Hare et al., 1978; Ogloff and Wong, 1990), as well as in several delayed fear and aversive conditioning paradigms (Hare, 1965; Hare and Quinn, 1971; Flor et al., 2002; Veit et al., 2002; Birbaumer et al., 2005; Rothemund et al., 2012). Interestingly, psychopaths are able to recognize the relationship between unconditioned and conditioned stimuli on a cognitive level, but lack to decode the emotional importance of this association (Sommer et al., 2006).

Using functional neuroimaging, Birbaumer et al. (2005) showed that the physiological, as well as the cortical and subcortical fear deficits in psychopaths are related with factor 1 of the PCL-R and directly linked to amygdala, insular and orbitofrontal dysfunctions. The integrity of the amygdala is crucial for successful fear conditioning (Ledoux, 2000). Interestingly, structural imaging studies investigating the relationship between changes in brain morphometry and psychopathy subtypes could show that amygdala abnormalities (Yang et al., 2009) and gray matter reductions in the insula (De Oliveira-Souza et al., 2008) are strongly related to the affective and interpersonal facet of psychopathy.

While most of psychopathy research is based on peripheral psychophysiological measures, studies regarding event related potentials (ERP) are rarely published and show partly ambiguous results. Moreover only a few of those studies used the PCL-R as a diagnostic criterion for psychopathy (Kiehl et al., 1999, 2000, 2006; Flor et al., 2002; Howard and Mccullagh, 2007; Rothemund et al., 2012). Regarding the various ERP components, the N100 is sensitive to early attentional processes and larger amplitudes can be interpreted as a marker of selective attention (Luck et al., 2000). Jutai and Hare (1983) found in high -in comparison to low- psychopathic individuals decreased N100 amplitudes to task irrelevant stimuli indicating different allocation of attention depending on the focus of interest. Using an early event related potential (P140), Baskin-Sommers et al. (2012) similarly showed that psychopathic individuals are superior in focusing selectively on goal-directed information, whilst less attention is paid to peripheral, but fear and threat relevant information. Interestingly, the increase in N100 amplitude was associated with higher scores in the affective-interpersonal factor when emotional pictures were presented with high but not low complexity (Sadeh and Verona, 2012). Referring to fear conditioning paradigms the results are quite ambiguous: while Flor et al. (2002) found a larger N100 response to the paired stimuli (CS+) compared to the unpaired stimuli (CS−) during the acquisition phase in psychopaths, but not in healthy controls, Rothemund et al. (2012) reported smaller N100 components in psychopaths than controls, independent of the CS type. Another major electrical component is the P300 amplitude which is particular sensitive to changing salience of information (Sutton et al., 1965). Moreover this measure provides information of late attentional processes independent of early attentional allocations (Schupp et al., 2004). In general, the P300 potential is linked to orienting responses and reduced amplitude was found in many behavioral and medical disorders and diseases. For example, a reduction of the P300 amplitude was often reported as a diagnostic marker in schizophrenia (Galderisi et al., 2009). Reduced P300 amplitudes were shown in psychopaths compared to non-psychopathic participants in response to target stimuli in an oddball paradigm (Kiehl et al., 2000). In the context of Pavlovian aversive conditioning, the P300 was only occasionally addressed. During aversive conditioning, a CS+/− differentiation was specific for the psychopathic group during the early conditioning phase (Flor et al., 2002). Regarding the association with subordinate psychopathy factors, it has been shown the late positive potential that is similar to the P300 was negatively correlated with the affective-interpersonal dimension during the presentation of highly complex emotional pictures (Sadeh and Verona, 2012). This reinforces the assumption of a complex emotionattention interaction which moderates emotional processing.

Another ERP component, which has been studied in psychopathy research, is the contingent negative variation (CNV). This slow changing cortical potential can be interpreted as a correlate of selective attention and arousal, but it is also sensitive to expectancy and motivational aspects (Tecce, 1972). The CNV occurs as a response to a two-stimulus paradigm which consists of a warning signal indicating the condition followed by an imperative signal. The resulting potential shift can be decomposed into an initial (iCNV) and a terminal (tCNV) component. While the initial orienting response to the warning stimulus is associated with evaluation of the stimulus (Rockstroh et al., 1982), the terminal CNV arises just before the onset of the second stimulus and is modulated by its emotional salience and particularly pronounced in anticipation of intense aversive stimuli, i.e. an electric shock (Birbaumer et al., 1990). Initially, it was reported that psychopaths display a diminished CNV to the warning stimulus (McCalloum, 1973). Later findings indicated that psychopaths showed even enhanced CNV responses when the task is sufficiently interesting (Jutai and Hare, 1983). Forth and Hare (1989) used a forewarned reaction time task in which the participants could win or lose money. They found an enhanced magnitude of the iCNV, but not the tCNV in psychopaths, compared to non-psychopaths. Studies exclusively focusing on electrophysiological correlates of fear conditioning are generally rare (e.g., Lumsden et al., 1986; Regan and Howard, 1995). In healthy participants, Regan and Howard (1995) reported increased tCNV amplitudes to paired stimuli (CS+) compared to unpaired stimuli (CS−) in anticipation of phobia-related animal pictures. Regarding psychopathic individuals, Flor et al. (2002) used foul odor as the (US) and emotionally neutral faces as conditioned stimuli. They found superior information processing in psychopaths indicated by increased tCNV independent of CS type. Recently, Rothemund et al. (2012) used a fear conditioning paradigm with electric shocks as the US and reported larger left lateralized CS+/CS− differentiation in the iCNV during the late acquisition phase in psychopaths. The terminal component revealed a mixed pattern with lower magnitudes in psychopaths compared to non-psychopaths at frontal sites and the reversed pattern at parietal sites. Nevertheless it is important to mention that it is still unclear, if the inconsistent results in the ERP findings are due to the different tasks, stimuli or subjects. Despite the strong evidence of deficient fear responses that were found in several experimental investigations in general psychopathy, to date, only little work has been done to investigate the relationship between the subtypes of psychopathy in relation to their possible psychophysiological correlates. Both Flor et al. (2002) and Rothemund et al. (2012) did not explicitly differentiate between lower psychopathy factors and assumed a general factor that contributes to the fear conditioning deficit. The study of Birbaumer et al. (2005)investigated only subjects with high values on factor 1 and led to the conclusion that factor 1 is the mediating factor, influencing fear reactivity during implicit learning. Despite the growing evidence of affective-interpersonal characteristics that modulate fear related learning (Patrick, 1994), there is only limited information on the subcomponents of psychopathy specifically regarding the facets lifestyle and antisociality. Furthermore, disentangling the different facets of psychopathy would, on one hand be fruitful to extend various theoretical models (and related focuses as well as etiologies) and on the other hand may shed more light on prognosis (Marcus et al., 2006).

The present study aimed to expand the limited knowledge of Pavlovian fear conditioning on the subcomponents of psychopathy in highly criminal psychopaths. Therefore, we wanted to assess to which extent the different factors and facets of Hare's psychopathy construct contribute to the conditioned fear deficit on the peripheral, subjective and electrocortical level. In subjects with increasing PCL-R scores in the affective and interpersonal facet we expected a diminished anticipatory skin conductance response to the CS+ compared to CS− during the acquisition phases in the classical conditioning task. Furthermore we wanted to investigate differences in the subjective and electrodermal fear responses at the facet level in more detail.

Another focus of the study were the electrocortical correlates of fear conditioning in relation to the subordinate psychopathy dimensions. We hypothesized in an exploratory manner, that the higher the subjects score on the antisocial facet of the PCL-R, the better they are able to differentiate between CS+ and CS− regarding the measured P300 and CNV components. Contrary, we expected higher scores on the affective and/or interpersonal facet to be associated with a deficit in CS+/CS− differentiation. Concerning the N100 component we hypothesized, that subjects with high scores on the affective or interpersonal facet will show more pronounced amplitudes to the CS+, than to the CS−.

# **MATERIALS AND METHODS**

## **PARTICIPANTS**

Fourteen adult psychopathic males (mean age: 43.14 ± 11.52 years, all right handed) with a history of violent and/or sexual offences participated in the study. All of them were long-time detained in one of two cooperating German maximum security forensic psychiatric institutions. Exclusion criteria were an age below 18 or over 55 years, an IQ below 80 or health problems. The participants were informed about the aim of the study and gave written informed consent. They were paid 20C in agreement with the forensic institutions. All subjects were scored using Hare's Psychopathy Checklist Revised (PCL-R) (Hare, 2003), resulting in a mean score of 30.14 ± 2.77 (range = 25–34). Nine out of fourteen participants reached the cut-off score of 30, which is commonly used in American studies to classify psychopaths. However, all participants exceeded the psychopathy score of German and European norms (Cooke et al., 2004). Regarding the two main factors of the PCL-R, the mean scores were 11.79 ± 2.51 for factor 1 and 15.86 ± 1.56 for factor 2. According to the four facet model the mean scores were: Interpersonal facet: 5.50 ± 1.65; affective facet: 6.29 ± 1.35; lifestyle facet 7.00 ± 1.35; antisocial facet 15.86 ± 1.56. All participants met the full criteria for a diagnosis of antisocial personality disorders in line with the DSM-IV criteria (Apa, 2000). The study was approved by the Ethics Committee of the Medical Faculty of the University of Tübingen. The classical conditioning experiment presented here was conducted in the context of a comprehensive study on criminal psychopaths.

## **EXPERIMENTAL DESIGN**

A classical delayed fear conditioning paradigm was applied, using a modified version of the design of Birbaumer et al. (2005). Two neutral grey-scale male faces were used as conditioned stimuli (Schneider et al., 1994). With pseudo-random assignment between participants, one of the faces was paired with the unconditioned stimulus (CS+), while the other face was never followed by the unconditioned stimulus (CS−). The (US) consisted of an electric shock and was administered using a Digitimer DS5 Isolated Bipolar Constant Current Stimulator (Digitimer Ltd., United Kingdom). The two electrodes were attached over the distal and proximal phalanx of the right thumb. The intensity of the electric shock was individually adjusted in a pre-experiment, immediately before the actual experiment, at a level where the participants estimated the electric shock as unpleasant ("8" on a visual analog scale in which "0" indicates not at all unpleasant and "10", extremely unpleasant). The faces were presented for 5 s and the US was administered after 4 s of the picture presentation and lasted 500 ms.

A 50% partial reinforcement schedule was chosen, indicating that only 50% of the CS+ were followed by US. The conditioning procedure consisted of a habituation phase (6 CS+, 6 CS−, 4 US), an acquisition phase (32 CS+, 32 CS−) separated in an early and late acquisition block, and finally an extinction phase (16 CS+, 16 CS−). During habituation and extinction, the Inter-Trial-Interval (ITI) varied between 4.5 and 7.5 s, during acquisition between 7.5 and 12 s.

Subjective ratings of emotional valence and arousal were collected on a 9-point scale using the Self Assessment Manekin (Bradley and Lang, 1994) four times (after habituation, early acquisition, late acquisition and extinction). Based on the inherent implicit learning mechanism in classical conditioning it was not intended to inform the subjects about the CS-UCS contingency. The only information that was provided was that male faces will be shown and electric shocks will be applied from time to time. However, the contingency of US and CS was assessed after both acquisition phases and after the extinction phase using a visual analogue scale ("How likely is it that an electric stimulus will follow this face?" (ranging from "1," indicating no association between CS and US, to "9," indicating an absolute certainty that US follows CS).

## **APPARATUS AND PHYSIOLOGICAL RECORDINGS**

Physiological data were acquired using a Theraprax Neurofeedbacksystem (NeuroConn GmbH, Ilmenau, Germany). Electroencephalography (EEG), electrooculography (EOG), and skin conductance response (SCR) were recorded using a sampling rate of 128 Hz and 40 Hz high cut-off-filtering. EEG activity was measured using four recording electrodes, located at Fz, FCz, Cz, and Pz according to the 10–20 system. Left mastoid was used as the reference and right mastoid as ground. SCR activity was recorded from the intermediate phalanx of the index and ring finger of the non-dominant hand with strap electrodes.

## **DATA PROCESSING**

Skin conductance data were analyzed using the Matlab-based software Ledalab 3.4 (Benedek and Kaernbach, 2010). The recordings were decomposed into their tonic and phasic components which resulted in phasic activity timelines with zero baselines. SCR was then extracted for each single trial using the integrated phasic SCR between 1 and 4 s after picture onset and averaged over paired and unpaired trials in the different conditioning phases. This novel approach allows an unbiased estimation of sympathetic fear-related activity compared to the conventional baseline to peak computation using minimum amplitude criteria for classifying SCR.

The EEG data were processed using BrainVision Analyzer Professional 2.01 (BrainProducts GmbH, Gilching, Germany). A notch filter at 50 Hz was applied. Thereafter the signals were filtered using a 0.1 Hz highpass and a 15 Hz lowpass filter. Ocular artifacts were adjusted using an eye-blink artifact correction method (Gratton et al., 1983). The data was then segmented in epochs of 5.5 s duration (−0.5 to 5 s relative to the onset of the CS). On the segmented data, a time interval of 300 ms was used to detect and reject remaining artifacts exhibiting (a) gradient changes more than 15µV/ms, (b) voltage differences of more than 100µV, or c) signal amplitudes over ±80µV. Baseline correction was performed using a 200 ms interval before trial onset. ERPs were extracted using the following parameters: For N100, amplitude and latency of the highest voltage peak between 100 and 220 ms was registered, while 300–800 ms was used for P300. CNV areas under the curve were calculated for the time intervals 800–1500 ms (iCNV) as well as 3500–4000 ms (tCNV).

## **STATISTICAL ANALYSIS**

The self-report and SCR data were analyzed by means of repeatedmeasures analyses of variance (ANOVA) with phase (habituation, acquisition and extinction) and CS-type (CS+ vs. CS−) as within-group factors. Contingency ratings were analyzed using the acquisition phase (early vs. late) and the CS-type as within subject factor. For each EEG parameter separate repeated measures ANOVAs were performed using electrode position (FCz, Fz, Cz, and Pz), and CS-type as within-subject factors during habituation and extinction. During acquisition, the first and second half of the phase (early and late acquisition block) were selected as an additional within-subject factor. Greenhouse-Geisser correction was applied if the sphericity assumption was violated. In order to improve signal-to-noise ratio in subsequent analysis, EEG amplitudes during early and late acquisition phase were averaged over corresponding CS type. Differences between the CS types during early and late acquisition as well as both acquisition blocks combined were calculated. The differential values were correlated with the individual total PCL-R score as well as with its two factors and underlying facets (Pearson's bivariate correlation). Apart from simple correlations a partial correlation approach was conducted using one facet as independent variable and the respective other facets as control variables. In addition, a forward stepwise regression analysis using the four facets as independent variables and the differential (CS+ vs. CS−) subjective, peripheral and EEG parameters as dependent variables. For the inclusion we used F-probability of 0.05 and for the exclusion a probability of 0.10. Furthermore, a multiple regression analysis with all four facets at once was computed to determine the account of variance. Because of problems during SCR recordings, data were available only for 11 participants. One participant was excluded from the EEG analyses due to artifacts during data acquisition.

## **RESULTS**

## **CORRELATION BETWEEN SUBJECTIVE, ELECTRODERMAL AND EEG MEASURES**

The correlation analyses revealed a close relationship between the differential SCR responses and N100 and P300 amplitudes during early and late acquisition. Reduced anticipatory fear responses were accompanied with degraded early attention to the CS+ compared to the CS−, as reflected in the N100 waveforms. Late attentional processes, as reflected in the P300 potential showed the opposite effect. In a similar vein, a smaller conditioned SCR was correlated with a more negative rating of CS+ faces (see **Table 1**).

#### **SUBJECTIVE AND SKIN CONDUCTANCE MEASURES**

Repeated measurement ANOVAs with the factors phase (habituation, early acquisition, late acquisition and extinction) and CS type revealed no significant main effects or interactions for valence and arousal as well as for SCRs during fear conditioning. However, a significant difference in contingency ratings between CS types [*F(*1*,* <sup>13</sup>*)* = 45*.*66, *p <* 0*.*0001] was found indicating, that processing of the paired and unpaired stimuli was intact on a cognitive level. The correlation analysis revealed that during early acquisition high total PCL-R scores were negatively associated with differential SCR amplitudes (*r* = −0*.*757, *p* = 0*.*007). Regarding the different facets, the effect was most prominently induced by the affective facet of psychopathy (*r* = −0*.*663, *p* = 0*.*036; see **Figure 1**, right). Partial correlation analysis using the affective facet as dependent variable and the other facets as control variable confirmed the observed association, although it became not significant due to the reduced degrees of freedom relative to the number of control variables (*r* = −0*.*604, *p* = 0*.*112). A stepwise regression analysis revealed that the reduced differential SCR responses during the early acquisition phase were solely predicted by the affective facet [*R*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*439, [*F(*1*,* <sup>9</sup>*)* <sup>=</sup> <sup>7</sup>*.*04, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*026]. The other facets did not explain additional variance. Concerning the differential valence ratings, we found the opposite effect. Participants scoring high on the affective facet rated the paired stimulus more negative than the unpaired stimulus in the early acquisition phase (*r* = −0*.*589, *p* = 0*.*027; see **Figure 1 left**). This effect was confirmed using partial correlation analysis (*r* = 0*.*608, *p* = 0*.*050). During late acquisition phase, no significant correlations between subjective measures, SCR and PCL-R scores were found. Contingency ratings revealed that the CS-UCS pairing was


*\*p <sup>&</sup>lt; 0.05, \*\*p <sup>&</sup>lt; 0.01. The correlations refers to the difference between CS*<sup>+</sup> *and CS*− *of the respective variables.*

correctly identified during the early [*t(*13*)* = 4*.*63, *p <* 0*.*001] and late [*t(*13*)* = 7*.*43, *p <* 0*.*001] acquisition phase. All zero-order correlations and partial correlations for the subjective and SCR measures are depicted in **Table 2**.

## **CORTICAL MEASURES (EEG)** *N100*

Repeated measures ANOVA revealed no main effect or interaction for the N100 amplitude during the habituation and acquisition phase. During extinction, a significant electrode effect was observed [*F(*3*,* <sup>36</sup>*)* = 4*.*172, *p* = 0*.*034] with larger amplitudes at Fz compared to Pz. The correlation analyses showed a significant positive covariation between the scores in the affective facet of the PCL-R and the differential N100 amplitude (CS+ minus CS−) at frontal locations (FCz *r* = 0*.*588, *p* = 0*.*035; Fz *r* = 0*.*585, *p* = 0*.*036) during the early acquisition phase, indicating decreased attentional allocation to the CS+ in comparison to the CS−. Correlation analyses using both acquisition phases combined revealed a tendency of decreased N100 amplitudes during CS+ compared to CS− with the interpersonal facet (*r* = 0*.*524, *p* = 0*.*066). A partial correlation analyses confirmed this highly significant association (*r* = 0*.*903, *p <* 0*.*001). During extinction, a negative correlation was observed between total PCL-R score and CS+ /CS− differentiation (*r* = −0*.*565, *p* = 0*.*044).

The grand averages during acquisition of the EEG recordings at FCz and Fz over all participants and trials are shown in **Figure 2**. All zero-order correlations and partial correlations between psychopathy scores and EEG measures are depicted in **Table 3**.

## *P300*

No significant differences were found during the habituation phase. During acquisition significantly increased P300 amplitudes at parietal compared to frontal sites were observed [*F(*3*,* <sup>36</sup>*)* = 8*.*18, *p* = 0*.*001]. In addition, a significant electrode × block interaction was found [*F(*3*,* <sup>36</sup>*)* = 4*.*95, *p* = 0*.*023]. During the extinction phase, a significant electrode effect [*F(*3*,* <sup>36</sup>*)* = 4*.*351, *p* = 0*.*010] with larger P300 amplitudes at parietal compared to frontal sites was found. The correlation analysis revealed that high PCL-R scorers had an augmented P300 amplitude to the


**Table 2 | Correlation between PCL-R scores and peripheral and subjective measures.**

*Each cell consists of simple correlation coefficients. Partial correlations using all facets as control variable are listed as the second value in the cells. R<sup>2</sup> including all facets. \*p < 0.05, \*\*p < 0.01.*


*Each cell consists of simple correlation coefficients. Partial correlations using all facets as control variable are listed as the second value in the cells. R<sup>2</sup> including all facets. \*p < 0.05, \*\*p < 0.01.*

CS+ compared to the CS− in the early condition phase at all recording sites, but most prominent at the parietal electrode (FCz: *r* = 0*.*575, *p* = 0*.*040; Fz: *r* = 0*.*599, *p* = 0*.*031; Cz: *r* = 0*.*560, *p* = 0*.*047; Pz: *r* = 0*.*816, *p* = 0*.*001). The opposite pattern was found in the late acquisition phase with decreased P300 amplitude to CS+ in relation to the CS− at fronto-central positions (FCz: *r* = −0*.*703, *p* = 0*.*007; Fz: *r* = −0*.*674, *p* = 0*.*011, Cz: *r* = −0*.*702, *p* = 0*.*007). The correlation analysis, using the combined early and late conditioning phase, revealed a positive association between the P300 amplitude differentiation (CS+ minus CS−) and the antisocial facet of the PCL-R (Fz: *r* = 0*.*801, *p* = 0*.*001; Cz: *r* = 0*.*736, *p* = 0*.*004). The interpersonal facet of the PCL-R covaried negatively with the P300 differentiation (Cz: *r* = −0*.*671, *p* = 0*.*010; see **Figure 2**). Partial correlation analysis showed that the antisocial facet was the strongest predictor (*r* = 0*.*774, *p* = 0*.*009) for the CS+/CS− differentiation in the P300 amplitude. In a similar vain, the stepwise regression analysis favored a model with antisocial facet as the solely predictor [*R*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*541, *<sup>F</sup>(*1*,* <sup>12</sup>*)* <sup>=</sup> <sup>12</sup>*.*98, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*004]. During extinction no correlation reached statistical significance.

#### **INITIAL CONTINGENT NEGATIVE VARIATION (iCNV)**

During habituation and extinction, no significant effects were found. During acquisition, a significant electrode effect [*F(*3*,* <sup>36</sup>*)* = 6*.*545, *p* = 0*.*004] with increased negativity at

CS+ trials are depicted in red, while CS− trials are shown in blue. Time 0 indicates the onset of the face stimuli.

fronto-central electrode position was observed. However, neither CS type nor acquisition blocks yielded statistically significant effects. The correlation analyses revealed a positive relationship between the CS type related iCNV differentiation (CS+ minus CS−) with the antisocial facet (Fz: *r* = 0*.*733, *p* = 0*.*004; FCz: *r* = 0*.*716, *p* = 0*.*006; Cz: *r* = 0*.*623, *p* = 0*.*023) as well as with the original factor 2 of the PCL-R (Fz: *r* = 0*.*583), and in particular during the early acquisition phase (Fz: *r* = 0*.*747, *p* = 0*.*003). High scores on the antisocial facet were associated with smaller negative and even positive shifts of brain activity in response to CS+ compared to the CS−. Moreover, the interpersonal facet of the PCL-R covaried negatively with the CS+/CS− differentiation (Fz: *r* = −0*.*561, *p* = 0*.*036; Pz: *r* = 0*.*602, *p* = 0*.*029). Thus, participants with high interpersonal deficiencies had larger negative shifts in their brain activity in response to paired (CS+) relative to unpaired (CS−) stimuli (see **Figure 4**). A partial correlation analysis revealed that the antisocial facet showed the strongest association (*r* = 0*.*698), just as the stepwise regression analysis yielded [*R*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*537; *<sup>F</sup>(*1*,* <sup>12</sup>*)* <sup>=</sup> <sup>12</sup>*.*77, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*004]. During extinction, a negative correlation between scores in the antisocial facet and CS+/CS− iCNV differentiation was observed (*r* = −0*.*588, *p* = 0*.*035).

### **TERMINAL CONTINGENT VARIATION (tCNV)**

ANOVA revealed no significant effects during habituation or extinction. During the acquisition phase, there was only a tendency of an electrode effect [*F(*3*,* <sup>36</sup>*)* = 2*.*64, *p* = 0*.*064] with pronounced tCNV at Cz and the smallest effect at Pz, but no CS type or blocks effect. Resembling the iCNV results, the correlation analyses showed that the CS+/CS− differentiation in the terminal CNV was raised by the antisocial and interpersonal psychopathy facet. While high scores of the antisocial facet correlated positively with CS+/CS− tCNV differentation at fronto-central sites (Fz: *r* = 0*.*645, *p* = 0*.*017; FCz: *r* = 0*.*654, *p* = 0*.*015, Cz: *r* = 0*.*595, *p* = 0*.*032), the interpersonal facet revealed a negative correlation centrally (Cz: *r* = 0*.*610, *p* = 0*.*027). Thus, high antisocial scores were again associated with smaller negative or even positive shifts in brain activity in CS+ trials, while an opposite pattern was found in participants with pronounced interpersonal deficits. A step-wise regression analysis showed that a model including the

interpersonal facet only, explained most variance [*R*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*373, *F(*1*,* <sup>12</sup>*)* = 6*.*53, *p* = 0*.*027]. During extinction, a negative correlation between scores in the lifestyle facet and the tCNV difference between CS+ and CS− was observed (*r* = −0*.*566, *p* = 0*.*044).

## **DISCUSSION**

The present study aimed to explore associations of psychopathy and the different facets in Hare's psychopathy construct with subjective, peripheral-physiological and cortical measurements in a classical fear conditioning paradigm. On the group level, without accounting for the individual PCL-R scores, we did not find a conditioning effect in the subjective and peripheral measures in these highly psychopathic criminals. The results of our study corroborate previous findings, demonstrating a psychopathy related deficiency in developing a conditioned fear response to aversive or fearful stimuli (Flor et al., 2002; Veit et al., 2002; Birbaumer et al., 2005; Rothemund et al., 2012). In addition, the correlation analyses indicated that the deficiency in fear conditioning is linearly modulated by total PCL-R scores. Participants with extremely high psychopathy scores showed weaker or absent conditioned electrodermal fear responses compared to lower scorers. A closer inspection revealed that this effect was most prominently modulated by the affective facet. The opposed pattern was found using the differences in valence ratings between CS+ and CS− as a subjective measure of successful conditioning. Interestingly, the correlation between the differential valence ratings and SCR responses revealed that subjects with profound fear deficit rated the CS+ faces more negative than the CS−. We postulate that participants scoring high on the affective facet either tried to mimic normal emotional behavior by responding more negatively to the paired stimuli or that they are indeed perfectly able to evaluate the expected "cognitive" dimension of fear. This matches the observation that core psychopaths are masters of deception and/or are cognitively quite aware of the contingency between face stimuli and painful electric shock. Recently, Lopez et al. (2013) assessed self-reported psychopathy (PPI-R, The Psychopathic Inventory Revised), (Lilienfeld and Widows, 2005) in a student sample and showed that high scores in the "fearless dominance" subscale, but not in the "impulsive antisociality" subscale, were associated with deficient fear conditioning. This is similar to what we found in highly criminal psychopaths. The diminished emotional responsiveness is a key finding in psychopathy research and it has been shown that the reduced startle response during presentation of negative emotions is closely related to factor 1 of the PCL-R (Patrick et al., 1993; Patrick, 1994; Vaidyanathan et al., 2009b). Those findings fit perfectly to the elaborated theoretical framework of Patrick et al. (2007), which describes a bifactor conceptualization of psychopathic syndromes with different underlying etiological mechanisms. Nonetheless, it is not yet clear whether the affective or the interpersonal facet contributes to the fear deficit or to which extent they might contribute. Based on the anticipatory skin conductance responses, our data indicate that in a fear conditioning paradigm, the affective facet modulates the deficit in fear reactivity.

Regarding the EEG measures, we found decreased N100 amplitudes to the CS+ compared to the CS− during the early acquisition phase in participants with pronounced affective deficits. The same association was found in participants with distinct interpersonal deficits, when regarding the early and late acquisition phase combined. This result might reflect a different early attention status, presumably arising from the conditioning procedure in psychopaths scoring high on the superordinate factor 1 of the PCL-R. Our findings are in contrary to Flor et al. (2002) who revealed increased N100 amplitudes in the psychopaths to CS+ compared to CS− trials in the early conditioning phase, but in line with Rothemund et al. (2012) who showed overall lower N100 amplitudes in psychopaths compared to healthy controls. In context of the response modulation theory, Newman et al. (2010) proposed attentional abnormalities in psychopathy as an alternative fear deficit explanation. Furthermore, in a recent study deficient fear responses in highly psychopathic individuals were only found when the attention was shifted to irrelevant information at early stages prior to the onset of fearrelevant stimuli (Baskin-Sommers et al., 2011). We could show that the phase specific CS type differentiation in the N100 amplitude was directly related to the differential SCR responses in the early and late acquisition phase. Higher N100 amplitudes to the CS+ compared to the CS− were accompanied with enhanced electrodermal reactivity to the CS+, supporting the importance of early sensory processing of conditioned stimuli in fear conditioning (Miskovic and Keil, 2012). Although a direct comparison between our findings and the results of Newman et al. (2010) and Baskin-Sommers et al. (2011) is not possible, due to critical differences ranging from the experimental paradigm (implicit vs. explicit learning) up to the data collection (anticipatory SCR's and ERP's vs. fear potentiated startle response), our findings highlight the influence of early attentional processes in fear related learning that might be different in participants with affective/interpersonal deficits.

Regarding the P300 component, correlational analyses revealed larger CS type differentiation in the P300 amplitudes in high PCL-R scorers during the early acquisition phase. Considering the late acquisition phase, however a negative correlation was found with the PCL-R total scores as well as with factor 1 and the P300 responses. Flor et al. (2002) found a comparable positive CS+/− differentiation at frontal leads only in the psychopathic group during the early acquisition period, while Rothemund et al. (2012) showed a CS type differentiation in both psychopathic and non-psychopathic individuals. Our findings are consistent with the view that high psychopathic individuals exhibit intact attentional processes in particular during the early conditioning phase, when the meaning of the situation must be conceived. With respect to the psychopathy facets we showed that the interpersonal facet was negatively correlated with the CS+/− evoked P300 responses throughout the acquisition, while the antisocial facet modulated the P300 responses in the opposite way. It has to be emphasized that the P300 does not reveal a uniform pattern in psychopathy and most studies reporting reduced P300 amplitudes in psychopathy (Kiehl et al., 1999, 2000, 2006; Gao and Raine, 2009) used paradigms that influence P300 pattern selectively. In a similar vein, Patrick et al. (2006) reported a strong association between the reduction in P300 amplitude and externalizing dimensions such as antisocial behavior, pathological gambling, drug abuse and disinhibition in a visual oddball paradigm. Accordingly, this approach explains the decreased ERPs in psychopaths, scoring high on the interpersonal but not on the antisocial facet. However, it is important to mention that the P300 potential is not a measure for successful or unsuccessful Pavlovian fear conditioning, instead it measures selective attention and expectancy, modulating emotional learning (Verleger et al., 1994). Regardless of psychopathy scores, we demonstrated that enhanced P300 to the conditioned stimuli was associated with diminished electrodermal fear responses, suggesting cognitive top-down modulation in affective learning (Olofsson et al., 2008).

Concerning the conditioned CNV responses related to expectancy, orientation (iCNV) and preparation (tCNV), we found increased cortical negativity in the CS+ compared to the CS− condition in frontal and central sites for participants scoring high on the interpersonal facet. On the contrary, high scorers on the antisocial facet showed increased negativity in CS− compared to CS+ trials. Augmented iCNV was observed in some studies investigating psychopaths (Forth and Hare, 1989; Flor et al., 2002), while others reported no differences (Raine and Venables, 1987) or even reduced CNV responses (Walter et al., 1964; McCalloum, 1973). Rockstroh et al. (1982) emphasizing cognitive rather than emotional aspects as main sources of the CNV. Therefore, the enhanced iCNV during CS+ compared to CS− trials in participants with pronounced interpersonal deficits reflected heightened attention or interest in the conditioned face stimuli, while the antisocial facet showed an inverse effect. With respect to the tCNV, the increased tCNV differentiation in subjects scoring high on the interpersonal facet might be interpreted in the context of preparedness, cognitive appraisal and contingency evaluation and can be considered as a further proof of superior cognitive processing without affecting emotional fear learning.

In line with previous studies, we found deficient fear conditioning in incarcerated, highly psychopathic offenders as indicated by diminished SCR differentiation between types of conditioned stimuli. Stepwise regression analysis revealed that only the affective facet is responsible for the low fear responses. Socialization is to a high degree based on the learning of stimulusresponse (classical conditioning) and stimulus-reinforcement (instrumental conditioning) associations to adequately adapt behavior. A weak SCR differentiation during fear conditioning can therefore be an indication of maladaptive learning and failed socialization. Indicated by the event related potentials, we found a rather inferior early (N100) attentional but superior late (P300) attentional processing in subjects scoring high on the affective facet. Regarding the CNV responses, supposedly reflecting cognitive processing (Rockstroh et al., 1982), the interpersonal facet was associated with stronger CNV responses to the CS+ compared to the CS−, while the antisocial facet revealed the opposite effect. Such a "diaschisis" between the emotional and cognitive processing was often proposed in descriptive (Cleckley, 1941/1982), legal (Sommer et al., 2006) and artistic (Musil, 1930- 1943) accounts and explanations of psychopathic criminals. This discrepancy between emotional and cognitive processing is not only mirrored by the reported discrepancy between the cognitive and emotional awareness of the aversive stimulus, but also obvious in empathy tasks with psychopathic individuals. Psychopaths demonstrate a complete failure to experience emotional empathy (Vollm et al., 2006) and this dysfunction was particularly evident in psychopaths with affective/interpersonal deficits (Decety et al., 2013). The psychopathic lack of emotional empathy seems to relate to disrupted affective processing and production. On the other hand, psychopaths are able to complete theory of mind tasks that require perspective taking without much difficulty (Richell et al., 2003).

# **LIMITATIONS**

Firstly, the generalizability of our results is limited by the small sample size and the lack of an adequate control group. By including more subjects, it would have been possible to validate the existent findings of differences between non-psychopathic and psychopathic subjects. In our study we attempted to capture psychopathic patients scoring high on PCL-R. A broader spectrum of psychopathy scores would be desirable to verify the dimensional relations we observed in the different subtypes of psychopathy.

We computed the Post hoc power analysis (two-tailed) for the significant bivariate correlations using the sample size, the effect size and the alpha error probability (0.05). Regarding the strongest correlation between PCL-R total scores and SCR, we calculated a power of 0.86 (one-tailed 0.92), and for the correlation with the affective facet a power of 0.70 (one-tailed 0.81). The power for the significant correlations between the EEG measures and psychopathy scores ranged from 0.58 to 0.88 (one-tailed 0.71–0.94). For that, we can assume that the actual power of the presented findings is moderate to large. Other critical points are the limited number of trials during the conditioning procedure. One reason for the relatively low number of trials was the fact that we used the classical conditioning design as a part of a comprehensive investigation conducted in the forensic institutions. Rothemund et al. (2012) used a quite similar design including the same face stimuli as CS and an electric shock as US. During the acquisition procedure they used 48 CS+ and 48 CS− trials, while in our study we presented 32 CS+ and 32 CS− trials. Rothemund and colleagues found remarkable differences between psychopathic participants and non-psychopathic control subjects in subjective, peripheral and electrocortical measures. In addition, Flor et al. (2002) and Birbaumer et al. (2005) showed in their conditioning experiments with psychopaths and non-psychopathic controls the same faces as CS as we used and both found successful conditioning in the relevant outcome measures in the control group. Therefore we can conclude that the fear conditioning deficit in terms of a reduced anticipatory

#### **REFERENCES**


Baskin-Sommers, A., Curtin, J. J., Li, W., and Newman, J. P. (2012). Psychopathy-related differences in selective attention are captured by an early event-related potential. *Personal. Disord.* 3, 370–378. doi: 10.1037/a0025593

Baskin-Sommers, A. R., Curtin, J. J., and Newman, J. P. (2013). Emotion-modulated startle in psychopathy: clarifying familiar effects. *J. Abnorm. Psychol.* 122, 458–468. doi: 10.1037/a0030958

SCR in psychopathic individuals, in particular with high scores on the affective facet, is specific to the group and not to the task.

The selection of the electrode placement already proofed to be sufficient in another task (Strehl et al., 2006) and refers to our specific questions concerning the ERP measurements, mainly focusing in the CNV changes in response to the CS (with both, iCNV and tCNV showing their maximal amplitude on FCz and Cz). Finally, the generalizability of our results is also limited by the fact that till now no study exists, investigating fear conditioning in female, psychopathic inmates. A corresponding study would help to understand the underlying mechanism of this psychopathy-related physiological manifestation.

## **CONCLUSION**

In conclusion the diminished peripheral-emotional response (SCR) to aversive events in our study seems to be attended by inferior sensory and superior cognitive processing in more affective/interpersonal deficient psychopaths. Therefore, especially the aberrant cognitive-emotional interaction in psychopathy seems to be the key in fear conditioning as indicated by the subjective, peripheral-physiological and electrophysiological data. The present findings hint at segregated emotional and cognitive processing during implicit fear learning in psychopathic subtypes. This is of special importance and could have profound implications for the research on psychopathy including externalizing psychopathology. Without doubt, more studies are needed to shed light on the different cortical as well as peripher-physiological processes associated with the subtypes, facets and related shortcomings of psychopathy.

## **ACKNOWLEDGMENTS**

Prof. Birbaumer was supported by the German Research Foundation (DFG, the Koselleck Program) and Open Access Publishing Fund of Tübingen University. Lilian Konicar was supported by the International Centre for Ethics in the Sciences and Humanities (IZEW) at the University of Tübingen. We would like to thank the staff and the study subjects of the two cooperating German forensic institutions. We must also express special thanks to Dr. Hedwig Eisenbarth from the University of Regensburg, Department for Forensic Psychiatry and Psychotherapy at the District Hospital Regensburg who enabled and supported us to carry out the study in the cooperating forensic institutions.


in psychopathy: a functional magnetic resonance imaging study. *Arch. Gen. Psychiatry* 62, 799–805. doi: 10.1001/archpsyc.62.7.799


picture processing: an integrative review of ERP findings. *Biol. Psychol.* 77, 247–265. doi: 10.1016/j. biopsycho.2007.11.006


Baltimore, MD: Urban and Schwarzenberg.


457–466. doi: 10.1016/ S0079-6123(06)56025-X


doi: 10.1001/archgenpsychiatry. 2009.110

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 08 April 2013; accepted: 04 October 2013; published online: 25 October 2013.*

*Citation: Veit R, Konicar L, Klinzing JG, Barth B, Yilmaz Ö and Birbaumer N (2013) Deficient fear conditioning in psychopathy as a function of interpersonal and affective disturbances. Front. Hum. Neurosci. 7:706. doi: 10.3389/ fnhum.2013.00706*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Veit, Konicar, Klinzing, Barth, Yilmaz and Birbaumer. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# What can we learn about emotion by studying psychopathy?

# *Abigail A. Marsh\**

*Department of Psychology, Georgetown University, Washington, DC, USA*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Tobias Brosch, University of Geneva, Switzerland Hedwig Eisenbarth, University of Colorado at Boulder, USA Alice P. Jones, Goldsmiths, University of London, UK*

#### *\*Correspondence:*

*Abigail A. Marsh, Department of Psychology, Georgetown University, 37th and O Streets NW, WGR 302A, Washington, DC 20057, USA. e-mail: aam72@georgetown.edu*

Psychopathy is a developmental disorder associated with core affective traits, such as low empathy, guilt, and remorse, and with antisocial and aggressive behaviors. Recent neurocognitive and neuroimaging studies of psychopathy in both institutionalized and community samples have begun to illuminate the basis of this condition, in particular the ways that psychopathy affects the experience and recognition of fear. In this review, I will consider how understanding emotional processes in psychopathy can shed light on the three questions central to the study of emotion: (1) Are emotions discrete, qualitatively distinct phenomena, or quantitatively varying phenomena best described in terms of dimensions like arousal and valence? (2) What are the brain structures involved in generating specific emotions like fear, if any? And (3) how do our own experiences of emotion pertain to our perceptions of and responses to others' emotion? I conclude that insights afforded by the study of psychopathy may provide better understanding of not only fundamental social phenomena like empathy and aggression, but of the basic emotional processes that motivate these behaviors.

#### **Keywords: psychopathy, emotion, amygdala, empathy, fear**

Emotion is the major driver of all human and animal behavior, including social behavior—it is emotion that literally moves us to seek or escape positive and negative consequences (LeDoux, 2012). Many unanswered questions remain about the nature of human emotion and are the topic of vibrant ongoing debates: are different emotions qualitatively distinct, emerging from separable neurobiological processes, or can emotions be more accurately described dimensionally in terms of arousal and valence (Russell and Barrett, 1999; Barrett et al., 2007; Izard, 2007; Panksepp, 2007; LeDoux, 2012)? If distinct neurobiological events contribute to the generation of different emotions, which brain structures are most relevant to the emergence of these emotions (Panksepp, 2007; Vytal and Hamann, 2010; Lindquist et al., 2012)? And finally, how do emotions we experience pertain to our perceptions of and responses to emotions in others (Zahavi, 2008; Heberlein and Atkinson, 2009)?

Answering these questions about human emotion presents a variety of challenges. Unlike the study of some other human cognitive processes, the study of emotion benefits from the now widely accepted fact that humans and non-human animals share many emotional processes, enabling more, and more diverse study paradigms on emotion (Panksepp, 2007; Panksepp and Lahvis, 2011; LeDoux, 2012). A benefit of studying non-human animals is that they enable critical experimental manipulations to be performed, such as environmental manipulations that cause intense, ecologically valid experiences like fear, and manipulations of subcortical brain structures involved in emotion, such as permanent or temporary lesions or genetic manipulations. Gray and McNaughton argue that such techniques are essential for drawing causal inferences about some emotional processes (Gray and McNaughton, 2000). However, animals can provide little information relevant other critical features of emotion, such as information about subjective experiences. Research in humans can target subjective experience, but, conversely, many critical experimental manipulations of emotion are not feasible or ethical to perform in humans, such as intense, ecologically valid environmental manipulations or lesions to subcortical structures.

One means of circumventing this conundrum is to conduct research in individuals affected by pathologies that provide "natural experiments" in which emotional processes are altered, enabling identification of the downstream effects. One example is the use of case studies of individuals with lesions to specific brain regions as a result of disease, injury, or surgical intervention, such as the orbitofrontal cortex (Hornak et al., 2004), insula (Phillips et al., 1997), or amygdala (Feinstein et al., 2011). Such cases can yield rich and detailed evidence about the emotional processes subserved by the damaged region. The downside is that individuals in whom lesions are neuroanatomically specific enough to yield meaningful evidence are rare. Thus, few researchers have access to these patients, and the possibility persists that certain response patterns result from patient-specific idiosyncrasies unrelated to the lesion. In addition, most brain lesions occur in late adolescence or adulthood, precluding an understanding of the developmental consequences of lesions to structures like the amygdala, damage to which may result in distinct behavioral outcomes in adulthood relative to infancy (Amaral, 2003).

An alternative to lesion-based case studies is the study of populations of patients affected by psychopathologies known to affect specific neurocognitive systems. Psychopathy, a cluster of behavior tendencies and personality traits associated with callousness and antisocial behavior, is one such form of psychopathology (Hare, 1993; Blair et al., 2006; Skeem et al., 2011). Evidence is accumulating to suggest impairments in the systems and processes supporting fear responding in psychopathy, leaving other systems largely intact (Lilienfeld et al., 2012; Patrick et al., 2012; Rothemund et al., 2012). Psychopathy may therefore be a useful empirical tool for understanding the nature of fear and perhaps emotion more broadly.

In this review, I will consider how understanding psychopathy can shed light on the three questions outline above: (1) Are emotions discrete, qualitatively distinct phenomena or quantitatively varying phenomena best described in terms of dimensions like arousal and valence? (2) What are the brain structures involved in generating specific emotions like fear, if any? And (3) how do our own experiences of emotion pertain to our perceptions of and responses to others' emotion?

# **PSYCHOPATHY**

Psychopathy is a disorder that is generally viewed as the confluence of core personality characteristics plus antisocial behavioral tendencies, and which, in its extreme form, affects 1–2% of the general population and as many as 50% of violent offenders (Hare, 1993; Rutter, 2012). The core personality features associated with psychopathy are callous and unemotional personality traits, which include a lack of empathy or remorse, weak social bonds, an uncaring nature, and shallow emotional responding (Cooke et al., 2005; Frick and White, 2008; Viding and McCrory, 2012). The antisocial behavior tendencies that tend to accompany these traits include poor control of anger, impulsiveness, irresponsibility, and a parasitic orientation toward others (Frick and Ellis, 1999). These factors are generally positively related, such that higher levels of callous and unemotional personality traits predict increased antisocial behavior (Viding et al., 2007; Kahn et al., 2013). The presence of psychopathic traits are particularly strong predictors of aggression that serves an instrumental goal, such as bullying, sexual violence, or assault during the course of a robbery (Blair, 2001; Woodworth and Porter, 2002). Debates persist as to whether the features of psychopathy are best classified using various two-, three-, and four-factor models that have been proposed (Jones et al., 2006; Skeem et al., 2011), and whether criminal or aggressive behavior is an essential part of the psychopathy construct (Hare and Neumann, 2010; Skeem and Cooke, 2010), however, the basic features that compose the construct of psychopathy are generally agreed upon.

Psychopathy is not a clinical diagnosis in the Diagnostic and Statistical Manual (DSM-IV-TR), although features of psychopathy are incorporated into the Axis II diagnosis Antisocial Personality Disorder (Lynam and Vachon, 2012). Various suggestions for updating the DSM 5 to reflect current conceptualizations of psychopathy in adults and children have been proposed (Frick and Moffitt, 2010; Skodol et al., 2011). That said, emerging evidence suggests that psychopathy is not taxonomic in structure. As is the case for traits that comprise other forms of mental illness (Markon et al., 2011), psychopathic traits appear to be continuously distributed in the population and can be most reliably and validly assessed when treated as a continuous rather than a discrete measure (Edens et al., 2006; Guay et al., 2007; Kotov et al., 2011). This is important because it suggests that information about psychopathy can be drawn from both clinically diagnosed samples and community samples (Malterer et al., 2010).

Psychopathy affects both children and adults. Markers of psychopathy may emerge early in childhood (Glenn et al., 2007; Wang et al., 2012), are moderately reliable predictors of adult psychopathy (Lynam et al., 2008), and the core affective features of psychopathy appear to be highly heritable (Larsson et al., 2006). The heritability coefficient of the core callous and unemotional features has been estimated to be at least 0.43 (Larsson et al., 2006) and as high as 0.71 (Viding et al., 2005, 2008). An individual's risk for engaging in antisocial behavior during childhood or adulthood can be increased by any number of life history events, including trauma exposure, low socioeconomic status, or delinquent peer groups (Lynam et al., 2008), but these factors do not seem to precipitate the emergence of psychopathic traits in children (often termed callous-unemotional traits). In fact, callous-unemotional traits may paradoxically serve as a protective factor against parental maltreatment: among children with callous-unemotional traits, there is little correspondence between the quality of parenting that children receive and the severity of their antisocial behavior problems (Wootton et al., 1997). Instead it appears that life stressors that result in heightened stress responding represent a distinct etiological route toward antisocial behavior (Blair, 2001). Among children without high levels of callous-unemotional traits, parental maltreatment is associated with increased antisocial behavior (Wootton et al., 1997). In addition, antisocial behavior in the absence of callous-unemotional traits does not appear to be highly heritable, supporting the role of environmental stressors in leading to antisocial behavior in the absence of callous-unemotional traits (Viding et al., 2005, 2008).

# **PSYCHOPATHY AND FEAR RESPONDING**

From the earliest formal clinical descriptions of psychopathy, the construct has been linked to deficient fear responding. Most modern conceptualizations of psychopathy are based on the work of Cleckley (1988), whose compiled observations of institutionalized psychopaths are described in *The Mask of Sanity*. He distinguishes psychopaths from other psychiatric patients as typically free from delusions or irrational thinking, suicidality, or other self-harm tendencies, and, in particular, from anxiety or fear. The second criterion Cleckley specifies for identifying psychopathy is an, "Absence of nervousness or psychoneurotic manifestations," and he describes the prototypical psychopath as "incapable of anxiety" (p. 340) showing "immunity from . . . anxiety or worry" (p. 339), and being "free from . . . nervousness" (p. 339).

Although Cleckley's descriptions of psychopathy reflect a psychodynamic orientation, his observations are consistent with more recent experimental data assessing fear responding in psychopathy. A focus on fear responding emerged from the observation that psychopathic offenders are particularly likely to re-offend, suggesting that the threat of future punishments is not sufficiently motivating for them (Corrado et al., 2004; Hare, 2006). Fear is, in essence, *the state that accompanies the anticipation of an aversive outcome* (i.e., punishment) *and promotes avoidance and escape behaviors* (Stein and Jewett, 1986; Panksepp, 1998; LeDoux, 2000). Fear being the emotion that promotes avoidance of behaviors that result in punishment (LeDoux, 2003), it is ostensibly is the mechanism by which punishing criminal behavior serves to deter it. Early hypotheses proposed that dysfunctional fear responding renders psychopaths less likely to avoid engaging in criminal behaviors that result in punishments like imprisonment, and were supported by laboratory findings that psychopaths are less likely to modulate their behavior in response to anticipated punishments ranging from electrical shock to loss of points in a computer game (Lykken, 1957; Hare, 1966; Newman and Kosson, 1986; Blair et al., 2004).

Abundant psychophysiological research supports the notion that psychopaths' responses to the threat of an aversive outcome are muted. Psychopathy impairs anticipatory skin-conductance responses (Lykken, 1957; Aniskiewicz, 1979; Herpertz et al., 2001; Birbaumer et al., 2005; Rothemund et al., 2012), fear-potentiated startle responses (Patrick et al., 1993; Levenston et al., 2000; Herpertz et al., 2001; Rothemund et al., 2012), and contraction of the corrugator muscle underlying the brows (Herpertz et al., 2001; Rothemund et al., 2012) during threat anticipation. Psychopathy also impairs aversive classical conditioning (Flor et al., 2002) as well as other fear-relevant responses such as the recognition of fear from the face, body, and voice (Marsh and Blair, 2008; Dawel et al., 2012). These differences are particularly evident for psychopathic offenders characterized as "primary" psychopaths who exhibit the core callous and unemotional personality features of the disorder (Lykken, 1957; Aniskiewicz, 1979; Dawel et al., 2012). This is in contrast to "secondary" psychopaths, in whom antisocial behavior may primarily reflect social disadvantage or maltreatment and who may present with increased anxiety (Newman et al., 2005; Kimonis et al., 2012).

Finally, both anecdotal reports and empirical evidence indicate that subjective experiences of fear are reduced in psychopathy. In *Without Conscience* (Hare, 1993), Hare describes an interview with a psychopathic offender who seemingly fails to understand the fundamental nature of fear:

Another psychopath . . . said that he did not really understand what others meant by "fear." However, "When I rob a bank," he said, "I notice that the teller shakes or becomes tongue-tied. One barfed all over the money. She must have been pretty messed up inside, but I don't know why. If someone pointed a gun at me, I guess I'd be afraid but I wouldn't throw up." When asked to describe how he *would* feel in such a situation, his reply contained no references to body sensations. He said things such as, "I'd give you the money"; "I'd think of ways to get the drop on you"; "I'd try and get my ass out of there." When asked how he would *feel*, not what he would think or do, he seemed perplexed. Asked if he ever felt his heart pound or his stomach churn, he replied, "Of course! I'm not a robot. I really get pumped up when I have sex or when I get into a fight" (pp. 53–54).

Also supporting reduced subjective experience of fear in psychopathy are the results of a recent study in which adolescents with psychopathic traits and healthy controls underwent an autobiographical recall paradigm adapted from a task developed to measure subjective experiences of emotion across cultures (Scherer and Wallbott, 1994). In the task, participants described recent emotionally evocative events and their subjective responses during these events. This paradigm has the advantage of using a single measure to assess responses to five emotional states. Relative to controls, adolescents with psychopathic traits reported reduced symptoms of sympathetic nervous system activation, such as changes in breathing or muscle tension, during fear-evoking events, even though judges rated the psychopathic adolescents' descriptions of the fearevoking events as no less inherently frightening than the events reported by controls. In addition, psychopathic adolescents reported that in daily life they experience fear less often and less intensely than did controls (Marsh et al., 2011). Two adolescents with psychopathic traits in this study reported *never* having felt fear, an experience not reported by any of the healthy adolescents.

In keeping with this pattern, many contemporary assessments of psychopathy specifically index items related to reduced anxiety and fearfulness. These measures include the Triarchic Psychopathy Measure, e.g., "I'm afraid of far fewer things than most people" (Patrick, 2010); the Youth Psychopathy Inventory e.g., "What scares others usually doesn't scare me" (Andershed et al., 2002); and the Psychopathic Personality Inventory, e.g., "I can remain calm in situations that would make many other people panic" (Lilienfeld and Andrews, 1996). Researchers who use psychopathy measures that do not explicitly include anxiety and fear-relevant items often supplement the scale with anxiety measures or clinical assessments of anxiety disorders (Sutton et al., 2002; Finger et al., 2008; Malterer et al., 2008; Marsh et al., 2008; Kimonis et al., 2012; Koenigs et al., 2012).

In contrast to fear, other forms of emotional responding in psychopathy appear to be spared. The clearest example is anger, which appears intact and perhaps enhanced in psychopathy. Anger can be defined as the *high arousal state that follows frustration or perceived threat and, behaviorally, is closely linked to aggression against the source of frustration or threat* (Blair, 2012). Elevated anger responding is intrinsic to many descriptions of psychopathy. Both Cleckley and Hare's case studies include numerous descriptions of psychopaths whose misbehavior included frequent temper tantrums and rage-induced aggression. And contemporary measures of psychopathy universally feature items that index frequent, heightened, or undercontrolled displays and experiences of anger. These measures include the youth and adult variants of the Psychopathy Checklist, e.g., "Poor anger control" (Forth et al., 2003); the Antisocial Processes Screening Device, e.g., "Becomes angry when corrected or punished" (Frick and Hare, 2001); the Levenson Self-Report Psychopathy Scale, e.g., "When I get frustrated, I often 'let off steam' by blowing my top" (Levenson et al., 1995), and the Psychopathic Personality Inventory, e.g., "From time to time I really 'blow up' at other people" (Lilienfeld and Andrews, 1996). That these criteria are positively correlated with the overall construct reinforces the positive relationship between psychopathy and anger experiences.

In psychopathy, anger is most likely to result from goal frustration rather than perceived threat (Blair, 2012), although it should be noted that considerably less empirical research has assessed anger responding in psychopathy compared to fear. That said, three recent studies have found psychopathy to be associated with intact or heightened anger responding both physiologically and subjectively. Hicks and Patrick (2006) evaluated angry responding using a series of self-report scales and found elevated anger responding in psychopathy, with closer associations found between angry responding and the antisocial behavior subscale. In a similar vein, Blackburn and Lee-Evans (2011) found that psychopathic participants anticipated that they would respond with greater anger than non-psychopaths to a variety of angerinducing scenarios. Lobbestael et al. (2009) performed an anger induction task in individuals with Antisocial Personality Disorder (who varied in psychopathic traits), Borderline Personality Disorder and controls. The induction task entailed recalling a situation in which subjects had experienced a conflict with another person and had felt very angry, after which subjects spent several minutes recalling the details of the event. Results indicated that neither total psychopathy scores nor callous and unemotional personality trait scores among individuals with antisocial personality disorder were predictive of physiological changes during the anger induction task, suggesting an intact anger response. Other studies have found no group differences in responses linked to anger, such as the study assessing subjective experiences of emotion in psychopathic adolescents and controls (Marsh et al., 2011), and the results of two meta-analyses assessing the recognition of anger from the face, body, or voice (Marsh and Blair, 2008; Dawel et al., 2012).

A second emotional state that appears to be intact in psychopathy is positive excitement. This state can be distinguished from happiness, which is more closely associated with goal attainment, as the *state that accompanies the anticipation of an appetitive outcome* (i.e., reward) *and promotes acquisition or achievement of the reward*—a state that is in some ways a mirror image of fear and that has been alternately termed wanting, seeking, or interest (Berridge et al., 2009). The quotation from the incarcerated psychopath above is suggestive of the presence of positive excitement in psychopathy, and is consistent with clinical observations and empirical data that psychopaths are positively motivated by the prospect of reward, particularly near-term reward. Cleckley's criteria include several items that describe unrestrained goalseeking in the context of money, sexual gratification, and other rewards (Cleckley, 1988). And, as is true for anger, contemporary measures of psychopathy feature items related to the experience of wanting, seeking, and excitement, including the Psychopathy Checklist, e.g., "Stimulation seeking" (Forth et al., 2003); the Youth Psychopathy Inventory, e.g., "If I get the chance to do something fun, I do it no matter what I had been doing before" (Andershed et al., 2002); the Levenson Self-Report Psychopathy Scale, e.g., "My main purpose in life is getting as many goodies as I can" (Levenson et al., 1995), and the Psychopathic Personality Inventory, e.g., "If I were a firefighter, I think I might actually enjoy the excitement of trying to rescue someone from the top floor of a burning building" (Lilienfeld and Andrews, 1996). Empirical behavioral data also exist to suggest that the motivational salience of rewarding stimuli is similar to that of comparison samples (Blair et al., 2004) or perhaps even increased (Scerbo et al., 1990; Bjork et al., 2012). Because positive excitement is not always included on lists of basic emotion it is subject to less focused research than emotions like anger and fear. However, what evidence exists suggests that this state is intact or heightened in psychopathy.

There is very little evidence available that describes other types of emotional reactions in psychopathy, although what evidence exists suggests that disgust responding remains intact, and there is little evidence for consistent impairments in happiness or surprise (Marsh and Blair, 2008; Marsh et al., 2011; Dawel et al., 2012). One emotion for which the present literature is genuinely ambiguous is sadness, with meta-analytic findings generally showing some deficits in recognizing sadness expressions in psychopathy, albeit less consistently and with generally smaller effect sizes than for fear. Very little literature explores sadness responses in psychopathy in other contexts, and results from these studies are equivocal (e.g., Blair et al., 1995; Brook and Kosson, 2013) In general, the neurobiological basis of sadness is not as well understood as that of fear, and further development of the neurocognitive basis of sadness may be required to develop targeted tasks assessing it in psychopaths.

It should be noted that among Cleckley's original criteria is "General poverty in major affective reactions" which is reflected in items measuring shallow affect in contemporary measures such as the PCL variants and APSD (Hare, 1991; Frick and Hare, 2001). However, Cleckley's emphasis is primarily the quality of the anger, excitement, etc. that psychopaths experience—how long-lasting these states are, how consistent, and how "mature" their expression. Thus, whereas psychopaths may display outward signs of rage and become "vexed," "peevish," or "resentful," Cleckley proposes that they do not experience "mature, wholehearted anger" (Cleckley, 1988, p. 348). The lability or consistency of affective reactions in psychopathy may be an important feature of the disorder. However, it remains the case that among basic emotions, only in the case of fear does strong, consistent empirical evidence support the existence of deficits in psychopathy.

# **ARE EMOTIONS DISCRETE NATURAL KINDS OR CONSTRUCTED USING DIMENSIONS OF CORE AFFECT?**

These patterns of observed emotional responding in psychopathy may help to explicate a central ongoing question about emotion, namely: can emotions be better described as *qualitatively* distinct, for example, as discrete "basic emotions" or "natural kinds" (Ekman et al., 1983; Izard, 1992; Panksepp, 2005) or as *quantitatively* distinct, for example, as points along a circumplex defined by dimensions like arousal and valence (Russell and Barrett, 1999; Barrett and Wager, 2006)? Recent years have seen a protracted debate in the literature about how to most accurately capture the nature of emotion (Barrett et al., 2007; Izard, 2007; Panksepp, 2007; Tracy and Randles, 2011), with proposed models of emotion including not only basic emotion and dimensional models, but also those that focus upon goal-relevant appraisals of emotional stimuli (Moors et al., 2013), emotions as coping responses (Roseman, 2013), and emotions as survival circuits (LeDoux, 2012). An extended conversation about the strengths and weaknesses of these various views will not be reviewed in full here, rather, the focus will be on the basic consideration of whether different emotions (e.g., fear, anger) are best viewed as qualitatively or quantitatively distinct.

Models that posit emotions to be qualitatively distinct, such as "basic emotion" models, holds that a limited number of emotions like fear, anger, and positive excitement emerge from dissociable neurophysiological processes (Ekman et al., 1983; Izard, 1992; Panksepp, 2005; Lench et al., 2011). These neurophysiological processes are generally linked to activity in the evolutionarily ancient subcortical structures of the midbrain, striatum, and limbic system most commonly linked to emotion (Panksepp, 2005; Vytal and Hamann, 2010). So, for example, the generation of positive excitement is linked to activation in a striatal circuit centered on dopaminergic neurons in the nucleus accumbens (Ikemoto and Panksepp, 1999), whereas the generation of fear is associated with activity in a circuit involving the periaqueductal gray, anterior and medial hypothalamus, and amygdala (LeDoux, 2000). In this view, finer gradations of experience result when basic emotions are modulated or elaborated by higher-level cognitive processes controlled by the cerebral cortex, but the emergence of qualitatively distinct emotions is not dependent on these cortically-controlled processes (Panksepp, 2005).

Models that posit emotions to be quantitatively distinct hold that emotions like fear, anger, and happiness are best described as points on one or more core dimensions. Core dimensions typically proposed to distinguish among emotions are physiological arousal or activation (low—high) and valence (bad—good) (Bradley et al., 2001). [Some have proposed a withdrawal approach dimension as a substitute or supplement to the valence axis (Wager et al., 2003; Christie and Friedman, 2004; van Honk and Schutter, 2006)]. Arranged orthogonally, these dimensions form a circumplex upon which emotions can be plotted and quantitatively compared (Barrett and Russell, 1999; Russell and Barrett, 1999; Colibazzi et al., 2010). Positive excitement is plotted as high in arousal and positive in valence, and sadness is low in arousal and negative in valence. Fear is typically plotted as high arousal and strongly negative, as is anger (Russell and Barrett, 1999). Further distinctions among emotions are thought to reflect differences in cognitive construals of the events surrounding the basic changes in arousal and valence. Thus, whether an individual experiences anger or fear (which are similar in terms of arousal or valence) may be shaped by interpretations of neurophysiological changes in valence and arousal in light of the eliciting stimulus and the individual's idiosyncratic stores of semantic knowledge, memories, and behavioral responses that shape the subjectively experienced state (Russell, 2003). Under this view, distinctions among experienced emotional states are highly dependent on these cognitively complex processes, which are subserved by a distributed network of regions of the cerebral cortex (Lindquist et al., 2012).

These models generate distinct predictions to the question of whether a disorder or lesion could result in a single emotion being disabled without affecting the experience of other emotions. The discrete emotions view would argue that a disorder or lesion that resulted in dysfunction in the specific structures subserving a particular emotion could affect the experience of one emotion while leaving others intact. In contrast, the dimensional view would require either that other emotions that are dimensionally similar to the affected emotion also be affected, or that deficits in a particular emotion would reflect dysfunction in cortically-driven higher-level cognitive processes.

The case of psychopathy lends clear support to notion that fear is qualitatively distinct from other emotions. In psychopathy, the bulk of the clinical and empirical evidence points toward the conclusion that fear responding is uniquely disabled, with other high-arousal (positive excitement, anger) and negatively valenced (anger, disgust) emotions remaining intact. The dimensional view cannot easily explain why in psychopaths the high arousal, negatively valenced state of anger is easily (perhaps too easily) generated, whereas the high arousal, negatively valenced state of fear is not. The problem cannot lie in a failure to fully engage neurocognitive systems underlying either the arousal or valence dimension, because psychopaths experience other higharousal emotions (positive excitement) as well as other negatively valenced emotions (disgust). It also cannot result from some difficulty arising at the interaction of these axes, because anger and fear are highly similar in terms of both dimensions. Models that substitute a withdrawal—approach axis for a negative—positive axis are no more successful; the two most strongly withdrawallinked emotions are disgust and fear, and there is no evidence for disgust-based impairments in psychopathy.

Can cognitive construals of emotion explain the patterns observed in psychopathy? Perhaps, one could argue, psychopaths under threat are less likely to construe their negative, high-arousal state as fear and more likely to construe it as anger compared to non-psychopaths. So, for example, the psychopath whose interview is transcribed above might interpret a pounding heart and churning stomach as the angry response that accompanies a tendency to respond aggressively. Another person might interpret the same body symptoms as the fear that accompanies a tendency to escape or submit. Theoretically, this explanation could explain both the deficits in fear and a concomitant increase in anger in this population. One could argue that, particularly for studies that focus on subjective reports of emotion, group differences in construal underlie the tendency of psychopaths to underreport experiencing fear and overreport experiencing anger.

This argument suffers two shortcomings. First, it is inconsistent with psychophysiological findings of overall reduced arousal during threat anticipation in psychopathy. As described above, there are two major categories of anger elicitors: perceived threat and goal frustration (Blair, 2012). The construal argument would require that psychopaths experience arousal in response to threat, but interpret this arousal as anger rather than fear. But the evidence is clear that psychopaths (particularly primary psychopaths) are no more likely than average to experience physiological arousal under conditions of threat (Blackburn and Lee-Evans, 2011)—and in fact, as described previously, show reduced physiological responses, including reduced skin conductance, potentiated startle, and corrugator muscle activity. This suggests that threat anticipation results in neither fear nor anger in this population. Psychopaths are, however, more likely than average to experience anger is in response to frustration (Blair, 2012). Thus, rather than being chronically likely to construe any high arousal state as anger, psychopaths appear more likely to experience anger primarily in response to frustrated attempts to achieve a reward. That both frustration-based anger and positive excitement (the state that reflects the anticipation of reward) are normal or elevated in psychopathy is consistent with the notion that in psychopaths the systems that govern anticipation of reward are functional and perhaps even overactive while the systems that govern threat anticipation are dysfunctional. A further concern is that the construal explanation of emotion leaves unclear *why* psychopathy might engender such a dramatic shift in emotional experience. Such a phenomenon is particularly difficult to explain in light of the high heritability coefficient found for psychopathy. Cognitive construals of emotional states are thought to reflect the individual's autobiographical memories and semantic knowledge of emotion prototypes, phenomena that are necessarily a result of learning, rendering it unlikely that the tendency to construe one's emotional response to an event as fear versus anger would itself be heritable.

The pattern of reduced fear responding to anticipated threat observed in psychopathy, then, is more consistent with the view that states like anger and fear reflect biologically coherent and qualitatively distinct responses to particular eliciting stimuli. Dimensions like valence and arousal are useful means of quantitatively describing differences among subjective feeling states like fear, anger, and positive excitement, but may not accurately reflect the neurobiological origins of those states.

# **WHAT ARE THE BRAIN STRUCTURES INVOLVED IN GENERATING SPECIFIC EMOTIONS LIKE FEAR?**

If psychopathy is associated with specific deficits in fear responding, this not only supports the idea that emotions are qualitatively distinct, it supports the corollary that specific neurophysiological processes that support the fear response are also affected. A key feature of models of discrete emotions is that distinct emotions have dissociable neurophysiological correlates (Vytal and Hamann, 2010). Ekman (1999) has argued:

The distinctive features of each emotion, including the changes not just in expression but in memories, imagery, expectations, and other cognitive activities, could not occur without central nervous system organization and direction. There must be unique physiological [CNS] patterns for each emotion (p. 50).

Limited evidence exists to suggest specific patterns of peripheral nervous system activity that accompany discrete emotions (Ekman et al., 1983; Christie and Friedman, 2004), however, assuming that the origins of basic emotions are in the central nervous system, most research in this vein has focused on the central origins of emotions, specifically, the structures or networks of brain structures in which activity supports the emergence of particular emotions (Panksepp, 2007; Vytal and Hamann, 2010; Lindquist et al., 2012).

The availability of non-human animal analogues has made fear one of the best-studied emotions on a neuroanatomical level. On the whole, the empirical data support the idea that the amygdala, along with its efferent projections, is an essential structure for the generation of conditioned fear responses, which account for the majority of experienced fear (Davis, 1992, 1997). [Unconditioned fear in response to specific events like carbon dioxide-induced air hunger may rely on distinct neural pathways (Johnson et al., 2011; Feinstein et al., 2013)]. Extensive early evidence demonstrated that the amygdala plays a crucial role in the creation of conditioned fear in rodents. For example, lesions to the amygdala prevent rats from developing a conditioned fear response, like freezing in response to a stimulus that predicts shock (Blanchard and Blanchard, 1972). Later studies clarified the roles of the various subnuclei of the amygdala, demonstrating that the lateral nucleus is primarily involved in the acquisition of the fear response whereas the central nucleus is involved in both the acquisition and the expression of conditioned fear responses (Davis, 1992; Wilensky et al., 2006). The amygdala's many efferent projections coordinate autonomic and behavioral responses to fear eliciting stimuli. Projections from the central nucleus of the amygdala to the lateral hypothalamus are involved in activating autonomic sympathetic nervous system responses, and projections to the ventrolateral periaqueductal gray direct the expression of behavior responses, such as defensive freezing (Davis, 1992; LeDoux, 2012). The amygdala's central role in coordinated fear responding can be demonstrated by electrical stimulation studies showing that complex patterns of behavioral and autonomic changes associated with fear responses result from stimulation of the relevant regions of the amygdala (Davis, 1992). Heavy reliance on animal models is justified in the study of fear responding and the amygdala given how strongly conserved the amygdala nuclei involved in responding to conditioned threats are across species ranging from reptiles to birds to rodents to primates (LeDoux, 2012).

Ethical and pragmatic considerations prevent experimental paradigms employing electrical stimulation or ablation of the amygdala from being undertaken in human subjects. However, the advent of neuroimaging technologies have enabled considerable assessments of subcortical responses to a variety of emotional stimuli, enough to provide a basis for seven meta-analyses that have been conducted to assess patterns of brain activation in response to specific emotions (Phan et al., 2002; Murphy et al., 2003; Kober et al., 2008; Sergerie et al., 2008; Fusar-Poli et al., 2009; Vytal and Hamann, 2010; Lindquist et al., 2012). The findings from four of these meta-analyses support the role of the amygdala in human fear responding. Phan and colleagues reviewed 55 PET and fMRI studies (including 13 that assessed fear responding) and found that fear specifically activated the amygdala relative to other emotions (Phan et al., 2002). Sixty percent of studies assessing fear responses observed an increased amygdala response whereas fewer than 25% of other emotional tasks resulted in amygdala activation increases. Murphy and colleagues reviewed 106 PET and fMRI studies (Murphy et al., 2003) and again observed the most consistent amygdala responses during the induction or perception of fear relative to other emotions, interpreting their data as consistent with amygdala specialization for fear. In neither meta-analysis was any other structure observed to be consistently and selectively activated during fear paradigms. Fusar-Poli and colleagues included only fMRI studies assessing responses to emotional faces, but again found heightened amygdala responses to fearful faces relative to other emotional faces (Fusar-Poli et al., 2009). Finally, Vytal and Hamann (2010) employed a more sensitive meta-analytic method, activation likelihood estimation (ALE), to analyze the results of 83 PET and fMRI studies of emotion (including 37 that assessed fear responding) and again found strong support that the amygdala is preferentially active during fear paradigms, and this activation in this region differentiated fear from happiness, sadness, and disgust.

Three recent meta-analyses did not yield findings that fear is preferentially associated with amygdala activation. Two were conducted by Feldman-Barrett and colleagues (Kober et al., 2008; Lindquist et al., 2012). In the more recent analysis, Lindquist and colleagues analyzed 91 fMRI and PET studies of emotion, including 42 assessing fear (Lindquist et al., 2012). The authors observed that, bilaterally, the amygdala was the most active brain region during fear perception paradigms (although not significantly more active during fear than other emotions), but that the amygdala was not preferentially active during fear experience paradigms. The selection of studies in this meta-analysis may account in part for the differential findings. For example, of the nine fear-experience studies included in this analysis, six were conducted by a group that uses primarily IAPS pictures (Lang et al., 1999) and similar images to elicit disgust and fear (e.g., Stark et al., 2003; Schienle et al., 2005). These studies may be problematic because many of the "fear" images they use explicitly depict strong non-fear emotional cues (human or animal anger expressions) or depict events like a car accident or lava covering a road that are unpleasant but not obviously frightening. These meta-analyses also omitted pain anticipation and mood induction tasks included in other meta-analyses that are more directly relevant to fear experience (Murphy et al., 2003; Vytal and Hamann, 2010). The third meta-analysis (Sergerie et al., 2008) also excluded pain anticipation and mood induction tasks, in addition to employing a distinct analytical approach, whereby the authors compiled the statistical effect sizes of all studies of emotion (148 in total) that reported any activation in the amygdala and its surrounding regions. This approach yielded results showing amygdala activation that was stronger in response to positive emotional stimuli than to any negative emotional stimuli. Clearly, the conclusions drawn from the various metaanalyses are divergent enough to leave questions remaining as to whether the amygdala is in fact specifically implicated in fear responding.

Can the study of psychopathy clarify the role of the amygdala in fear experience? Perhaps, given the prominence of dysfunctional fear responding in psychopathy, empirical support that amygdala dysfunction underlies aberrant fear responding in psychopathic participants would support the amygdala's role in fear. And indeed, early hypotheses about the brain basis of psychopathy focused on potential amygdala dysfunction (Patrick, 1994; Blair et al., 2001). More recently, the results of both functional and structural neuroimaging studies support these hypotheses. Several studies have observed that psychopathy is associated with reduced amygdala activation during the viewing of fearful emotional facial expressions but not other expressions like anger, a pattern that is independent of attentional processes (Marsh et al., 2008; Dolan and Fullam, 2009; Jones et al., 2009; White et al., 2012). A recent study also found that psychopathy assessed in a community sample was also associated with a failure to exhibit amygdala activation to fear-evoking statements (Marsh and Cardinale, 2012b). Again, no group differences were observed in this task when other emotionally evocative statements were presented. (In addition, no main effect of fear stimuli was observed in the amygdala across groups. This suggests that amygdala responses to fear may fail to emerge in neuroimaging studies when the sample contains an unusual proportion of high psychopathy scorers.) Finally, a fear-conditioning paradigm found that psychopaths' failure to exhibit skin conductance responses during the task was accompanied by reduced activation in the amygdala and functionally connected regions of the cortex, such as orbitofrontal cortex and insula (Birbaumer et al., 2005).

These patterns of dysfunction may stem from structural abnormalities in the amygdala, which have also been observed in psychopathy. Structural abnormalities across multiple nuclei in the amygdala have been observed in psychopathy (Yang et al., 2009, 2010; Ermer et al., 2012). Yang and colleagues observed not only significant bilateral volume reductions in the amygdalae of adult psychopaths relative to controls controls, but also surface deformations in the vicinity of the amygdala's basolateral, lateral, cortical, and central nuclei. A later study indicated that these deformities are more significant in "unsuccessful" psychopaths, or those who have been prosecuted for their criminal acts (Yang et al., 2010). Ermer and colleagues identified gray matter reductions in adult psychopaths' amygdalae, in addition to other paralimbic regions such as parahippocampal gyrus (Ermer et al., 2012). It should be noted that how specific nuclei of the amygdala are involved in psychopathy is not yet clear, in part due to insufficient spatial resolution of functional imaging scan. Various hypotheses have been proposed regarding the role of discrete nuclei in psychopathic symptoms (Blair, 2005a; Moul et al., 2012).

On the whole, the results of these studies directly link amygdala dysfunction to observed deficits in fear responding in psychopathy.

But perhaps the most compelling evidence that amygdala dysfunction underlies fear deficits in psychopathy emerges from the results of paradigms testing fear responding in psychopaths and individuals with lesions to the amygdala. As previously described, psychopathy has been found to impair anticipatory skin-conductance responses (Lykken, 1957; Aniskiewicz, 1979; Herpertz et al., 2001; Birbaumer et al., 2005; Rothemund et al., 2012), fear-potentiated startle responses (Levenston et al., 2000; Herpertz et al., 2001; Rothemund et al., 2012), aversive classical conditioning (Flor et al., 2002), subjective experiences of fear (Marsh et al., 2011) and the recognition of fear from the face, body and voice (Marsh and Blair, 2008; Dawel et al., 2012). Striking parallels to these deficits can be found in studies of individuals with amygdala damage. In these individuals, comparable impairments in each of these fear paradigms have also been observed (**Table 1**).



Because amygdala dysfunction has been observed in psychopathy during several of these tasks, and because amygdala lesions impair performance in all of them, these patterns generate a compelling case for the role of the amygdala specifically in fear responding. Consistent with this, researchers studying one patient with bilateral amygdala damage (SM) clarify that she has not only striking deficits in fear responding, but these deficits are limited to fear responding:

SM's reaction to fear-inducing stimuli was not characterized by a loss of responsiveness, but rather manifested as a heightened arousal and interest in the face of a near-complete lack of avoidance and caution . . . Our findings suggest that the amygdala's role in the induction and experience of emotion is specific to fear. To say that SM is emotionless or unable to feel emotion is simply false. Her emotional deficit is primarily circumscribed to the behaviors and experiences that characterize a state of fear (Feinstein et al., 2011).

The clear correspondence between patterns of fear dysfunction observed in psychopathy and following amygdala lesions, in the absence of other clear emotional deficits, provides strong support for the specific involvement of the amygdala in fear. Dysfunction in the amygdala, whether via acquired lesion or developmental psychopathology, impairs fear-related processes while leaving other forms of emotion, such as anger, positive excitement, and disgust, largely intact. In answer to our second question, then, research in psychopathy suggests that the amygdala—or, more likely, specific populations of neurons within the amygdala (LeDoux, 2012)—plays a critical role in generating fear but does not appear to be critical for other emotions like positive excitement and anger.

## **HOW DO OUR OWN EXPERIENCES OF EMOTION PERTAIN TO OUR PERCEPTIONS OF AND RESPONSES TO OTHERS' EMOTION?**

The findings reviewed thus far suggest answers to a third question of ongoing interest in psychology and neuroscience: how do our emotional experiences affect our responses to and perceptions of others' emotions?

As we have seen, the evidence is clear that psychopathy is associated with deficits in the experience of fear but not other emotions. Psychopathic individuals show reduced physiological responding during anticipation of an aversive event, are less apt to adapt their behavior in response to punishment, and report reduced subjective fear. In some psychopaths the experience of fear may be essentially absent but, in keeping with the idea that psychopathy is a continuum rather than a taxon, fear is likely muted to varying degrees rather than absent in most individuals with psychopathic traits. Finally, psychopathy impairs the recognition of others' fear. Three meta-analyses have now demonstrated that psychopathy impairs recognition of fearful facial expressions in the face, body, and voice (Marsh and Blair, 2008; Wilson et al., 2011; Dawel et al., 2012), a pattern that is particularly closely associated with the central affective deficits of psychopathy. Marsh and Blair (2008) found that responses to fear are impaired to a significantly greater degree than any other emotion, and Dawel et al. (2012) found that the core affective features of psychopathy impaired the recognition of fear but not other emotions. In addition, psychopathy impairs the ability to identify the circumstances under which others would experience fear, such as in response to threats of harm (Marsh and Cardinale, 2012a). The parallels between psychopathic deficits in emotional experience and emotion recognition are striking. The emotion that psychopaths appear not to feel strongly—fear—is the same emotion that they have the most difficulty recognizing in others. Associations between the experience and recognition of emotion have previously been observed for a number of emotions, including fear (Buchanan et al., 2010). These data suggest the possibility of a basic empathic failure in psychopaths—they have great difficulty understanding an emotion in others that they themselves do not feel (or at least, do not feel strongly). This breakdown appears to occur in primarily for fear, rendering others' expressions of fear essentially meaningless in individuals with psychopathic traits.

These patterns are consistent with the theory that we recognize others' emotions through a low-level empathic simulation process, exploiting our own experiences of an affective state to understand others' experiences (Goldman and Sripada, 2005). Empathic simulation has become a favored explanation among researchers studying empathy for pain, boosted by a voluminous literature that the perception or inference of others' pain results in increased activation in the same brain structures involved in processing affective and motivational features of felt pain (Lamm et al., 2011). It is now widely agreed that the experience of empathy for pain emerges from shared representations for personal and vicarious experiences of affective states (Bernhardt and Singer, 2012).

The neurobiological evidence that empathy for fear also results from shared neural representations is equally compelling: both experienced fear and perceived fear result in specific activation in the amygdala, a structure that, when damaged or dysfunctional (as in the case of psychopathy), leads to impairments in both felt fear and the ability to recognize when others are experiencing fear. And yet an extremely similar pattern of data to support amygdala-based shared representations of fear has been interpreted differently from evidence supporting shared insula and anterior cingulate cortex-based representations for pain.

Why might this be? For one, the functions of the amygdala were first articulated in animal models, with a historical emphasis on stimulus-reinforcement learning rather than social functions and subjective experiences. This emphasis may have resulted in early observations of amygdala activity in response to fear expressions being interpreted as indicating that fear expressions signal threat, akin to the CS+ in a conditioning trial (Breiter et al., 1996; Morris et al., 1996). However, there is little empirical data to support the idea that fear expressions are interpreted as primarily threatening. Indeed, fearful facial expressions have been shown to be more strongly appetitive than aversive (Marsh et al., 2005b), and to resemble the morphological appearance of an infantile face (Marsh et al., 2005a) consistent with the idea that others' fear elicits empathic concern. The assumption that fearful expressions signify threat because they elicit amygdala activation may be a case of erroneous reverse inference—an inference regarding the psychological significance of a stimulus on the basis of neural responses to it (Poldrack, 2008).

Alternate hypotheses exist as well, such as that amygdala responses to fearful expressions reflect the amygdala's role in directing attention to the eyes of these expressions, which is critical to correctly identifying these stimuli (Dadds et al., 2006; Han et al., 2012). This theory is supported by findings that instructing both patients with amygdala lesions and children with psychopathic traits to attend to the eyes of faces reduces fear recognition deficits (Adolphs et al., 2005; Dadds et al., 2006). But this theory is less clearly able to accommodate the facts that psychopathy also impairs pre-attentive recognition of fearful faces (Sylvers et al., 2011), that both amygdala lesions and psychopathy impair recognition of vocalized fear, auditory stimuli for which the relevance of attention directed to salient features is unclear (Scott et al., 1997; Blair et al., 2002), and that psychopathy impairs the recognition of written statements that evoke fear (Marsh and Cardinale, 2012a). No low-level features of fear-evoking statements distinguish them from any other emotionally evocative statement, so there is no obvious mechanism by which the redirection of attention would be relevant to identifying these stimuli. I suggest that the total available evidence can be more parsimoniously interpreted under the hypothesis that amygdala is essential to generating an internal representation of fear, and that amygdala dysfunction in psychopathy impairs this process, thereby impairing identification of others' fear across contexts (Marsh and Cardinale, 2012b). This theory has the benefit of being consistent with the vast and consistent literature on empathy for pain.

That low-level emotional processes may impair empathy for fear in psychopathy may be particularly germane to an understanding of empathic processes more generally. "Empathy" is a term plagued by multiple overlapping definitions that include low-level emotional contagion, cognitive perspective-taking, and empathic concern (de Waal, 2009). The form of empathy most notoriously impaired in psychopathy is empathic concern, sometimes called sympathy, the inverse of which is callousness (Hare, 1991; Blair, 1995). By contrast, the evidence is clear that cognitive empathy, or perspective-taking, is not impaired in psychopathy (Blair, 2008; Jones et al., 2010; Schwenck et al., 2012). But emotional contagion, defined as simple affectedness by another's emotional state (de Waal, 2009), is clearly affected, at least in response to others' fear. The accumulated literature on psychopathy thereby suggests the possibility of critical links among emotional contagion in response to others' fear, recognition of others' fear, and empathic concern (Nichols, 2001). It also reinforces the importance of resisting the temptation to conflate the various forms of empathy, which may rely on distinct neurobiological processes.

From a societal perspective, understanding empathic deficits for others' fear may be the most important question of all that the study of psychopathy helps to answer. Although amygdala lesion cases can illuminate the amygdala's role in fear, because these lesions usually occur in late adolescence or adulthood, their effects on the development of other brain regions and behavior is more limited. This may be why amygdala lesions in adulthood are not associated with heightened aggression, whereas the case of psychopathy suggests a strong relationship between developmental deficits in fear and aggression. Fear plays an important role in preventing or ending aggression during social encounters (Blair, 1995, 2005b), and fearful emotional facial expressions elicit empathic concern and the desire to help from people who perceive them, even subliminally (Marsh and Ambady, 2007). The rationale for much research on psychopathy is that individuals with this disorder are responsible for a disproportionate amount of suffering, as they engage in a variety of antisocial, criminal, and violent behaviors that cause others distress and fear (Hare, 1993; Rutter, 2012). There is limited evidence that failure to exhibit empathic responses to others' pain is related to lower self-reported empathic concern or aggressive or antisocial behavior (Singer et al., 2004, 2006). In contrast, the evidence linking the failure to exhibit empathic responses to others' fear, both on a neural and a behavior level, is abundant. Psychopaths, in whom the failure to recognize others' fear or to generate empathic activation in the amygdala and autonomic nervous system is a hallmark feature, exhibit profound impairments in empathic concern for others and notoriously commit antisocial acts. Thus, as important as the study of psychopathy is for answering fundamental psychological and neuroscientific questions about the nature of emotion and empathy, an improved understanding of emotion and empathy as they pertain to psychopathy may be critical to developing improved means of ameliorating psychopaths' harmful effects on others.

#### **CONCLUSIONS**

The study of psychopathy has generated information relevant to addressing three questions of central importance to emotion and affective neuroscience. Evidence collected from psychopathic populations supports the conclusion that fear is qualitatively distinct from other emotions and arises from discrete neurobiological processes, rather than the conclusion that emotions like fear and anger reflect quantitative variations in core dimensions like arousal and valence. Recent neurocognitive and neuroimaging evidence also supports the specific role of the amygdala in generating a fear response over the view that the amygdala plays a domain-general role equally relevant to the generation of multiple emotions. And finally, psychopaths'

## **REFERENCES**


Damasio, A. R. (1995). Double dissociation of conditioning and declarative knowledge relative to the amygdala and hippocampus in humans. *Science* 269, 1115–1118.


parallel deficits in experiencing fear and recognizing fear in others lend support to the notion that empathy for affective states results from shared representations for personal and vicarious experiences of fear, consistent with simulation-based theories of empathy. These conclusions may prove useful not only in furthering the neuroscientific studies of emotion, but in developing a better understanding of the fundamental nature of psychopathy, empathy and aggression.

deficits in psychopathy and autism. *Q. J. Exp. Psychol.* 61, 157–170.


Response and habituation of the human amygdala during visual processing of facial expression. *Neuron* 17, 875–887.


Davis, M. (1997). Neurobiology of fear responses: the role of the amygdala. *J. Neuropsychiatry Clin. Neurosci.* 9, 382–402.


Aversive Pavlovian conditioning in psychopaths: peripheral and central correlates. *Psychophysiology* 39, 505–518.


punishment. *J. Abnorm. Psychol.* 71, 25–29.


brief hypercarbic gas exposure. *J. Psychopharmacol.* 25, 26–36.


Psychophysiology, University of Florida.


right amygdala during anticipation of negative stimulus. *Behav. Modif.* 27, 607–619.


Patrick, C. J. (2010). Triarchic Psychopathy Measure (TriPM). PhenX Toolkit Online assessment catalog.


G., Federlein, J., Buttner, T., et al. (1999). Knowing no fear. *Proc. R. Soc. Lond. B. Biol. Sci.* 266, 2451–2456.


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 08 February 2013; paper pending published: 01 March 2013; accepted: 22 April 2013; published online: 10 May 2013.*

*Citation: Marsh AA (2013) What can we learn about emotion by studying psychopathy? Front. Hum. Neurosci. 7:181. doi: 10.3389/fnhum.2013.00181*

*Copyright © 2013 Marsh. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any thirdparty graphics etc.*

# Neuropsychology, social cognition and global functioning among bipolar, schizophrenic patients and healthy controls: preliminary data

*Elisabetta Caletti 1, Riccardo A. Paoli <sup>1</sup> \*, Alessio Fiorentini <sup>1</sup> \*, Michela Cigliobianco1, Elisa Zugno1, Marta Serati 1, Giulia Orsenigo1, Paolo Grillo2, Stefano Zago3, Alice Caldiroli 1, Cecilia Prunas 1, Francesca Giusti 1, Dario Consonni <sup>2</sup> and A. Carlo Altamura1*

*<sup>1</sup> Department of Neuroscience and Mental Health, Psychiatric Clinic, University of Milan, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico, Milan, Italy*

*<sup>2</sup> Epidemiology Unit, Department of Preventive Medicine, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico, Milan, Italy*

*<sup>3</sup> Neurology Unit, Department of Neuroscience and Mental Health, University of Milan, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico, Milan, Italy*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Agustin Ibanez, Institute of Cognitive Neurology, Argentina Nicola Dusi, Department of Public Health and Community Medicine Section of Psychiatry, Italy Annick Razafimandimby, University of Caen Basse-Normandie, France*

#### *\*Correspondence:*

*Riccardo A. Paoli and Alessio Fiorentini, Department of Neuroscience and Mental Health, Psychiatric Clinic, University of Milan, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico, 35, via F. Sforza, 20100 Milan, Italy e-mail: riccardo.paoli@guest.unimi.it; alessio.fiorentini@policlinico.mi.it*

This study aimed to determine the extent of impairment in social and non-social cognitive domains in an ecological context comparing bipolar (BD), schizophrenic (SKZ) patients and healthy controls (HC). The sample was enrolled at the Department of Psychiatry of Policlinico Hospital, University of Milan; it includes stabilized SKZ patients (*n* = 30), euthymic bipolar patients (*n* = 18) and HC (*n* = 18). Patients and controls completed psychiatric assessment rating scales, the Brief Assessment of Cognition in Schizophrenia (BACS) and the Executive and Social Cognition Battery (ESCB) that contains both ecological tests of executive function and social cognition, in order to better detect cognitive deficits in patients with normal results in standard executive batteries. The three groups differed significantly for gender and substance abuse, however, the differences did not influence the results. BD patients showed less impairment on cognitive performance compared to SKZ patients, even in "ecological" tests that mimic real life scenarios. In particular, BD performed better than SKZ in verbal memory (*p <* 0*.*0038) and BACS symbol coding (*p <* 0*.*0043). Regarding the ESCB tests, in the Hotel task SKZ patients completed significantly less tasks (*p <* 0*.*001), showed a greater number of errors in Multiple Errands Test (MET-HV) (*p <* 0*.*0248) and a worse performance in Theory of Mind (ToM) tests (*p <* 0*.*001 for the Eyes test and Faux pas test). Both patients' groups performed significantly worse than HC. Finally, significant differences were found between the two groups in GAF scores, being greater among BD subjects (*p <* 0*.*001). GAF was correlated with BACS and ESCB scores showing the crucial role of cognitive and ecological performances in patients' global functioning.

**Keywords: schizophrenia, bipolar disorder, social cognition, neuropsychological deficits, ecological tests**

## **INTRODUCTION**

Over the last two decades, there has been an increased interest in neurocognitive functioning and in social cognition (SC) in major psychoses, schizophrenia (SKZ) and bipolar disorder (BD) (Barch and Keefe, 2010; Samamé et al., 2012), diseases causing severe behavioral, relational, and socio-familial disabilities (Altamura et al., 2001). It is widely recognized that SKZ patients exhibit neuropsychological deficits in several cognitive domains, including memory, attention, and executive functions over time (Cornblatt and Keilp, 1994; Addington and Addington, 2000; Kuperberg and Heckers, 2000). Moreover, they experience low levels of performance and a reduced ability to live independently, despite the remission of acute symptomatology, with a negative impact on social and occupational functioning (Heinrichs and Zakzanis, 1998; San et al., 2007; Tuulio-Henriksson et al., 2011). Both neurocognitive deficits and limitations in the ability to carry out daily activities could contribute to poor circumstances in daily life, exaggerating negative attitudes, thus contributing to lower motivation, interest, and engagement in productive activities.

Neurocognitive dysfunction is also a key aspect of BD (Lewandowski et al., 2011), observable even during the remission of symptoms (Torres et al., 2007; Bora et al., 2011; Mann-Wrobel et al., 2011; Gama et al., 2013), with a strong impact on social functioning (Huxley and Baldessarini, 2007; Martino et al., 2009; Wingo et al., 2010). A meta-analysis by Kurtz and Gerraty (2009) considering 42 studies including euthymic BD patients (e.g., Bora et al., 2009a) stated that BD is characterized by an overall level of moderate cognitive impairment, that may exacerbate during acute phases, having a direct effect on rehabilitation outcome and an indirect effect on SC (Bell et al., 2009). Neuroimaging studies in SKZ have linked structural and functional abnormalities to symptoms and progressive structural changes to clinical course and functional outcome (Ahmed et al., 2013). Alterations in brain structures has been found also in BD, more pronounced in patients with repeated episodes (Gama et al., 2013).

Social cognition, defined as the mental operations underlying social interactions (Green et al., 2005), is considered a multidimensional domain, involving a complex set of processes allowing adaptive social interaction as the representation of internal somatic state, the awareness of the self-perception of others and interpersonal motivation (Fiske and Taylor, 1991; Kunda, 1999; Amodio and Frith, 2006). Both SKZ and BD patients show deficits in SC tasks, mainly in those requiring greater context sensitivity, performing normally in tasks that can be solved by explicit knowledge (Baez et al., 2013). Previously, Bromley and Brekke (2010), measuring social functioning in SKZ, highlighted how explicit knowledge is not enough to perform well in real life, identifying three dimensions of functioning: functional capacity, functional performance, and functional outcome. In particular functional capacity is the ability to perform a functional task (capacity) while functional performance is the ability to perform (performance) the same task in the community environment. Functional outcomes are the result of both capacity and performance; indeed, an individual may demonstrate a good functional capacity but may not be able to use it in his own social context. Recently, Pinkham et al. (2013) identified four core domains of SC: emotion processing, social perception, theory of mind/mental state attribution, and attributional style/bias. They focused on one particular aspect of SC, known as "Theory of Mind" (ToM) or "mentalizing" conceptualized as the ability to reflect upon one's own and other persons' mental states including desires, beliefs, knowledge, intentions, and feelings (Frith and Frith, 2003), repeatedly shown to be compromised in most SKZ patients (Lee et al., 2004) and evaluable with a variety of tasks and assessment methods (Brüne and Brüne-Cohrs, 2006). Most common tests utilized to assess ToM abilities are the Hinting Task (Corcoran et al., 1995), the cartoon method (Corcoran et al., 1997; Brüne, 2003), the pictorial tasks (Sarfati et al., 1997), ToM Advanced Test—composed of stories and drawings—created by Happé (1994), the "Moving Shapes" paradigm, used in early stages of SKZ (Abell et al., 2000; Koelkebeck et al., 2010), the Eyes Test designed to assess the capacity to re-attribute complex mental states in adults and adolescents in absence of severe mental retardation (Baron-Cohen et al., 2001; Serafin and Surian, 2004), the Faux Pas Test evaluating the ability to recognize a social faux pas (Baron-Cohen et al., 1999).

In SKZ neural mechanisms underlying metacognition, defined as the processes by which we monitor and control our own cognitive processes (Frith, 2012), include frontal lobe, in particular fronto-temporal and fronto-parietal circuits, premotor cortex, mirror neurons and dopaminergic reward circuits, involving neuropeptides such as oxytocin and vasopressin (Gallese and Goldman, 1998; Chafee and Goldman-Rakic, 2000; Mehta et al., 2013). Interestingly, most cortical abnormalities are subject to regional variations and differ from those observed in neurodegenerative diseases. Gray matter reductions in "social brain" areas of SKZ patients such as temporal and left occipital white matter regions, left posterior callosal region pole and left anterior hippocampus seem to be involved in socioemotional processing including ToM (Olson et al., 2007; Schobel et al., 2009; Miyata et al., 2010).

BD patients, both during mood phases and euthymic states, revealed impaired emotion processing with poor ability to distinguish facial emotions and impaired ToM (Bozikas et al., 2006; Summers et al., 2006; Lahera et al., 2008; Sánchez-Moreno et al., 2009; Montag et al., 2010; Martino et al., 2011). The processing of facial expressions of others relies upon the neural system of ventral prefrontal cortex (VPFC), amygdala and their interconnections, disrupted in BD patients (Blumberg et al., 2003; Lochhead et al., 2004; Adler et al., 2005; Stanfield et al., 2009; Gama et al., 2013; Lim et al., 2013).

Different brain regions seem to undergo different domains of SC: in some studies amygdala volume was correlated to impaired facial emotion recognition (FER) ability, whereas medial prefrontal cortex volume was correlated to impaired emotion attribution (Yamada et al., 2007; Matsukawa and Murai, 2013). Furthermore, ventral striatum, which is implicated in emotional and motivational aspects of behavior, seem to have an important function for SC ability (Adolphs, 2001). ToM studies in euthymic BD patients (Montag et al., 2010) revealed, independently from cognitive deficits, an insufficient performance in cognitive ToM with preserved emotional mentalizing abilities correlated with the number of manic episodes (Kerr et al., 2003; Olley et al., 2005; Lahera et al., 2008). A recent electrophysiological study by Ibañez et al. (2013) has found emotional N170 impairment in SKZ and BD patients, being cortical processing of emotional stimuli predictive of social-cognitive profile, indexed by measures of ToM, fluid intelligence, speed processing and executive functions. Previously, a comparison between euthymic BD patients and controls, pointed out abnormal facial modulation associated with individual profiles of ToM in BD patients (Ibañez et al., 2012).

In summary, neuropsychological and SC deficits are present both in SKZ and in BD, involving several brain areas, among which frontal lobes seem to play a crucial role.

Until now in literature neuropsychological findings have been mainly obtained with classical cognitive measures, however, more context-sensitive measures similar to real-life situations should be used when studying major psychoses (Baez et al., 2013). For this reason, in our study, we administered the Executive and Social Cognition Battery (ESCB), proved to be more sensitive in detecting executive and social cognitive impairments than conventional batteries, both in early behavioral variant of frontotemporal dementia (bvFTD) and BD (Torralva et al., 2009, 2012). Previous studies highlighted the importance of including ecological tests in the assessment of BD patients in order to provide a more realistic cognitive profile of this patient population, allowing better therapeutic and rehabilitation strategies able to minimize impact in real-life settings (Torralva et al., 2012).

## **AIM OF THE STUDY**

The objective of the study was to analyze neurocognitive abilities, SC and global functioning in a pharmacologically stabilized sample of SKZ, BD patients in comparison to HC, using a specific neuropsychological and SC battery in an ecological context to analyze a possible correlation with subjects' global assessment of functioning.

# **MATERIALS AND METHODS**

### **SAMPLE**

Forty-eight outpatients were enrolled at the Department of Psychiatry, Fondazione IRCCS Ca' Granda, Ospedale Maggiore Policlinico University of Milan: 30 stabilized SKZ patients (10 paranoid, 14 undifferentiated, 6 disorganized subtypes) and 18 euthymic BD patients (10 BD I, 8 BD II). Age-matched HC (*n* = 18) were recruited among volunteers who did not have a history of drug abuse. This study was approved by the Ethics Committee of Fondazione IRCCS Ca' Granda Maggiore Policlinico Hospital, Milan and informed consent was obtained from all subjects. Patients' inclusion criteria were a diagnosis of SKZ or BD according to Diagnostic and Statistical Manual for Mental Disorders-Text Revision (DSM-IV-TR). Exclusion criteria were: acute psychotic episodes in SKZ referring to Positive and Negative Symptom Scale (PANSS) (Kay et al., 1987) with a score *>*50; acute depression episodes in BD referring to Hamilton Depression Rating Scale scores (HDRS *>* 7) (Hamilton, 1960); acute mania episodes in BD referring to Young Mania Rating Scale scores (YMRS *>* 10) (Young et al., 1978); mental retardation or other neurological brain diseases.

# **EVALUATION TOOLS**

Trained psychiatrists conducted the Structured Clinical Interview for DSM Axis I (First et al., 2002) and rated patient functioning at baseline. The following psychometric scales were administered to SKZ patients: PANSS (Kay et al., 1987), Calgary Depression Scale for Schizophrenia (CDSS) (Addington et al., 1990) and Clinical Global Impression (severity of illness) (CGIs) (Guy, 1976). HDRS (Hamilton, 1960), YMRS (Young et al., 1978), and Hamilton Anxiety Scale (HAM-A) (Hamilton, 1959) were administered to BD patients. Global functioning (social, functional, and occupational) for both patients' groups and HC was measured with General Assessment of Functioning scale (GAF) included in DSM-IV-TR (American Psychiatric Association, 2000).

## **NEUROCOGNITIVE ASSESSMENT**

Cognitive status of both HC and patients (SKZ and BD) was assessed through standard neuropsychological battery: the Brief Assessment of Cognition in Schizophrenia (BACS) (Keefe et al., 2004; Anselmetti et al., 2008). In our study we applied a battery with normative data available for the Italian population.

Among cognitive batteries BACS (Keefe et al., 2004) can be easily administered and scored and it has been used in several SKZ clinical trials (Keefe et al., 2008). BACS assess different domains of cognitive function (Verbal Memory, Working Memory, Motor Speed, Attention, Verbal Fluency, and Executive Functions) with six tests, lasting about 35 min. Keefe et al. (2006) suggest that BACS scores are correlated with a performance-based measure of functional capacity and real-world functional outcome. It is noteworthy that BACS has also high test–retest reliability properties which are important for assessing alteration over time (Anselmetti et al., 2008).

Following is a description of the 6 subtests of the BACS:

– List Learning (Verbal Memory): Subjects are read a list of 15 words and then asked to recall as many as possible. This procedure is repeated five times and designed to measure episodic memory functions.


# **EXECUTIVE AND SOCIAL COGNITIVE MEASURES**

All participants (both HC and patients) completed ecological tasks included in the ESCB (Torralva et al., 2009). This battery was created in order to detect cognitive and social components of the early stages of bvFTD, consisting of five subtests. Some tasks were used largely in neurological and neurorehabilitation fields (e.g., Manes et al., 2009; Torralva et al., 2009). Shallice and Burgess (1991) first demonstrated that patients with frontal lobe damage may be specifically impaired in everyday situations that require planning and multitasking, despite normal performance on standard cognitive tests. These authors found that three patients with frontal lobe syndrome due to traumatic brain damage performed well on a wide range of conventional executive tests, but showed noticeable difficulties with two novel tasks, where they had to organize their behavior and set priorities in the face of competing demands. The use of ecological tests could be helpful in psychiatric studies considering that patients have a large range of neuropsychological impairments.

Below is a description of the subtests of the ESCB.

The Hotel Task: "Hotel task" is part of a number of ecologically valid tests of executive function, in which individuals are required to carry out five hotel-related tasks, e.g., making up guests' bills, sorting coins, proofreading a brochure. Patients were required to devote some time to each test having only 15 min. Patients also have to keep in mind to press a button at two pre-designated times that correspond to opening and closing the hotel garage gate. Performance is measured in three ways, number of tasks attempted, deviation from optimal time on each task, and opening and closing the garage gate (Manly et al., 2002; Torralva et al., 2009).

MET-HV: As regards to Multiple Errands Test (MET) (Shallice and Burgess, 1991), a multitasking test carried out in shopping context, we adapted the 2002 test version (Knight et al., 2002) created for hospital settings (MET-HV). Its strengths are the simplicity of administration and the few administration time, being a good indicator of functional performance. The test requires carrying out a number of tasks simulating "real life" situations in which minor inconveniences can take place. The test takes place inside the hospital: the patient has a card with several sets of simple tasks with 12 subtasks. The first set requires participants to attain six specific goals, which include collecting an envelope from the secretary, purchasing three items (a postcard, a letter, a bottle of mineral water), using the internal phone and posting something to an external address. The second set involves obtaining and writing down on a chart some information (the price of a snack, how many parking spaces are available for visitors in the hospital, at what time the hospital's bar opens and closes on Friday and Saturday). The participant is required to call the examiner about 20 min after the test has begun and state the time over the phone. The third task requires the participant to inform the examiner when the task has been completed. Rules are clearly stated in the instruction sheet and the participant's behavior, while carrying the tasks, is monitored by two examiners. Errors in this test were categorized as: (a) inefficiencies, where a more effective strategy could have been applied; (b) rule breaks, where a specific rule was broken; (c) interpretation failure, where the requirements of a task had been misunderstood; (d) task failures and (e) total fails, the sum of all the previous ones.

MET is often utilized as an assessment tool useful for the detection of deficits in real-life executive functioning in post-stroke patients (Rand et al., 2009) and among patients with vascular lesions (Manes et al., 2009). Dawson et al. (2005b) showed a good correlation of this task with self-report measures of every day ability and living skills in cerebrovascular accident patients.

1. IOWA Test: This test represents a version of Bechara and colleagues test (1994), firstly used in neurological settings (patients with prefrontal cortex damage). This is a gambling task that models real life personal decision making activities in real time that include reward and punishment and the uncertainty of outcomes. The task involves four decks of cards, called A, B, C, and D. Subjects must choose one card at a time from one of the four decks. Desks A and B are ultimately risky (large rewards and large punishment) while C and D are more conservative (small rewards and small loss). The task is completed after 100 selections. Net scores are calculated with following formula: [(C + D) − (A + B)]; positive net scores reflect advantageous performance whereas negative net scores reflect the disadvantageous performance (Bechara et al., 2000). The IOWA Gambling Task (IGT) was putatively associated with ventromedial frontal lesions, and show decision-making deficits manifest in consistent selection of risky decks (Bechara et al., 2000; Torralva et al., 2007). The task is influenced by cognitive functions besides reward coding and use, including learning, shifting, and spatial working memory (Dunn et al., 2006).


### **STATISTICAL ANALYSES**

Pairwise comparisons between groups were performed using χ<sup>2</sup> test for categorical variables and two-sample Wilcoxon (Mann-Whitney) rank sum for quantitative variables. Multiple regression models with robust standard error were also used to compare GAF, neuropsychological, and ecological measures across groups while adjusting for gender, age, and past use for alcohol or drugs.

Spearman's rho correlation coefficient was used to quantify the relationship between GAF and neuropsychological or ecological measures. Sidak correction for multiple correlations was performed.

Significance level for all statistical tests was set at 0.05, twotailed.

Statistical analyses were performed using Stata SE (StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX: StataCorp LP).

# **RESULTS**

SKZ patients differ significantly from HC subjects (*p* = 0*.*001) and BD patients (*p <* 0*.*001) for substance abuse and gender, with a greater percentage of males, while no gender differences were found between BD patients and HC (*p* = 0*.*45). A more frequent history of substance abuse was found among SKZ patients with respect to the other two groups (*p* = 0*.*02). No significant differences were found between the groups for age, age at onset, duration of illness, and duration of untreated illness (**Table 1**).

Statistical analysis adjusted for abuse, age, and gender confirmed that our results were not influenced by these variables (**Tables 2**, **3**).

Regarding neuropsychological tasks both SKZ and BD subjects showed in BACS test a significant worse performance compared to HC in verbal memory (for BD: *z* = 2*.*437; *p* = 0*.*01; for SKZ:


**Table 1 | Differences between healthy controls (HC), bipolar disorder (BD) and schizophrenia (SKZ) patients in demographic and clinical measures.**

*DUI* = *Duration of Untreated Illness.*

**Table 2 | Performances of healthy controls (HC), bipolar disorder (BD) and schizophrenia (SKZ) patients on brief assessment of cognition in schizophrenia (BACS).**


*P-c* = *P-crude; P-a* = *P-adjusted (Multiple regression model—for abuse).*

*Values are shown as Mean (SD).*

*n.v.* = *normal value, referred to an equivalent score 2 (Anselmetti et al., 2008).*

*z* = 4*.*782; *p <* 0*.*001), working memory (for BD: *z* = 2*.*279; *p* = 0*.*02; for SKZ: *z* = 4*.*496; *p <* 0*.*001), motor speed (for BD: *z* = 2*.*010; *p* = 0*.*04; for SKZ: *z* = 3*.*749; *p* = 0*.*0002), symbol coding (for BD:*z* = 3*.*102; *p* = 0*.*002; for SKZ:*z* = 4*.*898; *p <* 0*.*001) and verbal fluency (for BD: *z* = 3*.*546; *p* = 0*.*0004; for SKZ: *z* = 4*.*835; *p <* 0*.*001) (**Figures 1**–**5**).

Performances differed significantly between BD and SKZ in BACS verbal memory (*z* = −2*.*897; *p* = 0*.*0038) and symbol



*P-c* = *P-crude; P-a* = *P-adjusted (Multiple regression model - for abuse, age, gender).*

*MET-HV* = *Multiple Errands Test-hospital version;*

*Values are shown as Mean (SD).*

*\*Healthy controls values reported in Torralva et al.. study (2012).*

*\*\*Advantageous net scores as calculated in Bechara et al. (2000).*

*\*\*\*Mean value in 41–60 age population as reported in "Test degli Occhi" (Serafin and Surian, 2004).*

Tower of London test (*z* = 3*.*079; *p* = 0*.*0021) (**Figure 6**). With respect to SC tasks including ToM tests, SKZ patients performed worse than HC and BD in Eyes Test (*z* = 4*.*096, *p <* 0*.*001; *z* = −3*.*947; *p* = 0*.*001, respectively) and in Faux Pas

Test (*z* = 4*.*410; *p <* 0*.*001; *z* = −4*.*138; *p <* 0*.*001, respectively), while BD patients performed similarly to HC (**Figure 7**). In MET-HV test, included in ESCB, both BD and SKZ subjects attempted to perform less tasks (for BD: *z* = 3*.*614, *p* = 0*.*0003; for SKZ: *z* = 3*.*819, *p* = 0*.*0001) (**Figure 8**), failed to complete a greater number of tasks (for BD: *z* = −3*.*614, *p* = 0*.*0003; for

◦Significantly different from BD (*p <* 0*.*05).

SKZ: *z* = −3*.*819, *p* = 0*.*0001) (**Figure 8**), broke more rules (for BD: *z* = −4*.*916, *p <* 0*.*001; for SKZ: *z* = −3*.*893, *p* = 0*.*0001) (**Figure 9**) and showed more interpretation failures (for BD: *z* = −2*.*092, *p* = 0*.*0365; for SKZ:*z* = −3*.*037, *p* = 0*.*0024) compared to HC (**Figure 10**). Moreover, both BD and SKZ subjects committed more inefficiencies (for SKZ: *z* = −4*.*538, *p <* 0*.*001; for BD: *z* = −2*.*645, *p* = 0*.*0082) than HC (**Figure 9**). In the

**FIGURE 6 | Differences between HC and SKZ in neuropsychological tasks: BACS Tower of London test.** •Outliers. ∗Significantly different from HC (*p <* 0*.*05).

Hotel Task, both patients' groups attempted to complete significantly less tasks (for BD: *z* = 2*.*089, *p* = 0*.*0367; for SKZ: *z* = 5*.*241, *p <* 0*.*01) and obtained greater time deviations in all tasks (for BD: *z* = −2*.*081, *p* = 0*.*0375; for SKZ: *z* = −4*.*750, *p <* 0*.*001) compared to HC (**Figure 11**). When comparing SKZ and BD subjects, SKZ patients showed less tasks attempted (*z* = −4*.*300, *p <* 0*.*001) and a greater sum of time deviation (*z* = 3*.*760 and *p* = 0*.*0002) (**Figure 11**). Finally, SKZ subjects' performed worse than HC in IGT (*z* = 2*.*740, *p* = 0*.*0061), while we didn't find any significant difference neither between BD and HC nor between SKZ and BD patients (**Figure 12**).

Regarding global functioning, both BD and SKZ patients differed significantly from HC for GAF scores being inferior for

**FIGURE 8 | Differences between groups in ESCB tasks: MET-HV Task Attempted and Task Failures.** ∗Significantly different from HC (*p <* 0*.*05).

patients (BD: *z* = 4*.*843, *p <* 0*.*001; SKZ: *z* = 5*.*758, *p <* 0*.*001). A significant difference was found also between SKZ and BD, with greater GAF scores for BD subjects (*z* = −5*.*204, *p <* 0*.*001). GAF scores were positively correlated to the performance in all BACS tests, except for Tower of London (verbal memory: rho = 0.5866, *p <* 0*.*001; working memory: rho = 0.5225, *p* = 0*.*0001; motor speed: rho = 0.4461, *p* = 0*.*0036; symbol coding: rho = 0.5799, *p <* 0*.*001; verbal fluency: rho = 0.5250, *p* = 0*.*0001), in both ToM tasks (Eyes test: rho = 0.6066, *p <* 0*.*001; Faux Pas Test: rho = 0.5847, *p <* 0*.*001) and to number of tasks attempted in MET-HV (rho = 0.4347, *p* = 0*.*0181) and Hotel Task (rho = 0.7024, *p <* 0*.*001) (**Figure 13**). Instead, GAF

(*p <* 0*.*05).

scores were negatively correlated to the number of task failures (rho = −0.4347, *p* = 0*.*0181), inefficiencies (rho = −0.4655, *p* = 0*.*0058) and interpretation failures (rho = −0.4653, *p* = 0*.*0059) in MET-HV and to the sum of time deviations in Hotel Task (rho = −0.6174, *p <* 0*.*001) (**Figure 13**).

On the contrary, performance in Tower of London (rho = 0.3473, *p* = 0*.*0861), rule breaking in MET-HV task and IGT performance did not seem to predict global functioning (rho = −0.3656 and 0.2473, *p* = 0*.*1511 and 0.9364, respectively) (**Figure 13**).

## **DISCUSSION**

Among neuropsychological tests, a significant impairment was found both in SKZ and BD patients with respect to HC, with SKZ performances being worse than BD in the majority of tests. Data regarding impairment in verbal memory and executive functions were previously underlined by Altshuler et al. (2004) in both diseases. Verbal memory deficits, more pronounced in SKZ patients than in BD patients, have also been reported by Konstantakopoulos et al. (2011). Symbol Coding has been recently considered the most sensitive test in detecting attention and speed processes, seen as the early signs of deterioration in a variety of neurological disorders (Strauss and Brandt, 1986; Storandt and Hill, 1989; Lezak, 1995; Wechsler, 2008).

Interestingly our work shows differences among SKZ, BD, and HC performances in ecological tests, able to detect more subtle cognitive deficits, indicating that mainly SKZ but also BD patients could have a worse capacity for planning, low flexibility, and organization skills as previously reported by Torralva et al. (2009). Our findings are in line with some previous literature results, showing a worse performance for SKZ than for BD subjects both in standard cognitive domains, ToM tasks and experimental tasks (mainly in MET-HV and Hotel task) (Sanchez-Morla et al., 2009). It is important to note that our sample of SKZ patients present a significant disability in detecting both emotional (Eyes Test) and verbal (Faux Pas Test) ToM tasks as an evidence of global compromised ToM ability. In particular, in the Eyes Test SKZ patients are more compromised than BD patients in mental state decoding of other individuals' facial expressions, in contradiction to Donohoe et al. (2012), who state comparable levels of impairment in the two disorders. Furthermore, our findings showed a deficit in the Faux Pas Test in both groups, being SKZ patients more compromised; in line with previous studies reporting serious ToM deficits, both in BD and SKZ, demonstrated by poor performance in advanced ToM tasks, such as the recognition of a faux pas, a social misstep (Bertrand et al., 2007; Brüne et al., 2007; Bora et al., 2009a).

Failures in cognitive ToM (Faux Pas Test) were observed in a previous study by Ibañez et al. (2012) also among BD patients when compared to HC; in that test patients' performance was significantly reduced, whereas the Reading The Mind in the Eyes test, measuring mainly affective ToM, showed a non-significant difference. Cognitive ToM refers specifically to the ability to infer the mental beliefs and states of others, while "affective" ToM (emotional) refers to the ability to infer the emotions of others (Barrera et al., 2013).

In IOWA test the SKZ group showed a poor ability in reasoning before acting and insensitivity to future consequences with respect to BD patients, although statistical significance was not reached. In addition, a significant difference between SKZ and BD patients was found in GAF functionality, in part possibly explained by deficits reported in previous tests. According to our results, all the tasks, especially ecological tests, were significantly correlated with GAF scores, being a possible useful marker of social functioning in major psychoses. Although our sample is composed of stabilized patients, we have to take into account that sub-syndromal mood changes, mainly depressive, could alter the mechanisms of social understanding, thus, worsening the ability to detect faux pas or embarrassing social situations and recognize basic and complex emotions.

Numerous studies have reported that "real-world" situations, reproduced in ecological tests (e.g., MET-HV), assess everyday life ability better than traditional tests and could be more prognostic (Burgess et al., 1998, 2006; Wilson et al., 1998; Knight et al., 2002; Alderman et al., 2003; Dawson et al., 2005a,b). Studies involving individuals with bvFTD have underlined how patients can score normally on neuropsychological test and reveals no abnormalities in brain imaging, but demonstrate notable defects in social interactions therefore it is necessary to consider the context as an intrinsic part of SC. The social context network model (SCNM) has been linked to a fronto-insular-temporal circuit and seems to be involved in SC, attempting to update context, coordinate internal and external processes and associate previous information (Ibañez and Manes, 2012). Given the overlap of certain symptomatic dimensions between bvFTD and psychiatric disorders (apathy, disinhibition, depression, anhedonia, stereotyped behavior, and psychosis), such as late onset SKZ and BD (Pose et al., 2013), it could be hypothesized that the abnormal social context processing could explain SC deficits also in major psychoses.

We identified some problem areas in our study. Our sample, SKZ and BD patients differed significantly for substance (mainly cocaine, cannabis), alcohol abuse and gender. Alcohol and substance abuse have often proved to have an influence on cognitive performance, particularly on immediate verbal learning, processing speed and working memory (Meijer et al., 2012). Despite the high prevalence of substance abuse (particularly cannabis)

both in SKZ and BD disorders, study results are still inconclusive regarding the repercussions on neurocognitive functions (Coulston et al., 2007). Moreover, in our sample abuse occurred many years before the assessment and was substantially moderate. Regarding gender, our sample was mainly composed by males and a study by Bora et al. (2009b) indicated that female patients performed better in ToM tests. However, statistical analysis performed in our sample, proved that gender and abuse did not influence the performances. Furthermore, we should take into account that the current observed relations between SC, neurocognition, and clinical assessment, have been studied crosssectionally and may not necessarily represent the longitudinal outcome. The fact that all patients were medicated, constitutes an additional limitation of the study, given that patients received different classes of drugs to modulate neurotransmitters, which are known to affect specific aspects of ToM (Montag et al., 2008). Our sample did not present significant comorbidities, except for the presence of some personality traits although without the configuration of a personality disorder. Finally, a limitation of our study could be the diagnostic variability. It is accepted in literature that outcome of SKZ patients is worse than BD, but partially similar to BD type 1 patients (Lewandowski et al., 2011), suggesting the presence of a continuum between a typically psychotic disorder such as SKZ and affective disorders with important psychotic features such as BD1. In particular, among SKZ, 10 had a diagnosis of paranoid SKZ, 14 of undifferentiated SKZ and 6 of a disorganized subtype. It has been demonstrated that there is a difference in outcome between paranoid and non-paranoid SKZ (Kendler et al., 1984) and that patients with undifferentiated SKZ showed less clinical and cognitive recovery than the others (Seltzer et al., 1997). More recently, Salokangas et al. (2002) argued that the disorganized subtype presented poorer outcome and low quality of life, although data in literature are still inconclusive. However, in the last years, the multi-factorial etiology and the evidence of a continuum between the disorder and the general population (Kaiser et al., 2011) identified an ultra-risk population, supporting a dimensional approach. In light of these considerations and of the substantial homogeneity of the three groups, the diagnostic differences within our sample had more than likely not influenced our results.

# **CONCLUSION**

Our work substantially confirms data presented in literature of a more severe impairment of SKZ than BD in cognitive and SC tasks. To our knowledge, this is the first attempt to compare BD, SKZ patients and HC with the application of tools derived from neurological context considering SC as a mediator more closely related to community functioning than neurocognition and a target for psychosocial and pharmacological interventions. The originality of our work consists in a more specific and in-depth assessment of SKZ functioning in comparison to BD and HC.

The future goal will be to confirm these data in a larger sample, study bipolar subtypes in order to see if BD patients type I have a similar social cognitive functioning to SKZ as previously reported (Lewandowski et al., 2011) and to establish if specific cognitive remediation tasks can have an impact on outcome as suggested by Ryan et al. (2013). Furthermore, it would be of interest to study high risk populations for psychoses (HR) as in Whitney's study 2013, which reported in youths at high risk of BD a significant impairment in social reciprocity possibly due to innate differences in brain development governing socio-emotional functioning or to disruptions in normal development caused by mood regulation difficulties. The assessment of SC, besides traditional neuropsychological tests, could provide new insight into major psychoses, perhaps contributing to understanding the neural basis of these disorders, considering that human brain is influenced by emotions and social stimuli.

# **REFERENCES**


a thousand words, but is it saying anything important? *Curr. Psychiatry Rep*. 15, 345–356. doi: 10.1007/s11920-012-0345-0


DC: American Psychiatric Association.


syndrome. *J. Autism Dev. Disord*. 29, 407–418.


Neurocognition, social cognition, perceived social discomfort, and vocational outcomes in schizophrenia. *Schizophr. Bull*. 35, 738–747. doi: 10.1093/schbul/sbm169


*Biobehav*. *Rev.* 30, 437–455. doi: 10.1016/j.neubiorev.2005.08.001


Robertson, D., et al. (2012). Social cognition in bipolar disorder versus schizophrenia: comparability in mental state decoding deficits. *Bipolar Disord*. 14, 743–748. doi:


*Neurosurg. Psychiatry* 23, 56–62. doi: 10.1136/jnnp.23.1.56


in Schizophrenia (BACS). *Schizophr. Res*. 102, 108–115. doi: 10.1016/j.schres.2008.03.024


*Psychiatry Res*. 189, 379–384. doi: 10.1016/j.psychres.2011.04.033


frontotemporal dementia and several psychiatric disorders that appear in late adulthood. *Int. Rev. Psychiatry* 25, 159–167. doi: 10.3109/09540261.2013.769939


*Schizophr. Res*. 25, 199–209. doi: 10.1016/S0920-9964(97)00025-X


Neuropsychological functioning in euthymic bipolar disorder: a meta-analysis. *Acta Psychiatr. Scand.* 116, 17–26. doi: 10.1111/j.1600-0447.2007.01055.x


*Neuroimage* 35, 292–298. doi: 10.1016/j.neuroimage.2006.10.046

Young, R. C., Biggs, J. T., Ziegler, V. E., and Meyer, D. A. (1978). A rating scale for mania: reliability, validity and sensitivity. *Br. J. Psychiatry* 133, 429–435. doi: 10.1192/bjp.133.5.429

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 19 May 2013; accepted: 23 September 2013; published online: 17 October 2013.*

*Citation: Caletti E, Paoli RA, Fiorentini A, Cigliobianco M, Zugno E, Serati M, Orsenigo G, Grillo P, Zago S, Caldiroli A, Prunas C, Giusti F, Consonni D and Altamura AC (2013) Neuropsychology, social cognition and global functioning among bipolar, schizophrenic patients and healthy controls: preliminary data. Front. Hum. Neurosci. 7:661. doi: 10.3389/fnhum.2013.00661*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Caletti, Paoli, Fiorentini, Cigliobianco, Zugno, Serati, Orsenigo, Grillo, Zago, Caldiroli, Prunas, Giusti, Consonni and Altamura. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# What does Williams syndrome reveal about the determinants of social behavior?

# *Anna M. Järvinen\* and Ursula Bellugi*

*Laboratory for Cognitive Neuroscience, The Salk Institute for Biological Studies, La Jolla, CA, USA*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Jack Van Honk, Utrecht University, Netherlands Suzanne Avery, Vanderbilt University, USA*

#### *\*Correspondence:*

*Anna M. Järvinen, Laboratory for Cognitive Neuroscience, The Salk Institute for Biological Studies, 10010 North Torrey Pines Road, La Jolla, CA 92037-1002, USA e-mail: pasley@salk.edu*

Growing evidence on autonomic nervous system (ANS) function in individuals with Williams syndrome (WS) has begun to highlight aberrancies that may have important implications for the social profile characterized by enhanced social motivation and approach. In parallel, neurobiological investigations have identified alterations in the structure, function, and connectivity of the amygdala, as well as prosocial neuropeptide dysregulation, as some of the key neurogenetic features of WS. A recent social approach/withdrawal hypothesis (Kemp and Guastella, 2011) suggests that autonomic cardiac control may play a key role in regulating the relationship between oxytocin (OT) and social behavior. This article discusses evidence from these critical, new strands of research into social behavior in WS, to consider the extent to which data on WS may provide novel insight into the determinants of social behavior. Future research directions are suggested.

**Keywords: Williams syndrome, social motivation, social behavior, autonomic nervous system, heart rate, oxytocin, arginine vasopressin**

### **INTRODUCTION**

Elucidating the origins of human social behavior has relevance to both typical and atypical development. In this vein, the unusual social phenotype of Williams syndrome (WS) has been gaining momentum among the neuroscience community. WS provides an attractive model for social/cognitive neuroscience because the hemideletion of 25–28 genes on chromosome 7q11.23 is wellcharacterized (Korenberg et al., 2000). Further, the phenotype comprising distinct socially positive and dysfunctional behaviors that implicate several neural systems is observed with remarkable consistency. The neurocognitive profile of WS is associated with mean IQ of 50–60, with typically higher verbal than non-verbal abilities (Searcy et al., 2004; Mervis and John, 2010).

The unusual social behavior of WS spans three discrete dimensions: enhanced motivational social drive, atypical emotional sensitivity, and increased salience of social stimuli (Järvinen et al., 2013). Social limitations are underscored by paradoxes suggesting that although such individuals keenly instigate social engagement they lack the skill to sustain a conversation and make friendships (Davies et al., 1998), and while they seem socially uninhibited they suffer from diagnostically significant non-social anxiety, attentional problems, and social maladjustment (Davies et al., 1998; Leyfer et al., 2006). In short, the genetically determined expression of hypersociability of WS combines with inadequate tools and skills to navigate and act appropriately in the social world. The profile of WS raises several fascinating questions regarding the underpinnings of the enhanced social drive.

There has been a recent expansion of research into the social brain in WS (e.g., Haas and Reiss, 2012; Järvinen et al., 2013). This body of work has indicated alterations in the structure and function of the amygdala, fusiform face area (FFA), and insula. In addition, atypical connectivity between the amygdala and the FFA, the orbital-frontal regions, and the insula, as well as within the frontostriatal pathway, has been reported. At the same time, the role of the autonomic nervous system (ANS) function remains an overwhelmingly under-researched area among researchers addressing the social profile of WS. The link between the amygdala, ANS function, and subsequent social behavior is a significant one: the amygdala is critically involved in both appetitive and aversive affective processing (Aggleton, 2000) and in emotional evaluation that contributes to social behavior (Adolphs, 2009). The amygdala further mediates affective arousal (LeDoux, 2000; Laine et al., 2009), and direct amygdala stimulation results in a robust skin conductance response (SCR) in humans (Mangina and Beuzeron-Mangina, 1996). As evidence implicates aberrancies in both the amygdala (e.g., Meyer-Lindenberg et al., 2005; Haas et al., 2009; Haas and Reiss, 2012) and ANS responsivity (e.g., Doherty-Sneddon et al., 2009; Plesa Skwerer et al., 2009; Järvinen et al., 2012; Riby et al., 2012a) in WS, the aim of this mini-review is to examine the extent to which ANS function may contribute to the characteristic social behavior of WS. We begin by briefly discussing the role of the ANS function and its regulation by prosocial neuropeptides in social–emotional behavior generally, followed by a review of the relevant literature on WS. We will discuss how the landmark social characteristics of WS converge with the ANS features, to determine the extent to which WS may offer insight into the origins of social behavior.

#### **ANS FUNCTION AND SOCIAL BEHAVIOR**

The postulated relationship between sociability and ANS function reflects an old idea: for example, in the 1960s, Eysenck hypothesized that individual differences in the cortical processing of arousal are linked to emotional experience and social behavior. Specifically, whereas extraverted individuals are characterized by chronic under-arousal, which leads them to actively seek out stimulation (e.g., social engagement), introverted individuals display the opposite pattern of both ANS arousal and subsequent behavior (Eysenck, 1967, 1994, 1997, but see Beauducel et al., 2006). Thus, relative to introverts, extraverts have been described as inherently less aroused and arousable (Stelmack, 1990; Smith, 1994); exhibit decreased heart rate (HR) reactivity (Smith et al., 1995); lower skin conductance levels (SCL) (Smith et al., 1986); reduced phasic SCR (Smith et al., 1990); and faster electrodermal habituation (Smith et al., 1995).

Honing in on the role of the ANS in human sociability, the polyvagal theory (Porges, 2003, 2007; Porges and Furman, 2011) posits that specifically autonomic cardiac control is critically implicated in social behavior and attachment. An evolutionarily important dynamic regulatory system enables adaptive responses: when under threat, the "vagal brake" is released reflecting survival-promoting energy consumption. In contrast, the ANS promotes positive approach-related behaviors during secure times. The neural circuit known as the social engagement system, which is under cortical regulation, comprises a key component of the social ANS (Porges, 2007). The heart is innervated in a dual fashion by both the sympathetic and parasympathetic branches of the ANS, with an acceleration in HR being linked to greater sympathetic influence, and a decrease to greater parasympathetic involvement. Consequently, HR variability (HRV) is regarded as a direct index of parasympathetic NS activity (Bernston et al., 2005). Indeed, it has been hypothesized that resting state HRV is a biomarker reflecting an individual's capacity for approachrelated motivations for social interaction (Kemp et al., 2012a,b; Patriquin et al., 2013). For example, autism is associated with decreased HRV (Bal et al., 2010), and higher baseline HRV amplitudes have been linked to improved social behavior and receptive language abilities in such individuals (Patriquin et al., 2013). The link between social–emotional behavior and autonomic cardiac control is thought to lie in the abundant connectivity between brain regions modulating ANS activity and emotion perception (Smith and DeVito, 1984; Thayer et al., 2009). Indeed, this psychophysiological biomarker is a useful research tool since the key aspect of social behavior, the motivation to approach or withdraw, may not always be overt and observable (Kemp et al., 2012b).

A further rationale for focusing on the ANS function in WS in an attempt to illuminate the underpinnings of its unusual social– emotional behavior comes from a recent study implicating the endogeneous dysregulation of prosocial neuropeptides, oxytocin (OT), and arginine vasopressin (AVP), in the social phenotype of WS (Dai et al., 2012). More specifically, this investigation reported increased baseline OT levels together with increased OT and AVP responses to emotional stimulation, in individuals with WS contrasted with typical controls (Dai et al., 2012). A contrasting profile is reported in autism, characterized by low plasma OT levels (Modahl et al., 1998). These hormones are proposed to play a key role not only in transient social behaviors but also in broader states and orientations, such as anxiety, social motivation, and the salience of social stimuli (Churchland and Winkielman, 2012). The association between ANS function and social behavior is underscored by recent evidence suggesting the mediating effect of OT. Specifically, according to a recent social approach/avoidance hypothesis (Kemp and Guastella, 2011; Quintana et al., 2013), OT increases social approach behaviors and may either be adaptive or maladaptive. The paraventricular and optical nuclei of the hypothalamus are responsible for the synthesis of OT, with direct OT projections to the dorsal brain stem, which is vital for cardiac regulation (Buijs et al., 1978). OT receptors are widespread in the central and peripheral nervous system (NS), with pronounced concentrations in brain regions critically implicated in complex social behaviors (Landgraf and Neumann, 2004). Neuroimaging data pinpoint contingencies between the effects of OT and the nature of the stimulus: OT decreases amygdala responses for fearful faces, while increasing responses for happy faces (Gamer et al., 2010). Autonomic control may also be mediated by OT via its actions on the amygdala, which expresses OT receptors in high density (Tribollet et al., 1992), and mediates intricate ANS responses (Davis and Whalen, 2001). The theory of Kemp and Guastella (2011) is ultimately congruent with the polyvagal theory (Porges, 2007): increased HRV following extraneous OT administration is observed (Kemp et al., 2012a,b), and the socially withdrawn predisposition of autism is associated with reduced HRV (Kemp et al., 2010). Animal studies have also suggested the link between OT and HRV (Grippo et al., 2009). Further support to the link between ANS function and OT is provided by findings suggesting that intranasal OT administration elicits pupil dilation, which has been suggested to promote approach behaviors (Wiseman and Watt, 2010). The exact mechanism via which OT influences central brain structures implicated in autonomic cardiac control or social cognition is currently poorly understood (Quintana et al., 2013). However, as new evidence may suggest alterations in social reward, social salience, and social motivational functions in WS (Dai et al., 2012), in light of the above literature, the ANS emerges as an attractive candidate for aspects of the altered social–emotional behaviors associated with WS.

## **LINKING SOCIAL BEHAVIOR WITH ANS FUNCTION IN WS**

WS is characterized by a robustly established increased appetitive drive toward social interaction (see Järvinen-Pasley et al., 2008, for a review). Hallmark features of this characteristic include an unusually gregarious, friendly, un-shy, and peopleoriented personality (Klein-Tasman and Mervis, 2003), increased attraction specifically toward unfamiliar people (Bellugi et al., 1999; Doyle et al., 2004), and a bias toward viewing faces and eyes (Mervis et al., 2003; Riby and Hancock, 2008). Thus, social information appears atypically salient for individuals with WS, manifesting as an attentional bias toward social over non-social stimuli (e.g., Järvinen-Pasley et al., 2008; Riby and Hancock, 2009a,b), as well as more competent cognitive processing of social than non-social stimuli (Järvinen-Pasley et al., 2010). Taken at face value, these behavioral features may implicate ANS responsivity patterns in WS that correspond to the extraverted personality profile, increased HRV, and elevated plasma levels of OT, indexing increased approach-related motivation and heightened salience of social stimuli (Eysenck, 1967; Porges, 2007; Kemp and Guastella, 2011). As will become apparent below, studies addressing HR and/or electrodermal activity (EDA) in WS are sparse, and have produced mixed results. The aim of the section below is to determine the extent to which the social behavioral profile of WS appears in tune with what is known about the underlying ANS function in the syndrome.

#### **EDA-BASED FINDINGS ON ANS FUNCTION IN WS**

Initial evidence suggested reduced autonomic arousal to face stimuli in individuals with WS (Plesa Skwerer et al., 2009). In this study, participants with WS, CA-matched TD controls and those with intellectual disabilities with were presented with dynamic faces expressing anger, disgust, fear, happiness, sadness, surprise, and neutral expression, while SCR and HR were monitored (Plesa Skwerer et al., 2009). However, as a control condition included neutral nature scenes, SCRs to social stimuli were increased relative to the non-social stimuli. Moreover, a subsequent study hinted that the finding that suggested hypoarousal to faces in WS may reflect the artificial nature of the face stimuli: the stimuli used by Riby et al. (2012a) incorporated both live and videomediated displays of happy, sad, and neutral faces. Results showed that while video-mediated faces failed to increase the SCL in individuals with WS, live faces elicited the typically observed increase in arousal. Further, lower than typical SCLs were reported in participants with WS, which were interpreted as reflecting general hypoarousal in WS. Doherty-Sneddon et al. (2009) measured changes in SCR in individuals with WS and CA-matched TD controls during arithmetic tasks varying in both complexity and the degree of eye contact with the experimenter. Another task assessed the degree of gaze aversion related to cognitive load. The results indicated that while individuals with WS showed general hypoarousal and reduced gaze aversion in the naturalistic, live social interaction context, similar to the TD controls, their arousal levels elevated in response to face stimuli. This led Doherty-Sneddon et al. (2009) to suggest that atypically low general arousal level (Plesa Skwerer et al., 2009; Riby et al., 2012a) may underlie the tendency of individuals with WS to hold gaze for extended periods. At the same time, eye contact during cognitive processing leads to the typical decline in performance also in individuals with WS (Riby et al., 2012b), suggesting that holding direct gaze is taxing for such individuals as well. The finding of general hypoarousal in WS indeed appears consistent with that linked to the extraverted personality profile (Eysenck, 1967), as is that of reduced SCRs to social stimuli (Plesa Skwerer et al., 2009). The only significant EDA-related finding reported by Järvinen et al. (2012) showed a lack of typical habituation to faces in individuals with WS, indexing increased novelty value of face stimuli. In the visual component of the study, adults with WS and CA-matched TD individuals were presented with static images of happy, fearful, and neutral faces and non-social scenes. The authors suggested that the absence of habituation to faces may provide an ANS correlate for the increased interest in face stimuli observed in WS, as faces may appear atypically novel and original despite the repeated exposure in everyday life. This feature may thus contribute to the increased approach-related motivation in WS.

#### **CARDIAC-BASED FINDINGS ON ANS FUNCTION IN WS**

Plesa Skwerer et al. (2009) reported increased interest in faces in individuals with WS, on the basis of findings of increased HR deceleration to such stimuli. This finding is consistent with the WS social profile. By contrast, utilizing more complex HRderived analyses than those in the previous studies, Järvinen et al. (2012) found a general acceleration in mean HR for face stimuli in individuals with WS as compared to TD controls, together with decreased HRV to such stimuli. These results suggest increased emotional reactivity to the affective face stimuli in WS, as vagal control was diminished for social–affective information. This ANS profile is in fact in line with that associated with social anxiety (Elsesser et al., 2006; Wieser et al., 2009). This is surprising in light of findings that WS is specifically associated with anxiety that is non-social in nature (Leyfer et al., 2006). At the same time, approach-related motivation is also associated with increased autonomic arousal (Pönkänen and Hietanen, 2012). In the auditory modality, happy, fearful, and sad vocal relative to musical emotional stimuli elicited increased HRV in participants with WS only, suggesting reduced arousal to auditory social information. This pattern is in contrast to that reported in the visual domain. Additionally, WS was characterized by greater HRV as compared to the TD controls. Järvinen et al. (2012) interpreted the results to suggest that human vocalizations appeared more engaging than the music stimuli for individuals with WS, as HR deceleration reflects increased focused attention. Across the visual and auditory modalities, WS was further associated with elevated HRV to happy stimuli. This result indexing greater vagal involvement is in line with the positive bias frequently documented in individuals with WS (Dodd and Porter, 2010), as positively valenced emotional stimuli are specifically socially engaging promoting approach-related motivations (Porges, 2007).

#### **PUPIL DILATION AS AN INDEX OF ANS ACTIVITY IN WS**

Studies quantifying pupil dilation in WS have reported attenuated pupil dilation in response to social stimuli in such individuals relative to CA and mental age (MA) matched TD participants, suggesting decreased ANS arousal to social information (Plesa Skwerer et al., 2011). In this study, participants were presented with social and non-social images, and notably, all groups exhibited increased arousal to the social as compared to non-social visual stimuli. The participants with WS also showed reduced pupil dilation to negative facial expressions as compared to controls. This finding is consistent with both behavioral and neurobiological reports indicating insensitivity to negative social information in individuals with WS (Meyer-Lindenberg et al., 2005; Haas et al., 2009; Santos et al., 2010), a feature that is thought to contribute to the increased affiliation with unfamiliar people in WS. Taken together, the ANS findings suggest a complex pattern of ANS function indexed by EDA, cardiovascular reactivity, and pupil dilation, underpinning the social profile of WS.

#### **PROSOCIAL NEUROPEPTIDES AND ANS FUNCTION IN WS**

In this section, we attempt to consolidate the ANS data on WS with some relevant findings on OT and AVP. In the context of the broader literature on prosocial neuropeptides, the findings of elevated base line levels as well as peak release of OT and AVP to emotional stimulation in WS relative to TD (Dai et al., 2012) appear consistent with the social profile of WS that is associated with increased approach and proclivity toward engaging the eyes, as well as maladaptive behaviors. Importantly, Dai et al. (2012) reported a positive association between basal OT level and approach, and a negative correlation with adaptive social behaviors, for individuals with WS, suggesting that some aspects of the increased OT indeed are maladaptive. The finding linking intranasal OT administration to pupil dilation (Wiseman and Watt, 2010) appears surprising in light of the data of Plesa Skwerer et al. (2011) indicating reduced pupil dilation in WS, as perhaps the opposite could have been expected. Intranasal OT administration has also been suggested to be associated with increased HRV (Kemp et al., 2012a,b). Järvinen et al. (2012) reported decreased HRV within the visual domain, and increased HRV within the auditory domain, in individuals with WS, suggesting context-dependent or unstable HRV in WS. In the study of Dai et al. (2012), no significant associations between HR and blood pressure measures and neuropeptide function were observed, also suggesting a complex mechanism in WS. Future studies should thus establish HRV in WS in the resting state. Further, studies employing sensitive cardiac indices of ANS function in WS are acutely needed to clarify the inconsistencies in the current literature, and to allow the data to be linked to theories of social behavior. At the same time, the existing evidence may reflect some degree of heterogeneity in ANS function in WS, which may be further exacerbated by the fact that individuals with WS commonly present with hypertension and cardiac abnormalities (Pober, 2010), which may impact ANS function. In a similar vein, Dai et al. (2012) noted in their study that OT and AVP function was variable in their sample of individuals with WS.

## **DETERMINANTS OF SOCIAL BEHAVIOR: INSIGHTS FROM WS**

The picture of ANS function that is emerging from investigations of individuals with WS suggest that virtually in all studies, the typical elevation in arousal in response to (live) face stimuli in such individuals is present, despite the fact that baseline arousal levels may appear atypically low. This finding is typically seen in EDA-based analyses, while cardiac-based indices indicated

#### **REFERENCES**


*J. Autism Dev. Disord.* 40, 358–370. doi: 10.1007/s10803-009- 0884-3


hyperarousal to faces in WS (Järvinen et al., 2012). Thus, the evidence does not suggest hyporesponsivity to faces in WS *per se*. Further, individuals with WS were found to lack the typical habituation effect to face stimuli, suggesting that social information may retain its originality for those with the syndrome. Evidence further supported the uneven patterns of neural and behavioral responsivity across positive (preserved) vs. negative (compromised) social information (e.g., Haas et al., 2009) in WS, as such individuals demonstrated diminished arousal as indexed by pupil dilation to negative facial expressions (Plesa Skwerer et al., 2011), while within both visual and auditory social domains, increased HRV to happy stimuli was evident. This constellation of evidence fits in well with the social-behavioral characteristics of WS.

Future studies should determine the degree of heterogeneity within the WS population with respect to ANS function by testing sizeable sample of participants; this is crucial for being able to ultimately map social–emotional profiles in terms of behavior, and neural and hormonal characteristics, onto patterns of ANS function reliably. Contributing factors to some of the inconsistencies in the existing, scarce literature may include differences in experimental paradigms (ranging from arithmetic tasks to static/dynamic displays of affective faces), age ranges of participants, whether ANS activity was assessed using EDA vs. HR derived measures, and whether the effects of endogeneous vs. extrageneous OT were measured (cf. Churchland and Winkielman, 2012). Of the studies addressing ANS function in WS, only Järvinen et al. (2012) utilized indices of HRV, allowing more direct comparisons with the tenets of the polyvagal theory (Porges, 2007) and the social approach/avoidance hypothesis (Kemp and Guastella, 2011). Nevertheless, the evidence discussed in this article highlights that the study of ANS function in tandem with neuropeptide systems promises to open up an exciting avenue for the quest toward understanding the underpinnings of the social behavior of WS, including its positive as well as maladaptive features. Such studies may also prove helpful in identifying sensitive areas for intervention.

# **ACKNOWLEDGMENTS**

This research was supported by NICHD 033113, NINDS 22343, and The Oak Tree Philanthropic Foundation to Ursula Bellugi. We warmly thank all the participants, their families, and the Williams Syndrome Association for their generous cooperation.


behavior with oxytocin: how does it work. What does it mean? *Horm. Behav.* 61, 392–399. doi: 10.1016/j.yhbeh.2011.12.003


difficulties. *Br. J. Psychiatry* 172, 273–276. doi: 10.1192/bjp.172.3.273


social stimuli in Williams syndrome. *J. Neurosci.* 29, 1132–1139. doi: 10.1523/JNEUROSCI.5324-08. 2009


implications for cardiovascular risk. *PLoS ONE* 7:e30777. doi: 10.1371/journal.pone.0030777


cognition in Williams syndrome. *Nat. Neurosci.* 8, 991–993. doi: 10.1038/nn1494


decreased detection of angry faces in children with Williams syndrome. *Neuropsychologia* 48, 1071–1078. doi: 10.1016/ j.neuropsychologia.2009.12.006


and arousal: effects of attentional conditions on electrodermal activity. *Pers. Individ. Dif.* 7, 293–303. doi: 10.1016/0191-8869 (86)90004-8


eye to eye contact really threatening and avoided in social anxiety: an eye-tracking and psychophysiology study. *J. Anxiety Disord.* 23, 93–103. doi: 10.1016/j.janxdis.2008. 04.004

Wiseman, R., and Watt, C. (2010). Judging a book by its cover: the unconscious influence of pupil size on consumer choice. *Perception* 39, 1417–1419. doi: 10.1068/p6834

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 11 June 2013; published online: 28 June 2013.*

*Citation: Järvinen AM and Bellugi U (2013) What does Williams syndrome reveal about the determinants of social behavior? Front. Hum. Neurosci. 7:321. doi: 10.3389/fnhum.2013.00321*

*Copyright © 2013 Järvinen and Bellugi. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Integrating intention and context: assessing social cognition in adults with Asperger syndrome

# *Sandra Baez 1,2,3†, Alexia Rattazzi 4, María L. Gonzalez-Gadea1,2, Teresa Torralva1, Nora Silvana Vigliecca2,5, Jean Decety6, Facundo Manes <sup>1</sup> and Agustin Ibanez 1,2,7\**

*<sup>1</sup> Institute of Cognitive Neurology and Institute of Neuroscience, Favaloro University, Buenos Aires, Argentina*

*<sup>5</sup> Research Centre of the Faculty of Philosophy and Humanities, National University of Córdoba, Córdoba, Argentina*

*<sup>6</sup> Departments of Psychology and Psychiatry, and Center for Cognitive and Social Neuroscience, University of Chicago, Chicago, IL, USA*

*<sup>7</sup> Laboratory of Cognitive Neuroscience, Universidad Diego Portales, Santiago, Chile*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Chiara Fiorentini, University College London, UK Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland Tiziana Zalla, CNRS, France*

#### *\*Correspondence:*

*Agustin Ibanez, Laboratory of Experimental Psychology and Neuroscience, Institute of Cognitive Neurology and National Scientific and Technical Research Council, Pacheco de Melo 1860, Buenos Aires, Argentina. e-mail: aibanez@ineco.org.ar*

*†This work is part of the master dissertation (Baez S) ongoing by the author on the Italian Hospital at Buenos Aires, Argentina.*

Deficits in social cognition are an evident clinical feature of the Asperger syndrome (AS). Although many daily life problems of adults with AS are related to social cognition impairments, few studies have conducted comprehensive research in this area. The current study examined multiple domains of social cognition in adults with AS assessing the executive functions (EF) and exploring the intra and inter-individual variability. Fifteen adult's diagnosed with AS and 15 matched healthy controls completed a battery of social cognition tasks. This battery included measures of emotion recognition, theory of mind (ToM), empathy, moral judgment, social norms knowledge, and self-monitoring behavior in social settings. We controlled for the effect of EF and explored the individual variability. The results indicated that adults with AS had a fundamental deficit in several domains of social cognition. We also found high variability in the social cognition tasks. In these tasks, AS participants obtained mostly subnormal performance. EF did not seem to play a major role in the social cognition impairments. Our results suggest that adults with AS present a pattern of social cognition deficits characterized by the decreased ability to implicitly encode and integrate contextual information in order to access to the social meaning. Nevertheless, when social information is explicitly presented or the situation can be navigated with abstract rules, performance is improved. Our findings have implications for the diagnosis and treatment of individuals with AS as well as for the neurocognitive models of this syndrome.

**Keywords: Asperger syndrome, contextual social cognition, executive functions, individual variability**

# **INTRODUCTION**

Social cognition refers to specific information processing involved in the successful navigation of challenges related to survival and reproduction in social species (Adolphs, 1999). The construct of social cognition involves several domains, including emotional processing, theory of mind (ToM), decision-making, empathy, moral judgment, and social norms knowledge, among others. Despite the seemingly differences in these components, some of them require similar underlying processes. Multiple social cognition domains require the spontaneous perception of the relevant social elements of the situation and the interpretation of how these elements create a given social context (Klin, 2000), which depends on the implicit inference of contextual clues that bias the social meaning of an action (Ibáñez and Manes, 2012). For example, emotional recognition of a face usually occurs within a background that includes emotional body language and other convergent information such as prosody, gestures, and situational clues. In contrast, other processes may require the use of explicit and abstract rules about the general social setting in terms of conventions or expected behaviors (e.g., explicit social norms during specific social interactions). Thus, different strategies underlie the different social cognition domains. Here, we investigate different aspects of social cognition in adults with Asperger syndrome (AS).

AS is a pervasive developmental disorder characterized by severe and sustained impairments in social interaction and the development of restricted, repetitive patterns of behavior, interest, and activities. These disturbances must cause significant impairments in social, occupational, or other important areas of functioning (American Psychiatric Association, 1994; Matson and Wilkins, 2008). AS may be distinguished from autistic disorder by a lack of delay in early language development (Baron-Cohen et al., 2005). Because the main focus has been on early recognition and diagnosis, this syndrome has primarily been studied in children. However, given that AS is a chronic lifelong condition and nuclear symptoms persist, research in adults has recently received particular attention (Fombonne and Tidmarsh, 2003; Lugnegard et al., 2011).

*<sup>2</sup> National Scientific and Technical Research Council, Buenos Aires, Argentina*

*<sup>3</sup> Pontifical Catholic University of Argentina, Buenos Aires, Argentina*

*<sup>4</sup> Argentinean Program for Children, Adolescents and Adults with Autism Spectrum Disorders (PANAACEA), Buenos Aires, Argentina*

Recent reports suggest that adults with AS exhibit deficits in multiple social cognition domains including face recognition, emotional processing, ToM, empathy, and moral judgment (see below). Nevertheless, previous studies have not taken into account several factors that should be considered simultaneously in the social cognition research of these individuals. These factors include: (1) the simultaneous assessment of multiple social cognition domains, (2) the sample selection, (3) the assessment of executive functions (EF), and (4) the cognitive heterogeneity of the AS. In the present study, we considered all of these aspects, which are essential for establishing the underlying factors that contribute to the social cognition deficits of adults with AS.

### **SOCIAL COGNITION DISTURBANCES IN ADULTS WITH AS**

Emotional processing is an emerging topic of interest. There are numerous reports of individuals with autism spectrum disorders [autism, high functioning autism (HFA), and AS] being impaired in both recognition (Hobson et al., 1988; Ashwin et al., 2006; Hubert et al., 2007; Atkinson, 2009) and production of emotional expressions (Macdonald et al., 1989). Studies focused in adults with AS (Philip et al., 2010) show deficits on emotion recognition from faces [especially negative emotions (Ashwin et al., 2007; Falkmer et al., 2011)]. Thus, evidence suggests that emotional processing is affected in AS and other autism spectrum disorders.

ToM is another area of interest in AS research, since it requires the ability to infer the beliefs, intentions, and emotions of others (Baron-Cohen et al., 1985). Adults with AS have difficulty understanding the intentions (cognitive ToM) and emotional impact of others' actions (affective ToM) as assessed by the Faux Pas Test (FPT) (Zalla et al., 2009). However, reports of adults with AS with the reading the mind in the eyes test (RMET) have shown impaired (Baron-Cohen et al., 1997; Baron-Cohen et al., 2001) and preserved performance (Roeyers et al., 2001; Ponnet et al., 2004; Spek et al., 2010). These controversial results have been explained by the features of the RMET since correlations between RMET and other ToM measures are weak (Luzzatti et al., 2002; Spek et al., 2010).

Impairments in empathy, the capacity to share and understand the emotional states of others in reference to oneself (Decety and Moriguchi, 2007), are also a feature of the AS. Nevertheless, few studies have examined empathy in adults with AS. The majority of the studies (Baron-Cohen and Wheelwright, 2004; Rogers et al., 2007) have focused on self-report questionnaires. However, other reports (Dziobek et al., 2008) have represented an experimental assessment of empathy in adults with AS. These studies show that these patients are impaired in cognitive empathy but do not differ from controls in emotional empathy.

Finally, one study recently reported that participants with AS and HFA participants exhibit specific impairments in moral judgment. Participants made atypical moral judgments when they needed to consider the intention of harm (accidental vs. intentional) and the outcome (neutral vs. negative) of a person's actions (Moran et al., 2011). These participants were unable to judge the moral difference between accidental and attempted harms.

#### **RELEVANT FACTORS IN AS SOCIAL COGNITION RESEARCH**

As we mentioned above, to establish the underlying factors that contribute to the social cognition deficits of adults with AS, it is essential to consider several factors. First, to explore the social cognitive deficits in adults with AS, it is important to examine multiple domains with different tasks. Implicit social cognition tasks would require the spontaneous perception of the relevant contextual elements of the situation (Klin, 2000). Conversely, in explicit social cognition tasks the elements of the situation are clearly defined and these can usually be solved with relatively abstract and universal rules learned by explicit knowledge. Individuals with AS fail when they need to spontaneously apply social reasoning abilities to solve more naturalistic tasks, but when explicit information is provided, they improve the performance (Klin, 2000; Senju et al., 2009; Izuma et al., 2011). Thus, to assess several social cognition domains with different contextual clues involvement allows for a more comprehensive evaluation, and it makes it possible to establish whether there is a common factor that explains the adults with AS social cognition deficits. However, until now, only a few studies have simultaneously tested more than one social cognition domain.

Furthermore, most of previous social cognition reports (Baron-Cohen et al., 2001; Baron-Cohen and Wheelwright, 2004; Moran et al., 2011; Zalla et al., 2011) have included subjects diagnosed with AS and patients with other autism spectrum disorders (e.g., HFA). Therefore, the findings of these investigations can be biased by the sample selection. There is an ongoing debate about the differentiation among autistic subtypes, especially between AS and HFA. According to the DSM-IV criteria (American Psychiatric Association, 1994) for autism, not for AS, delay in language and qualitative impairments in communication must be evident. However, several studies suggest that there is not only a difference in language abilities among HFA and AS (for a review see Matson and Wilkins, 2008). Unlike HFA, individuals with AS do not have delay in early cognitive functioning (Frith, 2004). Furthermore, AS compared to HFA individuals have more accentuated visual-motor deficits (Klin et al., 1995; Noterdaeme et al., 2010), less strong impairments in verbal comprehension (Noterdaeme et al., 2010; Planche and Lemonnier, 2012), higher verbal than performance IQ (Klin et al., 1995) and less severe behavioral abnormalities (Gilchrist et al., 2001). These evidences suggest that both of these disorders should be studied as separate diagnostic entities (Matson and Wilkins, 2008).

On the other hand, EF are required for the processing of emotional stimuli and social cognition tasks (Pessoa, 2008; Uekermann et al., 2010). Emotional processing requires holding stimuli in the working memory, and irrelevant information needs to be inhibited. In the same vein, ToM and empathy entail holding information in the working memory and switching between one's own perspective and that of another person (Uekermann et al., 2010). Nevertheless, no studies on adults with AS have controlled for the effect of EF on social cognition performance.

Finally, adults with AS perform variably among multiple domains (Hill and Bird, 2006; Towgood et al., 2009). This variability is observed more frequently in EF but also in social cognition. Deficits in working memory, cognitive flexibility and inhibitory control have been reported (Morris et al., 1999; Ambery et al., 2006; Hill and Bird, 2006), while other studies in adults with AS (Just et al., 2007; Nyden et al., 2010) have found preserved executive functioning. Affected (Baron-Cohen et al., 1997; Zalla et al., 2009) and intact performances (Ponnet et al., 2004; Spek et al., 2011) on ToM tasks have also been reported. These mixed findings suggest that patterns of deficits vary from individual to individual and that the adults with AS population include patients with both sub-normal and supra-normal performance. Thus, AS is more likely to be associated with a complex pattern of deficits across and within domains rather than just a single primary processing deficit (Happe et al., 2006). The heterogeneity in AS individuals has been interpreted as an obstacle to research (Happe et al., 2006). Traditional group-study type of analysis is problematic for individuals with high variability in performance because of the *averaging artifact* (Shallice and Evans, 1978).

### **THE GOAL OF THIS STUDY**

The primary goal of this study was to examine the performance of adults with AS on multiple social cognition domains with different levels of contextual integration while assessing the influence of EF. The social cognition domains evaluated were emotion recognition, ToM, empathy, moral judgment, social norms knowledge, and self-monitoring behavior in social settings. We included some tasks that require the implicit perception and integration of the relevant social elements to solve a social situation, and other in which the elements of the situation are explicitly defined and can be solved with relatively abstract and universal learned rules. In adittion, we explored the individual variability in the AS group. For this purpose, we employed a methodology called *multiple case series analysis* (MCSA) (Hill and Bird, 2006; Towgood et al., 2009), that detects the domains in which a given individual displays an abnormal performance. Group comparison analyses requires homogeneity between subjects; however, individuals with AS exhibit performance variability, which is concealed in these analyses. Therefore, the lack of significant differences is not necessarily an index of intact performance in this population (Hill and Bird, 2006).

Taking previous findings into account, we predicted that adults with AS will have deficits in several social cognition domains. We hypothesized that the social cognition deficits of adults with AS would be more related to impairments in the capacity to implicitly integrate action intentions with contextual clues than to the inability to apply explicit social rules. We also hypothesized that the social cognition difficulties would not be explained by EF profiles. This hypothesis was based on the fact that deficits in social cognition seem to be a fundamental characteristic that is less affected by AS heterogeneity, while patterns of EF have shown high variability between individuals. Finally, we predicted that the MCSA should demonstrate that patterns of cognitive strengths and weaknesses vary within individuals.

## **MATERIALS AND METHODS PARTICIPANTS**

Fifteen adult's diagnosed with AS and 15 healthy subjects participated in the present study. All participants were selected from the outpatient population of the Institute of Cognitive Neurology. All adults with AS had an estimated IQ above 94 (SD ≤ 7.42). Patients were assessed by a psychiatrist and met the diagnostic and statistical manual of mental disorders (DSM-IV) criteria for AS (American Psychiatric Association, 1994). The diagnosis was made on the basis of the adult Asperger assessment (AAA) (Baron-Cohen et al., 2005). Before the clinical interview, patients are asked to complete autism spectrum quotient (AQ) and the empathy quotient (EQ) as screening questionnaires (see **Table 2**). The psychiatrist then sought to validate the symptom examples provided by the AQ and EQ and checked the other AS symptoms and criteria.

Healthy control participants matched with the adults with AS were recruited from a large pool of volunteers. No significant differences in age [*F(*1*,* <sup>28</sup>*)* <sup>=</sup> <sup>0</sup>*.*003, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*95], gender [*X*<sup>2</sup> *(*1*)* = 0.012, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*91], handedness [*X*<sup>2</sup> *(*1*)* = 0.00, *p* = 1*.*00] or years of formal education [*F(*1*,* <sup>28</sup>*)* = 1*.*36, *p* = 0*.*25] were observed between adults with AS and controls.

The following exclusion criteria were applied: (1) AS participants who met DSM-IV criteria for any axis-I diagnosis were excluded; (2) control subjects with a history of mental retardation, neurological disease, psychiatric disease, or any clinical condition that may affect cognitive performance were excluded; (3) adults with AS and controls with a history of drug or alcohol abuse were also excluded. All participants provided written informed consent in agreement with the Helsinki declaration. The study was approved by the ethics committee of Institute of Cognitive Neurology.

## **MATERIALS AND PROCEDURE**

A battery of neuropsychological tests was used to assess EF and social cognition (see below). Patients were also evaluated with the Wechsler abbreviated scale of intelligence (WASI). This scale includes vocabulary and matrix reasoning subtests and provides an estimated IQ (Weschler, 1999). All participants were individually evaluated in a quiet office of the Institute of Cognitive Neurology. A complete evaluation was administrated in one session that lasted approximately 2 h. Subjects were initially assessed with the social cognition tasks and then with the EF and intellectual level tests. The order of administration of the tasks was the same for each participant.

## *EF assessment*

All participants were evaluated with an EF battery which included measures of verbal fluency, inhibitory control, interference control, working memory, and cognitive flexibility. Verbal and design fluency tests (Delis and Kaplan, 2001) were used to assess recall, self-monitoring, and cognitive flexibility strategies. The trailmaking test (Partington, 1949) was employed to assess cognitive flexibility and processing speed, and the Hayling test (Burgess and Shallice, 1996) was used to measure inhibitory control. The Flanker test (Eriksen and Eriksen, 1974) was applied to evaluate the ability to inhibit responses to irrelevant stimuli and the executive control of attention. The set shifting task (Diamond and Kirkham, 2005) was used to assess cognitive flexibility and inhibitory control. Finally, a span counting task (Case et al., 1982) and the 1-back test (Gevins and Cutillo, 1993) were applied to evaluate working memory.

#### *Measures of social cognition*

A description of social cognition tasks is provided in **Table 1**. All participants were evaluated with a social cognition battery that included measures of emotion recognition, ToM, empathy, moral judgment, social norms knowledge, and self-monitoring behavior in social settings. The awareness of social inference test (TASIT) (McDonald et al., 2003, 2006; Kipps et al., 2009) was used to assess recognition of emotional states. This task introduces contextual cues (e.g., prosody, facial movement, and gestures) and additional processing demands (e.g., adequate speed of information processing, selective attention, and social reasoning) that are not taxed when viewing static displays. The RMET (Baron-Cohen et al., 1997) and the FPT (Stone et al., 1998) were applied to assess emotional and cognitive aspects of the ToM. An empathy for pain task (EPT; Couto et al., 2012) was employed to evaluate the empathy in the context of intentional and accidental harms. We also used the interpersonal reactivity index (IRI; Davis, 1983), a 28-item self-report questionnaire that measures both the cognitive and affective components of empathy. Finally, we included a moral judgment task (Young et al., 2010) and the revised self-monitoring scale (RSMS) (Lennox and Wolfe, 1984). A detailed description of the social cognition tasks is provided in supplementary data.

#### **Table 1 | Social cognition domain assessed and tasks employed.**


*TASIT, The awareness of social inference test; RMET, Reading the mind in the eyes test; FPT, Faux Pas test; EPT, empathy for pain task; IRI, index of interpersonal reactivity; SNQ, social norms questionnaire; RSMS, revised self-monitoring scale.*

## **DATA ANALYSIS**

The demographic, neuropsychological, and experimental data were compared between the groups using ANOVA and Tukey's HSD *post-hoc* test (when appropriate). The ANOVA results were also corrected for multiple comparisons using the Tukey's test. When analyzing categorical variables (e.g., gender), χ<sup>2</sup> square tests were applied. To control for the influence of EF on the performance on social cognition tasks, we applied an ANCOVA test that was adjusted for the cognitive flexibility score. The α value for all statistical tests was set at 0.05.

To assess individual differences, we conducted a MCSA and compared each participant with the control group on every performance measure. We followed the method of Towgood et al. (2009) and used a threshold of 2 standard deviations (SD) from the mean of the control group to define the normal range. First, we identified control subjects who displayed abnormal performance in each sub-measure, according to the 2 SD criteria, and removed them. Then, we recomputed the control means and SD excluding these subjects and identified adults with AS and control participants who were below (minus 2 SD) or above (plus 2 SD) the controls mean. We carried out frequency analyses in order to record the instances in which the performance of each subject was subnormal or supranormal. We then used non-parametric tests (Mann–Whitney tests) to compare the number of measures of impaired and supra-normal performance.

Finally, Pearson's correlations were performed to examine the association between the EF measures with the greatest variability, and the total scores on the social cognition tasks that were significantly different between groups.

# **RESULTS**

**Table 2** shows the overall results from the demographic and EF assessment.

## **EXECUTIVE FUNCTIONS ASSESSMENT**

The results showed that our groups have similar EF performance. No differences in verbal fluency, inhibitory control, interference control, or working memory were observed (**Table 2**). However, the adults with AS performed significantly lower than controls on the switching design fluency task [*F(*1*,* <sup>28</sup>*)* = 5*.*10, *p <* 0*.*05], suggesting subtle cognitive flexibility impairments. Given these results, we considered this measure as a covariate in the social cognition performance analysis.

#### **MEASURES OF SOCIAL COGNITION**

**Figure 1** summarizes the significant differences between groups.

#### *Recognition of emotional states*

No significant differences in the TASIT total score were observed [*F(*1*,* <sup>28</sup>*)* = 0*.*69, *p* = 0*.*41]. The per category analysis showed significant differences between groups [*F(*4*,* <sup>108</sup>*)* = 7*.*97, *p <* 0*.*01]. A *post-hoc* analysis (Tukey HSD, *MS* = 0.49, *df* = 134.13) revealed that adults with AS had difficulty with disgust categorization (*p <* 0*.*01). This effect was preserved (*p <* 0*.*01) after co-varying for cognitive flexibility (*p* = 0*.*35). No significant differences were observed for anger (*p* = 1),

#### **Table 2 | Demographic and executive functions assessment.**


*Note: The results are shown as the mean (SD). Statistical results are shown in the right column. Significant differences are in bold.*

*TMT, Trail Making Test; WAT, Word accentuation test.*

fear (*p* = 0*.*22), sadness (*p* = 0*.*11) or surprise (*p* = 0*.*74) categorization.

#### *Theory of mind*

For the ToM measures, the adults with AS scored significantly lower than controls on the FPT total score [*F(*1*,* <sup>28</sup>*)* = 20*.*62, *p <* 0*.*01]. This result did not change (*p <* 0*.*01) after adjusting for cognitive flexibility (*p* = 0*.*15). Significant differences were also observed on the hits [*F(*1*,* <sup>28</sup>*)* = 20*.*62, *p <* 0*.*01]. Differences were preserved (*p <* 0*.*01) after co-varying for cognitive flexibility (*p* = 0*.*13). The AS group also showed lower intentionality scores [*F(*1*,* <sup>28</sup>*)* = 74*.*21, *p <* 0*.*01]. This effect was preserved (*p <* 0*.*01) in the covariate analysis (*p* = 0*.*41). Furthermore, adults with AS scored lower on emotional attribution [*F(*1*,* <sup>28</sup>*)* = 29*.*08, *p <* 0*.*01]. This effect was maintained (*p <* 0*.*01) after adjusting for the covariate (*p* = 0*.*43). No significant differences were observed on the reject scores [*F(*1*,* <sup>28</sup>*)* = 0*.*007, *p* = 0*.*93].

Scores on IRI subscales. EC, empathic concern; PD, personal distress; PT, perspective taking; F, fantasy. **(D)** Empathy for pain task, ratings for Scores on RSMS subscales. SEBO, sensitivity for expression behavior of others; AMSP, ability to modify self-presentation; DMSP, difficulty to modify self-presentation.

No differences between the groups were observed on the RMET [*F(*1*,* <sup>28</sup>*)* = 0*.*09, *p* = 0*.*76].

#### *Empathy*

*Empathy for pain task.* The ratings of empathic concern were significantly different between groups [*F(*2*,* <sup>52</sup>*)* = 6*.*70, *p <* 0*.*01]. A *post-hoc* analysis (Tukey HSD, *MS* = 10*.*62, *df* = 55*.*08) revealed that the adults with AS rated the intentional pain situations with lower scores (*p <* 0*.*01), even controlling for cognitive flexibility (*p* = 0*.*65). Furthermore, the controls rated greater empathic concern for intentional harm situations than accidental harm situations (*p <* 0*.*01). However, this difference was not observed in the adults with AS. Moreover, significant group differences were observed in the punishment ratings [*F(*2*,* <sup>52</sup>*)* = 7*.*02, *p <* 0*.*01]. The *post-hoc* comparisons (Tukey HSD, *MS* = 6*.*87, *df* = 66*.*7) showed that the adults with AS tended to rate intentional harm situations with lower scores than controls (*p* = 0*.*06). This tendency did not change (*p* = 0*.*06) in the covariate analysis (*p* = 0*.*93). No differences were observed in the judgments of discomfort, intention to harm or correctness.

In addition, the RTs of the discomfort judgments were different between groups [*F(*2*,*52*)* = 4*.*72, *p <* 0*.*05]. The RTs of the discomfort judgments were longer for the intentional harm than the neutral (*p <* 0*.*01) and accidental (*p <* 0*.*05) harm situations. These differences were preserved (*p <* 0*.*05) in the covariate analysis (*p* = 0*.*17).

*IRI.* Adults with AS scored higher on PD subscale [*F(*1*,* <sup>28</sup>*)* = 6*.*02, *p <* 0*.*05] than controls. This effect was preserved (*p <* 0*.*05) after adjusting for the covariate (*p* = 0*.*60). No difference between groups [*F(*1*,*28*)* = 1*.*96, *p* = 0*.*17] were observed on the EC subscale. Furthermore, the AS group tended to have lower scores than controls [*F(*1*,*28*)* = 4*.*01, *p* = 0*.*055] on the PT subscale. This tendency was true (*p <* 0*.*01) after controlling for cognitive flexibility (*p* = 0*.*09). No difference between the groups was observed [*F(*1*,* <sup>28</sup>*)* = 0*.*17, *p* = 0*.*67] on the *F*-subscale.

## *Moral judgment*

In both groups, actions with neutral intentions [*F(*1*,* <sup>28</sup>*)* = 146*.*29, *p <* 0*.*01] and neutral outcomes [*F(*1*,* <sup>28</sup>*)* = 24*.*55, *p <* 0*.*01] were judged to be more permissible than actions with negative intentions and negative outcomes. Accidental harm was judged as being more permissible than intentional harm (Intention × Outcome Interaction) [*F(*1*,* <sup>28</sup>*)* = 7*.*40, *p <* 0*.*01]. The group × intention × outcome interaction [*F(*1*,* <sup>28</sup>*)* = 1*.*60, *p* = 0*.*21] was not statistically significant. Therefore, the adults with AS and controls did not differ in their judgments of morality. Specifically, the judgments of the neutral (neutral outcome, neutral intent), attempted harm (neutral outcome, harmful intent), accidental harm (harmful outcome, neutral intent), or intentional harm (harmful outcome, harmful intent) vignettes did not differ between groups.

#### *Knowledge of social norms*

No differences between groups were observed in the break [*F(*1*,* <sup>24</sup>*)* = 0*.*50, *p* = 0*.*48] and over-adhere [*F(*1*,* <sup>28</sup>*)* = 0*.*00, *p* = 1*.*00] scores of the SNQ.

### *Self-monitoring behavior in social settings*

Adults with AS obtained lower scores in the sensitivity for expression behavior of others compared to controls [*F(*1*,* <sup>28</sup>*)* = 29*.*26 *p <* 0*.*01], even after the covariate (*p* = 0*.*65). Adults with AS also received lower scores on the ability to modify self-presentation [*F(*1*,* <sup>28</sup>*)* = 19*.*40, *p <* 0*.*01]. This effect remained true after the covariate analysis (*p <* 0*.*01), even though a significant effect of cognitive flexibility (*p <* 0*.*05) on self-presentation was observed.

In summary, adults with AS showed impairments on measures of disgust recognition (TASIT), ToM (FPT), and empathic concern and punishment ratings for the intentional harm situations (EPT). Additionally, the adults with AS showed higher scores on the PD subscale. They also showed lower scores on subscales of the sensitivity to the expressive behavior of others and the ability to modify self-presentation (RSMS). All differences were preserved after covarying for cognitive flexibility. Overall, adults with AS seem to perform less well in tasks that require an implicit encoding of socially relevant information and automatic context integration. Nevertheless, they performed as well as controls in tasks in which the social information was explicitly presented and when the task could be solved with abstract rules. Finally, the difficulties experienced by the adults with AS were not explained by abnormalities in EF.

#### **MULTIPLE CASE SERIES ANALYSIS (MCSA)**

To explore the intra-individual variability in tasks performance of the AS group, we examined the ranges of *z*-scores based on the performance of the control group (Towgood et al., 2009). The maximum range of performance on each of the 78 measures in controls was 4.60. Among the adults with AS, more than 43% of the measures (34/78 sub-measures) showed a *z*-score range exceeding the maximum threshold observed in controls. Specifically, 27.78% (5/18) of the EF measures exceeded the maximum range of the control group, whereas 48.33% (29/60) of the social cognition measures exceeded this range.

A greater number of adults with AS performed atypically compared with the control group. The individual performance profiles of each AS and control participants are provided in Appendix (see **Tables A1a**, **A1b**, **A2a** and **A2b**). The measures that were the most variable are detailed in **Table 3**. Most of the adults with AS performed below normal (*<*2SD below control group mean) in both, EF and social cognition measures. With regard to EF, supra-normal (*>*2SD above control group mean) performance was observed only in the phonological fluency task. They also obtained supra-normal performance on several EPT measures. More specifically, the adults with AS showed supra-normal ratings in tasks involving neutral situations (e.g., discomfort, intention to hurt, and happiness ratings). In neutral scenarios in which the actions do not involve the intention to hurting someone, one would expect lower discomfort or intention to hurt ratings. Thus, the results suggest that the adults with AS are unable to discriminate between the neutral, accidental and intentional pain situations.

Consistent with the group analysis, the MCSA revealed that the adults with AS performed less well than the controls. Inter-individual variability (subnormal performance) was observed on: FPT (60%), TASIT (26.67%), empathic concern rating of intentional pain (33.33%), PD (33.33%), sensitivity of expression behavior of others (33.33%) and ability to modify self-presentation (53.33%).



To explore the inter-individual variability, we analyzed the performance of each participant and recorded instances in which the performance was 2 SDs below or above of the control mean. A non-parametric test was applied to compare the number of measures for subnormal and supra-normal performance (see **Table 4**). As expected, the adults with AS showed a greater number of abnormal measures than controls (Mann–Whitney *U* = 19*.*00, *p <* 0*.*01). The AS participants also showed a greater number of measures in which they performed below control performance (Mann–Whitney *U* = 14*.*00, *p <* 0*.*01). However, no significant differences were observed in the number of measures with supranormal performance (Mann–Whitney *U* = 82*.*00, *p* = 0*.*21).

In summary, the MCSA showed higher variability in the performance of the adults with AS compared with controls. A larger proportion of the social cognition measures compared to the EF measures exceeded the maximum range of the *z*-scores calculated based on the control group performance. In the AS group subnormal performance was higher than supra-normal.

#### **ASSOCIATION BETWEEN EF AND SOCIAL COGNITION PERFORMANCE**

Finally, we explored the influence of EF on social cognition performance. We examined the correlation between the EF measures with the greatest variability, and the total scores on the social cognition tasks that were significantly different between groups. No significant correlations were observed.

## **DISCUSSION**

The primary goal of this study was to examine the performance of adults with AS on tasks of multiple domains of social cognition, while assessing the influence of EF. The secondary goal was to explore individual variability in adults with AS performance on both the social cognition and EF tasks. Our results suggest that participants with AS have a fundamental deficit in several domains of social cognition. We also found that the AS participants showed a greater number of social cognition measures in which they performed below controls' performance. These deficits were not explained by abnormalities in EF.

Furthermore, our data suggest that a common mechanism underlies the deficits in multiple social cognition domains in the adults with AS. In brief, these participants performed poorly on tasks (TASIT, FPT, EPT) that imply the ability to implicitly infer the intentionality of actions and those that require the integration of mental states (intentions, beliefs, emotions) with contextual information.

This is the first study in adults with AS to explore the effect of EF on social cognition performance. Both AS and control groups



were similar regarding executive functioning. Moreover, to control for the effect of EF on performance during social cognition tasks, we conducted covariance analysis adjusted for cognitive flexibility, the only domain in which we found group significant differences. All significant differences in the social cognition measures remained significant. Moreover, we did not find significant correlations between scores on the EF measures with higher variability and those of the social cognition tasks that were different between groups. Because we selected tasks that were designed to assess specifically EF and have been utilized extensively to assess these domains (Partington, 1949; Eriksen and Eriksen, 1974; Case et al., 1982; Gevins and Cutillo, 1993; Burgess and Shallice, 1996; Delis and Kaplan, 2001; Diamond and Kirkham, 2005), we consider that the failure to find significant correlations could not be explained by the lack of the sensitivity of the executive measures. Instead, the lack of significant correlations may be explained by the low variability observed in the EF performance, since both groups had a similar executive functioning and low variability. Consequently, these results indicate that EF do not seem to play a major role in the social cognition impairments of adults with AS.

## **DEFICITS IN SOCIAL COGNITION**

We employed an ecological task of contextual inference of emotional states (TASIT) which requires the integration of cues from face, prosody, gesture, and social context to identify the emotions. Consistent with previous reports (Ashwin et al., 2007; Falkmer et al., 2011), our results showed that individuals with AS have difficulty recognizing expressions of disgust. It has been shown that the basal ganglia, in parallel with the insula, are involved in disgust recognition (Calder et al., 2000; Adolphs, 2002; Wang et al., 2003; Ibáñez et al., 2010a,b). Fronto-insular networks seem to be crucial for social cognition (Couto et al., 2012). Individuals with AS show reduced gray matter in the basal ganglia (McAlonan et al., 2002; Nayate et al., 2005). They also show abnormalities in the white matter between the basal ganglia and thalamus, which connects brain areas (amygdala and fusiform gyrus) (McAlonan et al., 2009). Moreover, adults with AS present smaller volumes in the insular cortex (Kosaka et al., 2010). Therefore, the deficits in disgust recognition may be associated with abnormalities in the basal ganglia and the insula.

As previously reported (Ponnet et al., 2004; Spek et al., 2011), no differences between AS individuals and controls were found in ToM as measured by the RMET. Nevertheless, our data showed that the adults with AS performed poorly on the FPT, which is consistent with other studies (Zalla et al., 2009; Spek et al., 2011). In this test, adults with AS failed to identify the faux pas and to understand them as unintentional actions. Furthermore, they had difficulties to understand the emotional impact generated by the faux pas. The discrepancy in the performance between both ToM tests in the AS group can be explained by the features of these tasks. First, the FPT presents social scenarios resembling daily life situations. These tasks that involve real-life social scenarios are more sensitive to detect the ToM deficits of individuals with autism and AS (Klin, 2000). Furthermore, an adequate performance in the FPT involves the capacity to implicitly integrate cognitive inferences about mental states with empathic understanding. This capacity is mediated by the appraisal of contextual clues and relevant social elements provided in the scene information. Conversely, the RMET can be solved using basic and general matching strategies to correctly pair the depicted eyes and emotions. Thus, taken together, the ToM results suggest that adults with AS have difficulty integrating implicit information from the context and using this information to infer the intentionality and the emotional impact of the others' actions.

We employed a more ecologically valid measure of empathy (EPT) than the self-report questionnaires. In this task, the adults with AS showed abnormal empathic concern ratings, punishment ratings, and RTs of discomfort judgments for the intentional pain situations. Consistent with previous findings (Klin, 2000; Zalla et al., 2009), our results indicate that these individuals have difficulty with inferring the intentionality of actions. Information about intentionality allows us to decide how bad or good an action is. The deficit in intention inference may have affected the empathic concern ratings and therefore, the punishment ratings of the adults with AS.

In addition, the adults with AS showed higher levels of PD and a trend toward lower levels of PT compared with controls on the IRI. These results are supported by previous studies (Rogers et al., 2007; Dziobek et al., 2008). The high PD scores indicate greater levels of discomfort in interpersonal settings. This finding may be related to the slower RTs in the AS group for discomfort judgments in the intentional pain situations. Furthermore, individuals with AS show higher levels of anxiety (Hurtig et al., 2009; Lai et al., 2011), which may increase their PD scores. The lower scores on the PT subscale suggest that individuals with AS have difficulty understanding the feelings and perspectives of others, which is congruent with the EPT results.

In summary, the pattern of performance on the empathy measures indicated that adults with AS are impaired when using contextual information to infer the intentions of others. These deficits are reflected by lower ratings of empathic concern and punishment. Moreover, these individuals show higher levels of discomfort in stressful interpersonal situations.

Interestingly, we found that adults with AS performed similarly than control participants on measures of moral judgment. Both groups judged accidental harm as being more permissible than intentional harm. The lack of difference between groups in this task may be due to the fact that information about intention, outcome, and context (scene information) were presented in an explicit way. Therefore, it was possible to understand the moral content using two abstract rules with a linear relationship. For example, if the protagonist had the intention of harming another person (negative intent) and in fact caused harm (negative outcome); then the protagonist's action should be morally forbidden. Our results are in line with previous studies in individuals with AS (Klin, 2000; Izuma et al., 2011) that have shown intact performance or subtle deficits on tasks where explicit information is available. However, a recent study (Moran et al., 2011) employing a similar paradigm reported atypical moral judgment in individuals with AS and HFA. The discrepancy between these results and the current findings may be explained by the sample selection criteria employed in each study. Moran and colleagues included both HFA and AS participants. Individuals with HFA have language delay and usually present impairments in verbal skills (Baron-Cohen et al., 2005; Matson and Wilkins, 2008). These difficulties can affect their performance on the task. Thus, moral judgment in adults with AS needs to be further studied using naturalistic social situations without explicit rules.

On the other hand, this is the first attempt to investigate selfmonitoring in social settings in an AS population. As expected, AS participants were less sensitive to the expressive behavior of other individuals, indicating that they had a low capacity for detecting implicit social and interpersonal cues. They also showed a diminished ability to modify self-presentation in social situations, suggesting that they had difficulty with adjusting their behaviors and with navigating novel or challenging social situations. Consistent with this idea, a negative correlation between self-monitoring and measures of social skills has been reported (Furnham and Capon, 1983). Furthermore, the ability to modify self-presentation is negatively correlated with social anxiety (Cramer and Gruman, 2002). Thus, the deficits in self-monitoring in social settings may be related to the lack of social skills and the high levels of anxiety (Hurtig et al., 2009; Lai et al., 2011) experienced by individuals with AS.

Moreover, our results revealed no differences between the AS participants and controls on the SNQ. This finding indicates that social rules knowledge is preserved in adults with AS. In accordance with our data, a study (Zalla et al., 2011) reported that AS and high-functioning individuals with autism are able to detect social rule violations. Furthermore, social norms can be learned in an explicit way. This explicit knowledge can be used by adults with AS to guide their behavior and can act as a compensatory strategy for their social cognition deficits.

Overall, consistent with our hypothesis, the adults with AS showed impairments in several social cognition domains (emotion recognition, ToM, empathy, and self-monitoring in social settings). Specifically, the adults with AS performed poorly on those social cognition tasks (TASIT, FPT, and EPT) that involve an implicit encoding of socially relevant information and the automatic integration of contextual information to solve a given social situation. Conversely, these individuals performed as well as controls in some tasks (RMET, moral judgment task, and SNQ) that had common features. In these tasks the elements of the situation are clearly defined and usually can be solved with relatively abstract and universal rules. This pattern of social cognition performance suggests that one underlying factor may explain the deficits. According to a recently proposed social context network model (Ibáñez and Manes, 2012), this factor seems to be the implicit encoding and the integration of contextual information in order to access to the social meaning.

In addition, our results suggest that adults with AS may benefit from the use of explicit information. However, in most reallife situations, the social demands are not explicitly formulated. Social situations involve implicitly inferring the meaning of the circumstance by integrating contextual cues. Therefore, the pattern of deficits presented here may partially explain the difficulties with social interaction that individuals with AS experience in their daily lives.

Adults with AS may use abstract rules to compensate for their impairments in social cognition. Previous reports have shown that individuals with AS have superior abstract reasoning abilities (Hayashi et al., 2008; Soulieres et al., 2011). This strength may contribute to the performance on social cognition tasks that require the use of abstract rules and the integration of explicit information. On the other hand, this superiority in abstract reasoning may not help in social situations that involve implicit social rules and the integration of contextual cues. In these situations, the meaning of social information is less predictable and relies heavily on context, which reduces the chances of inferring the meaning by applying explicit abstract rules.

#### **VARIABILITY IN THE PERFORMANCE OF ADULTS WITH AS**

Adults with AS showed heterogeneous performance on several EF and social cognition tasks. These participants obtained mainly subnormal performance among the measures with the largest variability. Furthermore, this intra-individual variability was higher for the performances of social cognition than for the EF tests. The decreased variability of the EF tasks can be explained by the intact or superior fluid intelligence in adults with AS (Hayashi et al., 2008; Soulieres et al., 2011). Fluid intelligence is a major dimension of individual differences and refers to reasoning, abstract though and novel problem-solving ability (Duncan et al., 1995; Gray et al., 2003). Previous studies have suggested that high fluid intelligence is associated with better scores on EF tasks (Gray et al., 2003; Burgess and Braver, 2010) and indirectly related to psychosocial cognition (Huepe et al., 2011).

The current study is the first to explore the intra-individual variability of social cognition measures in adults with AS. Consistent with the group analysis, these patients obtained subnormal performance on the same tasks (TASIT, FPT, EPT, IRI, and RSMS). Our data indicates that social cognition performance of adults with AS does not follow the same pattern of strengths and weaknesses reported in other cognitive domains (Hill and Bird, 2006; Towgood et al., 2009). Conversely, the social cognition patterns of individuals with AS is characterized by sub-normal performance, suggesting that these deficits are probably the core of the disorder.

# **CONCLUSIONS**

Our study documents multiple social cognition deficits as fundamental features of the AS diagnosis. Our results showed that adults with AS present deficits in the implicit integration of contextual information in order to access to the social meaning. However, when social information is explicitly presented and the situation can be solved with abstract rules, the individuals with AS usually perform as well as controls. We also found that individual profiles of adults with AS showed subnormal performance in social cognition measures.

This is the first report in adults with AS to evaluate multiple social cognition domains assessing the EF and exploring inter- and intra-individual variability. However, some limitations of this study should be acknowledged. First, our sample size is relatively small, but it is similar to previous social cognition studies (Dziobek et al., 2008; Zalla et al., 2009; Moran et al., 2011) and it is also similar to other reports that have explored the cognitive variability of adults with AS (Hill and Bird, 2006; Towgood et al., 2009) and other patient populations (Deloche et al., 1999; Ramus et al., 2003). Moreover, unlike other reports (Baron-Cohen et al., 2001; Baron-Cohen and Wheelwright, 2004; Moran et al., 2011; Zalla et al., 2011), we only included individuals diagnosed with AS. Second, given the ongoing debate about the differentiation among autistic subtypes, especially between AS and HFA, future studies should compare social cognition profiles of both conditions. Further research should also explore the variability patterns of adults with AS compared with HFA. Third, although AS will probably be formally excluded as a diagnostic category in the DSM-V, our findings are still relevant for studying individual differences within autism spectrum disorders and the subset of people who show a particular profile (previously diagnosed as individuals with AS). In the future, detailed scientific assessments on cognitive domains, such as the ones presented in this work, may help to identify subcategories of autism spectrum disorders.

From a theoretical perspective, our findings are relevant for discussions on social cognition domain specificity in adults with AS. As previously proposed (Stone and Gerrans, 2006a,b), our results support a social cognition profile involving different degrees of affectation and a heterogeneous profile. These results do not support a modular or the "all or nothing" structure of social cognition. Contextual processing seems to affect the social cognition profile of adults with AS in a dissimilar way. For instance, their performance on social cognition tasks may be partially explained by the interaction of low-level mechanisms with the general capacity to integrate contextual information.

From a clinical perspective, our findings may have important implications for the diagnosis and treatment of the AS. The deficits found in multiple social cognition domains seem to be the core feature of the AS. It is also important to promote the use of tasks involving real-life social scenarios because these assessments are more sensitive to AS impairments (Klin, 2000). "Ecological" measures are context-sensitive tools that should be applied in neuropsychiatry (Burgess et al., 2009; Torralva et al., 2009; Ibáñez and Manes, 2012).

In addition, the traditional social skills interventions for individuals with AS are based on learning explicit rules to build and foster relationships with others (Cappadocia and Weiss, 2011). However, the social skills acquired during those interventions do not generalize to situations outside of the treatment setting, which limits the efficacy of these programs (Rao et al., 2008; Cappadocia and Weiss, 2011). Thus, incorporating naturalistic environments into treatment may help individuals with AS generalize the learned social skills. Contextual integration of situated information seems to be crucial for several cognitive processes (Ibáñez et al., 2006, 2010a,b, 2011a,b, 2012; Hurtado et al., 2009; Aravena et al., 2010; Riveros et al., 2010; Amoruso et al., 2011, 2012; Barutta et al., 2011; Couto et al., 2012; Ibáñez and Manes, 2012). Although implementation would be challenging, intervention programs should be based on teaching implicit rules for interpreting unpredictable social contexts. Learning to assess implicit contextual clues may improve the social skills of adults with AS.

# **ACKNOWLEDGMENTS**

The authors thank Ralph Adolphs and Phil Baker for their helpful and insightful comments in an earlier version of the paper. This research was partially supported by CONICET, FONDECYT

## **REFERENCES**


asperger syndrome. *J. Child Psychol. Psychiatry* 34, 163–175.


(1130920) and INECO Foundation Grants. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of those grants.

the growth of short-term memory span. *J. Exp. Child Psychol.* 33, 386–404.


emotions and visual search strategies in adults with Asperger syndrome. *Res. Autism Spectr. Disord.* 5, 210–217.


spectrum disorders. *J. Autism Dev. Disord.* 37, 1386–1392.


video event-related potential design. *Psychiatry Res.* 191, 68–75.


of disgust in Chinese with Huntington's or Wilson's disease. *Neuropsychologia* 41, 527–537.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 27 July 2012; accepted: 19 October 2012; published online: 08 November 2012.*

*Citation: Baez S, Rattazzi A, Gonzalez-Gadea ML, Torralva T, Vigliecca NS, Decety J, Manes F and Ibanez A (2012) Integrating intention and context: assessing social cognition in adults with Asperger syndrome. Front. Hum. Neurosci. 6:302. doi: 10.3389/fnhum. 2012.00302*

*Copyright © 2012 Baez, Rattazzi, Gonzalez-Gadea, Torralva, Vigliecca, Decety, Manes and Ibanez. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# **APPENDIX MEASURES OF SOCIAL COGNITION RECOGNITION OF EMOTIONAL STATES**

The awareness of social inference test (TASIT). The TASIT is a test of social perception that involves videotaped vignettes of everyday social interactions (Kipps et al., 2009; McDonald et al., 2003, 2006). This task introduces contextual cues (e.g., prosody, facial movement, and gestures) and additional processing demands (e.g., adequate speed of information processing, selective attention, and social reasoning) that are not taxed when viewing static displays. We only considered part 1, called the emotion evaluation test (EET), which assesses recognition of emotional expression (fearful, surprised, sad, angry, and disgusted). In the EET, speaker demeanor (voice, facial expression, and gesture) together with the social situation indicate the emotional meaning. In some scenes, there is only one actor talking, who is either on the telephone or talking directly to the camera. Other scenes depict two actors and instructions are given to focus on one of them. All scripts are neutral in content and do not lend themselves to any particular emotion. The brief EET comprises a series of 20 short (15–60 s) videotaped vignettes of trained professional actors interacting in everyday situations. After viewing each scene, the test participant is instructed to choose from a forced-choice list the emotion expressed by the focused actor.

### **ToM**

### *Reading the mind in the eyes (RMET)*

This test (Baron-Cohen et al., 1997) assesses the emotional inference aspect of the ToM (or empathic accuracy). This is a computerized and validated test in which consist of 17 pictures of the eye region of a face. Participants are asked to choose which of four words best describes what the person in each photograph is thinking or feeling.

## *Faux pas test (FPT)*

The FPT assesses the emotional and cognitive inference aspects of the ToM. In this task, the participants read stories that may contain a social faux pas (Stone et al., 1998). After each story was read, the subject is asked whether someone said something awkward (in order to identify stories containing a faux pas). Each story was presented in front of the patient in order to decrease working memory load. Performance was scored regarding the adequate identification of the faux pas (hits) and the adequate rejection of those stories which did not contain a faux pas (rejects). The score was 1 point for each faux pas correctly identified (maximum: 10), or nonfaux pas correctly rejected (maximum: 10). A total score was computed (out of 20 total points) by adding the number of hits and rejects. When a faux pas was correctly identified, subjects were also asked 2 additional questions to measure intentionality—that is, recognizing that the person committing the faux pas was unaware that they had said something inappropriate (maximum 10)—and emotional attribution, in which participants should recognize that the person hearing the faux pas might have felt hurt or insulted (maximum 10).

# **EMPATHY** *Empathy for pain task (EPT)*

The EPT evaluates the empathy in the context of intentional and accidental harms. The task consists of 25 animated situations involving two individuals that are presented successively (Decety et al., 2011). The three following kinds of situations were depicted: intentional pain in which one person (passive performer) is in a painful situation caused intentionally by another (active performer), e.g., stepping purposely on someone's toe (pain caused by other); accidental pain where one person is in a painful situation accidentally caused by another; and control or neutral situations (e.g., one person receiving a flower given by

another). Importantly, the faces of the protagonists are not visible and there was no emotional reaction visible to the participants. We measured the ratings and reaction times (RTs) to situation comprehension (e.g., "press the button as soon as you understand the situation"). In addition, we assessed 7 questions about the following aspects of the scenarios: intentionality, e.g., the accidental or deliberate nature of the action; emphatic concern (how sad you feel for the victim); degree of discomfort (for the victim); harmful behavior (how bad was the purpose of the perpetrator); the valence behavior of the active perpetrator (how much positive emotion he/she felt in performing the action); the correctness of the action (moral judgment); and finally punishment (how much penalty this action deserves). Each question was answered using a computer-based visual analogue scale giving 7 different ratings by trial. Accuracy, RTs and ratings were measured.

Interpersonal Reactivity Index (IRI) (Davis, 1983). The IRI is a 28-item self-report questionnaire that separately measures both the cognitive and affective components of empathy. The instrument contains four scales: Perspective Taking (PT), Empathic Concern (EC), Fantasy (F), and Personal Distress (PD).

# **MORAL JUDGMENT**

#### *Moral judgment task*

Following the protocol reported elsewhere (Young et al., 2010), we presented participants with 24 scenarios. The four variations of each scenario followed a 2 × 2 design: (1) the protagonists either harmed another person (negative outcome) or did no harm (neutral outcome); (2) the protagonists either believed that they would cause harm (negative intent) or believed that they would cause no harm (neutral intent). Each possible belief was true for one outcome and false for the other outcome. The agent held true beliefs in the all-neutral and all-negative conditions and false beliefs in the accidental harm and attempted harm conditions. The participants saw one version of each scenario. In total, eight possible versions of the 24 scenarios with six trials of each of the four conditions were presented. The stimuli were presented in a pseudorandom order and the conditions were counterbalanced across participants. Each participant read six stories in each of the four conditions. After reading each story, the participants were asked to rate the scenario on a Likert-scale ranging from totally permissible (7) to totally forbidden (1).


**A1a | Individual profiles of executive functions tasks performance for each adult with the AS.**

**Table** 

**138**

*control mean*  *Note. P. Fluency, Phonological*

*(supra-normal*

 *performers).*

 *Fluency; Simple F.D., Simple Fluency Design Task; Switching F.D., Switching Fluency Design Task; RT, reaction Time; ACC, accuracy.*


**Table A1b |**


*Note.*

 *P. Fluency, Phonological Fluency; Simple F.D., Simple Fluency Design Task; Switching F.D., Switching Fluency Design Task; RT, reaction Time; Acc, accuracy.*

> | **139**


**TableA2a|IndividualprofilesofsocialcognitiontasksperformanceforeachadultwiththeAS.**

*Note. RTs* =

*reaction times; SC* =

*situation* 

*comprehension;*

 *EC* =

*empathic concern;* 

*D*=*discomfort;*

*IH*=*intention*

 *to hurt;* 

*C*=*correctness;*

*P*=*punishment;*

*I*=*intentionality.*


*Acc. H, accidental harm; Att. H, attempted harm; IH, intentional harm.*


**A2bIndividualprofilesofsocialcognitiontasksperformanceforeachcontrol**

**Frontiers in Human Neuroscience www.frontiersin.org** November 2012 | Volume 6 | Article 302



*Acc. H, accidental harm; Att. H, attempted harm; IH, intentional harm.*

## **SOCIAL NORMS KNOWLEDGE**

#### *Social norms questionnaire (SNQ)*

The SNQ questionnaire consisting of 20 yes–no questions was used (Rankin et al., 2009). The participants were asked to determine whether a behavior would be appropriate in the presence of an acquaintance (not a close friend or family member) according to the mainstream culture. Two scores were derived. The break score was defined as the total number of errors made in the direction of breaking a social norm, and the over-adhere score was defined as the total number of errors made in the direction of over adherence to a perceived social norm.

## **SELF-MONITORING BEHAVIOR IN SOCIAL SETTINGS** *Revised self-monitoring scale (RSMS)*

The RSMS is a 13-item instrument and assesses the tendency to regulate one's behavior to present a particular self in a social context (Lennox and Wolfe, 1984). The scale involves two styles of self-monitoring behavior: the ability to modify self-presentation (e.g., "in social situations, I have the ability to alter my behavior if I feel that something else is called for") and the sensitivity to the expressive behavior of others (e.g., "I am often able to read people's true emotions correctly through their eyes"). The participants responded using a 6-point Likert-scale. The ratings ranged from 0 = "strongly disagree" to 5 = "strongly agree."

# **REFERENCES**


# Neural responses to emotional expression information in high- and low-spatial frequency in autism: evidence for a cortical dysfunction

# *Corrado Corradi-Dell'Acqua1,2\*, Sophie Schwartz 2, Emilie Meaux2, Bénedicte Hubert 3,4, Patrik Vuilleumier 1,2 and Christine Deruelle4*

*<sup>1</sup> Swiss Center for Affective Sciences, University of Geneva, Geneva, Switzerland*

*<sup>2</sup> Laboratory for Neurology and Imaging of Cognition, Department of Neuroscience and Clinic of Neurology, University Medical Center, Geneva, Switzerland*

*<sup>3</sup> Hôpital Rivière-de-Praires, University of Montréal, Montréal, QC, Canada*

*<sup>4</sup> CNRS, Institut de Neurosciences de la Timone, Aix-Marseille Université, Marseille, France*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Nathalie George, Centre National de la Recherche Scientifique, France Martin Schulte-Rüther, Universityl Hospital RWTH Aachen, Germany*

#### *\*Correspondence:*

*Corrado Corradi-Dell'Acqua, Swiss Center for Affective Sciences, NCCR Affective Sciences, University of Geneva – CISA, Campus Biotech, Uni Durfour, 24 rue Géneral Dufour, CH-1211 Geneva, Switzerland e-mail: corrado.corradi@unige.ch*

Despite an overall consensus that Autism Spectrum Disorder (ASD) entails atypical processing of human faces and emotional expressions, the role of neural structures involved in early facial processing remains unresolved. An influential model for the neurotypical brain suggests that face processing in the fusiform gyrus and the amygdala is based on both high-spatial frequency (HSF) information carried by a parvocellular pathway, and low-spatial frequency (LSF) information separately conveyed by a magnocellular pathway. Here, we tested the fusiform gyrus and amygdala sensitivity to emotional face information conveyed by these distinct pathways in ASD individuals (and matched Controls). During functional Magnetical Resonance Imaging (fMRI), participants reported the apparent gender of hybrid face stimuli, made by merging two different faces (one in LSF and the other in HSF), out of which one displayed an emotional expression (fearful or happy) and the other was neutral. Controls exhibited increased fusiform activity to hybrid faces with an emotional expression (relative to hybrids composed only with neutral faces), regardless of whether this was conveyed by LSFs or HSFs in hybrid stimuli. ASD individuals showed intact fusiform response to LSF, but not HSF, expressions. Furthermore, the amygdala (and the ventral occipital cortex) was more sensitive to HSF than LSF expressions in Controls, but exhibited an opposite preference in ASD. Our data suggest spared LSF face processing in ASD, while cortical analysis of HSF expression cues appears affected. These findings converge with recent accounts suggesting that ASD might be characterized by a difficulty in integrating multiple local information and cause global processing troubles unexplained by losses in low spatial frequency inputs.

#### **Keywords: autism, facial expression, emotion expression, spatial frequency, fMRI**

# **INTRODUCTION**

Autism Spectrum Disorder (ASD) is a pervasive neurodevelopmental disorder characterized by dysfunctional socialization and communication, with the emergence of stereotyped and repeated behavior. Although this disorder is mostly known for its social symptoms, a wealth of studies converge in reporting atypicalities in elementary aspects of perception, as in the case of visual processing of (emotional and neutral) facial expressions (see Harms et al., 2010; Gaigg, 2012; Weigelt et al., 2012, for meta-analyses and reviews).

A wealth of studies based on abstract and geometrical stimuli, suggest that in ASD individuals have difficulties in processing visual stimuli in a global fashion, focusing instead on details and local information (e.g., Dakin and Frith, 2005; Happé and Frith, 2006; Mottron et al., 2006). These accounts can potentially explain also ASD atypical processing of faces, especially considering that facial properties (identity, gender, emotional expressions, etc.) are not usually processed by the analysis of isolated local features, but of how all different features relate one another at the global level (configural processing). In this perspective, tasks asking neurotypical individuals to assess the sameness of two faces usually report poorer performance when the standard spatial relation between the parts is distorted, as for upside-down faces (Valentine, 1988), composites made of two aligned half-faces from different people (Young et al., 1987), or faces with scrambled parts (Tanaka and Farah, 1993). However, studies implementing the same tasks in individuals with ASD have reported mixed findings with some describing them as not influenced (Van Der Geest et al., 2002; Joseph and Tanaka, 2003; Teunisse and de Gelder, 2003; Rondan and Deruelle, 2004; Riby et al., 2009) or less influenced than Controls (Hobson et al., 1988; López et al., 2004; Barton et al., 2007; Pellicano et al., 2007, see also Weigelt et al., 2012), but others describing equal effects as in neurotypical individuals (Teunisse and de Gelder, 2003; Rouse et al., 2004; Lahaie et al., 2006; Gross, 2008). Such variability could reflect the important heterogeneity of the ASD population, in which diagnostic symptoms are expressed differently across individuals, maybe confounded by age or attentional factors (Rondan and Deruelle, 2007), and/or possibly stem from the development of compensatory neuronal mechanisms (Gaigg, 2012; Dickstein et al., 2013).

To better characterize the face processing atypicalites observed in ASD, several studies have focused on the spatial frequency at which specific information is conveyed, suggesting that distinct frequencies might play different roles in face processing (Deruelle et al., 2004, 2008; Rondan and Deruelle, 2004; Boeschoten et al., 2007a; Vlamings et al., 2010). Indeed, local information can be processed only through high-spatial frequencies (HSF), whereas global configurations can be retained also from low spatial frequencies (LSF). It is well known that HSF visual information is carried by parvocellular pathways (see **Figure 1**, orange arrow) which reach the striate cortex and project almost exclusively to ventral occipito-temporal structures, including that part of the fusiform cortex which processes face stimuli (Fusiform Face Area [FFA], Kanwisher et al., 1997). LSF information instead is conveyed by magnocellular pathways (**Figure 1**, blue arrow) which project mostly to dorsal to parietal regions and, in less extent, to ventral cortical visual areas (Livingstone and Hubel, 1987, 1988). In addition, however, it has been proposed that the amygdala, a medial temporal structure critically involved in processing emotional expression in faces (Vuilleumier and Pourtois, 2007; Pessoa and Adolphs, 2010), may receive direct subcortical inputs from an additional collicular-pulvinar projection of magnocellular pathways (De Gelder et al., 1999; Morris et al., 1999), allowing the amygdala and ventral visual stream to receive coarse (LSF), but fast, information about facial emotional expressions (Vuilleumier et al., 2003, 2004; Winston et al., 2003b; Carretié et al., 2007; Vuilleumier and Pourtois, 2007). In this perspective, the frequent reports of atypical fusiform and/or amygdala responses to face stimuli in ASD (Baron-Cohen et al., 2000; Critchley et al., 2000; Schultz et al., 2000; Pierce et al., 2001; Hall et al., 2003; Hubl et al., 2003; Wang et al., 2004; Grelotti et al., 2005; Ashwin et al., 2007; Kleinhans et al., 2008; Scherf et al., 2010) raise the question of whether these effects might depend on differential visual frequency information conveyed by parvocellular (cortical) or magnocellular (also subcortical) pathways.

A number of studies have employed electrophysiological recording or behavioral techniques in children with ASD using high or low spatial filtered stimuli. While some results suggested that ASD affects preferentially the visual pathway conveying LSF (Deruelle et al., 2004, 2008; Boeschoten et al., 2007a; Vlamings et al., 2010), similar approaches in adults reported an ability to process LSF expressions comparable to that of neurotypical individuals (Rondan and Deruelle, 2004). To the best of our knowledge, no study has ever tested directly how, in ASD, the fusiform gyrus and the amygdala respond to HSF and LSF information in human faces.

In the present study, we showed to adults with ASD and matched Controls hybrid facial stimuli, which were generated according to a methodology used in previous studies (Schyns and Oliva, 1999; Winston et al., 2003b), by merging a HSF face with a LSF face of opposite gender (see **Figure 2**). These stimuli are particularly suited for our purpose as they offer to the observer both high- and low- extremes of the frequency spectrum of faces in the same stimulus, allowing us to determine

the band in which specific information is preferentially selected for face processing and responded to in different brain areas. Critically, in addition to mixing opposite genders in each spatial frequency band, our hybrid stimuli were made of the combination of a neutral expression and an emotional expression, with the latter being either fearful or happy and contained either in the HSFs or LSFs (counterbalanced across the gender dimension). In order to probe for differential responses to emotion expressions triggered by one or the other frequency bands, we engaged our participants in a gender discrimination task in which they had to report the apparent gender of each face. In a separate condition, our participants instead had to watch passively each stimulus, allowing us to determine any influence of different task demands. This led to a factorial design with group (ASD participants, Controls), frequency (emotional expressions conveyed by HSF, LSF), valence (fearful, happy expressions), and task (gender discrimination, passive viewing). For high-level baseline, we used (in each task) hybrid stimuli with no emotional expression (i.e., mix of neutral female and neutral male). Following previous studies on neurotypical individuals, we expected increased activity in ventral visual cortex (including the fusiform gyrus) when Controls discriminated faces in which emotions were conveyed by either HSFs or LSFs (as opposed to neutral expressions), as evidence of parvocellular cortical and magnocellular visual inputs respectively (Vuilleumier et al., 2001, 2003; Winston et al., 2003b; Rotshtein et al., 2007). The critical question, however, was whether these parvo- and magnocellular neural signatures were observable also in individuals with ASD. We reasoned that if ASD affects the direct subcortical inputs to the amygdala (Vuilleumier et al., 2003, 2004; Carretié et al., 2007; Rotshtein et al., 2007), ASD individuals should exhibit a reduced neural response to emotional expressions conveyed by LSFs. On the other hand, if ASD individuals present a reduced neural response to emotions mediated by HSFs, this can be interpreted exclusively as reflective of an effect to the cortical path.

# **MATERIALS AND METHODS PARTICIPANTS**

Two groups were included in the experiment. The first group comprehended 13 high-functioning adults males with autistic spectrum disorder (ASD) recruited from the database of the Specialized Clinic for Pervasive Developmental Disorders of Rivière-des-Prairies Hospital. Diagnosis of autism was established with the Autism Diagnostic Interview-Revised (ADI-R; Lord et al., 1994) and validated by a standardized assessment with the Autism Diagnosis Observation Schedule (ADOS-G, module 3 or 4; Lord et al., 2000). All participants from the clinical group met the diagnostic criteria for autism or Asperger syndrome according to both instruments. The second group comprehended 15 matched male participants with typical development recruited from the same database.

Some of the participants were excluded from the overall analysis due to technical problems occurred during the acquisition phase and due to head-movement artifacts in the BOLD signal. Therefore, the overall analysis was run on two homogeneous groups of 10 individuals each. Participants from both groups completed one of the Wechsler Intelligence scales (WAIS-R,

**FIGURE 2 | Schematic representation of how the hybrid stimuli were created.** Male and female faces were selected from a validated database and subjected to spatial frequency filtering. Hybrid faces were then created by overlaying a high-spatial frequency (HSF) face with a low-spatial frequency (LSF) face of opposite gender. Emotional expressions (in either the HSF or LSF face) were always overlaid with a neutral facial expression of opposite gender and opposite filtering. Three stimuli examples are also displayed on the lower part of the Figure. A HSF neutral face overlaid with a happy LSF face (condition LH), a HSF fearful face overlaid with a neutral LSF face (condition HF) and a control condition containing neutral faces in both HSF and LSF bands (N). Full details in the text.

WAIS-III) and the Edinburgh questionnaire (Oldfield, 1971). ASD and Control participants were group-wise matched according to their IQ [Manova on all IQ variables: Pillai's trace = 0.15, *F*(3,15) = 0.92, n.s.], age and handedness. **Table 1** summarizes the participants' demographic characteristics. Each participant gave informed consent to participate in the study and received monetary compensation. The study was formally approved by the ethics committee of Rivière-des-Prairies Hospital and the committee for ethics and research of Regroupement Neuroimagerie/Québec (CMER-RNQ). All participants were naïve to the purpose of the task.

### **STIMULI**

Our experimental stimuli were hybrid images built by combining a face composed only by HSFs with a face composed only by LSFs of opposite gender, whose expression was also separately manipulated (see **Figure 2**). The detailed procedure was adapted from the one used by previous studies (Schyns and Oliva, 1999; Winston et al., 2003b) and can be summarized as follows. We took pictures of emotional (happy and fearful) and neutral facial expression, displayed from a frontal point-of-view, from the Karolinska Directed Emotional Faces database (A series). Validation studies on stimuli from this database (Goeleven et al., 2008) confirmed that happy and fearful expressions were matched for intensity and arousal, and both associated with higher recognition scores than neutral expressions. These images were desaturated, scaled to a size of about 5.30◦ (horizontal) × 6.80◦ (vertical) of visual angle and, subsequently, filtered in Fourier space, using a Butterworth filter to remove either high spatial frequencies (above 24 cycles/face [c/fw], corresponding to about 4 cycles/degree of visual angle [c/deg]) or low spatial frequencies (below 6 c/fw, corresponding to about 1 c/deg). Hybrid stimuli were then created by overlapping one HSF face and one LSF face into a single stimulus (see **Figure 2**). The eyes and mouth position was matched between the LSF and HSF images in order to obtain a visual overlap yielding the percept of single face. Critically, faces in the LSF and HSF images were chosen so that they always had an opposite gender (one female, one male) and could display different emotional expressions. This manipulation led to the following five conditions of interest: high fearful stimuli, composed by a fearful HSF face and a neutral LSF face (HF); high happy stimuli,


*Average age and IQ values are displayed in average values with bootstrap estimated 95% confidence intervals. None of the displayed measures varied significantly across groups with* α = *0.05.*

composed by a happy HSF face and a neutral LSF face (HH); low fearful stimuli, composed by a neutral HSF face and a fearful LSF face (LF); low happy stimuli, composed by a neutral HSF face and a fearful LSF face (LH); neutral stimuli (high-level control), composed by a neutral HSF face and a neutral LSF face (N). For each of these conditions we built 32 different images, each of which was presented twice during the experimental session (32 images × 2 repetitions × 5 conditions = 320 face stimuli). In half of these 32 pictures of each emotional condition, the emotional expression was conveyed by the female face, whereas in the remaining half the emotional expression was conveyed by a male face.

#### **EXPERIMENTAL SETUP**

The 320 face stimuli used in the study were presented to participants in an event-related fashion. On each experimental trial, one hybrid face was shown to the participant for 83 ms and followed by an inter-stimulus interval of variable duration (range 2500–12500 ms) in order to improve sensitivity of fMRI BOLD measurements. To encourage participants to keep their gaze on the center of the screen, a fixation cross was present during the inter-stimulus interval. The whole experiment was organized into four experimental sessions, each comprehending 80 trials (16 per condition) and lasting about 4 min. Among these four sessions, two were associated with an active gender detection task: since participants were not aware that the hybrid stimuli were created by faces of opposite genders, they were requested to indicate, as fast as possible, its apparent gender by pressing one of two possible keys with either hand (e.g., left hand for male response, right for female, counterbalanced across participants). Previous studies using hybrid stimuli filtered at the same cut-offs found that, with such short stimulus presentations, participants rely with comparable likelihood on LSF and HSF information to make their gender judgments, as they report the gender of the LSF face on ∼50% of the trials (Schyns and Oliva, 1999; see also Winston et al., 2003b, in which the LSF face was chosen 60% of the trials). The two remaining sessions had no active task, and participants were simply requested to pay attention to each and every face. The order between passive and task-positive sessions was counterbalanced across subjects.

The stimuli were presented using E-Prime 1.0 (Psychology Software Tools, Inc.) and projected inside the scanner bore with a LCD projector on a screen subtending about 19◦ (horizontal) × 14◦ (vertical) of visual angle. Key-presses were recorded on an MRI-compatible bimanual response button box. Participants were instructed to press one of two possible keys, placed at each hand's reach, to indicate their responses.

## **FACE FUNCTIONAL LOCALIZER**

Our study aimed at investigating the sensitivity to band-filtered face information in key areas of the core face processing system, particularly fusiform cortex and amygdala (Haxby et al., 2000; Gschwind et al., 2012). To this aim, we mapped the face processing network in both groups with an unbiased (not band-filtered) set of face stimuli. We therefore carried out an independent scanning session adapted from previous studies (Schwarzlose et al., 2005; Spiridon et al., 2006) and structured as follows. Participants were presented four blocks of gray-scale full-band face photographs alternating with four blocks of gray-scale house photographs. Photos were displayed centrally, and had a size of about 9.82◦ × 9.82◦). Within each block, there were 18 face/house specimens each presented for 750 ms followed by an interstimulus interval of 500 ms. Each block lasted of about 22 s each and was immediately followed by another. Whilst perceiving these images participants performed a 1-back task, in which they had to signal through key-press whether the picture in the current trial was identical to the one in the previous trial. The experiment was built so that a positive response from the participant was expected only in two trials in each block. The whole localizer session lasted about 3 min and always followed the four main experimental sessions.

## **IMAGING PROCESSING** *Data acquisition*

The study was conducted in the neurofunctional imagery unit at the research center of the geriatric institute of Montreal. A Siemens Trio 3-T whole-body scanner was used to acquire gradient-echo planar T2-weighted MRI images with blood oxygenation level dependent (BOLD) contrast. The scanning sequence was a trajectory-based reconstruction sequence with repetition time (TR) of 2160 ms, echo time (TE) of 30 ms, flip angle of 90◦, in-plane resolution 64 × 64, 36 slices, slice thickness of 3 mm, and no gap between slices. A structural image was of each participant was also recorded with a T1-weighted MPRAGE sequence (176 slices, *TR* = 9.7 msec, *TE* = 4 ms, flip angle = 12◦, in-plane resolution = 256 × 256, 1 × 1 × 1 mm voxel size).

## *Preprocessing*

Statistical analysis was performed using the SPM software (http:// www.fil.ion.ucl.ac.uk/spm/). For each subject, all functional images were realigned, slice-time corrected to allow a whole volume to be treated as a single data point, normalized to a template based on 152 brains from the Montreal Neurological Institute (MNI), resliced at a voxel size of 3 × 3 × 3 mm, and then smoothed by convolution with a 8 mm full-width at halfmaximum (FWHM) Gaussian kernel.

#### *First-level analysis*

Data from each participant were analyzed using the General Linear Model (GLM) framework implemented in SPM. For the face localizer session, we modeled each of the two active conditions (faces, houses) with a boxcar function. For the main experimental sessions, the trial onsets from each condition of our design were modeled with a delta (stick) function. Critically, whereas in the two passive viewing sessions we modeled only the main five conditions of our design (HF, HH, LF, LH, N), in the gender discrimination task we also took into account participants' response on every trial (see Winston et al., 2003b). Thus, for each of the five main conditions, we modeled separately those trials in which participants made their gender judgments on the basis on visual cues conveyed by LSFs (e.g., HFL, HHL, LFL, LHL, NL), those trials in which participants judged gender based on HSFs (HFH, HHH, LFH, LHH, NH), and also those few trials in which responses were omitted (if any). Each regressor was convolved with a canonical hemodynamic response function as implemented in SPM. To account for movement-related variance, we included, for each session, six differential movement parameters [x, y, and z translations (in millimeters) and pitch, roll, and yaw rotations (radiants)] as covariates of no interest. Low-frequency signal drifts were filtered using a cutoff period of 128 s.

## *Second level analysis*

For the functional localizer, we calculated for each participant the contrast describing the differential activity *Faces* > *Houses*. These contrasts were fed in a second-level independent sample *t*-test, under the assumption of unequal variance between the groups. This test allowed us to investigate both effects of *Faces* vs. *Houses* in Controls and ASD participants, as well as cross-over interaction effects.

For the main experiment, we considered for each subject 15 contrast images. 10 of them were computed from the gender discrimination task, and concerned activity associated with the five main conditions and the two possible responses (i.e., HFL, HFH, HHL, HHH, LFL, LFH, LHL, LHH, NL, NH). The remaining five concerned activity in the five conditions of interest (i.e., LF, LH, HF, HH, N) during the passive viewing sessions. These contrasts were fed into second-level flexible factorial design with "conditions" as a within-subject factor, "group" as betweensubject factor and "subject" as random factor, using a random effects analysis (Penny and Holmes, 2004). In modeling the variance components, we allowed each of these three factors to have unequal variance between their levels. Activations in these analyses were considered as significant if exceeding an extent threshold allowing *p* < 0.05 correction for multiple comparison for the whole brain (corresponding to 59 and 63 consecutive voxels, for the localizer and main experiment respectively—Friston et al., 1993), with an underlying height threshold corresponding to *p* < 0.001 uncorrected [*t*(18) > 3.61 and *t*(250) > 3.13, for the localizer and main experiment].

# **RESULTS**

## **BEHAVIORAL RESULTS**

To obtain a measure of spatial frequency biases in face processing for different conditions in each group, we analyzed the rate at which participants selected the gender of the LSF face in the hybrid stimuli. In this measure, values greater than 0.5 reflect experimental conditions in which participants relied more on the LSF information to make gender judgments, whereas values smaller than 0.5 reflect conditions in which participants relied more on the HSF information.

We first analyzed the conditions in which an emotional expression was displayed through a 2 × 2 × 2 Repeated measures ANOVA, with the FREQUENCY containing the emotional expression (HSF, LSF) and the VALENCE of this emotional expression (Fearful, Angry) as within-subject factors, plus participant GROUP (ASD individuals, Controls) as between-subjects factor. We found a significant main effect of FREQUENCY [*F*(1, <sup>18</sup>) = 9.79, *p* < 0.01], reflecting that overall participants relied more on LSF information [average 0.56, bootstrapestimated 95% confidence intervals of the average (0.46, 0.64)], rather than on HSF [0.52 (0.40, 0.62)]. However, this LSF-bias also depended on the valence of the emotion expression (see **Figure 3**). Thus, whereas the VALENCE main effect was not significant [*F*(1, <sup>18</sup>) = 0.14], this factor interacted significantly with FREQUENCY [*F*(1, <sup>18</sup>) = 18.48, *p* < 0.001]. **Figure 2** shows that, in both groups, gender judgments were more LSF-biased when low frequencies conveyed happy expressions, as opposed to fearful [*LH* > *LF*: *t*(19) = 2.39, *p* < 0.05]. Instead, judgments were more HSF-biased when high frequencies conveyed happy, as opposed to fearful, expressions [*HF* > *HH*: *t*(19) = 2.96, *p* < 0.01]. The factor GROUP yielded no significant main effect nor interaction [*Fs*(1, <sup>18</sup>) < 1.00]. Visual inspection of **Figure 3** suggests that ASD individuals might be more LSF-biased than controls, although no significant effect of the factor GROUP was found. However, this initial analysis did not comprehend the high-level control condition in which neutral facial expressions were presented. We therefore also tested for putative group differences in LSF-rate, both when taking each of the five main conditions (thus including N) separately, and when averaging them together. None of these tests led to a significant effect [|*t*|(18) always < 1.60].

Furthermore, for each condition, we computed the median time [in milliseconds (ms)] necessary to deliver a response (Response Times) and analyzed it in a similar fashion as above. In this analysis we also tested for any putative effect of the participants' choice. We therefore analyzed the emotional conditions in a 2 × 2 × 2 × 2 repeated measures ANOVA with FREQUENCY (HSF vs. LSF), VALENCE (Fear vs. Happy), and CHOICE (HSF vs. LSF gender) as within-subject factor and GROUP (Controls vs. ASD individuals) as between subject factor. We found a significant main effect of FREQUENCY [*F*(1, <sup>18</sup>) = 4.96, *p* < 0.05], reflecting faster responses when the emotional expression was conveyed by LSF [825.50 ms (735.41, 908.49)], as opposed to HSF [870.51 (769.19, 944.95)]. No other main/interaction effect, including those associated with the factor CHOICE, was significant [*Fs*(1, <sup>18</sup>) < 4.27, *p*s > 0.05].

### **NEURAL RESPONSES**

*Face Localizer.* Data from the Face localizer are displayed in **Table 2** and **Figure 4**. We tested, in each group, whether there were significant differences in neural activity between the Face and House categories. The contrast *Faces* > *Houses* confirmed, in both neurotypical (**Figure 4**, red clusters) and ASD individuals (green clusters), an involvement of the amygdala and of the posterior portion of the superior temporal sulcus in the two hemispheres. Controls also exhibited activation the medial orbitofrontal cortex. No fusiform activation was found in either group at the whole-brain threshold. We therefore performed additional region-of-interest analyses restricted to those voxels that were part of these fusiform gyrus as described by predefined anatomical masks (AAL database—Tzourio-Mazoyer et al., 2002). In Controls, we found bilateral activation of the fusiform gyrus, ∼45 mm posteriorly from the anterior commissure, over and around the region usually identified as FFA. No suprathreshold activation was found in ASD participants, although at a less stringent height threshold (corresponding to *p* < 0.005 uncorrected) activation was found around the same FFA coordinates as defined in the Control group for both the right (12 consecutive voxels centered at the coordinates *x* = 42, *y* = −48,

*z* = − 21) and left hemisphere (three consecutive voxels centered at *x* = − 39, *y* = − 48,*z* = − 21). The opposite contrast (*Houses* > *Faces*) implicated large portions of the parahippocampal gyrus, extending to the calcarine cortex and the medial portion of the middle occipital gyrus, in both groups similarly. When testing for the interaction between the grouping factor and the stimuli employed (*Faces* vs. *Houses)*, we found no suprathreshold effects.

In sum, data from this localizer session successfully identified neural structures most sensitive to face stimuli, indicating the recruitment of similar portions of the fusiform cortex and amygdala in each group (although the evidence of FFA activity in ASD individuals was obtained with a more liberal threshold).

As the functional localizer aimed at mapping in our population those portions in fusiform cortex and amygdala that were most sensitive to full-band face stimuli in each group, we then used the results of the localizer session to create a mask which could serve as region of interest in all subsequent analyses. This mask was built following anatomical and functional criteria, as it included voxels which (1) were part of either the fusiform gyrus or the amygdala according to predefined anatomical masks (AAL database) and (2) exhibited significant [*t*(18) > 1.73, *p* < 0.05 uncorrected] increase of neutral activity for faces (as opposed to houses) in each group [conjunction ((*Faces* > *Houses*)Controls ∩ (*Faces* > *Houses*)ASD)]. The resulting mask, which was smoothed (8 mm FWHM Gaussian kernel) and subsequently re-binarized to minimize spatial inhomogeneities, encompasses that part of the fusiform-amygdala face network that is common to both groups.

#### *Effects of LSF emotional expressions*

We focused on that portion of the data in which Controls carried out the gender discrimination task and tested for increases of neural activity associated with LSF expressions, relative to neutral stimuli [(*LFL* + *LFH* + *LHL* + *LHH*)/2 − (*NL* + *NL*), hereafter *LSF* - *N*]. When correcting for multiple comparisons across the whole brain we found no suprathreshold activation. However, when applying small-volume correction on those portions of the fusiform gyrus and amygdala identified in the localizer (see above), we found a significant increase of neural activity in the right fusiform cortex (see **Table 3** and **Figure 5A**, red blob). This right fusiform activation was close, not only to the location previously identified by Winston et al. (2003b) in the same contrast (distance between the local maxima from the two studies ∼11 mm), but also to the right FFA cluster isolated in the same group during the face localizer and displayed in **Figure 4** (local maxima distance ∼15 mm). No suprathreshold effect was found in the amygdala (similar to Winston et al., 2003b, but see Vuilleumier et al., 2003 who used simple band pass filtered stimuli).

One of the key questions of the present study was to assess whether this increase of neural activity in FFA for LSFs (as found in Controls) was absent or preserved in ASD individuals. We therefore examined the sessions in which ASD participants performed the gender discrimination task and tested for the same contrast *LSF* − *N*: this revealed an activation of the left FFA, in a location very symmetrical to that identified in Controls (see **Figure 5A**, green blob—local maxima distance between this cluster and the left FFA cluster identified in the same group ∼6 mm). No effect was found in the right fusiform gyrus or in the amygdala even at the most liberal thresholds.

We further explored putative group differences in the neural response to LSF emotional expressions by testing the interaction between the contrast *LSF* − *N* and the grouping factor. In particular, we tested for regions in which the differential activity between LSF and neutral expressions in Controls was not only larger than 0 (as already tested above), but also larger than the same differential activity in ASD [i.e., (*LSF* − *N*)Controls − (*LSF* − *N*)ASD]. However, as this test also isolates regions with no difference between LSF and neutral expressions in Controls, but with reduced activity for LSF expressions (as opposed to N) in ASD individuals, we excluded from our search those regions that were implicated (*p* < 0.05 uncorrected) in the contrast *N* − *LSF* in ASD (exclusive masking). This test revealed no differential effect, neither when correcting for multiple comparisons for the whole brain, nor when focusing on the face-sensitive portions of fusiform gyrus/amygdala. With a similar logic, we tested for

#### **Table 2 | fMRI analysis: face localizer.**


*Regions showing significant activations associated with the 1-back task in the Face localizer session. Coordinates (in standard MNI space) refer to maximally activated foci: x* = *distance (mm) to the right (*+*) or the left (*−*) of the midsagittal line; y* = *distance anterior (*+*) or posterior (*−*) to the vertical plane through the anterior commissure (AC); z* = *distance above (+) or below (*−*) the inter-commissural (AC-PC) line. L and R refer to the left and right hemisphere, respectively. M refers to medial clusters.*

*†p* < *0.001; ‡p* < *0.01; \*p* < *0.05 corrected at the cluster level for the whole brain (underlying height threshold: p* < *0.001, uncorrected).*

*1p* < *0.05 corrected at the voxel level for the fusiform gyrus bilaterally.*

regions in which the differential activity between LSF and neutral expressions was larger in ASD individuals than in Controls [i.e., (*LSF* − *N*)ASD − (*LSF* − *N*)Controls]. Also this test led to no suprathreshold effects, including for the fusiform gyrus/amygdala at liberal thresholds.

In sum, not only we found reliable evidence in the neurotypical brain for a role of LSF inputs conveying emotional expression information to FFA (Winston et al., 2003b), but we also found equivalent (although contralateral) effects in ASD, suggesting that these LSF inputs are preserved in these participants. Furthermore, direct comparison of the effects identified in each group failed to reveal any significant difference.

#### *Effects of HSF emotional expressions*

We next tested for regions exhibiting suprathreshold activity when emotional face expressions were conveyed by HSFs. We first computed, in Controls, the contrast [(*HFL*+*HFH* + *HHL*+*HHH*)/2 − (*NL* + *NL*), hereafter *HSF* − *N*] which revealed enhanced bilateral activity in the fusiform gyrus, over and around FFA, as well as in the Amygdala (see **Table 3** and **Figure 5B**, red blobs). Critically, the fusiform clusters were proximal to the FFA clusters delineated with the functional localizer in the same group (**Figure 4**) and to the clusters identified by the main effect of LSF expressions (**Figure 5A**, red blobs). No effect was found for the contrast *HSF* − *N* when ASD participants carried the discrimination task.

We then tested directly whether the differential activity observed in Controls was larger, not only than 0, but also of its homologous in ASD individuals *via* an interaction test [(*HSF* − *N*)Controls − (*HSF* − *N*)ASD, excluding voxels sensitive to (*N* − *HSF*)ASD]. We found no suprathreshold activity, neither when correcting for the whole brain, nor when applying a small volume correction. It should be mentioned, however, that under an uncorrected extent threshold (underlying

#### **Table 3 | fMRI analysis: effects of HSF and LSF emotional cues.**


*Regions showing significant activation associated with the discrimination task.*

*†p* < *0.001; ‡p* < *0.01; \*p* < *0.05 corrected at the cluster level for the whole brain (underlying height threshold: p* < *0.001, uncorrected).*

*1p* < *0.05 corrected at the voxel level for FFA and amygdala bilaterally as described by the localizer data.*

height threshold of *p* < 0.001), we found five consecutive voxels on right FFA [local maxima: *x* = 33, *y* = −51, *z* = −21, *t*(162) = 3.57], proximal to the cluster previously implicated when testing effects of HSF expressions (local maxima distance <5 mm). No region exhibited HSF increases of activity specific for ASD individuals [(*HSF* − *N*)ASD − (*HSF* − *N*)Controls], neither when correcting for the whole brain, nor when inspecting the fusiform gyrus and the amygdala with more liberal approaches.

In sum, the analysis of HSF effects during the discrimination task revealed significant increases of neural activity in FFA and amygdala to emotional expressions in the Controls exclusively. For the right FFA, such increase was not only larger than 0, but also larger than the homologous (non-significant) effect measured in ASD individuals.

#### *Direct comparisons between LSF and HSF expressions*

We also compared differential responses to LSF or HSF expressions, not against the control neutral condition, but against each other. Unlike the analysis conducted insofar—which identified regions sensitive to one frequency band, irrespective of their sensitivity also to the other bands—these direct comparisons now probed for any region that would code *preferentially* for emotional information conveyed by specific frequencies.

When testing for differential responses to LSF expressions [contrast (*LFL* + *LFH* + *LHL* + *LHH*)–(*HFL* + *HFH* +

*HHL* + *HHL*), hereafter *LSF* − *HSF*], we found no suprathreshold effect in neither in Controls, nor in ASD individuals. We then tested the converse contrast *HSF* − *LSF*, which probed for any region processing emotional facial expressions from HSF cues preferentially to LSF cues. For Controls, this contrast elicited large activations within the ventral occipital cortex, including the posterior portions of the fusiform gyrus. Further activations were found in the right angular gyrus, the precuneus/posterior cingulate cortex, the ventral striatum bilaterally, the medial orbitofrontal cortex and, when applying small volume correction, the right amygdala (see **Table 3** and **Figure 6**, red blobs). No effects (even at the most liberal thresholds) were observed when the same contrast was run on the ASD group.

In sum, these data confirm in Controls the recruitment of a widespread network processing emotional face expression from HSFs preferentially to LSFs. Such network was not reported in ASD individuals, not even at the most liberal thresholds. It is therefore possible that the same regions processing preferentially HSF in Controls might exhibit different sensitivity to spatial frequency emotional cues in ASD. We formally tested this via a cross-over interaction contrast (*HSF* − *LSF*)Controls − (*HSF* − *LSF*)ASD, comparing the differential sensitivity between HSF and LSF emotional cues across groups. As fully described in **Figure 6** (yellow blobs) and **Table 3**, this analysis confirmed the role played by the lingual gyrus, the ventral striatum and the right amygdala. Furthermore, this analysis also implicated the left lateral occipital cortex and left FFA, thus confirming how this region seems more sensitive to HSF expressions in Controls and, concurrently, to LSF expressions in ASD individuals (see also **Figure 5**).

#### *Effects of the reported gender and of emotional valence*

All analyses conducted insofar were run regardless of the behavioral performance and of the emotional valence. **Figure 7** illustrates the activity parameters extracted from those FFA and Amygdalar voxels identified by the contrasts *LSF* − *N* and *HSF* − *N* (**Figure 5**). Visual inspection of these data suggests how in some cases the differential activity between emotional and neutral expressions described above might be biased by the task demands. In particular, the amygdala exhibited, in Controls, a differential increase in activity for HSF expressions; however, further in-depth analyses on the extracted parameters revealed a general marginal preference for all trials in which HSF were "preferred" for the gender discrimination [choose HSF vs. choose LSF: right Amygdala *t*(9) = 2.04, *p* = 0.072; left Amygdala *t*(9) = 2.05, *p* = 0.071]. Instead, amygdala activity seemed unaffected by the kind of emotion displayed by HSFs [fearful vs. happy: right Amygdala *t*(9) = 0.87, n.s.; left Amygdala *t*(9) = −0.13, n.s.]. Keep in mind that the contrast *HSF* − *N*, implicating the amygdala in our earlier analyses (**Figure 5B**), was calculated by weighting equally the two possible gender choices.

On the other hand, we found that FFA activity to HSF (in Controls) and to LSF (in Controls for the right hemisphere, and ASD individuals in the left hemisphere) was globally unaffected by participants' behavior or by emotional valence [|*t*|(9) always <1.60]. Visual inspection of **Figure 7D**, suggests that, in ASD individuals, the processing of LSF happy expressions might elicit larger activity in left FFA for those trials in which a HSF gender was chosen as opposed to a LSF gender. This was confirmed by an *ad-hoc* comparison [LHH vs. LHL, *t*(9) = 2.76, *p* < 0.05].

Finally, we extended the results obtained in FFA and amygdala to the whole brain, by assessing for each group the putative effects of the behavioral choice (HSF vs. LSF gender) and of emotional valence (fearful vs. happy). However, this analysis led, in its most relevant contrasts, to no suprathreshold activity. Specifically, neither Controls nor ASD individuals exhibited any suprathreshold effect associated with emotional valence, neither when testing the overall main effect [contrast (*LFL* + *LFH* + *HFL* + *HFH*) − (*LHL* + *LHH* + *HHL* + *HHH*) and inverse], nor when focusing only on those trials in which emotions were conveyed by specific frequency bands [contrasts (*LFL* + *LFH*) − (*LHL* + *LHH*), (*HFL* + *LFH*) − (*HHL* + *HHH*), and inverses]. No suprathreshold effect was found when testing whether there

were regions affected by participants' choice [contrast (*LFL* + *LHL* + *HFL* + *HHL*) − (*LFH* + *LHH* + *HFH* + *HHH*) and vice versa]. As in the case of the amygdala (**Figures 7A,B**), we tested putative effects of choice within those frequency bands conveying emotional information (choose HSF > choose LSF only for HSF emotional expressions, or choose LSF > choose HSF only for LSF expressions), but no significant effect was found in any of the groups. No suprathreshold effect was found associated with the interaction between the frequency conveying an emotional expression and the frequency promoting the gender choice, specifically when searching for regions with higher activity in trials in which the face with an emotional expression was chosen rather than neglected [contrast (*LFL* + *LHL* + *HFH* + *HHH*) − (*LFH* + *LHH* + *HFH* + *HHH*)]. Finally, in keeping with our behavioral finding that participants' response was affected by the emotional content of happy expressions only, we excluded from the interaction contrast those trials displaying fearful faces [contrast (*LHL* + *HHH*) − (*LHH* + *HHH*)], but even in this case we found no suprathreshold activity.

## *Passive viewing trials*

Finally, all effects associated with the passive viewing trials are displayed in **Table 4** and can be summarized as follows. No region was uniquely recruited by the perception of LSF emotional expressions as opposed to neutral stimuli (*LF* + *LH*)/2 − *N*, neither in Controls nor in ASD individuals. Instead, ASD individuals (but not Controls) exhibited increased activity in the right fusiform gyrus for HSF expressions [contrast (*HF* + *HH*)/2 − *N*], in proximity to the region identified through the same contrast in Controls when testing the gender discrimination sessions (see **Figure 8A**, green blob). We then inspected any effect of emotional valence, both as a global main effect [contrast (*LF* + *HF*) − (*LH* + *HH*) and inverse] and by analyzing separately each frequency band. Controls exhibited only enhanced activity of the most anterior portion of the left fusiform gyrus, extending to the parahippocampal cortex, for HSF happy (relative to HSF fearful) expressions (see **Figure 8A**, red blobs).

On the other hand, ASD individuals exhibited increased activity in the left middle-anterior insula for fearful (as opposed to happy) expressions, regardless of the spatial frequency. Such effect was not observed when repeating the analysis separately for each frequency band. Furthermore, in ASD individuals, LSF happy expressions triggered (compared to LSF fearful expressions) enhanced activity in the most ventro-lateral part of the right amygdala (**Figure 8B**, green blob). Finally, in ASD individuals, the contrast (*HH* − *HF*) elicited significant differential activation in the temporo-parietal junction (bilaterally), the posterior cingulate cortex and the left superior frontal sulcus.

In sum, in sharp contrast with the case of the Gender Discrimination task, during the passive viewing sessions ASD individuals exhibited increased neural responses in portions of the core face network for emotional facial expressions, including those conveyed by HSFs.

## **DISCUSSION**

We tested for the independent contribution of HSF or LSF visual inputs to brain regions critical for face processing, by engaging individuals with ASD and matched neurotypical Controls in a

**FIGURE 7 | Effects of the reported gender and of emotional valence.** Average parameters extracted from representative voxels of **(A)** right Amygdala, **(B)** left Amygdala, **(C)** right FFA and **(D)** left FFA. The left and right FFA voxels were chosen as exhibiting significant conjoint activity for the contrasts *LSF* − *N* (as shown in **Figure 5A**) and *HSF* − *N* (**Figure 5B**). Amygdalar voxels were those composing the clusters depicted in **Figure 5B**. The four regions are displayed in yellow on a ventral portion of an inflated human brain. For each of these four regions, average parameters estimates are displayed with bootstrap-based 95% confidence intervals of the mean. Data from different groups are displayed in separate subplots. Empty bars refer to trials in which participants choose the gender depicted

by the LSF, whereas dotted bars refer to trials in which the HSF gender was chosen. The average activity associated with the neutral condition is displayed as a gray horizontal dash-dotted line. The portions of the bars which exceed the activity of the neutral condition are colored according to the functional test with which the regions were defined. Regions isolated through the contrast *LSF* − *N* (**Figure 5A**) display the bars associated with LSF conditions colored in blue; instead regions isolated through the contrast *HSF* − *N* (**Figure 5B**) display the bars colored in orange. *HF*, HSF fearful expression; *LF*, LSF fearful expression; *HH*, HSF happy expression; *LH*, LSF happy expression; *N*, Neutral expression; *Amy*, amygdala; *FFA*, Fusiform Face Area; *r* and *l*, right and left hemisphere respectively.

gender discrimination task with hybrid face stimuli. We found that, compared to Controls, individuals with ASD exhibited a reduced sensitivity to emotional information conveyed by the cortical HSF pathway, but were as sensitive as Controls to information conveyed by the LSF pathway. This was observed both in the portion of fusiform gyrus sensitive to face stimuli (FFA), when measuring the neural responses to emotional expressions in either frequency against control neutral faces (**Figure 5**), and in both the ventral occipital cortex and the amygdala when testing HSFs against LSFs (**Figure 6**, red blobs). FFA, the ventral occipital cortex and the amygdala were also showed a significant interaction reflecting that the increased activity for HSF expressions observed in Controls was reliably larger than this effect in ASD individuals (**Figure 6**, yellow blobs). Furthermore, both FFA and amygdala responses to emotional cues seem independent from the emotional valence, whereas they were modulated by the participants' choices in the gender task—at least for the amygdala (**Figure 7**). Critically, these data cannot be interpreted as ASD being characterized by a generalized insensitivity to HSF cues conveyed by the cortical pathway *per se*, because posterior visual cortical regions responded to HSF emotional information in ASD individuals during the passive viewing sessions (**Figure 8**). Instead, the data suggest decreased sensitivity to HSF information when processing global facial features, such as during active gender discrimination.

#### **LOW- AND HIGH-FREQUENCY PROCESSING IN ASD**

The hybrid nature of our stimuli, and LSF and HSF cut-offs adopted in keeping with our previous studies (<6 c/fw and >24,


**Table 4 | fMRI analysis: regions showing significant activation associated with the passive viewing sessions.**

*†p* < *0.001; ‡p* < *0.01; \*p* < *0.05 corrected at the cluster level for the whole brain (underlying height threshold: p* < *0.001, uncorrected).*

*1p* < *0.05 corrected at the voxel level for FFA and amygdala bilaterally as described by the localizer data.*

**FIGURE 8 | Passive viewing sessions. (A)** Whole-brain maps showing significant increase of neural activity associated with the contrast *HH* − *HF* in Controls (red blob) the contrast *HSF - N* in ASD individuals (green blob). Activations are overlaid on an inflated brain surface. **(B)** Coronal sections (*y* = −3, 2) displaying the increase of neural activity for the contrast *LH* − *LF* in ASD individuals (green blobs) in the right amygdala. For both **(A,B)** portions of fusiform and amygdalar cortex implicated in the contrast *HSF* − *N* in earlier analysis on the gender discrimination sessions are displayed in yellow. *Fus*, Fusiform Gyrus; *r* and *l*, right and left hemisphere respectively. *dmA* and *vlA,* dorsomedial and ventrolateral portions of the amygdala.

c/fw respectively—Vuilleumier et al., 2003; Winston et al., 2003b; Pourtois et al., 2005) served two main purposes: first, they allowed co-occurrent, and yet dissociable, recruitment of parvocellular and magnocellular pathways; second they insured that spatial frequency information conveyed by each pathway provided coarse (LSF) and fine-grain (HSF) facial cues that were equally distant from optimality. With this set of stimuli, we found no behavioral difference between ASD individuals and Controls. Group differences were observed only when measuring neural responses, specifically for fine-grained information that is uniquely conveyed by the cortical pathway. Earlier studies using gratings stimuli of LSFs or HSFs found comparable contrast thresholds in ASD individuals and Controls (Bertone et al., 2005; De Jonge et al., 2007; but see Davis et al., 2006, for differences in HSF), but nevertheless documented atypical neural responses in ASD (for LSF, Boeschoten et al., 2007b; Vlamings et al., 2010; for HSF Boeschoten et al., 2007b; Milne et al., 2009).

Faces are much more complex stimuli as they are processed through the integration of co-occurrent HSF and LSF information arising from each pathway. Notably, earlier studies using simple band-pass filtered or hybrids faces often reported that ASD individuals might be more biased in favor of HSF than LSF (Deruelle et al., 2004, 2008), and exhibit atypical neural responses to LSF faces (Vlamings et al., 2010). It should be mentioned, however, that these studies differed from ours in many aspects, such as the recruitment of children (see Rondan and Deruelle, 2004, for a lack of effects in adults), the task employed (see Deruelle et al., 2008, who reported no HSF biases in gender discrimination task) and, critically, the use of a more liberal LSF cutoff (<12 c/fw). Indeed, psychophysical investigations in neurotypical individuals have consistently described that face information is optimally processed from intermediate frequency bands (between 8-16 c/fw – Costen et al., 1994, 1996; Gold et al., 1999; Näsänen, 1999; Parker and Costen, 1999; Boutet et al., 2003; Collin et al., 2006; Watier et al., 2010). In this perspective, previous studies should not be interpreted as showing atypical processing of LSFs *per se*, but of those intermediate frequencies optimal for face processing. Consistently with this conjecture, a study employing face stimuli filtered under a more stringent LSF cutoff (< 5 c/fw—thus, outside the range 8–16 c/fw) reported no difference in neural activity between ASD children and matched Controls (Boeschoten et al., 2007a).

#### **GENDER DISCRIMINATION IN ASD**

The gender discrimination task employed here served the purposes of our study in three critical aspects. First, this task chiefly requires the inspection of faces from a global point of view, as shown by decreased performance when the face stimuli are inverted, scrambled, or when the upper and lower halves are misaligned (Zhao and Hayward, 2010). Second, the gender of hybrid stimuli can be discriminated by relying on either LSF or HSF bands (equally from both bands in Schyns and Oliva, 1999; slightly LSF-biased in Winston et al., 2003b). Third, the discrimination is influenced by the (taskirrelevant) emotional expressions of one of the two faces composing the hybrid, as shown by our behavioral data: happy expressions bias the judgment toward the frequency bands in which these are conveyed (see **Figure 3**), suggesting that gender discrimination itself might actually be combined with a parallel and automatic extraction of the emotional information from the face, including its valence (Vuilleumier, 2007; Vuilleumier and Righart, 2011). In Controls, the increase of activity in the fusiform gyrus when either frequency band conveyed an emotional expression might be a neural signature of such extraction.

In the ASD group, no increase of neural activity was associated with HSF emotional expressions during active gender discrimination, suggesting lower use of emotional information from HSF in this condition, or alternatively increased efficiency at ignoring task-irrelevant information from one specific frequency band. In this perspective, one might expect ASD individuals to be conversely more biased toward LSFs than Controls in their judgments on hybrid faces, a pattern also suggested by visual inspection of behavioral data in **Figure 3**. Unfortunately, group differences in these behavioral results did not reach statistical significance. In any case, the reported differences in brain activation cannot merely be explained by performance, as LSF- and HSF-biased trials were modeled independently in each participant, and both weighted equally on all subsequent analytical stages regardless of individual idiosyncratic response-biases. We are therefore confident that our results truly reflect differences in visual perceptual processing.

#### **GLOBAL AND LOCAL PROCESSING IN ASD**

At a first sight, ASD's decreased sensitivity to high-frequency information (only during the discrimination task) might be considered at odds with a large body of evidence suggesting how ASD processing of visual stimuli might be biased in favor of detailed (fine-grain) information, at the expense of the global picture. Indeed, ASD individuals have been reported to be more proficient than Controls in tasks in which the global information conflicts with locally-displayed targets (Shah and Frith, 1983; Happé, 1996; Pellicano et al., 2005; Simmons et al., 2009; Almeida et al., 2013) but, at the same time, less proficient in detecting coherent global patterns when intermingled with distracting local information (Spencer et al., 2000; Milne et al., 2002; Pellicano et al., 2005; Spencer and O'Brien, 2006; Tsermentseli et al., 2008).

Please note, however, that the distinction between local vs. global information from earlier studies is not necessarily equivalent to a distinction between HSF vs. LSF information. Indeed, whereas local information is indubitably conveyed by HSF, global information can, at least in principle, be obtained by all frequency ranges, with some critical differences: on the one side, LSF provides global cues from visual stimuli (e.g., a face) *regardless* of local information, instead HSF can provide global cues by integrating multiple local details together. In this perspective, our findings of decreased HSF-related activity in ASD can be reconciled with earlier accounts only under the assumption that ASD local biases are reflective of a difficulty in seeing the whole through the integration of many details. Consistently with this assumption, recent studies investigated visual integration by using two independent kinds of stimuli: (1) stimuli whose global properties are retained regardless of the details (hierarchical figures, Navon, 1977), for which ASD individuals perform comparably to Controls Deruelle et al., 2006; Rondan and Deruelle, 2007; (2) stimuli whose global properties are retained only from the combined information of many local features (e.g., gestalt illusions of similarity, proximity, etc.), for which ASD individuals exhibit difficulties relative to Controls (Brosnan et al., 2004; Deruelle et al., 2006; Bölte et al., 2007; Rondan and Deruelle, 2007). Please note that in the former kind of stimuli, the global information was available at a coarse level of resolution, thus retainable even after low-pass spatial filtering. This is not necessarily the case for the latter kind of stimuli, in which the global information may also be obtained from information at a more fine-grain level (see also Dakin and Frith, 2005; Simmons et al., 2009 for similar arguments in contour integration tasks).

## **FUSIFORM AND AMYGDALA FUNCTION IN ASD**

Although many behavioral studies failed at documenting differences in face processing between ASD individuals and Controls, more systematic effects were reported by fMRI studies including reduced neural responses in the fusiform gyrus and the amygdala when processing (emotional or neutral) facial expressions (Baron-Cohen et al., 2000; Critchley et al., 2000; Schultz et al., 2000; Pierce et al., 2001; Hall et al., 2003; Hubl et al., 2003; Wang et al., 2004; Grelotti et al., 2005; Ashwin et al., 2007; Scherf et al., 2010). These results were first interpreted as ASD being characterized by an atypical development of the fusiform gyrus and/or the amygdala (e.g., Baron-Cohen et al., 2000; Schultz, 2005). However, as for other accounts that attempt to describe ASD symptomatology with the dysfunction of specific brain regions (e.g., the broken mirror hypothesis, Hamilton, 2013), these anatomical models are subjected to several critiques. First, some processes associated with the incriminated regions are often spared in ASD individuals (e.g., amygdala dysfunction should also impair emotional arousal, aversive conditioning, or reward contingency learning, but these impairments were not consistently found across studies testing ASD individuals; Gaigg, 2012; see also Zalla and Sperduti, 2013). Second, lesions in incriminated regions, even when occurring at early stages of life, do not lead to the same symptomatology of ASD (Amaral et al., 2003; Paul et al., 2010, but see Bachevalier, 1994). Third, recent studies often report comparable functional properties in the incriminated regions between ASD individuals and neurotypical Controls, when controlling for factors such attentional load, stimuli presentation time or eye movements (Hadjikhani et al., 2004, 2007; Dalton et al., 2005; Bird et al., 2006). In this perspective, ASD might not be associated with damaged fusiform gyrus or amygdala *per se*, but with atypical recruitment/modulation of these regions by high-order top-down control or attentional processes (Santos et al., 2008). Also in our ASD sample the fusiform gyrus and the amygdala did not appear to be generally impaired, e.g., due to either a regional dysfunction or a general atypicality in gazing behavior—but rather this group exhibited a selective hypoactivation for a specific class of information (HSF emotional expressions in hybrid images) and under specific task demands (gender discrimination).

Moreover, Kleinhans et al. (2008) reported decreased functional connectivity between fusiform gyrus and amygdala when ASD participants processed face stimuli, pointing to a dysfunction at the network level rather than at each of its constituent nodes. We concur with this interpretation, but also extend it by offering further insights on the nature of the dysfunction. As shown in **Figure 1**, the amygdala is thought to receive coarse (LSF) facial information from a direct subcortical (i.e., collicularpulvinar) path, which may then project back to the fusiform (Winston et al., 2003b), whereas in addition the fusiform gyrus also receives fine-grained (HSF) information from a feedforward (i.e., geniculo-striate and ventral occipitotemporal) cortical path. Critically, cortical and subcortical processing of faces are integrated with each other, as shown by enhanced functional connectivity between amygdala and fusiform gyrus during face processing (Morris et al., 1998), and by the impact of amygdala damage on fusiform sensitivity to facial emotional expressions (Vuilleumier et al., 2004). Thus, within this model, we can distinguish between two independent components of the amygdala-fusiform connectivity according to the direction of the information flow. Signaling *from the amygdala to the fusiform gyrus* is supported by the modulation of fusiform responses by LSF facial information initially processed in the amygdala (see **Figure 4A**, but also Vuilleumier et al., 2003; Winston et al., 2003b). Conversely, signaling *from the fusiform to the amygdala* is consistent with amygdala responses being also sensitive to HSF facial information conveyed by the visual cortex (see **Figure 5**). Our data provide novel evidence suggesting that it is the signal in the latter (but not the former) direction that exhibits atypical properties by ASD. This in turn suggests that integrative face processing functions mediated by higher level visual cortices might be more affected by ASD than lower level subcortical pathways providing inputs to the amygdala.

#### **LIMITATIONS OF THE STUDY AND FUTURE RESEARCH**

Like many other neuroimaging investigations on autism, including those reviewed in this article, our dataset is penalized by the limited number of participants and by an ASD population including both individuals affected by Asperger Syndrome and High Functioning Autism (see **Table 1**). Low power is not necessarily detrimental for positive results, which in our case were all obtained under corrected statistical thresholds (see also Friston, 2012), but it is problematic for those tests producing null or marginal results and for which an effect could potentially still be found with a larger sample. Also the heterogeneity of the clinical sample might be an additional source of noise with detrimental effects on the power of statistical analysis. Furthermore, some of the effects might be driven by only one of the two clinical sub-groups without a possibility of further verification on corresponding subsamples. It should be stressed, however, that the distinction between Asperger Syndrome and High Functioning Autism was removed in the last edition of the Diagnostic and Statistical Manual of mental disorders (American Psychiatric Association, 2013). In this perspective, putative heterogeneities in our clinical population should be treated as any within-group variability against which the significance of effects is estimated.

In particular, low power and sample heterogeneity might account for the weak effects of valence of emotional face expressions. Indeed, participants' behavior in both groups was significantly affected by valence, while the analysis of the fMRI signal did not reveal a similar effect in the brain. We should stress, however, that this consideration is not critical for our main results, since a lack of valence effects in the fusiform gyrus and the amygdala is plausible with respect to earlier accounts (Sander et al., 2003; Surguladze et al., 2003; Winston et al., 2003a). Interestingly, however, during the passive task, ASD individuals exhibited increased activity for LSF happy as opposed to LSF fearful expressions in the right amygdala. This activation arose in a ventrolateral portion of the amygdala, whereas earlier effects associated with the discrimination task arose in a more dorsal and medial location (**Figure 8B**). Parcellation of the human amygdala has been carried out with both cytoarchitectonic (Amunts et al., 2005) and connectivity-based approaches (Bzdok et al., 2013), and suggest that the different effects in **Figure 8B** might concern different sub-regions. Future research with high-resolution fMRI techniques is needed to investigate more specifically how ASD impairments in face and emotional processing might relate to different subregions of the amygdala.

Furthermore, caution should be used to interpret group differences in their response to LSF expressions relative to the neutral control condition because, unlike for HSFs, no significant interaction with the group factor was found. We can therefore not conclude whether the lateralization displayed in **Figure 5A** is truly reflective of different network-organization in the two groups. Please notice that, although left FFA was identified only when testing ASD individuals, visual inspection of the parameters extracted from this region suggests that a similar effect might be present also in Controls. It is therefore plausible to assume that, like for HSFs, the greater sensitivity of Controls to LSF expressions might extend to both hemispheres.

Finally, although we are quite confident that in the gender discrimination task participants focused their attention on global aspects of the face stimuli (Zhao and Hayward, 2010), we have little control on which processes were at play during the passive viewing session, in which the only instruction was to watch the stimuli attentively. Furthermore, even if participants focused on face stimuli, we do not know whether they preferentially attended to global or local properties or shifted between both. In this perspective, the increase of neural activity observed in ASD individuals for HSF emotional information in the passive condition (**Figure 8**) can only be taken as evidence for spared functionality of the cortical path outside the demands of the discrimination task (see Discussion section above). Future studies will need to extend these results by using other tasks in which participants are forced to focus on local facial details, thus allowing us to determine the neural signatures associated with featural facial processing in addition to the frequency content manipulation used here.

## **REFERENCES**


autism spectrum disorder view faces. *Neuroimage* 22, 1141–1150. doi: 10.1016/j.neuroimage.2004.03.025


emotional facial expressions. *Brain J. Neurol*. 121(Pt 1), 47–57. doi: 10.1093/ brain/121.1.47


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 30 October 2013; paper pending published: 10 December 2013; accepted: 14 March 2014; published online: 09 April 2014.*

*Citation: Corradi-Dell'Acqua C, Schwartz S, Meaux E, Hubert B, Vuilleumier P and Deruelle C (2014) Neural responses to emotional expression information in highand low-spatial frequency in autism: evidence for a cortical dysfunction. Front. Hum. Neurosci. 8:189. doi: 10.3389/fnhum.2014.00189*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2014 Corradi-Dell'Acqua, Schwartz, Meaux, Hubert, Vuilleumier and Deruelle. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The amygdala and the relevance detection theory of autism: an evolutionary perspective

# *Tiziana Zalla1\* and Marco Sperduti 2,3*

<sup>1</sup> Institut Jean Nicod, Centre National de la Recherche Scientifique, Ecole Normale Supérieure, Paris, France

<sup>2</sup> Laboratoire Mémoire et Cognition, Institut de Psychologie, Université Paris Descartes, Boulogne-Billancourt, France

<sup>3</sup> Inserm U894, Centre de Psychiatrie et Neurosciences, Université Paris Descartes, Paris, France

#### *Edited by:*

Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland

#### *Reviewed by:*

Sebastian B. Gaigg, City University London, UK Bhismadev Chakrabarti, University of Reading, UK Danilo Bzdok, Research Center Jülich, Germany

#### *\*Correspondence:*

Tiziana Zalla, Institut Jean Nicod, Centre National de la Recherche Scientifique, Ecole Normale Supérieure, 29 rue d'Ulm, 75005 Paris, France e-mail: tiziana.zalla@ens.fr

In the last few decades there has been increasing interest in the role of the amygdala in psychiatric disorders and, in particular, in its contribution to the socio-emotional impairments in autism spectrum disorders (ASDs). Given that the amygdala is a component structure of the "social brain," several theoretical explanations compatible with amygdala dysfunction have been proposed to account for socio-emotional impairments in ASDs, including abnormal eye contact, gaze monitoring, face processing, mental state understanding, and empathy. Nevertheless, many theoretical accounts, based on the Amygdala Theory of Autism, fail to elucidate the complex pattern of impairments observed in this population, which extends beyond the social domain. As posited by the Relevance Detector theory (Sander et al., 2003), the human amygdala is a critical component of a brain circuit involved in the appraisal of self-relevant events that include, but are not restricted to, social stimuli. Here, we propose that the behavioral and social–emotional features of ASDs may be better understood in terms of a disruption in a "Relevance Detector Network" affecting the processing of stimuli that are relevant for the organism's self-regulating functions. In the present review, we will first summarize the main literature supporting the involvement of the amygdala in socio-emotional disturbances in ASDs. Next, we will present a revised version of the Amygdala Relevance Detector hypothesis and we will show that this theoretical framework can provide a better understanding of the heterogeneity of the impairments and symptomatology of ASDs. Finally, we will discuss some predictions of our model, and suggest new directions in the investigation of the role of the amygdala within the more generally disrupted cortical connectivity framework as a model of neural organization of the autistic brain.

**Keywords: autism spectrum disorders, amygdala, ventromedial prefrontal cortex, self-relevance, social brain**

# **INTRODUCTION**

Autism spectrum disorders (ASDs) are pervasive developmental disorders characterized by a triad of symptoms including abnormal socio-emotional processing, verbal and non-verbal communication problems, and restricted interests and repetitive behaviors (American Psychiatric Association, 2000). Although there is now substantial evidence implicating genetic bases and brain mechanisms in ASD etiopathology (see Eapen, 2011), there is no apparent core neurocognitive dysfunction associated with a single structure that could esaustively explain the variety of symptoms observed in these disorders. Current data rather suggest that multiple perceptual and cognitive processes subserved by different neural systems are affected. However, it is possible that the dysfunction of a single structure of an interconnected neural circuit, such as a circumscribed damage to the amygdala, can influence other areas of the circuit and have widespread repercussions on multiple cognitive functions, especially if this occurs early in development (Bachevalier, 2005).

Several theories have been proposed to account for the atypical pattern of socio-emotional behavior in ASDs. The most influential are the *Theory of Mind* (Baron-Cohen et al., 1985; Baron-Cohen, 1995), the *Socio-emotional* theory (Hobson, 1993), the *Social Motivation* theory (Grelotti et al., 2002; Schultz et al., 2003; Schultz, 2005), and the *Fast-track modulator* model (Senju and Johnson, 2009). While a full description of these theories is beyond the scope of the present paper, and we direct the interested reader to two recent extensive reviews (Gaigg, 2012; Hamilton, 2013), what we intend to emphasize here is that all these proposals are compatible with a core deficit of the so-called "social brain" in which the amygdala is the key component.

Alternatively, in the present review, we will acknowledge that the function of the human amygdala is better characterized in terms of a *self-relevance detection* system (Sander et al., 2003) and, based on theoretical argument and experimental support taken from cognitive neuroscience and evolutionary biology, we argue that abnormalities in this structure associated with the disruption of this self-relevance detection system would potentially explain a larger variety of impairments and symptomatology of ASDs that include, but are not restricted to, the social domain. In this view, the amygdala, originally designed to automatically detect potentially threatening or dangerous environmental events under ancestral conditions, has enlarged its domain of specificity

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 1 — #1

in humans to respond to a broader range of self-relevant information in the physical and social environment, including intrinsic biological features and extrinsic context-dependant information. As previously defined (Sander et al., 2003), an event is relevant for an organism if it significantly influences (positively or negatively) the attainment of his or her goals, the satisfaction of his or her needs, the maintenance of his or her well-being within the physical environment and the social context. Following the theoretical account advocated by the *Relevance Detection Theory of the Amygdala* (Sander et al., 2003), the present proposal aims to specify the role of the amygdala in the ASD etiopathology by highlighting the notion that this structure is a key component of a larger interconnected fronto-limbic neural system. Because of its complex functional connectivity with the ventromedial prefrontal cortex (vMPFC), a stimulus is deemed relevant through two distinct processes of salience attribution: (a) the intrinsic salience of a stimulus, which is determined by its biologically innate (e.g., threat, food) or physical (e.g., intensity or novelty) features, via stimulus-driven bottom-up low level processes, and (b) the extrinsic or *context-dependent* salience which can be assigned flexibly through top-down evaluative processes.

In the present work, we will first discuss the early formulation of the "amygdala theory" of autism (Baron-Cohen et al., 2000), and review research on amygdala function in subjects with and without ASDs that calls into question the specific social view as stated by the "hard" formulation of the "amygdala theory." We will then propose a more general view based on the notion that the amygdala is a critical component of a brain circuit responsible for the detection of relevant stimuli or events, and crucially for the formation of a "salience map" that integrates and prioritizes salience signalsfrom various sources of information, in accordance with the motivations and the contextual goals of the perceiver.

#### **THE "AMYGDALA THEORY OF AUTISM"**

Based on research on animal lesion (Kling and Brothers, 1992), single cell recording studies (Brothers et al., 1990) and neurological studies, Brothers (1990) has proposed that a brain network including three regions, the amygdala, the orbitofrontal cortex (OFC), and the superior temporal gyrus (STG), constitutes the neural basis of social intelligence, the so called "social brain." Given that social perception impairment, abnormal gaze behavior and emotional processing are central to the autistic symptomatology, it is not surprising that a great emphasis has been placed on amygdala involvement in the etiopathology of this condition. Nevertheless, the exact role of this structure in the behavioral deficit of ASDs, above all in the social domain, is still a controversial issue.

Baron-Cohen et al. (2000) posited that damage or dysfunction of the amygdala should be at the root of social impairments in ASDs and proposed the *Amygdala Theory of Autism*. The "hard" formulation of this theory states that "...the amygdala is one of several neural regions that are *necessarily* abnormal in autism" (Baron-Cohen et al., 2000, p. 1; emphasis added). To sustain this claim, the authors presented converging evidence coming from animal models, post-mortem and structural studies showing abnormalities in the amygdala in ASDs, as well as behavioral similarity between subjects with ASDs and patients with amygdalotomy. Furthermore, the authors reported a fMRI study on adults with high functioning autism (HFA) and Asperger syndrome (AS) showing difficulties in identifying mental state/emotional information from the eyes of others (reading the mind in the Eyes task) that was associated with weaker amygdala activation, as compared to typically developing subjects. The reduced amygdala response to the intentional meaning of the emotional expressions in adults with ASDs is consistent with a large amount of studies reporting atypicalities in face processing in infants in the first 6 months of life (Maestro et al., 2002) and abnormal fixation to the eye region in adults (Klin et al., 2002; Pelphrey et al., 2002; Adolphs et al., 2005; Dalton et al., 2005).

However, evidence from animal research seems to challenge this hypothesis. In a series of studies, Emery et al. (2001) used the rhesus monkey as a model system to examine the role of the amygdala in conspecific social behavior, and showed that, in dyadic social interactions, adult monkeys with extensive bilateral lesions of the amygdala can decode and generate social gestures and initiate and receive more affiliative social interactions than control monkeys. Importantly, the monkeys exhibited abnormal response to normally fear-inducing stimuli such as snakes, and the normal reluctance to engage with a novel animal was eliminated. Reduced fear response and socially uninhibited behavior were also observed in primates at 2 weeks of age with bilateral lesions of the amygdala (Prather et al., 2001).

More recently, the specific role of the amygdala in social cognition in ASDs has been questioned in a study of two rare patient cases suffering from Urbach–Wiethe disease, which is characterized by a developmental selective atrophy of the bilateral amygdala (Paul et al., 2010). In fact, even if these patients reported some social deficit associated with ASD symptomatology, their overall performance on the standard diagnostic test for ASDs and in clinical examination did not reveal a clear association with ASD symptomatology. A direct evidence comes from a recent study by Birmingham et al. (2011) showing that the differences between individuals with amygdala lesions and ASDs are more striking than the similarities. Indeed, while patients with amygdala damage failed to attend to social features in stimulus-driven manner, but showed an intact modulation of eye gaze by the task, the ASD group exhibited a notable absence of such task-dependent modulation. The authors concluded that social disturbance in ASDs would be better understood in terms of a disruption of the complex network of structures with which the amygdala is connected rather than in the amygdala itself.

Because of its widespread functional connections with sensory, associative areas and autonomic systems, the amygdala is regarded as a "sensory gateway" and plays an important role in the integration of a wide array of visceral, sensory, and cognitive information (Freese and Amaral, 2009). The fact that the amygdala receives projections from both subcortical and cortical pathways confirms the view that multiple processes may be engaged depending on the type of information involved, but of particular interest for the present proposal are the functional connections with the vMPFC which relay amygdala input to regions involved in more deliberate forms of decision making reasoning and cognitive control. It is,

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 2 — #2

in fact, well documented that the amygdala has multiple connections to prefrontal areas, receiving from and relaying information to areas of insular, OFC, and lateral prefrontal cortex (Stefanacci and Amaral, 2000). These reciprocal connections extend the functionality of the amygdaloid structure which is responsive to the entire state of the organism and contextual information (Mosher et al., 2010).

In a recent review, Gaigg (2012) discusses results from studies on emotional arousal, aversive conditioning, and reward contingency learning in ASDs and concludes that the results are globally inconsistent with the view that only emotions relevant to social cognition are compromised in ASDs. Noteworthy, the author emphasizes that theories uniquely based on a dysfunctional social brain network ignore multiple aspects of the interpersonal emotional disturbance and the more widespread anomalies in the general domain of emotions in ASDs. Overall, current findings in subjects with and without ASDs challenge the "hard" formulation of the *Amygdala Theory of Autism*, primarily grounded on the *social function* view of the amygdala, and question its role in ASD symptomatology.

# **THE AMYGDALA: ANATOMICAL AND NEUROIMAGING FINDINGS IN TYPICAL DEVELOPMENT**

The amygdala is an almond-shaped group of subcortical nuclei belonging to the limbic system situated in the deep medial temporal lobe. Even though, in most neuroimaging studies, it is considered as a whole, the amygdala is composed of several subnuclei that present specific cytoarchitectonic features and different patterns of connectivity with several subcortical and cortical structures. The amygdala has been divided into three major subdivisions1: the laterobasal, the centromedial, and the superficial nuclei, each of them being associated with a specific coactivation profile (Bzdok et al., 2012).

While anatomical connectivity of the amygdala has been largely elucidated by non-human primate studies (Amaral and Price, 1984; Barbas and De Olmos, 1990), with the advent of non-invasive neuroimaging techniques, an increasing number of studies have been devoted to determining the functional connectivity of the human amygdala. The amygdala does not work in isolation, but rather serves as a complex node within multiple neural networks (Pessoa, 2008). Using a connectivity-based parcellation approach, Bzdok et al. (2012) identified three distinct clusters in human amygdala based on their brain-wide coactivation maps. These analyses revealed that the laterobasal nuclei group of the amygdala is linked with the integration of high-level sensory inputs (visual, auditory, gustatory, somatosensory, and, in part, olfactory environmental information), and the representation of stimulus-value associations. Its centromedial nuclei group is, in turn, functionally associated with attentional, vegetative, and motor responses, while the superficial nuclei group is found to process olfactory information.

Research in humans (Roy et al., 2009; Robinson et al., 2010) is fundamentally in agreement with studies in macaque monkeys showing widespread connections of the amygdala with cortical and subcortical regions encompassing the anterior cingulate cortex (ACC) and the inferior and medial prefrontal cortex, the hippocampus and the parahippocampal gyrus, the temporal lobe and the insula (Amaral and Price, 1984; Barbas and De Olmos, 1990; Stefanacci et al., 1996; Ghashghaei and Barbas, 2002). Using probabilistic diffusion tensor parcellation, Bach et al. (2011) have shown that the superficial portion, approximately corresponding to the centromedial and the superficial nuclei, and the deep portion, corresponding to the basal nucleus, are preferentially connected with OFC and the temporal pole, respectively. Recently, using resting state data Mishra et al. (2013) replicated this pattern of connectivity and additionally reported that the superficial nucleus shows greater connectivity with motor and MPFC regions, while the deep nucleus is strongly functionally coupled with the middle frontal gyrus and inferior parietal lobe.

Functional neuroimaging studies have demonstrated that the amygdala is implicated in a large variety of cognitive and behavioral functions, including fear conditioning (Adolphs et al., 2005), memory formation (Packard and Cahill, 2001), learning of stimulus–reward associations (Baxter and Murray, 2002), social and affective processing (Anderson and Phelps, 2000; Hariri et al., 2002), appraisal of positive (winning) and negative (losing) emotions elicited during a competitive contest (Zalla et al., 2000), as well as in a multiplicity of high-order cognitive functions ranging from emotional control (Ochsner et al., 2004; Goldin et al., 2008) to self-awareness and social perception. In the domain of social cognition, a large variety of stimuli and situations are associated with amygdala activation in typical development, including gaze direction (Kawashima et al., 1999; George et al., 2001;Wicker et al., 2003), eye contact (Emery, 2000), face identity, trustworthiness (Adolphs et al., 1998), facial familiarity (Dubois et al., 1999), racial outgroup faces (Hart et al., 2000; Phelps et al., 2000), body movements (Bonda et al., 1996), attribution of others' mental states and communicative intents (Baron-Cohen et al., 1999; Hart et al., 2000; Portas et al., 2000).

Alternative views have emphasized the role of the amygdala as a mechanism for more general vigilance and attention orientation (Davis and Whalen, 2001). Along this line, Vuilleumier (2005) showed that, while normal subjects exhibited enhanced brain activity in visual areas for fearful faces, patients with amygdala lesions did not show the same effect, suggesting that the role of the amygdala is to modulate the processing of sensory input that might be relevant for its vital significance, both directly and by top-down signals. This function has also been demonstrated using tasks in which emotional information is prioritized and receives privileged access to consciousness and attentional resources (Vuilleumier and Schwartz, 2001).

Remarkably, although the amygdala is involved in processing a wide range of emotions, comprising positive ones (Costafreda et al., 2008; Sergerie et al., 2008; Bzdok et al., 2012), it has been suggested that it specifically responds to the degree of arousal and not to the valence of the stimulus (Small et al., 2003; Costafreda et al., 2008), and studies that have independently manipulated valence

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 3 — #3

<sup>1</sup>This subdivision, originally proposed by Amunts et al. (2005), *using* post-mortem cytoarchitectonic mapping, was recently replicated *in vivo* using diffusion tensor imaging (Solano-Castiella et al., 2010; Bach et al., 2011), high field structural MRI (Solano-Castiella et al., 2011), functional connectivity-based parcellation (Mishra et al., 2013), and meta-analytic procedures (Bzdok et al., 2012) reporting either twoor three-cluster solutions.

and intensity have provided evidence that amygdala activity is preferentially involved in processing the affective intensity rather than the valence of the event (Anderson et al., 2003; Small et al., 2003). Interestingly, Cunningham et al. (2008) showed that, in concert with other neural components of evaluative processing, the amygdala may respond flexibly to the valence and intensity of stimuli in goal-congruent fashion, although it processes negativity in a less flexible fashion than positivity.

# **THE AMYGDALA: ANATOMICAL AND NEUROIMAGING FINDINGS IN ASDs**

A seminal post-mortem study on ASDs children reported abnormal cell packing primarily in the medial temporal lobe regions comprising the hippocampus and the amygdala (Bauman and Kemper, 1985). Preliminary volumetric *in vivo* studies of amygdala morphology in ASDs have reported contrasting evidence showing either increased (Abell et al., 1999; Howard et al., 2000), decreased (Aylward et al., 1999), or no difference (Haznedar et al., 2000) in amygdala volume. These contrasting results could be explained by several factors, such as the heterogeneity of the studied sample with respect to psychometric (e.g., IQ) and demographic (e.g., age) measures, differences in data analysis, or the small number of subjects in single studies. Recently, neuroimaging meta-analytic techniques have allowed researchers to partially overcome these limitations, allowing the pooling of large datasets.

In a meta-analysis of 19 voxel-based morphometry (VBM) studies, Duerden et al. (2012) reported a decrease in right amygdala volume in child/adolescent subjects with ASDs, but not in adults. However, more extensive meta-analyses have consistently reported volume decrease in the amygdala, particularly in the right hemisphere, even in adults with ASDs (Cauda et al., 2011;Via et al., 2011). In a more recent meta-analysis,Nickl-Jockschat et al. (2012) reported a significant decrease of gray matter volume in a cluster in the medial temporal lobe that did not include the amygdala. This is probably due to the fact that in the latter study authors used probabilistic cytoarchitectonic maps to localize anatomical regions corresponding to significant clusters of decreased gray matter volume. This approach is much more reliable, especially for structures whose anatomical borders are not as easily determinable as those of the amygdala, allowing assignment of cluster sites to histologically defined brain regions in a probabilistic fashion. Indeed, using the same localization approach, Ball et al. (2009) showed that only about 50% of peaks reported as amygdala activation across 114 functional neuroimaging studies could reliably be assigned with a probability ≥80% to this structure. A summary of the meta-analytic results described above is presented in **Table 1**.

It is noteworthy that, in comparison with studies reporting gray matter changes, there are few studies investigating the anatomical connection between the amygdala and other cortical–subcortical structures in ASDs. Preliminary diffusion tensor imaging (DTI) studies have reported, among other structures, altered fractional anisotropy (FA), a measure of fiber tracts integrity, in regions surrounding the amygdala (Barnea-Goraly et al., 2004; Noriuchi et al., 2010) or in tracts connecting the amygdala and the fusiform gyrus (Conturo et al., 2008). Other studies have shown reduced FA in specific fiber tracts, such as the inferior longitudinal fasciculus and inferior fronto-occipital fasciculus connecting the amygdala, the fusiform face area (FFA), and the superior temporal sulcus (STS; Jou et al., 2011), and in the uncinate fasciculus connecting the lateral and medial OFC with the anterior portion of the temporal lobe, including the amygdala (Radua et al., 2011).

In functional neuroimaging research, the involvement of the amygdala in the physiopathology of ASDs has been advocated either in terms of *hypoactivation* or *hyperactivation* of this structure. According to the "hypo-active models," the amygdala fails to process social stimuli as meaningful with the result that they do not receive preferential attention (Schultz, 2005), while in "hyperactive models," social stimuli are thought to cause an aversive over-arousal, with the result that they are actively avoided (Dalton et al., 2005; Corden et al., 2008). Structural and functional studies in ASD subjects failed, however, to report a systematic hypo- or hyperactivation of the amygdala. Baron-Cohen et al. (2000) found diminished amygdala activity in ASD subjects, but there is convincing evidence of amygdala hyperactivity in adults with ASDs when they gaze at the eye region (Dalton et al., 2005; Kliemann et al., 2012). In a more recent study by von dem Hagen et al. (2013), control subjects showed increased activation in the amygdala when contrasting neutral faces with direct *vs* averted gaze, while ASD participants showed an inverse pattern of activation.


Results for meta-analytic contrasts comparing gray matter volume between TD and ASDs. When available, results are separately reported for Ch–Ado and Adu. Only results concerning the amygdala are reported. Ch–Ado, children–adolescents; Adu, adults; Amy, amygdala; R, right; L, left; <, decreased gray matter volume in ASDs; =, no difference between groups.

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 4 — #4

Indirect behavioral evidence of amygdala functions in people with ASD are provided by fear conditioning protocols. For example, Gaigg and Bowler (2007) reported a pattern of abnormalities in differential fear conditioning in individuals with ASDs. In contrast, Hall et al. (2010) did not find any difference in brain activity between ASDs and control participants when presenting sub-threshold anxious expressions and, similarly, South et al. (2011)showed preserved fear acquisition in individuals with ASDs using aversive conditioning. Overall, these findings suggest that amygdala reactivity in ASDs is not absent, but response variability may depend on several factors, such as fixation to eye region, gaze avoidance and, as we argue in this review, more crucially on the abnormal fronto-amygdala connectivity associated with the diminished modulatory role of the vMPFC on this structure.

There is a growing consensus that the cognitive and behavioral disturbance in ASDs cannot be fully understood in terms of local dysfunction but are better viewed as impairments of functional networks (Kana et al., 2011). The fronto-amygdala disconnectivity explanation is consistent with a more general disrupted cortical connectivity framework (Belmonte et al., 2004), as a model of ASDs neural organization (Just et al., 2004; Geschwind and Levitt, 2007). Reduced activity of a fronto-parietal network was associated with a task requiring the flexible allocation of cognitive resources to guide goal-directed behavior in participants with ASDs (Solomon et al., 2009). Monk et al. (2010) reported altered connectivity between the right amygdala, subgenual vMPFC and middle temporal gyrus during emotional face processing, and diminished top-down modulation has been reported in studies using face processing and imitation (Cook et al., 2012). Disrupted connectivity between the OFC and the amygdala is supported by resting state data showing altered long-range connectivity in ASDs participants (Cherkassky et al., 2006; Anderson et al., 2011a,b) at both the structural and functional levels (Radua et al., 2011; Swartz et al., 2013).

Interestingly, Kleinhans et al. (2008) have shown that altered functional connectivity between the amygdala and the FFA during a face identification task correlates with the severity of social impairment in adults with HFA. In a subsequent study, Kleinhans et al. (2009) observed diminished amygdala habituation in response to neutral faces in subjects with ASDs, compared to subjects with typical development, and lower level of habituation correlated with the amount of social impairment. In accordance with these results, Swartz et al. (2013) showed that reduction of amygdala habituation to neutral and sad faces correlates with symptom severity, and that connectivity between the vMPFC and the amygdala was reduced in young subjects with ASDs. Overall, these data suggest that amygdala habituation correlates with symptom severity, and, that both phenomena could reflect the disrupted connectivity between the amygdala and the MPFC.

The idea of a key role of a single structure, the amygdala, seems difficult to reconcile with the view that the neuropathology of autism involves impaired widespread connectivity throughout the brain. However, as revealed by a recent study by Gotts et al. (2012), disrupted connectivity in high-functioning adolescents with ASDs, relative to typically developing adolescents, is most pronounced between limbic-related brain areas involved in affective processing, particularly in the amygdala and the vMPFC. More importantly, it has been shown that early damage to medial temporal lobe structures, including the amygdala, has widespread repercussions on other neural systems, such as a delayed maturation of the dorsolateral prefrontal cortex (Bertolino et al., 1997) and a dysregulation of striatal dopaminergic neurotransmission (Saunders et al.,1998). This view suggests that early developmental dysfunction in the medial temporal lobe (amygdala, hippocampus, and parahippocampus) in ASDs may cause a breakdown in brain connectivity that are normally recruited during complex cognitive tasks and trigger abnormal development of the prefrontal cortex (Dawson et al., 2002).

## **THE RELEVANCE DETECTION THEORY OF THE AMYGDALA**

The amygdala is a structure of the mammalian limbic system, shaped by evolution to rapidly and automatically detect potentially threatening or dangerous environmental events, and for learning about contingencies that are likely to predict similar events in the future (Amaral and Price, 1984; Öhman and Mineka, 2001; LeDoux, 2005). In virtue of its primary adaptive function, threatening or dangerous events are detected automatically and rapidly through the physiological mechanism of emotional arousal (Lang et al., 1993; Critchley et al., 2002). Emotional arousal allows recruiting additional cognitive and attentional resource allocation (Anderson and Sobel, 2003), facilitating access to awareness (Vuilleumier and Schwartz, 2001) and enhancing encoding and memory through an automatic process mediated by the sub-cortical amygdalar–hippocampal route (Kensinger and Corkin, 2004). Other findings indicate that the amygdala is important in the implicit processing of emotional stimuli. The inducing of amygdala responses by pre-attentively processed faces expressing threat (Vuilleumier et al., 2003) and fearful or happy faces (Juruena et al., 2010) presented by backward masking is thought to reflect the functioning of a primitive pathway specifically devoted to the rapid unconscious processing of socioemotional events encompassing explicit cognitive assessments (Sergent, 1994).

As posited by the "Relevance Detection Theory" (Sander et al., 2003), the human amygdala is a component of an extended neural cortico-limbic system involved in detecting stimuli by focusing attentional and physiological resources on cues that have special relevance for the safety or success of an organism within the broader context of its social life. As previously defined (Sander et al., 2003, p. 311), "*An event is relevant for an organism if can significantly influence (positively or negatively) the attainment of his or her goals, the satisfaction of his or her needs, the maintenance of his or her well-being, and the well-being of his or her species. According to this view, fearful and angry faces represent relevant information because they potentially obstruct one's goal and signal the presence of a danger for the organism and his or her con-specifics*."

From a phylogenetic perspective, in the primitive mammalian brain, the amygdala is part of a modular system shaped by evolution to detect potentially threatening physical events and biological stimuli (e.g., spiders, snakes), and to prepare the organism for action by facilitating escape and avoidance (LeDoux, 1996). MacLean (1970) provided an evolutionary explanation of emotion

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 5 — #5

and social intelligence. In particular, he proposed that emotions engage relatively primitive circuits that are conserved throughout mammalian evolution, along with the idea that structures in neocortex are specialized in cognitive and deliberative processing, such as action planning, decision making, and social cognition. Originally designed to signal potential threat and danger under ancestral conditions, the human amygdala has evolved, conjointly with the cortical structures, to serve to alert an organism toward a broader range of self-relevant information, including appetitive and aversive events coming from the internal milieu as well as from the physical and social environment to promote more adaptive behavior and flexible social exchanges. More crucially, it responds flexibly to stimuli whose relevance is contextually and cognitively modulated and is associated with various affective experiences (Cunningham et al., 2008).

Cross-species comparative studies have provided evidence of the co-evolution of the amygdaloid complex and the prefrontal areas in the neocortex (Barton and Aggleton, 2000) substantiating the view that this structure is a critical component of the integrative cortico-limbic network that constitutes an unitary evolved system for the detection of relevant events (Sander et al., 2003). The amygdala is involved in enhancing sensory processing and orienting visuo-spatial attentional resources toward salient features of the stimulus through both direct (amygdala–visual cortex) and indirect (amygdala–prefrontal cortex–visual cortex) connections, while the "quick-and-dirty" response relies on the activation of the arousal systems via the direct sub-cortical afferent route from all sensory modalities and the efferent connections with hypothalamic and brain-stem nuclei (LeDoux, 1995).

With respect to these distinct cortical pathways, one might distinguish the *intrinsic* and the *extrinsic* types of salience. While certain stimuli are intrinsically (or innately) self-relevant, because of their biological significance (e.g., threat, food, anger) or physical features (e.g., loudness, brightness, intensity, frequency of appearance, etc.), the extrinsic salience is flexibly acquired through context-dependant and conscious appraisal processes. Thus, the computational role of the human amygdala is twofold: on the one hand, it automatically and rapidly detects physically and biologically relevant information, via bottom-up processes, reflecting its more primary function; on the other hand, it integrates multiple salience signals originated via a top-down processes so as to create a priority map of intrinsically and extrinsically self-relevant information. Importantly, while the amygdala is specifically responsible for processing stimulus or event salience, which is a more fundamental feature since it is a measure of its importance, in a strict biological sense, value signals coding positivity for appetitive stimuli and negativity for aversive stimuli (Navalpakkam et al., 2010) are dynamically construed in vMPFC (Harris et al., 2011).

Direct evidence for this theory in humans is provided by neuroimaging studies. For example, food stimuli are more salient if we are hungry (LaBar et al., 2001) and very intense stimuli can lose their salience if they are repetitive, as shown by the habituation phenomenon (Marks and Tobeña, 1991). Morris and Dolan (2001) observed that amygdala activation was positively correlated with recognition memory scores for food items and that participants showed enhanced recognition of food stimuli (relative to non-food) in a fasting state. This enhanced recognition for food stimuli was significantly reduced when participants were in a satiated state. In accordance with this idea, Mohanty et al. (2008) investigated the neural mechanisms underlying attention toward food in participants when hungry and satiated, varying the relevance of the food stimuli. When hungry, participants showed increased amygdala activation to pictures of food and faster attentional orienting toward food cues as well as increased connectivity between limbic areas and parietal attention regions subserving attentional shifts, compared to when they were satiated.

Ousdal et al. (2008) reported increased amygdala activation toward letter stimuli, which are non-emotional and non-social, when the letters were targets in a go/no-go task and thus behaviorally relevant to participants' performance with respect to one's ongoing motivational state. In a further study using neutral taskdependant stimuli, Ousdal et al. (2012) suggested that when the relevance of a stimulus is determined by a specific task or context, the amygdala activity is modulated by cortical activity in the prefrontal cortex, based on context or prior knowledge.

Notably, results about people with ASD' performances in go/no-go tasks are mixed, with some studies reporting impaired performances (Ozonoff et al., 1994; Langen et al., 2012; Xiao et al., 2012) or only subtle difference (Geurts et al., 2009), while others showed comparable performances in this task (Happé et al., 2006; Schmitz et al., 2006; Lee et al., 2009). Xiao et al. (2012) showed that impairment in the go/no-go task was associated with decreased right prefrontal cortex activity during no-go blocks, while in Schmitz et al.'s (2006) study, increased prefrontal activity was found in ASD group for correct inhibited no-go trials. These results seem to suggest that in general prefrontal dysfunction is related to diminished performance in the go/no-go task, but that compensatory mechanism could be observed and lead to comparable performance thus explaining the contrasting behavioral results. The direct link between these studies in ASDs and that of Ousdal et al. (2012) is not straightforward since in classical go/no-go tasks the behavioral relevance of the stimulus (no-go) is not manipulated independently of its frequency. Thus, in this case, it is not easy to disentangle the role of frequency (that in our framework could be considered as "intrinsic salience") and behavioral salience (that in our framework could be considered as "extrinsic salience"). Overall, this handful of studies evidence the possibility that salient stimuli in these protocols (the no-go trials) are processed, at least in some cases, less efficiently in participants with ASD and that this might be linked to prefrontal cortex dysfunctions.

The amygdala also appears to be important in stimulus–reward association (Schoenbaum et al., 1998) or when magnitude of reinforcement needs to be maintained in working memory in order to accomplish a successful performance (Kesner andWilliams, 1995), in processing positive words (Hamann and Mao, 2002), positive pictures (Hamann et al., 1999, 2002; Canli et al., 2001; Garavan et al., 2001), pleasant tastes (O'Doherty et al., 2001), or expectation of pleasant tastes (O'Doherty et al., 2002).

In the socio-emotional domain, N'Diaye et al. (2009) have showed that amygdala response to facial emotion is modulated by interaction between the expressed emotion and gaze direction:

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 6 — #6

greater activation has been reported for fearful faces with averted gaze, signaling a possible threat, and for anger expression with direct gaze, signaling aggression. Similarly, increased activation in the amygdala was observed when contrasting neutral faces with direct *vs* averted gaze in control subjects indicating that an angry face is more *relevant* if the gaze is directed at the observer than if it is averted (von dem Hagen et al., 2013).

The role of the amygdala as a relevance detector is also consistent with neurophysiological findings in non-human primates showing that the neural response in this structure codes not only the raw value of a stimulus, i.e., the negative or positive representation of a stimulus, but also its "state value" (Morrison and Salzman, 2010). The latter takes into account the internal (e.g., hunger) and external (e.g., a specific rule) parameters of a given situation. One study also reported that amygdala is responsive to the subjective valence of emotional pictures, but not to the self-relatedeness of the same stimuli, which, however, did modulate the activity of MPFC (Phan et al., 2004). It has to be noted, however, that the two dimensions of self-relatedeness and selfreference are not clearly distinguished at both the conceptual and experimental levels. Moreover, the self-relatedeness task activated regions that are well known to be responsible for self-referential processing (e.g., self-representation, semantic and episodic autobiographical memory retrieval; Martinelli et al., 2013). Overall, the functional similarity of neuronal populations in the amygdala and the OFC and their strong reciprocal connectivity support the view that these two regions are pivotal for coding the state value of an event (Salzman et al., 2007; Morrison and Salzman, 2009; Salzman and Fusi, 2010).

Taken together, these findings show that the amygdala responds to stimuli whose relevance for the organism is contextually and cognitively modulated, regardless of their valence (positivity and negativity) and beyond their social dimension.

## **AN EVOLUTIONARY PSYCHOLOGICAL THEORY OF THE AMYGDALA**

As proposed by Brothers (1990), the amygdala, together with the OFC and the STG, is part of a network of neural regions that constitutes the"social brain."According to the social brain hypothesis (Dunbar, 2009), the size of the neocortex, which is mainly responsible for the expansion of the primate brain, has been found to be positively correlated with the increased complexity of social groups. Information-processing demands increase with the number of relationships as well as with the need to flexibly respond to the more complex scenarios of daily life. Within a large group, social interaction requires continuous on-line processing and monitoring of the dynamically and rapidly changing dispositions and intentions of conspecifics, as revealed by their bodily postures, facial expressions, or kinematics, and requires integration of this information with knowledge about their past actions, identity, and other social attributes.

Using a comparative method designed to detect coordinated evolution, Barton and Aggleton (2000) found that the amygdala and the neocortex volume correlated more strongly with each other, suggesting that these two distinct structures were conjointly tuned by natural selection to respond adaptively to particular lifestyles. Overall, the architecture of the prefrontal cortex is such that, on average, inputs from the amygdala attain approximately 90% of the prefrontal areas (Emery et al., 1997). Previous studies had already shown that between species amygdala volume was correlated with group size and the complexity of social networks. Cross-species comparative findings in nonhuman primates suggested that, when group size is taken as a proxy measure of social complexity, a significant positive correlation was found in 44 primate species between the relative amygdala volume (the ratio is estimated from total brain volume), and social group size, suggesting that this structure and in particular the *basolateral* nuclei, have evolved under evolutionary selectional pressure to increase the ability to manipulate information necessary to subserve sophisticated social relationships (Emery et al., 1997). More recently, Barger et al. (2007) also reported that larger amygdala, in particular the corticobasolateral complex, conjointly expanded with evolutionarily newer cortex under the pressure of the increased processing demands required by a complex social life. It has recently been shown that interindividual variability, both in humans (Bickart et al., 2010, 2012) and in primates (Sallet et al.,2011), is also linked with these parameters. Indeed, Bickart et al. (2010, 2012) showed that amygdala volume positively correlates with increasing network size and complexity (Bickart et al., 2010) and that stronger amygdala connectivity with other structures belonging to the social brain, such as the vMPFC, predicted group size and complexity. Importantly, this relationship was specific to the amygdala network and was not reported for other large scale functional networks, when controlling for age and correcting for multiple comparisons. Sallet et al. (2011) randomly assigned adult macaques to small or large social group housing conditions and found that several regions comprising the amygdala showed increased volume in the large social group. Taken together these findings suggest that interindividual differences in amygdala volume are strictly linked to social group size and complexity. Moreover, this variability seems sensible to environmental conditions and flexible to change even in adulthood. Even if no firm conclusion can be derived so far, the results of Sallet et al. (2011) suggest that reduced amygdala volume could be the consequence rather than the cause of individual social behavior. Although brain volume is an index of information-processing capacity, the fact that these two separate structures show closely correlated evolutionary changes in size reveals an increase in neural connectivity between them, in particular between the basolateral nuclei and the STG, the OFC and MPFC2.

Therefore, converging evidence suggests that the amygdala and the frontal cortex underwent expansion and evolved together by increasing neural connectivity. As we discussed above, from the perspective of the evolutionary psychology, the amygdala whose primary modular function was to rapidly and efficiently evaluate the environment for potentially threatening events (Amaral and Price, 1984; Öhman and Mineka, 2001; LeDoux, 2005) was constructed and adjusted in response to the statistical composite of situations encountered by our species in ancestral environments.

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 7 — #7

<sup>2</sup>Besides the well-recognized connections with medial and orbital regions of prefrontal cortex, the amygdala is also connected to the lateral prefrontal cortex, albeit to a lesser degree.

However, because of such increased connectivity in the frontolimbic neural circuit strongly characterizing the development of the human brain, the amygdala broadened its domain of specificity and enhanced the system's ability to regulate and generate more flexible and adaptive social behavior.

The same neural system is rarely capable of solving different adaptive problems fast and efficiently since different informationprocessing systems usually instantiate distinct procedures for their successful solution (Cosmides and Tooby, 1994). What we argue here is that the amygdaloid complex has preserved the primitive function of self-relevance detector by reshaping its internal modular structure, likely by weakening some of its modular properties (e.g., limited central accessibility and informational encapsulation3), to flexibly respond to a larger variety of self-relevant evolutionarily unprecedented circumstances.

## **THE RELEVANCE DETECTOR THEORY OF AUTISM**

In the following, we argue that the complex pattern of emotional and socio-behavioral impairments typically reported in individuals with ASDs reflects a disruption of the neural system devoted to the processing of self-relevant information, primarily relying on the functional and connectivity integrity of the fronto-amygdala circuit. Indeed, as we discussed above, although the amygdala can process relevant stimuli in a reflexive and unconscious manner (Vuilleumier et al., 2003; Juruena et al., 2010), it serves the function of bringing to consciousness awareness self-relevant information through the mechanism of emotional arousal (Vuilleumier and Schwartz, 2001). Thus, a disruption of the Relevance Detector System would lead to an impairment in the conscious appraisal of self-relevance emotions, which would compromise the ability to represent and communicate one's own internal states and feelings and lead to a reduce affective flexibility and emotional control (Cunningham et al., 2008).

A previous study on electrical stimulation suggested that the limbic system has a special role in bringing experience to a conscious level by associating affective and motivational significance with sensory information (Gloor et al., 1982). Neurobiological research has revealed that the neural substrates of self-awareness and subjective experience critically include the medial frontal cortex and the insula, both of which structures are functionally interconnected with the amygdala (Damasio, 1999; LeDoux, 2007). More recently, converging evidence from two studies (Kennedy and Courchesne, 2008; Lombardo et al., 2009) points to functional abnormalities in the vMPFC associated with self-related evaluative processing.

Research focused on emotional dysfunctions and theoretical accounts have emphasized the notion that the mechanisms mediating the self-regulation of behavior during social–emotional exchanges are severely impaired in ASDs (Yirmiya et al., 1992; Hobson, 1993). Pioneer studies reported that difficulties in children with ASDs might arise with both basic emotions (fear, disgust, anger) and social cognitive emotions (pride, embarrassment, shame) that are related to introspection and self-reflection (Capps et al., 1995; Loveland et al., 1997; Kasari et al., 2001; Heerey et al., 2003). It has been shown that children with autism have a less coherent representation of their own emotional experiences and failure to distinguish emotional experiences stems from a lack of reflective appraisal of those experiences (Harris et al., 1987). Despite preserved physiological responses and emotional empathy, they might often fail to show cognitive empathy (Rogers et al., 2007) or to generate and regulate emotionally laden situations introspectively (Rieffe et al., 2007). Recently, a consistent amount of evidence has pointed out that there is considerable overlap in the clinical presentation of persons with a diagnosis of ASD or of alexithymia (Hill et al., 2004; Hill and Berthoz, 2006), since both are characterized by disturbances in recognizing emotions and in the ability to use feelings to regulate interpersonal exchanges (Fitzgerald and Bellgrove, 2006). Remarkably, alexithymia can be regarded as a disrupted interaction between emotional arousal and the subjective experience of feelings (see Gaigg, 2012).

Different lines of behavioral research have reported disturbance in processing self-related information in individuals with ASDs, in terms of monitoring self-performed actions (Russell and Jarrold, 1999), or in correctly deciding whether an action had been produced by oneself or another agent (Russell and Jarrold, 1999). Millward et al. (2000) reported that children with autism have a specific difficulty with the recall of personally experienced events, as compared with memory for events experienced by a peer. Using a recognition test,Toichi et al. (2002)showed that a group of adults with HFA does not benefit from the self-reference effect since they are impaired in processing words in a self-related manner, in the absence of semantic and phonological impairments. More recently, Hare et al. (2007) found that adults with ASD demonstrate superiority for self-experienced events over events merely observed when the recall is cued whilst this superiority effect disappeared in free recall. On the same line, Zalla et al. (2010) have reported that adults with AS exhibited a reduced enactment effect for self-performed actions in free recall, as compared to a matched control group.

Although an abundant body of neuroimaging studies have related amygdala activation to the social dimension of stimuli (i.e., eye contact, gaze orientation, biological actions and intentions, trustful faces), the *Relevance Detection Theory of Autism* predicts that the amygdala specifically responds to self-relevant information. The literature is, however, extremely varied with respect to the physiological responses to socio-emotional events, associated with amygdala functionality in ASDs. In fact, ASD individuals are found to be either hyper- or hypo-aroused in response to simple sensory stimuli, or stimuli varying in emotional valence or social dimensions (Dalton et al., 2005; Schultz, 2005; Schoen et al., 2008; Anderson and Colombo, 2009; Ball et al., 2009). The *Relevance Detection Theory of Autism* posits that hyper-activation of the amygdala in response to potentially threatening or physically intense events in the environment is due to a disrupted interplay between a cognitive "top-down" attentional system and an automatic "bottom-up" attentional mechanism operating on raw sensory input. Since the amygdala

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 8 — #8

<sup>3</sup>According to Fodor (1983), a system is informationally encapsulated when the information is processed in a purely feedforward (bottom-up) manner: it is not affected by the person's expectations, beliefs, or desires and it is inaccessible to consciousness, and hence unavailable for explicit report. Limited central accessibility is a closely related property which involves restriction on the flow of information out of the system.

automatically and rapidly detects salient physical and biological features of potential importance by enhancing *bottom-up* attentional resources, reduced effective top-down control and attentional modulation exercised by the vMPFC on this structure would lead to the inability to form a "priority map of saliencies" that allows to regulate behavior and navigate the complex social world. Thus, in this view, reduced eye contact and social withdrawal are the result of adaptive avoidance responses to overcome excessive stimulation by a physically intense world or emotional hyperarousal and overresponsiveness to potentially aversive events (see Dalton et al., 2005; Kylliäinen and Hietanen, 2006; Kliemann et al., 2010).

Among the social stimuli, the eyes constitute an special source of relevant information. The ability to discriminate eye direction is thought to reflect an innate predisposition and a primitive function (Scaife, 1976). For many of species, direct gaze generally signals hostility and threat, and is associated with escape behavior (Emery et al., 1997, 2001). In monkeys, perceived eye gaze contact is associated with amygdala activation (Emery et al., 1997). In humans, eye gaze is a salient stimulus constituting an important source of information about other conspecifics (e.g., identity, age, gender, mental states, and internal emotional dispositions) but, depending on cultural and context-related factors, mutual eye contact and direct gaze may not necessarily be intrinsically threatening. Thus, because these signals can be ambiguous, their decoding may necessitate additional cognitive information and more conscious, evaluative processes (Engelmann and Pogosyan, 2013). The disrupted functionality of this integrative *Relevance Detection System* might lead to abnormal sustained activation of this subcortical route and to failure to detect meaningful aspects of the environment, in accordance of a "priority map" integrating intrinsic and extrinsic salience stimuli.

This explanation is in accordance with previous studies showing altered functional connectivity between vMPFC and amygdala, associated with diminished habituation of amygdala response to emotional faces (Swartz et al., 2013). Intriguingly, South et al. (2008) have shown that individuals with ASD exhibit a "threat advantage" effect (faster response time in detection of threatening stimuli as compared to neutral ones) and a typical anger superiority effect in visual search tasks employing face stimuli. In a recent neuroimaging study (Dalton et al., 2005), the amount of eye gaze fixation was strongly correlated with amygdala activation when viewing both emotional and neutral faces in participants with ASD, but not in control participants.

According to Liddell et al. (2005), an "innate alarm system," mediated by the primitive subcortical pathway, enables the organism to detect potentially threatening stimuli or unpredictable events in the physical environment, and thus promotes withdrawal and escape behaviors. In typically developed individuals, automatic fear-driven amygdalar responses are followed by activation in brain areas associated with controlled and reflective processes (Liddell et al., 2005). Indeed, amygdalar abnormalities typically associated with difficulties with fear extinction (Davis, 1992, 2000), also involve disturbances in social anxiety, hyperarousal, and sensory over-responsivity in ASDs (Amaral et al., 2008; Green and Ben-Sasson, 2010). Increased amygdala volume in children with ASDs was found to be positively correlated with anxiety and severity of social-communication deficits (Amaral et al., 2008) and higher scores for social anxiety show greater right amygdala response to negative emotional expressions in participants with ASDs (Kleinhans et al., 2010). While Mogg and Bradley (1999) regarded anxiety as preattentive bias toward threat, and argued that it results from an automatic encoding of threat without modulatory and elaborative processing, according to Davis and Whalen (2001), pathological anxiety may not be a disorder of fear, but a deficit in the ability to regulate vigilance and generalized hyperarousal in response to potential threat.

It is likely that reduced eye contact, perceived as potentially aversive stimuli, would preclude the development of perceptual expertise for faces, and hamper the ability to process different types of self-relevant social information acquired through faces, such as emotions, intentions, and trustworthiness (Begeer et al., 2008; Harms et al., 2010), and thereby trigger a cascade of deficits in this population in the domain of social interaction, such as initiated joint attention (Mundy and Newell, 2007), communication and attachment behavior (Hobson, 1993; Davies, 1994; Hobson and Lee, 1998).

Interestingly, the administration of oxytocin, a neuropeptide, which is known to be lower in individuals with autism (Modahl et al., 1998), enhances the salience and retention of social information in individuals with autism (Hollander et al., 2007; Andari et al., 2010) and decreases repetitive behaviors (Hollander et al., 2003). Recently, in a neuroimaging study, Domes et al. (2013) found that the oxytocin treatment promotes face processing and eye contact in individuals with ASDs and increases right amygdala activity. The medial nucleus of the amygdala, through the actions of the oxytocin, is a critical site for regulating approach and avoidance behaviors, for promoting social attachment (Ferguson et al., 2001), and for reducing anxiety (Bartz and Hollander, 2006).

Beyond the socio-emotional domains, our theory predicts that a disrupted functionality of this integrative *Relevance Detection System* might lead to the abnormal capture of attention by lowlevel, bottom-up visual properties of the stimuli (e.g., intensity, color, contrast, orientation), due to enhanced sensitivity of the physical attributes of the stimulus (Joseph et al., 2009). While this hyper-sensitivity is often associated with superior visual search abilities in ASDs, the enhancement of low-level visual processing and of physical salience of the events might lead to allocation of attention to irrelevant aspects of the visual environment (Joseph et al., 2009; Kaldy et al., 2011).

The hyper-sensitivity to the physical salience of external stimulation would lead to failure to shape a valid priority map of saliencies which could allow sensory stimuli to be integrated with current goals, personal needs, and contextual and prior knowledge.While bottom-up attention is driven by visually salient events in the environment (Itti and Koch, 2001), top-down attentional mechanisms implement longer-term cognitive strategies, biasing attention toward salient features as a function of the organism's internal needs and goals (Connor et al., 2004). This hypothesis is in accordance with recent findings showing that individuals with ASDs require a higher signal-to-noise ratio for the discrimination of visual or auditory presentations of fear *vs* disgust expressions (Charbonneau et al., 2013). Recently, Amso et al. (2013) found

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 9 — #9

that, relative to control children, children with ASDs rely more on bottom-up physical information for initial attention guidance, despite a similar orienting to faces in the two groups. Importantly, this finding suggests that reduced attention to faces and gaze in ASDs does not reflect disruption of an innate system devoted to the detection of eye contact, nor the lack of social motivation, but it would be the result of an unbalanced reliance on physical features of the environment. As also posited by the "Intense World Theory" (Markram et al., 2007), the hyper-emotionality, reflecting hyper-functionality of the limbic system, together with excessive responsiveness to environmental stimulation, result in perception of an aversive world, and social withdrawal in individuals with ASDs.

Taking into account the distinction we make between intrinsic and extrinsic context-dependent salience of the stimulus, a possible operationalization in an experimental setting would be to orthogonally manipulate these parameters to test which specific aspect of salience detection is impaired in people with ASDs and the corresponding response in the amygdala. Based on our current knowledge, we hypothesize that participants with ASDs would be more responsive to the bottom-up physically salient features associated with prolonged amygdalar activity, while diminished impact of the contextual contingency (extrinsic salience) may reflect reduced modulatory affect exercised by prefrontal regions on amygdala activity.

## **CONCLUSION**

In the present review, we have proposed that an early emerging neurological insult to the interconnected fronto-amygdala circuit disrupting the ability to flexibly and adaptively orient attention toward self-relevant stimuli might be a primary deficit of ASDs. Specifically, the amygdala is responsible, in concert with the vMPFC, of the formation of a *priority map* of selfrelevant events that might be accessible to and modulated by conscious evaluative processes. This priority map includes stimuli whose salience is determined by their intrinsic biological significance, the physical properties or the extrinsic contextual situation. In this view, physically intense stimulation and emotionally arousing events, associated with the amygdala hyperactivation, are actively avoided thus producing reduced attendance to meaningful aspects of the environment, including the social ones, and deficits in the self-regulation of behavior. At the neural level, our theory is in accordance with the frontoamygdala disconnectivity explanation and the hyper-active models which posit that the amygdala hyperactivation results from a defective top-down modulation by prefrontal areas involved in conscious evaluative processes. Moving away from the classical account of the amygdala as a *threat detector* or a socio-emotional processing submodule would favor the design of studies that might provide the opportunity to account for heterogeneities of cognitive phenotype and symptomatology across the autistic spectrum.

#### **ACKNOWLEDGMENTS**

The authors would like to thank the anonymous reviewers for their helpful and constructive comments that greatly contributed to improving the manuscript.

## **REFERENCES**


"fnhum-07-00894" — 2013/12/26 — 15:22 — page 10 — #10


"fnhum-07-00894" — 2013/12/26 — 15:22 — page 11 — #11


Hobson, R. P. (1993). *Autism and the Development of Mind*. Hove: Psychology Press.

Hobson, R. P., and Lee, A. (1998). Hello and goodbye: a study of social engagement in autism. *J. Autism Dev. Disord.* 28, 117–127. doi: 10.1023/A:10260885 31558


"fnhum-07-00894" — 2013/12/26 — 15:22 — page 12 — #12


"fnhum-07-00894" — 2013/12/26 — 15:22 — page 13 — #13


objects in monkeys with neonatal amygdala lesions. *Neuroscience* 106, 653–658. doi: 10.1016/S0306-4522(01)00445-6


"fnhum-07-00894" — 2013/12/26 — 15:22 — page 14 — #14


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 May 2013; accepted: 08 December 2013; published online: 30 December 2013.*

*Citation: Zalla T and Sperduti M (2013) The amygdala and the relevance detection theory of autism: an evolutionary perspective. Front. Hum. Neurosci. 7:894. doi: 10.3389/fnhum.2013.00894*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Zalla and Sperduti. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

"fnhum-07-00894" — 2013/12/26 — 15:22 — page 15 — #15

# On the representation and processing of social information in grounded cognitive systems: why terminology matters

# *Kendall J. Eskine\**

*Department of Psychological Sciences, Loyola University New Orleans, New Orleans, LA, USA \*Correspondence: kjeskine@loyno.edu*

*Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Tommaso Bruni, University of Milan, Italy Elisa Ciaramelli, Università di Bologna, Italy Hanah Chapman, The Ohio State University, USA*

The mechanisms underlying social behavior are indeed complex, yet researchers have made important contributions to our understanding of how people make judgments and behave across various social contexts. In particular, recent years has seen a proliferation of research spotlighting the guiding role of embodied and affective information in social processing. Grounded approaches to cognition offer an exciting opportunity for researchers throughout the cognitive sciences to work within a unified framework to shed light on traditionally nebulous and intractable psychological quagmires (e.g., symbol grounding).

In what follows I will describe how embodied and affective information influence some hallmark social processes (moral judgments and prosociality) and then clarify some misunderstandings about representation and processing in grounded cognitive systems. I will then argue that the term "directionality" in grounded accounts engenders misleading views about cognition and will conclude with recommendations that should improve our understanding of social behavior from the growing perspective of grounded cognition.

A growing body of literature indicates that embodied and affective states influence moral judgments and prosociality. In the domain of moral judgment, research has shown that inducing physical disgust (via visual, olfactory, and gustatory senses) can harshen moral judgments (Schnall et al., 2008b; Eskine et al., 2011). The conceptual overlap between physical and moral disgust has been further confirmed with physiological evidence (Calder et al., 2001; Moll et al., 2005; Borg et al., 2008; Chapman et al., 2009). In a similar vein, researchers have also demonstrated that feeling physically clean and pure can license people to judge others morally harsher than those feeling dirty (Zhong et al., 2010). However, physical cleansing can also attenuate people's own moral guilt. Zhong and Liljenquist (2006) found that people felt less guilty about their own transgressions (and were more likely to volunteer) after they had cleansed themselves with an antiseptic wipe, whereas those who did not receive a wipe showed increased volunteerism (but see also Fayard et al., 2009).

This line of research is often described as a *moral purity* metaphor, in which physical purity is metaphorically projected onto conceptual representations of morality. However, the direction of these effects travels both ways. A disgusting taste in the mouth can harshen moral judgments (Eskine et al., 2011), but thinking about moral transgressions, virtues, or control events can lead people to perceive a neutral tasting beverage as disgusting, delicious, or neutral-tasting, respectively (Eskine et al., 2012; see also Ritter and Preston, 2011). Similarly, cleanliness can attenuate harsh moral judgments (Schnall et al., 2008a), while committing moral transgressions can enhance the desirability of cleansing products (Lee and Schwarz, 2010). The implications of directionality for grounded theories will be discussed.

In the domain of prosociality, Schnall et al. (2010) explored whether emotional elevation affected volunteerism (Study 1) and helping behavior (Study 2). Overall, they found that those who experienced elevation, but not other positive emotions like happiness or amusement, were more likely to volunteer for an unpaid study and help experimenters with a boring task compared to those in control conditions. Similarly, Liljenquist et al. (2010) tested whether clean scents affect financial decisions in an economic trust game with the prediction that clean scents will prime purity and thus enhance altruism. Results confirmed that those in cleanscented rooms gave more money to an alleged team-mate compared with those in baseline rooms. They replicated these findings by showing that clean-scented rooms also encouraged volunteerism and monetary donations for helpful causes.

To determine whether there is any psychological truth in common taste metaphors like "she/he's a sweetie," Meier et al. (2012a) found that preferences for sweet foods significantly predicted prosocial behavior, and in another study they revealed that participants who consumed sweet foods (chocolate) were more likely to help another than those who consumed non-sweet foods (cracker) or nothing. These findings indicated that taste is an important embodied source domain that is projected onto abstract domains like prosociality, which bolsters a conceptual metaphor view that grounds abstract meaning in embodied source domains. But to what extent do views like these overlap in their theoretical assumptions with other views of grounded cognition (e.g., simulation theories)?

Broadly, principles of grounded cognition assert that sensorimotor and perceptual experiences are instrumental in the representation and processing of concepts. Simulation (Barsalou, 1999, 2008) and conceptual metaphor (Lakoff and Johnson, 1980, 1999) approaches represent the two dominant theories of conceptual grounding. Simulation models posit that conceptual processing recruits (roughly) the same perceptual states that were originally instantiated during one's initial embodied experiences, and metaphorical models contend that concrete, embodied experiences are projected onto abstract target domains. Although both views are "embodied," it remains unclear whether the origin and organization of conceptual knowledge ultimately reside in simulation-based models rooted in perceptual simulation or conceptual metaphors as explained by cognitive linguistics. I view this point as important but somewhat tangential with respect to the current debates on the structure of conceptual knowledge. Metaphorical theories are traditionally argued to have a unidirectional structure (concrete-to-abstract effects), whereas simulation theories imply bidirectionality (concrete-to-abstract and abstract-to-concrete effects). These views are compatible to the extent that they both ground meaning in embodied/affective states, yet the issue of directionality has caused concern among many researchers (Landau et al., 2010; IJzerman and Koole, 2011; Lee and Schwarz, 2012; Slepian et al., 2012).

Lee and Schwarz (2012) maintain that the manner in which concepts are generally represented (representational structure) does not necessitate how concepts will be processed in real-time cognition. They argue that conceptual representations can have unidirectional structure but can *still* reveal bidirectional effects when processed online. While their insights are accurate, this seems to be a point that should not require defending if one considers the interaction between conceptual representation and processing. Representational structure and online processing are intricately interwoven, so much so, that teasing the two apart, particularly in terms of causality (directionality), can seem at times like more of an exercise for the armchair than the laboratory. While classic findings from cognitive science have helped refine our understanding of representation and processing (e.g., the rejection of traditionally accepted semantic network models à la Collins and Quillian, 1969, and Collins and Loftus, 1975, in lieu of more complex connectionist models à la McClelland, 2000), this representation-processing distinction is a core component of Barsalou's (1999) perceptual symbol systems (PSS).

Sensorimotor activity that naturally accompanies various perceptual states becomes incorporated into the representational and processing structure of concrete (e.g., cats) and abstract (e.g., generosity) category domains. These embodied perceptual states are stored in memory and (partially) reactivated in bottom-up format during later conceptual processing. For example, many early experiences of interpersonal warmth naturally co-occur with physical warmth, such as cradling infants. Hence, perceptual experiences involved with physical and interpersonal warmth become part of the same representational and processing structure, which is one way to explain the nowpopular finding that experiencing physical warmth can promote interpersonal warmth toward a stranger (Williams and Bargh, 2008). Here, the representations themselves are the patterns of neural activity that span different regions of the brain, specifically perceptual and motor areas. While they may have some rough-andready structure, the task demands, social context, embodiment, top-down knowledge, affective states, etc. will re-construct representational-processing paths on a case-by-case basis. Thus, it should be no surprise that embodied states can affect abstract judgments and vice versa (implying bidirectional structure); they all participate in the same conceptual domain.

Therefore, while Lee and Schwarz (2012) are correct in arguing for bidirectionality in conceptual metaphors, a significant aspect of cognition is glossed over. I would assert that metaphorical knowledge is organized along lines of connectivity, not directionality. Both Lakoff and Johnson's (1999) and Lee and Schwarz's (2012) explanation of metaphor indicate that embodied and abstract domains are linked with each other and acquired through experiential coactivation, which is theoretically consistent with PSS. This truth in itself obviates the need for discussion of directionality because the manner in which these representations are activated (concrete-to-abstract or abstract-to-concrete) is simply a matter of the task-demands and context for a given embodied agent.

If this is indeed the case, then (1) why such emphasis on directionality and (2) what does this mean for social behavior? First, conceptual metaphor theory was born (in part) out of research in linguistics (Lakoff and Johnson, 1980), and directionality matters in linguistic metaphors. For example, calling a "butcher a surgeon" is very different than calling a "surgeon a butcher." The direction of the metaphor completely changes its meaning. Thus, directionality seems to be crucial to understanding/inferring meaning in linguistic metaphors, but it can be argued to be irrelevant to conceptual metaphors because brains do not process information in terms of rigid directionality; their processing is often determined by context-sensitive experiences that provide coactivations between various processing regions. Second, in addition to propagating misleading views about conceptual processing, another danger of overemphasizing directionality in grounded theories is that it can lead researchers down garden path research programs. Bidirectional and unidirectional effects can still be accommodated by simulation models, and both distract researchers from testing more specific models of groundedness. Third, though it is well-documented that metaphorical approaches can transfer embodied source domains to various dissimilar abstract/target domains (Landau et al., 2010), which implies directionality, context-sensitivity can still account for these differences, and it cannot be ruled out that more complex activation patterns in association areas of the brain undergird these effects. Therefore, since both PSS and conceptual metaphors appear to equally rely on such coactivations to create substrates for conceptual development, the use of the term "directionality" engenders misleading views about conceptual representation-processing and needlessly creates theoretical divisions among similar-minded researchers.

This clarification is particularly important for social behavior because conceptual knowledge has been demonstrated to be an important but *flexible* foundation that shapes people's social processing (see Lee and Schwarz, 2011, for evidence for context specificity). Therefore, I propose that we shift the language from "directionality" to "connectivity" to highlight the relative flexibility of conceptual systems, and how variability (culturally, contextually, and individually) helps determine how people think, judge, and act toward others. This view is also consistent with alternative metaphorical models that rely on "blends" to engender the kinds of temporary, embodied, contextually sensitive, and dynamical conditions that seems most consistent with what is currently known about cognitive processing (see Fauconnier and Turner, 1998, for a blending theory of metaphor that is more compatible with Lakoff and Johnson, 1999, than Lakoff and Johnson, 1980). Thus, while directionality is an appropriate tool for linguistic metaphors, it proves problematic for conceptual processing. Since these are separate domains, this proposed distinction will similarly be unable to accommodate linguistic metaphors because it is tooled for investigating conceptual processing.

In short, embodied/affective states not only help ground meaning and guide social processing but are also intricately linked to online processing and experience, upon which representationalprocessing states are founded. Along these lines, researchers have rightly argued that more attention should be given to cultural differences in metaphorical and embodied cognition (Meier et al., 2012b), which is better accommodated by "connective" rather than "directional" terminology, as the latter implies a certain amount of rigidity that fails to empirically occur in the brain or in social conceptual processing. In this way, there is considerable overlap in simulation and metaphorical views of grounded cognition.

Embodied effects that were once striking, intriguing, and perhaps confounding, are now commonplace, and it is indeed time to breathe life into these effects with systematic theory-building that is predicated on a deeper analysis of the mechanisms underlying grounded theories of cognition. By nature, embodied and affective information are flexible sources of information, which is evidenced by their context dependence (Lee and Schwarz, 2011), and researchers have just begun tapping into their malleable (and adaptive) properties. To better investigate the nuances of grounded theories of social behavior, I have proposed that we reconsider our terminology in simulationand metaphorical-based approaches. Rather than focusing on aspects of directionality that are couched in linguistic theories, highlighting the connective properties that develop through coactivation seems like a more promising path that better reflects how the brain actually processes information.

### **ACKNOWLEDGMENTS**

I sincerely thank Corrado Corradi-Dell'Acqua, the three reviewers, and Natalie Kacinik for their thoughtful feedback on an earlier draft of this manuscript.

#### **REFERENCES**


metaphor is specific to the motor modality involved in moral transgression. *Psychol. Sci.* 21, 1423–1425.


*Received: 24 February 2013; accepted: 25 March 2013; published online: 10 April 2013.*

*Citation: Eskine KJ (2013) On the representation and processing of social information in grounded cognitive systems: why terminology matters. Front. Psychol. 4:180. doi: 10.3389/fpsyg.2013.00180*

*This article was submitted to Frontiers in Emotion Science, a specialty of Frontiers in Psychology.*

*Copyright © 2013 Eskine. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Is it all about the self? The effect of self-control depletion on ultimatum game proposers

#### *Eliran Halali <sup>1</sup> \*, Yoella Bereby-Meyer <sup>1</sup> and Axel Ockenfels <sup>2</sup>*

*<sup>1</sup> Department of Psychology, Ben-Gurion University of the Negev, Beer-Sheva, Israel <sup>2</sup> Department of Economics, University of Cologne, Cologne, Germany*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Berna Güroglu, Leiden University, ˘ Netherlands Claudia Civai, University of Minnesota, USA Nicolas Silvestrini, University of Geneva, Switzerland*

#### *\*Correspondence:*

*Eliran Halali, Department of Psychology, Ben-Gurion University of the Negev, Ohayon Building 98, Beer-Sheva 84105, Israel e-mail: halali@bgu.ac.il*

In the ultimatum-game, as in many real-life social exchange situations, the selfish motive to maximize own gains conflicts with fairness preferences. In the present study we manipulated the availability of cognitive-control resources for ultimatum-game proposers to test whether preference for fairness is a deliberative cognitive-controlled act or an automatic act. In two experiments we found that a shortage of cognitive control (ego depletion) led proposers in the ultimatum game (UG) to propose significantly more equal split offers than non-depleted proposers. These results can be interpreted as resulting from an automatic concern for fairness, or from a greater fear of rejection, which would be in line with a purely self-interested response. To separate these competing explanations, in Experiment 2 we conducted a dictator-game in which the responder cannot reject the offer. In contrast to the increased fairness behavior demonstrated by depleted ultimatum-game proposers, we found that depleted dictator-game allocators chose the equal split significantly less often than non-depleted allocators. These results indicate that fairness preferences are automatically driven among UG proposers. The automatic fair behavior, however, at least partially reflects concern about self-interest gain. We discuss different explanations for these results.

**Keywords: social preferences, fairness, ultimatum game, dictator game, dual process, cognitive-control, selfcontrol, ego-depletion**

# **INTRODUCTION**

Behavioral decision-making research suggests that behavior is best understood as resulting from the operation of at least two underlying systems: the affective (system 1) and the deliberative (system 2). The affective system is generally described as fast, automatic, associative in nature, emotionally charged, and requires minimal cognitive resources. In contrast, the deliberative system is slow, deliberately controlled, analytical, affect free, and requires cognitive resources (e.g., Stanovich, 1999; Kahneman and Frederick, 2002, pp. 49–81; and for an overview: Evans, 2008). For individual decision-making tasks, such as inter-temporal choice, agreement exists among researchers about the behavior expected under the affective system, but for social decision-making the evidence is equivocal (e.g., Loewenstein et al., 2008). It is not clear whether economic self-interest or social preferences, such as fairness, are the primary motives (i.e., automatic) that need to be controlled by the deliberative system. In the current study, we contribute to this ongoing discussion by studying the role of cognitive-control in fairness behavior. Specifically, we examine whether fairness behavior is a deliberate act that requires self-control or whether it is evoked automatically. Answering this question is important because people often make social decisions under conditions of limited cognitive-control resources, such as exhaustion, sleep deprivation, cognitive load, and time pressure.

A well-known paradigm customarily used to study fairness perception and behavior is the Ultimatum Game (UG; Guth et al., 1982). In this game, two players are given an opportunity to split a sum of money. One player proposes how to split the sum, and another player responds. If the responder accepts the offer, the money is split as proposed. If the responder rejects the offer, neither player receives anything. The standard economic model dictates that the proposer should offer the smallest possible amount of money since the responder would accept any offer above zero. Contrary to this prediction, empirical results show that individuals consider fairness in their offers and choices. Proposers, on average, ask for less than 70% of the total sum, and responders usually reject unfair offers (for an overview: Camerer, 2003).

Models of social preferences address this fairness behavior. According to inequality aversion theories, people may not only care about their absolute outcome but also about their relative share (e.g., Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000). As a result, people may prefer to decrease the difference between their outcome and the outcome for others, even if this diminishes their absolute outcome. Alternatively, according to reciprocal fairness based theory, people care about the intention behind the offer and are willing to pay to punish (or reward) their opponents for their unfair (fair) offers (e.g., Rabin, 1993; Blount, 1995; Dufwenberg and Kirchsteiger, 2004; Bereby-Meyer and Niederle, 2005; Radke et al., 2012).

It has been suggested that fairness preferences result from deliberation processes (Moore and Loewenstein, 2004; see Knoch et al., 2006, for neurological support among UG responders). According to this view, egoism-based self-interest is the primary motive that needs to be constrained. In line with this suggestion, developmental studies have found that kindergarteners behave according to the standard economic model (e.g., Bereby-Meyer and Fiks, in press), while fairness preferences are most likely learned throughout life (Güroglu et al., 2009, 2011; Bereby-Meyer ˘ and Fiks, in press). However, the majority of neurological (e.g., Sanfey et al., 2003; Tabibnia et al., 2008) and behavioral (e.g., Cappelletti et al., 2011; Halali et al., in press) findings regarding UG responders suggest that by adulthood reciprocal fairness preferences become automatic relative to self-interest considerations. Thus, they are those that need to be controlled. Accordingly, Halali et al. (in press), found that a shortage of cognitive-control resources resulted in an increase in rejection rates of unfair offers in the UG, i.e., an increase in reciprocity behavior.

## **CURRENT RESEARCH**

In the current study, we examine the effect of cognitive control shortage on fairness behavior of UG proposers. By cognitive control (also termed "self-control" or "executive-control"; e.g., Schmeichel, 2007; Robinson et al., 2010) we mean the ability to "deliberately inhibit dominant, automatic, or prepotent responses," in order to maximize the long-term best interests of the individual (e.g., Mischel, 1996; pp. 197–218; Muraven and Baumeister, 2000). According to the deliberative approach to fairness preferences (e.g., Moore and Loewenstein, 2004) selfinterested behavior will increase under a shortage of cognitive control, i.e., an increased rate of unfair UG offers is expected. Contrary to that prediction, based on the automatic tendency of reciprocal fairness observed in the UG responders' behavior (Halali et al., in press), we expect an increase in fairness behavior under a shortage of cognitive control. Initial support for this hypothesis can be found in Rubinstein (2007) who found that equal split offers compared to non-equal offers are implemented faster, and by Cappelletti et al. (2011) who found that UG proposers offer more under time pressure.

To reveal the automatic tendency of UG proposers, in the current study, following Halali et al. (in press), we adopted the strength model suggested by Baumeister et al. (1998). According to this theory, self-control relies on a limited resource that gets depleted when one tries to inhibit competing behaviors, urges, or desires, just as a muscle tires after performing an effortful action. Consequently, an initial act of self-control impairs subsequent acts of self-control, even in unrelated tasks; this state is called *ego-depletion* (Baumeister et al., 1998; Muraven et al., 1998; Vohs and Heatherton, 2000; for a review, see Baumeister et al., 2007). The limited resource explanation has been disputed recently (e.g., Inzlicht and Schmeichel, 2012), however, there is agreement regarding the ego-depletion phenomenon, given the numerous experiments that support this finding (for a metaanalysis: Hagger et al., 2010). Thus, given that deliberate actions require cognitive-control resources, a state of depletion should increase automatic behavior (e.g., Vohs, 2006; Masicampo and Baumeister, 2008). In two experiments, we had our participants undergo an ego depletion manipulation and then examined the (un)fairness of their offers in the role of proposers in the UG. Given our assumption that fairness preferences are automatic, we expected an increased rate of fair offers by depleted participants compared to non-depleted participants.

# **EXPERIMENT 1**

# **METHOD**

## *Participants*

Twenty nine participants (14 Female and 15 Male) with no previous knowledge of the UG, participated in exchange to 20 New Israeli Shekels (NIS; approximately \$5). We informed participants ahead of time that we will randomly choose five participants and pay them according to their actual earnings in one random trial of the UG, which we actually did. We randomly assigned participants to one of two experimental conditions: depletion (*n* = 14; 7 Females, 7 Males), no-depletion (*n* = 15; 7 Females, 8 Males).

# **MATERIALS**

### *Depletion task*

We manipulated the cognitive-control resources depletion using Mead et al.'s (2009) procedure, which has been also used by Halali et al. (in press). Participants in the *depletion condition* completed 20 incongruent trials of the Stroop (1935) task. In each trial, participants had to name the color of the ink and suppress their automatic tendency to read the incongruent color word. In the *no-depletion condition*, the words matched the ink colors, making it unnecessary to ignore the words. Therefore, the incongruent condition required more cognitive-control resources than did the congruent condition.

## *UG task*

We randomly assigned participants to the role of proposers in a computerized version of a mini UG. We first thoroughly instructed participants about the nature of the rules of the UG. The task included 8 different independent trials with 8 different responders, who play the game in a different session of the same experiment. Other than that, we did not give the participants any other information regarding the responders. In each round, proposers had to make a one-time monetary offer of either a fair division, i.e., 50% of the stake for both players, or an unfair division, i.e., 80% of the stake to the proposer and 20% to the responder. Four different "Rejection-Outcomes" were associated with the different offers. As can be seen in **Figure 1**, these outcomes were: 0, 10, 20, or 30% of the stake to each player. For each Rejection-Outcome we implemented two different "Stake-Size": 100 NIS and 200 NIS (∼25 and \$50, respectively). We presented the 8 trials (4 Rejection Outcome ×2 Stake-Size) in a random order. We randomized the location of the equal split (50:50) on the screen (i.e., left/right) and its response-key within participants. To avoid outcome effects we did not give participants feedback about the responders' choices during the experiment. Note that the higher the Rejection-Outcome is, the lower are the consequences of a rejection for proposers' payoff. This tendency, however, is the same for the responders, which causes the risk of rejection to increase. Consequently, we do not expect the Rejection-Outcome to affect proposers' offers or to interact with the experimental condition. Thus, we were able to improve the statistical power of our test by presenting participants with several repetitions of the UG while minimizing the risk that participants will be bored.

#### *Mood and arousal*

Participants completed the Brief Mood Introspection Scale (BMIS; Mayer and Gaschke, 1988) that measures mood and arousal. The BMIS assesses participants' current mood based on their responses to 16 adjectives. In particular, participants rate how they feel in relation to each of the adjectives on a five-point Likert scale (1 = definitely do not feel, 5 = definitely do feel). The scale contains two subscales; mood valance and arousal.

#### **PROCEDURE**

We invited participants to different timeslots in groups of up to six participants each. With arrival participants were seated in a separate computer desks. All the assignments and questionnaires were computerized. After signing the consent form, and before starting the depletion task, we verbally informed participants that they would participate in a number of separate and independent experiments. Following the Depletion task, participants reported their mood and arousal levels on the BMIS, and following the UG task, they answered a questionnaire aimed at assessing suspicion regarding their partners in the UG. Participants had to indicate regarding the identity of the responders whether they are: "participants (a) from different sessions; (b) in the same lab with them; (c) different: \_\_\_\_\_\_."

#### **RESULTS AND DISCUSSION**

Two participants were excluded from the analysis since they indicated they did not believe they are playing a real game with real responders. The pattern of the following reported results was the same when these participants were included in the analysis.

We submitted the participants' offers (0: unequal, 1: equal split offers) to a three-way generalized probit estimation equations for binominal data with Condition (depletion, no-depletion) as a between-participant independent variable, Rejection-Outcome (0, 10, 20, 30%) and Stake-Size (100, 200 NIS) as withinparticipant independent variables, and the participants as a random factor.

We found a significant main effect for Condition: *Wald* <sup>χ</sup>2(1) <sup>=</sup> 4.12, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*042. As expected, on average, compared with the no-depletion condition (*M* = 52.7%), depletion state resulted in significantly higher rate of equal split offers (*M* = 68.3%). In addition, a two way interaction between Rejection-Outcome and Stake-Size was found, (*Wald* <sup>χ</sup>2(3) <sup>=</sup> 9.39, *<sup>p</sup>* <sup>=</sup> 0*.*025). Since we did not have a clear prediction regarding this interaction and it was not relevant to the significant main effect of Condition which is the focus of this experiment, we did not further analyze this interaction. No other effect or interactions were significant (all *p*s *>* 0.05).

The experimental condition did not affect Mood (nodepletion: *M* = 15.0, *SD* = 6*.*9, depletion: *M* = 12.1, *SD* = 6*.*9; *F <* 1*.*2, *n.s*.) or Arousal (no-depletion: *M* = 16.4, *SD* = 4*.*7, depletion: *M* = 17.4, *SD* = 6*.*0; *F <* 1, *n.s*.), thus, they are unlikely to account for the reported effect.

To summarize, the results of Experiment 1 showed that a shortage of cognitive-control resources led to an increase in the tendency to propose fair offers in the UG. This increase in fair offers proposals is in line with the tendency of UG proposers to propose more fair offers under time pressure (Cappelletti et al., 2011). In principle, the observed high proportion of fair offers may either be due to an automatic fear of rejection (selfish component) or to automatic fairness preferences. In Experiment 2, we tried to decide between these two motives.

#### **EXPERIMENT 2**

To further disentangle the two aforementioned explanations for the automatic tendency to propose fair offers in the UG, in Experiment 2 we use the dictator game (DG; Forsythe et al., 1994), a variant of the UG in which the responder cannot reject the offer. The advantage of the DG is that there is no fear of rejection in this game. At the same time, however, the DG removes the reciprocal relationship inherent in the UG—an observation that we get back to in the concluding section. If fairness preferences are automatic then we expect in the UG as well as in the DG to an increase in the rate of fair offers under a shortage in cognitive control. However, if selfish considerations associated with the perceived risk drive the automatic tendency to propose higher offers, then, we don't expect an increase in fair offers in the DG under a shortage in cognitive control. If something, we may even expect a decrease in the rate of fair offers in the DG. This prediction will be in line with the reduction in helping behavior observed among depleted participants (DeWall et al., 2008; Xu et al., 2012), and with a recent response-time study on DG allocators (Piovesan and Wengström, 2009) which found that self interested choices are made quicker than fair choices, both in a between and a within participants analysis.

Finally, one of the major dispositional factors related to decision-making in social situations is social value orientation (SVO; Van Lange et al., 1997). SVO are individual differences in how people evaluate outcomes for themselves and others (Messick and McClintock, 1968; Kuhlman and Marshello, 1975). Van Lange (1999) suggested that most people can be classified as being pro-socials, competitors, or individualists. Because individualists and competitors—both assign a higher weight to their own outcomes than to the outcomes of others they are usually taken together and defined as pro-self (e.g., Van Lange and Kuhlman, 1994). Regarding fairness preferences, Van Dijk et al. (2004) found that only pro-self participants are sensitive to the strategic aspect of the UG game. For example, in one of their experiments, when responders received incomplete information, pro-self participants took advantage of that and kept for themselves more money, while leading the responders to believe that they got a fair proposal. Yet, Van Dijk et al. (2004), have not examined the effect of SVO on automatic fairness behavior. In Experiment 2, therefore, we also assessed participants' SVO using the decomposed games measure suggested by Van Lange et al. (1997).

## **METHOD**

#### *Participants*

One hundred and seventeen undergraduate students (59 Female and 58 Male) with no previous knowledge on the UG or the DG, participated in exchange to extra course credit. We randomly assigned participants to one of four experimental conditions: UG depletion (*n* = 24; 11 Females, 13 Males), UG no-depletion (*n* = 25; 14 Females, 11 Males), DG depletion (*n* = 37; 18 Females, 19 Males), DG no-depletion (*n* = 31; 16 Females, 15 Males). We informed participants ahead of time that we will randomly choose 8 participants and pay them according to their actual earnings in one random trial of the UG/DG, which we actually did.

#### **MATERIALS**

#### *Depletion task*

We manipulated cognitive-control resources depletion using the Schmeichel's (2007) procedure. We instructed participants in the *no-depletion condition* to "Write a story about a recent trip you have taken. It may be a trip to a store, to some location in Israel, or to another country—wherever! Please keep writing until the computer program asks you to stop." For participants in the *depletion condition* we gave an additional instruction: "Very important! Please do not use the letters "*Aleph*" (equivalent to the English letter *a*) or "*Nun*" (equivalent to the English letter *n*) anywhere in your story." Hence, one group was required to regulate their writing by avoiding the use of two common letters, whereas the other group did not get any writing restrictions. The experimenter stopped all participants after 5 min of writing.

# *UG/DG task*

First we thoroughly instructed participants about the rules of the game they were assigned to. In the UG participants played in the role of proposers, offering one-time monetary offers to 4 different responders. Each offer involved different stake size: 100, 80, 50, and 20 NIS (∼25, 20, 12.5 and \$5, respectively), presented in a randomized order. We used a computerized version of the UG, in which the responders are participants from another academic institution who play the game in a different session of the same experiment. Other than that, we did not give the participants any other information regarding the responders. On each trial, participants first saw the stake amount for that trial, and then a response scale indicating the proportion of the stake size that they want to offer to their partner, from 0 to 50%, in increments of 10. In the DG, the task was the same as in the UG except for the fact that the responder has no decision to make.

## *Assessment of social value orientation*

As the last task, following an unrelated filler task, participants completed a nine-item Decomposed Games Measure (Van Lange et al., 1997). They chose among combinations of outcomes for themselves and for an anonymous other. These choices are made in a non-strategic setting (i.e., the outcomes depend only on what the participant chooses). Outcomes are represented by points, and participants are instructed to imagine that the points have value to themselves and to the other person. Each option represents a particular orientation. An example is the choice between alternative A: 500 points for self and 100 points for other, B: 500 points for self and 500 for other, and C: 550 points for self and 300 for other. Option A represents the competitive orientation because this distribution maximizes the difference between one's own outcomes and the other's outcomes (Choice A: 500–100 = 400, vs. B: 500–500 = 0, and C: 550–300 = 250). Option B represents the cooperative or pro-social orientation, because it provides an equal distribution of outcomes (i.e., 500 for self and other), and generates the highest number of collective outcomes (i.e., 1000). Finally, option C represents the individualistic option because one's own outcomes are maximized (550 vs. choice A: 500, and B: 500) irrespective of the other's outcomes. Participants are classified as pro-social, individualistic or competitive when at least six choices (out of nine) are consistent with one of the three orientations (e.g., Van Lange and Kuhlman, 1994). As in some prior research on SVO, we combined the individualists and competitors to form a group of pro-self individuals (e.g., Van Dijk et al., 2004).

#### *Mood and arousal*

We measured mood and arousal using the BMIS, in the same way as in Experiment 1.

#### **PROCEDURE**

We invited participants to different timeslots in groups of up to 6 participants each. Each participant sat in a separate computer desks. All the assignments and questionnaires were computerized. Following the UG/DG task, participants answered a questionnaire that assesses suspicion regarding their partners in the UG/DG. They indicated regarding the identity of the proposers whether they are: "participants (a) from other academic institutes; (b) from future sessions in the same institute; (c) in the same lab with them; (d) different: \_\_\_\_\_\_." Next, as a manipulation check, participants rated the difficulty of the writing task, on a scale from 1 (not at all difficult) to 7 (very difficult), and reported their mood and arousal levels on the BMIS.

#### **RESULTS AND DISCUSSION**

A total of 23 participants were excluded from all analyses. Fifteen participants were excluded because of their performance in the *depletion regulated-writing task*: 13 participants (5 in the UG, 8 in the DG) used the forbidden letters in over 10% of the words they wrote, and 2 participants (1 in the UG, 1 in the DG) did not write anything at all. Eight participants (3 in the UG and 5 in the DG) were excluded since they indicated they did not believe they were playing a real game with real responders. The pattern of the following reported results was the same when these participants were included in the analyses.

#### *Manipulation check*

Ratings of the difficulty of the initial writing task indicate that the instructions in the depletion condition indeed were more difficult to follow (*M* = 4.14, *SD* = 1.84) than the free writing instructions in the no-depletion condition (*M* = 2.42, *SD* = 1*.*49), *F(*1*,* <sup>92</sup>*)* = 25*.*04, *p <* 0*.*001, η<sup>2</sup> *<sup>p</sup>* = 0*.*21. This pattern is consistent with the assumption that the two writing instructions required different degrees of cognitive-control.

#### *Decision-making*

No main effects for Gender or for Stake-Size or any interactions with these factors were found in any of the analyses (all *p*s *>* 0.05) and thus these factors were not further analyzed.

We calculated for each participant the proportion of equal split offers (i.e., 50% of the stake). We submitted the proportion of equal split offers to a Two-Way ANOVA with Game (UG, DG) and Condition (depletion, no-depletion) as between-participant independent variables <sup>1</sup> . Consistent with previous results regarding the UG and the DG, we found a marginally significant main effect for the Game: *<sup>F</sup>(*1*,* <sup>90</sup>*)* <sup>=</sup> <sup>3</sup>*.*78, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*055, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*04, indicating that the proportion of equal split offers was higher for UG participants (*M* = 70.0%, *SD* = 40*.*9) compared to DG participants (*M* = 55.1%, *SD* = 45*.*9). While the main effect of Condition was not significant (*F <* 1, *n.s*.), the Game × Condition interaction was highly significant, *F(*1*,* <sup>90</sup>*)* = 12*.*25, *p <* 0*.*001, η<sup>2</sup> *<sup>p</sup>* = 0*.*12 (see **Figure 2**).

To probe the significant interaction, we conducted two simple contrast analyses, one for the UG and one for the DG. In each

contrast we compared the average proportions of equal split offers in the depletion and in the no-depletion groups. Consistent with Experiment 1's results, depleted UG participants proposed significantly higher rate of equal split offers (*M* = 85.3%, *SD* = 30*.*7) compared to the non-depleted participants (*M* = 58.7%, *SD* = <sup>44</sup>*.*3), *<sup>F</sup>(*1*,* <sup>90</sup>*)* <sup>=</sup> <sup>4</sup>*.*04, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*05, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*04. The pattern of results was reversed for the DG participants, i.e., depletion state resulted in significantly lower rate of equal split offers (*M* = 37.9%, *SD* = 45*.*1) compared to the no-depletion state (*M* = 72.2%, *SD* = <sup>40</sup>*.*6), *<sup>F</sup>(*1*,* <sup>90</sup>*)* <sup>=</sup> <sup>9</sup>*.*25, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*01, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*09.

The experimental condition, as in Experiment 1, did not affect Arousal as measured by the BMIS (no-depletion: *M* = 20.0, *SD* = 5*.*6, depletion: *M* = 19.4, *SD* = 5*.*7; *F <* 1, *n.s*.). Thus, Arousal is unlikely to account for the reported effects. However, for Mood valence (i.e., pleasant vs. unpleasant), we found a main effect of Condition, *<sup>F</sup>(*1*,*90*)* <sup>=</sup> <sup>4</sup>*.*75, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*05, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*05. Different than expected participants who performed the writing task in the depletion condition reported being in a more pleasant mood (*M* = 10.57, *SD* = 9*.*61) than participants who performed the free-writing task in the no-depletion condition (*M* = 6.24, *SD* = 9*.*44), with no main effect for Game nor an interaction between Game and Condition (both *F*s *<* 1*, n.s*.). To rule out the possibility that the mood accounts for the differences in the proposals we repeated the analysis on the proportion of equal split offers, while including as covariates Mood valence and Arousal. Neither Mood (*F <* 1, *n.s*.), nor Arousal (*F <* 1*.*25, *n.s*.), reliably predict the proportion of equal split offers, whereas the marginally significant main effect for Game [*F(*1*,* <sup>88</sup>*)* <sup>=</sup> <sup>3</sup>*.*47, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*066, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = 0*.*04], and the significant Game × Condition interaction [*F(*1*,* <sup>88</sup>*)* = 10*.*69, *p <* 0*.*002, η<sup>2</sup> *<sup>p</sup>* = 0*.*11] obtained in the original analysis were hardly affected. Hence, although there were unexpected differences in self-reported Mood valence between the two experimental conditions, these differences did not account for the pattern of results in the UG and the DG tasks.

#### *Social value orientation*

In the present experiment, out of 94 participants included in the previous analysis, nine participants (4 in the UG, 5 in the

<sup>1</sup>A Probit regression with subjects as a random variable, participants' offers as the dependent variable and Condition, Game and the interaction between them as independent variables, revealed the same results.

DG) made fewer than six consistent choices according to one of the three orientations (e.g., Van Lange and Kuhlman, 1994) in the nine-item Decomposed Games Measure. Hence, they could not be classified and were therefore excluded from further analyses. Of the 85 remaining participants, 51 (60.0%) were classified as pro-social and 34 (40.0%) as pro-self. The distribution of pro-socials and pro-selfs in each experimental condition was as follows: UG depletion (*n* = 15: 10 pro-socials, 5 pro-selfs), UG no-depletion (*n* = 21: 11 pro-socials, 10 pro-selfs), DG depletion (*n* = 27: 16 pro-socials, 11 pro-selfs), DG no-depletion (*n* = 22: 14 pro-socials, 8 pro-selfs). There were no significant differences in the proportion of pro-socials in each of the four experimental conditions (all χ<sup>2</sup> *<* 1, *n.s.*).

We repeated the analysis for the proportion of equal split offers with participants' SVO (pro-self, pro-social) as an additional between-participant independent variable. The Game × Condition interaction remained significant, *F(*1*,* <sup>77</sup>*)* = 10*.*07, *p <* 0*.*003, η<sup>2</sup> *<sup>p</sup>* = 0*.*12, with the same pattern as previously reported. We also obtained a significant main effect for SVO, *F(*1*,* <sup>77</sup>*)* = 6*.*21, *p <* 0*.*05, η<sup>2</sup> *<sup>p</sup>* = 0*.*07, indicating that, across Game and Condition, pro-social participants had a higher rate of equal split offers (*M* = 70.0%, *SD* = 40*.*3) compared to pro-self participants (*M* = 43.4%, *SD* = 47*.*8). No other effect was significant (all *p*s *>* 0.10). Therefore, SVO did not moderate the depletion effect for UG or DG proposers.

To summarize, in Experiment 2 we replicated the results of Experiment 1 for UG proposers, using a different manipulation for ego-depletion and a different structure of the game. Specifically, a shortage of cognitive-control resources resulted in an increase of fair behavior. For depleted DG allocators however, we found the reversed pattern, i.e., they demonstrated a decrease of fair behavior compared to non-depleted allocators. Further, in line with previous findings (Van Dijk et al., 2004), pro-social participants tended overall to care more for fairness than pro-self participants. Yet, participants' SVO did not moderate the effect of ego depletion in the UG or in the DG. It is worth noticing, however that the number of participants within each orientation (i.e., pro-self/pro-social) in each condition is relatively small.

#### **GENERAL DISCUSSION**

Is the automatic fairness tendency of UG proposers due to automatically elicited fairness preferences, or is it due to an increased fear of rejection, i.e., automatic strategic selfish preferences associated with risk perceptions?

The observation in Experiment 2 that depleted DG allocators became more selfish compared to the non-depleted allocators indicates that participants were less concerned with fairness when the fear of rejection was absent. The automatic selfish behavior demonstrated by depleted DG allocators is consistent with the findings of a recent study which demonstrated that ego-depletion reduced the willingness to help others (DeWall et al., 2008). This effect was mediated by decreases in guilt feelings (Xu et al., 2012). Notably, if fairness preferences drive the high proposed offers of depleted UG proposers, we should have observed higher offers from depleted, compared to non-depleted DG allocators as well. Given the reversed observed pattern for depleted DG allocators, the increase in fair behavior of the depleted UG proposers probably reflects an automatic selfish fear of rejection.

Interestingly, the increase in fairness behavior among UG proposers matches the automatic behavior of UG responders documented in most studies on that matter (e.g., Cappelletti et al., 2011; Halali et al., in press). Specifically, it corresponds with the increase in negative reciprocity of UG responders following a shortage of cognitive control resources (Halali et al., in press). The results of the current study, however, suggest that this match in behavior is probably driven by different motivations, namely, depleted UG proposers are motivated by automatic selfish preferences rather than automatic fairness preferences, that probably motivate depleted responders in the UG.

At first glance, given the reasoning aspect assumed to be involved in strategic thinking it sounds contradictory that strategic considerations are revealed under a shortage of cognitive control. We suggest that different types of emotions, which are affected differently by a shortage in cognitive control, may explain this counter intuitive result (Halali et al., in press). Specifically, strategic considerations of UG proposers are driven by fear that their offer will be rejected (e.g., Nelissen et al., 2011). In contrast, fair behavior of DG allocators is suggested to be driven by guilt (e.g., Ellingsen et al., 2010). While fear is an immediate experienced emotion that is viscerally driven (Loewenstein, 1996, 2000) and therefore is dominant under a shortage of cognitive control resources (Wagner and Heatherton, 2013; Vohs et al., submitted), guilt is an anticipated emotion that is likely to be reduced under ego depletion (e.g., Xu et al., 2012). While Nelissen et al. (2011) suggested that guilt is an additional motivation for fair behavior of UG proposers, given the finding that UG offers are in most cases higher than DG offers, the strategic component in the UG (i.e., fear of rejection) is probably more dominant in this game. The current results suggest that this fear of rejection in the UG is even more pronounced under ego depletion.

Another possible explanation for the seemly mixed results for UG proposers and DG allocators may be the characteristics of the games, which may trigger different motives for being fair. Much research has concluded that UG behavior is mainly driven by reciprocity (e.g., Rabin, 1993; Blount, 1995; Dufwenberg and Kirchsteiger, 2004; Bereby-Meyer and Niederle, 2005; Radke et al., 2012)—a motive that cannot explain DG behavior, because the receiver in the DG cannot reciprocate (reward or punish) the allocator's offer. This seems to suggest an alternative hypothesis: in games of reciprocity the automatic response is to behave in a reciprocally fair way, while in games without reciprocal interaction, selfishness is the automatic response. This hypothesis is consistent with automatic fairness on both the proposer's and responder's side in the UG, as well as with automatic selfishness in the DG. Further research is needed to distinguish between these possible explanations.

# **ACKNOWLEDGMENTS**

This research was supported by the Israel Science Foundation grant number 11/1337 to Yoella Bereby-Meyer. Eliran Halali was partly supported by the ISEF foundation. Axel Ockenfels gratefully acknowledges support from the German Science Foundation through the Leibniz-program and through the research unit "Design & Behavior."

# **REFERENCES**


*Econ.* 114, 817–868. doi: 10.1162/ 003355399556151


*Psychol.* 32, 922–931. doi: 10.1037/ 0022-3514.32.5.922


Fear and guilt in proposers: using emotions to explain offers in ultimatum bargaining. *Eur. J. Soc. Psychol.* 41, 78–85. doi: 10.1002/ ejsp.735


*Soc. Psychol.* 40, 697–707. doi: 10.1016/j.jesp.2004.03.002


orientations and impressions of partner's honesty and intelligence: a test of the might versus morality effect. *J. Pers. Soc. Psychol. J. Pers. Soc. Psychol.* 67, 126–141. doi: 10.1037/0022-3514. 67.1.126


depletion increases emotional reactivity in the amygdala. *Soc. Cogn. Affect. Neurosci.* 8, 410–417. doi: 10.1093/scan/ nss082

Xu, H., Bègue, L., and Bushman, B. J. (2012). Too fatigued to care: ego depletion, guilt, and prosocial behavior. *J. Exp. Soc. Psychol.* 48, 1183–1186. doi: 10.1016/j.jesp. 2012.03.007

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 28 February 2013; accepted: 16 May 2013; published online: 13 June 2013.*

*Citation: Halali E, Bereby-Meyer Y and Ockenfels A (2013) Is it all about the self? The effect of self-control depletion on ultimatum game proposers. Front. Hum. Neurosci. 7:240. doi: 10.3389/ fnhum.2013.00240*

*Copyright © 2013 Halali, Bereby-Meyer and Ockenfels. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# The effect of partner-directed emotion in social exchange decision-making

#### *Iveta Eimontaite1 \*, Antoinette Nicolle1, Igor Schindler <sup>1</sup> and Vinod Goel 1,2\**

*<sup>1</sup> Department of Psychology, University of Hull, Hull, UK*

*<sup>2</sup> Department of Psychology, York University, Toronto, ON, Canada*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Vera Shuman, University of Lausanne, Switzerland Sergio Agnoli, University of Bologna, Italy*

#### *\*Correspondence:*

*Iveta Eimontaite, Department of Psychology, University of Hull, Cottingham Road, Hull, HU6 7RX, UK e-mail: i.eimontaite@2011.hull.ac.uk; Vinod Goel, Department of Psychology, York University, 4700 Keele St., Toronto, ON M3J 1P3, Canada e-mail: vgoel@yorku.ca*

Despite the prevalence of studies examining economic decision-making as a purely rational phenomenon, common sense suggests that emotions affect our decision-making particularly in a social context. To explore the influence of emotions on economic decision-making, we manipulated opponent-directed emotions prior to engaging participants in two social exchange decision-making games (the Trust Game and the Prisoner's Dilemma). Participants played both games with three different (fictional) partners and their tendency to defect was measured. Prior to playing each game, participants exchanged handwritten "essays" with their partners, and subsequently exchanged evaluations of each essay. The essays and evaluations, read by the participant, were designed to induce either anger, sympathy, or a neutral emotional response toward the confederate with whom they would then play the social exchange games. Galvanic skin conductance level (SCL) showed enhanced physiological arousal during anger induction compared to both the neutral and sympathy conditions. In both social exchange games, participants were most likely to defect against their partner after anger induction and least likely to defect after sympathy induction, with the neutral condition eliciting intermediate defection rates. This pattern was found to be strongest in participants exhibiting low cognitive control (as measured by a Go/no-Go task). The findings indicate that emotions felt toward another individual alter how one chooses to interact with them, and that this influence depends both on the specific emotion induced and the cognitive control of the individual.

**Keywords: anger, sympathy, decision-making, social exchange, prisoner's dilemma, trust game, cognitive control, emotion**

# **INTRODUCTION**

Economic theory commonly follows a normative approach to understanding human decision-making. That is, humans are assumed to be rational beings, motivated purely by the goal of maximizing gains and minimizing losses (Camerer, 1997). Recently, however, economists have taken a more descriptive approach, incorporating psychological findings of the way humans actually behave into their models. Since humans must commonly make decisions within a social context, it is important to explore the factors that influence our social decision-making. It has been found that social interactions are driven not only by logic (Camerer, 1997; Burks et al., 2003; DeSteno et al., 2010) but also by factors such as descriptive framing (Camerer, 1997; De Martino et al., 2006), fairness equilibrium (Camerer, 1997), consideration of the beliefs and desires of other players (Dubey et al., 1987; Mellers et al., 2010), perceived trustworthiness (Cox, 2004; King-Casas et al., 2005; Charness et al., 2011), and other aspects of the players' perceived character (De Dreu and McCusker, 1997). Moreover, social decision-making is influenced by our emotions (Frank, 1988; Elster, 1999). While normative economic theories rely on the view that humans are purely rational agents computing the best possible outcome, descriptive economic theories are beginning to incorporate emotion into their models (Frank, 1988; Loewenstein, 2000; Lerner et al., 2004; Andrade and Ariely, 2009).

The role of emotion in social decision-making can be explored using social exchange games. The Prisoner's Dilemma and the Trust Game are two games which are commonly used to measure decision-making in which the outcome depends on the interaction between two players. In the Prisoner's Dilemma [developed by Flood and Dresher in 1950 (Kuhn, 2009)], each of two players simultaneously choose to cooperate or to defect against the other player. If one player cooperates and one defects, then the defector wins money while the cooperator loses money. If both players choose to defect, then both will lose money, but the amount lost is less than if one is the sole defector. The pay-off matrix is such that the "rational" choice (in a one-shot game) is to defect; however in a repeated game a better outcome is received by both players when they both cooperate. In the Trust Game (Berg et al., 1995), the two players make their decisions sequentially. The first participant must choose either to cooperate and share an amount of money with the other player (in this case the amount of money the other player receives is multiplied by a certain coefficient), or to defect and keep everything for themselves. If they choose to cooperate, the other player can then either reciprocate by returning half of the received money or they can keep everything for themselves. The fact that the Prisoner's Dilemma involves simultaneous interaction, while the Trust game involves sequential choices, may result in the two games loading differently on the decision-makers' cognitive resources. Specifically, players of the Prisoners Dilemma must keep in mind four possible outcomes of the interaction and to anticipate what their opponent might chose, while players of the Trust Game must consider only two possible options and have greater influence on the end result of interaction. Cognitive load is known to influence the level of cooperation in such socially-interactive decision games. For example, when participants must memorize 7 digits (highcognitive load) instead of 2 digits (low load), they are found to cooperate more in the Prisoner's Dilemma, particularly as the end of the game approaches (Duffy and Smith, 2012).

From studies investigating decision-making in the Prisoner's Dilemma and the Trust Game it is evident that people do not always make the "rational" choice (Dawes and Thaler, 1988). One possible factor explaining these deviations from rationality is that emotions influence our decisions in these games (Frank, 1988; Elster, 1999). As such, participants may be seen to make decisions more so with an aim of regulating their emotional responses than to maximize monetary reward. Moreover, emotions can also aid decision-making by providing information relevant for choice valuation, motivating those behaviors which are most in line with personal values as well as moral and social norms (Peters et al., 2006a; Pfister and Böhm, 2008). Emotions can also focus the decision-maker's attention onto the most salient (or personally relevant) aspects of the decision scenario, thus adjusting which information will be used most for the decision (Pfister and Böhm, 2008).

Another factor influencing what choices individuals make is cognitive control capacity. In a study by De Neys et al. (2011) performance on the Go/no-Go task was compared between individuals who rejected a high number of unfair offers in the Ultimatum Game with those who rejected a low number of unfair offers. The results showed that those who rejected a low number of unfair offers had higher cognitive control than those who rejected a high number, suggesting that judgments of fairness have a greater effect on choice behavior when cognitive control is low. Cognitive control also has influence in logical reasoning, where individuals with higher cognitive control are found to reason in line with logic while low cognitive control participants make their choices more intuitively (Stanovich and West, 2000). In addition to this, an imaging study with the Ultimatum Game showed that recipients of unfair offers had a higher activation in brain areas related to cognitive control (dorsolateral prefrontal cortex) and emotional processing (anterior insula) (Sanfey et al., 2003) suggesting that both cognitive control and emotion processing are involved in making decisions in economic games.

Here we explore the effects of two partner-directed emotions predicted to influence social exchange decision-making sympathy and anger. Sympathy is defined as an emotional response that results from awareness of another person's undesirable experiences. Its subjective experience consists of feelings of sorrow and concern for the other, and is also associated with heightened awareness of the plights of others, and a desire to help (Eisenberg and Strayer, 1987; Eisenberg, 1991). Many researchers have considered sympathy and empathy as synonyms (Rosenberg and Towers, 1986; Eisenberg and Miller, 1987; Eisenberg and Fabes, 1990; Decety and Chaminade, 2003) and here we also do not distinguish them. On the other hand, anger is related to hostility and aggression, and varies in intensity from mild irritation to fury or rage (Spielberger et al., 1983). For sympathy to be induced, past studies have shown that the subject must adopt the other's perspective or to place at least a moderate value on the welfare of the other (Smith, 1992; Lishner et al., 2011). For anger to be induced unexpected and apparently real frustrating events, with negative impacts on wellbeing, are required (Stemmler, 1997; Clore and Centerbar, 2004; Lobbestael et al., 2008; Winterich et al., 2010; Deffenbacher, 2011).

With their differential antecedents, it is unsurprising that sympathy and anger promote differential behavioral tendencies. Sympathy has been known to induce helping behavior in students sharing their lecture notes with another student for whom illness has prevented them from taking their own notes (Reisenzein, 1986). It also promotes willingness to help a family whose son has cancer (Harmon-Jones et al., 2003) and to help a multiple sclerosis patient even after receiving an insulting comment from him (Harmon-Jones et al., 2004). Sympathetic concern also encourages higher donations when a victim (a starving child in Africa) is identifiable (where participants receive a photo and description of the child), than when the victim is presented as a non-identifiable single victim or merely as a statistic (Small et al., 2007). It is also found to encourage more generous decisions toward the other person in economic decision-making games when the outcome of interaction depends on two individuals, such as in the Prisoner's Dilemma (Batson and Moran, 1999; Batson and Ahmad, 2001; Duersch and Servatka, 2007) and "Ring Measure of Social Values" (Van Lange, 2008). In these two games, higher cooperation rates are promoted when participants perceive their opponent to be in need and when they adopt their opponent's feelings (Batson and Moran, 1999; Van Lange, 2008). In a study by Batson and Moran (1999), relating to and being aware of a partner's current difficulties, results in higher cooperation in the Prisoner's Dilemma, compared to a control condition. In a follow-up study by Batson and Ahmad (2001), this increased cooperation was apparent even when the opponent had made previous decisions in the game that were against the interests of the participant.

In contrast, anger is found to encourage higher defection rates in social-exchange games, including the Power-to-Take Game (Bosman and van Winden, 2002; Ben-Shakhar et al., 2004), the Prisoner's Dilemma (Duersch and Servatka, 2007) or the Ultimatum Game (Sanfey et al., 2003). Using the Power-to-Take Game, Bosman and van Winden (2002) found that the more anger participants felt about their opponent's decision, the more often they destroyed income even if that was costly to the participant themselves. Moreover, the intensity of felt anger is found to be positively related to the defection rate in an economic game with punishment (De Quervain et al., 2004). In addition, it has been found that anger, induced through perceptions of character, elicits violent behavior toward the anger-inducing individual (Harmon-Jones and Sigelman, 2001). Similar results emerge from studies investigating the effect of emotion on negotiation decisions. In a study by Van Kleef et al. (2004), participants acted as a phone seller and were asked to negotiate with a potential buyer about price, warranty, duration of the service contract etc., Van Kleef et al. found that, when facing angry buyers, participants made lower demands (offered lower price, longer warranty, etc.) and more often accepted bigger concessions requested by the buyer (Van Kleef et al., 2004). On the other hand, when individuals received information about the buyers' own emotional responses to either the offers or to the individuals themselves (e.g., "this [offer/person] makes me really angry"), anger directed toward their behavior was found to have different effects compared to emotions directed toward the person. Specifically, anger induced by the individual's previous offers resulted in larger concessions and lower demands compared to behavior-oriented happiness. Conversely, buyers who felt person-directed anger (i.e., buyers who said "this person makes me really angry") encouraged individuals to make lower concessions and higher demands in the negotiation process compared to person-directed happiness (Steinel et al., 2008). In a study by Kopelman et al. (2006), participants made higher demands (when playing the role of seller), while interacting with buyers displaying negative emotions, offering higher phone price, shorter warranty period, etc., and were less likely to sign a deal compared to positive and neutral emotions (Kopelman et al., 2006). These studies not only show that anger results in reduced cooperation with others compared to other emotions (neutral and happy), but also indicate that person-directed emotions and behavior-directed emotions can have different effects on social behavior.

The current study explored how the emotions of sympathy and anger affect decision-making in the Prisoner's Dilemma and the Trust Game in a within-subject design. We hypothesized that sympathy and anger would have different effects on social decision-making, such that sympathy would reduce defection rates and anger would increase defection rates, compared to neutral emotion. Given the possibility that the two games load differently onto cognitive resources, we also explored how individual differences in cognitive control moderate emotional influences on decision-making in the Prisoner's Dilemma and the Trust Game. We expected participants with lower cognitive control to have different defection rates than those with higher cognitive control.

To test the efficacy of our emotional manipulations we used galvanic skin conductance measures and subjective reports. Skin conductance is commonly used as an indication of physiological and psychological arousal, by observing electrical conductivity responses in the skin. In accordance with past literature, we expected to find higher skin conductance levels (SCLs) to be associated with anger and sympathy emotion-induction conditions compared to neutral (Rustichini, 1966; Ben-Shakhar et al., 2004; Hein et al., 2011). We also collected self-report data and used a cluster analysis to examine the subjective experience associated with each emotion induction condition.

## **METHODS PARTICIPANTS**

Thirty-eight participants took part in the study. All participants had normal or corrected-to-normal vision and were not undergoing any psychopharmacological treatment (one participant was removed after self-declaring that they had an anxiety disorder). Another eight participants were removed after declaring that they were aware of the deception, leaving 29 participants for the final analysis (14 females) (mean age = 23 years, *SD* = 4*.*4). The study was approved by the Department of Psychology ethics committee, University of Hull, and was carried out in accordance with the ethical guidelines published by the British Psychological Society, the American Psychological Association and the Declaration of Helsinki.

## **PROCEDURE**

Participants were asked to come to the experiment with an essay they had written about something that was important to them. They also believed that three "other participants" had done the same and would be participating in the experiment at the same time, though the participant never met these other individuals and, indeed, they were not real. Participants were told that, for reasons of anonymity, all participants would complete the experiment in separate rooms. During the experiment participants would read the other participants' essays and would evaluate them one by one (while they believed their own essay was also being evaluated by each other participant).

Participants always began the experiment by completing the Go/no-Go task, to measure their cognitive control. Following this, they were presented with their first opponent's essay to read and evaluate. Once this essay was evaluated, participants played two distractor games while the experimenter left the room (the participant believed to collect the opponent's evaluation). The Wason Card Selection task (Wason, 1968) and the THOG task (Wason and Brooks, 1979) were used as distractors in order to make the aims of the study less obvious to participants. Performance in these tasks was not analysed further. The participant then received his opponent's evaluation of his own essay, and then immediately played three rounds of the Prisoner's Dilemma and three of the Trust Game with this same opponent. This was followed by new versions of each distractor task.

This procedure of essay reading/evaluation, distractor tasks, receipt of one's own evaluation and social-exchange game playing was then repeated for the remaining two emotion conditions (i.e., with the remaining two "other participants"). The order of emotion conditions (sympathy, anger, and neutral) and the order of the social decision-making tasks were counterbalanced between subjects to avoid order effects (**Figure 1A**). At the end of the experiment, participants completed the emotion questionnaire (see below). Finally, the experimenter asked questions to determine whether the participant suspected deceit or the aim of the experiment. While deception/harm to the participant was transitory, full debriefing, and contact details for a university counselor were given to participants at the end of the experiment.

## **STIMULI**

#### *Emotional manipulation*

The emotion manipulation was achieved by presenting participants with pre-constructed essays, which they believed were written by their partner participants, and with subsequent evaluations of the participant's own essays, which they believed were also written by their partners. The evaluation

forms consisted of ratings of the essays on six 9-point bipolar scales (unintelligent–intelligent; thought provoking–boring; friendly–unfriendly; illogical–logical; respectable–unrespectable; irrational–rational), along with a space for free comments.

Together, the essays and evaluations were designed to induce sympathy toward one of the partner participants, anger toward another and neutral emotion toward the third. The emotion in the sympathy condition was induced through the essay, and in the anger condition was due to the negative evaluation of the participant's own essay. The sympathy-inducing essay was modified from Harmon-Jones et al.'s (2003) and concerned a young person coping with cancer. The essay was re-written according to UK education and healthcare standards. After reading this essay, the participant received a neutral evaluation of their own essay, consisting of neutral ratings (between 4 and 7 on the evaluation scales) and a hand-written positive comment "I can understand why a person would think like this." In the anger-inducing condition the participant read a poorly written essay (grammatical mistakes, badly structured arguments) and subsequently received a negative evaluation of their own essay (Harmon-Jones and Sigelman, 2001). The anger-inducing evaluation consisted of ratings that were weighted toward negative words (e.g., illogical or unacceptable). An insulting comment was also hand-written underneath the evaluation ("This is the stupidest thing I have ever read"). In the neutral condition they received an emotionally neutral essay, written in an unemotional and grammatically correct way, followed by a neutral evaluation of their own essay (consisting of neutral evaluations between 4 and 6, and no additional hand-written comments).

The three essays/evaluations were written in clearly differentiable handwriting, and were piloted before the study to check that they triggered the targeted emotion (15 participants were monitored with galvanic skin conductance measurement and later reported what emotions the essays triggered). Galvanic skin conductance serves as an objective measure of emotional arousal, since participants cannot exert top-down control on their skin conductivity responses (Ben-Shakhar et al., 2004; Lin et al., 2005). However, we realize that a drawback of such measures is that they do not allow us to address the subjective content and direction of the emotional experience which is why we also included a selfreport emotion questionnaire which participants completed at the end of the experiment.

*Self-report emotion questionnaire.* Here participants were presented with a list of 36 emotion words and, for each word, indicated which (if any) "other participant" they had felt it toward. The questionnaire was analysed with a hierarchical and k means cluster analysis.

*Galvanic skin conductance.* Galvanic skin conductance was continuously recorded through the experiment using a second computer, connected to a Biopack MP100A digital skin conductance amplifier with a constant voltage of 0.5 V. Electrodes were placed on the non-dominant hand and attached to the medial phalanx surfaces of the middle and index finger. An electrodermal gel (GEL101) was used as an electrolyte for conductance.

Galvanic SCL was calculated individually for each emotioninduction condition. The skin conductance measurements were analysed from the time when participants received the essays and evaluations (with baselines collected at rest periods before each of these critical time windows). That is for the sympathy condition, SCL was analysed while participants read the sympathetic essay and for the anger condition while reading the negative evaluation on the participant's own essay. For the neutral condition, galvanic skin conductance was averaged from reading the neutral essay and receiving the neutral evaluation on the participant's own essay. The mean SCLs were computed for each condition, using Acknowledge 3.9.1 for Windows.

#### *Decision-making tasks*

The following tasks were completed by participants separately for each emotion condition (with three repetitions of each task per fictional partner). The tasks were presented on a computer, using Cogent 2000v1.32 (www*.*vislab*.*ucl*.*ac*.*uk) through Matlab (version R2011.a). Participants were guided through the rules of these games, and the experimenter asked questions to make sure that the participants understood the game. To reduce participant's expectations and any reputation effects in the games, participants were told that they may or may not play some games more than once.

*Prisoner's Dilemma.* The task was developed by Flood and Dresher in 1950 (Kuhn, 2009). Participants are asked to imagine that they are two criminals who are hiding money. They have been caught by the police, separated, and each given two options: betray/defect or keep silent/cooperate. If one cooperates and the other one defects, the defector is able to keep all the money, while the cooperating player must pay a fine. If both remain silent, however, they both get half of the money. If both choose to defect, they will both have to pay half of the fine. This pay-off matrix is illustrated in **Figure 1B**.

*Trust Game.* In the Trust Game (Berg et al., 1995) participants can be either player A or player B. Player A has an amount of money and may decide to either send it to player B or to keep it all for himself/herself. If the money is sent to player B, the total is multiplied by four and then player B must choose to either send half back to player A or keep it all. During this experiment participants played both as player A and player B, with the order counterbalanced across the runs of the game. The pay-off matrix is given in **Figure 1C**.

#### *Cognitive control task*

Participants also completed a Go/no-Go task to measure cognitive control abilities (see De Neys et al., 2011). This task was administered once at the start of the experiment (i.e., prior to any essay reading/evaluation). At trial onset, a central fixation point was shown for 500 ms followed by a single letter for 500 ms (the target letter was either "W" or "M," counterbalanced across participants) with an intertrial interval of 1 s. Participants were instructed to respond as fast and as accurately as possible with a keypress whenever the target letter was present. A warning message appeared if they took longer than 500 ms or the response was incorrect. In total 100 trials were presented with 80% of the trials showing the target.

#### **ANALYSIS**

The key-dependent measures in this study were defection rates in the Prisoner's Dilemma and the Trust Game (for each game, participants could defect a total of 0, 1, 2, or 3 times per emotioninduction condition). These dependent measures are ordinal, and Kolomogorov–Smirnov and Shapiro–Wilk tests showed that the data were not normally distributed. As a result, we used nonparametric statistical tests (as has been done previously, Brosig, 2002; Falk et al., 2005). The data were analysed with a two-way mixed non-parametric design (2 cognitive control groups × 3 emotion conditions) with defection rate as the dependent variable (Field et al., 2012). This analysis was performed separately for the Prisoner's Dilemma and the Trust Game and *post-hoc* comparisons were carried out using Bonferroni corrected Wilcoxon Signed-Rank tests (two-tailed, alpha = 0.017) to explore any differences further.

The number of errors in the Go/no-Go task was used to calculate *d* for each participant as a measure of cognitive control ability. Using a median split, participants were divided into two groups according to this measure; a low (*d*- = 2*.*21–3*.*08, *N* = 14) and a high cognitive control group (*d*- = 3*.*24–8*.*60, *N* = 15). Planned Mann–Whitney U tests were then used to analyse whether the effect of emotion on socialexchange decision-making depended on between-subject differences in cognitive control, as measured by the Go/no-Go task. Within each cognitive control group, a Wilcoxon Signed-Rank test for two related samples was used to test for within-subject differences between the effects of emotion-induction condition on defection rates (Bonferroni corrected alpha = 0.017, two-tailed).

Individual SCL scores were z-transformed for subsequent analyses with a mixed design ANOVA comparing the three emotion induction conditions (within-subject) and cognitive control (between-subject). *Post-hoc* comparisons were performed using paired *t*-tests with Bonferroni corrected alpha (two-tailed, *p* = 0*.*017).

## **RESULTS**

# **EMOTIONAL MANIPULATION**

#### *Galvanic skin conductance*

A significant main effect of emotion condition on z-scored galvanic SCL (zSCL) was found [*f(*2*,* <sup>50</sup>*)* = 6*.*13, *p* = 0*.*004]. However, there was no main effect of cognitive control and there was no emotion condition × cognitive control group interaction (*p >* 0*.*05). *Post-hoc* analyses with paired *t*-test revealed that zSCL during the sympathy condition did not differ significantly from the neutral condition (*p >* 0*.*05). However, in the anger condition zSCL was significantly higher compared to the sympathy condition and to the neutral condition [*t(*28*)* = 2*.*63, *p* = 0*.*014, and *t(*28*)* = 4*.*12, *p* ≤ 0*.*001, respectively, Bonferroni corrected]. These findings show that anger induction, but not sympathy, is associated with a higher zSCL compared to the neutral emotional induction (**Figure 2**).

In order to evaluate whether zSCL was related to the effect of cognitive control on defection rate, Spearman's correlation analyses were performed separately for low and high cognitive control individuals. There were no significant correlations between defection rate and zSCL neither in high nor low cognitive control participants (*p >* 0*.*05).

#### *Self-report questionnaire*

We used a hierarchical cluster analysis procedure to determine the number of clusters that could be extracted from participants' responses on the self-report emotion questionnaire. This analysis was based on the Squared-Euclidian distance following Ward's method (Willebrand et al., 2002; Bigne and Andreu, 2004) and determined the number of clusters according to an agglomeration schedule as suggested by Burns and Burns (2008). We selected a three cluster solution, on the basis that adding further clusters

was significantly higher than that of the neutral induction (*p* = 0*.*005) and sympathy induction (*p* = 0*.*029). Sympathy SCL was not significantly higher than that of the neutral condition (*p >* 0*.*05). The asterisks highlight significant paired comparisons after Bonferroni correction (*p* ≤ 0*.*017). Error bars represent ±1 SEM.

had minimal additional effect on the agglomeration coefficient. Accordingly, a three cluster analysis was then performed using a *k* means approach, which grouped all 36 self-report emotion questionnaire items according to their similarity across participant ratings (Bigne and Andreu, 2004). The words found to be associated with each cluster are presented in **Figure 3A**, along with each cluster's Cronbach's alpha. **Figure 3B** illustrates these clusters according to the number of participants reporting words specific to each cluster in each emotion condition. *T*-tests showed that words from cluster 1 were more often reported to be experienced during the neutral condition than the anger [*t(*11*)* = 7*.*18, *p* = 0*.*015, Bonferroni corrected] or sympathy conditions [*t(*11*)* = 2*.*89, *p* ≤ 0*.*001]. In contrast, words from cluster 2 were more often experienced during the anger condition, compared to the sympathy and neutral condition [*t(*14*)* = 4*.*38, *p* = 0*.*001 vs. *t(*14*)* = 6*.*94, *p* ≤ 0*.*001]. Cluster 3 words were more often reported in the sympathy condition than in the anger [*t(*8*)* = 3*.*07, *p* = 0*.*015, Bonferroni corrected] or neutral condition [*t(*8*)* = 4*.*06, *p* = 0*.*004].

## **SOCIAL EXCHANGE TASKS** *The Prisoner's Dilemma*

The 2 (high and low cognitive control) × 3 (anger, sympathy, and neutral conditions) mixed design non-parametric analysis yielded a significant main effect of emotion (*Q* = 0*.*454, *p* = 0*.*002) and a significant interaction between cognitive control and emotion (*q* = 5*.*06, *p* = 0*.*01). The main effect of cognitive control was not significant (*p >* 0*.*05). *Post-hoc* Wilcoxon Signed-Rank tests showed a significantly higher defection rate after anger induction compared to sympathy induction (*Z* = −3*.*21, *p* = 0*.*001, Bonferroni corrected). While there was no significant difference between the defection rates following neutral and sympathy

asterisks highlight significant paired comparisons after Bonferroni correction

(*p* ≤ 0*.*017). Error bars represent ±1 SEM.

**FIGURE 3 | (A)** The table shows all 36 words from the self-report emotion questionnaire grouped into three different clusters identified by the results of the cluster analysis, along with each cluster's associated Cronbach's alpha. **(B)** For each cluster of words (as identified by the cluster analysis), the figure

induction (*p >* 0*.*05), the anger induction resulted in higher defection rates compared to the neutral induction (*Z* = −2*.*84, *p* = 0*.*004, Bonferroni corrected) (**Figure 4**).

To explore the interaction effect further, within-subject comparisons with Wilcoxon Signed-Rank tests were then performed for each cognitive control group separately. Defection rates did not differ significantly between emotion-induction conditions in high cognitive control participants (*p >* 0*.*05). In contrast, low cognitive control participants showed a significantly higher defection rate in the anger condition, compared to both neutral and sympathy inductions (*Z* = −2*.*98, *p* = 0*.*003, and *Z* = −2*.*90, *p* = 0*.*005, respectively, Bonferroni corrected). The increased defection rate for the neutral, compared to the sympathy condition, was not significant (*p >* 0*.*05) (**Figure 5**).

#### *The Trust Game*

The same non-parametric mixed design analysis was performed for the Trust Game. The results showed a significant main effect of emotion (*Q* = 9*.*10, *p* = 0*.*001), but no main effect of

cognitive control and no significant cognitive control × emotion interaction (*p >* 0*.*05). *Post-hoc* comparisons with Wilcoxon Signed-Rank tests yielded a significantly higher defection rate in the neutral condition compared to sympathy induction (*Z* = −2*.*45, *p* = 0*.*014, Bonferroni corrected) and a significantly higher defection rate after anger induction compared to sympathy induction (*Z* = −3*.*36, *p* = 0*.*001, Bonferroni corrected). The difference in defection rates between the neutral and anger conditions was not significant (*p >* 0*.*05) (**Figure 4**).

#### *Additional analyses*

Kruskall–Wallis tests for more than 2 independent samples did not find any influence of emotion-induction order on participants' defection rates for either game (*p >* 0*.*05). Additionally, defection rates did not differ depending on the "other participants" previous choice (defect or cooperate) in either game for any emotion (Wilcoxon Signed Ranks; *p >* 0*.*05).

The asymmetry of the effects of anger and sympathy (compared to neutral) on defection rates were tested using Wilcoxon Signed-Rank tests. There was no significant interaction effect of [anger – neutral] vs. [neutral – sympathy] for either game (*p >* 0*.*05). These results suggest that, despite sympathy and anger exerting opposite effects on decision-making (compared to neutral), the relative strength of these effects was symmetrical.

Finally, to evaluate whether the Prisoner's Dilemma and the Trust Game have different cognitive demands, the overall defection rates in both games were compared. Although participants chose defection more often in the Prisoner's Dilemma than in the Trust Game, the Wilcoxon Signed Ranks test did not reveal a significant difference between the cognitive control groups; neither in overall defection rates nor separately in each emotion condition (*p >* 0*.*05).

#### **DISCUSSION**

This study investigated the influence of partner-directed emotions on social decision-making. The experiment compared the effects of two emotion inductions (anger and sympathy) and one baseline (neutral) emotional condition, and assessed their differential impacts on decision-making in two social-exchange games—the Prisoner's Dilemma and the Trust Game.

The results of the self-report questionnaire indicated that the three emotion induction conditions were associated with distinct affective experiences. The feelings most associated with the anger induction were all negative and in keeping with common definitions of anger (see **Figure 3**). The cluster most associated with our sympathy induction included a mix of positive and negative feelings, suggesting that sympathy may be a more complex (or mixed) emotional experience. Specifically, some of the feelings are associated with empathic understanding of others (e.g., upset and also feeling strength in the knowledge that people can cope with a disease), while others may be more linked to heightened concern for others (e.g., feeling attentive and alert), or with the effect the other person's psychological state has on oneself (e.g., feeling inspired and interested). The cluster of feelings most strongly associated with the neutral condition was positive and relatively placid, which was also in keeping with our expectations. While this cluster was significantly more associated with the neutral condition than both emotional conditions, the sympathy condition did also load somewhat onto this cluster (clearly more so than the anger condition), suggesting that there may be a certain level of overlap between our neutral and sympathy conditions. It is worth noting, however, that the neutral condition showed no closer relationship than the anger condition with the cluster that was most associated with sympathy (i.e., Cluster 3).

Our skin conductance findings show clearer evidence of overlap between the sympathy and neutral conditions, in that our anger induction was associated with increased SCL, while our sympathy induction was not. This result is consistent with findings by Frodi and Lamb (1980) (see also Frodi et al., 2006), who showed that sympathy-oriented emotions had no significant impact on physiological responses. On the other hand, threat related stimuli such as angry faces, spiders, or snakes are detected faster due to evolutionary reasons (Öhman and Mineka, 2001; Öhman et al., 2001). This idea has received criticism, however, by those who suggest that the speeded responses to fearful or threatening stimuli are due to the relevance of the stimuli to the individual rather than its negative valence (Sander et al., 2003; Brosch et al., 2007, 2008, 2010). In the context of this study, it is possible that participants perceived the anger inductioncondition to be more relevant to their current situation which resulted in a stronger emotional response and inducing a desire in participants to do something to change their feelings. In contrast, induced sympathy may not always promote such strong action tendencies. Accordingly, anger results in higher arousal, while sympathy is more neutral in terms of the evoked physiological response. Another explanation may be that sympathy *does* have a physiological impact, but that this was simply not measurable through SCL in our experiment.

The results of the social-exchange games indicated that, although the sympathy and neutral conditions did not differ noticeably in their effects on physiological arousal, both the anger and the sympathy inductions had significant (and opposing) effects on participants' social decision-making. The direction of this effect was consistent with past findings—sympathy triggered lower defection rates and anger triggered higher defection rates compared to the neutral condition (Batson and Moran, 1999; Bosman and van Winden, 2002; Ben-Shakhar et al., 2004; Duersch and Servatka, 2007; Van Lange, 2008). Moreover, the strengths of these impacts were found to be more or less symmetrical compared to the neutral condition, despite only the anger condition having significant influences on participant's physiological arousal.

Though the defection rate tended to show at least some increase from sympathy to neutral and from neutral to anger in both games, there were subtle differences between the two games: in the Prisoner's Dilemma significant differences were found between anger and neutral, and in the Trust Game between sympathy and neutral. Therefore, both games were affected by the emotion manipulations, but in slightly different ways. One possible explanation for this pattern of results could be the different framing of the choices in the games. The Prisoner's Dilemma holds a loss frame, because one possible outcome of the game is that the participant might lose money. In contrast, the Trust Game holds a gain frame, since the participant can either gain money or else they will neither lose nor gain. Framing effects have yielded conflicting results in different studies. Though there are a wide range of experiments showing that such framing does influence individuals' decisions (De Dreu and McCusker, 1997; Tversky and Kahneman, 1981; Frank and Claus, 2006) other studies find that not all people are affected by the framing effect (Peters et al., 2006b). The results of the current study hint that framing effects may interact with emotion in social decision games. In the Prisoner's Dilemma participants are generally more driven to avoid loss, and the anger condition may make these losses more salient and the option to defect even more tempting. Conversely, the Trust Game rewards cooperation, and this may be further promoted by sympathy rather than anger. Future studies could explore these possible effects of framing on the influence of emotion in social decision-making.

A particularly interesting finding from this study is that the effect of anger on decision-making in the Prisoner's Dilemma depended on cognitive control ability—as determined by performance in a Go/no-Go task. The effect of anger was driven almost exclusively by the low cognitive control group. This is consistent with De Neys et al. (2011), who found that participants showing high defection rates in the ultimatum game also made more mistakes in a Go/no-Go task, compared to the low defection rate participants. Our SCL analysis, however, did not indicate a difference in the strength of experienced emotions between low and high cognitive control participants. It is possible that high cognitive control participants were better at focusing on the game itself, and were therefore less affected by their emotions. Kollock (1998) as well as Komorita and Parks (1999) note that, in the long-term, cooperation can bring bigger benefits to the players than defection, and high cognitive control participants may be more likely to use this logic while playing the game. On the other hand, low cognitive control participants may be relying more on intuition (Stanovich and West, 2000; Sunstein, 2005), and in particular an "outrage heuristic," which promotes a desire to punish others as retribution for their anger (Kahneman and Frederick, 2002).

One strength of the present study lies in its within-subject design, whereby the influence of both emotions—sympathy and anger—were measured and compared to a neutral baseline within the same group of participants. The value of a within-subject design results particularly in the reduction in variance when comparing our emotional manipulations. In between-subject designs, such comparisons may be confounded by variance due to individual differences or context effects, giving us less power to address the effect of the emotional responses we are interested in. Moreover, in exploring the effect of our between-subject measure of cognitive control, a within-subject emotional manipulation allows us to address not only the role of cognitive control on the effect of one emotion (e.g., anger) on decision-making, but importantly to address its role in the *change* in decision-making between two emotion conditions. In addition to this, the study assessed the influence of emotions directed *toward* the other player with whom participants were playing, rather than being purely incidental to the decision-scenario. To our knowledge, this is the first study that uses a within-subject design for investigating two different emotions directed toward the other person.

One limitation to the current study may be possible reputation effects induced through multiple repetitions of the games. Although game order was counterbalanced, and we did not find effects of reputation in the three sequential runs of each game, future studies might randomize the trials completely, such that participants play multiple games against the three partners in a fully interleaved manner. Another limitation may be in the selfreport emotion questionnaire used, where sympathy and anger conditions had different amounts of words representing the possible emotions (indeed while the word "anger" was present in the list, "sympathy" was not). Finally, our data show that anger and sympathy differ in their experiential complexity and associated physiological arousal, as well as in their valence. As such, it is yet unclear precisely which of these components of anger and sympathy best explain their differential effects on social decisionmaking. Future studies could directly compare the motivational effects of emotions that are of similar levels of experiential complexity but differ in terms of valence and/or arousal, or conversely that are of similar valence and arousal but differ in their complexity.

This study shows the differential effects of sympathy and anger (directed toward the opponent) on socially-interactive decisionmaking. Emotions can be beneficial when making decisions especially when people do not have time to consider all the possible choice options and their possible outcomes carefully. Specifically, emotions can help us to solve a problem more efficiently, and in better accordance with our personal goals and moral and social norms, than can decision-making in the absence of emotional influence (Peters et al., 2006a; Pfister and Böhm, 2008). Indeed, the results of this study show that sympathy and anger, directed toward ones opponent, can have emotion-specific influences on our social interaction, further reflecting the goaldirected nature of emotion influences on decision-making. If a person feels angry, and is motivated to use this emotion in the decision process, their tendency to defect increases. In contrast, if they are motivated to help their partner (as is typical of sympathy) then their level of co-operation will increase. In

#### **REFERENCES**


our Prisoner's Dilemma game, healthy individuals with higher cognitive control tended to rely less on their anger felt toward others in their decision-making, while individuals with lower cognitive control tended to be more heavily influenced by feelings of anger and chose to defect more often, perhaps as punishment or to express their anger. These findings provide support for complex, and likely bidirectional, interactions between emotion and cognition in decision-making. Heuristic-based thinking styles have also been suggested to account for judgments and decisions made in many moral and social contexts (Sunstein, 2005) such as in the Trolley Dilemma, emission trading or Asian Disease problem. Moreover, emotion-based heuristics (or "affect heuristics") have been proposed to provoke judgments and decisions that are heavily biased by our emotional responses without the involvement of significant cognitive deliberation (Slovic et al., 2007). In accordance with such accounts, feelings of anger would be expected to provoke behaviors that can express this anger and seek retribution (as can be done through defecting). On the other hand, sympathy promotes a desire to help the person in need, and this motivation leads to enhanced co-operation. In keeping with accounts that emotions can bias judgment and decision-making through a heuristic route, our findings suggest that people who are more likely to utilize heuristic processing styles (as in the case of our low cognitive control participants) will be more heavily influenced by their emotional responses.

## **ACKNOWLEDGMENTS**

This work was supported by Wellcome Trust grant (ABH00FA032YBH064) to Vinod Goel. The authors would like to thank the hosts of this special issue, Corrado Corradi-Dell'Acqua, Susanne Leiberg, Leonie Koban, Patrik Vuilleumier, and Ernst Fehr, along with two anonymous reviewers for their helpful comments on a previous version of this manuscript.


reputations in an investment game. *Games Econ. Behav.* 72, 361–375. doi: 10.1016/j.geb.2010.09.002


brain. *Science* 313, 684–687. doi: 10.1126/science.1128356


Cambridge: Cambridge University Press.


University Press), 49–81. doi: 10.1017/CBO9780511808098.004


through the eyes of the players. *J. Exp. Psychol. Gen.* 139, 743–755. doi: 10.1037/a0020280


Stemmler, G. (1997).Selective activation of traits: boundary conditions for the activation of anger. *Pers. Individ. Dif.* 22, 213–233. doi: 10.1016/S0191- 886900189-4

Sunstein, C. R. (2005). Moral heuristics. *Behav. Brain Sci.* 28, 531–541.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 05 July 2013; published online: 25 July 2013. Citation: Eimontaite I, Nicolle A, Schindler I and Goel V (2013) The effect of partner-directed emotion in social exchange decision-making. Front. Psychol. 4:469. doi: 10.3389/fpsyg. 2013.00469*

*This article was submitted to Frontiers in Emotion Science, a specialty of Frontiers in Psychology.*

*Copyright © 2013 Eimontaite, Nicolle, Schindler and Goel. This is an openaccess article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Reappraising social emotions: the role of inferior frontal gyrus, temporo-parietal junction and insula in interpersonal emotion regulation

#### *Alessandro Grecucci <sup>1</sup> \*, Cinzia Giorgetta2, Nicolao Bonini <sup>3</sup> and Alan G. Sanfey4*

*<sup>1</sup> Department of Psychology and Cognitive Science, University of Trento, Trento, Italy*

*<sup>2</sup> Centro Nazionale Della Ricerca, Istituto di Scienze e Tecnologie Della Cognizione, Trento, Italy*

*<sup>3</sup> Department of Economics and Management, University of Trento, Trento, Italy*

*<sup>4</sup> Donders Institute for Brain, Cognition and Behavior and Behavioral Science Institute, Radboud University, Nijmegen, Netherlands*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Jan B. Engelmann, University of Zurich, Switzerland Antonia New, The Mount Sinai Hospital, USA Mascha Van 'T Wout, Brown University, USA*

#### *\*Correspondence:*

*Alessandro Grecucci, Department of Psychology and Cognitive Science, University of Trento, Corso Bettini 31, Rovereto 38068, Italy e-mail: alessandro.grecucci@unitn.it*

Previous studies have reported the effect of emotion regulation (ER) strategies on both individual and social decision-making, however, the effect of regulation on socially driven emotions independent of decisions is still unclear. In the present study, we investigated the neural effects of using reappraisal to both up- and down-regulate socially driven emotions. Participants played the Dictator Game (DG) in the role of recipient while undergoing fMRI, and concurrently applied the strategies of either up-regulation (reappraising the proposer's intentions as more negative), down-regulation (reappraising the proposer's intentions as less negative), as well as a baseline "look" condition. Results showed that regions responding to the implementation of reappraisal (effect of strategy, that is, "regulating regions") were the inferior and middle frontal gyrus, temporo parietal junction and insula bilaterally. Importantly, the middle frontal gyrus activation correlated with the frequency of regulatory strategies in daily life, with the insula activation correlating with the perceived ability to reappraise the emotions elicited by the social situation. Regions regulated by reappraisal (effect of regulation, that is, "regulated regions") were the striatum, the posterior cingulate and the insula, showing increased activation for the up-regulation and reduced activation for down-regulation, both compared to the baseline condition. When analyzing the separate effects of partners' behavior, selfish behavior produced an activation of the insula, not observed when subjects were treated altruistically. Here we show for the first time that interpersonal ER strategies can strongly affect neural responses when experiencing socially driven emotions. Clinical implications of these findings are also discussed to understand how the way we interpret others' intentions may affect the way we emotionally react.

**Keywords: interpersonal emotion regulation, decision-making, social interactions, mentalizing**

# **INTRODUCTION**

Perspectives on affective neuroscience suggest that brain structures which generate emotional responses can be successfully regulated by control regions when subjects are asked to apply cognitive strategies to emotion eliciting stimuli such as unpleasant pictures (Golkar et al., 2012; Ochsner et al., 2012). Emotion regulation (ER) refers to a set of different strategies by which "individuals influence which emotions they have, when they have them, and how they experience and express these emotions" (cf. Gross, 2007). Although mechanisms of basic emotion self-regulation have been, at least in part, recently uncovered, surprisingly little empirical work exists on an important topic: the specific neurocognitive mechanisms behind interpersonal emotion regulation (IER), a particular form of ER applied to socially driven emotions.

ER can refer not only to people's capacity to manage their own emotions, but importantly can also extend to regulating emotions that result from the interaction with others (Grecucci, 2012; Grecucci et al., 2013b). Previous studies have examined the processes that individuals use to influence which emotions they generate, when they do so, and how these emotions are experienced or expressed (Gross, 1998), and therefore we know that different attentive, behavioral, emotional, or interpretative strategies can be used also at an interpersonal level (Fonagy, 2006). Of particular interest for the present paper are studies examining the use of a strategy to regulate an existing or ongoing emotional response, typically known as reappraisal. This strategy involves reinterpreting the meaning of a stimulus to change one's emotional response to it (Gross, 1998). Subjects are usually asked in this context to build an interpretation of the emotional stimulus in such a way as to increase or decrease their emotional response (respectively, up- and down-regulation), and behavioral studies have shown that reappraisal is one of the most efficient strategies for regulating negative emotional responses (Gross, 2002). However, reappraisal applied to interpersonal contexts, that is, focusing on the interpretations of others'intentions, is relatively neglected in the literature. Despite the existence of an extensive literature on emotion "self-regulation," focused primarily on the regulation of basic emotions such as fear and disgust in relation to visual stimuli (Ochsner and Gross, 2005, for a review), research on regulation in social interactive situations (e.g., IER) is scant (e.g., Koenigsberg et al., 2011; Grecucci et al., 2013a,b; Grecucci and Sanfey, 2013).

Notably, processing socially cued emotions engages differential networks than does non-socially cued emotion (Britton et al., 2006; Harris et al., 2007; Lestou et al., 2008), thus motivating further exploration of the regulation of socially induced emotions. The interest on such "social regulation" has been explored in a recent study examining the ER of subjects while looking at pictures depicting social vs. non-social scenes (Koenigsberg et al., 2011). This study had subjects observing emotional vs. neutral pictures while applying reappraisal strategies, but the novelty of the study was in the usage of a subset of International Affective Pictures depicting scenes with social features (e.g., people in situations of loss, abuse, aggression*...* ) instead of simple emotional pictures. Interestingly, exposure to pictures depicting social situations activated brain areas partially involved in social cognition, such as the superior and middle temporal gyri, in addition to emotional and cognitive structures similar to previous non-social studies.

However, though in this study people were asked to reappraise emotions elicited by pictures depicting social scenarios, they were not exposed to real social interactive situations. Studying the neural systems involved in the regulation of actual interpersonal situations is particularly important given the relevance of failure in regulating interpersonal responses in psychiatric disorders (Phillips et al., 2003; Ochsner and Gross, 2008; Grecucci, 2012; Grecucci et al., 2013c). Moreover, in the aforementioned study (Koenigsberg et al., 2011) a particular form of reappraisal was used, namely "distancing," a "self-focused" strategy in which subjects view an emotional stimulus from the perspective of a detached and distant observer (Koenigsberg et al., 2011; Ochsner and Gross, 2013). This strategy may be reasonable when looking at a picture but its use may be detrimental when interacting with a real person. In contrast, in the present study we aimed to use a reappraisal strategy focused on the "intention of others," which involves a reinterpretation of the meaning of the other person's mind, behavior and intentions. One advantage of the latter strategy is that reinterpretation can be in both more or less negative directions thus providing the opportunity to study both up- and down-regulation effects, whereas distancing is only intended to down-regulate one's emotions. This is of notable importance given that some clinical populations (e.g., paranoid and borderline personality disorders, anxiety, schizophrenia, etc.) are characterized by interpreting the intentions of others in a malevolent way, thus causing inappropriate interpersonal and emotional reactions (Grecucci et al., 2013c). Clinicians of different schools defined this process as "projective identification" (Klein, 1946; Clarkin et al., 2006) or "hypermentalizing" (Allen and Fonagy, 2006).

In a previous study we tried to fill this gap by employing research paradigms designed to explore social economic decisions, and we evaluated whether interactive emotion regulation can occur through the mechanisms involved in self-regulation of negative emotions. These studies (Grecucci et al., 2013a,b) showed that an IER strategy of *reappraising the intentions of the other player as less negative*, or mentalizing-reappraisal (a particular kind of "reappraisal"), is effective in changing both interpersonal decisions (i.e., rejection rates of unfair offers in the context of a socio-economic game) (Grecucci et al., 2013a), as well-subjective responses to emotion themselves (Grecucci et al., 2013b). The task used in one of these experiments (Grecucci et al., 2013a) was the classic Ultimatum Game, where participants played the role of responder (Guth et al., 1982). The study showed that subjects' decisions were strongly modulated by the reappraisal strategy used: less rejections of unfair offers when down-regulating their emotions and more rejections when upregulating their emotions. The modulation was visible in an area of the brain previously involved in the aversive reactions elicited by unfair offers, namely the insula. The posterior part of the insula showed a similar pattern of activation as was shown behaviorally (less activity for down- and more for up-regulation as compared to the neutral baseline). A limitation of that study was that the task required subjects to respond to economic offers with the possibility of rejecting the bad proposals, thus leading to lesser gain for the proposer him or herself. That is, subjects could punish proposers for the bad behavior directed toward them, and indeed, one of the primary emotions subjects reported when treated unfairly was anger. Therefore, it could well be that the punishment that subjects could inflict on proposers was itself a way to show their feelings and thus modulate their own emotional states. In other words, behavioral and neural responses showing modulation according to the reappraisal strategies could have been more concerned with the decision than with the socially induced emotions themselves. To further examine, at the neural level, how purely socially induced emotions are regulated, the same subjects played another socio-economic game called the Dictator Game (DG) (Kahneman et al., 1986), once again as responders. In the DG players must passively accept socioeconomic offers, usually both fair and unfair, and therefore do not have the possibility to punish the proposers' unfair behavior and to potentially vent their anger. Importantly, in this task we can focus more on the neural activations, without the complication of having subjects involved in both making a decision and providing a motor response. In other words we have the unique opportunity to observe the neural effects of the regulation of socially elicited emotions without the involvement of other decision processes.

In terms of the neural structures involved in ER, the literature on "self" regulation typically distinguishes regions that implement the strategy (Regulating regions) and regions that are modulated by the strategy (Regulated regions). According to a recent model of the cognitive control of emotions (MCCE, Ochsner et al., 2012), the regions involved in emotion generation that can be regulated, are, in order of importance: the amygdala, with less evidence for other regions such as the ventral striatum, the ventromedial prefrontal cortex (wmPFC), and the insula. At the same time, other regions appear to act as control systems that implement the regulatory strategy. These regions are primarily the dorsolateral prefrontal cortex (dlPFC), the anterior cingulate cortex (ACC), the ventrolateral prefrontal cortex (vlPFC) and the dorsomedial prefrontal cortex (dmPFC). However, we do not know if this model can be applied to the context of IER. Social emotions rely on different mechanisms and activate different brain areas as do non-social emotions (Britton et al., 2006), and therefore IER may be of a qualitatively different nature from self-emotion regulation (Grecucci and Sanfey, 2013).

Thus, the first goal of the present study is to identify neural correlates and possible modulations of the regulation of socially induced emotions stemming from interactive situations. In particular we aim to uncover how dedicated brain areas respond to the implementation of mentalizing-reappraisal strategies (we define them: "Regulating regions") when regulating socially induced emotions such as those elicited by selfish and altruistic behaviors during a DG. Given the particular interactive task used in the present study, we expect that brain areas more connected with building an interpretation of others' minds and intentions will be activated, specifically the temporo-parietal junction (TPJ). In recent years TPJ activation has been connected to both social perception (Allison et al., 2000; Kourtzi and Kanwisher, 2000) as well as to attributing intentions (Van Overwalle, 2009) and mental states to others, namely theory of mind (Frith and Frith, 2003). These results can extend a useful model of ER (e.g., the MCCE) by adding social—interpersonal mechanisms.

A second goal of the present study is to explore brain regions that are modulated by these strategies ("Regulated regions"). We expect social interactions to involve different neural structures as compared to those of observing "scenes of humans interacting in a negative way" (such as scenarios of aggression or mourning). A recent study found that an emotional structure involved when looking at social emotional pictures was the amygdala (Koenigsberg et al., 2011), which is likely connected with the unpleasantness of those scenarios themselves than to the interpersonal reactions. In contrast, previous studies involving fair and unfair socioeconomic behaviors have shown that the insula may be responsible for negative reactions when treated unfairly by another player in the Ultimatum game (Sanfey et al., 2003), and thus we expect that the insula will be active in the present study when subjects are treated unfairly. We will test explicitly for the emotions invoked by assessing affective reactions following the game play. Based on the two previous goals we aim to determine the neural circuitry underlying interpersonal regulatory processes. In line with previous studies we expect a network of areas working together in order to produce successful regulation of emotions elicited by social situations. This will be formally tested in a dynamic causal modeling (DCM).

Strictly related with goal one and two, the third goal of the study is to inquire what happens when we reappraise in a negative way the intentions of others. The vast majority of the previous studies focused their attention on the effect of down regulating one's emotion. However, understanding what happens when we up-regulate emotions is of critical importance. The up-regulation of the emotion is commonly observed in psychiatric patients (in the form of excessive emotional reactivity or inappropriate emotionally laden behaviors), and it has been hypothesized to be caused by failures in the way we interpret others'intentions (Allen and Fonagy, 2006; Clarkin et al., 2006). The way we interpret others' mind, indeed affects the way we emotionally respond. This is of undeniable relevance as it covers many clinical phenomena associated with negative style of thinking and its effect on interpersonal emotional reactions as visible in paranoid, borderline patients and related disorders. The paradigm used in this experiment gives us the opportunity to have subjects reappraising events in a more or less negative way, thus providing the opportunity to study both up- and down-regulation effects on the brain and on emotional perception.

Finally, a fourth goal of this study is to detect both common and different brain regions and subjective experience, associated with the experience of being treated fairly, moderately unfairly or very unfairly.

# **METHODS**

## **PARTICIPANTS**

Twenty-one participants (11 males, mean age: 23.5 ± 3.6 years) participated in the study. Participants had normal or corrected to normal vision and had no history of psychiatric, medical or neurological illness, as verified by a semistructured interview by a physician. All participants provided written informed consent, as approved by the local ethical committee, and were paid 35 euros for participation.

### **ASSESSMENT, TRAINING PROCEDURE, PARADIGM, AND FORMAL DEBRIEFING**

The experimental procedure comprised of four phases. A general cognitive and emotional assessment (including the Emotion Regulation Questionnaire, ERQ, Gross and John, 2003), followed by training and testing in ER techniques. Then, subjects underwent scanning with fMRI while playing rounds of the Ultimatum Game and DG under conditions of ER in two separate runs intermingled by a break. Finally, there was a formal debriefing phase. Importantly, the sequence of the 4 phases was fixed having the subjects performing first the training, than the UG, followed by the DG and finally the debriefing. Participants were told they will be playing with every partner twice in two different games (UG and DG). They were also told that partners were real and that they made two independent offers (one per game) recorded before running the experiment. The two offers were randomly assigned to every player in a way to avoid carry over effects of reputation from one game to the other. In a previous paper we reported results on Ultimatum Game (Grecucci et al., 2013a), therefore in the present paper we concentrate on the results of the DG task.

In line with the previous formal operationalization of mentalizing-reappraisal (see Grecucci et al., 2013b), participants were asked in the training phase to reappraise the social situation following formal instructions. "When you are required to "up-regulate" you should interpret the intentions and behavior of your partner as more negative or potentially bad (instruction: "increase"); when you are required to "down-regulate" you should interpret the intentions and behavior as less negative or potentially good (instruction: "decrease"), when you are required to "look" you should try to perceive the situation spontaneously as it is without any effort to build any particular interpretation of it." They were given an example of a common negative situation and how it can be reinterpreted (reappraised) in such a way as to make it either more or less negative (See Grecucci et al., 2013b). To ensure subjects understood the instructions and were successfully applying the required reappraisal strategies, they were asked to reappraise while viewing pictures from the IAPS picture set (Lang et al., 1997). Eighteen unpleasant IAPS pictures were selected and divided into three subsets to be used across the reappraisal conditions (up, down, and look). After a picture was presented for 5 s, participants rated them according to valence and arousal dimensions using the Self-Assessment Manikin procedure (Lang, 1994). If the experimenter was satisfied by the reappraisal strategies used, the participant was introduced to the last part of the training, the DG. First, instructions were given on the DG (see **Figure 1A** for a timeline). The task instructions emphasized that the different partners in the game would play the game independently of each other, and participants were led to believe the games would be played for real with the set of partners they saw.

After the basic DG instructions, subjects were given instructions on how to apply reappraisal to DG. In the DG-training phase, each participant played three practice rounds of the DG as responders, twice in which they were asked to reappraise (according to the strategies indicated), and once in which they played without any reappraisal instruction (baseline condition). The instructions given on how to apply reappraisal strategies were as follows: "It is very important that you now try to apply the reappraising strategies learned in the IAPS-training to the situations evoked by the DG. In particular, you should try to come up with possible interpretations of the intentions and behaviors of the proposer in a way to make it more (up regulation) or less negative (down regulation). For example, when instructed to "increase" you may think the player is a selfish person (intentions) and wants to keep all the money (behavior). Whereas, when you have to "decrease," you may think that the player has financial problems and is giving you the best offer they can." In the "look" condition they were asked to read and emotionally respond to the offer in the most natural and spontaneous way. Participants were debriefed following these three practice trials and asked to report their strategies for each trial. After the training, participants entered the scanner and played a block of 20 rounds for each of the three regulation conditions counterbalanced across participants, for a total of 60 rounds as recipients, with each trial proposal involving a division of C10.

The set of offers received by each participant was pre-assigned. The set of 20 offers comprised of 7 fair offers (C5 to each player) and 13 unfair offers, defined as offering the participant less than half of the money. The unfair set was composed of 7 very unfair offers of C1, and of 6 mid-range values (2 offers of C2, 2 offers of C3 and 2 offers of C4). Half of the offers were made by a male partner, and half by a female partner. The order of partners and the pictures associated with each offer was completely randomized. Participants first saw a picture of the proposer on that round, followed by the offer of that player. After the offer was made, participants applied the reappraisal strategy required. To encourage participants to pay attention to the task it was emphasized that they would be paid according to the other players choice in the game (even though for local ethical reasons they were paid the same), and to make them responsive, they were required to press a button to advance to the next trial. In a post scan session participants were exposed to two samples of rounds (specifically

on involving the fair 5:5 offer and one involving the unfair 1:9 offer) used during the scanning session and asked to evaluate the strength of emotions elicited (anger, sadness, disgust, surprise, and happiness) on a 9-point Likert scale. After each of these rounds they were also asked to indicate whether they felt their emotions were modulated according to the strategy when asked to apply up- and down-regulation on each of these sample trials.

#### **SCANNING PROCEDURE**

Whole brain distortion-corrected EPI with 32 axial slices (3-mmthick, 1-mm gap) were collected at 4T (Bruker MedSpec MRI) with a T2∗-sensitive gradient echo spiral pulse sequence (TR of 2.2 s, TE 33 ms, 75◦ flip angle, 64 × 64 data acquisition matrix). T2-weighted spin-echo scans were acquired for anatomical localization using the same slice prescription. Stimulus presentation and data acquisition were controlled using E-prime software. Responses were made with the index and middle fingers of the right hand using two buttons on a four button MRI-compatible response box.

## **fMRI DATA PRE-PROCESSING AND GENERAL LINEAR MODEL ANALYSIS**

Functional images were slice time corrected and motion corrected using SPM8 (Wellcome Department of Cognitive Neurology, London). For all participants, we acquired 738 volumes (246 each fMRI-run); the first 3 volumes were discarded for each run. In preprocessing of the data, the EPI volumes were spatially realigned to correct for movement artifacts (Ashburner and Friston, 2003) and motion corrected by distortions interactions (Andersson et al., 2001), and smoothed using 9-mm Gaussian kernel to account for residual intersubject differences (Worsley and Friston, 1995). For statistical analysis, we used the general linear model implemented in SPM8 as an eventrelated design and we modeled the onset of each category and convolved with the canonical hemodynamic response function (HRF, event duration = 0), then we estimated the effect size for each participant for each of the relevant 9 conditions (fair offers down-regulate, fair offers look, fair offers up-regulate, mid offers down-regulate, mid offers look, mid offers up-regulate, unfair offers down-regulate, unfair offers look, unfair offer up-regulate) using the general linear model. Because our main question concerned the regulation of the behavior of the partners in the DG, activation onsets were aligned with the display of the proposed monetary offer on each trial. Finally, the first-level analyses included also the parameters of the realignment (motion correction) as covariates of no interest. Next, we obtained 9 contrast images per participants, corresponding to the 9 conditions of interest. Statistical threshold were set to p-corr. = 0.05 corrected for multiple comparisons at the cluster level (cluster size estimated at p-unc. = 0.001), considering the whole brain as the volume of interest. Furthermore, region-of-interest (ROI) analyses were also carried out with the aim to provide additional information confirming the statistically valid inferences based on main effects and simple main effects off the random effects analysis. Each ROI consisted of a sphere of 8 mm of diameter centered around the peak of activation using Marsbar toolbox (Brett et al., 2002).

### **DCM AND BAYESIAN MODEL SELECTION**

DCM (Friston et al., 2003) was used to explore experimentally induced modulations (Stephan et al., 2007) in key regions of interest to better understand the effects uncovered in the general linear model analyses. DCM models can shed light on how the neural dynamics are shaped by experimentally controlled manipulation. With DCM we aimed to test which regions were involved in the effect of ER of social interactive situations. To ensure compatibility, the choice of subject-specific coordinates was guided by group maxima as derived by the GLM analyses, and adapted to each individual by adjusting for closest maxima. Regional time series of each subject was extracted as the first eigenvariate of all activated voxels within a 8 mm radius around the maxima. BMS was based on the same GLM model of the RFX analyses described above.

# **RESULTS**

## **RATINGS RESULTS**

We first examined if the affective ratings when reappraising IAPS pictures were different across conditions in the training phase (also see Grecucci et al., 2013a). To calculate the ability to reappraise the stimuli, we calculated the fluctuations of both arousal and valence over the baseline "look" condition (see **Figure 1B**). We ran paired sample *t*-tests, with participants' subjective ratings separately for both arousal and valence as dependent variables. Both comparisons were all significant, indicating that participants appeared to have learned reappraisal abilities—Valence: down vs. up [*t(*19*)* = 549, *p <* 0*.*001]; Arousal: down vs. up [*t(*19*)* = −419, *p <* 0*.*001]. Subjects rated their arousal as increasing in the upregulation and decreasing in the down-regulation, while valence was decreased in the down-regulation (meaning it was less negative), and increased in the up-regulation (more negative).

To understand which were the emotions that might be involved when reappraising the social situation of DG, and to check for confidence when applying the strategies, we analyzed the debriefing questionnaires. Notably, this debriefing exposed subjects to the same kind of stimuli taken from the scanning session, but, added questions to understand (1) the emotions involved, (2) the level of emotional strength and (3) the perceived ability to reappraise. One participant was excluded due to non-completion of the ratings. First, we performed an ANOVA with factors being Fairness (C1 vs. C5) and Type of emotion (anger, sadness, disgust, surprise, happiness). This returned a significant main effect of Fairness [*F(*1*,* <sup>19</sup>*)* = 15*,* 000, *p <* 0*.*001], of Type of emotion [*F(*4*,* <sup>76</sup>*)* = 7466, *p <* 0*.*0001], as well as the interaction [*F(*4*,* <sup>76</sup>*)* = 39*,* 920, *p <* 0*.*0001]. Then dependent-sample *t*-tests were performed using subjective ratings for every couple of emotions per time as dependent variables. Results demonstrate that the level of anger significantly differed from most of other emotions [anger-disgust: *t(*19*)* = 2058, *p <* 0*.*05; anger-surprise *t(*19*)* = 2868, *p <* 0*.*01; anger-happiness: *t(*19*)* = 6064, *p <* 0*.*001; anger-sadness: *t(*19*)* = 296, *p <* 0*.*05]; disgust differed from happiness [*t(*19*)* = 4807, *p <* 0*.*001] but not from surprise [*t(*19*)* = 1539, *p* = 0*.*14], and from sadness [*t(*19*)* = 847, *p* = 0*.*408]; surprise differed from happiness [*t(*19*)* = 4578, *p <* 0*.*001], but not from sadness [*t(*19*)* = −607, *p* = 0*.*55]; happiness differed from sadness [*t(*19*)* = −4188, *p <* 0*.*001]. However, when correcting for multiple comparisons (Bonferroni, *p* = 0*.*005) anger did not differ anymore from disgust and from sadness, and surprise did not differ from happiness. Overall, these results indicate that the emotion elicited by the unfair offers in a post scan session identical to the one used in the scanning session, and presumably modulated by the reappraisal strategies when subjects reappraised the DG rounds, was anger followed by sadness and disgust (see **Figure 1C**). Finally, in a manipulation check, participants were asked to indicate whether they felt their emotions changed according to the strategy adopted (see **Figure 1D**). Results were computed as deviations from the mean (5 point in a scale from 1 to 9) using dependent-sample *t*-test with subjective ratings for each of two offers as dependent variables. Participant ratings indicate that in the "Down" condition, both fair (5:5) and unfair (1:9) offers were modulated in the predicted direction [respectively, *t(*1*,* <sup>20</sup>*)* = −2416, *p <* 0*.*05 and *t(*1*,* <sup>20</sup>*)* = −3141, *p <* 0*.*05], while in the "Up" condition only the unfair offer was modulated in the expected direction [*t(*1*,* <sup>20</sup>*)* = 2234, *p <* 0*.*05; *t(*1*,* <sup>20</sup>*)* = 576, *p >* 0*.*05 for the fair offer]. Please note that these results were also partially presented in a previous study (Grecucci et al., 2013a).

# **fMRI RESULTS**

### *Main effect of strategy*

To begin with, the main effect of regulation strategy (down + up *>* look across all trial types) was computed to explore the brain structures involved when applying the strategy reappraisalmentalizing to the social situation of the DG as compared to the baseline condition of merely observing the offers. This analysis showed activations of, in order of significance, the left middle frontal gyrus, a swathe of temporo-parietal regions bilaterally, the insula bilaterally and the left inferior frontal gyrus. (see **Figure 2** and **Table 1**). In addition, the IFG positively correlated with ERQ measures and insula was positively correlated with the perceived change in emotional response as an effect of up-regulating and negatively when down-regulating (*p <* 0*.*05), supporting the insula's role in IER.

#### *Separate effects of up- and down-regulation strategies*

In order to test for differences between the two regulatory strategies, we separately computed the effects of up- and down-regulation. These contrasts were each computed by comparing to the baseline look condition. Down-regulation strategy involved significant activation of the TPJ bilaterally, the left middle and right superior temporal gyrus and the left inferior frontal gyrus (**Table 2A**), whereas, the up-regulation strategy revealed the right middle temporal gyrus, the left insula, the right superior temporal gyrus, the left striatum, the left inferior frontal gyrus and the left inferior parietal gyrus (**Table 2B**). In other words, the way we interpret others' intention (mentalize), affect the activity of brain regions associated with unpleasant emotional reactivity (insula), and with the perception of others (semantic areas in temporal regions).

#### *Regulation effects*

Similarly to results of a previous study (Grecucci et al., 2013a), where some activations were reduced when down-regulating and others increased when up-regulating, we expected the effects of the applied strategies to produce varied effects across key brain


*ˆinc. TPJ sites, \*p* <sup>=</sup> *0.05 FEW.*

# **Table 2A | DOWN regulation (DOWN** *>* **LOOK for unfair + midfair).**


**Table 2B | UP regulation (UP** *>* **LOOK for unfair + midfair).**


*\*p* <sup>=</sup> *0.05 FWE.*

regions. To test for this hypothesis we computed the contrast down *<* look *<* up. The regions modulated by the strategies were, in order of significance, the striatum bilaterally, the posterior cingulate cortex and the insula. Of particular interest for the present paper are the insula for its well-known role in socioeconomic games, the striatum, often modulated in reward experiments (Staudinger et al., 2009), and the posterior cingulate cortex.

To better understand the activity patterns of these three regions, we extracted the signal from the voxels from a sphere of 8 mm around the peak of activity using Marsbar toolbox (Brett et al., 2002). As shown in the bar plots, the insula, the cingulate and the striatum were clearly modulated by the strategies, each showing down *<* look *<* up behavior (See **Figure 3**, **Table 3**). Notably, insula activity was correlated with the level of anger experienced by subjects when receiving a very unfair offer (*p <* 0*.*05), thus confirming the hypothesis derived from clinical observations that if we perceive in a negative way the intentions of others this will affect our interpersonal emotions and reactions.

#### *Interaction of strategy with different types of social behaviors*

To examine how the regulation strategies were applied across different types of social behavior observed by the subjects (fair, moderately unfair, very unfair), we computed three separate contrasts for each set of behaviors when regulating the associated emotions (up and down vs. baseline for each of fair, moderately fair and very unfair behaviors). This set of analyses demonstrated several areas commonly activated independent of offer type, but, also some differences. This result was further confirmed when computing conjunction and disjunction analyses for the three contrasts (see **Figure 3**, as well as **Table 4**). A conjunction analysis returned the common areas active for all the three types of behaviors, and a disjunction analysis was computed by collapsing between unfair and mid fair (previously exploratory analyses had shown that they were very similar), and contrasting them to the fair condition with exclusive contrast. These analyses returned common areas: the inferior frontal gyrus, the middle temporal and parietal cortices, together with the occipital gyrus (**Figure 4**, on the bottom left), and also specific areas: the middle frontal gyrus, the TPJ, the insula, and loci on the temporal cortex were only active during moderately fair and very unfair offers (**Figure 3** on the bottom right), confirming and extending previous results on this topic (Sanfey et al., 2003; Grecucci et al., 2013a).

#### **DYNAMIC CAUSAL MODELING**

Following the contrast results presented above and based on the previous literature on this topic, we assume that when subjects reappraise their emotions, some regions in the brain are responsible for the implementation of the reappraisal strategy that is they act as "Regulating regions" and some other regions responsible for the emotional appraisal becomes regulated, in other words they can be considered as the "Regulated regions." Building on this observation we aimed at discovering which region is modulated by the regulating regions that may subserve the regulation of interpersonal emotions. This was done by testing three different models (DCMs) that keep constant the regulating regions (more active regions in the "effect of strategy" contrast, IFG and TPJ), while varying the regulated region (striatum, insula, posterior cingulate). We assume that the model that shows the stronger connection parameter between the regulating regions and the regulated regions is the model that better explain the regulatory effects observed in this experiment.

To begin with, we used the same GLM design used for all the contrasts in this paper. Inputs were modeled with the same design matrix of the GLM used in the main analyses. There were three regressors for strategy (down, look, up) multiplied by three regressors for level of fairness (fair, mid fair, unfair), with a total of nine regressors. The contrasts that entered the DCM were the effect of strategy and the effect of regulation (see previous paragraphs), for both unfair and mid fair offers that showed a similar result in previous analyses. Then we selected the meaningful

Grecucci et al. Emotion regulation of social situations

**Table 3 | Effect of regulation DOWN** *<* **LOOK** *<* **UP for unfair and mid fair.**


*\*p* <sup>=</sup> *0.05 FWE.*

regions to put in the models to test. We extracted time series from spheric volume of interests (VOI) of 8 mm from these five regions using the coordinates derived from the **Tables 1**, **3**, though adjusted for local maxima. We included the two key regions found in the main effect of strategy (down + up *>* look contrast), namely the IFG (−54, 8, 22) and the TPJ (−54, −46, 28), that reasonably are the structures implementing the reappraisal process and act as modulators. Whereas, from the regulation contrast, the striatum (−21, 14, −17), the posterior cingulate (27, −46, 37) and the insula (−39, 5, 1) were found to be the regions regulated (down *<* look *<* up contrast). Previous exploratory analyses reported similar results for separate IFG and TPJ so we assume they are acting in a similar or in concert and thus, we computed three separated DCMs as follow: (1) Regulating regions: IFG + TPJ, Regulated: Striatum, (2) Regulating regions: IFG + TPJ, Regulated: insula, (3) Regulating regions: IFG + TPJ, Regulated: posterior cingulate, in order to test the idea of which region is modulated by IFG and TPJ. Moreover, we tried different combinations of connections (feedforward and backforward), tested for both up and down regulation conditions. However, results derived from different types of connectivity and for both regulations, led to similar results. For the matter of simplicity, we reported only results derived from the up-regulation for unfair and mid fair offers, and feedforward connections (hierarchically

**Table 4A | Interaction effects (conjunction of all offers).**


**Table 4B | Interaction effects (disjunction between unfair + mid vs. fair offers).**


organized from IFG and TPJ to each of the three target regions) analyses. The three models were estimated with a Bayesian model comparison. Results reported in **Figure 5**, show clearly a preference for model 2 (Regulating regions: IFG + TPJ, Regulated region: insula).

## **DISCUSSION**

In the present study we show the neural correlates of IER, that is, regulatory strategies applied to socially evoked emotions. As detailed below, this study extends previous studies on this topic, exploring for the first time whether cognitive regulation strategies modulate brain responses of social emotions (e.g., affective response to being treated well or poorly by another). Previous findings on the neural substrates of cognitive reappraisal were

and in circles the "Regulated regions." Three models were tested. Results showed that a model considering the IFG and TPJ acting as modulators and the insula as the regulated regions is the one that better explains the data.

replicated, while also extended to uncover brain structures more generally involved in both mentalizing and interpreting other's intentions in a more or less negative way.

#### **BRAIN CORRELATES OF INTERPERSONAL EMOTION REGULATION**

During the acquisition of reappraisal strategies subjects were capable of successfully modulating their perception of the valence and arousal levels of training stimuli. Unpleasant pictures taken from the IAPS database were rated as more arousing and more negative in the up-regulation condition as compared to baseline, and conversely less arousing and less negative in the downregulation, again compared with the baseline condition. Further assessment of strategy application revealed significant modulation of the ability to down- and up-regulate on command. This allowed us to address four primary questions here. Firstly, we sought to confirm previous studies on ER that have outlined a role for inferior frontal gyrus in implementing reappraisal strategies (see Wager et al., 2009; Ochsner et al., 2012, for reviews). We confirmed this point, showing clear activation of the IFG when asking which brain regions were generally responsible for reappraising the intentions of others. This finding further extends the role of this region in reappraising, by demonstrating its involvement in interpreting another emotional state, this time anger when treated unfairly in a social interactive context. This region has also been observed in a previous study about socioeconomic decision-making using a different task (see Grecucci et al., 2013a).

Using for the first time a social interactive task independent of a decision-making situation allows for exploration in more detail of brain regions associated with different kinds of behavior. This manipulation showed strong involvement of social and mentalizing related regions. The temporo-parietal areas, as well as the medial prefrontal cortex including the paracingulate cortex, have been implicated in mentalizing (Frith et al., 1991; Frith and Frith, 2003) and intention-detection, and may be particularly important here when considering that reappraisal strategies specifically lead participants to reinterpret the intentions of their opponents, as assessed by self-report measurements taken after scanning. Making sense of social interactions requires inferring intentions, beliefs, and desires, that is attributing mental states (i.e., mentalizing; see Frith et al., 1991). This was exactly what players were doing when applying the reappraisal strategies, and other recent studies have pointed out that mentalizing abilities are involved when making socially valued decisions (Evans et al., 2011). In sum, this study can extend actual model of ER such as the MCCE of Ochsner et al. (2012), suggesting that TPJ should be included in the list of regions acting as modulators, in addition to the previously cited dlPFC, ACC, vlPFC, and dmPFC.

Another goal of the present experiment was to study brain responses when facing different kinds of social behaviors from another, from a fair interaction based on equity to increasingly unfair scenarios based on inequity and selfishness. Insula was found to be the key region in differentiating the selfishness of another's social behavior.

#### **MENTALIZING INTERPERSONAL EMOTIONS**

Another finding of this paper was the detection of areas potentially responsible for appraising and reappraising social emotions. The regions implicated here were the striatum, the posterior cingulate cortex and the insula. Interestingly, the striatum has been involved not only in primary or secondary rewards, but also to more abstract, social rewards (van den Bos et al., 2013). One hypothesis is that when subjects engage in social interactions such as the one induced by the DG, the associated social reward value is changed according to the success of this interaction. Therefore, the regulation strategies may affect this region's response in such a way as to adjust the social value when treated unfairly, depending on the reappraisal strategy used. Importantly, when mentalizing in a negative way, activity in the striatum is increased. This mechanism may serve to evaluate and "label" the unfair partner and adjust future interactions with the same partner. Indeed, it was recently proposed that striatum plays a role in reputation formation, another aspect of regulating our reactions when interacting with others (Engelmann and Hein, 2013).

Another region, modulated by the strategy was the posterior cingulate. This is in accordance with previous findings on perceiving negative emotions, especially anger (Murphy et al., 2003), and on regulating emotions induced by simple visual stimuli (Ochsner et al., 2004a,b; Goldin et al., 2008), thus extending the role of these areas into regulating more complex socio-economic emotions.

Last but not least, the insula has been previously reported in the context of the UG, and shown to be involved in responses to unfair offers in particular (Sanfey et al., 2003), and also when modulating the associated decision to reject them (Grecucci et al., 2013a). Consistent with previous studies (Pillutla and Murnighan, 1996; Xiao and Houser, 2005), post-scanning debriefing indicated that anger was the primary emotion elicited by a selfish interactions. Interestingly, neural evidence of the involvement of the insula in the emotion of anger has recently been shown (Denson and Nandy, 2009). One difference with the previous study mentioned above is that in Grecucci et al. (2013a) two regions of the insula where found to be active, one more anterior and one more posterior. In the present study only the anterior insula was modulated by the strategies. Activation of bilateral anterior insula to unfair behavior when interacting with a partner is particularly interesting in light of this region's association with negative emotional states (Sanfey et al., 2003). This region has also been implicated in studies of emotion, in particular involvement in the evaluation and representation of specific negative emotional states (Calder et al., 2001). With respect to emotion-processing systems, it has been hypothesized that reappraisal would modulate the processes involved in evaluating a stimulus as affectively significant (Goldin et al., 2008). Reappraisal effectively down-regulates emotion related neural responses that together modulate ongoing emotion experience in emotion-appraisal brain systems, including the amygdala, subgenual ACC, ventromedial PFC, and insula (Ochsner et al., 2004a,b; Ochsner and Gross, 2005; Eippert et al., 2007; Grecucci et al., 2013a). If the activation in the anterior insula is a reflection of the responders' negative emotional response to an unfair offer, we might expect activity in this region to correlate with the degree to which subjects apply the reappraisal strategies, which is indeed what was found. The better subjects are at down-regulating their emotions, the less the insula is active, whereas, the better subjects are at up-regulating their emotions the more this regions is active. The role of this region in reappraising social emotions was also confirmed by further tests using DCM as a way to explore the network implied in effective regulation. These analyses showed that a circuit including IFG and TPJ acting as modulatory structures and the insula as the region modulated, is responsible for the regulation of socially induced emotions. One hypothesis is that the insula represents the mean by which cognitive strategies can modulate the arousal associated with emotions (Grecucci et al., 2013a). Indeed, other regions found to be modulated by the strategies in the GLM analysis (cingulate cortex and striatum) were not found to be modulated by IFG and TPJ when considering DCM. It is typically assumed that the beneficial effects of reappraisal are accomplished via interactions between PFC regions and subcortical networks related to emotional responding (Beauregard et al., 2001; Ochsner et al., 2004a,b; Kalisch et al., 2005; Phan et al., 2005; Urry et al., 2006; Eippert et al., 2007; Kim and Hamann, 2007; van Reekum et al., 2007; Goldin et al., 2008; Wager et al., 2009). In particular, Wager and collaborators showed with pathway-mapping analysis that a circuit including the ventro-lateral prefrontal cortex (close to the IFG of the present study) and target emotional regions (nucleus accumbens and amygdala) are responsible for regulation strategies.

#### **CLINICAL IMPLICATIONS**

The present study has also relevance for understanding some clinical phenomena such as paranoid thinking and interpersonal skills deficits. Psychotherapists as well as psychiatrists, know that the way we interpret the intentions of others can deeply affect emotional reactions (Allen and Fonagy, 2006; Clarkin et al., 2006), and interpersonal behavior (Linehan, 1993). The more we perceive the intentions of others as malevolent, the more negative emotions we feel, and the more we respond to others in a bad way. In the present study we studied what happens when subjects mentalize in a negative way the intentions of others (up regulation condition). We found that this thinking strategy (implemented in the IFG and TPJ) increases activity in brain structures responsible for emotional reactions (such as the insula and the striatum), and areas associated with the perception of others' mind (middle and superior temporal gyrus?). Notably, when mentalizing in a negative way, insula's activity correlates with the level of anger when treated selfishly. Overall, these data confirm clinical previous observations stating that interpreting others' intentions in a negative way, increases inappropriate interpersonal emotional reactions by affecting the perception of others.

#### **LIMITATIONS AND FUTURE DIRECTIONS**

Lastly, we acknowledge some of the limitations that characterize the present study. First, the lack of internal emotional rating during the scanner limits the connection of the neural results with the corresponding subjective level. However, it should be pointed out that there is supporting evidence that the manipulation was affecting the subjective-behavioral level, as the same

# **REFERENCES**


subjects also played the Ultimatum Game in which we showed strong behavioral modulation of subjects' decisions when applying the strategies (see Grecucci et al., 2013a). Future studies will have to assess at a more behavioral-subjective level the effect of reappraisal strategies in regulating social emotions (Grecucci et al., 2013b). Moreover, in the present study we did not include a measure to assess the quality of interpersonal transaction, though a previous study used the percentage of rejection rates of the partners' proposals (Grecucci et al., 2013a). Future studies may include subjective or behavioral indexes in order to have a quantitative measure of this. Last but not least, the DCM results should be considered as exploratory and more complex models may be addressed in future research.

## **CONCLUSION**

Previous studies have reported the effect of ER strategies in the self, however, the effect of regulation on socially driven emotions was still unclear. Here we show for the first time that IER strategies can strongly affect neural responses when experiencing socially driven emotions, thus extending actual models of ER.

(Chichester; John Wiley and Sons Ltd), 53–100.


implications for affect, relationships, and well-being. *J. Pers. Soc. Psychol.* 85, 348–362. doi: 10.1037/ 0022-3514.85.2.348


*Affective Ratings* (NIMH Center for the Study of Emotion and Attention). Gainesville, FL: University of Florida.


meta-analysis. *Hum. Brain Mapp.* 30, 829–858. doi: 10.1002/hbm. 20547


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 08 April 2013; accepted: 13 August 2013; published online: 03 September 2013.*

*Citation: Grecucci A, Giorgetta C, Bonini N and Sanfey AG (2013) Reappraising social emotions: the role of inferior frontal gyrus, temporo-parietal junction and insula in interpersonal emotion regulation. Front. Hum. Neurosci. 7:523. doi: 10.3389/fnhum.2013.00523*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Grecucci, Giorgetta, Bonini and Sanfey. This is an openaccess article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Does interoceptive awareness affect the ability to regulate unfair treatment by others?

# *Mascha van 't Wout 1,2 \*, Sara Faught <sup>2</sup> and David Menino2*

<sup>1</sup> Department of Psychiatry and Human Behavior, Alpert Medical School Brown University, Butler Hospital, Providence, RI, USA <sup>2</sup> Department of Cognitive, Linguistic and Psychological Sciences, Brown University, Providence, RI, USA

#### *Edited by:*

Susanne Leiberg, University of Zurich, Switzerland

#### *Reviewed by:*

Barney Dunn, University of Exeter, UK Peter Sokol-Hessner, New York University, USA

#### *\*Correspondence:*

Mascha van 't Wout, Department of Psychiatry and Human Behavior, Alpert Medical School Brown University, Butler Hospital, 345 Blackstone Boulevard, Providence, RI 02906 USA e-mail: Mascha\_vant\_Wout@ brown.edu

In this study we aimed to investigate how awareness of bodily responses, referred to as interoceptive awareness, influences decision-making in a social interactive context. Interoceptive awareness is thought to be crucial for adequate regulation of one's emotions. However, there is a dearth of studies that examine the association between interoceptive awareness and the ability to regulate emotions during interpersonal decision-making. Here, we quantified interoceptive awareness with a heartbeat detection task in which we measured the difference between subjective self-reports and an objective psychophysiological measurement of participant heart rates. Social decision-making was quantified using a tworound Ultimatum Game. Participants were asked to first reject or accept an unfair division of money proposed by a partner. In turn, participants could then make an offer on how to divide an amount of money with the same partner. Participants performed 20 rounds of the two-round Ultimatum Game twice, once during baseline condition and once while asked to reappraise emotional reactions when confronted with unfair offers from partners. Results showed that after reappraisal participants (1) accepted more unfair offers and (2) offered higher return divisions, as compared to baseline. With respect to interoceptive awareness, participants with better heartbeat detection scores tended to report less emotional involvement when they applied reappraisal while playing the Ultimatum Game. However, there was no reliably significant relationship between heartbeat detection and the acceptance of unfair offers. Similarly, heartbeat detection accuracy was not related to return offers made in the second round of the Ultimatum Game or the habitual use of emotion regulation. These preliminary findings suggest that the relationship between interoceptive awareness and behavioral changes due to emotion regulation in a social decision-making context appears to be complex.

**Keywords: interoceptive awareness, decision-making, social, unfairness, regulation, emotion, Ultimatum Game, reappraisal**

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 1 — #1

# **INTRODUCTION**

Recently there has been increasing attention towards the role of affective responses when people make strategic decisions in interpersonal contexts. Decision-making in a social interactive context has been particularly well-studied in a well-known game known as the Ultimatum Game (Guth et al., 1982). In the Ultimatum Game two people are asked to divide a certain amount of money. The first player makes a proposal of how to split the money in any way she likes. The second player then has to make a choice. She can accept the division of money in which case the money is split as proposed by the first player. The alternative is that she rejects the division in which case neither player receives any money. In this scenario a "rational" second player who solely cares about the money will accept any offer (as something is more than nothing), and the first player, realizing this, will offer as little as possible. However, in actuality second players typically reject 50% of unfair offers that are 20% or less of the total money amount to be divided (Camerer, 2003).

It has been proposed that this rejection of unfair offers reflects the importance that people place on fairness and punishment associated with being treated unfairly (Fehr and Gachter, 2002). For instance, the (negative) emotional reactions to unfair offers might be a robust reason why people reject these offers (Pillutla and Murnighan, 1996). A neuroimaging study in which people were playing in the role of second player while being scanned showed that activation of the insula was predictive of subsequent rejection of unfair offers (Sanfey et al., 2003). Activation of the insula has been associated with feelings of disgust (Phillips et al., 1997) and (negative) arousal in general (Kuhnen and Knutson, 2005; Nitschke et al., 2006; Nielen et al., 2009; Caria et al., 2010). This lead to the suggestion that insula activation in response to the to-be-rejected unfair offers reflects negative emotional feeling states associated with unfair treatment (Sanfey et al.,2003). Studies that use psychophysiological methods, such as skin conductance responses or heart beat variability, to directly quantify (emotional) arousal have replicated the relationship of higher (emotional) arousal and a tendency to subsequently reject unfair offers (van 't Wout et al., 2006; Osumi and Ohira, 2009). These findings are consistent with the idea behind the somatic marker theory, which proposes that arousal-based bodily signals can guide decisionmaking (Damasio, 1994; Damasio et al., 1996). In his early work, James (1884) already highlighted the importance of awareness of bodily changes in response to stimuli for the generation of an emotional experience.

Interestingly there is variability between people in how likely they are to reject an unfair offer, ranging from those who reject every offer that is not an equal split, to those who never reject any non-zero offer. The decision to accept has been associated with the implementation of cognitive strategies frequently aimed to reduce negative emotional arousal, i.e., emotion regulation (van 't Wout et al., 2010). More specifically, emotion regulation refers to a diverse set of cognitive processes by which "individuals influence which emotions they have, when they have them, and how they experience and express these emotions" (c.f. Gross, 1998). In the study by van 't Wout et al. (2010) participants accepted more unfair offers when asked to reappraise their emotions in response to unfair offers that were 20% or less of the total sum as compared to when they were not reappraising or using suppression as a regulatory strategy. Given that we often interact multiple times with the same person, we had adapted the Ultimatum Game to allow examining whether after reappraisal people were also less likely to retaliate, i.e., to propose a similar unfair offer in return. Our data showed that, after reappraisal, people proposed a fairer split when they were able to divide a sum of money with a partner even after this same partner had treated them unfairly previously. Yet we also noted that there were individual differences in how successful people were at reappraising their emotions.

An important prerequisite for successful emotion regulation is interoceptive awareness. Interoceptive awareness is the awareness of bodily signals and has been highlighted as important in many early theories of emotion (James, 1884; Schachter and Singer, 1962). Füstös et al. (2012) report that interoceptive awareness facilitated the use of reappraisal as an emotional regulation strategy to decrease subjective negative affect and electrophysiological responses associated with emotion regulation (P3 and slow wave). Other studies have validated the presence of an association between interoceptive awareness and emotion arousal (Pollatos et al., 2005), emotion processing, and activation of the insula (Craig, 2002, 2003, 2004; Critchley et al., 2004; Pollatos et al., 2007a), the same region that was predictive of rejecting unfair offers in the Ultimatum Game. Interestingly, Kirk et al. (2011) showed that experienced Buddhist meditators accept the most unfair offers (i.e., 5 and 10% of total sum) more often than control participants. Compared to controls, meditators displayed a different neural activation pattern associated with interoception, including the (posterior) insula. Whether interoception is related to Ultimatum Game behavior was more directly examined by Dunn et al. (2012). In their study, Dunn et al. (2012) demonstrated that as interoceptive abilities increase, people reported more anger in response to unfair offers andfound these offers more unfair. Moreover, those with better interoceptive ability showed a larger difference in psychophysiological arousal, i.e., skin conductance, to rejected relative to accepted offers. This difference in arousal further predicted higher rejection rates in people with better interoception, but this relationship was absent for people with poorer interoception. These data were interpreted as being consistent with emotion regulation explanations for rejection decisions in the Ultimatum Game. However, emotion regulation was not explicitly measured in the study by Dunn et al. (2012). Examining whether people with better interoceptive ability are better at applying emotion regulation when confronted with unfair offers in the Ultimatum Game might provide more insight into the relationship between emotion regulation, interoception, and reactions to unfair treatment. Moreover, there is no investigation on whether interoceptive ability influences Ultimatum Game behavior when interacting with the same person for a second time (who may have been unfair the first time).

In this study, we directly wanted to test whether there is a relationship between interoceptive ability and the ability to apply emotion regulation, i.e., reappraisal, when treated unfairly by others in the Ultimatum Game. In addition, we were interested in testing whether there is a relationship between interoceptive awareness and emotion regulation ability when proposing offers to others who previously had treated them unfairly in the Ultimatum Game. In the experiment, we opted for the use of reappraisal as a regulatory strategy. During reappraisal, people actively try to rework the meaning of emotion-inducing situations, and it has been shown to be effective in lowering emotional experience and reducing the associated psychophysiological processes, such as heart rate, skin conductance responses, and neural activity (Gross, 2002; Ochsner et al., 2002; Gross and John, 2003; Goldin et al., 2008). Moreover, in our previous study on regulation during the Ultimatum Game, reappraisal seemed to be most effective in influencing decision-making (van 't Wout et al., 2010). We predicted that people who are better at (interoceptively) accessing their bodily signals would accept more unfair offers proposed by others and would be less emotionally involved during regulation as compared to baseline. This was based on the above mentioned research showing (1) the importance of interoceptive awareness for successful emotion regulation (Füstös et al., 2012), and (2) that those who typically are better regulators, i.e., meditators, accept more unfair offers and show neural patterns indicative of interoception (Kirk et al., 2011). Our hypotheses with respect to an association between interoceptive awareness and proposal behavior in the Ultimatum Game while applying emotion regulation as compared to baseline were exploratory. A potential positive correlation between interoceptive awareness and proposed offers in the second round after regulating (as compared to baseline) suggests that people with better interoceptive awareness are better at limiting the influence of negative feelings from the first encounter on behavior during a second interaction. We measured interoceptive awareness using a heartbeat detection task in which we computed the difference between subjective self-report and an objective psychophysiological measurement of one's heart rate (Schandry, 1981). Second, we tested the exploratory hypothesis of a positive relationship between the self-report habitual use of reappraisal and interoceptive awareness.

#### **MATERIALS AND METHODS PARTICIPANTS**

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 2 — #2

Thirty three healthy people aged 18–46, mean age 25.36 years (SD 6.85), 23 females were recruited from the general and Brown University community and participated in the study. The Mini International Neuropsychiatric Interview (MINI, Sheehan et al., 1998) was used to confirm the absence of current psychological illnesses or the use of any psychotropic medication. In addition to the MINI, we administered the Beck Depression Inventory (Beck et al., 1988) and the Beck Anxiety Inventory (Fydrich et al., 1992) to measure self-reported levels of depression and anxiety. The Emotion Regulation Questionnaire (ERQ: Gross and John, 2003) was also administered to examine self-reported levels of the habitual use of suppression and reappraisal as emotion regulatory strategies.

Out of these 33 participants, one participant demonstrated high scores on the BDI and BAI indicative of moderate depression and severe anxiety. In addition, two participants provided dubious answers on questionnaires (i.e., answered every question on the questionnaire with the same response which led to serious doubt about task performance). Due to software malfunctioning, we lack data on heartbeat detection for two participants. This resulted in a group of 30 participants for Ultimatum Game data analyses and a group of 28 participants for analyses regarding heartbeat detection.

The order of task administration was fixed and started with the MINI, after which participants played the Ultimatum Game, performed the interoception task, and completed the questionnaires. The study was conducted in a quiet room at the Cognitive, Linguistic, and Psychological Sciences Department, Brown University. Except for the MINI, all tasks were administered on a computer. Participants were compensated for their time and earned some additional money based on their performance on the Ultimatum Game (see below for details). The local ethics committee approved the study and all participants provided written informed consent after the procedures had been fully explained, in accordance with the Declaration of Helsinki.

#### **ULTIMATUM GAME**

Participants completed a total of forty trials of the two-round Ultimatum Game (van 't Wout et al., 2010). On each trial, participants were first shown a picture of their partner with whom they would be interacting for that round. Pictures of partners were obtained from a previously used database of undergraduate students from a different US university (age range 18–30 years, half of these pictures portrayed a female face; van 't Wout et al., 2010). Although we do not have exact demographics of each face (due to IRB regulations), the faces should closely match the demographics of the undergraduate sample recruited for this study.

Participants first interacted in the role of responder, i.e., they received an offer on how a partner wanted to split \$10 with them and they could accept or reject that offer. If the participant accepted the offer, the money was split as proposed and allocated accordingly to each player. If the participant rejected the offer, neither player received any money. Monetary outcomes after the participant's decision were shown for both the participant as well as their partner.

Immediately after the completion of this interaction, participants interacted again with this same partner, but this time the participant was the proposer and thus in the position to make an offer on how to split \$10 with the same partner. Similar to the first interaction, monetary outcomes to both players were shown immediately after the partner decided to reject or accept the offer proposed by the participant. The partner's response to the participant's offer was predetermined and based on close to typical rejection rates of unfair offers. This means that all \$0 were rejected; \$1 and \$2 offers were rejected 60% of the time; \$3 and \$4 offers were rejected 20% of the time; offers of \$5 and higher were always accepted. See **Figure 1** for a graphical representation of the two-round Ultimatum Game.

Participants were told that the offers they would receive as responders had been collected previously. In reality the range of offers being presented to participants was: \$1, \$2, \$3, \$4, or \$5 out of \$10 and was predetermined so that each offer occurred eight times. To further encourage participants to be more cognizant of their decisions, they were instructed that they would play for real money and that a percentage of the total earnings in the game would be paid out to them.

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 3 — #3

Across the entire game, participants made an additional \$5. Study personnel confirmed before the onset of the Ultimatum Game that none of the participants had prior experience with the game.

The 40 two-round Ultimatum Game trials were divided equally across two blocks of 20 identical trials each. During one twenty trial block, participants were asked to apply reappraisal when they received the offer of their partner, whereas during the other block they could play normally (i.e., baseline). The order of reappraisal or baseline was counterbalanced across participants. Out of 30 participants, 14 performed the baseline first-reappraisal second order and 16 participants completed the reappraisal first-baseline second order. Participants were given instructions before beginning any of the trials on how to reappraise. All participants practiced reappraisal on two mildly negative pictures from the International Affective Picture System (Lang et al., 1999) and performed two practice rounds of the Ultimatum Game. Key instructions for reappraisal can be summarized as follows: "It is very important to us that you try your best to adopt a neutral attitude as you watch the offers. To do this, we would like for you to view the offers with detached interest or try to come up with possible reasons for why someone might give you a certain offer" (see also van 't Wout et al., 2010).

After completion of all Ultimatum Game trials, participants were asked to fill out a debriefing questionnaire. Three questions about their emotional involvement were asked: (1) how emotionally involved they were while playing the Ultimatum Game regardless of the offers, (2) how emotionally involved they were when confronted with unfair offers during the trials in which they were asked to regulate, and (3) how emotionally involved they were when confronted with unfair offers during baseline. Answers were given on a −2 (not at all) to +2 (very much) rating scale. Additionally, participants reported how likely they thought it was that they played with a real person on a −2 (not at all) to +2 (very much) rating scale. Ratings on emotional involvement were completed after completion of both versions of the Ultimatum Game (reappraisal and baseline) in order to reduce potential impact of these questions on participant's reactions and performance.

#### **INTEROCEPTIVE AWARENESS TASK**

Interoceptive awareness was measured by having people estimate their own heart rate, which we compared to their actual heart rate. Participants' heart rate was monitored with a pulse oximeter (PulseOximeterOnline.com) to obtain their average heart rate. At the same time that their heart rate was measured, participants were instructed to press a key on the computer keyboard every time they thought their heart beated. The task ended after 60 key presses on the keyboard. Accuracy of heart beat detection was calculated using the formula: 1 − (|recorded heart beats − counted heart beats|)/recorded heart beats (Pollatos et al., 2007a). This measure allows a range of scores between 0 and 1, with higher scores indicative of better heart beat detection.

#### **STATISTICAL ANALYSES**

The effect of emotion regulation and no regulation (i.e., baseline) on Ultimatum Game responder behavior, that is rejections of offers (a binary variable), was analyzed with a generalized

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 4 — #4

estimating equation (GEE) model. The main reason for the implementation of a GEE model was that it allows adjusting for correlations due to repeated (binary) observations within each participant over the different offers. The Decision to reject (or accept) was entered as the binary dependent variable. The variables Offer (four level: \$4, \$3, \$2, \$1), Condition (two levels: reappraisal, baseline), Order (two levels: baseline first, reappraisal first) and their two-way and three-way interactions were added as predictors (factors). The variable Subject was entered as a repeated effects variable.

For the analysis of offer amount proposed in return (second Ultimatum Game round), we performed a linear mixed model to examine the effect of regulation and no regulation on return offers proposed (a continuous variable) while again taking into account the repeated and correlated nature of observations within participants. The proposed Offer amount in return was the dependent variable. The following variables were included as fixed effects: Condition (two levels: baseline, reappraisal), Initial offer in first Ultimatum Game round (five levels: \$5, \$4, \$3, \$2, \$1), Decision of initial offer (two levels: accepted or rejected), and Order (two levels: baseline first, reappraisal first) were added as predictors (factors). Additionally, we included the two- and three-way interactions analogous to the data analyses on proposer behavior. The variable Subject was entered as a correlated random effects variable.

Data on emotional involvement (debriefing) was tested using (paired sample) *t*-tests. The relationship between heartbeat detection performance and Ultimatum Game behavior (rejection rates and return offers) while applying regulation, no regulation or the difference between regulation and baseline was examined using multiple regression analyses. The reason for using multiple regression was that we observed a single data point on heart beat detection accuracy, which was entered as the dependent variable in all regression analyses. Additionally, the use of multiple regression instead of bivariate correlations reduces the number of tests performed and thus the likelihood for type I error. In the regression models for Ultimatum Game responder behavior (rejection rates), we performed three separate regression analyses. First, we examined whether there was an association between heart beat detection accuracy (dependent variable) and rejection rates of unequal offers (four independent variables: rejection rate for \$4, \$3, \$2, and \$1 offers) during baseline. Similarly, a regression analysis was performed to test for an association between rejection rates of unequal offers (same four independent variables) during reappraisal and interoceptive awareness. Finally a third regression analyses was performed to test for an association between interoceptive awareness and the calculated difference between rejection rates of the four unequal offers during reappraisal minus baseline (positive scores suggest higher acceptance rates during reappraisal relative to baseline). These three regression analyses were repeated for the analyses of offer amount returned in the second interaction (proposer behavior). In these regression analyses heart beat detection accuracy was again entered as the dependent variable. Return offer amounts after being confronted with a \$4, \$3, \$2, or \$1 offer were entered as four separate independent variables. Important to note here is the potential for multicollinearity in these analyses as some of our listed independent variables are (highly) correlated. In order to assess multicollinearity, we measured the Variance Inflation Factor (VIF). A VIF cut-off of five or greater was interpreted that collinearity was associated with that variable and we subsequently removed this variable from the analyses. Data was analyzed using SPSS v21.

#### **RESULTS**

#### **ULTIMATUM GAME: RESPONDER**

To confirm the effectiveness of reappraisal on acceptance behavior of participants in this version of the Ultimatum Game, we first performed a GEE model to predict the binary variable rejection of the received offer by the participant. We first added the variable Offer consisting of four levels: \$4, \$3, \$2, \$1 to predict rejection rate. We excluded \$5 offers as these equal offers were typically almost always accepted (99%). The second variable we added was Condition with the levels baseline and reappraisal. A third variable included was the Order in which participants played the games, i.e., baseline first-reappraisal second or reappraisal first-baseline second. Finally we included the interactions Offer × Condition, Offer × Order, and Condition × Order as well as the Offer × Condition × Order interaction.

This analysis resulted in a significant main effect for Offer [*F*(3,26) = 48.19, *p* < 0.0001), a significant main effect for Condition [*F*(1,28) = 4.65, *p* = 0.03], a non-significant main effect for Order [*F*(1,28) = 0.01, *p* = 0.91], a non-significant Offer × Condition interaction [*F*(3,26) = 1.32, *p* = 0.72], a non-significant Offer × Order interaction [*F*(3,26) = 1.51, *p* = 0.68], but a significant Order×Condition interaction [*F*(1,28)=12.48, *p* =0.0004]. The three-way interaction Offer × Condition × Order interaction was non-significant [*F*(3,26) = 1.99, *p* = 0.57].

The main effect for Offer was due to acceptance rates declining as offers became more unfair: *M*\$4 = 0.79 (SE = 0.06), *M*\$3 = 0.50 (SE = 0.08); *M*\$2 = 0.35 (SE = 0.07); and *M*\$1 = 0.25 (SE = 0.06). This replicates the pattern of rejection rates documented for responders in the Ultimatum Game (Camerer, 2003; Sanfey et al., 2003; Harlé et al., 2010; van 't Wout et al., 2010). The main effect for Condition showed that participants accepted unfair offers more often after reappraisal (*M* = 0.52, SE = 0.07) as compared to no regulation (baseline: *M* = 0.43, SE = 0.07). The non-significant main effect for Order demonstrated that across the two order groups (baseline first or reappraisal first) there was no difference on acceptance rates, namely *M*baseline first 0.47 (SE = 0.10) and *M*reappraisal first 0.48 (SE = 0.09).

The non-significant Offer × Condition interaction showed that acceptance rates declined as offers became less fair in both the baseline as well as the reappraisal condition, see **Figure 2**. Similarly the non-significant Offer × Order interaction revealed that acceptance rates declined as offers became less fair regardless of whether participants played baseline first or reappraisal first, see **Figure 2**. Finally the Order × Condition interaction was significant due to a larger difference in accepting unfair offers during reappraisal as compared to baseline in participants who first played during baseline and reappraisal second [*M*baseline = 0.36 (SE = 0.08) and *M*reappraisal = 0.58 (SE = 0.11)]. In contrast, those who reappraised first and then performed under baseline showed a smaller difference in acceptance rates between conditions

[*M*reappraisal = 0.45 (SE = 0.09) and *M*baseline = 0.51 (SE = 0.10)], see **Figure 2**. Indeed, the effect of reappraisal on acceptance rates was significant when selecting only those participants who played baseline first and reappraisal second (paired sample *t* = −3.04, df = 13, *p* = 0.01). For those participants who played reappraisal first and baseline second, the effect of reappraisal on acceptance behavior was non-significant (paired sample *t* = 1.00, df = 15, *p* = 0.33).

#### **ULTIMATUM GAME: PROPOSER**

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 5 — #5

To test whether there was an effect of reappraisal on return offers made by participants in the second part of the Ultimatum Game, we performed a linear mixed model to predict return offer proposed by participants. We used a linear mixed model to allow for repeated measurements (i.e., multiple Ultimatum Game trials) per participant. We included the following predictors: Condition (Baseline or Reappraisal) to test whether regulation affects return offers beyond the initial interaction; Initial offer received when acting as responder (\$5, \$4, \$3, \$2, \$1), as we expected that participants would propose lower return offers after being treated more unfairly; Decision of initial offer (accepted or rejected), based on the hypothesis that rejected initial offers would result in higher return offers than accepted initial offers (van 't Wout et al., 2010); Order (baseline first-reappraisal second or reappraisal second-baseline first), to examine whether the effect of playing while applying reappraisal first or second might influence return offers. We further included the analogous interaction terms as those added to the analysis on responder data, namely the twoway interactions Initial offer × Condition, Initial offer × Order and Condition × Order, and the Initial offer × Condition × Order three-way interaction.

This analysis showed a significant main effect for Condition [*F*(1,1151.94) = 5.36, *p* = 0.02] suggesting that participants proposed a higher return offer after they applied reappraisal

[*M*reappraisal = 4.02 (SE = 0.12)] during a previous interaction with the same person as compared to baseline [M *rmbaseline* = 3.83 (SE = 0.12)]. The main effect for Initial offer was also significant [*F*(4,1160.08) = 29.84, *p* < 0.0001] demonstrating that return offers were lower when initial offers were less fair, see **Figure 3**. We further observed a significant main effect of Decision [*F*(1, 1133.73 = 12.54, *p* < 0.0001] suggesting that participants proposed higher return offers after they had rejected (as compared to accepted) their partners' initial offer previously, *Mrejected* = 4.14 (SE = 0.13) and *M*accepted = 3.72 (SE = 0.12). The main effect of Order was not significant [*F*(1,27.99) = 1.96, *p* = 0.17] suggesting that average return offers were comparable across the "baseline first" and "reappraisal first" groups.

The interaction between Initial offer × Condition was nonsignificant [*F*(4,1151.16) = 0.87, *p* = 0.48] suggesting that return offer amount declined as initial offers were less fair in both the baseline as well as the reappraisal condition. The interaction Initial offer × Order was also non-significant [*F*(4,1151.14) = 0.17, *p* = 0.96] demonstrating that return offer amount declined as initial offers were less fair regardless of whether participants played baseline first or reappraisal first. The Condition × Order interaction was significant [*F*(1,1153.38) = 4.22, *p* = 0.04]. Data showed that there was a larger difference in return offer amount during reappraisal as compared to baseline in participants who first played during reappraisal and baseline second [*Mbaseline* = 3.57 (SE = 0.08) and *Mreappraisal* = 3.91 (SE = 0.09)]. In contrast, those who performed under baseline first and reappraised second showed a smaller difference in return offer amount between conditions [*Mreappraisal* = 4.03 (SE = 0.07) and *Mbaseline* = 4.07 (SE = 0.08)]. The three-way interaction Initial offer × Condition × Order was non-significant [*F*(4,115.14) = 0.51, *p* = 0.73].

#### **DEBRIEFING**

Participants reported to be only somewhat emotionally involved while playing the Ultimatum Game (regardless of offer), *M* = 0.17 (SE = 0.24) on a −2 (not all emotionally involved) to +2 (very emotionally involved) scale. Participants reported to be less emotionally involved when confronted with unfair offers during trials in which they were asked to reappraise as compared to their emotional involvement during baseline trials, *M*reappraisal = −0.73, *M*baseline = 0.07, paired sample *t*-test = −2.89, df = 29, *p* = 0.007.

Given that we observed an interaction between Condition and Order on Ultimatum Game acceptance rates, we tested whether playing baseline or reappraisal first affected emotional involvement in the game. There was a trend for participants who played baseline first to be more emotionally involved in the game [*M* = 0.53 (SE = 0.27)] as compared to those who played reappraisal first [*M* = −0.27 (SE = 0.37)], *t*-test = 1.73, df = 28, *p* = 0.09.

With respect to whether participants thought their partners were real, 10% (*N* = 3) of participants thought their partners were not at all real (−2 on rating scale); 27% (*N* = 8) of participants reported that their partners were most likely not real (−1 on rating scale); 20% (*N* = 6) reported that they were not sure about whether their partner was real or not (0 on rating scale); 23% (*N* = 7) thought their partner is most likely real (+1 on rating scale) and 20% (*N* = 6) of participants reported that they thought their partner for sure was real (+2 on rating scale).

#### **INTEROCEPTIVE AWARENESS**

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 6 — #6

Average heart rate recorded was 74.95 beats/min (SD = 12.76). The average number of taps on the keyboard in order to estimate heartbeat by participants was 55.92 taps/min (SD = 19.97). The mean calculated heartbeat detection score was 0.66 (SD = 0.21) with a range between 0.28 and 0.98.

#### **INTEROCEPTIVE AWARENESS AND ITS RELATIONSHIP WITH ULTIMATUM GAME BEHAVIOR AND HABITUAL REGULATION**

As mentioned in the analysis section, we tested the presence of an association between interoceptive awareness and rejection rates of unfair offers (<\$5) during different Ultimatum Game conditions using multiple regression analyses. Results did not support a relationship between interoceptive ability and acceptance rates during baseline (all *p*'s > 0.13). Similarly, multiple regression analysis did not support a relationship between interoceptive ability and acceptance rates during reappraisal (all *p*'s > 0.14). However, VIF analyses demonstrated the presence of multicollinearity (VIF statistic: 5.32) for the predictor "rejection rate of \$2 offers during reappraisal." A regression analysis without this predictor (i.e., remaining three predictors were rejection rates of \$4, \$3, and \$1 offers during reappraisal) resulted in a positive relationship between interoceptive ability and rejection rate of \$1 offers during reappraisal, β = 0.48, *t*(23) = 2.23, *p* = 0.04. To directly test whether there was a relationship between interoceptive ability and difference in acceptance rates due to reappraisal relative to baseline, we calculated a "regulation difference score" by subtracting acceptance rates during baseline from acceptance rates during reappraisal. Positive scores suggest higher acceptance rates during reappraisal relative to baseline. When looking at the specific predictors, we observed a negative relationship between interoceptive ability and regulation difference score for \$4 offers only, β = −0.47, *t*(23) = −2.59, *p* = 0.02. For all other unfair offers *p*s>0.46. The significant association between interoceptive awareness and increased acceptance of \$4 offers during reappraisal compared to baseline is based on 10 participants who actually showed a difference in acceptance behavior due to regulation. Therefore this observed association needs to be interpreted with extreme caution.

We repeated these three regression analyses to test the relationship between interoceptive awareness and return offers during (1) baseline, (2) reappraisal, and (3) reappraisal relative to baseline. Reappraisal relative to baseline was examined using a regulation difference score for return offers in which positive scores suggest higher return offers after reappraisal compared to baseline. In all of these three regression analyses, a significant association between interoceptive awareness and return offers proposed was not observed (all *p*'s > 0.15).

Using linear regression, we tested whether there was a relationship between interoceptive ability and emotional involvement while playing the Ultimatum Game during baseline and reappraisal. This was non-significant for baseline (*p* = 0.75). The relationship between interoceptive awareness and emotional involvement during reappraisal approached significance [β = −0.34, *t*(25) = −1.73, *p* = 0.09]. This suggests that those who had better interoceptive awareness tend to report less emotional involvement in the game when they applied reappraisal.

Finally, we tested whether heartbeat detection accuracy was correlated with the self-reported habitual use of two regulation techniques: reappraisal and suppression, as measured with the ERQ. A linear regression in which the two regulation styles (reappraisal and suppression) were added to predict heartbeat detection accuracy demonstrated that the use of suppression did not significantly predict interoceptive awareness [β = 0.03, *t*(25) = 0.17, *p* = 0.86]. Reappraisal on the other hand seemed to significantly predict interoceptive awareness [β = 0.41, *t*(25) = 2.21, *p* = 0.03]. However these results seem to be explained by an outlier on the ERQ and when removing this data point from the analyses the results are no longer significant (*p*s > 0.28). Other factors such as behavior on the Ultimatum Game, whether it being acceptance rates or return offers, were not significantly related to reappraisal or suppression on the ERQ as tested using a linear regression approach (all *p*'s > 0.46).

### **DISCUSSION**

In this study we aimed to examine whether people who are better at interoceptive awareness were better at regulating unfair treatment by others in a social interactive decision-making context, i.e., the Ultimatum Game. This hypothesis was based on the idea that being aware of one's emotions is essential for the regulation of these emotions. Interoceptive awareness was quantified using a commonlyused heartbeat detection task in which participants were asked to approximate when their heart was beating (Schandry,1981). Regulation was accomplished by providing instructions to participants b how they could reappraise an emotional reaction in response to unfair offers in the Ultimatum Game. Reappraisal success was based on (1) increased acceptance rates of unfair offers during reappraisal as compared to baseline when participants played in the role of responder in the first part of the two-round Ultimatum Game, and (2) higher monetary return offers when interacting in the role of proposer after participants applied reappraisal as compared to baseline in the second part of the two-round Ultimatum Game.

First, it was important to show that we were able to replicate our previous findings of increasing acceptance rates of unfair offers when participants were asked to reappraise an emotional reaction to such offers in this Ultimatum Game compared to no reappraisal (van 't Wout et al., 2010). We were also able to replicate the typical finding of a decline in acceptance rates as offers became more unfair (Camerer, 2003; Sanfey et al., 2003). This is important as acceptance rates may be influenced by the knowledge that people will interact again with the same person, albeit in a different role, in this two-round Ultimatum Game. Acceptance rates appeared to be rather similar to other studies using a standard Ultimatum Game (Harlé et al., 2010), but potentially somewhat lower (Sanfey et al., 2003; Koenigs and Tranel, 2007). In both the baseline and reappraisal condition, acceptance rates decreased as offers became less fair. This pattern was not affected by whether participants played baseline first or reappraisal first. We did however find that participants who played the game while applying reappraisal first (and baseline second) accepted unfair offers to the same degree regardless of whether they applied reappraisal or not (i.e., baseline). Participants in the "baseline-first" group on the other hand did show a significant difference in acceptance rates after they applied reappraisal as compared to no reappraisal. One possible explanation for this finding might be a combination of (1) participants who first played the game while applying reappraisal may have continued doing this to some extent while playing baseline the second time, and (2) experience with the game, i.e., playing the game twice, may result in reduced affective responses to unfair offers and subsequent increased acceptance rates. For instance, we observed a trend for participants who played reappraisal first to be less emotionally involved in the game as compared to those who played baseline first. Such a reduction in emotional involvement when playing the game for the second time might make reappraisal all the more effective for those in the "baseline-first" group, as the to-be-regulated responses might be less intense and which could have facilitate the effect of reappraisal. We did not observe a three-way interaction between order, offer amount and condition.

We further replicated the effect of increased return offers after reappraisal as compared to baseline in a second interaction with the same partner (van 't Wout et al., 2010). Additionally, we replicated the effects of larger return offers after initially proposed offers were rejected as compared to accepted. Finally, we replicated the observation that participants proposed larger return offers to their partners if partners had initially proposed a more fair distribution of the sum. A surprising finding was the significant Condition × Order interaction showing a larger difference in return offer amount during reappraisal as compared to baseline in participants who first played during reappraisal and baseline second. In contrast, those who performed under baseline first and reappraised second showed a smaller difference in return

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 7 — #7

offer amount between conditions. This is opposite from what we demonstrated for responder behavior, i.e., participants who first played baseline and reappraisal second showed a larger difference in acceptance rates during reappraisal as compared to baseline. Furthermore, when looking at the means of the Condition × Order interaction for return offers one notices that the return offers are numerically higher for the "baseline first" group both during baseline as well as reappraisal. It should however be noticed that the main effect of Order on return offer amount was non-significant. Besides this last unexpected interaction, our data on rejection rates and return offers during reappraisal as compared to baseline mostly replicated the effect of emotion regulation on Ultimatum Game behavior (van 't Wout et al., 2010). The regulatory mechanism is most likely due to a reduction in (negative) feelings associated with unfair treatment. This is further supported by our finding of reductions in emotional involvement during reappraisal as compared to baseline.

With respect to interoceptive awareness and regulation, the main goal of this study, we observed a trend for participants with better heartbeat detection accuracy to report less emotional involvement while applying reappraisal during the game. This is in line with our hypothesis as we had predicted that those with better interoceptive awareness would be better at regulating their emotions, which should result in a reduction of subjective (negative) affect (Füstös et al., 2012). We however did not observe an association between interoceptive awareness and differential Ultimatum Game behavior during reappraisal as compared to baseline. Interoceptive awareness was also not associated with baseline Ultimatum Game acceptance rates. After removal of one variable due to multicollinearity in the analysis, we observed a positive association between interoceptive awareness and acceptance rate of the most unfair offer (\$1) during reappraisal. This suggests that people with better interoceptive awareness accept more very unfair offers (\$1) during reappraisal. This result needs to be interpreted with caution due to the presence of multicollinearity in the full regression model.

We did not observe significant relationships between interoceptive awareness and return offers made in the second round of the two-round Ultimatum Game, whether this was during baseline, reappraisal, or the difference between reappraisal and baseline. These data suggest that interoceptive abilities did not predict reappraisal success in order to change their behavior in a social interactive context. After removing an outlier, we also did not observe a significant association between interoceptive awareness and self-reported daily use of reappraisal or suppression. This is further evidence that people with better interoceptive abilities do not necessarily apply regulatory strategies more often in their everyday life.

Previous research demonstrated an association between interoceptive awareness and cognitive functions including decision making in the Ultimatum Game (Dunn et al., 2010) and selfregulation during physical exercise (Pollatos et al., 2007a). More specifically, the relationship between arousal (skin conductance) in response to offers and the rejection of offers was moderated by interoceptive accuracy (Dunn et al., 2012). These findings highlight that the relationship between interoceptive awareness and social interactive decision-making is not a simple one. We did

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 8 — #8

not examine psychophysiological variables, such as skin conductance, when confronted with (unfair) offers during reappraisal and baseline. The addition of such measures would have allowed examination of biological markers of bodily arousal in response to offers in the game and which have been modulated by reappraisal. Based on previous studies, it may actually be changes in these bodily responses due to reappraisal of the Ultimatum Game that could be mediated by interoceptive ability (Dunn et al., 2012).

Our aim was to examine the potential association between interoceptive awareness and emotion regulation abilities in interpersonal decision-making. The explicit instructions provided to participants on how to apply emotion regulation might have obscured the potentially subtle association between interoceptive awareness and emotion regulation capabilities in such a social context. Moreover, feedback provided to participants from the decisions made in the game could have further resulted in difficulties with observing more subtle influences of interoceptive ability on emotion regulation in the Ultimatum Game. It should be noted however that heightened interoceptive sensitivity has also been associated with symptoms of anxiety (Pollatos et al., 2007b; Domschke et al., 2010), which in turn is associated with reduced emotion regulation capacity (Suveg and Zeman, 2004). Thus, the association between interoceptive awareness and emotion regulation might follow a reverse U-shaped function.

As is often the case with null results, there is the potential that our study is underpowered. A lack of power reduces the generalizability of the results, could result in both type I and II errors and should therefore be taken serious. Our sample of 28 participants is on the smaller end of the spectrum and this is an important limitation. Nevertheless, Füstös et al. (2012)report data on 28 participants of a relationship between psychophysiological measures during regulation and interoceptive awareness. In addition, we succeeded in replicating previous findings of the effects of reappraisal on Ultimatum Game responder and proposer behavior using the same task in a different group of participants. This suggests we are not underpowered to detect changes due to regulation on Ultimatum Game behavior. Nonetheless, we might have been underpowered to detect a more subtle association between regulation and interoceptive awareness and the lack of significant findings should be interpreted cautiously.

Other limitations of this study are that a portion of our participants report that they did not take the computerized interactions in the Ultimatum Game as real social interactions, which could have influenced our data. However given that we explicitly mentioned that people would receive additional money based on the decisions made in the game, one would expect that if participants were not engaged by the social nature of the game, they would accept more, if not all, unfair offers. Instead, acceptance rates declined as offers became more unfair, which is in line with data typically observed in the Ultimatum Game (Camerer, 2003). An additional limitation is the implementation of a heart beat detection task to measure interoceptive awareness. It has been acknowledged before that awareness of heartbeat alone might be an incomplete index of interoceptive awareness (Khalsa et al.,2008; Mehling et al., 2012). We also did not measure aspects that could have influenced heartbeat detection such as body mass (Rouse et al., 1988; Cameron, 2001) or screened people on heart rhythm abnormalities. Furthermore, we provided limited practice with heartbeat detection and the application of reappraisal. Heartbeat detection accuracy is rather low, although within the range of reported findings on such a task (Pollatos et al., 2005). We believe that more practice with heart-beat detection, the application of reappraisal, a more extensive quantification of interoceptive awareness including psychophysiology, and an even more realistic social interactive decision-making context may provide different results. Because of the preliminary nature of this study, results should be interpreted with caution, and as with any scientific result, replication is needed. The investigation of an association between the awareness of bodily states and self-regulation in a social context is important for the generation and application of treatment options for psychiatric phenomena including social anxiety and those with regulation difficulties such as aggression and alexithymia.

#### **ACKNOWLEDGMENT**

This research was supported by the Netherlands Organization of Scientific Research (NWO) Rubicon Grant (No. 446.05.003) to Mascha van 't Wout.

#### **REFERENCES**


"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 9 — #9


van 't Wout, M., Kahn, R., Sanfey, A., and Aleman, A. (2006). Affective state and decision-making in the Ultimatum Game. *Exp. Brain Res.* 169, 564–568. doi: 10.1007/s00221-006-0346-5

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 04 November 2013; published online: 29 November 2013.*

*Citation: van 't Wout M, Faught S and Menino D (2013) Does interoceptive awareness affect the ability to regulate unfair treatment by others? Front. Psychol. 4:880. doi: 10.3389/fpsyg.2013.00880*

*This article was submitted to Emotion Science, a section of the journal Frontiers in Psychology.*

*Copyright © 2013 van 't Wout, Faught and Menino. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited andthatthe original publication inthis journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

"fpsyg-04-00880" — 2013/11/27 — 18:04 — page 10 — #10

# Economic and evolutionary hypotheses for cross-population variation in parochialism

#### *Daniel J. Hruschka1 \* and Joseph Henrich2*

*<sup>1</sup> School of Human Evolution and Social Change, Arizona State University, Tempe, AZ, USA*

*<sup>2</sup> Departments of Psychology and Economics, University of British Columbia, Vancouver, BC, Canada*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Jack Van Honk, Utrecht University, Netherlands Jung-Kyoo Choi, Kyungpook National University, South Korea*

#### *\*Correspondence:*

*Daniel J. Hruschka, School of Human Evolution and Social Change, Arizona State University, 900 S Cady Mall, PO Box 872402, Tempe, AZ 85287-2402, USA e-mail: daniel.hruschka@asu.edu*

# **INTRODUCTION**

In the last 200 years, the half million Iban living on Borneo's northwest region have undergone a remarkable transformation. When first encountered by colonizers in the 19th century, Iban lived in communal long-houses of 100–200 people and made a living from farming rice and hunting (Freeman, 1970). According to their festivals and mythology, Iban worked toward a community that was harmonious, rich in rice, flush with children, and endowed with an abundance of spiritual energy (Jensen, 1974; Heppell et al., 2005). A key way of fostering such flourishing communities was the taking of human heads—to cure a member of one's group or to rescue a member's soul from limbo or from spiritual slavery in another region (Klokke, 2004). It is important to note here that indiscriminate killing was not acceptable among the Iban. Tribal groupings were defined in part as those people who did not take each other's heads. Killing a fellow group member was considered a major transgression on the order of incest. It could upset the universal order and could lead to sterility in terms of offspring and rice production as well as the future taking of heads (Freeman, 1970; Jensen, 1974; Sutlive, 1992).

Fast forward to today. After the forceful imposition of colonial and state laws banning head-hunting, the practice is effectively dead, and only a few elderly men still wield the hand tattoo used to mark a successful headhunter (Freeman, 1970; Laukien, 2005). Iban engage in far-flung wage labor opportunities alongside members of other ethnic groups with which they have prior histories of war (Lumenta, 2003). They seek formal education, consume Malaysian mass media, and many have converted to dominant world religions, including Christianity and Islam. Many Iban now also identify as citizens of Malaysia in addition to being Iban (Lumenta, 2003; Postill, 2006). At times, violence reminiscent of earlier times flares up (BBC News, 2001), but after two centuries, most Iban have a very different way of defining insiders

Human populations differ reliably in the degree to which people favor family, friends, and community members over strangers and outsiders. In the last decade, researchers have begun to propose several economic and evolutionary hypotheses for these cross-population differences in parochialism. In this paper, we outline major current theories and review recent attempts to test them. We also discuss the key methodological challenges in assessing these diverse economic and evolutionary theories for cross-population differences in parochialism.

**Keywords: parochialism, in-group favoritism, cross-cultural, market integration, religion, institutions, parasite stress, closeness**

> and outsiders and very different views about appropriate social behavior with other groups.

> The Iban transformation illustrates three points. First, the ways that people behave toward others can depend heavily on how those others are classified—as kin, friends, and community members or outsiders, strangers and foreigners. Second, human populations can vary dramatically in: (1) how they define closeness and distance of a social partner and (2) how these qualities of a partner influence social behavior. Third, these population differences are not fixed or static. Populations can change quite dramatically within several generations, in this case, from hunting the heads of neighboring groups to participating relatively peacefully in a much larger nation-state and world system.

> How people socially and psychologically construct boundaries between insiders and outsiders or plot gradients of social distance and how these models of boundaries and distance shape behavior toward others are critical questions for a number of fields. Current models for the evolution of human social behavior, and of large-scale cooperation specifically, rely on the construction of groups that can contain the fruits of cooperation, exclude outsiders, and compete with other groups (Boyd et al., 2003; Choi and Bowles, 2007). Paradoxically, the same tribal instincts that may have fostered the human capacity for large-scale cooperation today pose problems for building peaceful and just societies at ever larger scales (Bernhard et al., 2006; Richerson and Henrich, 2012). They also underlay many currently recognized problems in today's world, including favoritism, racial and ethnic discrimination, armed ethnic conflict, and genocide (Levine and Campbell, 1972).

> In the past decade, researchers have proposed a number of theories to account for these population differences in parochialism and to explain historical changes like those observed among Iban. However, these diverse approaches are relatively scattered

across the social and behavioral sciences, they encompass a wide range of motivations and behaviors under the broad rubrics of in-group favoritism, ethnocentrism, xenophobia, and parochial altruism, and these different theories rarely come into contact in the same paper or analysis. In this paper, we clarify the diverse ways that scholars have operationalized parochialism, we outline and synthesize current hypotheses for cross-population variation in parochialism, and we discuss key methodological challenges in assessing these diverse economic and evolutionary hypotheses.

# **VARIETIES OF PAROCHIALISM**

Humans do not have a general tendency to help, protect, or harm others. Rather, these behaviors are conditioned by many contextual factors (Bekkers and Wiepking, 2011), including the perceived need of the recipient (Taormina and Messick, 1983; Engel, 2011), the legitimacy of the request for help (Bickman and Kamzan, 1973), the degree to which someone deserves harm or help (Skitka and Tetlock, 1992), genetic relatedness or kinship with a person (Rachlin and Jones, 2008; Alvard, 2009), and whether the individual or group are perceived to pose a threat (Semyonov et al., 2004). The degree to which an actor feels socially close to another individual also reliably guides social behavior, whether social closeness is determined by subjective assessments of a spatial metaphor (e.g., closeness or insideness) or by common membership in a group (Leider et al., 2009; Goeree et al., 2010; Mathew and Boyd, 2011; Branas-Garza et al., 2012). Here, we refer to the broad tendency to rely on cues of social closeness in guiding behavior as *parochialism*, a concept which encompasses a number of related concepts including xenophobia, ethnocentrism, and parochial altruism.

The social and behavioral sciences have a long tradition of studying the proximate mechanisms by which social closeness and group membership influence behavior toward others and how groups emerge in experimental settings (Sherif, 1961; Tajfel et al., 1971; Brewer, 1979; Glaeser et al., 2000; Hewstone et al., 2002; Dovidio et al., 2005; Goette et al., 2006). All of these approaches are united in studying how our decisions to help, protect or harm someone are shaped by perceptions of social closeness. However, these approaches also differ in two key respects: (1) in how social closeness is operationalized, and (2) in what behaviors, preferences and motivations are considered. We review these differences here.

#### **OPERATIONALIZING SOCIAL CLOSENESS**

Social closeness has been operationalized as both an ordinal and categorical concept. As an ordinal concept, researchers have assessed social closeness to a partner or a group in several ways, by asking participants: (1) to rate partners on a Likert scale in terms of "emotional closeness," "we-ness," or spatial overlap (Aron et al., 1992; Myers and Hodges, 2012), (2) to rank partners in terms of relative closeness (Rachlin and Jones, 2008), and (3) to indicate to what degree one sees oneself as a member of a group (Inglehart et al., 2006). A spatial metaphor is used to describe and assess this concept in many, but not necessarily all languages (as in English, Hruschka, 2010).

Operationalized as a categorical concept, social closeness is based on participation in a relationship (e.g., close friend, family) or on membership in a common group. This can be operationalized categorically in terms of the existence of a recognized face-to-face relationship, including different kinds of kinship, friendship, and acquaintanceship (Hruschka, 2010). It can also be operationalized categorically in terms of common membership in a larger group, such as a religion, denomination, nationality, region, city, neighborhood, language, university, ethnicity, or race (Hruschka and Henrich, 2013).

## **BEHAVIORS, PREFERENCES AND MOTIVATIONS**

Parochialism is manifest in a number of behaviors, preferences and motivations, which we classify here as avoidance, trust, favoritism, permission to harm, and ingroup bias.

First, one can accept or avoid individuals of different groups in everyday interaction (henceforth, *avoidance*). One of the first attempts to assess parochialism, the Bogardus social distance scale, used this approach by asking how much a respondent would accept someone from another ethnic or religious group as a close relative by marriage, as a close personal friend, as a neighbor on the same street, as a co-worker, as a fellow citizen, and as a visitor to one's country (Bogardus, 1933; Inglehart et al., 2006). Second, social closeness correlates with how much people report trusting others. This creates different "radii of trust," where people generally report trusting family members more than personally known others and neighbors, who in turn are trusted more than individuals from other regions, ethnicities and countries (Allik and Realo, 2004; Whitt, 2010; Delhey et al., 2011). Third, social closeness can influence how we distribute resources or protect others (*favoritism),* whether in allocating jobs (Van de Vliert, 2011) or money (Fershtman and Gneezy, 2001; Bahry et al., 2005; Habyarimana et al., 2007; Whitt, 2010), violating a rule to help others (Trompenaars and Hampden-Turner, 2000; Hruschka et al., submitted) or acting to protect others (Bernhard et al., 2006). Fourth, social closeness can shape how morally acceptable it is to harm others or how hostile one feels toward others (*permission to harm*) (Sutlive, 1992; Cashdan, 2001; Mathew and Boyd, 2011). Fifth, people tend to rank socially close friends, family and community members as better than others. This *ingroup bias* can be expressed as pride in family or country or relative ratings of competence, intelligence, or other positive qualities (Brown, 1986; Evans and Kelley, 2002). Researchers have measured these different behaviors, motivations and preferences in several ways, as self-reported attitudes (Evans and Kelley, 2002), behavior in hypothetical scenarios (Trompenaars and Hampden-Turner, 2000; Whitt, 2010), behavior with real monetary stakes (Fershtman and Gneezy, 2001; Bahry et al., 2005), and real-world behavior (Gazal-Ayal and Sulitzeanu-Kenan, 2010).

In addition to these specific manifestations of parochialism, researchers have also deployed several general measures derived from factor analyses intended to capture investment in one's local group. Perhaps the best known measure is collectivism, or the tendency to care about the consequences of one's behavior for ingroup members and to be willing to sacrifice personal interests for collective gains (Triandis et al., 1988; Hofstede, 2001). Schwartz's measure of embeddedness also falls into this category and captures restraint of actions or inclinations that might disrupt group solidarity or the traditional order (Schwartz, 2006).

Little research has focused on how these diverse measures of parochialism covary across individuals and populations. In a sample of 186 small-scale societies, between-society variation in hostile attitudes toward other ethnic groups was not correlated with the degree of belonging to one's own ethnic group (Cashdan, 2001). However, a number of measures of avoidance, favoritism, and ingroup bias are highly correlated across countries, and these also correlate with other non-specific measures of collectivism and embeddedness (Hruschka and Henrich, 2013). Interestingly, the tendency to favor socially close others appears to extend across diverse social scales, all the way from family to nation. For example, increased population levels of parochialism at one level (e.g., the immediate family) are moderately to strongly associated with parochialism at other levels (e.g., extended relatives, friends, compatriots) (Hruschka and Henrich, 2013). Measures of parochialism also appear to be associated with a more general syndrome of social and psychological tendencies, including tighter adherence to norms (Gelfand, 2011), greater concerns about obedience and authority (Inglehart et al., 2006), greater religiosity (Fincher and Thornhill, 2012), and more concerns about purity violations (Haidt and Graham, 2007).

Thus, many measures of in-group favoritism appear to correlate, although out-group hostility may constitute an independent dimension (Cashdan, 2001). Parochialism at one social scale (e.g., immediate family) appears to be associated with parochialism at other scales (e.g., extended family, community, and country). And parochialism appears to be one part of a syndrome of other tendencies toward conformity and obedience.

# **CROSS-POPULATION VARIATION IN PAROCHIALISM**

In the last two decades, psychologists and economists have begun to identify key cognitive and neurobiological mechanisms underlying parochialism, including perceptions of threat (Reik et al., 2006) and the role of oxytocin and brain circuits in modulating behavior toward in- and out-group members (De Dreu et al., 2010; Baumgartner et al., 2011; De Dreu, 2012). Researchers have also identified specific kinds of activities which can increase social closeness to others, including focused conversations (Aron et al., 1997), synchronized movement (Vacharkulksemsuk and Fredrickson, 2012), and synchronized multisensory inputs (Paladino et al., 2010). Moreover, it appears that the capacity and propensity to differentiate social groups arises early in development (Kinzler et al., 2007). However, researchers have only recently begun to explore why these psychological capacities for parochialism are recruited differently in different human populations and across different cultural settings (Miller and Bersoff, 1998; Buchan et al., 2009; Gelfand, 2011; Van de Vliert, 2011; Fincher and Thornhill, 2012; Hruschka and Henrich, 2013; Hackman and Hruschka, 2013b).

There are several ways that populations differ in parochialism. First, what counts as a kin tie, a friendship, or an in-group and what counts as appropriate behaviors with different social partners is informed by local cultural categories and norms. For example, most populations in the US do not have a cultural category of blood brother, and so there is no clear set of norms or expectations applied to being in such a relationship (Hruschka, 2010). Second, the social techniques available to organize and maintain in-groups of varying sizes and scales constrain the kinds of in-groups to which people can belong. Mass media and formal schooling makes it much more likely that people can identify with groups as large as those encompassed by modern nation-states. World religions disseminate and enforce common languages, symbols, and rituals which can forge large populations into a single in-group (Atran and Henrich, 2010). These social techniques permit the creation of new in-groups that may have never been possible before. Third, the most salient in-group category can change quickly based on local practices and contexts. Among Enga horticulturalists in Papua New Guinea, rituals aimed at dehumanizing members of another group can swiftly recast allies as enemies (Wiessner, 2006), and among the Nuer of Sudan, changing patterns of competition over resources can realign in-groups and out-groups (Evans-Pritchard, 1940). Finally, and most relevant to this article, given in-groups of similar scales, individuals from different populations differ remarkably in several crucial ways, including how much they trust and avoid outsiders and how much they favor friends, family, and community members (Fukuyama, 1995; Inglehart et al., 2006; Delhey et al., 2011; Hruschka and Henrich, 2013).

From the perspective of neurobiology, cross-population variability provides an opportunity to establish and distinguish those aspects of human brains and psychology that are reliably developing products of pan-human genes from those that depend on particular culturally-constructed niches (e.g., institutions) or ecological conditions. Grounded in culture-gene coevolutionary theory (McElreath et al., 2003; Henrich and Henrich, 2007), there is now substantial cross-population and developmental evidence suggesting that humans come equipped with cognitive abilities and psychological motivations to preferentially attend to, learn from, and interact with co-ethnics—individuals who share their markers for dress, dialect, language and bodily ornamentation. For example, infants and young children from diverse societies readily use dialect and dress to distinguish ingroup members/coethnics (Kinzler et al., 2011, 2012; Mahajan and Wynn, 2012; Corriveau et al., 2013). On the basis of these rather sparse cues, infant and children preferentially learn from these individual (Shutts et al., 2013), seek interaction with them, and punish them for norm-violations (Schmidt et al., 2012). As expected from the theory, such ethnic cues can even trump racial differences in both young children (Kinzler et al., 2009; Corriveau et al., 2013) and sometimes in adults (Kurzban et al., 2001).

However, such reliably developing features of human cognition and motivation have to be understood in the light of two emerging lines of theory and evidence. First, growing up and ontogenetically adapting to very different environments means that different populations of humans have different brains and biologies, even when no genetic differences exist between populations (Henrich et al., 2010b,c). Within neuroscience, both training studies and cross-population research indicates that brains and bodies develop somewhat differently, in different environments, and yield distinct patterns of activation and hormonal responses to identical stimuli (Nisbett and Cohen, 1996; Kitayama et al., 2009; Woollett and Maguire, 2011). Second, mounting evidence indicates that cultural evolutionary processes have systematically shaped the physical and social (institutional) environments that developing humans face. This implies that cultural evolution has shaped our brains ontogenetically and over historical time (Henrich et al., 2012; Richerson and Henrich, 2012). For example, behavioral studies of children from ages 3 to 14 and adults across six diverse societies, ranging from Congo foragers to Westwood Los Angelenos, reveals the emergence of distinct developmental trajectories for social behavior in different places (House et al., 2013). This pattern is broadly consistent with the presence of market institutions in these societies. Several theories suggest that cultural evolution has harnessed and extended aspects of our innate parochialism in forming nations and religions.

These developments suggest that, rather than attempting to make potentially dubious inferences by generalizing from WEIRD undergraduates (Chiao and Cheon, 2010; Henrich et al., 2010c), neuroscientists need to develop collaborations that take advantage of both the existing theories discussed in this paper and then tap the now well-establish psychological diversity in our species.

# **THEORIES OF CROSS-POPULATION VARIATION IN PAROCHIALISM**

Several theories have been proposed to account for crosspopulation differences and historical changes in parochialism. These theories vary along two major axes. First, they vary in the specific mechanisms by which individuals and populations change in response to their environment. Second, they vary in the specific ecological and social conditions which are posited to shape parochialism. We first review proposed mechanisms and then outline the different proposals for relevant environmental conditions, including market integration, religion, and environmental uncertainty.

#### **MECHANISMS**

Parochial behaviors and motivations might change in response to the environment in several ways. These include genetic adaptation, learning over development, immediate facultative responses, and social learning (Schaller and Murray, 2010).

One recent example of a genetic mechanism is Chiao and Blizinsky's proposal that differences in collectivism may result from allelic variation in the serotonin transporter functional polymorphism (5-HTTPLOR). Specifically, collectivist nations had higher frequencies of the short allele which is associated with heightened anxiety, harm avoidance, fear conditioning, and attentional bias to negative information (Chiao and Blizinksy, 2010). Furthermore, their analyses suggested that these genetic differences may reflect adaptations to infectious disease prevalence. However, a re-analysis of these data suggests that their findings can be accounted for by a model of neutral genetic and cultural change with migration (Eisenberg and Hayes, 2011).

At short time scales, individuals may respond relatively immediately to changing environmental conditions. For example, a vast body of experimental work indicates that cuing uncertainty in a number of domains, including mortality, disease, and social exchange, makes people more likely to favor in-group members (Kollack, 1994; Navarrete et al., 2004; Heine et al., 2006; Hohman, 2011). Conversely, priming individuals with terms related to safety and security make them less likely to favor in-group members (Mikulincer and Shaver, 2001). Thus, parochial motivations and behaviors can respond quite rapidly to environmental cues.

At longer time scales that are still shorter than a lifespan, parochial motivations and behaviors may change in response to environmental cues during specific windows of development. For example, Fincher and Thornhill propose that individual's may learn about disease risk from the local environment through recurring immune system activation, which in turn affects social behaviors and motivations (Fincher and Thornhill, 2012). Recent studies of exposure to war, suggest that specific parochial motivations and behaviors are sensitive to violence between ages of 7 and 20, but not before or after that window (Bauer et al., forthcoming). In addition to direct learning through exposure to their environment, individuals may also learn from others about key aspects of the environment, such as local disease risk, threat of mortality, and risk of inter-group conflict (Fincher and Thornhill, 2012).

In addition to learning environmental cues which may shape parochialism, individuals may also learn relevant social norms about who are members of one's in-group and how one should treat insiders and outsiders under different conditions (Henrich et al., 2010a). For example, individuals frequently engaging in market interactions may learn and eventually internalize norms about dealing fairly with relative strangers and anonymous others (Henrich et al., 2010a).

Each of these mechanisms would lead to different expectations about the time scale of response, from months, to decades, to centuries (Schaller and Murray, 2010). Apparent behavioral fit with specific environments may also result from a combination of co-evolutionary feedback loops involving these mechanisms. For example, infectious disease risk, which is proposed by some theories to be a driver of parochialism, is not simply an exogenous element of the environment. Rather it has changed in response to the emergence of public health institutions, which were in turn the outcome of early large-scale collective attempts to improve others' health. Such feedback between environments and behavior can lead to significant co-evolutionary trajectories.

#### **MARKET INTEGRATION**

The market integration hypothesis proposes that market norms emphasizing fair treatment of anonymous others have culturally evolved to sustain mutually beneficial exchanges in contexts demanding frequent interaction with strangers or ephemeral interactants. As, individuals increasingly interact with markets, they adopt and internalize these norms, and markets spread more successfully in places where such norms are already in place (Henrich et al., 2010a). Thus, individuals with greater market exposure will be more likely to have adopted or internalized these norms and thus will treat anonymous others more fairly. This hypothesis has been tested, replicated, and extended in two separate projects covering 24 different societies from Siberia to New Guinea. Overall, more market integrated societies tend to split pots of money more evenly with anonymous others, independent of the threat of punishment, income, wealth, education, community size, sex, and age (Henrich et al., 2005, 2010a). Since such equitable behavior arises even when punishment is not possible, and anonymity is assured, the authors argue it is guided by internalized local norms. More recent studies among 57 communities in Ethiopia which are tied to their land by customary rights suggests that the relationship between market integration and prosocial behavior with anonymous others is not due to selective migration (Rustagi et al., 2010; also see Voors et al., 2012 for findings from Burundi). And, recent experimental work on "giving" by Westerners show that such responses are automatic (Rand et al., 2012) and rely on the brain's reward circuitry (Fehr and Camerer, 2007; Harbaugh et al., 2007), suggesting that they do reflect internalized patterns of behavior.

### **RELIGION**

Many religious traditions emphasize the importance of helping strangers and treating others fairly, and thus enculturation in specific religions may reduce parochialism—either within one's religion or even across religions. One current theory holds that modern world religions, such as Christianity and Islam, were able to spread precisely because they effectively enculturated norms of prosocial behavior which galvanized large-scale cooperation among relatively anonymous strangers (Atran and Henrich, 2010). According to this view, followers of modern world religions, such as Christianity and Islam, will be more likely to have internalized these norms of prosocial behavior and will thus treat anonymous others with greater fairness and generosity. Findings from the cross-society studies described earlier are also consistent with this hypothesis (Henrich et al., 2010a), showing that adherents to modern world religions offer more in bargaining experiments. Similar experiments among Western populations have shown that unconsciously priming Christians, but not atheists, with "God" causing them to be more equitable in bargaining games, cheat less, cooperate more and sometimes punish selfishness to a greater extent (Randolph-Seng and Nielsen, 2007; Shariff and Norenzayan, 2007; Ahmed, 2009; McKay et al., 2011; Laurin et al., 2012).

World religions may also exhibit variation in how strongly they affect parochialism. Experiments meant to measure trust in anonymous transactions show that religious people are trusted more, especially by other religious people. Consistent with this, work from psychology suggests Christians trust each other more because they believe other Christians know God is watching (Gervais et al., 2011). Ritual participation seems to have effects independent of belief in God: participation in rituals increases ingroup favoritism, in the form of cooperation (Sosis and Ruffle, 2003; Ruffle and Sosis, 2006), and is associated with support for out-group aggression (Ginges et al., 2009).

Protestantism may be of particular interest here. Weber, and more recently Fukuyama, have argued that a key effect of Protestantism was to "shatter the fetters" of the extended family (Weber, 1951; Fukuyama, 2011), and recent authors have pinned this on Protestant core values of self-reliance and individualism which potentially led to less investment in family, friends, and local in-groups (Lipset and Lenz, 2000; Treisman, 2000). Consistent with this, cross-national analyses show that majority Protestant countries consistently report less favoritism, in-group bias, and out-group avoidance, after adjusting for economic security and government effectiveness, than countries with other religions in the majority—including Orthodox Christianity, Catholicism, and Islam (Hruschka and Henrich, 2013).

## **GLOBALIZATION**

The globalization hypothesis proposes that as people are increasingly exposed to individuals outside their community through new forms of mass media, including newspapers, the internet, social media, television, and movies, and through new forms of social interaction, they are less likely to think in terms of ingroups and out-groups and more likely to imagine humankind as a "we" where there are no "outsiders" (Buchan et al., 2009). Thus, individuals with greater interactions with global communication (e.g., televisions, print media, and employment in transnational firms) will be more inclined to engage in collective action with individuals outside of their immediate in-group. This hypothesis overlaps with the market integration hypothesis, but proposes that many kinds of interactions, including mere exposure to people from other countries through mass media, can change responses to outsiders. Consistent with this hypothesis, Buchan et al. (2009) found that contribution to global public goods increases with increasing exposure to different forms of mass media.

## **EXISTENTIAL OR MATERIAL SECURITY HYPOTHESES**

Here we group three related hypotheses that focus on the effects of various form of material or existential security on individual decision making, development, and cultural evolution. The first, generalized insecurity, casts a broad net by proposing that insecurity will influence parochialism, while the others suggest that individuals respond selectively to specific kinds of threats, such as pathogens, inter-group conflict, and thermic stress.

#### *Generalized insecurity*

Variants of the institutional quality hypothesis propose that informal and formal institutions change the costs and benefits of parochialism, which in turn shape social norms and behavior by a number of potential mechanisms. Public services, global markets, and social safety nets that mitigate material threats and guarantee safe interaction with anonymous partners may render investments in an expansive network of kith and kin less necessary as alternative forms of social insurance. It may also foster greater interaction and trust with a larger set of individuals (Inglehart and Welzel, 2005; Inglehart et al., 2006; Hruschka, 2010; Hruschka and Henrich, 2013). Ample experimental and observational evidence demonstrates the role of economic, existential, and symbolic security on parochial attitudes and behaviors (Kollack, 1994; Navarrete et al., 2004; Heine et al., 2006; Canetti-Nisim et al., 2008; Proulx and Heine, 2010; Hohman, 2011; Kaplan et al., 2012). Conversely, priming individuals with terms related to safety and security make them less likely to favor in-group members (Mikulincer and Shaver, 2001). And a body of work in political science and economics has examined how norms and institutions reduce barriers to trust, encourage cross-group cooperation and discourage parochialism in ethnically-divided societies (Knight, 1992; Jackman and Miller, 2004; Whitt, 2010). Several lines of observational evidence are also consistent with this hypothesis that stronger institutions and less exposure to generalized risk of famine, disease, and inter-group conflict are associated with reduced in-group favoritism (Cashdan, 2001; Inglehart et al., 2006; Whitt, 2010; Hruschka and Henrich, 2013).

#### *Pathogen stress*

The above hypothesis proposed that parochialism responds to existential or material insecurity, in general. However, there are other, more domain-specific, hypotheses that propose that specific forms of insecurity may have parochial effects. Recently, several evolutionary researchers have proposed that parochialism constitutes a form of behavioral immune system against the spread of pathogens. According to this hypothesis, in regions with high risk of infection by dangerous pathogens, individuals will preferentially interact with in-group members in a way that insulates them from infection by out-group members (Schaller and Murray, 2010; Fincher and Thornhill, 2012). Though originally predicting avoidance of and hostile attitudes toward outgroups, the theory has been extended to account for other aspects of parochialism as well, including ingroup favoritism and bias (Fincher and Thornhill, 2012). This hypothesis differs crucially from other hypotheses by positing that the adaptive mechanisms responsible for this effect are specific to pathogen risk and were designed to impede the spread of pathogens or to provide social support specifically in case of infection. Different mechanisms have been proposed, including sensitivity to immune system activation, social learning of local disease risks and direct observation of parasitic infections, all of which would lead to relatively fast facultative responses. Other longer-term mechanisms include culturally evolutionary processes by which groups which have social norms preventing and mitigating threats of infection (e.g., parochial social interaction) are more likely to spread and persist in regions of high endemic pathogen threat (Schaller and Murray, 2010).

Emerging experimental evidence suggests that people do indeed adjust some social motivations and behaviors (i.e., conformism) to specific cues of pathogen threats over and above generalized threats (Murray and Schaller, 2012). However, crossnational and cross-state studies have shown mixed support for this hypothesis as an explanation for extant cross-population variation in parochialism (Currie and Mace, 2012; Fincher and Thornhill, 2012; Cashdan and Steele, 2013; Hruschka and Henrich, 2013; Hackman and Hruschka, 2013a; Hruschka et al., submitted).

## *Inter-group conflict hypothesis*

Another insecurity hypothesis focuses narrowly on how the threat of, or experience of, intergroup conflict may strengthen in-group preferences, including egalitarianism. Using simple choice tasks in two post-conflict societies, the Republic of Georgia and Sierra Leone, Bauer et al. (forthcoming) show that the experience of war creates an enduring increase in individuals' in-group egalitarian motivations, while not influencing their motivations toward outgroup individuals. However, the effect of war only left an enduring mark on motivation if individual experienced the war during a developmental window from roughly age 7 to 20. The effect of war experience had no impact on those under about age 7, and only small effects on those who experience the war past roughly age 20. These results are supported by other work showing that senior Israeli citizens were more willing to punish norm-violators in a bargaining game during the conflict with Hezbollah, compared to both pre- and post-war measures (Gneezy and Fessler, 2012). Working in Burundi, Voors et al. show that victimization in war increases people altruism toward their neighbors, as well as their temporal discounting and risk preferences. This work also examines the effects of non-war related shocks to security, including draught, flooding, and pestilence. This work shows that the experience of droughts also increased altruism toward ingroup members, an independent effect, but did not alter temporal discounting or risk preferences. This suggests that war-related insecurity vs. drought-related insecurity may produce somewhat different psychological effects (Voors et al., 2012), supporting the notion that these are distinct domains. However, aside from this finding, all of these data are also consistent with the generalized insecurity hypothesis.

# *Thermic stress hypothesis*

The climate-economics hypothesis proposes that much of human culture is an adaptive response to thermic stress—either extreme cold or extreme heat—but that this can be buffered by economic resources. In the case of in-group favoritism, Van der Vliert argues that populations facing extreme temperature stress without the economic resources needed to adapt to that stress respond psychologically in a number of ways, including greater preferences for authoritarian leadership and for favoring members of one's in-group (Van de Vliert, 2011; Van de Vliert and Postmes, 2012).

## **METHODOLOGICAL ISSUES IN ASSESSING CROSS-POPULATION HYPOTHESES**

In the last decade, the observation of substantial betweenpopulation differences in parochialism has inspired considerable theoretical work on the possible causes of these betweenpopulation differences. This is exciting progress, and this review describes a number of promising theories that may account for cross-population variation.

However, there are serious challenges in efforts to discriminate between these different hypotheses and to identify the specific mechanisms by which parochialism rises and falls in societies. Most studies have relied on observational cross-population designs, raising concerns about causality, identification of specific mechanisms, the direction of effects, and the time-scale of adaptation. Several strategies can provide some check against these issues.

The first task is to begin culling hypotheses through strategic model comparison rather than testing each hypothesis against a straw man null model. This involves identifying different predictions across models and then finding appropriate crosspopulation data which can discriminate between these predictions. For example, in a recent study of population-level parochialism across countries, Hruschka and Henrich directly compared the parasite stress hypothesis with the material insecurity hypothesis using novel checks against regional autocorrelation, new longitudinal data to assess reverse causation, and an instrumental variable to check for the effects of omitted variables. These results provided consistent support for the material insecurity hypothesis. It also challenged prior studies supporting the parasite stress hypothesis which had not included these methodological checks. In another paper, Hackman and Hruschka re-assessed analyses of US data which had previously found an association between pathogen stress and collectivism. With new data stratified by race, they showed the observed associations across states were due exclusively to substantial differences across US Whites and US Blacks. They also found support for an alternative hypothesis related to the material insecurity hypothesis (Hruschka and Henrich, 2013; Hackman and Hruschka, 2013a). Of course, such model comparison using observational data does not definitively show that the "winning" hypothesis is correct. However, it helps winnow the playing field.

Another important check can come from combining psychological experiments with cross-population studies in order to triangulate between potential psychological processes and the macro-scale correlates of cross-population variation. The findings of experiments alone may not scale up easily to account for cross-population differences, and cross-population correlations without grounding in established psychological mechanisms can easily be explained away as spurious associations. Integrating these two orders of data can ensure that hypotheses are consistent at both the individual and population level. A number of theories, including the market integration, religion, institutional quality, and pathogen stress hypotheses have begun to accrue data at both of these levels.

To mitigate some concerns about causality, mechanism, and directionality, the social sciences offer a number of tools that provide further checks on findings from cross-population observational data. Instrumental variable analysis commonly used in economics provides one additional check by identifying quasi-experimental assignments in observational data. For example, Acemoglu et al. used the mortality rates of early settlers in European colonies (1600–1875) as an instrumental variable which is expected to affect contemporary government effectiveness—an important variable in the material security hypotheses of parochialism. There is ample historical evidence that European colonizers avoided settling in places with high mortality rates, such as in the Belgian Congo (McNeill, 1977; Acemoglu et al., 2001), and instead of settling, they set up extractive systems. In situations of low mortality, on the other hand, colonizers settled in larger numbers and brought with them institutions, such as respect of private property, checks and balances in government, and equality of opportunity, which in turn fostered greater government effectiveness that persisted even after independence (Acemoglu et al., 2001). These measures of settler mortality act in some ways as quasi-experimental assignments of countries to different levels of government effectiveness, and Acemoglu et al. used this quasi-experimental assignment to examine the effect of government institutions on economic growth. More recently, Hruschka and Henrich have used the same reasoning to examine the effect of government institutions on parochialism (Hruschka and Henrich, 2013).

As access to longitudinal data increases with longer running cross-national surveys, it will also be possible to assess the temporal precedence and coincidence of different changes within populations (Inglehart et al., 2006; Hruschka and Henrich, 2013). For example, between 1925 and 2005, US samples have shown steadily decreasing avoidance of other ethnic groups in a number of domains—as in-laws, friends, neighbors, and fellow citizens (Bogardus, 1933; Parrillo and Donoghue, 2005). Long-term longitudinal data like this may provide insights into what factors most readily account for long-term changes in parochialism and how rapidly changes occur. Migration studies, originally developed in epidemiology, but now applied in economics, also show some promise in identifying the time-scale by which different aspects of parochialism change across generations who are put into novel contexts (Guiso et al., 2006; Fisman and Miguel, 2007; Giuliano and Alesina, 2010). For example, Giuliano and Alesina used such a design to show that second generation immigrants carry "cultural baggage" from their home country. Specifically, even after two generations, immigrants from countries with greater stated investment in family ties moved less and lived with their parents longer (Giuliano and Alesina, 2010).

Another approach is to look for natural experiments, as Bauer et al. did with their investigation of the effects of war on parochialism (Bauer et al., forthcoming). They looked around the globe for situations in which the effects of war on individuals, households, and communities were—at least plausibly—random with respect to individuals' own parochial motivations. Refugees and soldiers would be relatively easier to access compared to the approach they took, but both fleeing and being alive might be caused by their particular social motivations (therefore endogenous). As checks on the natural experiment assumption, they also (1) examined whether observables, like ethnicity or age, predicted experiencing war (they did not) and (2) performed their analyses just on those who were children at the time of the conflict (and thus have less control). These analyses support the idea that the experience of war was imposed exogenously, and thus provides a natural experiment.

Despite all of these possible checks and triangulations, observational data is still plagued by concerns about endogeneity and non-random assignment of cases which can threaten causal interpretations. Thus, once hypothesis are culled and honed through the above-mentioned techniques, a growing body of field experiments in economics, public health, and development currently used to understand health and development holds promise in assessing specific mechanisms by which economic, social, and environmental conditions inhibit or foster parochialism (Banerjee and Duflo, 2011). With this combination of model comparison, cross-level confirmation, statistical checks on temporal precedence and causality, and ultimately field experiments of different hypotheses, this exciting and crowded field of theories for parochialism will hopefully lead to a clearer understanding of the specific mechanisms and time scales by which population differences in parochialism emerge and sustain themselves.

## **ACKNOWLEDGMENTS**

We thank Joe Hackman for helpful comments on the manuscript. Daniel J. Hruschka acknowledges support from the National Science Foundation grant BCS-1150813, jointly funded by the Programs in Cultural Anthropology, Social Psychology Program and Decision, Risk, and Management Sciences, as well as support from the University of Chicago and Templeton Foundation New Science of Virtues Grant. Joseph Henrich acknowledges support from the Canadian Institute For Advanced Research (CIFAR).

## **REFERENCES**


924–973. doi: 10.1177/08997640103 80927


random assignments to real social groups. *Am. Econ. Rev.* 96, 212–216. doi: 10.1257/000282806777211658


*Conflict, Ethnic Attitudes, and Group Behavior*. New York, NY: Wiley.


*Pers. Relatsh.* 19, 663–679. doi: 10.1111/j.1475-6811.2011.01382.x


contingency model of distributive justice. *J. Exp. Soc. Psychol.* 28, 491–522. doi: 10.1016/0022-1031 (92)90043-J


35, 94–95. doi: 10.1017/S0140525X1 1001075


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 08 April 2013; accepted: 22 August 2013; published online: 11 September 2013.*

*Citation: Hruschka DJ and Henrich J (2013) Economic and evolutionary hypotheses for cross-population variation in parochialism. Front. Hum. Neurosci. 7:559. doi: 10.3389/fnhum.2013.00559 This article was submitted to the journal*

*Frontiers in Human Neuroscience. Copyright © 2013 Hruschka and Henrich. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Neurocognitive mechanisms underlying value-based decision-making: from core values to economic value

# *Tobias Brosch\* and David Sander*

*Department of Psychology, Swiss Center for Affective Sciences, University of Geneva, Geneva, Switzerland*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Fabian Grabenhorst, University of Cambridge, UK Patricia Bado, Universidade Federal do Rio de Janeiro, Brazil*

*\*Correspondence: Tobias Brosch, Department of Psychology, University of Geneva, 40, Boulevard du Pont d'Arve, CH-1205 Geneva, Switzerland e-mail: tobias.brosch@unige.ch*

Value plays a central role in practically every aspect of human life that requires a decision: whether we choose between different consumer goods, whether we decide which person we marry or which political candidate gets our vote, we choose the option that has more value to us. Over the last decade, neuroeconomic research has mapped the neural substrates of *economic value*, revealing that activation in brain regions such as ventromedial prefrontal cortex (VMPFC), ventral striatum or posterior cingulate cortex reflects how much an individual values an option and which of several options he/she will choose. However, while great progress has been made exploring the mechanisms underlying concrete decisions, neuroeconomic research has been less concerned with the questions of why people value what they value, and why different people value different things. Social psychologists and sociologists have long been interested in *core values,* motivational constructs that are intrinsically linked to the self-schema and are used to guide actions and decisions across different situations and different time points. Core value may thus be an important determinant of individual differences in economic value computation and decision-making. Based on a review of recent neuroimaging studies investigating the neural representation of core values and their interactions with neural systems representing economic value, we outline a common framework that integrates the core value concept and neuroeconomic research on value-based decision-making.

#### **Keywords: decision making, core values, neuroimaging, value-based decision making, value**

"All sciences are now under the obligation to prepare the ground for the future task of the philosopher, which is to solve the problem of value, to determine the true hierarchy of values"


Value is arguably one of the most central concepts governing human life, as it is involved in practically every aspect that requires a decision: whether we choose between different consumer goods, whether we decide which person we marry or which political candidate gets our vote, whether we ask ourselves if something is beautiful, morally right, or sacred, value plays a crucial role. Value reflects the importance that something holds for us, what doesn't have any value is of no interest. Consistent with the central role of value in our lives, ever since Plato scholars have been trying to understand what value is and where it comes from. Today, the investigation of value is central to many disciplines studying human feeling, thinking and behavior, such as philosophy, psychology, sociology, economics, or neuroscience (Brosch and Sander, forthcoming).

Interestingly, the different disciplines are all focusing on somewhat different aspects and conceptualizations of value. According to the Oxford Dictionary of English, the word value in its broadest sense refers to the "importance, worth, or usefulness of something." This general definition is followed by several subdefinitions, the first of which describes value as "the material or monetary worth of something." This definition reflects how economists and neuroscientists think about value: A "common currency" that people use to compare different types of goods or experiences on the same scale when deciding between several options. *Economic value* is related to the amount of reward that a person expects to obtain from the choice. Over the last decade, neuroeconomic research has substantially increased our knowledge of the neural substrates representing value and the neurocognitive mechanisms underlying decision-making (Schultz, 2006; Rangel et al., 2008; Kable and Glimcher, 2009; Grabenhorst and Rolls, 2011; Padoa-Schioppa, 2011; Rushworth et al., 2011; Lee et al., 2012). While making great progress exploring the mechanisms underlying concrete decisions, neuroeconomists have put less emphasis on the questions of why people value what they value, and why different people value different things. This aspect is addressed in the second subdefinition of value in the Oxford Dictionary of English, "principles or standards of behavior, one's judgment of what is important in life." This definition resonates with how social psychologists and sociologists think about value: A broad motivational construct at the core of the self-image that guides choices and behaviors across situations, often framed as a shared belief about ideal objectives (Rohan, 2000). Value research in social psychology and sociology focuses on the role of universals and individual and cultural differences in *core value* systems, and has shown that people in many different cultures use and recognize the same set of core values, but may differ in terms of their relative value priorities.

Thus, research on economic value has produced many insights into the neurocognitive mechanisms that drive decisions in concrete situations, whereas research on core value allows explaining interindividual differences in decision situations as well as intraindividual consistency across decisions over time. Whereas these different facets of the value concept so far have been investigated more or less in isolation from each other, we feel that an integration of the two perspectives would be extremely useful. In this contribution we review (a) neuroeconomic research delineating the neurocognitive mechanisms underlying economic value computations and (b) social psychological and sociological research concerning the universal structure of core values and the role of individual core value differences in decisions and behaviors. We then propose a common framework that aims at integrating the core value concept into a neuroscience of decisionmaking, and support our idea by a review of recent neuroimaging studies investigating the neural representation of core values and their potential interactions with neural mechanisms underlying value computation and decision-making.

# **ECONOMIC VALUE: A COMMON CURRENCY FOR DECISION-MAKING**

In economic and neuroeconomic theory, value is conceptualized as a measure of the benefit that people can gain from choosing an option. When having to decide between several options, the person will compute the value of each option, and then choose the one with the highest value. The value of an option is derived from a person's behavior, i.e., the observable choices of the individual: If a person chooses option A over option B, it is inferred that option A has higher value. At the computational level, value depends on how much reward a person expects to receive from choosing an option, e.g., from eating a piece of chocolate or from receiving an amount of money. The notion of value as a common currency (Samuelson, 1947) is central to many economic theories of decision-making, as it allows to conceptualize how people can compare and choose between different types of rewarding objects. To illustrate the problem, whereas the decision between different amounts of the same rewarding object is relatively straightforward (e.g., "Would you prefer one piece of chocolate or two pieces?" or "Would you prefer \$5 or \$10?"), choosing between two objects with several different reward-related attributes is more complex (e.g., "Would you prefer a piece of chocolate or a salad?"), as different dimensions (e.g., considerations pertaining to taste and to health, respectively) need to be taken into account and weighed against each other. In these cases, a common value currency allows integrating and combining the different dimensions into one representation that can be used as a basis for individual decisions.

Over the last decade, the brain network representing economic value has been delineated using neuroimaging methods (Padoa-Schioppa and Assad, 2006; Schultz, 2006; Kable and Glimcher, 2009; Grabenhorst and Rolls, 2011; Padoa-Schioppa, 2011; Rushworth et al., 2011; Lee et al., 2012; Levy and Glimcher, 2012) as well as single neuron recordings (in primates, Platt and Glimcher, 1999; Tremblay and Schultz, 1999). In a typical neuroimaging experiment, participants view different stimuli (for example different consumer objects) and are asked to choose one of them (or to indicate how much they like each option). The individual choices (or preferences) are then used to derive a measure of economic value, which is used as a parametric regressor to identify brain regions that show systematic activation changes as a function of the value of the presented objects. A large number of converging studies have identified a network of brain areas representing subjective economic value for many different types of rewarding stimuli, consisting of ventromedial prefrontal cortex/orbitofrontal cortex (VMFPC/OFC), ventral striatum, posterior cingulate cortex, amygdala, insula and posterior parietal cortex (see, e.g., Kawabata and Zeki, 2004; O'Doherty, 2004; Kim et al., 2011; Levy and Glimcher, 2012).

Studies comparing neural activation to different classes of rewarding stimuli in the same subjects (e.g., to food, consumer goods, money, or social reputation gains) have observed overlapping activations in VMPFC/OFC, striatum, and insula, suggesting that these regions indeed represent a common currency for different types of rewarding stimuli that allows comparing and deciding between objects with very different properties (Izuma et al., 2008; Chib et al., 2009; Grabenhorst et al., 2010; Kim et al., 2011; Lin et al., 2012). This neural system representing economic value can implement computations of considerable complexity, such as a cost-benefit analysis (when participants are choosing between options that imply both rewarding and punishing aspects) in interactions of VMPFC/OFC and insula (Talmi et al., 2009), and value discounting during delay of gratification (when participants are choosing between a smaller reward right now and a higher reward later) in VMPFC/OFC and ventral striatum (McClure et al., 2004).

Activation in this network should thus allow to infer preferences and to predict choices: When two different objects elicit neural activation of equal magnitude, the two objects should be equally desirable for a person. In contrast, when activation is increased toward one object compared to another, this object should be preferred. And indeed, measurements of brain activation in regions of this network allow predicting which of two items an individual prefers and choses, at least when the subjective value difference between the two items is fairly large (FitzGerald et al., 2009; Lebreton et al., 2009; Levy et al., 2011).

To sum up, neuroeconomic research has reliably identified a brain network representing economic value that allows predicting individual preferences and choices. However, whereas much progress has been made identifying the neurocognitive mechanisms underlying concrete choices, neuroeconomic research has mostly neglected questions such as why people choose (and thus value) what they choose, or why different people choose (and thus value) different things. At the proximal level, this question has been addressed by looking at the impact of individual reinforcement learning histories (see Lee et al., 2012, for a review) However, more research on the distal motivational principles that can predict decisions across situations is clearly needed. Moreover, neuroeconomic research is largely restricted to relatively simple decisions, such as choices between two consumer goods, and rarely investigates more complex decisions and life choices. Such issues are however addressed by researchers interested in core value, mainly from social psychology and sociology. In the following section, we will summarize some key concepts and findings from this field.

# **CORE VALUE: A STABLE CONCEPT OF WHAT IS DESIRABLE**

Core value refers to stable motivational constructs or beliefs about desirable end states that transcend specific situations and guide the selection or evaluation of behaviors and events (Rohan, 2000). An individual's core values form an internal compass that people refer to when they are asked to explain and justify their preferences, decisions, or behaviors. For example, a person may frequently donate money to charitable causes and explain this behavior by their altruistic core values. Core values are thus instrumental in providing the individual with meaning in the world. They provide an organizational principle for an individual's self-schema (Roccas and Brewer, 2002), forming the core of one's identity (Hitlin, 2003).

Cross-cultural research has shown that certain core values are universal, meaning that people in many different cultures can recognize and use the same core values to describe their personal core value hierarchy (see **Table 1**; Schwartz, 1992).

These 10 core values can be grouped in a circumplex where they form clusters organized along two core value dimensions, which reflect conflicts between opposing classes of human interests (see **Figure 1**). The first dimension is labeled "selfenhancement vs. self-transcendence," and reflects the conflict between outcome maximization for the individual vs. outcome maximization for the social group. Individuals with highly selfinterested core values emphasize power and achievement-related goals and choices, whereas individuals with self-transcending values emphasize universal and benevolent goals and choices. The second dimension is labeled "openness to change vs. conservation," and reflects the conflict between following one's interests in uncertain directions vs. preserving the status quo embedded in existing relationships. Individuals with conservative values

**Table 1 | The 10 universal core values and their conceptual definitions (Schwartz, 1992).**


emphasize conformity, security, and tradition, whereas individuals with open-to-change values emphasize self-directive and stimulating goals and choices (Schwartz, 1992).

Importantly, core values are not only used to give orientation and stability to the self, but allow predicting individual differences in concrete decisions and behaviors. For example, a person emphasizing conservation-related values more frequently observes traditional customs on religious holidays than a person who does not hold these values in high esteem. A person who emphasizes self-transcending values more frequently uses environmentally friendly products than a person who emphasizes self-enhancing values (Bardi and Schwartz, 2003). Core value differences have furthermore been shown to be powerful predictors of voting behavior (Schwartz et al., 2010). Thus, the core value concept is a powerful construct that may explain why different people value different things and why different people choose differently in the same situation, and thus may be fruitfully combined with neuroeconomic research on value computation and decision-making.

However, so far not much research has attempted to investigate the neural mechanisms underlying the role of core value in decision-making. In a first attempt to integrate core value into current neuroimaging research, we aimed at identifying the neural regions involved in the representation of core value (Brosch et al., 2012). To this end, we showed our participants examples of behaviors that reflect different core values (e.g., "correcting injustice," "respecting traditions") and asked them to indicate on a scale from 1 to 4 how important the behavior (and thus the related core value) is for them (*core value condition*). In order to directly compare the neural regions representing core value to the regions representing economic value, these behaviors were intermixed with examples of potentially

rewarding concrete activities (such as "eating an apple," "playing tennis"), for which participants indicated (using the same scale from 1 to 4) how much they like performing this activity (*economic value condition*). The economic value condition activated the expected neuroeconomic value network, including regions such as VMPFC, posterior cingulate cortex, and posterior parietal cortex. In contrast, the core value condition led to increased activation in medial prefrontal cortex (MPFC) and in the dorsal striatum. MPFC has frequently been linked to processes involving self-reflection (Macrae et al., 2004; Northoff and Bermpohl, 2004; Mitchell et al., 2005; Lieberman, 2010), both when explicitly reflecting about one's self and when implicitly processing selfrelated information (Rameson et al., 2010), and has furthermore been shown to be activated when thinking about future goals, which are closely tied to one's core values (D'Argembeau et al., 2009). The observed activation of MPFC is thus consistent with the conceptualization of core value as an integral part of the selfschema (Hitlin, 2003). However, given that so far this is the only neuroimaging study linking core value to MPFC, it would be important to replicate this finding in future studies.

## **FROM CORE VALUES TO ECONOMIC VALUE: A COMMON FRAMEWORK FOR VALUE-BASED DECISION-MAKING**

As outlined in the previous sections, economic value and core value both refer to evaluative representations that guide decisions and behaviors. They are however conceptualized at different levels of situational concreteness, with economic value referring to a common currency that operates in concrete choice situations, and core value referring to motivational constructs that guide choices and behaviors across many situations. Despite the conceptual similarities, there has not been much integration and cross-fertilization between the two research traditions. We suggest combining the two value concepts into a common framework for decision-making. In linking these two concepts, neuroeconomic research may be enriched by an elaborate and empirically validated concept that allows predicting and explaining individual differences in value-based decision-making. Furthermore, integrating the set of core values and the related behaviors into neuroeconomic research goes beyond the kind of choices that are usually investigated empirically, moving from simple choices between consumer goods to a more diverse and complex array of choices. In return, core value research may gain a deeper understanding of the underlying mechanisms by which core values impact on decisions and behaviors. In this context, several core value researchers have suggested that the effects of core value on decisions and behaviors are relatively indirect, being exerted by changing the beliefs and norms of the individual (Dietz et al., 2005) or by exploiting one's need for consistency between beliefs and actions (Rokeach, 1973).

Here we want to evaluate the possibility that, in addition to these indirect effects, a more direct connection links core value, economic value, decision-making and behavior. Our hypothesis is that individual differences in core value may be determinants of how much economic value is given to the different options in concrete choice situations. Thus, the behavioral effects of core value differences may—at least partly—be implemented by neural mechanisms underlying the computation of economic value. In what follows, we will review the relevant neuroimaging evidence against which our hypothesis can be evaluated. Whereas to our knowledge only two studies have so far directly addressed the impact of core values on neural activation (Brosch et al., 2011, 2012), a number of other neuroeconomic studies have investigated the neural correlates of a specific behavior that is relevant to the core value dimension of self-enhancement vs. self-transcendence: egoistic vs. altruistic behavior expressed by charitable donations. The first neuroimaging study to investigate the neural correlates of charitable donations (Moll et al., 2006) presented participants with a series of choices on whether to donate money to a charitable organization related to a major societal cause (such as children's rights, gender equality, or nuclear power). In other trials, participants received money for themselves. Results revealed increased activation of the striatum, a central part of the neural system representing economic value, both when participants received money for themselves and when they decided to donate for a good cause. In further research, the perceived value of charitable donations has been shown to be represented in VMPFC/OFC as well (Hare et al., 2010). Taken together, these findings suggest that receiving money and donating money are both rewarding experiences, as expressed by a shared anatomical system of value representation. These findings were extended by demonstrating that increased striatal responses to charitable money transfers also occur when the transfer is mandatory (similar to an income tax), but that the striatal response is even higher when people voluntarily decide to make a donation (Harbaugh et al., 2007). In another study, participants were matched into pairs and presented with a series of unequal monetary distributions, where one participant received a large monetary endowment and the other one nothing (Tricomi et al., 2010). Participant who had already received a lot of money in previous trials showed a stronger neural response in VMPFC/OFC and ventral striatum when they observed a money transfer to the other participant (who had previously received less money), compared to when they received money themselves, indicating that the neural value regions also represent value related to distributive fairness. Finally, in a study on moral dilemmas, participants were confronted with scenarios where they had to make decisions that sacrificed the lives of some people in order to save others. The expected "moral reward value" (i.e., the ratio of lives saved/lost) was tracked by VMPFC/OFC and ventral striatum, suggesting that decisions based on self-transcending values may involve the same neural systems that represent economic value (Shenhav and Greene, 2010). Taken together, these results suggest that the neural regions representing economic value are involved in decisions and behaviors that are related to core values.

But are individual differences in the activation of these regions related to actual differences in altruistic decisions and behaviors? In the taxation-donation study by Harbough and colleagues described above, participants who showed a stronger striatal response when receiving money for themselves opted less frequently to donate money to charity (Harbaugh et al., 2007). Furthermore, in a study looking at individual differences in preferences for distributive fairness, participants who generally choose equal distributions of money showed increased amygdala activation when confronted with very uneven distributions (Haruno and Frith, 2010). These two studies suggest that behavioral differences that are relevant to core values may indeed be driven by differences in activation of neuroeconomic value regions.

As a final step in our argumentative chain, it remains to be shown that different neural activation patterns in economic value regions are actually related to individual differences in the core value hierarchy. To address this issue, we measured the core value hierarchies of individuals who participated in a donation task (Brosch et al., 2011). In some trials, participants could gain money for themselves, in other trials they decided whether they wanted to donate some of their money to charity. Analysis of the decisions made during the task showed that participants with selfcentered core value hierarchies donated less money to charity, demonstrating that more self-interested core values are actually reflected in more selfish behavior (see **Figure 2**). At the neural level, all our participants showed increased activation of the striatum when receiving money. However, the activation was more pronounced for participants with a more self-centered core value hierarchy, suggesting that egoistic behavior is potentially more rewarding for participants with self-centered core values than for less self-centered participants.

Participants with self-centered core values furthermore showed a stronger neural response of the amygdala when having the opportunity to gain money for themselves, consistent with the suggestion that the amygdala acts as a relevance detector that is sensitive to the motivational salience of a stimulus given the current needs, goals and values of the organism (Davis and Whalen, 2001; Sander et al., 2003; Pessoa, 2010; Cunningham and Brosch, 2012).

Somewhat surprisingly, participants showed decreases in striatal activation when deciding to donate their money to charity, consistent with striatal deactivations observed during financial loss (Delgado et al., 2003), which is in contrast to studies reviewed above that reported increased striatal activation during altruistic donations (Moll et al., 2006; Harbaugh et al., 2007). The difference between our results and the results by Harbaugh et al. (2007) and Moll et al. (2006) may be due to contextual or methodological differences. For example, in the study by Moll and colleagues, participants were confronted with a different charitable organization in each trial, which included also organizations whose goals were not endorsed by the participants, whereas in our study, participants always donated to the same charitable organization that was chosen by the participant in advance. Furthermore, in the study by Harbaugh and colleagues, the monetary payoff to the charity was not correlated with the financial loss by the participant (i.e., the experiment contained trials where the participant lost USD 45, but the charity only received USD 15, as well as trials where the participant lost USD 45 and the charity received all of it). The striatal response reflects increased activation to increased monetary payoff to the charity; this analysis is thus not sensitive to the effects of the financial loss by the participant.

Taken together, striatal activation differences have been shown to be linked to behaviors reflecting self-interested as well as selftranscendent core values. Furthermore, our results point to an additional neurocognitive process involved in self-transcendent behavior that involves social cognition mechanisms: In our study, when facing the opportunity to donate money, the more generous participants showed increased activation in dorsomedial prefrontal cortex (DMPFC), which, together with temporoparietal junction (TPJ) and precuneus forms a social cognition network that is involved in forming impressions of others and in thinking about the needs, goals, and beliefs of others (Frith and Frith, 1999; Van Overwalle, 2009). Thus, altruistic behavior may be related to a more thorough evaluation of the needs and goals of others rather than one's own needs. Consistent with this notion, another donation study observed that activation in right TPJ was correlated with the participants' willingness to donate money to a charitable organization (Hare et al., 2010). Furthermore, neuroanatomical differences in gray matter volume in TPJ have been shown to be strongly associated with altruistic behavior (Morishima et al., 2012), providing a potential biological substrate that may underlie the stability of altruistic choices.

Taken together, the findings reviewed here suggest that core values may indeed exert their effects on decisions and behaviors via modulations of the neural regions involved in the

**FIGURE 2 | Impact of self-centered core value hierarchies on neural regions representing economic value and on charitable behavior. (A)** Participants with a self-centered core value hierarchy kept more money for themselves instead of donating it to charity. **(B)** The same participants

showed increased activation in the ventral striatum when receiving monetary rewards. **(C)** Correlation between self-interest value and parameter estimates for ventral striatum (reproduced with permission from Brosch et al., 2011).

computation of economic value: Participants with a value hierarchy dominated by self-centered core values make more selfish decisions and show a concurrent stronger activation of the ventral striatum (Brosch et al., 2011). Thus, participants with selfcentered core values may perceive selfish choices and behaviors as more rewarding, and as a consequence will show these behaviors more often than participants with less self-centered core values. Altruistic behaviors may also be reflected in differential activation of the ventral striatum (Moll et al., 2006; Harbaugh et al., 2007), but may additionally involve an increased recruitment of social cognition regions such as DMPFC (Brosch et al., 2011) and TPJ (Hare et al., 2010), which are involved in perspective-taking and thinking about the needs and goals of others. During charitable choices, social cognition regions show increased connectivity with regions representing economic value (Hare et al., 2010), and may thus increase the expected economic reward value of selfless actions.

Thus, when a person with a given hierarchy of core values faces a concrete decision situation, these core values may exert their influence on individual choices and behaviors by directly modulating the computations of the expected reward value for the different options. Previous theorizing in social psychology and sociology has conceptualized the link between core value and behavior as relatively indirect, by postulating that core values impact on the beliefs and norms of an individual which then result in behavioral differences (Dietz et al., 2005) or by assuming that value-congruent behavior is mainly driven by an individual's need for consistency between one's beliefs and actions (Rokeach, 1973). We propose that, in addition to these indirect pathways, a more direct path may underlie the impact of core value on behavior. By modulating the economic value computations for different behavioral options, core values may directly impact on the perceived reward value of the different behavioral options (see also Feather, 1995). Of course, it must be noted that all neuroimaging studies cited here have used financial decisions, and have linked core value related decision-making to higher sensitivity to monetary reward only. There are many different types of rewards, including primary rewards such as food or erotic stimuli, as well as secondary rewards such as money or power. It remains to be shown that the findings reviewed here can generalize to other situations and types of rewards. A recent meta-analysis (Sescousse et al., 2013) confirmed that the neural network computing reward value is similarly activated by different kinds of primary and secondary rewards. However, it would be highly interesting to investigate individual differences in sensitivity to different types of rewards as a function of the individual core value hierarchy (e.g., comparing the reward value of erotic stimuli in participants with highly conservative values vs. participants with pronounced stimulation and hedonism values)<sup>1</sup> .

In addition to this direct impact of core values on neural representations of economic value in the striatum and VMPFC, as well as their modulations via social cognition regions such as TPJ and DMPFC, a more indirect pathway by which core values impact on individual beliefs and norms may play an important role: Core values form an important part of our self-concept, i.e., they help us define ourselves. Thinking about oneself as "a person who values benevolence" represents a motivationally important long-term goal that may promote core value-congruent behaviors even in the absence of concrete choice situations or rewarding options. For example, a person who values benevolence may frequently make efforts to select situations and environments in which concrete altruistic behaviors can be realized, such as going to fund-raisers or charity sales, in order to act accordingly to his beliefs.

The findings reviewed here furthermore suggest a new perspective on the mechanisms that may underlie the development of differences in individual (or cultural) core value hierarchies: Some groups or individuals may habitually show stronger sensitivity of economic value regions when receiving valued objects, which may be due to either genetic factors or epigenetic factors such as social reinforcement. Habitually stronger reward sensitivity may lead to an increase in self-interested behavior via positive reinforcement and to a more positive evaluation of prospective outcomes of such a behavior in related decision-making processes. This may result in an increased probability of choosing selfish alternatives. Similar to the role of self-perception in attitude formation (Bem, 1972), habitual choice of selfish behaviors may crystallize in an accordingly adjusted core value hierarchy that emphasizes self-centered values. Once these values become integral part of the self-concept, the explicit representation of the importance of certain classes of behavior may furthermore drive decisions and behaviors, by combining explicit and implicit reinforcing mechanisms. The model outlined in this paper should be considered as a starting point only, as research on the neural correlates and mechanisms of core values is at an early stage. We hope, however, that our contribution will stimulate further research that focuses on the role of individual differences in decision-making and the underlying neural mechanisms. In this context, economists recently have begun investigating the impact of individual differences in personality traits (e.g., "Big Five") on economic decision-making (Rustichini, 2009), suggesting, for example, that neuroticism is linked to a lower willingness to accept risks, and extraversion to a reduced aversion to ambiguity. Somewhat related to the egoism-altruism dimension discussed in the present paper, it has been suggested that the personality dimension of Agreeableness may be related to higher cooperation with others. Initial data from a Prisoner's Dilemma game seems to support this link Rustichini et al. (2011). Future research on core values should aim at measuring personality dimensions and the individual core value hierarchy simultaneously, to assess which constructs are more powerful predictors of individual decisions and behaviors.

Whereas the model outlined here focuses on the core value dimension "self-enhancement vs. self-transcendence," the model by Schwartz (1992) contains a second dimension, labeled "openness to change vs. conservation." While there is hardly any neuroimaging research directly investigating this core value dimension, a number of studies have investigated neural correlates of political liberalism vs. conservatism (Jost and Amodio, 2012), a dimension that can plausibly be related to the core value dimension. Interestingly, these findings suggest that political conservatism is associated with more persistence-related errors and

<sup>1</sup>We thank one of our reviewers for providing this example.

reduced neural responses of an error-detection system centered on anterior cingulate cortex (ACC) during a Go/No-Go task (Amodio et al., 2007). Similar results have been observed for highly religious individuals (Inzlicht et al., 2009). Whereas in these studies differential neurocognitive effects are found after a decision/behavior, when the consequences of the decision are assessed and errors are detected, it remains to be explored whether "openness to change vs. conservation"-core values may also be related to a differential weighing of the perceived economic value of different options before a decision is made.

Taken together, in this contribution we aimed at demonstrating the feasibility and usefulness of an integration of economic value research and core value research. We have suggested potential mechanisms by which core values, explicitly represented as long-term goals anchored in the self-schema, may drive concrete decisions and behaviors by acting on neural regions representing economic value. Core value research provides a theoretically

# **REFERENCES**


elaborate and empirically validated concept that allows predicting and explaining individual differences in value-based decisionmaking. The theoretical integration of the different concepts opens up several new and exciting topics of research, some of them with the potential for considerable societal impact. For instance, the links between core values and behavior are sometimes relatively weak (Bardi and Schwartz, 2003). As an example, many people claim that for them the protection of the environment is an important value, but do not show consistent environmentally friendly behavior (Dietz et al., 2005). Neuroimaging research may contribute to developing targeted interventions that aim at increasing the effect of environmental core values on the corresponding behavior by exploring how situations need to be framed to elicit a high economic value of the desired behavior. Many other examples are possible. We hope that the ideas outlined here will be valuable for many researchers who care about value, and will stimulate further integration of the different value literatures.

basis. *Science* 286, 1692–1695. doi: 10.1126/science.286.5445.1692


neuroimaging studies. *Neurosci. Biobehav. Rev.* 37, 681–696. doi: 10.1016/j.neubiorev.2013.02.002


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 12 April 2013; accepted: 07 July 2013; published online: 24 July 2013. Citation: Brosch T and Sander D (2013) Neurocognitive mechanisms underlying value-based decision-making: from core values to economic value. Front. Hum. Neurosci. 7:398. doi: 10.3389/fnhum. 2013.00398*

*Copyright © 2013 Brosch and Sander. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# The role of interest in the transmission of social values

# *Fabrice Clément 1,2\* and Daniel Dukes 1,2\**

*<sup>1</sup> Cognitive Science Centre, University of Neuchâtel, Neuchâtel, Switzerland*

*<sup>2</sup> Swiss Center for Affective Sciences, University of Geneva, Geneva, Switzerland*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Tom Ethofer, University Tubingen, Germany Antony S. Manstead, Cardiff University, UK*

*\*Correspondence: Fabrice Clément and Daniel Dukes, Cognitive Science Centre, Université de Neuchâtel, Espace Louis-Agassiz 1, CH-2000*

*Neuchâtel, Switzerland e-mail: fabrice.clément@unine.ch; daniel.dukes@unine.ch*

The environment is so rich with information that our cognitive system would be overloaded without a way to evaluate what is relevant for our needs and goals. Appraisal theory has shown how emotions, by "tagging" the environment with differential values, enable the attribution of our attentional resources to what is most relevant in any given circumstances. Most often, however, the different cues triggering the allocation of attention are thought of as purely individualistic, like physiological needs or past encounters with certain stimuli. This approach is perfectly appropriate for objects, organisms or events that, by their intrinsic properties, affect the organism's well being. But for humans, many aspects of the environment are culturally or temporally dependent: a soccer game may be highly relevant to some, but not at all to others. This paper contributes to a better understanding of the processes by which different elements of our social environment acquire value through our socialization process. We recruit different concepts proposed by developmental psychologists to shed some light on this social acquisition of relevance. The notion of "joint attention," for example, is particularly important to understand how we are sensitive to the other's focus of attention. Similarly, the term "social referencing" has been used to describe the process of taking into account the affective reaction to a given stimuli, in order to direct our behavior. At the core of this process, called "social appraisal" by Manstead, we propose that a specific emotion plays a major role: interest. Someone else's expression of interest, which seems to be detectable from a very early age, is extremely useful in gauging what is worthy of attention among stimuli that are not inherently interesting. The paper highlights how external sources of information (the life experiences of community members) indicate what is relevant, thus giving access to the social values of that group.

#### **Keywords: appraisal theory, interest, social appraisal, social values, social learning**

## **INTRODUCTION**

In the long empiricist tradition that characterizes the Western conception of humankind, our inner experience of the world results from successive perceptual contacts with our biological, physical and social environment. In this context, it is hard from a developmental perspective not to think of a baby's mental life as "one great blooming, buzzing confusion," (James, 1890/1981, p. 462). How could it be otherwise in a world of objects and events that overlap and coincide in so many ways? Many of the recent breakthroughs in developmental psychology have been aimed at understanding the relative placidity of babies confronted with a plethora of information that they in fact, master impressively quickly. Everything happens as if babies were naturally equipped for such "cognitive digestion," either because they can rely on some evolutionary modules to make sense of the information due to some core knowledge—naive physics, biology, psychology, or even sociology (Baillargeon, 1994; Wellman and Woolley, 1990; Hirschfeld and Gelman, 1994; Spelke, 1994; Xu and Carey, 1996), or because they rely on powerful "pattern detectors" which enable them to detect correlations and forge hypotheses about the structure of the world (Gopnik, 2010).

These cognitive predispositions, however, are just part of the solution. Even if we imagine that babies are equipped to process in specific ways, how can they assess which information should be processed at any given time? Without a system enabling them to prioritize how to distribute their cognitive powers, the risk of behavioral paralysis is too high.

Appraisal theory provides a possible answer since it describes how not all stimuli are equal to our cognitive system, and how our emotions play the role of "radar antennae scanning the environment," (Scherer, 1994, p. 230). This relative saliency in our environment can be due to the fact that, given our organism's needs, certain stimuli have a high biological significance and are therefore automatically prioritized by the attentional system (Brosch et al., 2007). For instance, when hungry, the value of food is very high and priority is given to any process leading to the satisfying of this biological need. Other factors are more personal and depend on the individual's history: a given odor or taste for instance, can mobilize a person's attention because it reminds her of a special time with her grandmother. Finally, certain elements of our environment are socially relevant because they are considered as important for the members of the individual's reference group. In certain communities for example, a soccer World Cup Final will thus be lived as though it is a religious ritual, whereas in others, it will be ignored. For all these values, the amygdala seems to play an essential role in filtering what is relevant to the organism, and an event is relevant if it can "significantly influence (positively or negatively) the attainment of his or her goals, the satisfaction of his or her needs, the maintenance of his or her own well-being, and the well-being of his or her species" (Sander et al., 2003, p. 311).

In this paper, we will focus on social relevance and, more specifically, on how socio-cultural values pass from socialized to newly arrived social members, i.e., babies. After a brief summary of the component process model (CPM) proposed by Klaus Scherer to describe how we appraise incoming stimuli, we will concentrate on a dimension of the model that has not been given much attention: normative significance. By specifying the nature of this check, we will propose that recent attempts to describe social appraisal are particularly relevant to this topic. More specifically, we will show how the emotion of interest can be essential for the baby to discern what is valued in his or her social environment when it is displayed by significant others. To specify the competences required for "using" expressed interest to detect what is socially relevant, we will revisit three important notions that underlie the way infants can be influenced by others in their development: joint attention, social referencing and social appraisal. Consequences for the study of the role of interest will then be discussed.

# **THE SOCIAL SIDE OF APPRAISAL**

Nowadays, the CPM is one of the most complex, empirically supported and heuristic models of emotion processing (Scherer, 1984, 2001; Ellsworth and Scherer, 2003). In this model, emotions are regarded as psychological episodes that have a felt character and are evaluative of particular objects (Deonna and Teroni, 2012). Its central idea is that emotions play a key role in the way our brains scan the environment and prepare our organism for action (Leventhal and Scherer, 1987). This evaluation is performed by a series of different checks that occur in a sequential order—Stimulus Evaluation Checks (SEC). Everything happens as if the emotional processes respond to the following questions: "(a) How relevant is this event for me? Does it directly affect me or my social reference group? (relevance); (b) What are the implications or consequences of this event and how do they affect my well-being and my immediate or long-term goals? (implications); (c) How well can I cope with or adjust to these consequences? (coping potential); (d) What is the significance of this event for my self-concept and for social norms and values? (normative significance)" (Scherer, 2009, p. 1309). These checks, that often occur automatically, unconsciously and effortlessly, are supposed to follow this order.

In this model, relevance detection plays an essential role because it is considered to be a first selective filter that a stimulus or event needs to pass through in order to merit further processing (Sander et al., 2005, p. 322). However, the role played by social factors at this stage remains unclear. On the one hand, the significance of the stimulus or the event with respect to social norms and values is clearly relegated to one of the final evaluation checks. It consists in checking the compatibility with external standards: social norms, values, beliefs about justice, or moral principles (Scherer, 2009, p. 1313). By definition, it requires highlevel and comprehensive information, and even "comparison with high-level propositional representation" (Sander et al., 2005, p. 322). On the other hand, the social nature of the relevance process is acknowledged: relevance is about how a given event affects oneself or one's *social reference group* (Sander et al., 2005, p. 319). Indeed, the role of social context in appraisal has been highlighted by recent work underlining, in the CPM perspective, the role played by the amygdala for relevance. Sander et al. (2003) suggested that the human amygdala works as a "relevance detector" and is activated in presence of social signals such as gaze direction, intentions, group adherence, trustworthiness and facial familiarity. Other works have highlighted the role played by the amygdala when individuals have to evaluate trustworthiness in their social exchanges (Adolphs et al., 1998; Winston et al., 2002; Todorov and Engell, 2008). Given the rapidity of these relevance detection processes driven by the amygdala (Vuilleumier, 2005; Brosch et al., 2008), one can therefore conclude that at least a part of what is social in the CMP does not need reflexive and/or propositional processes.

If the processing of another person's facial or bodily expressions triggers the amygdala, especially when the person is looking directly at you (Conty and Grèzes, 2012), it is unlikely that these intersubjective situations are the only way that appraisal processes are influenced by the "social." Indeed, relevance is evaluated according to the significance events have for what is valued by the organism. Certain stimuli have high biological significance and are automatically prioritized by the attention system (Brosch et al., 2007), while others trigger specific evaluations of the environment as a result of personal needs and values (Ellsworth and Scherer, 2003). Social relevance aims precisely at understanding how these personal values are shaped during the development of a person. Admittedly, values do not emerge in a social void. On the contrary, children most likely develop a certain number of their personal preferences and values as a result of the contact they have with their social referents (parents, friends, teachers, coach, etc.). Once this transmission period ends, specific parts of their environment will more or less automatically trigger their attention and interest. For a child born into a family of cyclists, for instance, a champion like Eddy Merckx would be immediately detected in a crowd and this episode would be remembered forever, while on the contrary, the cyclists' hero would stay unnoticed by a family of soccer players. Therefore, depending on the "attentional" priorities of their social milieu, an individual's environment tends to be colored by different "social lenses" that will render certain elements of the world as valuable and worthy of attention, at the same time sentencing the others to indifference and invisibility.

Therefore, it seems very important (even from a psychological point of view) to understand how values, inherent in a given form of life (Clément, 1996), pervade our appraisal system, to the point that it can influence the very beginning of the SEC. As the following section will show, developmental psychology appears to offer some indications.

## **ATTENDING TO OTHERS**

In many senses, the human species is fundamentally a social species. As the anthropologist Clifford Geertz wrote: "human behavior is so loosely determined by intrinsic sources of information that extrinsic sources are so vital" (Geertz, 1973/1993, p. 93). Even if not credulous, children learn most of what they know via others' testimony (Clément et al., 2004; Clément, 2010). Moreover, the human species is fundamentally altricial: offspring are highly dependent on others for a very long period of time. It is therefore not very surprising that other people's appraisal systems can influence one's own evaluation of events and stimuli.

From a cognitive perspective, newborns seem to be "prewired" for attending to human-like faces (Johnson et al., 1991; Heron-Delaney et al., 2011). This preference may well be due to the fact that there is an evolutionary advantage for babies in treating other human faces as particularly relevant to making sense of their surroundings, given their richness as a source of information. Furthermore, from an early age, babies are increasingly able to follow someone else's gaze (Scaife and Bruner, 1975). For instance, 2–5-day-old newborns can discriminate between direct and averted gaze, and 4-month-old infants' brain activity shows specific neural activity when presented with faces with direct (as opposed to averted) eye gaze (Farroni et al., 2002). A few months later, evidence begins to emerge that infants start looking at the world via others' perspectives. At 12 months, infants are able to detect selective attention when an adult looks at several things but attends only to some part of them (Tomasello and Haberl, 2003). This capacity, called *joint attention*, plays a crucial role in development, notably in language learning. Indeed, it is thanks to joint attention that caregivers and infants can establish what is being referenced, and learn that certain sounds match with objects, persons or events in the shared environment (Tomasello, 2003). From a perceptual point of view, it is therefore very likely that infants are prone to select their objects of attention, at least partially, by aligning their own attention with others' attention.

Beside the ability to take into account others' objects of interest and to be driven to be attentive to the same objects, babies around that age (the end of their first year) start to move around on their own, and they gain new ways of feeding their appetite for exploration. Facing all kinds of new objects, they have ever more opportunities to create mischief. Fortunately perhaps, it is also at this age that they start to rely more on their caregivers' cues (facial expressions, body language and tone of voice) to appraise ambiguous and new events (Klinnert et al., 1986). As Feinman and Lewis (1983) put it, caregivers serve from now on not only as a base of security but also as a base of information. Such social information gathering "is rooted in the ability to appreciate that another individual can function as a conduit for information about the world" (Baldwin and Moses, 1996, p. 1917). In the famous visual cliff experiment, infants dared to crawl over a simulated cliff when the mothers expressed joy or interest (Sorce et al., 1985). This ability that the infant has to disambiguate the emotional meaning of objects in the environment by actively seeking out emotional information from significant others (Hertenstein and Campos, 2004) has been called *social referencing*. It is important to highlight the fact that, contrary to joint attention where infants' focus of attention is driven by another's gaze direction, the object of concern for social referencing pre-exists in the infant's conscious field. In a way, the child has already evaluated the object as relevant, and the social information she obtains is essentially used to modulate her behavior toward that object. We propose therefore, a slightly less inclusive definition than Feinman and Lewis (1983, p. 878), who define social referencing as the use of one's perception of someone else's interpretations of a situation to form one's own understanding of that situation. We agree more readily with Pelaez et al.'s (2012) definition of social referencing as "a behavior chain in which the presence of an ambiguous object or event signals the gaze shift of an infant toward another person, typically the mother, whose facial, vocal, and gestural expressions may then serve as discriminative stimuli for a subsequent approach response" (p.23). We therefore endorse the view that social referencing directs behavior, rather than forms an understanding.

In contrast to the aforementioned cases, there are situations where the focus of a given object or event does not pre-exist the social interaction, or when the evaluation of the object itself is modified by the nature of the social information. Children, for instance, can be intrigued by the way adults are captivated and excited by a soccer game on television (Demers et al., 2013); in those families, we can expect children to be sensitive to future soccer related events. On the contrary, the appraisal that underlies an activity like stuffing oneself with ice cream can be modified by a strong and negative emotional parental reaction. Even occurrent emotional reactions triggered by an individual appraisal, for instance, bursting out laughing when seeing an old man stumbling in a bus, can be re-evaluated once the emotional reactions of the other, disapproving witnesses are taken into consideration (Jakobs et al., 1997). To refer to these cases where the value of events or objects are modified by the observation of other people's emotional reactions, Manstead and his colleagues (Manstead and Fischer, 2001; Evers et al., 2005) have proposed the concept of *social appraisal*. One of Manstead's objectives is to highlight the fact that most appraisal theorists tended to favor research that focuses on relatively socially isolated individuals, and on values that are essentially independent from the socio-cultural environment. In contrast, social appraisal highlights the fact that "the behaviors, thoughts or feelings of one or more other persons are often appraised in addition to the appraisal of the event *per se*" (Manstead and Fischer, 2001, p. 222).

Social appraisal can be expected to play a considerable part in a child's socialization, given that there are many events and objects in our social environment that are not relevant in terms of their intrinsic properties, and the fact our social environment is full of objects that arouse considerable interest for certain groups of people, but not to others.

It is only via others' appraisal that the relevance of a particular artistic form, sport, hobby, political engagement, or environmental consciousness becomes salient for the children. However, while there is an abundance of developmental research on joint attention and social referencing, the role of social appraisal has not really been identified in infancy. Compared to social referencing, where others' emotions seem to play a regulatory role in the expression of a behavior by encouraging or discouraging the on-going action, it is most likely that social appraisal necessitates a much finer understanding of the expressed emotion. To play the role of relevancy detectors, others' faces have to be "read" by the children: only a rather subtle interpretation of others' appraisal can help them to detect if an object or event is worthy of attention, on a scale going from "abhorrent" to "highly desirable." In this context, we hypothesize that certain expressed emotions play an essential role by "tagging" certain stimuli with a given emotional valence for the children. Such a transmission of values can be intentional and explicit: parents, for instance, may find that is very important to transmit their love of the arts, or the virtue of politeness, to their heirs. In these cases, parents may resort to what Gergerly and Csibra call *natural pedagogy* with ostensive communication to indicate new and relevant information (Gergely and Csibra, 2006; Csibra and Gergely, 2009). It has been shown in such pedagogical contexts that mothers adapt their voice and speech when talking to young children, speaking "motherese" (Snow, 1972). More generally, adults also modify their movements when interacting with infants such that their actions simultaneously enhance infant's attention and highlight meaningful units within the flow of motion (Brand et al., 2002). More specifically, when mothers show objects to young children, relative to showing them to other adults, their actions are notably characterized by closer proximity to the partner, greater enthusiasm, a larger range of motion, greater repetitiveness, longer gazes, more turn-taking and greater simplification (Brand et al., 2007). However, we argue that not all social transmissions of values rely on such ostensive cues. By observing others, children (and adults) can detect what captures their attention or, on the contrary, what they disregard: an expression of awe or an expression full of scorn, even expressed by an anonymous by-stander, can still be very socially relevant.

For this third-party influence in the ontogenesis of social relevance, we suppose that certain expressed emotions will play an essential role, notably disgust, contempt, and interest. We will focus here on interest because (1) it is an emotion that has not yet been extensively studied, (2) it should indicate to an observer what another person appraises as being "worthy of interest", i.e., as *relevant*. Interest is therefore an emotion of crucial importance for social appraisal.

## **THE ROLE OF INTEREST**

Given the scope of this paper, we cannot discuss here all the aspects of interest (but see Silvia, 2006). Briefly, interest is the emotion associated with curiosity, exploration, and information seeking (Tomkins, 1962; Berlyne, 1966; Izard, 2009). According to Silvia, interest as a felt emotion consists of appraisals of novelty (factors related to the unfamiliarity and complexity of an object or event) and appraisals of coping potential (the ability to understand the new object or event) (Silvia, 2005). Its function is to motivate seeking behaviors, learning and exploration (Panksepp, 2005; Silvia, 2008). One of the important questions in studying interest concerns the existence of a specific expression of interest. This aspect is especially important given our problematic: social appraisal could not take place without cues that enable children to detect others' interest.

The expression of interest has apparently no place among the most renowned and widely used basic emotion stimuli that Ekman considered as universal: happiness, sadness, anger, fear, disgust, and surprise (Ekman and Friesen, 1971). Actually, it appears from later research that Ekman had considered including both "interest" and "contempt" in the series, but presumably he was unable to find suitable static photographs (Ekman, 1992, 1993). However, even if "momentary expressions" are particularly efficient from an evolutionary perspective, Ekman did not deny the important role that "extended expressions" might play (Ekman, 1993). Interest seems precisely to be one of these extended emotions and it is therefore not surprising that even adults cannot recognize static stimuli of interest. This was first identified in a 1964 study where interest was one of eight expressions presented to participants (neutrality, surprise, distress, enjoyment, fear, anger, disgust, shame, and interest) and where interest was only more frequently recognized than surprise (Tomkins and Mc Carter, 1964). Interestingly, Tomkins and Mc Carter report that the actors they had hired to pose the expressions complained particularly about how difficult it was to pose "interest." We had a similar difficulty with the actors that were hired for a study we are currently conducting with adults, as if playing a static interest, contrary to a dynamic one, was impossible. In our experiment, participants are asked to watch pictures or movies and to recognize the staged emotions based on Ekman's basic expressions, including a neutral/calm expression and the expression of interest. Static headshots are not well recognized for interest, particularly when compared to six-second films of dynamic headshots and dynamic whole body shots (Dukes et al., in preparation). Similar results were found in another study in which four positive emotions (pride, pleasure, joy, and interest) were compared (Mortillaro et al., 2011). Facial expressions of each emotion were taken from the Geneva Multimodal Emotion Portrayal corpus in which each actor was asked to express each emotion several times [see Bänzinger et al. (2012) for details]. Representative facial expressions were then coded using the Facial Action Coding System (FACS). While the four positive emotions could not be differentiated on the basis of the presence or absence of particular Actions Units (AU), they could be differentiated in terms of their temporal dynamics—the sequence and timing of the unfolding expression [see Krumhuber et al. (2013) for a review of the dynamics of facially expressed affect].

From the perspective of studying the role of expressed interest in the ontogenesis of social appraisal, it will therefore be essential to expose infants to dynamic stimuli. The other important point to be assessed is the "contagion" of the interest. In other words, does the observation of someone being interested by an object cause children to also appraise this object with interest? One of the dimensions is *behavioral*. As interest is the emotion that underlies curiosity, seeking and exploratory behaviors are expected toward an object that has been considered with interest by a third-party. A more subtle behavioral dimension is eye gaze pattern: by using an eye-tracker, it should be possible to detect whether an object of someone's interest triggers more curiosity and becomes more visually explored by the participant. Another dimension is the transmission of the emotion itself: when an infant sees a person being interested in something, will he/she start being interested in the same object? Such an inquiry is rather complicated because one has to identify on babies' faces the signs of interest. Actually, several early studies have described the facial expression of interest in babies. One such study argued that babies as young as 9 months were able to express interest (Izard, 1980), while another study suggested that infants may facially signal emotions, including interest (Oster, 1978). A further study described several facial movements as indicative of an expression of interest, such as brows raised, brows knit, eyes widened and rounded, eyes squinted, cheeks raised, mouth opened and relaxed, tongue moved, lips pursed [(Izard, 1979); Izard, as cited in Langsdorf et al. (1983)]. Using these indications, Langsdorf et al. (1983) showed that facial expressions of interest predicted the time that the infants spent viewing human or inanimate objects while Izard et al. (1995) show that expressions of interest were morphologically stable between the ages of 2.5 and 9 months. Another important facet of the expression of interest appears to be the "body stilling and facial sobering" (Camras et al., 2002) or "freezing" (Scherer et al., 2004) that characterizes a reaction to a novel stimulus: the whole body and facial expression remains motionless for a moment after the stimulus becomes known. It has been argued that freezing is a normal reaction to an ambiguous situation, as the person is unsure how to react, and that this is more likely to occur in very young infants who "do not yet have the necessary cognitive mechanisms (nor the stored experiences) to conclusively appraise highly unusual events and to prepare appropriate action tendencies" (Scherer et al., 2004, p. 399). It is as if the organism is "buying time" to disambiguate the situation before reacting. Of course, the importance that freezing may play in an expression of interest shows again why interest might be more recognizable when presented dynamically.

By putting together what we currently know about (a) the way interest is expressed and (b) the different cues indicating that infants are experiencing the emotion of interest (behavior, eye gaze patterns, and felt interest), it is possible to conceive of studies that seek to understand when and how children are able to take into account the attentional parsing of the environment performed by their caregivers. Given that joint attention is assumed to emerge around the age of 12 months (Carpenter et al., 1998; Moll and Tomasello, 2004), we suspect that this ability emerges during the second year. Furthermore, as children have been shown to learn a lot about their social environment by observation alone (Rogoff, 2003), it is likely that this third-party appraisal does not require any ostensive signals from the adults in order to be accomplished.

#### **CONCLUSION**

The main objective of this paper was to consider more carefully the role played by others in the determination of what is relevant in our surrounding. Apart from some very basic biological values, most of the objects, events, and phenomena we consider as worthy of investment in time, energy or resources, we in fact inherit from our social and cultural environment. It seems therefore important to study, in an appraisal theory perspective, the last step of the SEC proposed by Klaus Scherer: normative significance.

Our feeling is that, via social information gathered by children from very early on, norms and values are so deeply embedded in the appraisal process that even the first evaluation check—how relevant is this event for me?—is marked by the social history of the individuals.

When scanning the environment, some objects or events seem more salient than others. These objects/events are often more salient because they are relevant to the individual's goals. We have shown the importance of an individual's life experience in the detection of what is relevant and therefore their "choice" of object about which they will appraise. This does not mean that this experience has not been tainted by numerous encounters with significant others who shared, explicitly or not, what they considered as interesting, appalling, or insignificant. But, at a given time, all these life experiences can act as an *internal* source of information when appraising an event. In other situations, the importance of the third person's perspective can be brought to the forefront in the appraisal process: in this case, it is some *external* source of information (the life experience of another person) that will influence the appraisal. Clearly, this third person directs her attention as a consequence of her own life experiences and values, which again were elaborated in contact with others. When these multiple social appraisals happen in a relatively interconnected circle, nothing less than a culture is transmitted and perpetuated.

To study the very beginning of this cultural process and to prepare the ground for experimental studies with infants, we proposed a conceptual gradation in the way children take advantage of social information in the early stages of their development. From a *perceptual* point of view, babies are sensitive to others' direction of gaze. At the end of their first year, they can detect others' selective attention and join their own attention to those of their caregivers—*joint attention*. From a *behavioral* point of view, they can, at around the same period, actively seek emotional information from significant others to modulate their own behavior—*social referencing*. Eventually, most likely in the second year, infants are able to take into consideration an emotion expressed by another person to appraise an event, object or person—*social appraisal*.

The emotion of *interest* appears to be particularly relevant for studying the onset and development of social appraisal by children. In expressing interest, adults offer important cues about what is salient for them in their environment. We hypothesize that every expression of interest that children can detect on an adult's face and body, enables them to "tag" their environment with different levels of saliency. Social appraisal plays therefore a crucial role for children: it enables them to enter a given society by gaining access to the values that are essential to the members of their reference groups.

#### **ACKNOWLEDGMENTS**

The authors would like to thank the two anonymous reviewers for their helpful suggestions and Joseph J. Campos and Philippe Rochat for their helpful advice. This work was supported by the National Centre of Competence in Research (NCCR) "Affective Sciences: Emotion in Individual Behavior and Social Processes", financed by the Swiss National Science Foundation (SNSF, 51NF40-104897), and hosted by the University of Geneva.

## **REFERENCES**


eds N. J. Enfield, and S. C. Levenson (Oxford: Berg Publishers) 229–255.


with mother present. *Dev. Psychol.* 22, 427–432. doi: 10.1037/0012- 1649.22.4.427


appraisal mechanisms in emotion. *Neural Netw.* 18, 317–352. doi: 10.1016/j.neunet.2005.03.001


the puzzle of infants' expressive reactions to expectancy violation. *Emotion* 4, 389–402. doi: 10.1037/ 1528-3542.4.4.389


*Neurosci.* 5, 277–283. doi: 10.1038/ nn816

Xu, F., and Carey, S. (1996). Infants' metaphysics: the case of numerical identity. *Cogn. Psychol.* 30, 111–115. doi: 10.1006/cogp.1996.0005

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 16 April 2013; accepted: 29 May 2013; published online: 17 June 2013.*

*Citation: Clément F and Dukes D (2013) The role of interest in the transmission of social values. Front. Psychol. 4:349. doi: 10.3389/fpsyg.2013.00349*

*This article was submitted to Frontiers in Emotion Science, a specialty of Frontiers in Psychology.*

*Copyright © 2013 Clément and Dukes. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# *Claudia Civai\**

*Department of Economics, College of Liberal Arts, University of Minnesota, Minneapolis, MN, USA \*Correspondence: claudia.civai@gmail.com*

*Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Luke J. Chang, University of Colorado, USA*

In the following paragraphs, I am arguing that rejecting inequality, even when it means sacrificing available resources, could be interpreted as a default response that occurs when there is no other reason to choose otherwise. Moreover, I am reviewing some of our latest findings suggesting that emotions might not be the sole mechanism that ultimately explains this response, as claimed instead by the most accredited account (e.g., Sanfey et al., 2003; van't Wout et al., 2006; Crockett et al., 2008; Tabibnia et al., 2008). The idea that a 50-50 share is preferred over other distributions, when there is no reason to support one of the contending parties, is not new to the psychological debate: it has been suggested that people use equality heuristically, because it has psychological advantages, such as being a cognitive simple strategy, easy to use and to be understood by everyone, quickly implemented, defensible, and, moreover, a useful starting point from which, in case, adjustments can be made (Messick and Schell, 1992; Messick, 1995). Furthermore, less-equal distributions are consistently rejected more often among different human populations (Henrich et al., 2006). The central claim of Bicchieri's book *The Grammar of Society* (2006) is that an equal-division norm plays a critical and under-appreciated role in driving behavior in bargaining games (Bicchieri, 2006; Nichols, 2010). Research in the field of behavioral economics has demonstrated that the model of *homo economicus* often fails to predict human behavior: the Ultimatum Game (UG) (Güth et al., 1982), a widely employed tool to investigate socio-economic decisionmaking, perfectly shows how people do not always make decisions driven by the principle of maximizing monetary payoff. In this game, a proposer has to share some money with a responder, who can either accept or reject the offer: if he accepts, the money is divided as the proposer has established, otherwise both of them get nothing. To maximize their payoffs, the proposer should offer the smallest amount of money, and responder should always accept, as even one is better than zero. However, numerous findings show that the proposer tends to make fair offers, around 50% of the share, while the responder prefers to reject a sure amount of money rather than accepting an unfair division. Models of social preferences provide a formal explanation for the apparently irrational behavior. Two families of theories, i.e., theories of negative reciprocity (e.g., Rabin, 1993; Falk and Fischbacher, 2006) and theories of inequality aversion (e.g., Fehr and Schmidt, 1999) tried to explain rejections: the former focuses on intentions and describes rejections as a tool to punish the unfair proposer, whereas the latter focuses on the outcome and claims that people are naturally averse to unequal distributions, especially when disadvantageous. Recently, Tricomi et al. (2010) found support for this claim, showing that basic reward brain structures, such as ventral striatum and ventromedial prefrontal cortex (vMPFC), are involved in both advantageous and disadvantageous inequity. From a psychological perspective, negative emotions, such as anger and frustration, elicited by the unfair treatment, are accounted to cause rejections (Pillutla and Murnighan, 1996), and a number of neuroscientific findings support this hypothesis: for example, van't Wout et al. (2006), using the skin conductance response (SCR) as a measure of emotional activation, reported that people were more emotionally aroused, showing a higher SCR when rejecting, as opposed to accepting, unfair offers. Moreover, areas known to be involved in emotional control, such as vMPFC, and in processing negative emotions, such as anterior insula (AI), are found also to be activated by rejections, and not acceptances, of UG unfair offers (e.g., Sanfey et al., 2003; Koenigs and Tranel, 2007).

However, if it is true that the accounts described above, i.e., negative reciprocity, inequality aversion and emotional involvement, explain responder's behavior in the standard UG paradigm, it is hard to develop a psychological interpretation of broad inequality perception based on the evidence collected using this standard version. First of all, UG is a self-centered task: perception of unfairness is confounded with self-serving bias, questioning whether responder is actually rejecting disadvantageous outcomes rather than a general idea of unequal division; also, it is unclear if anger and frustration are elicited by unfairness or by self-involvement. Second, the proposer is always the source of the unfair division confounding outcome and intentions concerns. Many studies have addressed the issue of intentions (e.g., Sutter, 2007; Falk et al., 2008); in particular, Blount (1995) compares the rejection rate of allocations decided by either a person or an algorithm that shares a sum of money randomly between two players, finding higher rejection rates in the first case compared to the second. However, the demands of this task were different, in that the experimenter asked the participants to indicate their expectations on the two distributions prior to the choice period, and this may have biased the responses (Sanfey, 2009). Nonetheless, rejection rate for the algorithm condition was not zero, confirming that outcome still plays a role as well. Third, the proposer, who decides how to allocate the money, always benefits from one part of the share, thus the responder never faces outcomes which exceed the 50% of the pie, confounding rejections of unequal outcomes with rejections of disadvantageous payoffs, and leaving questions concerning advantageous inequality unanswered.

Our research aimed at understanding the nature of a general inequality aversion, if any, employing manipulations of the traditional UG. First, we addressed the issue of the self-serving bias, by asking participants to play as responders both for themselves (myself-MS-condition), and on behalf of a thirdparty (TP condition), in which their payoff is not affected by their decision. Borrowing a famous expression coined by Adam Smith in his work *The Theory of Moral Sentiments* (1759), this manipulation put the participant in the condition of the "impartial spectator,"1 in that the decision made by the participant affected someone else's pockets; this way, it was possible to disentangle between the two hypotheses, i.e., rejections and negative emotions as elicited by the perception of unfairness itself, or rejections and negative emotions as related to the fact of being the target of the unfair division. We employed this paradigm in two studies: in the first study, we recorded the behavior, as the percentage of rejected offers (RR), and the SCR, to get a measure of emotional arousal (Civai et al., 2010), and in the second study we investigated neural activation by the mean of functional magnetic resonance (fMRI) (Corradi-Dell'Acqua et al., 2012). In both studies, behavioral analysis showed no difference between MS and TP: specifically, RR was higher for unfair offers, and decreased as the offers became fairer, both in MS and in TP. However, behavior dissociated from both psychophysiological and neural activations. In the first study, participants were more aroused, showing higher SCR and higher subjective emotional ratings, when rejecting, compared to accepting, offers in MS, but not in TP, where, instead, there was no effect of response on SCR. These results suggested that, albeit emotional arousal clearly enters the decision-making process, it should not be held as being the only mechanism that triggers rejections, in that rejections occurred also when there was no sign of it. Neuroimaging data of the second study revealed a dissociation between the medial prefrontal cortex, specifically associated with rejections in MS condition, thus confirming its role in self-related emotional responses, and the left AI, associated with rejections in both MS and TP conditions, supporting the hypothesis of a role played by this area in promoting fair behavior also toward third-parties (Spitzer et al., 2007; King-Casas et al., 2008). In both studies, findings in TP condition support the idea that people are concerned about unfairness among others, as showed by previous studies (Fehr and Fischbacher, 2004).

In two subsequent studies, we asked responders to decide whether to accept or reject allocations made by an external proposer, which could be either a person or a random number generator; MS-TP manipulation was maintained. This design rules out the possibility of using rejections to punish the source of unfairness; the idea is that if rejections still occur, then they have to be driven by the outcome and not by the unfair intentions. Moreover, responders were presented with allocations which were unequal but, at the same time, advantageous for them, allowing disentangling between decisions on disadvantageous and advantageous inequality. In both studies, participants rejected unequal offers, showing to care about the outcome rather than specifically about the intentions. In particular, unequal allocations in TP were mostly rejected, as well as unequal disadvantageous offers in MS, but unequal advantageous offers were mostly accepted (Civai et al., 2012). Imaging results showed a higher activation of the MPFC for disadvantageous, as opposed to advantageous, offers in MS, but not in TP, and this activation was negatively correlated with rejections; activation in the AI, instead, was higher for unequal offers, both disadvantageous and advantageous, irrespectively of the target (MS and TP) (Civai et al., 2012). Behavioral results confirmed that people prefer equal divisions and care about equality among third-parties; however, these findings also suggest that people change their preference when involved in first person, accepting inequality when it brings them an advantage on the other player. In terms of neural activations, the involvement of MPFC in MS rejections was confirmed; this activation extended more dorsally with respect to the MPFC activation found in the previous imaging study, supporting a recent account which claims that dorsal MPFC may be involved in shifting preference from a default option, represented in this case by rejecting the outcome, to a new one (Boorman et al., 2013). Interestingly, AI was activated by the perception of inequality, and was by no means related to the advantageousness of the offer in MS, supporting the idea of a crucial role played this area in signaling deviations from the norm, or expected outcome (King-Casas et al., 2008; Xiang et al., 2013).

In conclusion, our findings support an account that considers the rejection of inequality as a cognitive heuristic, a psychological anchor, which is a useful starting point that can be easily adjusted when salient contextual cues enter the environment and influence the decision. In our studies, third-party condition can be considered as the neutral situation, designing a context in which participants have no particular reason to accept inequality, except for maximizing the total payoff; in this neutral condition, people apply the simple strategy of equal split. First-person involvement (MS condition) is a salient contextual cue that modifies the environment and shifts the preference from 50-50 shares to outcomes that favor the responder. This interpretation is in line with recent findings about expectations, which showed that participants are more prone to reject offers when primed with expectations of fairness (Sanfey, 2009); moreover, a formal model that considers expectations outperforms models of inequity aversion in predicting behavior (Chang and Sanfey, 2013). In this framework, expectations can be considered as the contextual cues that shift preferences

<sup>1</sup>Although the semantic expression in this context is appropriate, the concept of impartial spectator as described by Adam Smith in his work The Theory of Moral Sentiments is different: here, the author argues that, in order to go beyond our own presuppositions to judge a situation, we do not have to rely literally on a third-party impartial spectator, but rather we have to "remove ourselves, as it were, from our own natural situation, and endeavor to view them at a certain distance from us" (Smith, The Theory of Moral Sentiments, 1759, III, I, 2).

away from the default 50-50. Interestingly, it seems that emotional arousal is limited to disadvantageous unfairness; however, the rejection of equality norm's violation occurs despite the lack of emotional arousal, suggesting the cognitive nature of the equal split heuristic. As far as the neural correlates are concerned, results suggest that the activation of AI in the UG be interpreted as a signal of deviation from an expected outcome (Chang et al., 2013; Xiang et al., 2013), which is, in this case, the equal split, rather than just a sign of emotional arousal; this interpretation also offers a straightforward and parsimonious way to account for the variety of cognitive and emotional tasks in which the AI has been found to play a role (see Craig, 2009 for a review on the tasks).

### **REFERENCES**


*Received: 11 February 2013; accepted: 21 March 2013; published online: 05 April 2013.*

*Citation: Civai C (2013) Rejecting unfairness: emotiondriven reaction or cognitive heuristic? Front. Hum. Neurosci. 7:126. doi: 10.3389/fnhum.2013.00126*

*Copyright © 2013 Civai. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# All inequality is not equal: children correct inequalities using resource value

# *Alex Shaw\* and Kristina R. Olson*

*Social Cognitive Development Lab, Department of Psychology, Yale University, New Haven, CT, USA*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Jessica Sommerville, University of Washington, USA Nikolaus Steinbeis, Max-Planck Society, Germany*

#### *\*Correspondence:*

*Alex Shaw, Social Cognitive Development Lab, Department of Psychology, Yale University, 527 Chapel St. Apt A2, New Haven, CT 06511, USA e-mail: alex.shaw@yale.edu*

Fairness concerns guide children's judgments about how to share resources with others. However, it is unclear from past research if children take extant inequalities or the value of resources involved in an inequality into account when sharing with others; these questions are the focus of the current studies. In all experiments, children saw an inequality between two recipients—one had two more resources than another. What varied between conditions was the value of the resources that the child could subsequently distribute. When the resources were equal in value to those involved in the original inequality, children corrected the previous inequality by giving two resources to the child with fewer resources (Experiment 1). However, as the value of the resources increased relative to those initially shared by the experimenter, children were more likely to distribute the two high value resources equally between the two recipients, presumably to minimize the overall inequality in value (Experiments 1 and 2). We found that children specifically use value, not just size, when trying to equalize outcomes (Experiment 3) and further found that children focus on the relative rather than absolute value of the resources they share—when the experimenter had unequally distributed the same high value resource that the child would later share, children corrected the previous inequality by giving two high value resources to the person who had received fewer high value resources. These results illustrate that children attempt to correct past inequalities and try to maintain equality not just in the count of resources but also by using the value of resources.

#### **Keywords: fairness, value, inequity aversion, social norms, social exchange**

Fairness is certainly important to human society (Boyd and Richerson, 2005), but deciding how best to be fair is no easy task. Adults and children must balance many different obligations and norms when deciding what is fair: they need to consider past inequalities, reciprocity, the value of resources, social relationships, and the amount of work that others have done when deciding how to fairly distribute resources (Fiske, 1992; Mills and Clark, 1994; Fehr and Schmidt, 1999; Olson and Spelke, 2008; Moore, 2009; Shaw and Knobe, 2013). Indeed, sometimes doing the fair thing requires, counter intuitively, distributing resources unequally. For example, imagine two employees both did a good job on a project and their boss rewarded them with baseball tickets, but one employee was given three tickets while the other was given only one. It would be fair for the boss to give two additional baseball tickets to the employee who had received fewer tickets, at a later date, but it would not be fair for the boss to give the employee two new company cars to address the ticket-based inequality. Giving unequally is fair in the former case because it corrects the past unequal distribution, but not in the latter case because this would over-correct the past inequality, actually increasing the overall inequality. The reason for this difference is that a car is substantially more valuable than a ticket to a baseball game. Adults recognize that all inequalities are not equal; they will share unequally themselves in order to correct or minimize inequality between others (Dawes et al., 2007; Xiao and Bicchieri, 2010), and do so by taking resource value into account (Cook and Hegtvedt, 1983; Brown, 1984; Gurven, 2006).

One goal of the current research is to examine whether children focus on the norm of sharing equally themselves, or on trying to make the overall distribution of resources equal by correcting existent inequalities. If they do correct previous inequalities in order to equate outcomes, a second goal of the current research is to investigate whether children take resource value into account when trying to minimize inequality between others. Do they correct inequalities by trying to make the count of resources equal—giving cars to make up for unfair ticket-giving—or do they attempt to make the overall value of the resources as equal as possible?

We know that children are biased toward equal distribution of resources, but there has been very little research on how children respond to existent inequalities. Research with infants using looking time measures suggests that by the second year of life infants expect resources to be distributed equally between two agents, as long as both agents are highlighted as possible recipients (Geraci and Surian, 2011; Schmidt and Sommerville, 2011; Sloane et al., 2012; Sommerville et al., 2012). By 3 to 4 years of age, children themselves share resources equally with third parties when they can (Olson and Spelke, 2008), and are reluctant to share unequally when an equal option is possible, even when they know that one recipient was mean in a previous interaction (Kenward and Dahl, 2011), or did more work than the other recipient (Baumard et al., 2012). Indeed, if an equal option is possible, children default to giving equally rather than based on merit until they are 6 years old (e.g., Lerner, 1974; Hook and Cook, 1979; Sigelman and Waitzman, 1991). We also know that by age six children dislike those who share unequally (Shaw et al., 2012). While we know that by age six children will distribute resources unequally themselves based on merit and dislike those who share unequally, we do not know if children will distribute resources unequally themselves to correct previous inequalities in service of making overall outcomes equal [the one exception is Libby and Garrett (1974), but this experiment conflates correcting previous inequalities with reciprocity]. Because in the real world not everyone will always share equally, it is important to know how children respond to existent inequalities. Do they simply maintain these inequalities by trying to share equally themselves, or do they attempt to correct these inequalities by sharing unequally? This is one question the current research will address.

We also know very little about how children divide resources that differ in value, despite the fact that many forms of exchange involve resources that are not equal in value. Most research on children's and even adults' equality concerns has focused on decisions that involve distributing a single type of resource, for example, distributing a sum of money or a sum of cookies rather than having a person divide some money and some cookies (for reviews, see Damon, 1977; Walster et al., 1978; Hook and Cook, 1979). Using a single resource is useful because it minimizes random variation in preference for different resources. However, using a single resource fails to capture an important aspect of real world exchanges, and certainly does not capture exchanges in pre-agrarian human societies since fungible currency is a relatively recent human invention (Burgoyne and Lea, 2006). Indeed, there are very few fungible resources—one cannot substitute a unit of iPod for a unit of yoyo. Equally sharing non-fungible resources requires some recognition of value: how many units of resource A could be exchanged for how many units of resource B (Fiske, 1992). Being able to equate the value of varied resources is not only important in modern societies, but is also important for bartering and food sharing in smaller hunter-gatherer societies—even within the same animal carcass, different regions of the animal vary in value (Hill and Kaplan, 1993; Gurven, 2004). Since real world exchanges require some understanding of value, it is important to understand how value influences children's intuitions about how to share with others.

What little work that has been done on children's understanding of value has examined how children's preference for a resource influences their willingness to give resources to another person. We know that children, like adults, demonstrate preferences for some goods over others (Harbaugh et al., 2001), and use their preferences when deciding how many resources to give away (Birch and Billman, 1986). Blake and Rand (2010) had 3 to 6-year-old children identify their least favorite sticker (Low Value) and their most favorite sticker (High Value). They found that children shared more of their least favorite sticker than their most favorite sticker [for a similar effect in adults, see Novakova and Flegr (2013)]. This result importantly demonstrates that children weigh the value of being generous or fair against the personal value they place on a resource. However, this result does not tell us if children take value into account when deciding how to minimize inequality between others. Doing so requires children to use their preferences or some other information to make guesses about what others would want, and try to minimize inequality between others based on this dimension. This assumes that children have a belief about the value of resources that goes beyond their own idiosyncratic preferences. If children really understand value, they should distribute resources to two third parties in a way that minimizes the discrepancy in the value of the resources that the two third parties have. In terms of the example above, if children believe that it is inappropriate to give someone two company cars to correct a past inequality of two baseball tickets, this would suggest that they understand value and use it to guide their judgments.

#### **EXPERIMENT 1**

In Experiment 1, we first investigated whether children correct existent inequalities in order to minimize inequality in outcomes. Children were asked to share two resources with two non-present recipients who had already received resources from the experimenter. The experimenter gave three resources to one of the recipients and one to the other recipient. If children try to make outcomes equal, then they should give both erasers to the recipient with fewer resources (giving unequally but correcting the inequality) rather than giving one to each recipient (maintaining the inequality by giving equally themselves). We investigated this question in 6- to 8-year-old children because past research has demonstrated that it is at this age that children become comfortable sharing unequally with third parties, at least based on merit (e.g., Hook and Cook, 1979; Shaw and Olson, 2012).

If children do give more resources to those who currently have fewer resources, it would be unclear if they do so in order to keep the count of the resources equal, or in order to keep the value of the resources equal. To investigate if children use value to determine how to equalize outcomes, we included two conditions in which children were sharing resources that were slightly more valuable (jar of Play-Doh, Medium Value Condition) or much more valuable (\$20 bill, High Value Condition) than the resources that were initially shared unequally by the experimenter. If children want to keep the count of resources equal, then children should respond similarly in all conditions, by giving two resources to the recipient with fewer resources and thus equalizing the count of resources. If instead children care about keeping the value of resources as equal as possible, then, as the value of the resources to be shared increases, children should become increasingly likely to share equally themselves by giving one resource to each recipient. We investigated these questions in this study.

#### **METHODS**

#### *Participants*

Participants included 84 children aged 6 to 8 years old. Of these participants, 28 were in the Equal Value Condition (*M* = 7 years, 6 months, *SD* = 12 months; 15 females), 28 were in the Medium Value Condition (*M* = 7 years, 4 months, *SD* = 11 months; 8 females), and 28 were in the High Value Condition (*M* = 7 years, 0 month, *SD* = 12 months; 15 females).

#### *Procedure*

Two buckets were placed in front of the participant and the experimenter said (modeled on Shaw and Olson's (2012) method):

Thanks for playing this game with me. Earlier today, two kids named Mark and Dan did a great job cleaning up their room and we want to give them erasers as a prize. The problem is I don't know how much to give them; can you help me with that? We are going to decide how many erasers Mark and Dan will get. Mark's erasers go in this bucket and Dan's erasers go in this bucket. We have six erasers. I am going to give these four erasers and you are going to give these two erasers. I'll go first. We have one for Mark, one for Dan. One for Mark, and one more for Mark. Now it's your turn; here are two erasers. Give them however you want.

Each time Mark or Dan's name was used, the experimenter pointed to the corresponding bucket. During the allocation phase of the task, the experimenter placed an eraser into the corresponding bucket when noting who was receiving the eraser (Mark or Dan). The erasers were colorful and shaped like fun things children like, such as turtles, sports balls, and ice cream cones, and have been used in previous research on decision-making in children (Shaw and Olson, 2012; Shaw et al., in press). After the allocation phase, children were handed two erasers that they could place in the two buckets however they wanted. Children were always given two resources to distribute, and distributed until there were no resources remaining. On half of the trials Mark's bucket was on the left, and on half of the trials Mark's bucket was on the right.

In order to investigate the influence of value on children's decisions, we had two additional conditions in which children distributed resources that were slightly more valuable (Medium Value Condition) or much more valuable (High Value Condition) than the resources (erasers) that were shared unequally by the experimenter; all other aspects of the design of these conditions was identical to the condition described above. The slightly more valuable object in the Medium Value Condition was a 3 oz jar of Play-Doh, and the much more valuable resource in the High Value Condition was a \$20 bill. Children could not make the value of resources equal in these conditions, since both Play-Doh and a \$20 bill are presumably worth much more than two erasers (for empirical verification that children see these items as more valuable than erasers, see Experiment 4), but they could ensure that the inequality did not increase. Specifically, they could give one high value resource to each recipient rather than giving two to the person with fewer low value resources, if they were interested in maintaining the smallest overall inequality.

#### **RESULTS**

Because no children chose the option of giving two erasers to the person with more erasers, we conducted analyses with just the two strategies that children used—sharing equally by giving one to each recipient, or giving two to the person with fewer resources. We first conducted a 3 <sup>×</sup> 2 Yates-corrected <sup>χ</sup><sup>2</sup> test on children's responses in the Equal, Medium, and High Value Conditions, which revealed a main effect of condition, χ2*(*2*, N* = 84*)* = 23*.*37, *p <* 0*.*001.

**with fewer resources in the Equal, Medium and High Value Conditions from Experiment 1.**

We then examined whether children's responses differed between pairs of conditions by conducting Yates-corrected χ<sup>2</sup> tests. A 2 <sup>×</sup> 2 Yates-corrected <sup>χ</sup><sup>2</sup> test revealed that children in the Equal Value Condition were more likely to give two erasers to the disadvantaged recipient than children in the High Value Condition, <sup>χ</sup>2*(*1*, <sup>N</sup>* <sup>=</sup> <sup>56</sup>*)* <sup>=</sup> <sup>23</sup>*.*17, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*001, and children in the Medium Value Condition, <sup>χ</sup>2*(*1*, <sup>N</sup>* <sup>=</sup> <sup>56</sup>*)* <sup>=</sup> <sup>4</sup>*.*29, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*<sup>038</sup> (see **Figure 1**). As the resources became more valuable than those involved in the original inequality, children were more likely to give one to each recipient than to give two to the recipient who had fewer resources. Next, we examined if children's responses differed in the Medium and High Value Conditions using a Yatescorrected χ<sup>2</sup> test, which revealed that children in High Value Condition were more likely to give one resource to each child than children in the Medium Value condition, <sup>χ</sup>2*(*1*, <sup>N</sup>* <sup>=</sup> <sup>56</sup>*)* <sup>=</sup> <sup>7</sup>*.*62, *p* = 0*.*006. Again, as the value of the resources increased, children shifted their preference from giving two to the recipient with fewer resources to giving one resource to each recipient.

We next conducted binomial tests to compare children's responses to chance responding. The binomial test on the Equal Value Condition revealed that children gave two to the recipient with fewer erasers (24 out of 28) more often than giving one to each recipient (4 out of 28), *p <* 0*.*001. This result indicates that children preferred to make the total amount of resources equal between recipients, rather than to give equally themselves, when all resources were of equal value. A binomial test on children's choices in the Medium Value condition revealed that children did not show a preference for how to distribute the medium value resources, with about half the children giving two to the recipient with fewer resources (16 out of 28), and about half of the children giving one to each recipient (12 out of 28), *p* = 0*.*572. However, the binomial test on the High Value Condition revealed that children gave one to each recipient (23 out of 28) more often than giving two to the recipient with fewer erasers (5 out of 28), *p* = 0*.*001.

#### **DISCUSSION**

Children corrected inequalities created by an experimenter, and did so by attempting to equate the value, not just the count, of the resources distributed. When children distributed resources that were equal in value to those shared unequally by the experimenter, children gave more resources to a recipient who had received fewer resources in order to correct the existent inequality. Children could have ignored the inequality that was created by the experimenter and simply focused on the norm of giving equally themselves, since we know from past research that children have a tendency to share equally with others (Damon, 1977). However, this result indicates that children can inhibit their tendency to give equally to others when they are confronted with someone who had received fewer resources previously. Rather than simply defaulting to giving one resource to each recipient, children wanted to ensure that both recipients received an equal number of resources, at least when those resources were of equal value.

We next asked whether, when children try to make outcomes equal, they try to simply make the count of resources equal, or whether they consider the value of the resources. Our results suggest that children do use value when deciding how to share. Children behaved differently when sharing resources that were much more valuable (e.g., \$20 bills) than the unequally shared erasers, giving the resources equally themselves rather than attempting to correct the past inequality. Perhaps children distributed the more valuable resources differently because they realized that giving the disadvantaged child two \$20 bills would actually make the outcome even more unequal, though now in the other recipient's favor.

Although we interpret these results as indicating that children minimize inequality between others by using value, this is not the only possible interpretation. One alternative possibility is that children were confused in the Medium and High Value Conditions because they were required to match distributions involving multiple resources. However, the fact that children differentiated between the Medium and High Value Conditions speaks against this alternative—the resources used in both the Medium and High Value Conditions were different from the resources that were distributed by the experimenter, yet children treated these two conditions differently, suggesting they used some sense of value to guide their decisions. However, this by itself does not provide enough evidence to rule out the possibility that children were confused in these conditions. Perhaps children were indeed confused in the Medium Value Condition, and only behaved differently in the High Value Condition because there is something special about money that makes children more likely to share equally or pay attention to the value of resources. Previous research with adults indicates that when people distribute money, as compared to other resources, they are likely to think in terms of market exchanges (DeVoe and Iyengar, 2010) and this may cause them to think about resources in terms of their value (Fiske, 1992). In Experiment 2, we controlled for this possibility by having children divide high value resources that were not money. We also attempted to further rule out the possibility that children responded differently in the higher value conditions because they were confused by having to distribute resources that were different from those shared by the experimenter. To do this, in Experiment 2, both conditions had children dividing resources that were different from the resources shared by the experimenter. However, in one condition children were sharing a higher value resource and in the other condition they were sharing a lower value resource. If children were merely confused in Experiment 1 by having to divide resources that were different than those shared by the experimenter, then they should respond similarly in both conditions of Experiment 2 because children are dividing different resources in both conditions. If instead, as we predict, children were using value to guide their decision of how to equate outcomes, then they should be less inclined to give one to each recipient when they are dividing a resource of lower value as compared to one of higher value.

# **EXPERIMENT 2**

# **METHODS**

#### *Participants*

Participants included 56 children aged 6 to 8 years old. Of these participants, 28 were in the Lower Value Condition (*M* = 7 years, 5 months, *SD* = 10 months; 13 females) and 28 were in the Higher Value Condition (*M* = 7 years, 1 month, *SD* = 8 months; 19 females).

#### *Procedure*

The procedure for Experiment 2 was very similar to that used in Experiment 1. Again the experimenter gave out four resources unequally, giving three to one recipient and one to the other, using the script described in Experiment 1. Then, the participant was told to share two resources with the two recipients. We did, however, make two changes. First, we used different resources in Experiment 2. In the Higher Value Condition, the experimenter gave out four lower value resources (four small fruit-flavored candies) and the participant gave out two higher value rewards (two full-sized chocolate candy bars). This method was similar to the Medium and High Value Conditions from Experiment 1, so we predicted a similar pattern of results—that participants would be less willing to give more resources to the person who had fewer resources. In the Lower Value Condition, the experimenter gave out high value resources (four full-sized chocolate candy bars), and the participant gave out two lower value rewards (two small fruit-flavored candies). If children's responses in the previous experiment were merely being driven by confusion about how to distribute a resource different than the one involved in the original inequality, or by money priming them to think about value, then they should respond at chance or give one lower value resource to each recipient as they did in the Medium and High Value Conditions from Experiment 1. However, if children try to equate value to minimize inequality of outcomes between others, then they should instead give two low value resources to the recipient who received fewer higher value resources. Giving more low value resources to the recipient with fewer resources would not make the distribution equal, but is the most equal option available to children. We deliberately used different types of candy because we wanted to ensure that children were responding to value, not merely thinking about the resources in terms of large and small quantities of the same resource (for empirical verification that children see the chocolate bars as more valuable than the small fruit candies, see Experiment 4).

A second change from Experiment 1 to Experiment 2 was that we now presented the resources on pieces of paper (5 × 8--) rather than placing them in buckets. We modified this aspect of the design to reduce the memory load required to complete this task.

#### **RESULTS**

Again, because very few children chose the option of giving two resources to the person with more resources (only one child who was in the Higher Value Condition and no children in the Lower Value Condition), we again conducted our analyses focusing on the strategies that children used—sharing equally themselves by giving one to each recipient, or giving two to the person with fewer resources<sup>1</sup> .A2 <sup>×</sup> 2 Yates-corrected <sup>χ</sup><sup>2</sup> test revealed that children in the Higher Value Condition were more likely to give one resource to each child than children in the Lower Value Condition, <sup>χ</sup>2*(*1*, <sup>N</sup>* <sup>=</sup> <sup>55</sup>*)* <sup>=</sup> <sup>6</sup>*.*28, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*013, see **Figure 2**. When the resources children shared were more valuable than the resources shared unequally by the experimenter, children shifted their preference from giving two to the person with fewer resources to giving one resource to each recipient.

We next conducted binomial tests to compare children's responses to chance responding. A binomial test on children's choices in the Higher Value Condition revealed that children did not show a preference, with about half the children giving two

<sup>1</sup>We analyze the results here without including the one child who gave the non-standard response of giving more to the person with more resources, but the pattern of results remain the same if we conservatively run the analyses counting this child as having given more to the person with fewer resources.

to the person with fewer resources (15 out of 27) and about half of the children giving one to each recipient (12 out of 27), *p* = 0*.*701. However, the binomial test on the Lower Value Condition revealed that children chose to give two resources to the recipient with fewer resources (25 out of 28) more often than giving one to each recipient (3 out of 28), *p <* 0*.*001.

## **DISCUSSION**

We again found that children correct previous inequalities in order to minimize inequalities in outcomes between recipients, and do so by using the value of the resources at their disposal. When children were presented with an inequality involving high value resources, but only had a few low value resources with which to address it, children gave two to the person with fewer resources since this was the best way to minimize inequality. However, when children were presented with an inequality involving low value resources, but only had a few high value resources with which to address it, children were much less likely to give two to the person with fewer resources. Importantly, in both conditions children were dividing resources that were different from the resources that were distributed by the experimenter, so the results cannot be explained by confusion involving the distribution of different resources (which was common to both conditions). In fact, the same resources were used in both conditions; what differed between conditions was which resource was distributed by the experimenter and which was distributed by the participant. These results suggest that children focus on trying to equalize outcomes, and that they do so by using the value of resources.

Although the results thus far are consistent with children using value to determine how to minimize inequality in outcomes, children could be using an even simpler heuristic—size of resource. In Experiments 1 and 2, the high value resource was physically larger than the low value resource, and so children may have been using resource size, not value, to guide their decisions. In Experiment 3, we dealt with this confound by matching the volume and surface area of the high and low value resources.

# **EXPERIMENT 3**

#### **METHODS**

#### *Participants*

Participants included 56 children aged 6 to 8 years old. Of these participants, 28 were in the Lower Value Condition (*M* = 7 years, 5.5 months, *SD* = 12 months; 13 females) and 28 were in the Higher Value Condition (*M* = 7 years, 3.5 months, *SD* = 11 months; 12 females).

#### *Procedure*

The procedure for Experiment 3 was the same as Experiment 2, except that we used different resources: chocolate bars, and pieces of cardboard cut to the same size as the chocolate bars. In the Lower Value Condition, the experimenter gave out high value resources (four chocolate bars; three to one recipient and one to the other) and the participant gave out two lower value rewards (two pieces of cardboard). In the Higher Value Condition, the experimenter gave out lower value resources (four pieces of cardboard; three to one recipient and one to the other) and the participant gave out two higher value rewards (two chocolate bars). If children in the previous experiments were trying to equate the volume or surface area of the resources, then children should behave similarly in the Higher and Lower Value Conditions here. However, if children in the previous experiments were trying to minimize inequality in outcomes by using value, then they should be more likely to share equally themselves by giving one resource to each recipient when distributing the higher value reward as opposed to the lower value reward (for empirical verification that children see the chocolate bars as more valuable than cardboard, see Experiment 4).

#### **RESULTS**

Again, because no children chose the option of giving more resources to the recipient with more resources, we conducted our analyses on children's two strategies of giving more to the person with fewer resources and giving one to each recipient. A Yates-corrected χ<sup>2</sup> test revealed that children in the Higher Value Condition were more likely to give one resource to each recipient than children in the Lower Value Condition,χ2*(*1*, <sup>N</sup>* <sup>=</sup> <sup>56</sup>*)* <sup>=</sup> <sup>6</sup>*.*3, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*012. As the value of the resource increased, children shifted their preference from giving two to the recipient with fewer resources to giving one resource to each recipient (see **Figure 2**).

We next conducted binomial tests to compare children's responses to chance responding. A binomial test on children's choices in the Higher Value Condition revealed that children did not show a preference, with about half the children giving two to the recipient with fewer resources (13 out of 28) and about half of the children giving one to each recipient (15 out of 27), *p* = 0*.*850. However, a binomial test on the Lower Value Condition revealed that children chose to give two to the recipient with fewer resources (23 out of 28) more often than giving one to each recipient (5 out of 28), *p <* 0*.*001.

#### **DISCUSSION**

These results again indicate that children are motivated to create equal outcomes, not just to give equally themselves, and that they use value, not just the volume or surface area of resources, to decide how to create equal outcomes for others. When a resource was of lower value than the resources involved in the original inequality, children gave more to the recipient who received fewer resources originally in order to correct the previous inequality. However, children were much less likely to give more to the recipient with fewer resources if the resources they were distributing were more valuable than those involved in original inequality, presumably because they understand that this would make things more unequal.

However, one limitation of Experiments 1 through 3 is that we did not have an empirical measurement of value. We deliberately chose resources that seemed more valuable to adults; however, we do not know if children actually think these resources are more valuable. In Experiment 4 we ask children explicitly about which items they think that another child would prefer and how many of the less preferred items they think one would need to trade in order to get the more preferred item.

One other open question from the previous experiments is whether children distributed high value resources differently then low value resources because they treat high value resources differently in general or because they noticed that the high value resources were more valuable than the originally distributed resources. Perhaps children just maintain the status quo by sharing equally when they are given certain resources to share, regardless of the value of resources shared by the experimenter. To examine this possibility in Experiment 4 we had children distribute the high value resource from Experiment 3 (a chocolate bar) in a situation in which equal sharing was not the option that minimized inequality—where an experimenter shared three chocolate bars with one recipient and one chocolate bar with the other. If children treat certain resources differently regardless of context, then they should give one chocolate bar to each recipient as they did in Experiment 2 and 3 when sharing chocolate bars. However, if what matters is the relationship between the value of resources already distributed and the resource children are sharing, then they should now give two chocolate bars to the person with fewer resources because this would minimize inequality between the two recipients.

### **EXPERIMENT 4**

#### **METHODS**

#### *Participants*

Participants included 28 children aged 6 to 8 years old (*M* = 7 years, 4 months, *SD* = 11 months; 13 females).

#### *Procedure*

The procedure for Experiment 4 was the same as Experiment 1 Equal Value Condition, except that the equal value resource was now the chocolate bars from Experiments 2 to 3 rather than erasers. That is, the experimenter gave out four chocolate bars, three to one recipient and one to the other, and the participant gave out two of the same kind of chocolate bar. In the previous experiments chocolate bars were treated as a high value resource in comparison to small fruit candies and cardboard. Therefore, if children are simply more inclined to maintain the status quo when distributing objectively valuable resources like chocolate bars, then we should see children giving one chocolate bar to each recipient as they had in Experiments 2 and 3. However, if what children are attempting to do is to equate value, then we should see them giving two candy bars to the recipient with fewer candy bars.

After completing the Equal Value Condition, children completed an explicit measure of value. We asked children to decide which resource they thought Mark would prefer, resource *X* or resource *Y*—which corresponded to the pairs of resources used in Experiments 1 through 3. Children were asked about the four resource pairs in the following order: eraser vs. Play-Doh, eraser vs. \$20 bill, chocolate bar vs. small fruit candy, chocolate bar vs. piece of cardboard (or the reverse order, counterbalanced between participants). The items were placed in front of children and children were asked, "Which do you think Mark would rather have?" Children indicated their choice by pointing at one of the two resources. After answering which one Mark would prefer, children were asked how many of the resource they did not choose (the one they thought Mark would not prefer) Mark would need to trade in order to get one of the chosen resource. The trading measure was designed to produce a rough estimate of how much more valuable children thought one resource was.

#### **RESULTS AND DISCUSSION**

Again, because no children chose the option of giving more resources to the recipient with more resources, we conducted our analyses on children's two strategies of giving more to the recipient with fewer resources and giving one to each recipient. We conducted binomial tests to compare children's responses to chance responding. A binomial test on children's choices in the equal value condition revealed that that children chose to give two to the recipient with fewer resources (24 out of 28) more often than giving one to each recipient (4 out of 28), *p <* 0*.*001. This result indicates that children do not simply maintain the status quo when distributing chocolate bars, a high value resource from Experiment 2 and 3. Children are perfectly willing to share unequally by giving more to the person with fewer resources, disrupting the status quo, when the chocolate bars are the same value as the resource shared unequally by the experimenter.

We next conducted binomial tests on children's responses to which resource Mark would prefer. Children thought that Mark would prefer: a jar of Play-Doh to an eraser (26 out of 28), *p <* 0*.*001, a \$20 bill to an eraser (28 out of 28), *p <* 0*.*001; a chocolate bar to a piece of cardboard (27 out of 28), *p <* 0*.*001; and a chocolate bar to a small fruit candy (24 out of 28), *p <* 0*.*001. These results indicate that our intuitions about children's valuation of these objects was correct—children thought that the items we labeled as higher value resources in Experiments 1 through 3 were in fact more preferred than the items we labeled as lower value resources. We next examined how many low value resources children thought one would need to trade to get one of the high value resources. We found that children thought that one would need to trade, on average, 8.5 erasers to get one jar of Play-Doh, 17 erasers to get one \$20 bill, 10 small fruit candies to get one chocolate bar, and 26.5 pieces of cardboard to get one chocolate bar. It is worth noting that in order to reduce skew on the cardboard/chocolate item we had to code five of the children's responses as "100" because they either gave very large numbers (two children said one thousand and one child said one billion) or they stated that no amount of the cardboard could be traded for a chocolate bar (*N* = 2).

## **GENERAL DISCUSSION**

These experiments demonstrated that 6- to 8-year-olds are more concerned with making the outcome of a resource distribution equal than with giving equally themselves. They also demonstrated that children consider value when responding to inequalities. Experiment 1 showed that children will give unequally themselves in order to minimize inequality of outcome. Children gave two resources to the recipient with fewer resources so that both recipients would have three resources, rather than giving equally themselves and maintaining the inequality. This result is consistent with past research demonstrating that children will give unequally in some circumstances, such as when others have done more work (Damon, 1977; Sigelman and Waitzman, 1991); however, these results are the first direct demonstration that children will correct unequal distributions by sharing unequally with others.

Experiments 1 and 2 also investigated what measure children use to determine how best to minimize inequality. These experiments illustrated that children use the *value* of resources, not just the count, to minimize inequality between others. They did not opt to give one person two high value resources (equalizing the count of resources) to correct past unequal sharing of a low value resource, and instead were more likely to give one high value resource to each recipient. Experiment 3 further confirmed that children were using value, not resource size, as a guide for how to share resources with others. In Experiment 4 children were asked to make explicit judgments about which resources they thought another child would prefer. These explicit judgments provided an empirical confirmation that our high value resources were actually valued more highly than resources that we labeled as lower value resources.

It is worth noting that while children became less likely to give both resources to the recipient with fewer resources as the value of the new resources increased, in Experiments 2 and 3 about half the participants still attempted to equate resource count rather than resource value when sharing the high value resource (large chocolate bar). It is unclear why children gave mixed responses in this case, though there are several possibilities. One possibility is that some children placed different value on the items they were asked to share. If children thought the chocolate bars were about as valuable as the fruit candies or cardboard, then it would be unsurprising that they attempted to equate count rather than value. However, a more likely possibility is that children did not know which norm to apply to this situation and so were forced to choose between two conflicting norms: should I equalize the count or value of the resources? This conflict in norms may have made children confused about what to do and led to their chance responding when distributing the higher value rewards. However, what is important about these results is that children did differentiate between distributing resources that had higher and lower value than the original inequality, suggesting that at least some children take resource value into account when deciding how to minimize inequality in outcomes between others.

The current findings are interesting to consider in light of recent work demonstrating that children are fair partly in order to signal to others that they are fair. Shaw et al. (in press) found that 6- to 8-year-old children were very fair when the other option was to appear unfair to an experimenter (see also Blake and McAuliffe, 2011), but were considerably less fair when they did not risk appearing unfair. The paradigm developed here could be used to investigate if children will use ambiguous norms, like those investigated in the current studies, to their advantage in order to appear fair while getting more for themselves. For example, imagine we repeated the High Value Condition from Experiment 1, but the participant was the recipient who received more erasers (low value rewards). We could then ask the participant to distribute two \$20 bills (high value rewards) between him- or herself and another recipient. Here, it seems likely that children would do the fair thing and give one \$20 bill to themselves and one to the other recipient because they would have no way to justify giving two \$20 bills to themselves. Next, imagine we repeated the High Value Condition from Experiment 1, but that this time the participant received fewer erasers than the other recipient and was again asked to distribute two \$20 bills. In this case, children might be more likely to take the two \$20 bills for themselves since they would have the plausible justification that they were simply trying to equate the count of resources. These results would demonstrate that children can use different norms and plausible deniability to justify their own selfishness, just as adults do (Dana et al., 2007; Andreoni and Bernheim, 2009). Future research should investigate this possibility.

The results of our experiments demonstrate that the value of resources influences children's sharing behavior, but they do not address how children determine the value of resources in the first place. The first strategy that children likely use to determine value is to simply use their own preferences as a guide for how to share with others. That is, they know what they like, and think that the things they like are valuable and that the things they dislike are not valuable. This strategy is likely a large part of children's early understanding of value, but as they get older they may use more sophisticated variables to determine how resources are valued. One possibility is that children use some aggregate sense of others' preference for a resource, analogous to the adult concept of demand—recognizing that the more others want a resource, the more valuable it is (Baumol, 1972). A second possibility is that children use resource scarcity to determine value, recognizing, as adults do, that rare things are more valuable than things that are commonly available (Lynn, 1991). Yet another possibility is that children use effort expended to obtain a resource to determine resource value; all else being equal, they may assume that if a person worked harder to make or obtain a resource, that resource is more valuable. It is likely that children, like adults (Baumol, 1972), use some combination of these factors to determine a resource's value, and as they get older they incorporate more of these sophisticated principles to determine resource value.

#### **REFERENCES**


*Cognition* 120, 215–224. doi: 10.1016/j.cognition.2011.04.006


Now that we know that children can use value to guide their equality judgments, we can investigate whether or not children use value in other domains such as trade. Trade is ubiquitous in modern society and simpler forms of bartering were also very prevalent before the advent of currency (Fagan, 1969; Hill and Kaplan, 1993). Being able to equate resources that differ in value is essential for participating in trade, both modern forms of trade between nations and simpler forms of bartering (Krugman, 1979; Hill and Kaplan, 1993). Without some sense of value it would be impossible to determine when one should and should not trade with another person. Anecdotally, children trade a number of resources, from baseball cards to lunchtime snacks to Silly Bands—children seem well acquainted with trade. Yet it is unclear whether these trades are simply based on personal preference or on some understanding of resource value. How do children reconcile others' personal preferences with the objective value of resources? Is subjective or objective value given more weight? Can children capture gains from trade (Krugman, 1979)? Understanding children's early notions of trade may provide some insight into how they grow into adults who perform more sophisticated exchanges.

Despite remaining questions, the current research demonstrates that children do not treat all inequalities equally—they use resource value, rather than just resource count, when deciding how to share with others.

#### **ACKNOWLEDGMENTS**

We would like to thank Nina Slywotzky, Anna Merrill, Melanie Fox, Danielle DeLee, Matt Choy, Zoe Liberman, Suzanne Horowitz, Alia Martin, Kelcey Wilson, and Alex Chituc for assistance running the participants in these studies. This research was supported by a grant from the University of Chicago's ARETE Initiative/A New Science of Virtue Program.


*Q. J. Econ.* 114, 817–868. doi: 10.1162/003355399556151


kids: on the development of rational choice behavior. *Am. Econ. Rev.* 91, 1539–1545. doi: 10.1257/aer.91.5.1539


*Psychol. Market.* 8, 43–57. doi: 10.1002/mar.4220080105


*Behav.* 33, 736–745. doi: 10.1016/ j.evolhumbehav.2012.06.001


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 11 June 2013; published online: 19 July 2013.*

*Citation: Shaw A and Olson KR (2013) All inequality is not equal: children correct inequalities using resource value. Front. Psychol. 4:393. doi: 10.3389/fpsyg. 2013.00393*

*This article was submitted to Frontiers in Emotion Science, a specialty of Frontiers in Psychology.*

*Copyright © 2013 Shaw and Olson. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Fairness overrides reputation: the importance of fairness considerations in altruistic cooperation

# *Sule Güney ¸ \* and Ben R. Newell*

*School of Psychology, University of New South Wales, Sydney, NSW, Australia*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Claudia Civai, University of Minnesota, USA Sunhae Sul, Korea University, South Korea Ozan Aksoy, Utrecht University, Netherlands*

#### *\*Correspondence:*

*Sule Güney, School of Psychology, ¸ University of New South Wales, Mathews Building, Botany Street, Sydney, NSW 2052, Australia e-mail: s.guney@unsw.edu.au*

Behavioral findings in several strategic games indicate that people punish others if they think they are being treated unequally, even at the cost of minimizing their own material payoff. We investigated the primary driving force behind such non-self-regarding behavior, so-called, altruistic cooperation. In all of our studies, a mini ultimatum game was played either one-shot (in Experiment 1a and 1b) or repeatedly (Experiment 2), and rejections of inequitable distribution were taken as a measure of altruistic cooperation. In Experiment 1a, we replicated previous findings indicating that the key mechanism contributing to the emergence of altruistic cooperation is fairness considerations. In Experiment 1b, we delved into the relative importance of two aspects of fairness considerations (i.e., outcome fairness and intentions) and showed that both aspects were effective in determining the level of altruistic cooperation, with the contribution of intentions being more important. In Experiment 2, we investigated the effect of the opportunity for reputation building and future interaction on altruistic cooperation. We found that these factors became influential only when fairness considerations were weakened, particularly, as a result of the removal of the possible intentions behind an offer.

**Keywords: altruistic cooperation, mini ultimatum game, fairness, reputation building, future interaction, intentions**

# **INTRODUCTION**

Human altruistic cooperation presents a puzzle from the perspectives of both the standard economic models of the "self-interested actor" and the evolutionary models of the "self-regarding individual" because it involves some characteristics that are difficult to reconcile with the predictions of standard game theoretical and evolutionary analyses. One form of altruistic cooperation is to reward cooperators (i.e., costly rewarding) and to punish norm violators (i.e., costly punishment) at a personal cost, even though the probability that this cost will be repaid (either by third parties or by that specific agent in the future) is very low (Gintis et al., 2003) 1 .

Evidence for the existence of altruistic cooperation largely comes from laboratory experiments in which the respective behavioral pattern has been observed through economic games. One of the best-known economic games used to demonstrate altruistic cooperation, particularly costly punishment, is the Ultimatum Game (UG) (Güth et al., 1982). In this game two players are presented with a sum of money; one of them is assigned to the role of Proposer while the other one to the role of Responder. The Proposer is asked to offer any portion of the money to the Responder. If the Responder accepts the amount offered, the money is distributed in accordance with the proposal. If the Responder rejects the offer, both get nothing.

According to standard economic theory of self-interest, a rational Proposer offers the minimum possible amount, and a rational Responder never rejects any amount unless it is zero (Binmore, 2007). The underlying assumption in this prediction is that both parties care only about how much money they get. However, the vast majority of experimental studies has shown that the modal offers by the Proposers lie between 40–50% of the total amount and the Responders frequently reject offers below 25% (Güth et al., 1982; Roth, 1995; Henrich et al., 2005). This pattern of results has been replicated cross-culturally (Henrich et al., 2005) and shown to be robust with large stakes (Cameron, 1999).

The experiments reported here aimed to investigate the differential contributions of fairness considerations and perceived opportunity of reputation building (RB)/future interaction to the emergence of costly punishment as a form of altruistic cooperation in experimental contexts.

Some researchers argue that the underlying mechanism of such non self-regarding behaviors in the UG (i.e., high offers by the Proposers and frequent rejections by the Responders) is not only to get as much money as possible, but also to maintain fairness norms among players (Fehr and Gachter, 2002; Gintis et al., 2003). In other words, the players have a preference for fairness, along with the preference for material benefits (Falk et al., 2008). In fact, the motivation behind the Proposers' high offers can be explained with or without the involvement of fairness considerations: they simply may not want to offer an amount that can be easily turned down by the Responder, so they are willing to distribute the money in a relatively fair way. Thus, the Proposers' main concern still might be getting as much as possible in the end, rather than treating the Responders fairly (Declerck et al.,

<sup>1</sup>We acknowledge that there are other forms of altruistic cooperation, such as cooperation in public goods games without any involvement of punishment of rewarding. However, the main interest of the current studies is costly punishment in Ultimatum bargaining games as a form of altruistic cooperation.

2009). However, for Responders, the role of fairness concerns is more apparent and must be stronger because they seem to accept ending up with nothing rather than being treated unfairly. Even though the Responders could have been better off by accepting any amount offered, they prefer to punish the Proposer's unfairness, at a cost to themselves. This pattern of response indicates that the Responders engage in costly punishment in response to the unfairness of the *outcome* proposed by the Proposer<sup>2</sup> .

A special version of UG has been used to demonstrate how much the Responders care about (un)fair *intentions* of the Proposers. The structure of the so-called Mini UG (see **Table 1**) is the same as the standard UG, with an exception: the Proposer is again asked to distribute an amount of money but unlike the standard UG, only in one of two ways. Both players participate in four consecutive Mini UGs, and throughout all these games one way of distribution is always fixed while the alternative distribution is always different across games. The fixed distribution is a relatively inequitable one (i.e., the Proposer can take \$8 for himself, and offer \$2 to the Responder, see **Table 1**).

However, the available alternative distribution varies in terms of the outcome fairness, sometimes yielding a more equitable outcome (i.e., the Proposer can take \$5 for himself, and offer \$5 to the Responder, see **Table 1**), and sometimes yielding an even more unequal outcome (i.e., the Proposer can take \$10 for himself, and offer \$0 to the Responder, see **Table 1**). Under the standard assumptions, rejection rates for the fixed distribution (8/2) are expected to be the same regardless of its alternatives, as its monetary value stays unchanged across games (Falk et al., 2003a). However, this particular distribution was rejected much more frequently when the Proposer intentionally ignored the more equitable alternative distribution [i.e., the (5/5) distribution] than when he ignored the more unequal alternative distribution [i.e., the (10/0) distribution] (Falk et al., 2003a; Sutter, 2007). Thus, the rejection decisions made by the Responders seem not to be determined by the absolute amount of the offer (i.e., \$2), but by whether the offer is seen as relatively unfair [i.e., in comparison to (5/5) split] or fair [i.e., in comparison to (10/0) split]. [See **Table 1** for the perceived fairness of the fixed distribution (8/2)

## across four games]. These findings indicate that the Responders punish the unfairness of the Proposers by rejecting an amount of money in one case and appreciate the fairness of the Proposer by accepting the very same amount in another case. It has been argued therefore that fairness considerations must be the underlying motive behind altruistic cooperation, especially in the context of costly punishment (Fehr and Gachter, 2002; Gintis et al., 2003).

Although the importance of fairness considerations in such bargaining games has been widely accepted, the real reasons for altruistic cooperation (i.e., the Responders' rejection/acceptance behaviors in the UG) have been a source of much debate (Declerck et al., 2009). As mentioned earlier, by rejecting a nonzero offer, the Responders seem to engage in actions that are opposite to their self-interest, in order to maintain the fairness norms between parties. Thus, fairness considerations seem to override the self-regarding/rational motives.

Confidence in such a conclusion mainly comes from the two critical features of the above-mentioned experiments: identities and the decision histories of both players are kept hidden (i.e., anonymous) and they will never meet again in another round (i.e., one-shot encounter). Anonymous and one-shot encounters eliminate the possibility of reputation building (henceforth, RB) and future interaction (henceforth, FI) respectively, as potential sources of this seemingly fairness-driven behavior (Fehr and Fischbacher, 2003). Involvement of *any* of these possibilities either RB *or* FI-would be especially critical in this context because the costly behavior obtained in these experiments could then be explained within the boundaries of self-regarding motives: it is rational and adaptive to reject unfair offers if the possibility of reencountering the same game partner in the future is high enough *or* if the possibility of building a reputation among other players is at stake. The underlying reason for this claim is that rejecting unfair offers protects the player from being offered with unequal distributions by the same game partner in the future or by third parties, and thus this behavior serves the player's self-interest (Burnham and Johnson, 2005; Hagen and Hammerstein, 2006).

This argument goes further in the direction that people engage in altruistic cooperation in one-shot and anonymous encounters simply because they confuse the experimental settings with the more familiar environments where interactions are normally repeated and non-anonymous (Burnham and Johnson, 2005). In fact, the participants might still be responding to implicit cues suggesting that future interaction is possible or that their reputation is at stake. One finding that supports this

#### **Table 1 | General structure of Mini Ultimatum Games.**


*\*The numbers in the parentheses denote how much the Proposer could get/how much the Responder could get.*

*\*\*The Proposer seems to have an excuse for offering the more inequitable distribution (8/2), because otherwise he would be unfair to himself [i.e., by offering the (2/8) distribution, he would give 8 to the Responder, and take 2 himself].*

<sup>2</sup>We argue that rejections of *inequitable distributions* in UGs can be interpreted as costly punishment because rejection of any non-zero offer (even any unevenly distributed offer) is (1) costly to the Responder himself because as a results he ends up with a zero outcome, and (2) a form of punishment to the Proposer's unfairness as the Proposer also gets nothing when a rejection is made.

interpretation is that the presence of eyespots on the computer desktop, which triggers the sense that participants are being watched, leads to increased generosity in another money allocation game (Haley and Fessler, 2005). Some other studies suggest that even the perception of being involved in a situation where FI and RB is possible triggers altruistic cooperation in one-shot, and anonymously played economic games (Kiyonari et al., 2000). Thus, behaving in an altruistically cooperative manner in the UGs might not solely result from the concern for the maintenance of fairness norms, but from the mis-perceived opportunity of RB and FI (Haley and Fessler, 2005; Bateson et al., 2006).

In the set of studies reported here, we aimed to investigate how important these two factors, namely fairness considerations (in Experiment 1a and 1b) and the possibility of RB and FI (in Experiment 2), are in the emergence of altruistic cooperation in general and costly punishment in particular. Experiments 1a and 1b were designed to understand the role of fairness considerations in costly punishment. Note that, as pointed out previously, fairness considerations have two major aspects, one being related to *outcomes* (Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000), and the other to *intentions* (Rabin, 1993; Dufwenberg and Kirchsteiger, 2004). Although, several economic models have been developed with a specific focus on outcome fairness (Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000), empirical evidence suggests that intentions are just as important as outcomes (Blount, 1995; Falk et al., 2008) for the maintenance of fairness norms. This is why we thought it was necessary to incorporate both of these important aspects of fairness consideration into our investigation and hence we used the Mini UG, instead of the standard UG, in all of our experiments.

Previous studies have already established the importance of intentions behind an action (i.e., offer) in the Mini UG: the (8/2) distribution is rejected at different levels depending on whether the alternative distributions are perceived as fair or not (i.e., highest rejections observed when the alternative was more equitable). However, findings diverge in terms of rejection rates of the (8/2) distribution when the alternative distribution was more inequitable. More specifically, 9% of the Responders rejected the (8/2) distribution in the (10/0) game in Falk et al.'s (2003a) study whereas almost 28% rejected it in Sutter's (2007) study. Considering these differences in previous findings, Experiment 1a was conducted to re-establish the basic phenomenon observed in the Mini UG (presented in **Table 1**). We found it preferable to observe the standard rate of rejections in all Mini UGs in our own subject pool first, in order to provide a standardized baseline before incorporating the subsequent manipulations (Experiments 1b, 2) (and potential implications to be drawn from these manipulations).

Experiment 1b was designed to clarify the relative impact of these two aspects of fairness considerations in the Mini UG. Two special features of this specific version of UG enable us to separate the effect of intentions from that of outcomes (Falk et al., 2008): the Proposer has two available options to distribute the allocated money, with one option always being more equitable or yielding a fairer outcome (compared to the other option). Importantly, the choice of one distribution over the other is under the Proposer's full control [except for the (8/2) game, see **Table 1**]. In order to differentiate the effect of intentions from the effect of outcome fairness, we removed the latter feature from the Mini UG and thus made any potential attribution of intentions impossible, but kept the former and thus made the evaluation of outcome fairness possible. If the rejections of the (8/2) distribution are primarily determined in response to the (unfair) intentions of the Proposer, then we should not obtain any differences in these rejections rates across the games because the intentions behind the offers are not assessable. However, it has been already shown that the Responders react to the fairness of outcomes as well (Blount, 1995; Falk et al., 2008). Thus, different rejection rates among different Mini UGs were expected but this manipulation would enable us to examine if these differences would be as strong as those observed in Experiment 1a where intentions were assessable.

In Experiment 2, we aimed to understand the combined effect of the *real* possibility of RB and FI in the Mini UG. The main reason for testing the combined effect of RB and FI was that in the above-mentioned studies demonstrating fairness driven responses (i.e., different rates of rejection of an inequitable distribution across Mini UGs), the features of "one-shot-interaction" and "anonymity" are inseparable. Therefore, it is difficult to identify whether the obtained responses could actually be the product of the (mis)perception of one-shot encounters as repeated (and thus players behave as if re-encountering the same game partner in the future is possible in order to maximize their material pay-offs) *or* that of the (mis)perception of anonymous encounters as non-anonymous (and thus Responders behave as if building a reputation among other players is possible in order to maximize their material pay-offs). Thus, incorporation of both possibilities of RB and FI through repeated and non-anonymous game play would make the two previously mentioned explanations (fairness-driven responses via one-shot/anonymous encounter vs. self-regarding responses via misperceived one-shot/anonymous encounter) commensurable. A second and even a more explicit reason was that the possibility of RB and FI are highly interrelated (i.e., repeated encounters with the same partner, by default, bring along the opportunity of building reputation as each player would know what the other player has done so far).

We predicted that if the main reason behind the rejections in one-shot and anonymously played games is the *misperceived* possibility of RB or FI, then an increase in the level of altruistic cooperation should be expected when the actual possibility of RB and FI is added to the context. Although such an additional effect of the possibility of RB and FI has not been investigated in the Mini UG, there are two main reasons for expecting such an increase. First, the importance given to equality is expected to be elevated (Rottemberg, 2008) because the equality norm (i.e., distributing the allocated money evenly) is strengthened in presence of the possibility of RB and FI (Hertel et al., 2002). Second and more importantly, the sanctions inflicted upon the unfairness of a game partner through altruistic cooperation might be considered as an effective tool for maximizing future gains (Kiyonari et al., 2000).

The structure of Mini UG allows us to examine how the possibility of RB and FI, along with the fairness concerns, contributes to the Responders' rejections especially when costly punishment is expected to take place (i.e., when the alternative offer yielded a more equitable distribution). In addition, in the Mini UG, there is one special game [the (8/2) game, see **Table 1**] in which the Proposer has no choice, but to offer the fixed amount. This particular case would enable us to detect the sole effect of the possibility of RB and FI on the Responders' decisions when an unequal distribution was offered without any (un)fair intentions of the Proposer involved. For all these reasons, to the best of our knowledge, Experiment 2 is the first attempt to understand the effect of the real possibility of RB and FI on costly punishment, particularly in the presence and absence of Proposer's intentions.

## **EXPERIMENT 1A**

We expected the rejection rate of the (8/2) distribution to be different across different Mini UGs. More specifically, the highest rejection rate was expected to be in the (5/5) game. In addition we expected to find statistically significant differences between the rejection rates of the (8/2) distribution in the (5/5) and the (10/0) games.

### **METHOD**

#### *Participants*

Fifty first year psychology students (*M* age = 19.5, 36 female) at UNSW participated in the experiment as a part of their course requirement, and were informed that they would be paid, contingent on the outcome of their choices. UNSW Human Research Ethics Advisory Panel approved the study.

#### *Procedure*

There were 10 experimental sessions in total, and five participants were tested at a time in each experimental session. Participants were seated in separate rooms and their identities were kept hidden throughout the whole experiment. All participants played the Mini UG as the Responders since our main interest was to see whether we would be able to replicate the choice pattern of the Responders obtained in previous studies (i.e., Falk et al., 2003a). However, each participant was told that only one participant in each group of five would be assigned to the Responder role and that the rest would be playing as Proposers. This procedure made them believe that the offer in each game would come from an actual but different participant (Proposer) rather than from the computer. The offers made by the computer mimicked the actual rate of proposals offered by real Proposers in the study of Falk et al. (2003a). For instance, in that study, the (8/2) distribution was offered by 31% of the Proposers in the (5/5) game, and 73% in the (2/8) game. Thus, the Responders in Experiment 1a were offered the (8/2) distribution with the probability of 0.31 in the (5/5) game, and that of 0.73 in the (2/8) game. The participants played the games for real money, but currency was defined as Monetary Unit (MU), where 1 MU was equal to 0.5 AUD. The experiment was conducted and run with the Runtime Revolution Software.

#### *Design*

The Responders participated in all four Mini UGs presented in **Table 1**. They were asked to indicate their acceptance/rejection decisions for each of the two possible distributions in each game before hearing the actual distribution offered [see Falk et al. (2003a) for further information regarding this strategy method]. For example, in the (10/0) game, the Responders were asked whether they would accept or reject if the Proposer offered them the (10/0) distribution instead of (8/2); and they were subsequently asked whether they would accept or reject if the Proposer offered the (8/2) distribution instead of (10/0). If the game was (8/2), they were simply asked what they would do if the Proposer had no choice but to offer the (8/2) distribution. Once the Responders indicated their rejection/acceptance decision for each possible distribution, they simply moved on to the next game. After the completion of all four games, the Responders were informed about the overall outcomes and debriefed about the real set-up of the experiment (i.e., the offers were not made by actual proposers). The presentation order of the Mini UGs and that of the possible distributions in each game were randomized.

#### **RESULTS**

**Table 2** shows the rejection rates of (8/2) distribution in different games. The main pattern observed in the previous studies (i.e., Falk et al., 2003a; Sutter, 2007) was replicated in our participant pool. To test the overall rejection rate differences across four games, we ran Cochran's *Q*-test. The test confirmed that the rejection rates of the (8/2) distribution were significantly different across four games (*p <* 0*.*0001). The rejection rate of the (8/2) distribution in the (5/5) game was the highest among four games. McNemar change tests were performed for the pairwise comparisons and they showed that the rejection rate in the (5/5) game was significantly higher than that of the (10/0) (*p <* 0*.*0001). The rejection rate of the (8/2) distribution was also significantly higher in the (5/5) game than in the (2/8) and the (8/2) games, *p* = 0*.*049, and *p <* 0*.*0001, respectively. In addition, the differences between the (2/8) and the (8/2) games, and the (2/8) and the (10/0) games were significant, *p* = 0*.*001 and *p* = 0*.*004, respectively. These results confirmed the previous findings that the rejections to an (unfair) offer were not determined by the absolute amount of money, but by how fair or unfair that offer was perceived in comparison to the other available offers<sup>3</sup> .

## **EXPERIMENT 1B**

Our main manipulation in this experiment was to eliminate the possibility of any attributions to intentions of the Proposer. To do so, the participants were informed that there were two distributions to be offered but that the actual offer would be determined by a random mechanism (Blount, 1995). Thus, the decision was not under the Proposer's full control, and therefore, it was impossible to evaluate the intentions behind the offer (Falk et al., 2008). When the fairness of intentions cannot be evaluated, the response should then only be determined by the outcome fairness if fairness considerations are the underlying force behind the Responder's responses. Thus, we expected that the rejection rates of the (8/2) distribution would still vary depending only on the relative fairness of the alternative outcomes but the

<sup>3</sup>The rejection rates for the alternative distributions (5/5), (2/8), and (10/0) were 2%, 6%, and 82%, respectively in Experiment 1a.



*R1, R2, R3, and R4 correspond to Round 1, Round 2, Round 3, and Round 4 of Experiment 2, respectively. Percentages reported for the row "average" are the rejection rates collapsed across rounds of Experiment 2.*

differences across games should not be as large as they were in Experiment 1a [i.e., the rejection rate of the (8/2) distribution in the (5/5) game should not be as high as it was in Experiment 1 because now the alternative (5/5) offer was not intentionally ignored in the same way that the unequal (8/2) was not intentionally offered]. However, the rejection rates of the (8/2) distribution should still be the highest in the (5/5) game because the (5/5) distribution yields a more equitable *outcome* for each player.

#### **METHOD**

Forty-five first year psychology students (*M* age = 19.8, 20 female) at UNSW participated in the experiment in return for a course credit. They were paid in accordance with the outcome of their decisions. The design and procedure of the experiment were almost the same as that of Experiment 1, with two exceptions. First, the participants were told that the offer of the Proposer would be determined by a coin flip [i.e., if it came up heads, the Proposer was going to offer the (8/2) distribution, otherwise the (5/5) distribution in the (5/5) game]. Second, the actual offer was indeed determined randomly [i.e., the (8/2) distribution was offered by the computer with probability of 0.5 in each game]. UNSW Human Research Ethics Advisory Panel approved the study.

#### **RESULTS**

The rejection rates of the (8/2) distribution across games are shown in **Table 2**. The rejection rates were only weakly different across four games, (*p* = 0*.*055, Cochran's *Q*-test). The only significant difference in terms of the rate of rejections for the (8/2) distribution within the games was between the (5/5) and the (10/0) games (*p* = 0*.*04, McNemar change tests) <sup>4</sup> . Crossexperimental analysis showed that the (8/2) distribution was rejected in the (5/5) game less frequently in Experiment 1b than in Experiment 1a, <sup>χ</sup><sup>2</sup> (1, *<sup>N</sup>* <sup>=</sup> 95) <sup>=</sup> 4.57, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*033.

The results demonstrated that when the intentions of the Proposer were not assessable, (1) rejection rates of the (8/2) distribution significantly decreased when its alternative was a more equitable one [i.e., (5/5)], and (2) the overall difference in the rejection rates of the (8/2) distribution across four games was not strongly significant. Even then, however, the rejection rates were still not identical across the four games. This is an indication that these rejections were shaped by how unfair the outcome was perceived in comparison to the outcomes yielded by the alternative distributions. Overall, these results support the idea that two aspects of fairness considerations are involved in the emergence of costly punishment, but the contribution of the intention aspect of fairness considerations seems greater, especially when the alternative distribution yields a more equitable outcome [i.e., the (5/5) game].

# **EXPERIMENT 2**

In order to test the effect of the real possibility of RB and FI we changed the structure of the Mini UG from being one-shot and anonymously played to being iterated and non-anonymously played. We predicted that the rejection rates of the (8/2) distribution in the Mini UG should be (1) even higher (than in the one-shot, anonymous version) when its alternative was the (5/5) distribution because it is adaptive to build the reputation that one is a tough bargainer who rejects unfair offers, and (2) even lower when its alternative was the (10/0) distribution because it is adaptive to give the message for future interactions that one is capable to discern and will reward fair intentions.

### **METHOD**

#### *Participants*

One hundred and ninety-two first year psychology students (*M* age = 19.76, 120 female) at UNSW participated in the experiment as a part of their course requirement and were informed that they would be paid depending on the outcome of their choices. Four participants were tested in each experimental session and there were 48 experimental sessions<sup>5</sup> in total. UNSW Human Research Ethics Advisory Panel approved the study.

#### *Instructions phase*

First, the participants were randomly allocated to their roles, (with 2 being Proposers, and the other 2 being Responders) and warned against revealing their allocated roles to the others. Individual players were then given detailed verbal instructions

<sup>4</sup>The alternative distributions (5/5) and (2/8) were rejected by only 1 Responder in each game while the (10/0) distribution by 78% of the Responders in Experiment 1b.

<sup>5</sup>We ran these 48 sessions in two separate blocks of 24 sessions. The second block of 24 sessions was conducted after one of the anonymous referees asked us to collect more data for this experiment. The demographics of participants in the two testing blocks was very similar and participants were tested by the same experimenter in the same laboratory. The data from these two blocks of sessions were first analysed separately to check whether patterns of responding were the same across sessions. This was confirmed via statistical analysis showing no significant differences in overall rates of rejection across the two session blocks. Thus data from both sessions was combined for the analyses presented in the body of the paper. Please see Footnote 8 for the details regarding the results of statistical analysis.

(along with a written instructions document) regarding the general structure of the game play, what their roles required them to do, and what the consequences of their accept/reject decisions would be. They were informed that they would play the game for more than one round with the same partner, and that their decision would be announced to other players right before they switched their partners. However, the players were not given any information about (1) the possible distributions available to the Proposer in Mini UGs [i.e., in order to eliminate the possibility of the (un)fairness of subsequent offers confounding the players' current decisions], (2) how many rounds they would play in total (i.e., in order to make the "shadow of the future" long enough), and (3) when exactly they would switch partners (i.e., in order to make the possibility of RB stronger). In order to eliminate a potential wealth effect, the participants were told that the overall amount that they would receive would be determined by a coin flip at the end of the experiment. If the coin toss came up heads, then they would get paid the amount that they earned in the first half of the experiment, and if tails, the amount earned in the second half (please see the Appendix section for the complete instructions). Afterwards, the instructions documents were collected, and the players were taken to the separate rooms to complete a short quiz (included in Appendix) measuring whether all the instructions were understood clearly.

#### *Design*

Each experimental session consisted of four consecutive rounds and in each round the participants played a different Mini UG game [i.e., the (5/5) game in Round 1, the (8/2) game in Round 2 and so on. Note that the allocation of the games into particular rounds was randomized] <sup>6</sup> . Each player was matched with his/her first game partner (i.e., Proposer 1 with Responder 1) before Round 1 and played two consecutive rounds (e.g., Round 1 and Round 2) with the same partner. After the completion of Round 2, they switched their partners (i.e., Proposer 1 started playing with Responder 2) and played the following two rounds (Round 3 and Round 4) with their new partners. At the end of each round, the decisions of both players (and the resulting outcomes) were announced to the players. These announcements were done privately (i.e., only between the pairs) after Round 1 and after Round 3; but publicly (i.e., to all players) after Round 2 and Round 4. For example, the decisions of Responder 1 and Proposer 1 were announced only to these two players after they completed Round 1, but their overall decisions in Round 1 and Round 2 were announced to all players just before they switched their partners.

#### *Game play*

In all Mini UGs, the Proposer was asked to choose one of the two available distributions (see **Table 1**). Simultaneously the Responder, without knowing what the Proposer had chosen to offer, was asked to indicate his/her acceptance/rejection decisions for each of the two possible distributions. (If the Responder had accepted the offer that the Proposer had chosen, the amount was distributed in accordance with the proposal. Otherwise, both got nothing). Both players were informed about the outcome right after the game was over, and then they moved on to the next game. The currency in the experiment was defined in MUs, where 1 MU equals 0.5 AUD. The experiment was conducted and run with *z*-Tree (Fischbacher, 2007). After the game play was over, both players received a questionnaire that was simultaneously prepared on the basis of the players' actual decisions in the experiment. The Proposers were asked to indicate why they offered the amount they offered and the Responders were asked for each game why they rejected/accepted the (8/2) distribution (please see the Appendix section for the respective questionnaires).

#### **RESULTS**

All participants passed the quiz distributed before the game play, thus all responses were included in the analysis. We first examined the extent to which the possibility of RB and FI influenced the Responders' overall rejection rates of the (8/2) distribution in each game in order to see how the opportunity of RB and FI could change this overall rejection pattern in each game. **Table 2** (the bottom row) presents the rejection rates of the (8/2) distribution in different games, collapsed across rounds. The highest rejection rate was obtained in the (5/5) game and the lowest in the (10/0) game. These rejection rates of the (8/2) distribution were significantly different across four groups (*p <* 0*.*0001, Cochran's *Q*-test)<sup>7</sup> . Interestingly, almost half of the participants rejected the (8/2) distribution in the (8/2) game. McNemar change tests indicated that the rejection rate in the (5/5) game was significantly higher than that in the (10/0) and the (2/8) games, *p <* 0*.*0001, and *p* = 0*.*035, respectively, but not than those in the (8/2) game, *p* = 0*.*75<sup>8</sup> .

A cross-experimental comparison demonstrated that the rejection rates of the (8/2) distribution between Experiment 1a and Experiment 2 did not significantly differ in the (5/5) games [χ2(1, *<sup>N</sup>* <sup>=</sup> 146) <sup>=</sup> 0.83, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*36], the (2/8) games [χ2(1, *<sup>N</sup>* <sup>=</sup> 146) <sup>=</sup> 0.16, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*68], and the (10/0) games [χ2(1, *<sup>N</sup>* <sup>=</sup> 146) = 0.47, *p* = 0*.*49]. Contrary to our expectations, the rejection

<sup>6</sup>In order to eliminate a potential confounding sequence effect, we kept the round order of four Mini UGs different in each experimental session. Initially, there were 24 (i.e., 4! = 24) possible different orders, and thus we ran 24 sessions with a different sequence of Mini UGs in each. As we stated above, we later conducted another block with 24 experimental sessions as requested by one of our referees. For that reason, each of those 24 sequences had to be realized twice in total in the experiment.

<sup>7</sup>The rejection rates of the alternative distributions in the (5/5), (2/8), and (10/0) games were as follows: the (2/8) distribution was rejected by 3%, and the (5/5) distribution by 2%. Almost 91% rejected the (10/0) distribution.

<sup>8</sup>As stated in the Method section, we collected additional data from 96 participants (i.e., in an additional 24 sessions with four participants in each) in accordance with the suggestion of one of our anonymous referees. We ran additional statistical analyses in order to detect if there was any significant differences between these two blocks of 24 sessions (i.e., between the initial 24 sessions and the second 24 sessions). In particular, we compared these two separate blocks in terms of the overall rejection rates obtained in each Mini UG, and we found no significant differences. For the (5/5) game, the rejection rates were 52% vs. 52%, [χ2(1, *<sup>N</sup>* <sup>=</sup> <sup>96</sup>*)* <sup>=</sup> <sup>0</sup>*.*00, *<sup>p</sup>* <sup>=</sup> 1]; for the (2/8) game 41% vs. 35%, [χ2(1, *<sup>N</sup>* <sup>=</sup> 96) <sup>=</sup> 0.40, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*53]; for the (10/0) game 18% vs. 27%, [χ<sup>2</sup> (1, *<sup>N</sup>* <sup>=</sup> 96) <sup>=</sup> 0.94, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*33]; and for the (8/2) game 50% vs. 48%, [χ<sup>2</sup> (1, *<sup>N</sup>* <sup>=</sup> 96) <sup>=</sup> 0.04, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*84] in the first block and second blocks, respectively.

rates of the (8/2) distribution did not increase when the alternative distribution was (5/5), and did not decrease when the alternative distribution was (10/0). However, the (8/2) distribution was rejected in the (8/2) game much more frequently in Experiment 2 than Experiment 1a, <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 146) <sup>=</sup> 15.14, *p* = 0*.*0001. Similar patterns of differences were obtained in the comparison of Experiment 1b and Experiment 2. No significant differences were found between Experiment 1b and Experiment 2 in terms of the rejection rates of the (8/2) distribution in the (5/5) [χ2(1, *<sup>N</sup>* <sup>=</sup> 141) <sup>=</sup> 2.49, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*11] <sup>9</sup> , the (2/8) [χ2(1, *<sup>N</sup>* <sup>=</sup> 141) <sup>=</sup> 0.36, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*55], and the (10/0) games [χ2(1, *<sup>N</sup>* <sup>=</sup> 141) <sup>=</sup> 1.00, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*32]. However, rejection rates for the (8/2) distribution were much higher in Experiment 2 than Experiment 1b in the (8/2) game, <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 141) <sup>=</sup> 8.61, *<sup>p</sup>* <sup>=</sup> 0*.*003. We will return to the interpretation of these results in the General Discussion10.

We then investigated round by round rejection rates in all games of Experiment 2. The rationale of the round-wise analysis was (1) to investigate the reason behind the unexpectedly high levels of rejections in the (8/2) game, and (2) to see the effect of the possibility of RB and FI more clearly. We first focused on patterns (in rejections) indicating any type of signaling from the Responders to the Proposers, in terms of what Responders would not like to be offered. Even though it was possible to see the effect of RB and FI in all four rounds (i.e., because the players did not know how many rounds they would play in total nor when exactly they would switch partners, they should have incentive to build reputation for future interactions in all rounds), the rates of rejections of the (8/2) distribution were especially important in Round 1 and Round 3. Because the Responders would have a chance to give a message to their newly matched partners, they would (presumably) perceive these rounds (1 and 3) as the most suitable time to signal their preferences to their game partners for the following rounds. **Figure 1** depicts the rejection rates (in percentages) of the (8/2) distribution across four rounds in each game of Experiment 2.

For the analysis of round by round rejection patterns, we first conducted a logistic regression by including dummy variables for different rounds, and correcting the standard errors for the clustering on participants (i.e., because we had independent samples for round-wise comparisons but matched samples for four Mini UGs). Afterwards, we tested the pairwise round differences for each game. **Table 3** demonstrates the round by round differences in all games and both the across-rounds and the pairwise significance levels. For the (5/5) game, there was no significant differences across rounds in terms of the rejection rates of the (8/2) distribution [LR <sup>χ</sup>2(3, *<sup>N</sup>* <sup>=</sup> 96) <sup>=</sup> 3.39, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*33]. A similar pattern obtained in the (10/0) game as well: except for the difference between Round 1 and Round 3 [LR <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 48) <sup>=</sup> 3.92, *p* = 0*.*05]. The rate of rejections in the (2/8) game in Round 1 was significantly higher than in Round 2 [LR <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 48) <sup>=</sup> 6.52, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*01], but not in Round 3 than Round 4 [LR <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 48) = 1.39, *p* = 0*.*24]. However, the rejection pattern for the (8/2) game was different: the rejection rates of the (8/2) distribution were especially high in Round 1 and Round 3 (see **Figure 1**). The (8/2) distribution was rejected much more frequently in Round 1 than Round 2 [LR <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 48) <sup>=</sup> 5.18, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*02], and in Round 3 than Round 4 [LR <sup>χ</sup>2(1, *<sup>N</sup>* <sup>=</sup> 48) <sup>=</sup> 9.31, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*0002]. This pattern indicates that the effect of RB and FI was especially prevalent in the (8/2) game.

We also compared the round-wise rejection rates of the (8/2) distribution in each game in Experiment 2 with the rejection rates in the corresponding games in Experiment 1a and Experiment 1b (see **Table 4** for complete lists of significance values revealed through comparisons of round-wise rejections in Experiment 2 with Experiment 1a and 1b for each game). Nevertheless, such analyses did not reveal anything different than the abovementioned results demonstrating the null effect of RB and FI on rejections, except for the (8/2) game. For the (5/5) and the (10/0) games, the rejection rates of the (8/2) distribution in none of the rounds in Experiment 2 were significantly different than those in Experiment 1a and 1b. For the (2/8) game, only the rejection rate in Round 1 of Experiment 2 was marginally higher than that of Experiment 1b (*p* = 0*.*05). However, for the (8/2) game, rejection rates of the (8/2) distribution, especially in Round 1 and Round 3 were significantly higher in Experiment 2 than those in both Experiment 1a and 1b (see the last two columns of **Table 4**).

#### **GENERAL DISCUSSION**

In Experiment 1a, we confirmed that people (negatively) respond to intentional unfairness in a Mini UG at a cost to their own material payoff. The difference in the pattern of results between Experiment 1a and 1b showed the relative impact of the aspects of the fairness consideration in shaping altruistically cooperative behaviors. We found that when a distribution yielding an unequal outcome between players was intentionally offered in the presence of a more equitable distribution [i.e., the (5/5) distribution], that distribution was rejected much more frequently (i.e., in Experiment 1, 60%) than when it was unintentionally offered (i.e., in Experiment 1b, 37%). This pattern of results indicates that from the perspective of the Responder, the intentions of the Proposer matter significantly. However, even when the intentions of the Proposers are not involved, another aspect of the fairness considerations is still present: it is the (un)fairness of the outcome distribution that governs the Responders' rejection behavior. The rejection rates of the unequal distribution were changed depending on the relative equitability of the alternative distribution.

<sup>9</sup>Note that the cross-experimental comparison between Experiment 1a and 1b demonstrated that the rejection rates of the (8/2) distribution in the (5/5) game were significantly lower when the intentions of the Proposer were not assessable. Thus, if the effect of RB and FI is dependent on the absence of intentions, then we would observe significantly higher rates of rejections in Experiment 2 as compared to Experiment 1b for the (5/5) game. Even though this expectation was met (i.e., rejection rates were 33% in Experiment 1b and 52% in Experiment 2), the respective difference was not statistically significant (*p* = 0*.*11), presumably due to the relatively low number of observations in

Experiment 1b (*<sup>n</sup>* <sup>=</sup> 45). 10Due to the variation in proportion of female and male participants in all our experiments, we checked to see if there was any effect of gender in general, or gender by Mini UG interaction on rejection rates of the (8/2) distribution. Even though male participants tended to reject the (8/2) distribution more, there was no significant main effect of gender, *p* = 0*.*42, *p* = 0*.*35, *p* = 0*.*14 in Experiment 1a, 1b, and 2, respectively. Also, there was no interaction effect of gender and the type of Mini UG.

**Table 3 | Significance levels (i.e.,** *p***-values) of across-round and pair-wise round differences in rejection rates of the (8/2) distribution obtained through logistic regression analysis for each Mini UG.**


*For example, the intersection of the first column and the third row corresponds to the significance level of difference between Round 1 and Round 3 in terms of the rejection rates of the (8/2) distribution in the (5/5) game. The rejection rate was 58% in Round 1 and 37% in Round 3 in the (5/5) game (see Table 2), and these rates did not significantly differ, p* <sup>=</sup> *0.15. [\*] sign corresponds to significant differences (i.e., p < 0.05).*

In the literature, there are two distinct approaches to the fairness preferences over material benefits (Falk et al., 2008). These are the intention-based approach to the fairness concept (Rabin, 1993; Dufwenberg and Kirchsteiger, 2004) in which the emphasis is on the (fair/unfair) intentions behind an offer; and the outcome-based models (e.g., Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000) in which the fairness is interpreted as the consideration of ending up with equitable material payoffs. Our results demonstrated that these two aspects of fairness considerations are differentially effective in determining the decisions of players. Thus, these findings provide convincing support for the idea that the economic models of preference for fairness that exclusively focus either on the intentions or on outcome fairness fail to capture altruistically cooperative behavior as a whole (Falk et al., 2008). Our results are rather compatible with the models that take both intentions and concerns for equitable outcomes into account (i.e., Falk and Fischbacher, 2006).

However, contrary to our predictions, the results of Experiment 2 indicated that the additional effect of the possibility of RB and FI did not lead to an increase in altruistic cooperation: rejection rates of the (8/2) distribution did not change when the Responders were expected to punish unfair offers (i.e., the 5/5 game) or to appreciate fair offers (i.e., the 10/0 game). Cross-experimental comparisons of the rejection rates obtained in Experiment 1a (i.e., involving both intention and outcome fairness) and Experiment 1b (i.e., involving outcome fairness only) with the overall rejections rates in Experiment 2 (i.e., involving RB and FI opportunity along with fairness considerations) confirmed that there were no changes in the rejections of an inequitable distribution in the (5/5) and the (10/0) games when the possibility of RB and FI was incorporated in to the context.

Two potential but competing explanations of this pattern of results can be offered. One is that the possibility of RB


**Table 4 | Significance levels (i.e.,** *p***-values) of differences in rejection rates of the (8/2) distribution obtained through the comparison of each round of Experiment 2 for each Mini UG with Experiment 1a and Experiment 1b for corresponding Mini UG.**

*For example, the intersection of the fourth column and the second row reads the significance level (0.28) of the difference between the rejection rate obtained in Round 2 of the (2/8) game in Experiment 2 (i.e., 21%—see Table 2) and the rejection rate obtained in the (2/8) game in Experiment 1b (i.e., 33%). [\*] sign corresponds to significant differences (i.e., p < 0.05).*

and FI is indeed (mis)perceived in one-shot and anonymously played games, and thus did not lead to any differences in the pattern of responses when it was explicitly incorporated into the context (Haley and Fessler, 2005; Bateson et al., 2006). The other is that the explicit incorporation of the possibility of RB and FI did not have any additional effect on the responses in the presence of the influence of fairness considerations (that are already effective enough to determine the rates of rejection). Unexpectedly high rejection rates of the (8/2) distribution observed in the (8/2) game in Experiment 2, as well as the round by round analyses of these rejection rates in each game strongly provide supporting evidence for the latter explanation.

The possibility of RB and FI led to an increase in the overall rejection rates only in a particular game where the intention of the Proposer was not assessable [the (8/2) game], but not in the other games in which the intentions were assessable [the (5/5), the (10/0), and the (2/8) games] (please see **Table 2**). This is the first indication of the effect of RB and FI being too weak to overcome the effect of fairness considerations. The Responders might only be taking the perceived intentions of the Proposers into consideration as a determinant of their accept/reject decisions for an unequal offer, and thus might not need to have additional reasons/concerns to change those decisions even when RB and FI are possible.

Round-wise analyses of the games in Experiment 2 support the claim that the possibility of RB and FI per se was not effective in changing the rejection responses, especially when the intentions were assessable: there was no variation across rounds in terms of the rejection rates of the (8/2) distribution, especially in the (5/5) and (10/0) games 11. However, the effect of RB and FI did become effective once the fairness consideration is weakened as a result of the removal of the possible intentions behind an offer in the (8/2) game: it makes the Responders overly react against the unfairness of the outcome of the (8/2) distribution, most likely, in order to increase the possibility of being treated fairly in the future (Kiyonari et al., 2000; Hertel et al., 2002). This interpretation is mainly supported by the comparison of the rejection rates obtained in the (8/2) game across rounds in Experiment 2. The round-wise analysis of Experiment 2 (see **Figure 1**) showed that the rejection rates were significantly higher both in Round 1 (than Round 2) and Round 3 (than Round 4) only in the (8/2) game, where the intentions of the Proposer was not assessable. As stated previously, these two rounds were particularly important for the Responders to convey their message for future encounters. The implicit message given under such condition could be that they do not like to be offered an unequal distribution by *the same or the next game* partner in the following rounds. The Responders' self-reports collected after the game play also confirmed that the main purpose of the rejections in this game was indeed to tell the Proposers that "I will reject again if you ever propose such an unequal distribution."

The results indicate that the absence of fairness intentions was the primary reason for the possibility of RB and FI becoming effective. However, the comparison between Experiment 1b and Experiment 2 revealed the importance of "outcome fairness" aspect of fairness considerations as well. This is because, except for the (8/2) game, the rejection rates of the (8/2) distribution obtained in none of the games were significantly different between Experiment 2 and Experiment 1b. This finding implies that even the presence of outcome fairness [i.e., perceived fairness of the (8/2) distribution relative to its alternative distribution] itself is strong enough to make the rejection rates reach a certain level—a level that could not get increased [i.e., for the (5/5) game] or decreased [i.e., for the (10/0) game] by the explicit incorporation of RB and FI. The possibility of RB and FI matters only when there is no intention behind the distribution offered, and there is no (more equitable or inequitable) alternative distribution to be offered [i.e., the (8/2) game].

These results shed light on when and how the possibility of RB and FI influence the responses in bargaining games. The possibility of RB and FI is normally expected to increase the rejections

<sup>11</sup>Here one could argue that if the opportunity of RB and FI was effective only in the absence of intentions [i.e., in the (8/2) game], why then were the rejection rates in Round 1 marginally significantly higher than those in Round 2 in the (2/8) game—a pattern that indicates an effect of RB and FI in the presence of intentions. Note that the effect of RB and FI was still not strong in this particular game: First, the overall rejection responses for the (8/2) distribution did not differ with or without the involvement of RB and FI [compare the overall rejection rates in Experiment 1a, 1b, and 2 for the (2/8) game, rows 1st, 2nd, 8th in **Table 2**]. Second, the difference between Round 3 and Round 4 (see the last row of the 2nd column in **Table 3**) was not significant in terms of the rejection rates. This is a further indication that the effect of RB and FI was not strong.

in the Mini UG as the Responders want to build the reputation of being a "tough bargainer." This is why such opportunity has been sometimes thought to be the source of conflict (i.e., reduction in overall pay-off-wise efficiency) in strategic environments (Falk et al., 2003b). Our findings suggest that the possibility of RB and FI can lead to the respective conflict only when the fairness intentions are not assessable (i.e., the absence of the evaluation of intentions seems to increase the rejections in Mini UG, and thus leads to a "zero" outcome for both parties).

The current set of studies demonstrates the importance of fairness considerations, especially that of (un)fair intentions, in interactive economic decisions, particularly in the ultimatum bargaining games. The main conclusion drawn from the experiments reported here is that when both outcome and intention fairness considerations are involved in decisions they have a strong

#### **REFERENCES**


combined effect on the emergence of costly punishment as a form of altruistic cooperation, and thus override the potential influence of RB and FI. When an important aspect of fairness concerns, namely intentions, are absent, RB and FI may play an important role.

## **ACKNOWLEDGMENTS**

Experiment 1a and 2 have been submitted (in the form of a short report) to the 35th Annual Conference of the Cognitive Science Society for presentation. This research was supported by a UNSW Overseas Postgraduate Research Scholarship awarded to the first author and Australian Research Council grants (DP110100797; FT110100151) to the second author. We are also grateful to our three referees and the editor for their helpful comments.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 17 February 2013; accepted: 20 May 2013; published online: 07 June 2013.*

*Citation: Güney ¸S and Newell BR (2013) Fairness overrides reputation: the importance of fairness considerations in altruistic cooperation. Front. Hum. Neurosci. 7:252. doi: 10.3389/fnhum.2013.00252 Copyright © 2013 Güney and Newell. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# **APPENDIX**

## **A. INSTRUCTIONS**

## *Interactive decision making in economic games*

Welcome and thank you for participating in this experiment. This experimental session consists of several parts. It's as follows:


*(The experimenter asks participants to follow the written instructions while she is giving the instructions verbally)*


# *(Assignment of the roles)*


## *(Rules of the game)*


acceptance/rejection decision both for the \$6 offer, and for the \$3 offer.


# *(Important details about the experimental procedure)*


### **B. QUIZ**

Please answer to the following questions.

	- Yes \_\_\_ No \_\_\_ (Please indicate what will happen then)
	- if the Responder accepts the offer, the Proposer takes \_\_\_ MUs, and the Responder gets \_\_\_ MUs.
	- if the Responder rejects the offer, the Proposer takes \_\_\_ MUs, and the Responder gets \_\_\_ MUs.

Yes \_\_\_ (Please indicate how many) No \_\_\_

l. Were your roles randomly assigned? If not, how? Yes \_\_\_ No \_\_\_ (Please indicate how)

### **C. QUESTIONNAIRE FOR PROPOSERS**

*(Former brackets were filled in with the actual offers by the Proposer, and the latter with the alternative offers in corresponding rounds)*


Please indicate your

Age:

Gender:

Thanks for your participation. Please see the experimenter for getting paid (and debriefing).

## **D. QUESTIONNAIRE FOR RESPONDERS**

*[Former brackets were filled in with the actual decisions (either accept or reject) by the Responder, and the latter with corresponding rounds in that particular session.]*


## Please indicate your

Age:

Gender:

Thanks for your participation. Please see the experimenter for getting paid (and debriefing).

# *Bojana Kuzmanovic 1,2\*, Anneli Jefferson1, Gary Bente3 and Kai Vogeley2,4*

*<sup>1</sup> Institute of Neuroscience and Medicine - Ethics in the Neurosciences (INM-8), Research Center Juelich, Juelich, Germany*

*<sup>2</sup> Department of Psychiatry and Psychotherapy, University Hospital Cologne, Cologne, Germany*

*<sup>3</sup> Department of Psychology, University of Cologne, Cologne, Germany*

*<sup>4</sup> Institute of Neuroscience and Medicine - Cognitive Neurology (INM-3), Research Center Juelich, Juelich, Germany*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*James S. Uleman, New York University, USA Laurence Kaufmann, University of Lausanne, Switzerland*

#### *\*Correspondence:*

*Bojana Kuzmanovic, Institute of Neuroscience and Medicine - Ethics in the Neurosciences (INM-8), Research Center Juelich, 52425 Juelich, Germany e-mail: b.kuzmanovic@fz-juelich.de*

Interpersonal impression formation is highly consequential for social interactions in private and public domains. These perceptions of others rely on different sources of information and processing mechanisms, all of which have been investigated in independent research fields. In social psychology, inferences about states and traits of others as well as activations of semantic categories and corresponding stereotypes have attracted great interest. On the other hand, research on emotion and reward demonstrated affective and motivational influences of social cues on the observer, which in turn modulate attention, categorization, evaluation, and decision processes. While inferential and categorical social processes have been shown to recruit a network of cortical brain regions associated with mentalizing and evaluation, the affective influence of social cues has been linked to subcortical areas that play a central role in detection of salient sensory input and reward processing. In order to extend existing integrative approaches to person perception, both the inferential-categorical processing of information about others, and affective and motivational influences of this information on the beholder should be taken into account.

**Keywords: person perception, impression formation, social inference, affective influence, reward**

## **INTRODUCTION**

For humans, other people are one of the most important sources of joy and sorrow. Our perception of others has far-reaching consequences for immediate reactions as well as for the likelihood and nature of future interactions. Thus, in everyday life, impression formation is crucial not only for private relationships but also for decisions regarding economic or political affairs (Delgado et al., 2005; Uleman et al., 2008; Antonakis and Dalgas, 2009). Due to its relevance, "person perception" in its broadest sense, i.e., covering sensory, cognitive, and affective processing of information about others, has generated intense research interest in a variety of disciplines. This widespread interest is in part due to the fact that the perception of persons fundamentally differs from that of objects insofar as it involves recognition of some other as an epistemic and moral subject and the possibility of reciprocity (Sturma, 1997).

Empirical approaches toward person perception describe different sources of person-related information and different kinds of impact of this information on the decoder. Within the classic social psychological research on categorical representation of others and inferences about others' current mental states and enduring personality traits, the person-related cues are seen as providers of diagnostic knowledge (Kelley, 1967; Trope, 1986; Mitchell et al., 2006; Freeman et al., 2010b). But psychological and neuroscientific research also demonstrated that interpersonal social cues have an intrinsic affective and motivational value and are able to influence sensory, inference and decision processes, as well as response selection (Klin et al., 2003; Winkielman et al., 2005; Vuilleumier and Pourtois, 2007; Schilbach et al., 2010). The relative involvement of these well-documented inferential-categorical and affective-motivational processes in person perception, however, crucially depends on the specific information format, e.g., whether information is conveyed verbally or non-verbally (Freeman et al., 2010b; Zaki et al., 2010; Kuzmanovic et al., 2012).

Critically, all these different approaches to person perception do not refer to processes that run independently, rather, they reciprocally modulate each other and the final information integration (Vuilleumier and Pourtois, 2007; Pessoa, 2008; Freeman and Ambady, 2011; Freeman et al., 2012). Nevertheless, empirical and theoretical approaches mostly focus only on one selective aspect of person perception, and are only beginning to develop integrative and comprehensive models (Freeman and Ambady, 2011; Freeman et al., 2012). For instance, the dynamic interactive theory of person construal integrates insights from social psychology and functional neuroimaging research related to face-processing by emphasizing complex interactions of cognitive processes underlying initial activation of categories (and corresponding stereotypes) and further higher order social reasoning (Freeman and Ambady, 2011; Freeman et al., 2012). This theory provides an excellent framework based on a recurrent connectionist model (Freeman and Ambady, 2011) and does refer to "top-down influences that originate in the perceiver (e.g., existing knowledge structures and motivations) and (. . . ) bottom-up influence of factors that originate in the target of perception (e.g., overlapping visual cues)" (Freeman et al., 2012, p. 3). However, while this framework considers bottom-up influences of sensory input on category and stereotype activation, it does not explicitly address affective and motivational properties of social stimuli resulting in enhanced and prioritized processing and reward-dependent learning effects. These effects have been extensively documented in the emotion and reward-related neuroimaging research (see section Affective Influences of Person-related Information on the Decoder), even in newborns without fully developed propositional knowledge (Farroni et al., 2002), and do not have to relate to categorical organization of social cognition that is described in the person construal theory (Vuilleumier and Pourtois, 2007). By delineating distinct approaches to person perception related to cognitive inferential-categorical processing (section Person-related Knowledge as a Basis for Social Reasoning) and affective-motivational influences (section Affective Influences of Person-related Information on the Decoder), respectively, and by specifying relative processing differences for verbal and nonverbal formats of information (section Processing Differences for Distinct Formats of Person-related Information), the present paper aims to extend existing integrative views on person perception by emphasizing the critical role of salience and rewardrelated effects within the dynamic processing of social others (section Integration of Distinct but Interactive Person Perception Aspects).

## **PERSON-RELATED KNOWLEDGE AS A BASIS FOR SOCIAL REASONING**

Traditionally, social psychology has been primarily interested in how we form high-level impressions of others and how knowledge about others is represented within a categorically organized semantic system (Freeman and Ambady, 2011). Originally, it was supposed that people are trying to causally explain the observed behavior of others. Attribution Theory defined conditions in which logical and objective reasoning leads to the assumption of internal, i.e., disposition-related, or external, i.e., situation-related, causes for actions, dependent upon the available information about the target person (Heider, 1958; Jones and Davis, 1965; Kelley, 1967). Novel approaches additionally integrate initial lower-level perceptual interpretation and categorization processes in order to account for top-down and bottomup dynamic interactions within social reasoning (Freeman and Ambady, 2011). Such models comprehensively explain how categories and corresponding stereotypes along with individuating information are used to form impressions of others and to understand their personality characteristics and current mental states. Furthermore, the general ability to attribute mental states such as beliefs and intentions to oneself and others in order to understand and predict behavior has been investigated with reference to "theory of mind" (ToM) (Premack and Woodruff, 1978). A prominent way to assess this ability is to use "false belief tasks" with social scenarios or non-verbal cues where test persons have to differentiate between their own perceptions, attitudes, or beliefs from those of others (Wimmer and Perner, 1983). Neuroimaging studies were able to associate this inferential and category-based social processing with a network consisting of the medial prefrontal, the retrosplenial and the temporo-parietal cortices, among others (Vogeley et al., 2001; Saxe and Kanwisher, 2003; Harris et al., 2005; Mitchell et al., 2005; Schiller et al., 2009), by using mostly, though not exclusively, verbal stimulus material (Walter et al., 2004; Freeman et al., 2010a).

While these studies obviously focus on types of knowledge and reasoning we all use extensively in our everyday life, it has become increasingly clear that other ways of processing personrelated information are equally or sometimes even more significant. For example, individuals with high functioning autism, who can pass explicit experimental false belief tasks as well as controls, are still impaired in their daily social life and are unable to transfer this knowledge into more complex and ecologically valid situations (Klin et al., 2003). The remaining impairments are supposed to relate to difficulties in spontaneously attending to socially meaningful stimuli in real world environments and in experiencing social stimuli as significant (Klin et al., 2003; Senju et al., 2009; Kuzmanovic et al., 2011). This example highlights the importance of taking into account influences of the affective value of social cues in a comprehensive investigation of person perception. While the outcome of the initial categorization and of inferential analyses of person-related information also crucially affects subsequent evaluations and behavior toward the target person (Freeman and Ambady, 2011), this category and inference-dependent influence can be distinguished from affective influences of salient social cues on the decoder. Such affective and motivational effects are present before categorical social knowledge has fully developed (Farroni et al., 2002), and can act independently of top-down attention control (see next section).

## **AFFECTIVE INFLUENCES OF PERSON-RELATED INFORMATION ON THE DECODER**

In general, humans' decision making is influenced by emotional factors (Slovic and Peters, 2006; De Martino et al., 2008). Such influences can be triggered by stimuli that have a predispositional or primary affective value, such as food and social cues including attractive faces or emotional expressions (Aharon et al., 2001; Bray and O'Doherty, 2007; Lin et al., 2012). Alternatively, stimuli can acquire an affective value through classical conditioning (also "Pavlovian conditioning"), i.e., when neutral stimuli acquire a positive or negative value due to repeated pairing with other unrelated positive or negative stimuli (Hermans et al., 2002). Especially during person perception, social cues are necessarily present and can influence the decoder due to their intrinsic affective value. This influence may concern an enhanced and prioritized processing relating to selective attention to, and recognition and representation of stimuli. Furthermore, social cues may also function as incentives and thus lead to reward-dependent learning.

Regarding the influence related to prioritized processing, neuroimaging studies have shown that face processing is enhanced for emotionally expressive as compared to neutral stimuli as a result of the modulatory influence of the amygdala (Vuilleumier and Pourtois, 2007). Known to play a central role in detecting affectively significant sensory input (Sander et al., 2003), the amygdala is able to modulate activity in brain networks associated with visual face processing and with other cognitive and affective responses in favor of the more salient emotional information via massive reciprocal connections (Vuilleumier and Pourtois, 2007; Pessoa, 2008; Robinson et al., 2010). Consistent across a line of studies, non-verbal person-related information including facial and bodily expressions as well as invariant facial features such as attractiveness or race elicits enhanced activity in the amygdala (Phelps et al., 2000; Hariri et al., 2002; Lieberman et al., 2005; Sergerie et al., 2008; Kuzmanovic et al., 2012). Moreover, increased activation of the amygdala has also been found for social stimuli in general, i.e., for neutral stimuli too and is thus independent of their valence, when compared to non-social stimuli (Vrticka et al., 2012). Critically, amygdala-driven modulation of visual processing by emotional information is also detectable for non-attended stimuli, i.e., without voluntary control or conscious awareness (Vuilleumier and Schwartz, 2001; Vuilleumier et al., 2001). Moreover, patients with lesions in the visual cortex who cannot achieve conscious visual experience can still discriminate facial expressions of emotions—presumably via a subcortical circuit including the superior colliculus, thalamus, and the amygdala (Adolphs, 2002). Similarly, autonomic measures demonstrate that patients with prosopagnosia, who are unable to recognize familiar faces, can still discriminate familiar from unfamiliar faces on an unconscious level (Ellis and Lewis, 2001). Thus, this subcortical processing may mediate affective influences independently of the activation of category-organized knowledge structures.

The described sensitivity of the amygdala to salient stimuli may play a central role in attracting the attention toward meaningful social cues. When this function is impaired, as in patients with amygdala lesions, spontaneous recognition of emotional expressions of faces is reduced (Adolphs et al., 2002), unless these patients have been explicitly instructed to attend to the informative eye region (Adolphs et al., 2005). Furthermore, in contrast to healthy or hippocampus-lesioned persons, patients with amygdala lesions do not demonstrate increased activation in visual face-related brain regions for fearful relative to neutral faces, but they show increased activation in these regions when faces are presented in a task-relevant relative to task-irrelevant position (Vuilleumier et al., 2004). Hence, the top-down attentional modulations by task demands can act independently of the modulations by affective significance of the stimuli via the amygdala (Vuilleumier and Pourtois, 2007).

Beyond these general affective influences related to prioritized processing, social cues can modulate affective responses to unrelated, but simultaneously or subsequently presented stimuli. Extending prior neuroimaging findings that face attractiveness or gaze following induce activity in reward-associated neural areas such as the orbitofrontal cortex and the ventral striatum (Aharon et al., 2001; Kranz and Ishai, 2006; Schilbach et al., 2010), pleasant social stimuli were also able to establish affective values in abstract and initially neutral stimuli by means of classical conditioning (Bray and O'Doherty, 2007). Thus, just like other types of reward such as food or money, person-related information can influence our evaluations of arbitrary stimuli when paired with them. Although the study by Bray and O'Doherty refers to classical conditioning and demonstrates the involvement of the ventral striatum, which has previously been associated with learning based on this principle, the fact that the measured effect relates to an evaluative attitude and not only to an affective response calls for a more precise reference to the similar, but not identical evaluative conditioning. This field of research provides further behavioral empirical evidence for influences of valent social stimuli such as likeable and dislikeable faces on the evaluation of neutral facial stimuli (Baeyens et al., 1992; Walther et al., 2005). In addition to their effect in classical and evaluative conditioning, which change the attitude and the affective reaction to previously neutral stimuli, positive and negative facial expressions were also shown to modulate complex consumption behavior. Subliminally presented smiling faces increased the consumption of and the willingness to pay for beverages while frowns had the opposite effect (Winkielman et al., 2005). Interestingly, these effects on overt behavior occurred without eliciting changes in conscious feelings. Such effects, "in which the motivational characteristics of a predictor influence the vigor of an action with respect to which it is formally completely independent are called "Pavlovian-Instrumental Transfer" (PIT) (Talmi et al., 2008, p. 360). The PIT has been shown to be mediated by the ventral striatum and the amygdala, in concordance with the regulatory role of these regions in integration of affective-motivational, cognitive, and motor processing in the brain (Talmi et al., 2008).

Taken together, these findings suggest that humans are equipped with motivational predispositions to respond to person-related information, among other biologically relevant stimuli, presumably due to its significance for adaptive social behavior and, in consequence, survival (Dunbar, 2009). In consequence, social cues may be more efficiently detected for the purpose of prioritized processing, memorization, and evaluation (Klin et al., 2003; Vuilleumier and Pourtois, 2007). Furthermore, this affective or motivational value of social cues can also influence simple approach-avoidance (Chen and Bargh, 1999) and complex instrumental behaviors (Winkielman et al., 2005) via reward-dependent learning.

## **PROCESSING DIFFERENCES FOR DISTINCT FORMATS OF PERSON-RELATED INFORMATION**

While the basic sensory and cognitive processing of verbal and non-verbal information is generally associated with distinct neural areas, there are also format-dependent differences specifically related to social processing. From very early on, psychological theories of interpersonal communication assumed that non-verbal information has been assumed to have a greater impact on the affective, relational level of communication (Watzlawick et al., 1967). Furthermore, linguistically encoded information always requires the processing of an explicit semantic code with a complex logical syntax and is thus necessarily linked to high-level cognitive processing, while non-verbal information lacks such an explicit interpretation code (Bente and Kraemer, 2008; Kraemer, 2008).

Recently, neuroimaging studies could provide empirical support for these assumptions by showing that the processing of non-verbal person information consistently recruits the amygdala (Hariri et al., 2002; Winston et al., 2002; Sergerie et al., 2008; Todorov and Engell, 2008; Todorov et al., 2008). By contrast, social inferences based on verbal descriptions of other persons' actions or traits, as well as explicit categorization of facial stimuli involved medial prefrontal, retrosplenial and temporo-parietal cortical brain regions (Mitchell et al., 2002, 2005; Harris et al., 2005; Schiller et al., 2009; Freeman et al., 2010a; Zaki et al., 2010). Only a few studies directly compared the processing of verbal and non-verbal information and further substantiated the view that different neural networks show relatively stronger links to the one or the other information format (Freeman et al., 2010b; Zaki et al., 2010; Kuzmanovic et al., 2012). For instance, during a person evaluation task, evaluations of increasing intensity based on non-verbal information recruited the amygdala to a greater extent, whereas the same pattern was observed in the retrosplenial cortex for verbal information (Kuzmanovic et al., 2012). This finding confirms qualitatively different cognitive processes underlying person perception with a closer relation of non-verbal information and salience-dependent processing on the one hand, and of verbal information and the high-level social cognition on the other.

In addition to the differences outlined for the information format, the context and the content of social cues modulate social processing in a complex manner. For instance, it has been supposed that social cognitive processes fundamentally differ when people engage in direct interpersonal interactions than when merely observing social cues, a topic that is beyond the scope of the present article [for a discussion on the second-person approach see Schilbach et al. (2013)]. Taken together, the exact format, context and content of the information that is available during person perception can critically determine the kind of cognitive processes recruited. While both inferential-categorical processing and affective influences may take place for all sources of information, possible effects have to be differentially weighted for distinct kinds of person stimuli.

## **INTEGRATION OF DISTINCT BUT INTERACTIVE PERSON PERCEPTION ASPECTS**

Although different processes with distinct functional implications can be delineated for person perception, they also have to be considered as embedded in a strongly interconnected neural network, thereby reciprocally modulating each other. For instance, the fact that person-related information is able to influence the observer due to its affective value does not mean that this influence is absolutely automatic and unfiltered, without the modulation by reflective and goal-directed processes such as voluntary attention and conscious intentions, appraisals and attitudes (Pessoa, 2008). Exactly this dynamic interactive nature of social processing including the influence of sensory cues has been described previously (Freeman and Ambady, 2011). However, in this model the impact of salient affective social cues is defined in terms of categories and

## **REFERENCES**


corresponding stereotypes without taking into account their ability to prioritize processing and act as a reward, thereby constituting a category-independent source of influence on the beholder.

An example of the top-down modulation of affective responses to monetary rewards has been provided both on the behavioral and neural level. When people believe that their trading partner has a praiseworthy moral character, they rely less on the actual behavior resulting in monetary losses or wins (Delgado et al., 2005). On a neural level, this effect corresponds to reduced differential activity in the ventral striatum in response to positive and negative outcomes for trading partners with positive or negative moral character evaluations as compared to neutral partners (Delgado et al., 2005). Thus, personality trait inferences can greatly influence the reward-dependent, motivational responses within person perception. On the other hand, however, facial attractiveness and expression may influence deliberate high-level judgments about unrelated performances such as political votes or management success (Ballew and Todorov, 2007; Antonakis and Dalgas, 2009), an effect mediated by the amygdala (Rule et al., 2010).

Thus, it can be assumed that there are different concurrent and mutual modulations during person perception, arising from both top-down cortical attentional and inferential networks, as well as from bottom-up primary and secondary sensory areas (Freeman and Ambady, 2011). We argue, however, that the intrinsic affective and motivational value of social stimuli exerts an additional influence over the general person perception via subcortical regions associated with salience and reward processing.

# **CONCLUSIONS**

In the light of the diversity of findings relating to person perception, person-related information has to be regarded as a basis for inferences and categorizations as well as a source of a potential affective-motivational influence on the decoder. As a complex spontaneous constructive process, and not a simple one-to-one representation of available cues, person perception includes a huge amount of uncertainty and is thus prone to biases. In order to prevent oversimplifications and deficient interpretations of empirical findings, an integrative theoretical reflection on person perception should carefully consider affective and motivational effects of person-related information in addition to the dynamic of interactive cognitive processes previously described in integrative frameworks.


unreflective face judgments. *Proc. Natl. Acad. Sci. U.S.A.* 104, 17948–17953. doi: 10.1073/pnas. 0705435104


*J. Neurophysiol.* 97, 3036–3045. doi: 10.1152/jn.01211.2006


*Neuroimage* 17, 317–323. doi: 10.1006/nimg.2002.1179


individuals. *Nat. Neurosci.* 8, 720–722. doi: 10.1038/nn1465


P., et al. (2001). Mind reading: neural mechanisms of theory of mind and self-perspective. *Neuroimage* 14, 170–181. doi: 10.1006/nimg.2001.0789


R. J. (2004). Distant influences of amygdala lesion on visual cortical activation during emotional face processing. *Nat. Neurosci.* 7, 1271–1278. doi: 10.1038/ nn1341


*and Paradoxes.* New York, NY: Norton.


*J. Neurosci.* 30, 8481–8488. doi: 10.1523/JNEUROSCI.0382-10.2010

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 March 2013; accepted: 24 May 2013; published online: 11 June 2013.*

*Citation: Kuzmanovic B, Jefferson A, Bente G and Vogeley K (2013) Affective and motivational influences in person perception. Front. Hum. Neurosci. 7:266. doi: 10.3389/fnhum.2013.00266*

*Copyright © 2013 Kuzmanovic, Jefferson, Bente and Vogeley. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# How interpersonal power affects empathic accuracy: differential roles of mentalizing *vs.* mirroring?

# *Dario Bombari 1,2\*, Marianne Schmid Mast 1,2, Tobias Brosch2,3 and David Sander 2,3*

*<sup>1</sup> Department of Work and Organizational Psychology, University of Neuchâtel, Neuchâtel, Switzerland*

*<sup>2</sup> Swiss Center for Affective Sciences, University of Geneva, Geneva, Switzerland*

*<sup>3</sup> Laboratory for the Study of Emotion Elicitation and Expression, Department of Psychology, University of Geneva, Geneva, Switzerland*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Oliver C. Schultheiss, Friedrich Alexander University, Germany Sylvia A. Morelli, Stanford University, USA*

#### *\*Correspondence:*

*Dario Bombari, Department of Work and Organizational Psychology, University of Neuchâtel, Emile-Argand 11, 2000 Neuchâtel, Switzerland e-mail: dario.bombari@gmail.com*

Empathic accuracy (EA)—the correct assessment of the affective states and thoughts of a social partner—affects social behavior and the outcome of interpersonal interactions. Growing evidence has shown that interpersonal power of a perceiver affects EA when assessing a target. This picture, however, is not obvious; there is evidence supporting both the idea that power can improve EA or impair it. Moreover, the mechanisms through which high power individuals are more (or less) accurate at reading others' minds are unknown. The present article provides a new perspective on the power-EA link by investigating how two core abilities involved in EA, mentalizing and mirroring, can explain when and how power is related to EA. The inclusion of findings from neuroimaging studies on mentalizing and mirroring adds a cognitive neuroscience perspective to the power-EA research that has traditionally been conducted in a social psychological framework. The extent to which a given EA-test requires mentalizing or mirroring and the way power affects both of them could explain the contrasting findings. In addition, the analysis of the neural substrates of mentalizing and mirroring may provide new insight into the relationship between power and EA.

#### **Keywords: power, empathic accuracy, mentalizing, mirroring, interpersonal sensitivity**

# **INTRODUCTION**

Power affects how people perceive their interaction partners (e.g., high power people perceive social interaction partners as a means to an end) (Magee and Smith, 2013) and how they interact with others (e.g., powerful people assert themselves by talking a lot and interrupting others) (Schmid Mast, 2002; Hall et al., 2005). In this contribution, we refer to power interpersonally, as the degree to which an individual can exert control over another person (Schmid Mast et al., 2009). We focus on the psychological properties of power that can be evoked not only by a real hierarchical relationship, but also simply through cues related to power. Other definitions are sometimes used in the literature. Structural power refers to the hierarchical differences in functions or positions (Ellyson and Dovidio, 1985). Status is a group acknowledgment of respect awarded to a specific individual or can be the power derived from membership in a specific social group (Sidanius et al., 2004). Dominance can reflect both an enduring trait of personality (Ellyson and Dovidio, 1985) or a more transient behavior related to the intention of seeking control over others (Schmid Mast, 2002). Because our review focuses on experimentally manipulated power and its effect on EA, the articles we cite define power in a similar way as we do (i.e., control over other people).

Power does not only affect how others are perceived and acted upon; it also affects the degree to which the assessment of a perceiver is correct (Hall et al., under review). Correct assessment of other people's traits and states is called interpersonal sensitivity or interpersonal accuracy (Hall and Bernieri, 2001; Schmid Mast et al., 2012). One aspect of interpersonal sensitivity is empathy, which has been defined as the ability of a perceiver to recognize, understand, and share the emotions, intentions, and feelings of a target (Zaki et al., 2009). Empathic accuracy (EA) is the correspondence between perceiver's judgments and target's states and feelings (Ickes, 1997; Zaki et al., 2009). In the present paper, we investigate the link between power and EA.

Research shows that high power individuals are better at correctly assessing others' emotions and thoughts (Schmid Mast et al., 2009). However, this finding is not unequivocal in that opposite effects have been documented as well (Galinsky et al., 2006). A recent meta-analysis (Hall et al., under review) revealed a small (M *r* = 0*.*07) but significant effect showing that high power individuals are more interpersonally accurate than low power individuals. It is noteworthy that the way power was operationalized (i.e., dispositional trait, structural power, or experimentally induced power) had no significant effect on this relationship. The huge heterogeneity of the effect sizes extracted from the literature suggests that there are moderators at work, affecting the power-EA link. One such moderator might be the different accuracy tests which require different skills or are sensitive to different underlying cognitive processes. Interpersonal accuracy tests tend to correlate only weakly, if at all, with each other (Hall, 2001; Zebrowitz, 2001), which corroborates the idea that different tests might require different skills or cognitive processing styles.

The mechanism through which power affects accurate interpersonal perception is unknown. Previous studies found that powerful people are more *prosocially oriented* (Cote et al., 2011) but also less *motivated to be accurate* (Stevens and Fiske, 2000) and more *socially distant* (Magee and Smith, 2013). Schmid Mast et al. (2009) showed that *feeling respected and proud* partially explained the high power individuals' greater EA. The metaanalysis by Hall et al. (under review) showed that trait dominance was related to more interpersonal accuracy when measured as empathic/responsible compared to egoistic/aggressive. Another trait aspect of power that might moderate the power-EA link is the implicit need for power (nPower) (Winter, 1973), which can influence the perceived saliency (Schultheiss and Hale, 2007; Wang et al., 2011) and the motivational response (Schultheiss et al., 2008) toward emotional faces. Since people high in nPower are faster at recognizing emotions (Donhauser et al., under review), it is possible that nPower positively affects EA. These examples of the potential mechanisms linking power to EA do not provide a comprehensive explanation of the contrasting findings mentioned above. This is why we propose a framework that might tie together the results of previous studies.

Historically, two approaches have been put forward to explain how we read other people's minds (Goldman and Sripada, 2005). The "theory-theory" explains mindreading as an extraction of meaning from targets' behavior, mental state, and context. The "simulation theory" instead postulates that we understand others through an internal simulation of their mental state. Although these two approaches have been developed quite independently, neuroimaging studies have shown that indeed EA involves two different mechanisms, mentalizing, and mirroring (Zaki et al., 2009) and in the present article, we aim to discuss their potential role in relation to how power affects EA.

*Mentalizing* typically means extracting and understanding another person's goals by making inferences about his/her mind state (Amodio and Frith, 2006; Spunt et al., 2011). It relies on the ability to distinguish between one's own mental perspective and that of others (i.e., theory of mind). Mentalizing skills are often tested through false-belief paradigms where participants read short stories about two characters and need to make inferences about others' minds based on the knowledge available to other people. There is evidence (Lieberman, 2010) that the different mentalizing tasks converge in that they all activate one specific brain area, the dorsomedial prefrontal cortex (dmPFC). Other regions (e.g., the temporo-parietal junction and the temporal pole) might be more contingent on task demands. *Mirroring* typically means simulating the state of the other to understand the content of his/her mind (Zaki and Ochsner, 2012). The rationale is that observing another person activates the corresponding motor and mental representations in the observer, enabling him/her to understand the other's mind (i.e., neural resonance). Mirroring is supposed to rely on the mirror neuron system, which was first discovered in the macaque brain (Gallese et al., 1996). Although the existence of such a system in humans is now quite commonly assumed, the topic is still debated (e.g., Kilner, 2011). According to Lieberman (2010), the mirror system relies on the bilateral posterior ventrolateral PFC and bilateral anterior inferior parietal lobule (IPL). Mirroring is considered a rather automatic, unconscious response based on shared mental representations whereas mentalizing is a rather cognitive aspect of empathy that necessitates an explicit representation of the subjectivity of the social interaction partner (Decety and Jackson, 2004). The two systems cooperate closely, because the mirror system helps provide an early identification of the facial expressions and the mentalizing system processes this input in order to make causal attributions about emotions (Spunt and Lieberman, 2012).

*Accuracy* in assessing others' emotions has been documented to be related to both of the aforementioned brain systems: regions within the mirror neuron system (i.e., the middle frontal gyrus and the IPL) and areas involved in mentalizing (i.e., the superior temporal sulcus and medial PFC) (Zaki et al., 2009). To the extent that the cues about a target's feelings and thoughts become multimodal and dynamic, concurrent activation of both systems might be crucial (Zaki et al., 2009).

In the present article we propose a new perspective on the relationship between power and EA by bringing together two strands of research that have so far been relatively unconnected: the study of power and interpersonal accuracy from a social psychological point of view and the study of EA and its neural bases from the cognitive neuroscience approach. In particular we argue (i) that different EA tasks might require predominantly mirroring or mentalizing skills, (ii) that power might influence both mentalizing and mirroring, and (iii) that power might affect the flexibility to switch between mentalizing and mirroring skills.

## **HOW MIRRORING AND MENTALIZING MAY BE DIFFERENTLY INVOLVED IN EA TASKS: A HYPOTHESIS**

Different EA tasks may require a perceiver to infer emotions of others, guess what they are thinking, and understand what their intentions and motives are, among others. Mirroring and mentalizing may differently affect each of these aspects. In this section, we illustrate how the tasks used in the studies assessing EA of high and low power people may require mentalizing or mirroring skills (see **Table 1** for a summary of the studies on this topic).

Some studies used simple recognition of facial expression of emotions to assess EA. This is a very simplistic measure of EA that might not take into account its entire complexity. Simple expression recognition might rely more on mirroring than on mentalizing. Indeed, a number of studies (Dimberg et al., 2000; Hess and Blairy, 2001) found that when participants are presented with pictures of emotional expressions, a facial mimicry response, which is supposed to rely on the mirroring system (Catmur et al., 2008; Heyes, 2011), is automatically elicited. Mentalizing might be less critical than mirroring for facial expression recognition. Even though contrasting findings have been reported (Uljarevic and Hamilton, 2013), some studies showed that children with autism can recognize facial expressions as accurately as typically developing children (Castelli, 2005; Rosset et al., 2008). Autistic children typically have impaired mentalizing skills and the fact that they are able to correctly recognize others' emotions suggests that mentalizing may not play a crucial role in emotion recognition.

In studies in which participants are tested in real-time faceto-face interaction settings, one interaction partner infers the other's feelings during the interaction. Even though this is perhaps a more naturalistic way of testing social variables, there is no control of the mimicry response of participants and therefore of

#### **Table 1 | An overview of the studies investigating the relationship between power and EA.**


*In the power section, we list how power was manipulated. Role play means the assignment of a participant to a high or low power role in an interaction with a social partner. Priming refers to an implicit manipulation by means of exposure to social cues related to power. In the EA-related assessment section we describe how the components that might influence EA were measured. The setting reports whether EA assessment relied on a face-to-face live interaction or on a computer-based task (e.g., recognition of pictures of emotional expression). In the following column we describe the measure that was used to assess perceivers' behavior. For each study, we report which skills (i.e., mentalizing and/or mirroring) we hypothesize to be predominantly involved in the task that was used. In the last column we report the main finding of the study.*

the involvement of the mirroring system. The mimicry response is contingent upon situational factors (e.g., attitudes toward the social target, type of social interaction) and can influence both EA and perceived power. For instance, competitive interactions decrease the mimicry response (Lanzetta and Englis, 1989; Weyers et al., 2009). This could be relevant because hierarchical interactions might be competitive and the inhibition of facial mimicry can in turn impair emotion recognition (Oberman et al., 2007). Moreover, facial mimicry can also influence perceived power: people who mimic more in an interaction are perceived as more likeable (van Baaren et al., 2009) and therefore may be less dominant because likeability (Farley, 2008) and agreeableness (Lippa and Arad, 1999) are negatively related to perceived dominance.

Yet other paradigms might require mentalizing skills to a higher degree. In a paper by Galinsky et al. (2006, Study 1) participants were asked to draw an E on their foreheads right before a live interaction with a partner. Powerful people were less likely to draw the letter by taking the perspective of the interaction partner. Even though perspective taking is not a measure of EA per se, it has been suggested as a mechanism through which power might hamper accuracy in social judgments. Perspective taking can be considered a more inferential type of thinking and might therefore rely mostly on mentalizing skills. Muscatell et al. (2012) found that lower social status was related to greater activity in the mentalizing system while encoding social information. This might explain why low power people engage more in perspective-taking strategies than high power people.

In many of the studies that use experimental manipulation of power, participants are asked to recall autobiographical events related to power (Galinsky et al., 2006; Schmid Mast et al., 2009). With this type of priming, the strategy participants use to recall the events is not controlled, which may represent a confounding factor. Some participants might choose spontaneously to focus on contextual information of the recalled event, a strategy that would foster mentalizing skills and advantage those participants in subsequent tasks requiring inferential reasoning (e.g., perspective taking). This idea is supported by a neuroimaging study by Morelli et al. (2012), which shows that focusing on the context of an emotional event involves the mentalizing system more than the mirroring system. Instead, people focusing more on their own bodily sensations (e.g., the stress of being powerless) might elicit a mirroring response and therefore be more accurate in a subsequent task requiring mirroring (e.g., simple emotion recognition).

Taken together, differences in the tasks used to assess EA and to manipulate power might contribute to explain the contrasting finding concerning their relationship.

## **HOW POWER MAY INFLUENCE MENTALIZING**

Construal Level Theory (CLT; Liberman and Trope, 1998) draws on the concept of psychological distance. Distal entities (e.g., events far in time or space or hypothetical) are more remote from direct experience and therefore need a higher level of construal (i.e., the missing information needs to be taken from more proximal entities). CLT makes specific predictions about power and these have also been taken up by the social distance model of power by Magee and Smith (2013). Powerful people should feel more psychological distance and more dissimilar to powerless individuals (Liberman et al., 2007; Magee and Smith, 2013). Indeed a study by Lammers et al. (2012) provided support for this hypothesis by showing that high power primed people were less willing to collaborate with a social partner on a series of games than low power people. There is evidence that when people are judging targets similar to them, a more ventral region in the medial PFC is activated compared to when people are judging targets that are less similar to them (Mitchell et al., 2006). If high power people perceive low power targets as less similar to them, they might show reduced activation in the ventral medial PFC region, but increased activation in the more dorsal region identified by Mitchell et al. (2006), which correspond to an area typically involved in mentalizing. To the extent that the social distance between high and low power individuals increases, the high power individuals might therefore rely more on mentalizing skills to correctly assess others' thoughts and feelings.

It could also be argued that powerless people rely more on mentalizing than powerful people. Fiske's continuum model of power (Fiske and Neuberg, 1990; Goodwin et al., 2000) predicts that low power people focus their attention on high power people, whereas the latter are more self-focused. A meta-analysis by Denny et al. (2012) showed that a more ventral region of the medial PFC is associated with self-related judgments, whereas a more dorsal region is related to judgments about others. The dmPFC activation suggests mentalizing and indeed its activity is greater in low than high social status people when encoding social information (Muscatell et al., 2012). Further experimental research is therefore necessary in order to specifically test the effects of power on mentalizing.

## **HOW POWER MAY INFLUENCE MIRRORING**

Studies on facial mimicry can support the hypothesis of an influence of power on mirroring. Theories of embodied cognition claim that emotion recognition is achieved through an internal simulation of the perceived expression (Goldman and Sripada, 2005; Niedenthal et al., 2010). Even though mimicry might not be strictly necessary for emotion recognition (Hess and Blairy, 2001; Sander et al., 2007; Rives Bogart and Matsumoto, 2010; Mumenthaler and Sander, 2012), mimicry responses predict the perceived intensity of facial expressions (Sato et al., 2013). Moreover, interfering with mimicry can hamper emotion recognition accuracy (Oberman et al., 2007; Neal and Chartrand, 2011). Research on mimicry is important if we consider that high power individuals tend to be more expressive than low power people in live interactions (Snodgrass et al., 1998). Hsee et al. (1990) found that powerful individuals are more likely to display subordinate's feelings than vice versa. These findings suggest that high power people might engage the mirroring system to a higher degree than low power people when reading others' minds and this might explain their high accuracy on facial expression recognition tests (Schmid Mast et al., 2009, Study 3). In addition, one could argue that low power people mimic less because they might have more negative mood or attitudes (e.g., toward superiors) and this is known to reduce the mimicry response (Likowski et al., 2011). Again, further experimental research is necessary in order to test the effects of power on mirroring.

# **POWER AND FLEXIBILITY**

Hirsh et al. (2011) proposed a model that draws on Keltner et al. (2003) explanations of approach and inhibition. According to these authors, powerful people would be more approachoriented. This would activate the Behavioral Approach System, which in turn would decrease the conflict between competing responses (i.e., disinhibition). Thus, powerful people would behave according to their most salient response, which can be externally driven, when strong contextual cues are present, or internally driven, in absence of such cues. When the response is internally triggered, power can reveal the internal dispositions of the person. When there are strong contextual cues, Hirsh et al.'s model predicts that powerful people will be more responsive to the affordances required by the task. This is in accordance with the Situated Focus Theory by Guinote (2007), which posits that power fosters the attunement to the situation and increases flexibility. Thus, this suggests that powerful people may be more

# **REFERENCES**


studies of self- and other judgments reveals a spatial gradient for mentalizing in medial prefrontal cortex. *J. Cogn. Neurosci.* 24, 1742–1752. doi: 10.1162/jocn\_a\_00233


able to switch flexibly between mentalizing and mirroring skills, according to what is more relevant to task demands.

# **CONCLUSIONS**

Whereas current models of power do not seem to properly account for the heterogeneity found in the literature on the power-EA link, in the present paper we speculate that a focus on the differential role of mirroring vs. mentalizing could. This approach can guide future research in at least two directions. First, empirical studies might test specific hypotheses based on the mechanisms we propose here. We expect that the ambivalent effects of power on EA will be teased apart once the effects of mirroring and mentalizing on power and EA are taken into account. Second, future studies might be more cautious in the choice of the specific EA tasks. Indeed, assessing EA through tasks focusing mostly on mirroring or on mentalizing might dramatically influence the outcomes of a study.

of face-based emotion recognition. *Cognition* 94, 193–213. doi: 10.1016/j.cognition.2004.01.005


to dynamic emotional facial expressions and their influence on decoding accuracy. *Int. J. Psychophysiol.* 40, 129–141. doi: 10.1016/S0167- 8760(00)00161-6


effects on observers' vicarious emotional responses. *J. Pers. Soc. Psychol.* 56, 543–554. doi: 10.1037/0022- 3514.56.4.543


facial feedback modulates emotion perception accuracy. *Soc. Psychol. Pers. Sci.* 2, 673–678. doi: 10.1177/1948550611406138


of mimicry. *Philos. Trans. R. Soc. B Biol. Sci.* 364, 2381–2389. doi: 10.1098/rstb.2009.0057


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 May 2013; accepted: 28 June 2013; published online: 19 July 2013.*

*Citation: Bombari D, Schmid Mast M, Brosch T and Sander D (2013) How interpersonal power affects empathic accuracy: differential roles of mentalizing vs. mirroring? Front. Hum. Neurosci. 7:375. doi: 10.3389/fnhum.2013.00375 Copyright © 2013 Bombari, Schmid Mast, Brosch and Sander. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any thirdparty graphics etc.*

# Understanding and accounting for relational context is critical for social neuroscience

# *Elizabeth Clark-Polner 1\* and Margaret S. Clark2 \**

<sup>1</sup> Department of Psychology, University of Chicago, Chicago, IL, USA

<sup>2</sup> Department of Psychology, Trumbull College, Yale University, New Haven, CT, USA

#### *Edited by:*

Leonie Koban, University of Colorado Boulder, USA

#### *Reviewed by:*

James A. Coan, University of Virginia, USA Ellen De Bruijn, Radboud University Nijmegen, Netherlands Ivana Konvalinka, Technical University of Denmark, Denmark

#### *\*Correspondence:*

Elizabeth Clark-Polner, Department of Psychology, University of Chicago, 5848 South University Avenue, Chicago, IL 60637, USA e-mail: clarkpolner@uchicago.edu; Margaret S. Clark, Department of Psychology, Trumbull College, Yale University, 2 Hillhouse Avenue, New Haven, CT 06511, USA e-mail: margaret.clark@yale.edu

Scientists have increasingly turned to the brain and to neuroscience more generally to further an understanding of social and emotional judgments and behavior. Yet, many neuroscientists (certainly not all) do not consider the role of relational context. Moreover, most have not examined the impact of relational context in a manner that takes advantage of conceptual and empirical advances in relationship science. Here we emphasize that: (1) all social behavior takes place, by definition, within the context of a relationship (even if that relationship is a new one with a stranger), and (2) relational context shapes not only social thoughts, feelings, and behaviors, but also some seemingly non-social thoughts, feelings, and behaviors in profound ways. We define relational context and suggest that accounting for it in the design and interpretation of neuroscience research is essential to the development of a coherent, generalizable neuroscience of social behavior.We make our case in two ways: (a) we describe some existing neuroscience research in three substantive areas (perceiving and reacting to others' emotions, providing help, and receiving help) that already has documented the powerful impact of relational context. (b) We describe some other neuroscience research from these same areas that has not taken relational context into account. Then, using findings from social and personality psychology, we make a case that different results almost certainly would have been found had the research been conducted in a different relational context. We neither attempt to review all evidence that relational context shapes neuroscience findings nor to put forward a theoretical analysis of all the ways relational context ought to shape neuroscience findings. Our goal is simply to urge greater and more systematic consideration of relational context in neuroscientific research.

#### **Keywords: relational context, attachment styles, relationship histories, relationship types, relationship character, relationship stages**

## **INTRODUCTION**

What makes a thought, feeling, or behavior "social"? A reasonable criterion is that the thought, feeling, or behavior is social if it arises from an individual's interdependence with another person. In other words, a person's thoughts, feelings, and/or behaviors can be considered "social" if they influence and/or are influenced by another person's thoughts, feelings, and/or behavior. This definition highlights an important point: social acts cannot, by definition, be understood by focusing on one actor alone; they must be examined as *multi*-person processes – processes that are powerfully shaped by the nature of the interdependence that exists, or is desired, between persons – or, in other words, by what we here call relational context.

The study of social behavior using *any* methodology, including neuroscience methodologies, must take relational context into account. Researchers must consider not only the nature of the actor him or herself, but also the nature of the person with whom he or she is interacting, and, crucially, *the nature of the existing (or desired) relationship between them*. It is this last aspect of multi-person processes on which we focus in this paper.

Whereas large literatures have accumulated in which researchers explain interpersonal interactions in terms of the characteristics of the actor (the participant in a research study), or, alternatively, the characteristics of person with whom he or she is interacting (the target person), far less frequently do researchers account for, or attempt to explain behavior in terms of the characteristics of the relationship *between* the actor and the target. These characteristics are, however, patently important: many crucial social variables inhere primarily in relational ties (or perceived relational ties), rather than in individuals or in targets themselves. Variables such as interpersonal similarity, trust, commitment, empathy, hostility, felt obligation, and prosocial behavior make no sense outside the context of a relationship. Moreover, most variability in such contacts occurs between the different relationships within individuals' sets of relationships, rather than between individuals themselves. (We hasten to acknowledge, of course, that chronic differences in peoples tendencies to feel such things as trust, commitment or empathy, and to elicit them from others do exist. Yet even these differences are inherently relational in nature, in that they are best understood as having developed in the context of specific

relationships, and then continuing to have their impact in similar relationships.)

Failing to account for relational context in our research, we suggest, has consequences beyond simply precluding researchers from maximizing their knowledge about social behavior. By collectively neglecting relational context as a variable of interest, while letting it vary across studies, researchers risk producing a confusing literature. Furthermore, by ignoring relational context in their empirical work, individual researchers risk drawing conclusions that will have limited generalizability.

Many current theoretical and empirical neuroscience models of social behavior have been developed largely upon the basis of research in which the participants are interacting with, or acting in the presence of, other individuals whom they have not met before and likely never expect to see again (e.g., Sanfey et al., 2008). These models may or may not be valid when it comes to predicting the actions and interactions within the context of established relationships – which are among the most common, and consequential, actions and interactions people, execute every day. To be fair, a growing amount of social neuroscience research has been conducted in the context of ongoing relationships. Moreover, some has involved participants actually interacting with one another as neuroscientific measures have been collected. Yet often, even work such as this is not accompanied with a careful conceptual consideration of the nature of the relational context involved, nor does it involve intentional, experimental manipulation of relational contexts within studies, or even comparison of results collected within different relational contexts. Conceptual analyses and consideration of relational contexts, as well as studies involving manipulation of relational contexts and comparison of results across distinct studies that have addressed the same questions in differing relational contexts, are all necessary to allow the researchers to determine how relational contexts shape psychological processes.

In what follows, we make a case for greater consideration of relational context in neuroscience research. We begin with a brief overview of some ways in which relational context influences social behavior.We then illustrate our point – that relational context matters for social neuroscience – with empirical examples drawn both from the social neuroscience and behavioral literatures. Broadly speaking, we posit that: (a) much neuroscientific research (and much behavioral research as well) neglects relational context; (b) the large behavioral literatures on topics listed above provide strong evidence that relational context matters, and (c) that neuroscientists who have taken relational context into account have provided us with additional strong evidence that relational context matters.

We have chosen to focus primarily on research in three domains of social behavior to illustrate our points: (a) the expression and perception of emotion, (b) giving social support, and (c) receiving social support. In each case, the point we make is simple: relational context matters. We believe, however, that the importance of relational context goes far beyond these three domains and that it is influential not only with regard to many other social behaviors (e.g., attitude formation and change, conformity, prejudice and stereotyping, etc.), but also with regard to some seemingly non-social domains as well such as perception (Schnall et al., 2008)

and intelligence (Woolley et al., 2010). We elaborate on the latter point a bit below.

## **RELATIONAL CONTEXT**

As stated above, all social behaviors are, by definition, interpersonal (as the saying goes, it takes two to tango!"). Anything that is interpersonal, furthermore, takes place within the context of a *relationship*. This is true even if that interaction is between people who have never met before, will never meet again, and who have no acquaintances in common – this is simply *one type* of relationship (one that exists between strangers who expect to remain so). This does not mean that interactions can only occur between two individuals (they could take place, for example, between an individual and a group, or between two groups), but rather that they *cannot* occur within one person, in isolation. The other person may be (and typically is) present but need not be; for instance, one can act in such a manner so as to benefit a person who is not present. The point remains that the study of social behavior must take into account not only the actor, but also the person with whom she or he is interacting, and the nature of the existing, desired or past relationship between them.

The relationship aspect of this equation is particularly important because the nature of the relationship that exists between people is a key determinant of the norms governing interactions, how partner behavior will be interpreted, what duties are felt toward the partner, how much attention will be paid to the partner and the list could go on. Relational context influences not only if, when and how people will act socially toward one another, and how they will respond to that other's behavior but, importantly, it also defines what counts as"pro-,""anti-," or neutral social behavior.

Consider a simple example: what happens when a beautiful bouquet of flowers is delivered to a woman's home? How will she react? It depends on relational context. If the flowers come from a suitor to whom she is attracted, and if she has been hoping the attraction is mutual, acceptance of the flowers and joy will result. If they come from her spouse of 30 years who has sent flowers every week for all those years, she will accept the flowers but may have no emotional reaction. If they come from a suitor who is nice enough but in whom this woman personally has no interest, reluctance to accept them, distress and perhaps feelings of guilt may arise. If they come from a person who has been stalking the woman and against whom she has a protective order, she will refuse the flowers and feel fear and distress. The point is straightforward: behavior, cognition, and emotion depend on relational context.

Before we can examine how relational context influences social behavior in more detail, however, it will be useful to understand six ways in which relational context can vary.

#### **SIX WAYS IN WHICH RELATIONAL CONTEXT VARIES**

When relational context has been noted in recent neuroscientific research, it is often referred to in terms of relationship types, with those types labeled in lay terms. Researchers have examined, for example, mother–child relationships and romantic relationships (e.g., Ortigue et al., 2010). Relational context certainly *can* be defined in lay language, yet that leaves open the questions of

just how these relationships differ, *conceptually*, from one another, whether there is meaningful variation within a group of relationships labeled with the same term, and how relationships given different names overlap with one another in conceptual ways. We believe understanding relationships in conceptual terms to be crucial to the study of social life generally, and to social neuroscience in particular (Clark et al., in press). We specifically suggest that social and affective neuroscience (and, indeed, many areas of psychology) will benefit by first considering six ways in which relational context varies, and then considering conceptual variation within each. (Often researchers will be able to capture the same conceptual variation in different methodological ways if they explicitly consider all six ways in which relational context varies. For instance, trust of partners will vary both within an individual according to relationship type, between individuals' in terms of chronic tendencies to trust other people, and within one individual's specific relationship with another as that relationship develops.)

The ways in which relational context vary are: (1) the type of relationship; (2) the character (or "personality") of the relationship (as distinct from the personalities of individuals involved in the relationship); (3) chronic individual differences in members' orientations toward relationships; (4) the history of the relationships; (5) the developmental stage of the relationships; and (6) the broader relationship network within which a particular relationship being studied is embedded (Clark et al., in press). In what follows we provide a short description of each way of thinking about relational context.

#### *Relationship type*

As noted above people (including many researchers) tend to think of relationship types in lay terms. They talk about, for example, their friendships, romantic relationships, parent–child relationships, and work relationships. Yet relationship types have been defined differently within the social psychological literature that deals specifically with studying and characterizing these different types of interactions. One way relationships are commonly characterized by relationship scientists is in terms of the norms, implicit and explicit, that govern interactions with others in those relationships (Clark and Mills, 1979, 1993, 2012; Fiske, 1992). These norms arise from the social function(s) people play in one another's lives (Bugental, 2000; Clark and Mills, 2012). Alternatively these can be conceptualized as the goals that people pursue, ultimately or proximately, in a given relationship.

In many cases, the goals that people pursue in given relationships and the social functions that relationships serve in their lives do differ along traditional lay-defined relationship lines. Friendship, for example, often serves the function of providing both members with a sense of security based on each member following an implicit rule to provide the other with non-contingent support aimed at maintaining and promoting the other's welfare. A romantic relationship may serve this same function but, importantly, it serves another function as well, providing for sexual gratification. So too does it serve the (ultimate) function of preserving genes by helping people reproduce and to raise children to the point of sexual maturity and reproduction themselves (see Bugental, 2000). Thus, lay terms do capture some important

variance in relationship type, *but* lay language does not make it clear just what is being captured, conceptually. If researchers think more in terms of social functions of relationships and of how those functions manifest themselves, they will be better off scientifically.

For example, as just stated, people rely on some relationship partners to be non-contingently responsive to their needs. These relationships (known as communal relationships) provide people with a sense of security and flexibility in seeking as well as giving support. Friendships, romantic relationships, and family relationships often (but not always!) exemplify communal relationships. In other relationships people do not assume nor desire such non-contingent responsiveness to needs and desires. Yet they may still wish to seek and give support to partners in a different, less committing may. Imagine that a person's drains are clogged and that person seeks a plumber's assistance. In such a case the person may wish to form what has been called an exchange relationship (Clark and Mills, 1979, 1993; Clark and Aragon, 2013) in which that person can seek support and provide compensation. In exchange relationships, individuals provide benefits to each other with the expectation that these actions will be repaid; the benefits are given contingently, and the individuals feel no particular non-contingent and ongoing responsibility for each other's welfare, beyond that which they feel for any other person (Clark and Mills, 1979; Mills and Clark, 1982).

The communal/exchange distinction, however, is just one conceptualization of relational context among many that may prove useful to neuroscientists. Other typologies capturing distinct social functions – ones that categorize relationships in terms of power or authority differences, in terms of sexual orientation, in terms of genetic relatedness, for example – will prove useful for different purposes. Our point is not to cover them all but just to urge researchers to think more in conceptual terms about relationship types and less in lay language terms, and to use the extant literature on relationships in so doing. Although there currently is no one scientific typology of relationship types that will adequately serve all research purposes (the research questions must guide the selection or generation of useful conceptualizations), we refer readers to Bugental (2000) for an example of a particularly clearly laid out typology regarding the various social functions a relationship may serve. She distinguishes relationships in terms of those which serve to keep us safe, those that allow collective acquisition and defense of resources and territories, those that promote mating, those involving reciprocity to maximize joint outcomes and those that allow us to optimize welfare by unequal distribution of power. She also discusses what sorts of information, neuro-hormonal regulators and social–emotional responses are relevant to each, as well as issues of development relating to each.

When thinking about relationship types, it is important to keep in mind that for purposes of conducting empirical work they can be captured both in terms of the distinct natures of existing, ongoing relationships as some researchers have done (e.g., Ortigue et al., 2010) but also that enacted, expected, and/or desired relationship types can effectively experimentally manipulated [see, for instance, Clark (1986) for a description of an experimental manipulation that effectively varies whether participants desire a communal or

exchange relationship with a target person or De Bruijn and von Rhein (2012),Koban et al. (2010), and Radke et al. (2011)for other examples of effective experimental manipulations of relationship type]. Each approach has advantages and disadvantages. Use of existing relationships captures naturally occurring differences between relationships but often lacks control and leaves room for alternative explanations of observations. Conducting true experiments in which expected or desired relationship types are manipulated provides for more control but likely will not be feasible for studies of relationship types that take days, months, or years to develop.

Regardless of what strategy is used, relationship type is likely to account for a great deal of variability in how we express and perceive emotions, how and when we give and accept support and empathize with others, and in how we interact with others in economic or strategic situations. Even so, there is more to relational context beyond just these distinctions.

#### *Relational character*

To truly capture relational context one must also account for what amounts to the personality of a relationship, or*relational character*. Just as there are many aspects of an individual's personality, so too are there many aspects of relational character that vary across, as well as within, relational types.

Take,for instance, communal relationships, which we described above. Clark and Mills (1979) identified communal relationships as those relationships in which partners provide benefits non-contingently, in support of one another's welfare. Within this general category, however, relationships can vary in terms of communal strength, or, in other words, in the degree to which one assumes responsibility for the other's welfare (Mills and Clark, 1982; Mills et al., 2004; Clark and Mills, 2012). Communal strength can be indexed by the effort, time, and cost one is willing to expend, in the service of (non-contingently) promoting the partner's welfare. Communal strength also can be indexed by the relative priority one assigns to caring for a specific partner when one has multiple, communal relationships, the demands of which may conflict. This aspect of relationship character is central to determining the giving of support and feelings of guilt and to determining levels of distress when support is not given.

Communal strength, however, is just one of many aspects of relational character to which neuroscientists might profitably attend in their studies of social and emotional phenomena. Interdependence theorists, for instance, have discussed variation between relationships in terms of the degree to which people's routines of thoughts, emotions and behaviors are dependent upon those of their partner. They index that degree of interdependence by how frequently, strongly, and in how many distinct ways members of the relationship influence each other's routines (Kelley, 1979; Berscheid et al., 1989). Other examples of relational character include two persons' *similarity* (along any of many possible dimensions; Amodio and Showers, 2005), the *trust* that exists between them (Simpson, 2007), the *certainty* that each person has regarding the existence and nature of their relationship (Clark et al., 1998), their *commitment* to remaining together (Rusbult, 1983), and the degree to which the partners are *satisfied* with the relationship (Hendrick, 1988). These dimensions of relational character overlap somewhat empirically as well as conceptually. There are not right or wrong ways in which to characterize relational character. What is important is to consider relational character in conducting social neuroscience work and to consider how it may shape whatever aspects of affect, cognition, and/or behavior are being studied.

As with relationship types, it is important for researchers to keep in mind that relational character can be measured and that it also can be effectively manipulated in many cases. We refer readers to Lamm et al. (2009) for a neuroscientific study in which the similarity of participants' pain experiences to those of a target person was effectively varied (and in which similarity did have significant effects on participants' empathic reactions to targets).

#### *Individual differences in approaches to relationships*

The third aspect of relational context is the chronic nature of individuals' orientation toward their relationships *in general*. People are known to vary in prosocial orientation (e.g., Batson and Shaw, 1991; Grant and Mayer, 2009), in chronic levels of relationship insecurity captured by anxious and avoidant attachment styles (Mikulincer and Shaver, 2007), in rejection sensitivity (Downey and Feldman, 1996), in empathic self-efficacy beliefs (Alessandri et al., 2009), in communal orientation (Clark et al., 1987), in self-esteem (Leary and Downs, 1995) and the list could continue. These individual differences (many of which are interrelated though, to date, no one has documented fully the extent to which they overlap and/or are independent of one another) manifest themselves in the ways people relate to more than one relationship, perhaps to all social relationships or to all social relationships within a category of relationship types. These individual differences also may interact with situational factors, including relationship types, to determine attention to others' feelings, needs and desires, and reactions to receiving support.

Individual differences also may influence a person's ability to effectively form and carry out the functions of any given relationship type. For instance, a person characterized by avoidant attachment may have difficulty forming and carrying out the functions of a communal relationship (see Simpson et al., 1992 for a particularly clear and compelling illustration of this point). This highlights another reason for researchers to be cautious regarding using lay language terms to characterize relationship types. All people may say they have friends, for instance. Yet not all relationships called friendships will be characterized by members feeling secure in partner responsiveness and being able to effectively give and receive responsiveness. In other words, not all relationships called friendships can be assumed to be the same conceptually. The very nature of "friendships" will almost certainly differ with the individual differences that people bring with them to those relationships (see Clark and Lemay, 2010 for a full description of this).

These individual differences constitute a part of what we mean by relational context, and constitute a distinct way of considering relational context both from relationship type and from the relational character of specific relationships. Still, in many cases, the individual differences that people bring to relationships will blend and interact with relationship type and relationship character to predict psychological reactions and processes1.

#### *Relationship histories and anticipated relationship futures*

Existing relationships also have histories that must be taken into account. A relationship's history plays a role in shaping both partners' behavior, their perceptions of events, and their reactions to those events. History with a partner creates expectations, and these expectations influence how one behaves toward a partner (Baldwin, 1992). An established pattern of interdependence with a partner, for instance, leads to firm expectations for future behavior, and can set the stage for feelings of emotions (both positive and negative) when those expectations are broken (Berscheid and Ammazzalorso, 2001).

Relationship history with one partner can also influence thoughts, feelings, and behavior in a different relationship (Coan et al., 2013a), especially if a current partner reminds a person in some way of a past partner (Chen et al., 2013). Relationship histories that influence new relationships more generally may sometimes be best conceptualized as individual differences in approaches to relationships (our category #3 here).

The anticipated *future* of a relationship also can shape social desires, emotional judgments, and behavior in the present (cf. Clark and Mills, 1979). Receiving a gift from someone with whom one anticipates forming a friendship or romantic relationship, for example, will elicit a different response than receiving the same gift from someone whom, one is certain, one will not be seeing again in the future.

#### *Developmental stage of relationships*

The developmental stage of a relationship also may have important consequences when it comes to social behavior2. Relationships have a time course that interacts with their functions and goals. All relationships change over time (Mitnick et al., 2009). Friendships or business relationships between peers, for example, will have establishment, maintenance, and – perhaps – deterioration stages. Another (overlapping) way of thinking about relationship stages is in terms of there being a deliberative stage of a relationship (involving deciding whether one wishes to be in the relationship and what type of relationship one desires) and an implemental stage (involving implementing the appropriate behaviors within an established relationship type; cf. Gollwitzer et al., 1990; Gagne and Lydon, 2001a,b). The stage of a relationship is an important predictor of the social and emotional processes that will occur in that relationship (Gagne and Lydon, 2001a,b; Beck and Clark, 2010; Clark and Beck, 2011). Social and affective reactions to others change across time as relationships unfold.

#### *Placement of a relationship within wider relationship networks*

The last aspect of relational context worthy of mention concerns the placement of a particular relationship within each person's larger set of relationships – i.e., in their social *network*. Just as individual interpersonal interactions occur within the context of specific relationships, so too do relationships function within the context of a person's larger social network, and this also will influence social and emotional behavior. For instance, attention to, and favorable judgments of, the physical attractiveness of potential romantic relationship partners have been shown to be decreased by the existence of, and commitment to, an existing romantic relationship partner (Johnson and Rusbult, 1989) and this moderation is manifest even in very fast, automatic, and non-conscious processes (Maner et al., 2008). A familiar face should be more comforting when spotted in the context of many unfamiliar faces in a new social situation (in which people tend to be anxious) than when seen in the context of many other familiar faces in established social situations in which people are happy and comfortable (cf. Vanbeselaere, 1980; Mikulincer et al., 2002; DeVries et al., 2010). A person who is perfectly comfortable with a friend when they are alone as a pair may be embarrassed by being associated with that same friend when in the company of additional peers (cf. Fortune and Newby-Clark, 2008).

In sum, relational context is complex. It includes relationship type, relationship character, individual differences in orientations toward relationships, relational history and stage, and the placement of a given relationship in the wider context of a person's other relationships. No matter what a researcher's substantive interests, we believe it will be useful for that researcher to consider all these aspects of relational context. A person interested in empathy as a process, for instance, may wish to consider variation in empathic processes between types of relationships, within types of relationships, at different points in a relationship's history and stage as well as how empathy is influenced by the wider social network into which a relationship fits. We have little doubt that an explicit consideration of conceptual variables as they are captured in each type of relational context will prove to be useful and, when taken into account in planning and conducting research, will make it easier to integrate findings, both within neuroscience, and across neuroscience and relationship science more broadly.

## **THE IMPORTANCE OF CONSIDERING SOCIAL NEUROSCIENCE FINDINGS IN RELATIONAL CONTEXT**

Here we have already argued that (1) all social behavior takes place, by definition, within the context of relationships, and that (2) relational context often affects the nature of results obtained in neuroscientific and other studies of psychological processes. In order to illustrate these points, we have chosen to discuss, as examples, a few specific types of neuroscience research on a few measures of social behavior. First, we examine relational context and the expression and perception of emotion. Next we look at the impact it has on empathy, and then we consider the impact it has on the giving and receiving of social support. Finally, we will briefly comment on how relational context influences even some seemingly non-social thoughts, feelings and behaviors. In each case, the point we will make is simple: relational context matters.

<sup>1</sup>Social neuroscientists more frequently have taken this aspect of relational context into account than other aspects of relational context (see, for instance, Coan, 2008; Vrticka et al., 2012), but attending to this one aspect of relational context cannot completely substitute for attending to the other aspects as well.

<sup>2</sup>There is overlap in the history of a relationship and the developmental stage of a relationship, of course. Still, there is value in considering them separately.

Specifically, we suggest that: (a) much neuroscientific research has ignored relational context; (b) the largely separate behavioral literature in this area provides strong evidence that relational context matters; and (c) those neuroscience studies that *have* taken relational context into account generally show that relational context does matter.

#### **PERCEIVING AND REACTING TO OTHERS' EMOTIONS**

There is now a large literature on the neural correlates of emotion perception. Within this literature, there is growing variation in the paradigms used (see discussions in, for example, Barrett, 2006; Scherer et al., 2011). Of particular importance is the fact that, in the vast majority of cases, the stimuli utilized in emotion perception paradigms have been depictions of people who are strangers to the participants (see Ebner et al., 2012; Montoya et al., 2012 for some recent examples but many more exist). Moreover often these are strangers whom participants actually are not meeting and whom they never expect to see again. This is the case in studies of reactions to facial (e.g., Xaoyun et al., 2009), vocal (e.g., Baum and Nowicki, 1998), and bodily (e.g., Coulson, 2004) expressions of emotion, and in cross-cultural (e.g.,Yik et al., 1998), developmental (e.g., Camras et al., 2002), and clinical (e.g., Anderson and Phelps, 2000) research as well. This is troubling, because the broader psychological literature on relationships (as well as a growing amount of the neuroscience literature) provides good reason to believe that relational context has a major impact on how people perceive and react to others' emotions.

#### *Perceiving emotion: some findings that may not extend beyond perceiving emotions in strangers whom one will never see again*

When people encounter *strangers*, the primary function that emotion perception may serve is to protect or promote the self's well-being by, say, avoiding angry people, approaching smiling people, and looking around to detect problems and to protect the self when someone else seems fearful (Klinnert et al., 1986). However, the functions that others' emotional expressions serve become more complex when relational context is considered.When people know one another, when they are interdependent, and when they assume responsibility for one another's welfare, others' facial expressions continue to be signals that can be used to protect or promote the self, but they take on additional functions as well – one important one being that they serve as signals of the other person's welfare and, as such, as signs that the perceiver should provide care to the other person. If a person is fearful perhaps one might reassure the other person, help the person distance him or herself from the feared stimulus, or remove the feared stimulus from the environment. If a person is happy, one might ask why and celebrate with the person, thereby prolonging the person's happiness, leave the person alone to continue enjoying whatever is causing the happiness, or repeat one's own actions if they were the source of the happiness as is appropriate to the situation.

For example, consider the implications of taking account of whether there is an existing caring (or communal) relationship – or the desire to establish a relationship – between a perceiver and an emotional target person for interpreting one recent study of reactions to others' facial expressions. N'Diaye et al. (2009) had 24 individuals look at target faces expressing happiness, fear, or anger, in either mild or intense forms. They also varied the direction in which the target face was gazing, such that the person was either looking directly at the perceiver, or off to the side. Functional magnetic resonance imaging (fMRI) scans were collected. The researchers predicted that, for the faces expressing fear, the perceived self-relevance of the expression would be higher when the gaze was averted (which suggests that there is something in the environment for the self to fear) rather than when it was direct (suggesting that the person feared the participant), but that, for faces expressing anger, the perceived self-relevance would be greater when the target face was looking directly at the participant (which suggests that the emotional target is angry at the perceiver) than when it was averted. These predictions were supported. Behaviorally, ratings of emotion intensity were greater when fear was expressed with an averted gaze than with a direct gaze, and when anger was expressed with a direct gaze than with an averted gaze. The same pattern was reflected in terms of the neural correlates of emotion perception, in the amygdala, as well as in the fusiform and medial prefrontal cortices.

These results are intriguing, and the explanations makes good sense *given* that the target persons were strangers to the participants, which, in this study, they were. We watch out for and protect *ourselves* when with strangers. However, what if the target person had been a friend, a romantic partner, or a family member? What if he or she were simply someone attractive, with whom a participant desired a relationship? Then, viewers' reactions to the emotions expressed, and the interaction effect with gaze direction that has been observed, we suggest, almost certainly would have been different.

Importantly, those who willingly express emotions to us are choosing to convey rather than to suppress information about their own well being (Clark et al., 2001). They are often in need of our support and, if we have assumed responsibility for their welfare (or wish to do so going forward), people often switch their relational focus of attention from themselves and the implications of others for themselves to the partner and what they can do for that partner (Clark et al., 2008). Perceivers then respond to others' negative emotions with care (Clark et al., 1987; Graham et al., 2008). To give an obvious example, parents typically respond to an infant's cries by shifting attention to the child, focusing on the child's needs and providing care. Spouses and friends would likely do the same.

When a fearful face looking right at us belongs to a person for whom we have assumed or would *like*to assume responsibility, that should trigger a shift in relational focus of attention (Clark et al., 2008) *from* the self *to* the other person. The face then becomes an implicit request for help,*especially* in the direct gaze condition. Our reactions ought to be just as intense (or perhaps more intense), and likely different in nature both from those captured when looking at a fearful stranger, and from those captured when the person is gazing somewhere else (then, just as in the original study, we may react on our behalf as well as on their behalf). And what about angry gazes? In an intimate relationship in which another's emotion signals us that our partner has a need to which we should attend, an angry person's averted gaze may be interpreted as a call for help from us. In the context of a communal relationship, an angry averted gaze may elicit just as much of a reaction from us

as does an angry direct gaze and, importantly, a distinct kind of reaction. We may wish to come to the aid of the person with an angry diverted gaze; we may still wish to protect ourselves when we perceived an angry direct gaze (suggesting our partner is mad at us) but when we care for the other's welfare that reaction may be mixed with some feeling of responsibility for the other (especially if we did cause the anger and are in a secure, well-functioning communal relationship; cf. Yoo et al., 2011).

#### *Perceiving and reacting to emotion (or events likely to have elicited emotion): data showing relational context does matter*

A case for routinely taking relational context into account in interpreting studies of reactions to others' emotional faces, and in planning for new studies also can be made on the basis results of existing neuroscience studies, that already have incorporated facets of relational context into their designs. For instance, although some have suggested that responses to self-related emotion do not overlap a great deal with responses to others' emotion (e.g., Jackson et al., 2006), a recent study reported by Beckes et al. (2012) found that whereas threats directed at strangers produced neural responses quite distinct from those directed at the self; threats directed at *friends* produced neural responses that overlapped far more with those produced by threats directed at the self. Relationship type mattered. As Aron et al. (1991) have shown using behavioral measures, when we care about close others we often "include the other in the self " cognitively.

Work by Barrett et al. (2012) and Singer et al. (2006) also is instructive in terms of the import of relational context. Barrett et al. (2012) used fMRI to examine the effects of infant facial expressions (both positive and negative) on adults. In conducting this research they considered three of the six dimensions of relational context highlighted earlier in this article: relationship type was varied experimentally (i.e., participants viewed their own infant or someone else's infant who was unknown to them), and aspects of relationship history (in this case participants' history of mood and anxiety during their own postpartum period while relating to their child) as well as the participant's own orientation toward close relationships (in this case attachment styles) were measured.

Both relationship type and participants' histories of relationships with their own infant influenced neural reactions to pictures of infants. Regarding relationship type, participants showed greater BOLD responses in the postcentral gyrus, subgenual anterior cingulate gyrus, ventral putamen, and superior temporal gyrus in response to their own infant's negative expression than in response to an unknown infant's negative expression.

More interesting is another finding the authors reported. It was that that poorer postpartum quality of the participants' maternal experience (a variable that picks up both relational history and, likely, chronic individual differences in participants' orientation toward the relationships with their children) was significantly related to reduced amygdala response to participants' own infants' positive facial expressions relative to participants' reactions an unknown infant's positive expressions. This is a fascinating result. Perhaps all mothers, depressed and anxious or not, simply must attend to distressed infants but a history of stress, depressed moods

and anxiety during the postpartum period selectively reduces some mothers' tendencies to see their own child's positive emotional expressions as significant. That is an important because other research suggests a child's happiness is an important signal to caretakers (Clark and Monin, 2014), and that happiness does capture most people's attention (Becker et al., 2011) and holds it (Power et al., 1982). Positive expressions suggest that a child is enjoying whatever is going on, thereby conveying information about what activities, foods and people a child enjoys and should be repeated or made use of when a child is not happy. The work reported by Barrett et al. (2012) suggests that all these functions of a child expressing happiness may be jeopardized among those with a history of postpartum depression, stress and anxiety. These specific results are interesting. For the present purposes, however, the overarching lesson remains that the meaning of facial expressions and neural responses to them are qualified by relational context.

Consider also the implications of the Barrett et al. (2012) findings for interpreting the results another recent study reported by Montoya et al. (2012). Montoya et al. (2012) collected neural scans that suggested that adults (in this case non-parents) experience happy infant faces as rewarding. Perhaps we are just generally built to find such faces rewarding but Barrett et al.'s (2012) results show us that in the situation in which such responses are surely most important (reacting to our *own* happy infant) that a poor relational history may cause this to go awry.

Research by Singer et al. (2006) also demonstrates that relational context matters to people's perceptions of others' emotional states. These researchers were interested in how research participants would react to watching a confederate in a laboratory who is experiencing pain and presumably distress. Others have examined this as well but Singer et al. (2006) added a twist to the study by *experimentally* varying relational history. Specifically, prior to viewing the confederate experiencing pain the experimenter randomly assigned half the participants to be treated fairly by the confederate in an economic game and half to be treated unfairly. That manipulated relationship history made a difference. Both male and female participants exhibited empathy-related activation in pain-related areas (fronto-insular and anterior cingulate cortices) when seeing previously fair players experience pain. However, when viewing the unfair confederate in pain and, presumably, distress, these responses were significantly lower among the male (but not the female) participants. Not only that, these males also showed increased activation in *reward* related areas when seeing the previously unfair confederate in pain and this activation correlated with these male participants' expressed desire for revenge. Relationship history – in this case, very recent relationship history – mattered a lot for these males.

Also worthy of note is a study by Vrticka et al. (2009). It too demonstrates that relational history can be manipulated and that people's histories of interactions with others can influence how they perceive faces going forward. These researchers manipulated people's exposure to smiling or angryfaces in the context of a game. Later they exposed the same people to these faces (now with neutral expressions) along with other neutral faces while using fMRI to record responses. As one might now expect, relational history made a difference. The results revealed that regions involved in recognizing familiar faces – the fusiform cortex, posterior cingulate gyrus, and amygdala, as well as motivational control areas such as the caudate and anterior cingulate cortex (ACC), were differentially modulated as a function of whether prior encounters with the face had been in a friendly versus unfriendly context. These results illustrate the impact of relational history well, and also show that it can be effectively experimentally manipulated.

Perhaps the most frequent way that relational context has been taken into account in this area of neuroscience research is by measuring individual differences in orientations to relationships in studies, and the most commonly examined individual differences are attachment styles. In this regard, we would simply note that when these measures have been added to several studies of reactions to others' facial expressions, they have been shown to make a difference in the neural activity observed (see Vrticka et al., 2008; Suslow et al., 2009 for a few examples).

#### **GIVING SOCIAL SUPPORT**

Psychologists – particularly social psychologists, but also sociologists, anthropologists, developmental psychologists, and economists – have long studied helping and other forms of what is commonly called prosocial behavior (Dovidio et al., 2006; Schroeder and Graziano, in press). More recently, a large neuroscientific literature has emerged on this topic. There are, for instance, studies using fMRI that have focused on identifying the neural correlates of reactions to signs of others' needs (e.g., reactions to a picture of a sad other for instance; Kim et al., 2009). There also has been research aimed at eliciting actual prosocial behavior, so as to identify the conditions under which is it likely (and unlikely) to occur (e.g., Kosfeld et al., 2005; Zak et al., 2007; Izuma et al., 2010). As is the case for many studies of reactions to others' emotions, the neuroscientific measures in much of this work (including all studies cited in the paragraph above) also have involved the collection of people's reactions to strangers whom participants have never seen before and likely never expect to see again. Further, it is quite common for researchers to select their stimuli from one of just a few standardized sets of stimuli and these stimuli often depict stereotypical expressions that may not adequately capture the nature of facial expressions that occur in normal, everyday, social interactions3.

#### *Indirect evidence that relational context matters*

We know from a now large behavioral research literature, that relational context plays a huge role in the degree to which we respond (or fail to respond to) other people's needs as well as in how we respond to others' needs (Clark and Aragon, 2013). We also know that relational context matters in terms of what elicits responsiveness, in the nature of responses that are elicited, and in how people act, after having provided support to another person (see Clark and Aragon, 2013 a review). People give more support to kin than to non-kin (Segal, 1984; Essock-Vitale and McGuire, 1985; Borgida et al., 1992; Burnstein, 2005), especially when support is

needed in life-threatening situations (Burnstein et al., 1994). They also give more support to those with whom they desire an ongoing relationship than to others (Clark et al., 1987). Motivations for giving support also vary – sometimes liking drives support giving; sometimes duty does so. Evolutionarily determined forces seem to drive some support giving; desire to establish business-like ties drives other forms of support giving (see Clark et al., in press). Relationship stage matters as well. We tend to give more support than we ask for as we strive to form desired relationships. In established relationships giving and seeking support tends to even out (Beck and Clark, 2009). This suggests that very early on in voluntary relationships – e.g., in friendships – help is, perhaps, given for selfish, strategic reasons; later on, it may be motivated primarily on the basis of partner need. The consequences of support giving also vary considerably. Sometimes we feel good about having helped (when it promotes relationships that are desired); sometimes we regret or feel gullible for having helped for instance when we help someone who is not special to us in any way (Williamson and Clark, 1989, 1992).

Individual differences in relationship orientations between people also matter a great deal. A striking example of this comes from research reported by Simpson et al. (1992). These scholars found that a person's response to a close other's needs depended on that person's own attachment style. Those who were low in avoidance reacted to partner anxiety as one might expect and as is socially functional. That is, the more anxious their partners were, the more support they offered. However, those high in avoidance reacted in an entirely different manner. The more anxious their partners, the *less* support they provided.

All this means that, in interpreting neural correlates of giving support we should carefully consider in what relational context the data have been collected and, consequently, what their limitations might be. It also recommends intentionally including relational context in the design of neuroscience studies.

#### *Direct evidence that relational context matters*

Indeed, when neuroscience research *has* included relational context in designs, it typically matters. Consider some results recently reported by Telzer et al. (2011). These researchers scanned people's brains while the people were choosing to give or not to give monetary rewards to family members. They found that decisions to give family members money were linked with activation in areas of the brain associated with self-control and mentalizing. More importantly for our point they also found that quality of the family relationships, or, in the terms we used earlier in this article, the relational character of the family relationships, mattered. Individuals who felt greater obligation to family members also showed greater functional coupling between regions related to self-control and those related to mentalizing within the ventral striatum, an area that, the authors report, is involved in reward processing. This suggests that it may well be effortful to support family members but that so doing is also rewarding for those who feel strong obligations to those family members.

For the present purposes the details of studies such as those reported by Telzer et al. (2011) and other studies in which relational context has been incorporated as a variable (see,for instance,

<sup>3</sup>We certainly recognize the value of standardizing stimuli within studies for purposes of experimental control and across studies for purposes of being able to compare results. Our point is simply that results from studies utilizing only strangers as stimuli and only stereotypical expressions run the risk of not being very generalizable.

Kim et al., 2011; Musser et al., 2012; Seifritz et al., 2013) are less important than are the overarching lessons taught by this research. Variations in relational context often will be associated with variations in the patterning of neural responses to partner needs and to actual responsiveness to partners. Moreover, only by combining knowledge about support giving, relational context and neutral responses may we come to understand just what neural responses to partner distress really mean.

Consider also lessons learned about the importance of relational context from studies on the neuropeptide oxytocin. Studies to date on the effects of experimentally varying intranasal exposure to oxytocin provide one of the most striking examples of the importance of taking relational context into account in neuroscientific studies of support giving. Early work on exposure to oxytocin and prosocial behavior involved the administration of oxytocin (or not) to individuals who were strangers to one another. Intriguing and dramatic effects emerged. One study utilized the now common ultimatum game paradigm in which one person is given money and then offers to split it any way he or she chooses between the self and the partner. Then the partner chooses to accept the split – in which case each person gets what was offered – or to refuse it, in which case each person gets nothing. In this study, participants who had been administered oxytocin intranasally made more generous proposals to "partners" (strangers) than did those who had not been exposed to oxytocin (Kosfeld et al., 2005; Zak et al., 2007). Some follow-up studies yielded similar effects (e.g., Israel et al., 2012). The initial study appeared in a prominent journal and received much media attention. Indeed, one author gave a widely shared TED talk (Zak, 2011) in which he declared oxytocin to be the "love hormone" and then wrote a popular book advocating a similar view (Zak, 2012). Subsequent papers urging therapeutic usage of oxytocin quickly appeared (cf. Striepens et al., 2011). Although many researchers were cautious in claims made, as just noted other researchers and many members of the media were not.

The problem was (and is) that if one-steps back and considers the literature more broadly, oxytocin does *not* always increase prosocial behavior. Indeed, not infrequently, it decreases prosocial behavior (Radke and de Bruijn, 2012 and see Bartz et al., 2011 for a review of literature on this topic). About 20% of the time intranasal administration of oxytocin actually seems to promote antisocial behavior (Bartz et al., 2011).

What predicts when oxytocin increases prosocial behavior and when it does not? The Bartz et al.'s (2011)review makes it clear that relational context is key. Intra-nasal administration of oxytocin seems to promote trust in and prosocial behavior toward benign strangers (Kosfeld et al., 2005) and toward liked others generally but *not* to outgroup members (De Dreu et al., 2011) or people suspected of being outgroup members (Radke and de Bruijn, 2012) or, necessarily, among people generally low in trust of others (Bartz et al., 2011). In other words, relational context (in this literature captured by variation in relationship types and by individual differences in people's orientations toward relationships) can flip the effects of oxytocin on participants' behavior changing them from pro- to anti-social in nature.

Genes also relate to variability in who does and does not respond to oxytocin with increased prosocial behavior (e.g., Poulin

et al., 2012). A nucleotide polymorphism involving a guanine (G) to adenine (A) substitution is linked to sensitivity to relationship context. People homozygous for the G allele (or, sometimes, having at least one copy of the G allele) seem to benefit more from partners' positive emotions and caring for them than those lacking G alleles (cf. Bakermans-Kranenburg and Van Ijzendoorn, 2008; Rodrigues et al., 2009) *and also* to be harmed more by a negative social context (e.g., abuse by others in early childhood (cf. Bradley et al., 2011) than those lacking G alleles. Although more research is needed, again, relational context, this time in the form of the relational character/relational histories in interaction with individual differences (here in genes), seem to matter a lot.

Importantly evidence that relational context makes a difference to the effects of oxytocin on social behavior spurred theorists to move away from declarations that oxytocin increases trust in and generosity toward others *per se*, toward more carefully considered explanations of just what the function and effects of oxytocin are. Perhaps, researchers now suggest, oxytocin increases the salience of social stimuli resulting in more positive reactions to safe, trusted, liked, or smiling people and more negative reactions to distrusted, disliked, or angry people (Rimmele et al., 2009; Sharnay-Tsoory et al., 2009). Alternatively, perhaps oxytocin increases approach tendencies (in both positive and negative ways) and damps down avoidant tendencies (Kemp and Guastella, 2011). No matter what the explanation turns out to be, it was the variability of results across relational contexts that forced theorists in this area to think about the oxytocin in a more nuanced way and that has already (and will continue) to result in better theory.

Once again, we would say: to understand neural correlates of prosocial behavior one must consider relational context. To build a good science of prosocial responding researchers must also combine the now large and rapidly expanding behavioral findings in this field by social psychologists, health psychologists, and developmentalists that takes relational context into account with neuroscience studies that also take relational context into account. It is happening to some extent in the literature; it needs to happen more and in more sophisticated ways.

#### **RECEIVING SOCIAL SUPPORT**

People not only behave prosocially toward others; they receive support or are the target of other prosocial acts from others most typically, in daily life, in the context of friendships, romantic relationships, and family relationships. As with the topics we have covered already, reactions to receiving various forms of prosocial behavior have begun to be studied by neuroscientists, often outside the context of ongoing relationships. Examples include two recent studies, one on the impact of receiving an apology (Strang et al., 2012) and another on the impact of receiving supportive text messages after having suffered exclusion (Onoda et al., 2009). Occasionally relational context has been taken into account in such work (see Coan et al., 2006 for an example of a study in which relational context was taken into account).

Can we generalize the results of studies of neural responses to support received from strangers to what is likely to happen when support is received from well known others? We suspect not. Consider Onoda et al.'s (2009)study of having received supportive text messages from a stranger after having been rejected by a different group of strangers. These authors adopted a manipulation of social exclusion used by many previous researchers (Williams and Jarvis, 2006). That is, participants played a game of cyberball with two other players. They were included in the ball tosses at first. Later they were excluded. The exclusion elicited activity in the ACC as has been found in other studies. Later they received emotionally supportive text messages. Those whose reported social pain was reduced also showed lowered ventral ACC activation and heightened left lateral prefrontal cortex activation. The authors suggest that for these people social support enhanced prefrontal cortex activity, which, in turn, dampened activity within the ventral ACC. But when and for whom will rejection be most and least painful? When and for whom will supportive messages be most and least helpful? The answers certainly depend in important ways upon relational context. Thus, can we safely generalize results such as those of Williams and Jarvis (2006)? Probably not.

One neuroscience study of support already provides proof that relational context can shape neural responses to receiving support (Coan et al., 2006 and see also Coan et al., 2013b for a follow-up on the original study). In these researchers' experiment, married women were told that they were in a study involving receiving shocks. On some trials they would be safe from shocks on other trials they might receive them. This occurred while participants were in an fMRI scanner. The experimenters varied whether the women received support in the form of handholding or not. Relational context also was varied in two ways. First relationship type was varied. Sometimes participants held a stranger's hand, sometimes they held their husband's hand, and sometimes they held no one's hand. In addition the relational character of participants' marriages was measured prior to the scanning session. In particular the women filled out measures of marital quality tapping satisfaction cohesion, consensus, and affection in their marriages.

Both relationship type and relational character moderated the women's neural responses to the prosocial behavior of someone holding their hand while they were under stress. Activation in the neural systems known to underlie emotional and behavioral threat responses was most attenuated when the women held their husbands' hands. A similar but less attenuated neural response was observed when they held the hand of a stranger compared to holding no one's hand. In addition, among those whose husbands held their hand, the higher the quality of the woman's marriage to her husband the less these neural areas were activated.

The behavioral literature on reactions to receiving support is long standing and fits well with Coan et al. (2006) and Coan et al. (2013b) findings. Beyond that it clearly shows that it is not just relationship type and quality that matter to people's reactions to receiving support. People welcome non-contingent social support when they are open to a close relationship with the support giver but prefer contingent support when we prefer a more formal business like relationship (Clark and Mills, 1979). In addition, if people are secure, they seek support when it is needed; but if they are avoidantly attached they retreat when they most need the support (Simpson et al., 1992). If they are secure they perceive support as having been given voluntarily; if they are avoidant, they seem biased to see support received as having been involuntarily given (Clark and Beck, 2011). If they want to be responsive to another they are more likely to see others as more responsive (holding

objective responsiveness constant) than if they do not have that desire (Lemay et al., 2007; Lemay and Clark, 2008) and the list could go on.

#### **RELATIONAL CONTEXT IS LIKELY TO PROVE IMPORTANT IN MANY ADDITIONAL DOMAINS**

In this paper, we have chosen to emphasize the importance of taking relational context into account by focusing on research in the topical areas of perceiving and reacting to others' emotions, giving social support, and receiving social support. These are among the most obvious substantive research areas for which relational context should matter. Yet it is important to note that the extant behavioral literature reveals that relational context can influence many, many types of thoughts, feelings, and behaviors, even ones that do not seem very social at all. Consider perceptions of the nature of one's physical environment, for instance. Schnall et al. (2008) conducted two studies in which relational context was found to influence how steep perceivers judged a hill to be. In a first study people who judged the steepness of a hill when with a friend judged the hill to be less steep than people asked to judge the same hill in the absence of a friend. In a second study participants were assigned to think about a supportive friend, a neutral person or a disliked person and then to judge the steepness of a hill. The hill was judged to be less steep after thinking about a friend than after thinking about a neutral or disliked other. Here relational context as indexed by relationship type mattered.

Alternatively, consider ability to perform a non-social task. Woolley et al. (2010) asked groups of people to solve tasks. The average individual IQ of group members and the highest IQ of any group member positively predicted performance, but weakly. However two relational context measures were substantially better positive predictors: the group members' ability to read one another's emotions [as indexed by Baron-Cohen et al.'s (2001) "reading the mind in the eyes" task and how evenly distributed participation in the task was (as indexed by the group members' turn-taking)]4. Here relational context as captured by individual differences in relational skills (in reading the mind in the eyes) and the relational character of groups (how evenly they shared tasks) influenced the groups' problem solving abilities.

The point of this brief section is simple. Relational context may influence outcomes (including outcomes assessed by neuroscientists) in many, many types of tasks including ones that may not appear to be social in nature. Thus, whereas taking relational context into account in neuroscience studies that are very clearly social in nature may be especially important, so too may taking it into account more broadly in neuroscience (and in other types of research as well) prove fruitful.

#### **CONCLUSION**

The overall points of this paper are simple: by definition *social* thoughts, feelings, and behaviors involve other people. All social thoughts, feelings, and behaviors occur within the context of a relationship with another person (even if the relationship is one between strangers interacting with one another for the first time who never expect to see one another again.) Much behavioral

<sup>4</sup>Performance also was better when more females participated.

social, developmental and clinical work and a growing body contemporary social neuroscience research provide evidence that the relational context within which people interact (together and in interaction with other variables) is a powerful factor when it comes to shaping social feelings, thoughts, and behaviors. Yet many neuroscientists (and many other types of researchers alike, including, somewhat surprisingly, social psychologists) still study humans in isolation or, at best, when they are interacting with strangers. Often the "other people" are simply pictures of strangers. Even when people are studied while actively interacting with others and even when those others are in an ongoing relationship with the participant, often the nature of the relational context is not varied within the study nor are results of studies conducted in such relational contexts compared across studies in which the same processes were observed in other relational contexts with the explicit goal of considering how relational context may have shaped the results. It is not sufficient simply to move toward studying effects in richer more naturalistic contexts. Relational context must be considered a variable that may (and often does) shape people's cognitions, emotions and behaviors.

If researchers are to build a coherent body of scientific knowledge about such things as empathy, support giving, support receipt (and the list could go on to cover many other topics), they must attend to relational context (including types of relationships, the character of relationships, individual differences in orientation toward relationships, histories and anticipated futures of relationships, relationship stage and where relationships sit within broader networks of other relationships). In so doing neuroscientists (and others) would be well advised to utilize, build upon and contribute to the now substantial relationship science literature. Researchers who have produced that literature have developed theory and solid empirical bodies of research characterizing these contexts in conceptual terms (Clark and Lemay, 2010; Simpson and Campbell, 2013).

Taking this literature into account is necessary to build a solid, generalizable, and integrated body of social neuroscience. In making this point we should (and do) acknowledge that, before us and continuing to the present, two other social psychologists, Ellen Berscheid and Harry Reis have urged the entire field of psychology to take relationship context into account in establishing and in integrating psychological knowledge (Berscheid and Reis, 1998; Reis et al., 2000; Reis, 2006, 2010). They have used somewhat different terms and arguments than we do here but, essentially conveying the same message. Moreover, others such as Guroglu et al. (2009) have called for studying neural correlates of social behaviors in relational context (in their case they specifically urge researchers to study these behaviors longitudinally across development as individuals acquire crucial social decision making skills). Still, this point is neglected sufficiently often that we feel it is well worthwhile to note the still largely individualistic nature of psychology studies generally (including neuroscientific studies) and to call for greater consideration of relational context.

Most recently, as we were concluding preparation of this manuscript for *Frontiers*, we read two other papers whose authors join us in making a call for more attention to relational context in research. First, Beckes and Coan (2013) published a review article on social neuroscience findings relevant to relationships in *The*

*Oxford Handbook on Close Relationships* (Simpson and Campbell, 2013). Within this paper they too call for integrating knowledge of relationships into social neuroscience. So too do Schilbach et al. (2013) call for more neuroscience work done in social context5.

We are delighted to have company. We endorse several of Beckes and Coan's (2013) suggestions for future efforts in this regard, namely that researchers should: (a) utilize relational context and, in particular "move toward the measurement of a larger variety of emotional and cognitive tasks in a relational context" because "many processes may diverge from current findings once they are tested in the presence of a loved one." (Beckes and Coan, 2013, p. 705), (b) realize that relationship processes unfold across time and consider conducting longitudinal research, and (c) realize that taking relational context into account will enhance the chances of social neuroscience contributing to the development of clinical interventions emerging from our work.

In concluding this paper, we would re-emphasize two additional points. Social neuroscientists and other researchers alike should not be content to think of relational context just in terms of relationship types identified in lay language terms (e.g., nonparents vs. parents, or loved ones versus non-loved ones). Neither will it be sufficient to think of relational context in terms of just one type of individual difference in people's approaches to relationships (e.g., attachment styles). We should be thinking of relationship context in clearly laid out conceptual terms and consideration should be given relational context in all its complexity including not just relationship type but also relationship character, relationship histories, relationship stages (this is *why* longitudinal studies are needed), as well as the placement of particular relationships in the context of other relationships.

Many of the conceptual variables that will prove central to our research (for instance, the trust between two people) will vary within relationships (as relational character), between relationship types, between individuals with different relationship orientations, with relational history and with the placement of relationships in larger networks of relationships. Thinking of how any construct in which a researcher is interested varies in each of these ways will allow that researcher to design studies that test hypotheses in different and more sophisticated ways and, in turn, to build a better and more integrated sets of findings.

As stated at the start of this manuscript, if social neuroscientists ignore relational context they risk establishing a body of social neuroscience that is narrow, of limited generalizability, and confusing. If relational context is taken into account as researchers build a social neuroscience of relationships, they stand a better chance of producing a coherent, integrated body of knowledge that will be intrinsically and practically valuable.

#### **REFERENCES**


<sup>5</sup>We thank Reviewer 2 (Ivana Konvalinka) of this manuscript for calling this article to our attention.


importance of the decision. *J. Pers. Soc. Psychol.* 67, 773–789. doi: 10.1037/0022- 3514.67.5.773


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 30 April 2013; accepted: 19 February 2014; published online: 25 March 2014. Citation: Clark-Polner E and Clark MS (2014) Understanding and accounting for relational context is critical for social neuroscience. Front. Hum. Neurosci. 8:127. doi: 10.3389/fnhum.2014.00127*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2014 Clark-Polner and Clark. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# The ethology of empathy: a taxonomy of real-world targets of need and their effect on observers

#### *Stephanie D. Preston1 \*, Alicia J. Hofelich1 and R. Brent Stansfield2*

*<sup>1</sup> Ecological Neuroscience Laboratory, Department of Psychology, University of Michigan, Ann Arbor, MI, USA <sup>2</sup> Department of Medical Education, University of Michigan, Ann Arbor, MI, USA*

*Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Melissa Duff, University of Iowa, USA Michael Lifshitz, McGill University, Canada*

#### *\*Correspondence:*

*Stephanie D. Preston, Ecological Neuroscience Laboratory, Department of Psychology, University of Michigan, 530 Church Street, Ann Arbor, MI 48104, USA e-mail: prestos@umich.edu*

Empathy is inherently interpersonal, but the majority of research has only examined observers. Targets of need have been largely held constant through hypothetical and fictionalized depictions of sympathetic distress and need. In the real world, people's response to life stressors varies widely—from stoicism to resilience to complete breakdown—variations that should profoundly influence the prosocial exchange. The current study examined naturally-varying affect in real hospital patients with serious chronic or terminal illness during videotaped interviews about quality of life. Participants viewed each video while psychophysiological data were recorded and then rated each patient's and their own emotion. Patients displayed three major emotion factors (disturbed, softhearted, and amused) that were used to classify them into five basic types (*distraught, resilient, sanguine, reticent, wistful*). These types elicited four major emotions in observers [personal distress (PD), empathic concern (EC), horror, pleasure], two of which were never discovered previously with fictionalized targets. Across studies and measures, *distraught* targets usually received the greatest aid, but approximately as many observers preferred the positive and likeable *resilient* patients or the quietly sad *wistful* targets, with multiple observers even giving their greatest aid to *sanguine* or *reticent* targets who did not display distress or need. Trait empathy motivated aid toward more emotive targets while perspective taking (PT) motivated aid for those who did not overtly display distress. A second study replicated key results without even providing the content of patients' speech. Through an ecological examination of real need we discovered variation and commonality in the emotional response to need that interacts strongly with the preferences of observers. Social interactions need to be studied in ethological contexts that retain the complex interplay between senders and receivers.

#### **Keywords: empathy, altruism, perception-action, prosocial, sympathy, compassion, helping**

The prosocial response has been studied in social, personality, and developmental psychology for decades, revealing largely consistent findings across researchers and populations (reviewed in Eisenberg and Miller, 1987; Preston and de Waal, 2002; Batson, 2011). In order to reliably elicit prosocial responses in the laboratory, virtually all studies used sympathetic, fictional, single targets of need depicted through written narratives, confederates, or actors featuring blameless young children, orphans, or adults in acute pain (e.g., Mehrabian and Epstein, 1972; Batson et al., 1988; Eisenberg et al., 1991). This approach allowed researchers to successfully predict observers' prosocial response from their trait or state empathic concern (EC), personal distress (PD), perspective taking (PT), emotion regulation, and similarity to the target (among other things; see review in Piliavin and Charng, 1990). While this was a highly successful approach to studying observers of distress and need, it did not allow us to understand how real people exhibit need or how their naturally varying responses influence prosocial behavior.

There are significant theoretical reasons to assume that how people display need influences the help they receive. For example, because the willingness to help is known to increase with the salience of the target's need (Dovidio and Gaertner, 1999; Preston, 2013), observers are unlikely to know someone needs help if they do not overtly express distress (Zaki et al., 2008). However, people also withdraw support when they become personally distressed by targets or cannot regulate their own emotional response (e.g., Batson et al., 1983; Eisenberg et al., 1994, 1998)—conditions that increase with the target's level of distress. Thus, individuals in need face a conundrum in which small displays of distress may not make their need salient enough but larger ones may overwhelm observers. Taken together, empathy-based motivational theories of altruism could assume that help is optimally elicited by intermediate levels of distress, but impeded by too little or too much.

Despite this delicate but seemingly logical situation, empathy and emotional resonance can also occur for positive states that can sometimes be even more motivating, fulfilling, and rewarding to observers (Preston and Hofelich, 2012). For example, research on altruism from economics and evolutionary biology that rarely interacts with the empathy-altruism research described above suggests that people should direct resources toward those who can provide substantial return benefits to the giver (Trivers, 1971; Seyfarth and Cheney, 1984; Andreoni, 1990; Noë and Hammerstein, 1994; Brosnan Sarah and de waal Frans, 2002; Fehr et al., 2002, 2005; Gintis et al., 2003; Fehr and Rockenbach, 2004; Preston, 2013). In this framework, clearly distressed targets may actually be passed over in favor of more positive ones when the latter are viewed as offering greater potential return rewards, such as a more enjoyable prosocial interaction, a shared bond, and the feeling that the target's resilience may render them better able to benefit from the aid and to return the help later. Thus, unlike empathy-altruism theories, models that emphasize cooperation and reciprocity (e.g., Trivers, 1971; Gintis et al., 2003) or a costbenefit analysis (Dovidio et al., 2006) may actually favor positive over distressed targets, especially when their need is similar.

Such complexities are exacerbated by the fact that people have different display rules guiding how emotions should be expressed (Ekman, 1971; Matsumoto, 1990; Zeman and Garber, 1996; Brody, 2000), which in turn influence how much emotionality (particularly negative emotion) they permit in others (Zeman and Garber, 1996; Brody, 2000). For example, people from more stoic cultures may be expected to silently endure the pain of illness, while those from more expressive groups may welcome the opportunity for a "good cry," while still others may want targets to cover their concerns with jokes or "gallows humor." Mismatches between the display rules of targets and observers would make it even harder for targets to maximize their potential aid. For example, expressive observers may not realize when a stoic target is in pain while expressive targets may make stoic or suppressive observers feel uncomfortable or judgmental, even if each of those displays would produce a strong response from someone in their own subculture.

Of course, distress is often a typical and honest signal of need that should promote aid in emergency situations, like those studied in the bystander apathy (Darley and Latané, 1968; Latané and Rodin, 1969) or empathy-altruism (Batson, 2011) paradigms. In such situations, positive affect would be incongruous in targets and unlikely to promote aid. Thus, aid in acute cases should be given in proportion to the target's distress or need when the observer can help (see Preston, 2013). However, such acute and immediate need—the focus of most existing research—may not actually be the most frequent form that we encounter in the real world.

Much of our daily altruism is in response to the sustained difficulties of familiar people that we often learn about indirectly during the natural course of conversation. For example, one parent may chat with another at the playground or coffee shop about the stress associated with an illness or pending move, in their own family or that of a common friend. The receiver may subsequently offer support through meals or childcare while further sharing this information with others who may also come to offer help, and so on. These less acute displays have yet to be examined, despite pervading daily life and making the difference between spending one's weekend working or shopping at the mall vs. cooking or babysitting for an ailing or overwhelmed relative or friend.

In sum, there are important reasons to assume that the display of affect during need is a complex problem that is solved in different ways by psychological vs. economic or biological theories. On the one hand, the overt expression of distress or need engenders empathy and altruism, but in a tenuous manner that is easy to under- or overshoot. On the other hand, a positive and resilient response may actually elicit more aid from those seeking to enjoy and build social bonds. Because past research largely aimed to prove the existence of a "pure" form of altruism, and only examined observers, we know little about these potential real-world interactions between targets and observers, which have great practical importance in situations like patient care, parental responding, and cross-cultural interactions.

The first goal of the current study was to document natural variation in the display of need in real-world targets of need that are in a more typical and conversational setting rather than one of acute need. Real hospital patients were used as the targets because illness is a common stressor that people are likely to encounter in relatives, friends, and neighbors who also likely display this need in various ways. Videotaped interviews with patients about their quality of life were used as the stimuli because they displayed real affect and mimicked the more conversant and less acute way that people often learn about need in real life. Hospital patients are also generally regarded as deserving help and differential observer responses to their emotions would have important implications for public health. We hypothesized that there would be variation in the way that the targets presented their need, which could be generalized to include at least (1) a highly distressed type that clearly displayed need and negative affect related to that need, (2) a highly positive type that remained socially engaged and engaging throughout the conversation even when need was clearly present, and (3) a more laconic or reticent type that did not openly express emotion, positive or negative.

The second goal of the current study was to document changes in the way that observers responded to these different natural types of patients, with some main effects observed across observers, such as increased helping for patients with more clear need, and some interaction effects with the observer's own trait propensities. The positive patients were expected to engender high levels of empathy and helping despite not appearing as distressed or in need because they would be more attractive as social partners and would not overwhelm observers the way that highly distressed patients could (e.g., Batson et al., 1983; Eisenberg et al., 1989). Trait empathy was expected to promote giving to the targets who displayed the most clear need (the distressed ones) while PT was expected to promote empathy and helping for targets who were less expressive and do not clearly display their need (the reticent ones) (Preston and de Waal, 2002; Preston and Hofelich, 2012).

Study 1 first measured the response of participants to patient videos that included all visual, sound, and semantic cues, in order to group patients into natural affective types and study the response of observers to each type. To minimize effects of variables other than affect on the prosocial response, videos only included answers to the same four semi-structured questions about quality of life. Information about patients' diagnoses and illness severity was not provided. All research was approved by the Institutional Review Board at the University of Michigan and all participants provided informed written consent before participation.

# **STUDY 1**

In Study 1, participant observers (hereafter, "observers") viewed 14 videos of hospital patients being interviewed about their quality of life (hereafter, "targets"). During each video, continuous measures of observers' heart rate, respiration, skin conductance, and facial muscle activity were recorded. After each video clip, observers self-reported the emotion they perceived in the target [*other*], felt themselves [*self* ], and their prosocial response (after Batson et al., 1997). At the end of the study, observers filled out demographic information including trait empathy. A combination of factor and cluster analysis was used to classify the targets into display "types" (hereafter, "target types") based on their displayed affect in the *other* ratings. Differences in the emotional, psychophysiological, and prosocial response to each target type were examined.

### **MATERIALS AND METHODS**

#### *Targets*

The targets were hospital patients with a variety of serious chronic or terminal conditions (cancer, heart disease, Hepatitus C, liver malfunction requiring dialysis). They were videotaped in their hospital room during interviews for an unrelated public health study (e.g., Zickmund et al., 2003). Patients faced the camera while seated or partially reclined with their upper body, face, and head visible along with some surround (e.g., edge of bed, wall behind). Interviews were edited to contain patient responses to the same four questions, which evoked the largest range of affective responses: (1) What has been the impact of your illness on your quality of life? (2) What are your health-related worries? (3) What has been the hardest thing for you to cope with related to your illness? and (4) What in your life are you the most proud of? The questions and their answers were always played in that order, separated by a brief fade. The average clip length was 88 s (range 31–150 s). The specific illness was not mentioned and subjects were unaware of patients' prognoses.

#### *Observers*

Observers were recruited through advertisements in the daily newsletter of a university hospital and paid for their participation. Fifty-one adults were tested (27 women; mean age = 29.9, range: 19–56), excluding those with a history of neurological or psychiatric illness.

#### *Questionnaire data*

After each video, observers answered Likert scale questions from 1 [*not at all*] to 7 [*extremely*]. Observers either rated 26 emotion adjectives on how the patient appeared to feel (*other*) and then how they themselves felt (*self* ), or vice versa (order counterbalanced across subjects). Adjectives were taken from Batson et al. (1997) including those normally associated with EC (sympathetic, softhearted, warm, compassionate, tender, moved) and PD (alarmed, grieved, troubled, distressed, upset, disturbed, worried, perturbed) as well as adjectives that are traditionally collected but not analyzed (happy, amused, afraid, concerned, disconcerted, horrified, panicked, sorrowful, bothered, pleased, sad, angry). Observers also rated other reactions to the patients on a scale from 1 [*not at all*] to 7 [*extremely*] (except where noted) including "How much do you like the person in this clip?" "How severe do you think this person's illness is?" "How compelled do you feel to help this person?" and "How much help would you offer this person?" [the highest response to this was labeled with the anchor (*as much as possible*)].

After viewing and rating all 14 target videos, observers reported on their gender, age, career, and prior experience with illness. None of these variables had results that were both significant and interesting for the current aims and are not reported here. Participants also completed three trait empathy scales: The Mehrabian and Epstein Scale of Emotional Empathy (ME; Mehrabian and Epstein, 1972), the Interpersonal Reactivity Index (IRI, with subscales for EC, PT, PD, and fantasy (FS); Davis, 1983), and the Jefferson Scale of Physician Empathy (JS, designed to measure empathy for patients; Hojat et al., 2001). All three scales were administered because they tap different aspects of empathy that may be relevant in response to different target types.

## *Psychophysiological data*

Psychophysiological variables were averaged across the length of each target video. Mean heart rate [in beats per minute (BPM)] was collected using lead II EKG, with one electrode attached inferior to the costal margin and the other anterior to the sternocleidomastoid muscle. The number of peaks in the skin conductance response (SCR) was measured using electrodes attached to the thenar and hypothenar areas on the palms of both hands and was smoothed and averaged between left and right hands. Facial electromyogram (EMG) responses were recorded with pairs of electrodes attached to the *zygomaticus major* and *corrugator supercilli* muscles and were root-mean-square transformed before averaging. Data were sampled at a rate of 200 Hz using a Biopac MP100WS system (Biopac Systems, Santa Barbara, California) and were analyzed with AcqKnowledge III software for Mac (Biopac Systems). For all measures, the average response across the video was standardized within participant, across the 14 videos, to provide observers' relative response across targets.

# **ANALYSIS AND RESULTS** *Overview*

Patient emotions were first determined through principle components analysis (PCA) of *other* ratings, which were then classified using cluster analysis into target types. Next, we compared the response of observers to each target type, after averaging all targets of a type together (comparing PCA-reduced *self* emotion ratings, psychophysiology, and prosocial responses). Lastly, we attempted to determine if prosocial responses could be predicted from observers' trait empathy. Detailed statistical information for each test is provided with the result below. All tests were evaluated at alpha = 0.05 and *post-hoc* comparisons were Bonferronicorrected; any comparisons not reported were nonsignificant (*ns*, *p >* 0*.*05).

All analyses that included emotion adjective ratings included the order of presentation—*other* or *self* ratings first—because *other* ratings were statistically higher when administered before vs. after *self* ratings (*M* = 2*.*85, 2.12, respectively), *F(*1*,* <sup>49</sup>*)* = 15*.*88, *p <* 0*.*001, and the effect of order differed by adjective, *F(*25*,* <sup>1225</sup>*) >* 6*.*9, *p <* 0*.*001.

## *Which emotions do targets display? factor analysis of other emotion ratings*

*Other* emotion ratings were standardized within subjects across videos, creating relative differences for each observer across targets that were factor analyzed with PCA. Factors with an eigenvalue *>*1 were Varimax (orthogonally) rotated. *Other* emotion ratings produced three primary factors explaining 69% of the variance (**Figure 1**). We report all adjectives that loaded *>*0.5 on each factor from highest to lowest coefficient, with the adjective bolded if it was used as the factor label. The first, highest adjective was used as the label whenever possible but the third *other* emotion factor uses the third adjective so it can be differentiated from the *self* factor with the same first adjective (below).

The first *other* emotion factor represented the degree to which the target felt "disturbed" (**disturbed**, upset, afraid, bothered, panicked, distressed, disconcerted, troubled, perturbed, worried, sad, horrified, angry, sorrowful, grieved, alarmed, concerned). The second factor represented the degree to which the target appeared "softhearted" (**softhearted**, compassionate, tender, warm, sympathetic, moved). These two factors largely replicate the PD and EC factors found in prior work (respectively, e.g., Batson et al., 1983; Eisenberg et al., 1989; Batson, 2011), but because they refer to qualities of the target and not the observer, those terms are not used here. The third factor represented the degree to which the target appeared "happy" (amused, pleased, **happy**), which is a novel factor that has never been reported in prosocial behavior research using similar methods with fictionalized stimuli.

### *Can the target emotions be used to group them into affective types?*

To group the 14 patients by their affective displays, mean *other* emotion factor coefficients were submitted to cluster analysis using the Ward Method (Ward, 1963). The saved PCA coefficients for each extracted *other* factor (above) were averaged across observers per target to create a single mean coefficient per PCA factor, per target type. The resulting profile of emotion factors displayed by each target type was then used to characterize each target type. To statistically characterize them, repeated-measures (RM) ANOVA compared *other* emotion factors within and across target types. The target types were named to best capture their global appearance and the emotions differentiating them, attempting to use terms from the literature whenever possible (esp. *sanguine* and *resilient*).

The clustering technique grouped targets into five types (means and *post-hoc* comparisons in **Table 1**). From within-type comparisons, the first included three *distraught* targets who were significantly more disturbed than softhearted or happy, and less happy than softhearted, *F(*2*,* <sup>98</sup>*)* = 46*.*81, *p <* 0*.*001. *Distraught* targets often broke into tears while describing their situation and at points had to stop talking to regain their composure. The second target type consisted of four *resilient* targets who were

outside of and inside of each pie slice (respectively), with the unexplained variance left out of the pie. Similar factors between *other* and *self* are shaded the same (i.e., disturbed and personal distress are black, softhearted and empathic concern are unfilled, happy and amused are dark

factor loadings averaged across targets within a type). (Horror emerged before amused but is represented last to preserve the similar mappings of emotion factors between targets and observers). Levels of significance are reported in **Table 1**.


#### **Table 1 | Mean factor scores, psychophysiological (psychophys.) responses, and ratings by target type in Study 1.**

*Superscript numbers represent statistical comparisons of emotions within target types (between row comparisons; used for other scores to characterize target types). Subscript letters represent statistical comparisons between target types for each measure (i.e., between column comparisons of other and self emotion factors and prosocial responses across types). Demographic information about the targets in each type are provided under each target type number (F, female; M, male; YA, Young adult; OA, Older adult; C, Caucasian; AA, African American).*

more happy than softhearted and more softhearted than disturbed, *F(*2*,* <sup>98</sup>*)* = 35*.*47, *p <* 0*.*001. *Resilient* targets talked about their struggles, but remained positive and made lighthearted comments or smiled during the interview. The third target type consisted of three *sanguine* targets who were more happy than disturbed or softhearted, *F(*2*,* <sup>98</sup>*)* = 8*.*96, *p <* 0*.*001. *Sanguine* targets were less emotional than *distraught* or *resilient* targets; they talked at length without conveying major health concerns and sometimes made jokes. The fourth target type consisted of one *reticent* male who was less softhearted than disturbed or happy, *F(*2*,* <sup>98</sup>*)* = 9*.*92, *p <* 0*.*001. The *reticent* patient was laconic, giving only the briefest of responses (e.g., single words such as "fine" or "none"), and did not express overt emotion. The fifth and final type consisted of three *wistful* targets who were more disturbed than happy, *F(*2*,* <sup>98</sup>*)* = 3*.*73, *p* = 0*.*03. *Wistful* targets talked quietly about their health problems or fears of dying but did not exhibit overt negativity or distress as *distraught* targets did. The five target types also exhibited differential levels of each emotion factor from one another, as expected from the clustering technique, *F*s*(*4*,* <sup>196</sup>*) >* 27*.*58, *p*s *<* 0.001 (**Figure 1**, **Table 1**). In general, *distraught* targets appeared more disturbed and less happy than all others, *resilient* targets were conversely less disturbed and happier than all others, and the *reticent* target was less softhearted than all others.

# *Which emotions do observers feel in response to targets? Factor analysis of observers' self emotion ratings*

To examine how observers responded to the five target types, *self* emotion ratings were classified into factors as above. After standardizing the *self* emotion ratings within subjects and across videos, PCA reduced the 26 *self* adjectives into four factors that explained 74% of the variance (**Figure 1**). Again, factors are presented with adjectives ordered from the highest to lowest coefficient (including any *>* 0.5), with the adjective bolded when used as the label. The first two emotion factors were again similar to Batson's "PD" (troubled, distressed, worried, upset, afraid, grieved, sad, disturbed, bothered, concerned, sorrowful, alarmed, disconcerted) and "EC" (compassionate, sympathetic, softhearted, tender, warm, moved). In this case, we did use his terms as the factor labels because the *self* factors represent observer affect as in the classic empathy studies making it more parsimonious to use those terms rather than the highest loading adjective. The third novel positive emotion factor again emerged, referred to as "amused" (**amused,** pleased, and happy) along with an additional novel negative emotion factor representing an intense negative, alienating response of observers to targets, referred to as "horrified" (**horrified,** perturbed, angry, panicked). Again, using the same emotion adjectives and a similar rating procedure as in prior studies, we found two completely novel emotional responses of observers through the depiction of naturally-occurring affect.

## *Do observer emotions differ across target types? Comparing self emotion factors across target types*

Using mixed ANOVA observers' *self* emotion factor responses were compared across target types. As was done for the *other* emotion factors, observer factor loadings were first averaged across targets of the same type, including target type as a repeated measure.

Observers responded with significantly different emotions to target types (means and *post-hoc* comparisons in **Table 1**, **Figure 1**, main effects reported here). As predicted, *distraught* targets elicited more PD than any other, *F(*4*,* <sup>196</sup>*)* = 60*.*58, *p <* 0*.*001, and tended to elicit more horror than *wistful* patients, *F(*4*,* <sup>196</sup>*)* = 2*.*64, *p* = 0*.*035. *Resilient* targets elicited more amusement than all other types, *F(*4*,* <sup>196</sup>*)* = 103*.*79, *p <* 0*.*001. The *reticent* target elicited less EC than all other types, *F(*4*,* <sup>196</sup>*)* = 25*.*27, *p <* 0*.*001.

## *Do observers have different psychophysiological responses across target types?*

Corroborating the self-reported differences reported above, observers' physiological responses also differed across target types (means and *post-hoc* comparisons in **Table 1**, main effects below). Observers' sympathetic and heart rate responses differed across types, *F(*4*,* <sup>188</sup>*) >* 8*.*70, *p <* 0*.*001, because *distraught* targets evoked more SCR peak counts than all other types and the *reticent* patient evoked a smaller heart rate response than *resilient* and *sanguine* patients. *Resilient* patients also evoked more positive facial muscle activity than the other types, *zygomatic* EMG, *F(*4*,* <sup>188</sup>*)* = 14*.*15, *p <* 0*.*001, and less negative facial activity than *distraught* and *sanguine* patients, *corrugator* EMG, *F(*4*,* <sup>188</sup>*)* = 5*.*09, *p* = 0*.*001. Respiration rates did not differ, *F(*4*,* <sup>188</sup>*)* = 0*.*30, *p* = 0*.*71.

#### *Do observer prosocial responses differ across target types?*

Mixed ANOVA compared the remaining observer responses across target types, including how likeable they were, how sick they seemed, how much help they would offer them, and how compelled they felt to help (averaging each observer's response to all targets within a type as above; means and *post-hoc* comparisons in **Table 1**). All ratings differed significantly across the five target types, *F*s*(*4*,* <sup>196</sup>*) >* 18*.*32, *p*s *<* 0.001. The *reticent* target was less well liked, seemed less sick, received lower offers of help, and elicited a lower compulsion to help than all other types. After the *reticent* target, *distraught* targets were also significantly less well liked than the remaining three more positive types (*resilient, sanguine, wistful*). *Distraught* and *wistful* targets were also perceived as being sicker than all others.

Although helping differed across target types, observers offered highly similar amounts of help to each type, *r(*51*) >* 0*.*84, *p <* 0*.*0001. To examine relative preferences, two types of frequency data are reported. After removing 9 observers who offered identical amounts of help to all types and 7 who offered their highest amount to more than one (i.e., ties that do not indicate a singular preference), 35 observers gave a higher level of help to one particular type. The greatest number (11) preferred *distraught* targets, but almost as many preferred *resilient* (10) and *wistful* (9) targets and still some preferred *sanguine* (4) and *reticent* (1).

We also compared how often a particular type received more aid than the observer's mean (difference *>* 0), which includes more data by allowing ties and can be used to intercorrelate preferences. The frequency of preferences over an observer's mean was fairly evenly distributed across target types (*distraught* (29), *wistful* (29), *sanguine* (27), and *resilient* (25) patients), but many fewer observers offered more than their average aid to the *reticent* patient (6). Thus, as with the emotion data, observers did not so much approve of one particular type as much as they failed to empathize with the *reticent* patient. However, observers who preferred some target types gave systematically less to others. Those who offered more help to the calm, *sanguine* patients also offered significantly less to the overtly *distraught* patients [and *vice versa*, *r(*51*)* = −0*.*37, *p <* 0*.*01] while observers who offered relatively more to the *reticent* patient also offered less to all three positive types [*resilient*: *r(*51*)* = −0*.*50, *p <* 0*.*001; *sanguine*: *r(*51*)* = −0*.*59, *p <* 0*.*0001; *wistful*: *r(*51*)* = −0*.*55, *p <* 0*.*0001]. Thus, observer preferences promote aid for some targets while inhibiting it for others.

## *Is the help observers offer to each target type a function of their trait empathy?*

Even though observers offered similar aid across target types, their offers were still associated with trait empathy (detailed statistics provided in **Table 3**), particularly for *distraught* and *resilient* targets but less so for *wistful* ones. Offers of help increased for *distraught*, *resilient*, and *reticent* targets across trait empathy measures (i.e., ME, JS, IRI-EC), *r(*44*) >* 0*.*30, *p <* 0*.*05 and increased toward *resilient* and *reticent* patients with PT (IRI-PT), *r(*44*) >* 0*.*30, *p <* 0*.*05. *Sanguine* patients also tended to receive greater offers from those with higher trait empathy or PT (IRI-EC, IRI-PT, JS), *r(*44*) >* 0*.*28, *p <* 0*.*06, while *wistful* targets only received marginally more from those with higher empathy for patients (JS), *r(*44*)* = 0*.*28, *p* = 0*.*06. Both *wistful* and *distraught* targets received marginally less help from observers with higher PD (IRI-PD), *r(*44*) <* −0*.*25, *p <* 0*.*1. These effects were largely replicated with how "compelled to help" observers reported feeling, particularly for trait empathy (ME, JS, IRI-EC; see **Table 3**).

Observers' trait empathy also predicted preferences to offer more than their average help to specific types (using the difference scores from above). The *reticent* target received relatively more help from those with greater PT, *r(*44*)* = 0*.*36, *p* = 0*.*018, suggesting that active participation in his plight could compensate for his minimal affect. In contrast, *wistful* targets actually received less help from observers with higher PT and EC, *r(*44*) <* −0*.*321, *p <* 0*.*04, suggesting that these subscales may access the response to clear distress that is absent in *wistful* patients. *Resilient* targets were offered more relative help as observers' empathy for patients increased (JS), *r(*44*)* = 0*.*33, *p* = 0*.*03, which measures the extent to which people believe in empathic patient care.

#### **DISCUSSION**

For the first time we measured the affect of real hospital patients to assess how people typically convey need in such serious situations in a fairly natural and conversational setting. These individuals clearly expressed emotion in different ways, but we were also able to group them into a few major types from their displayed emotion, which elicited distinguishable responses in experimental observers.

The targets were classified through their expressed affect into five types: *distraught*, *resilient*, *sanguine*, *reticent*, and *wistful*. Using similar self-report techniques as before (e.g., Eisenberg and Miller, 1987; Eisenberg and Strayer, 1987; Batson et al., 1988, 1997), we discovered that not only do observers normally feel distressed or empathic toward targets, targets also express these emotions, supporting a perception-action (Preston and de Waal, 2002) or emotional contagion (Rapson et al., 1993) view of empathy.

These real targets expressed a surprising amount of positive emotion and elicited very positive feelings in observers—a fact that clearly influenced observers' response even though such feelings are almost never emphasized in typical experiments. Moreover, using real targets of need, we identified another novel emotional response in observers: horror. While the patients in the videos were not currently or acutely experiencing pain, their conversation elicited this very negative response in observers, particularly to the most dysregulated and distressed *distraught* targets. Such a response is understandable, but has gone unrecognized to date even though it would have important implications for support in the real world. For example, horror could predict the withdraw of aid better than PD, since PD actually predicts giving in many cases and is usually intercorrelated with EC. The fact that we revealed two novel emotion factors is particularly striking given that we used similar methods and the same 26 self-report adjectives as in prior work; only the stimuli differed. Moreover, these novel factors—positive emotion and horror each explained as much variance as EC, suggesting that they are equally important components of the response. Note that the horror factor only emerged in observer*self* ratings and not their *other* ratings of the targets. It is expected that horror could be expressed by targets of need in other contexts, but while discussing one's personal experience with illness it appears more likely to be felt by observers and merged with the other disturbed emotions in the targets.

*Distraught* patients were seen as highly disturbed, distressed, and severely ill and elicited PD, autonomic arousal and negative facial affect in observers. Observers also did not like these patients as much and tended to offer them less help when they were more prone to feel PD. However, their high display and elicitation of PD did not preclude them from receiving help—indeed, these patients actually received the highest offers of help across measures. Thus, people do seem sensitive to need above and beyond the rewards they expect to receive from the target and PD should not be considered as a solely inhibitory response to giving. Of course, the offers of help in this case were hypothetical and did not require interaction with the disliked individuals; moreover, observers could still experience a "warm glow" from helping them if that reward were yoked to the patients' level of need or how difficult it was to help them. Yet, it is striking that almost a third of observers gave the most help to these patients, despite having multiple more likeable ones to choose who had similar illnesses.

That being said, and in support of economic and biological theories of altruism, almost as many observers offered their greatest aid to the *resilient* targets who were perceived as amusing and likeable, elicited positive facial affect, and seemed less sick. *Sanguine* targets were also perceived as happy and amusing, but did not elicit the same positive facial affect, reported liking, or offers of help as *resilient* targets, presumably because they displayed less positivity and need.

Patient preferences also interacted such that observers who preferred to help the calm, *sanguine* patients offered less to overtly *distraught* ones and those who preferred the *reticent* patient offered less to the positive patients. Thus, not only do targets of need differ from one another, and elicit different responses in observers, observers also prioritize certain affective styles and penalize opposing ones, based on the degree to which the targets exhibit overt emotion. These preferences sat atop generally similar offers of help across targets, but even small preferences have realworld consequences as people typically can only help one person at a time while ignoring others. Moreover, despite limited variance, these preferences could also be predicted by observers' trait empathy and PT. In general, more empathic observers offered more help across all types, but particularly toward the emotive *distraught* and *resilient* ones. PT also seemed to help observers identify less salient target need, such as in *resilient* and *reticent* and to some extent *sanguine* targets.

Taken together, real people express their need in a variety of ways, even under highly similar situations, and these differences interact with the affective traits and preferences of observers. Of course, there are limitations. While all patients were hospitalized for serious or life-threatening illness, they had a variety of illnesses at different stages. Thus, the differential responses to the patients could have been influenced by inferences about their illness or what they said and not just their emotion. Notably, even though the *distraught* and *wistful* patients were rated as the most sick and in need, we do not believe they were actually the most sick, using the threat of death as the metric of severity. Multiple patients in the more positive *resilient* and *sanguine* target types had much more life threatening illnesses than the *distraught* ones. One sanguine patient died in the same week as the interview despite not even displaying enough need to be classified as *resilient*. However, there could be lawful relationships between the type and severity of people's illness and their affect. For example some cancer patients who are regarded as resilient also engage actively in meaning making processes (Park, 2010), which may be more pronounced in those close to death. However, the prosocial response of other people to them is expected to be more powerfully driven by their expressed affect over and above their need state. To demonstrate the power of affect alone, apart from any cues about their illness or situation imparted during the interviews, a second study was performed.

Study 2 showed new observers the same videos, but with the semantic content stripped through an audio filter that made the words too garbled to understand while preserving the emotional prosody and facial affect. The observers also rated each patient's apparent health status to use as a covariate. This way, any replication of the emotion factors and patient classification without sound and taking apparent health into account could be directly attributed to their expressed affect. In addition, to address unrelated concerns that *self* and *other* ratings in Study 1 influenced one another (e.g., subjects giving the same rating for both, anchored to the one they did first), observers in Study 2 only rated one or the other. Finally, real monetary donations were added to determine whether offers of support would be similar when the offer was not hypothetical. A rank-ordering of patients was also added in case offers did not vary strongly across types. Most of the results from Study 1 were expected to replicate, but fine-grained distinctions among the targets were expected to be lost in the total absence of semantic cues.

## **STUDY 2**

#### **INTRODUCTION**

Study 2 aimed to verify that (1) similar emotion factors and target types would emerge when only visual and affective cues were available (without verbal content and when people only rate patients' or their own emotion), (2) observers would have similar affective and prosocial responses to the targets under these conditions, (3) the results would hold after controlling for perceived patient health, and (4) offers would show similar patterns when observers had to donate real money. To increase the sample size for statistical power, Study 2 was conducted online so psychophysiological data were not collected.

#### **MATERIALS AND METHODS**

#### *Targets*

The same 14 patient videos from Study 1 were used in Study 2 but the sound channels were modified to render the spoken words unintelligible. Sound was removed from the portions of the interview where the interviewer spoke offscreen and his questions were printed on the screen so participants could understand the context of patients' responses. Audio from the patients' responses was then altered with a band pass filter between 102 and 750 Hz and a +9*.*5 dB band at 270 Hz (*Q* = 1*.*0); this eliminated high frequency sounds while preserving emotional prosody and tone of voice. Participants were explicitly told that the sound had been altered to be difficult to understand because we were interested in their perception of and response to patient emotion, above and beyond their speech content. As a manipulation check, all participants rated how much verbal content they understood after responding to each video (1, *nothing*; 2, *one or two words*; 3, *a few words here and there*; 4, *a few partial sentences*; 5, *most of the content*; 6, *all of the content*).

#### *Observers*

Ninety-nine adult participants were recruited from Amazon Mechanical Turk (aka, "Mturk"; https://www*.*mturk*.*com/mturk/ welcome) to watch and rate the videos. Forty-nine participants rated only the emotions of the patients (*other*; 32 women; mean age: 37.1, range: 18–74) and fifty different participants rated only their emotional response to the patients (*self* ; 35 women; mean age: 33.56, range: 18–59). Participants were compensated \$0.75 for participation, plus any money they chose not to donate (described below).

#### *Perception of targets*

After each video, participants in the *other* condition rated the targets on their displayed emotion through the same 26 adjectives as in Study 1. They also rated them on aspects related to the patient's perceived health: how sick the patient seemed, how energetic, their apparent prognosis from recovering to dying, how much emotional support they needed ("e.g., talking to them, giving advice, soothing, spending time with them") and how much practical support they needed ("e.g., getting prescriptions, changing sheets, watering plants, grocery shopping").

#### *Observer response to targets*

Participants in the *self* condition rated their emotional response to each patient using the same 26 adjectives as Study 1 as well as how much emotional and practical support they were willing to give each patient, and how much they liked them. After these ratings, participants were told that the patients were interviewed in exchange for monetary donations to help with their illness and to promote awareness for their disease. They were allotted five tokens per patient and told that they could donate any number of them to the patient. They were explicitly told that any tokens they did not donate would be converted to cash at the end of the study and paid to them as a bonus in Mturk. The token exchange rate was intentionally not provided because research in our lab found that participants who perceive the total amount to be low give all tokens, precluding the variance necessary for analysis. In the event that observers again gave highly similar amounts across all patients, we added a ranking task after all videos in which participants drag-and-dropped thumbnail images of patients into order from the one they most-to-least wanted to help (1–14, respectively). To focus on the relationship between observers' emotional and prosocial response, only ranking data from observers in the *self* condition were analysed.

### *Trait scales*

At the end of the study, participants completed the IRI as in Study 1 to assess individual differences in trait empathy and completed the Berkley Expressivity Questionnaire (BEQ; Gross et al., 2000) to determine if differences in expressivity could predict target preferences.

#### **ANALYSIS AND RESULTS**

Confirming that the sound was successfully altered, participants reported only understanding one or two words across all videos (*other*: *M* = 2.39, *SD* = 1*.*21; *self* : *M* = 2*.*21, *SD* = 1*.*25). Next we determined if the patients were perceived and responded to similarly in this condition.

#### *Do similar patient types emerge when verbal content is eliminated?*

Analysis was as in Study 1, with PCA factor analysis reducing *other* emotion ratings into factors, which were clustered with the Ward method into target types. Three target emotion factors again emerged, replicating those in Study 1 and explaining 57% of the variance (listed with all adjectives with *>*0.5 loadings from highest to lowest coefficient). The first was the "disturbed" factor (panicked, horrified, upset, afraid, distressed, worried, bothered, sorrowful, sad, grieved, perturbed, concerned, troubled, alarmed, disconcerted, angry). The second factor had only strong negative loadings indicating feeling *not amused*, capturing the "happy" factor from Study 1 (amused, funny, pleased, happy). The third factor replicated the "softhearted" factor (softhearted, tender, compassionate, warm, sympathetic, engaging, likable).

As in Study 1, a five-cluster solution was extracted and similar target types emerged, particularly the distinction between lacking affect, high distress, and positive affect. The *reticent* patient again separated from the rest; the *distraught* patients again clustered together (although now split across two clusters); and one large cluster combined the positive patient types into one group (*resilient*, *sanguine, wistful*). One *resilient* and one *sanguine* target formed a new cluster.

#### *Do emotional responses to the original patient types vary when content is eliminated?*

To determine if observer responses to patient types remained after verbal content was eliminated and observers only rated their own emotion, PCA factor analysis reduced the 26 *self* emotion adjectives into factors, which were averaged across targets in the five original types. Observer responses were modeled by averaging responses into the five clusters from Study 1 because the goal was to determine if observer responses to these original types would replicate when observers only had access to their affect. Based on the scree plot, four factors best explained observer *self* emotions, accounting for 45.4% of the variance. The first factor combined the "PD" and "EC" factors from Study 1 (PD/EC: sad, sorrowful, worried, concerned, sympathetic, moved, upset, softhearted, bothered, troubled, grieved, distressed, compassionate, tender). The second factor was similar to the "horrified" factor from Study 1 (perturbed, panicked, horrified, afraid, angry), and the third and fourth factors divided the positive emotional response into two factors: "happy" (warm, likable, happy) and "amused" (funny, amused).

RM-ANOVA compared observer responses (*self* emotion factors) within and across the five original target types, which again differed significantly (means and *post-hoc* comparisons in **Table 2**). The five original types still elicited significantly different PD/EC and horror in observers, *F*s*(*4*,* <sup>188</sup>*) >* 6*.*51, *p <* 0*.*001, as *distraught* targets elicited more PD/EC than any other type and elicited more horror than *reticent* and *wistful* targets. *Resilient* targets also made observers feel more happy than *distraught* patients, *F(*4*,* <sup>188</sup>*)* = 3*.*91, *p* = 0*.*004, and more amused than *wistful* patients, *F(*4*,* <sup>188</sup>*)* = 2*.*48, *p* = 0*.*045.

To determine if these effects actually reflect target emotions rather than perceived health status, additional analyses replicated these results after controlling for perceived health. A health composite index was derived from ratings by participants who only rated the targets (the *other* condition) averaging the apparent sickness, energy level, and prognosis within each target and then across all targets in a type to create a single health status index per type. This health status composite was then entered as a covariate into a linear mixed model comparing observers' emotional responses (*self* emotion factors) to the types, nested within observer, with observer as a random factor. All effects remained significant, *F*s*(*4*,* <sup>188</sup>*) >* 2*.*97, *p*s *<* 0.021.

#### *Prosocial self-reported responses*

Observer ratings of how much they liked the patient, wanted to give them emotional, practical, and monetary support, and their ordinal ranking were averaged for all targets in the five original types and compared with RM-ANOVA (means and *post-hoc* comparisons in **Table 2**). Again, *resilient* patients were liked more than all others (except for *wistful*), *F(*4*,* <sup>188</sup>*)* = 2*.*90, *p* = 0*.*02 and the *reticent* patient received less emotional and practical support than any other, *F*s*(*4*,* <sup>188</sup>*) >* 3*.*61, *p*s *<* 0.007. These effects were still significant after controlling for targets' perceived health status in the linear mixed model, *F*s*(*4*,* <sup>188</sup>*) >* 2*.*90, *p <* 0*.*02. The order in which observers wanted to assist the target types also differed significantly, *F(*4*,* <sup>188</sup>*)* = 10*.*12, *p <* 0*.*001, with *distraught*, *resilient*, and *reticent* patients being ranked higher than *wistful* and *sanguine*. These rankings were also predictable from observers' trait data (**Table 3**) as the relatively calm *sanguine*


*Subscript letters represent statistical comparisons between the display types for each emotion (between column comparisons).*



*Study 1 used self-reported help ("how much help would you offer"; "how compelled do you feel to help"). Study 2 used different measures to more precisely estimate target preferences and avoiding intercorrelated gifts, including an ordinal target ranking and real monetary donations (using the difference from each observer's mean offer). All measures were first averaged across targets within a type per observer. ME, Mehrabian and Epstein scale of emotional empathy; IRI, Interpersonal Reactivity Index with subscales for Empathic Concern (EC), Perspective Taking (PT), Personal Distress (PD) and Fantasy (FS); JS, Jefferson Scale of empathy for patients; Pos. Exp., Neg. Expr., and Impulse Str. Refer to the positive and negative expressivity and impulse strength subscales of the Berkeley Expressivity Questionnaire (BEQ, respectively). Significance level noted as follows: \*\*p <* 0*.*01*, \*p <* 0*.*05*,* <sup>∼</sup>*p <* 0*.*1*.*

targets were ranked higher by observers with greater PT (IRI-PT), *r(*46*)* = −0*.*38, *p* = 0*.*007 and lower by those who overtly display more negative emotion in life (negative expressivity subscale of BEQ; rank close to 14 out of 14), *r(*46*)* = 0*.*34, *p* = 0*.*02. There were no other significant relationships, *r*s*(*46*) <* 0*.*22, *p*s *>* 0.13.

#### *Monetary donations*

Nearly half of observers (20 of 48) gave the same number of tokens to all patients. The most common offer was giving all tokens and the next most common was giving zero tokens, which precluded significant overall differences across types, *F(*4*,* <sup>188</sup>*)* = 1*.*08, *p* = 0*.*37. Key differences could still be replicated using the frequency analyses from Study 1. Of the 23 observers who exhibited a singular preference (gave more to one group), the greatest number again preferred *distraught* targets (10); the remaining observers had preferences that were evenly spread across remaining types (3 preferred *resilient*, 2 *sanguine*, 4 *reticent*, and 4 *wistful*). Comparing how often observers gave more than their mean amount to a target type, the greatest frequency again preferred *distraught* targets (16), but almost as many preferred *resilient* (14) with a fairly even distribution across the remaining three (9 *wistful*, 8 *sanguine*, 9 *reticent*). We also replicated the intercorrelated preferences across target types from Study 1, as observers who donated more money to *distraught* patients again gave less to *sanguine* patients, *r(*46*)* = 0*.*53, *p <* 0*.*001, and those who gave more to the *reticent* patient again gave less to the three positive types, *resilient*: *r(*46*)* = 0*.*50, *p <* 0*.*001; *wistful*: *r(*46*)* = 0*.*40, *p* = 0*.*005; *sanguine*: *r(*46*)* = 0*.*20, *p* = 0*.*17. *Sanguine* patients received more help to the extent that observers reported not expressing positive emotion in their own life (positive expressivity subscale of BEQ; **Table 3**), *r(*48*)* = −0*.*3399, *p* = 0*.*0181. Further affirming the validity of the self reported offers of help, emotional and practical support were significantly correlated with the number of tokens donated over each observer's mean, *r(*240*) >* 0*.*19, *p <* 0*.*002.

#### **DISCUSSION**

Study 2 attempted to replicate the results from Study 1 even after eliminating all spoken words and controlling for how sick the patient seemed and requiring offers of real money.

As expected, some fine gradations between target types were lost without the semantic cues (e.g., differences between the *resilient*, *sanguine*, and *wistful* patients), but Study 1 was largely replicated, particularly the distinctions among high negative affect (two *distraught* types), high positive affect (one large type that combined *resilient*, *sanguine, wistful* patients), and a lack of affect (*reticent*). Of course, in the real world, our ability to discriminate people employs both verbal and bodily affect, but the effects from Study 1 were surprisingly robust to the perturbations in Study 2.

Importantly, observers had similar reactions to the targets in Study 2. Similar emotional responses (from the *self* factors) emerged in the observers, and our novel emotion factors were even more salient, as the traditional PD and EC combined into a single factor while the positive emotion factor divided into two distinct factors. Without the semantic information, observers again offered less help to the *reticent* patient while liking the *resilient* patients the most. Observers also showed similar target preferences, with more people preferring to help the most needy but disliked *distraught* patients but almost as many preferring the *resilient* patients and some preferring the other types. It is remarkable that people can exhibit such similar patterns of disliking the *reticent* patient, offering the most aid to the less well-liked *distraught* patients, and liking and offering almost equal levels of help to the *resilient* patients, even with the sound so distorted.

The *sanguine* patients across studies elicited particularly interesting interactions with observers' trait tendencies. In Study 1 they tended to be helped more by those with greater PT and in Study 2 they were ranked higher by those with higher PT and lower by those who express a lot of negative emotion. *Sanguine* patients also received relatively larger monetary donations from those who express less positive emotion. Thus, the need of these calm and collected patients may have been too subtle for those who associate need with distress, but was perhaps ascertained by those who carefully attended to them or valued their understatement.

These differences among target types were upheld even after controlling for how sick the patients seemed on multiple dimensions. Moreover, we can anecdotally attest to the lack of connection between how sick patients actually were and how sick they seemed since, for example, a *sanguine* patient died shortly after the videos were taken and multiple *resilient* and *wistful* patients had life-threatening diseases while multiple *distraught* patients had chronic but treatable illnesses. Future research can further examine these relationships in the event that chronic illness is lawfully associated with high negative affect or terminal illness with more detached and sagacious sentiments.

The data generally support an interactionist view of social behavior (Griffiths and Scarantino, 2009; Van Kleef, 2009; Preston and Hofelich, 2012), in which it is not just the observer or the target who dictate the prosocial response, but rather their interaction. For example, emotion-regulation skills influence observers' response to need (e.g., Eisenberg and Fabes, 1992; Eisenberg et al., 1994, 1998) and, thus, those with lower regulatory skills may be more likely to avoid *distraught* patients, even when they have more personal experience with the state. In addition, people from less expressive cultures could punish or avoid *distraught* targets more than those who believe negativity is natural and common. As support, our observers with high trait PD tended to offer less to sad *wistful* and *distraught* patients while people who display a lot of negative emotion were less inclined to help calm, *sanguine* patients, and people who don't display positive emotion were more inclined to help them. *Distraught* patients also evoked the most variable responses; those who preferred the positive *resilient* or *sanguine* targets simultaneously gave less to *distraught* patients. These preferences may reflect observer expectations about how people are expected to react to illness or strife, which could serve as a rich source of data on interpersonal and cross-cultural differences (Preston and Hofelich, 2012). People may also have more intuitive vs. rational or practical decision styles that influence their relative aid across types. For example, *distraught* patients should receive the most aid if observers emphasize need in a simple way while *resilient* patients should be preferred if deciders consider both absolute need and the potential for change, as *resilient* patients may be better able to build upon support to help themselves. These hypotheses are in keeping with cost-benefit views of altruism (Dovidio et al., 2006) but require additional research that offers a rich source of ideas for future work.

# **FINAL DISCUSSION**

In daily life we are surrounded by people who could use our help. Everyone has needs that would benefit from some help, most of which are not immediate, but many of which are equally or more serious and problematic than the electric shocks or ice buckets of water that are often used in experiments. The neighbor next door has a baby that cries most of the night, an unmarried uncle suffers from cancer and has no one to take care of him, the school needs someone to organize a fundraiser, and a spouse needs help practicing for a job interview. All of these are concurrent requests for our resources—material or nonmaterial—and people must make routine decisions to help only some of them. What predicts these choices?

Most research in psychology has focused on the emotional correlates of helpful observers while examining only a few target qualities like need salience, culpability, similarity, relatedness, age and vulnerability (see reviews in Piliavin and Charng, 1990; Preston and de Waal, 2002; Batson, 2011; Preston, 2013). However, people also vary a lot in how they express need, even in the same situation—variance that influences who wants to help them and how much. The goal of this study was to examine this natural variation and how it affects and interacts with observers and their own preferences.

With a relatively small sample of fourteen real hospital patients, suffering from a variety of serious chronic and terminal illness, we were able to detect at least five subtypes of displayed affect during a time of need: *distraught, resilient, wistful, sanguine, and reticent*. The main affective differences across targets were even replicated in the absence of spoken text, again identifying targets who express a lot of negative affect (*distraught*), a lot of positive emotion infused with some discussion of their plight (*resilient, wistful, sanguine*), and a lack of emotion or desire to discuss personal problems (*reticent*). Our typology is likely not exhaustive, and a sample that is larger or taken from another need context will surely find additional types. However, the complexity of the emotions represented by even just a handful of patients attests to the degree to which people's response to need varies and affects observers in predictable ways.

While observers agreed that *distraught* patients needed the most help, they were also disliked by most observers and even elicited a novel and negative state of feeling horrified, perturbed, angry, and panicked. On the one hand, these results accord with theories that predict the greatest aid for the most salient need (e.g., see Dovidio and Gaertner, 1999; Zaki et al., 2008; Preston, 2013). However, they clearly indicate that high levels of observed distress in targets and PD in observers does not preclude giving (Preston and Hofelich, 2012).

Our results also support economic and biological views that emphasize altruism as a collaborative force in group life (Seyfarth and Cheney, 1984; Noë and Hammerstein, 1994; Brosnan Sarah and de waal Frans, 2002; Fehr and Rockenbach, 2004). Almost as many observers preferred to help the more positive *resilient* patients over the ones in the most need because they still have some need but were better liked. Moreover, nontrivial numbers of observers even preferred the three remaining types even though they were not the most in need or the best liked, including fairly even preference distributions over *sanguine*, *reticent*, or *wistful* patients, oftentimes predictable by their PT skills. The differential response to *distraught* vs. *resilient* patients provides a particularly promising way to examine observer-target interactions since both have serious need and elicit aid, but the former displays largely negative affect and the latter largely positive. The patient videos and transcripts will be shared with other researchers, and variables that had important effects in this context can be extended to more controlled settings, to further our understanding of these interactionist effects.

Taken together, the light of scientific investigation has been shown for decades upon the traits and emotions of the people who observe need, leaving information about how people express need largely in the dark. By studying prosocial behavior in the context of a naturally-occurring social interaction, which reflects both the quality of the target and

## **REFERENCES**


observer, we can better illuminate human giving as it occurs in everyday life.

# **ACKNOWLEDGMENTS**

Susan Zickmund facilitated acquisition of the patients who were recruited as part of her larger study. Thanks to Emily Recknor, Victoria Lauer, and Christopher Black for help testing subjects and to the individual who performed the interviews and who videotaped them. Melissa Fasteau, Monsi Mehta, Jennifer Porter, and Amy Ross helped with data processing and analysis. Thank you to two reviewers for their helpful comments. Support for this project was provided by Antoine Bechara, Hanna Damasio, Antonio Damasio, Tom Grabowski, The Institute for Research on Unlimited Love and the Templeton Foundation.

and related behaviors. *Psychol. Bull.* 101, 91–119. doi: 10.1037/0033- 2909.101.1.91


proximate bases. *Behav. Brain Sci.* 25, 1–71.


*Biol.* 46, 35–57. doi: 10.1086/ 406755


and pain: it depends on who is watching. *Child Dev.* 67, 957–973. doi: 10.2307/1131873

Zickmund, S., Masuda, M., Ippolito, L., and Labrecque, D. R. (2003). "They treated me like a leper." *J. Gen. Intern. Med.* 18, 835–844. doi: 10.1046/j.1525-1497.2003.20826.x

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 03 April 2013; accepted: 01 August 2013; published online: 22 August 2013.*

*Citation: Preston SD, Hofelich AJ and Stansfield RB (2013) The ethology of empathy: a taxonomy of real-world targets of need and their effect on observers. Front. Hum. Neurosci. 7:488. doi: 10.3389/fnhum.2013.00488*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Preston, Hofelich and Stansfield. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# *Agency matters!* Social preferences in the three-person ultimatum game

## *Johanna Alexopoulos 1,2\*, Daniela M. Pfabigan2, Florian Göschl 3, Herbert Bauer <sup>2</sup> and Florian Ph. S. Fischmeister 2,4*

*<sup>1</sup> Department of Psychoanalysis and Psychotherapy, Medical University of Vienna, Vienna, Austria*

*<sup>2</sup> Social, Cognitive and Affective Neuroscience Unit, Faculty of Psychology, University of Vienna, Vienna, Austria*

*<sup>3</sup> Department of Neurophysiology and Pathophysiology, University Medical Center Hamburg-Eppendorf, Hamburg, Germany*

*<sup>4</sup> Study Group Clinical fMRI, Department of Neurology, Medical University of Vienna, Vienna, Austria*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Pablo Billeke, Universidad del Desarrollo, Chile Takashi Nakao, University of Ottawa, Canada Roman Osinsky, University of Wuerzburg, Germany*

#### *\*Correspondence:*

*Johanna Alexopoulos, Department of Psychoanalysis and Psychotherapy, Medical University of Vienna, Waehringer Guertel 18–20, Vienna 1090, Austria e-mail: johanna.alexopoulos@ meduniwien.ac.at*

In the present study EEG was recorded simultaneously while two participants were playing the three-person ultimatum game (UG). Both participants received different offers from changing proposers about how to split up a certain amount of money between the three players. One of the participants had no say, whereas the other, the responder, was able to harm the payoff of all other players. The aim of the study was to investigate how the outcomes of the respective other are evaluated by participants who were treated fairly or unfairly themselves and to what extent agency influences concerns for fairness. Analyses were focused on the medial frontal negativity (MFN) as an early index for subjective value assignment. Recipients with veto-power exhibited enhanced, more negative-going, MFN amplitudes following proposals that comprised a low share for both recipients, suggesting that responders favored offers with a fair amount to at least one of the two players. Though, the powerless players cared about the amount assigned to the responder, MFN amplitudes were larger following fair compared to unfair offers assigned to the responder. Similarly, concerns for fairness which determined the amplitude of the MFN, suggested that the powerless players exhibited negative and conversely the responders, positive social preferences.

**Keywords: altruism, spite, social preferences, MFN, ultimatum game**

# **INTRODUCTION**

Comparative processes are essential to assess the emotional meaning assigned to a given situation. Whether we perceive something as pleasant or unpleasant depends on the alternatives and their accessibility (Ben-Ze'ev, 2000). For example, a rewarding stimulus might get devalued in situations associated with feelings of anger or envy. Thus, the nature of emotions elicited by the reception, omission, or termination of reward or punishment depends on what we expect and on what others receive in comparison to oneself (Festinger, 1954; Rolls, 2005). This circumstance becomes apparent when looking at recent findings in the field of neuroeconomics investigating how people evaluate specific situations associated with reward or punishment in relation to significant others using simple experimental games.

One [besides several others, for a review see, Rilling and Sanfey (2011)] commonly used experimental game to study reward related decision processes and the underlying neural substrates in a social context is the ultimatum game (UG; Güth et al., 1982). In its original version a proposer is endowed with a sum of money he/she has to share with a responder. He/She can send any positive amount to the responder, who in turn has the possibility to reject or accept the proposed division of money. If the proposed distribution is accepted by the responder, the money will be allocated accordingly. Otherwise, if rejected by the responder, both receive nothing. The proposer can make only one proposal, all players are anonymous to each other, and the game ends after the responder has made his/her decision. Of course, the aim of each player in this bargaining game is to maximize his/her share of the money. Nevertheless, most responders are willing to abandon their division if it is smaller than 20% of the total amount and proposers offer about 40–50% of the total amount (Güth et al., 1982; Thaler, 1988; Güth and Van Damme, 1998). Though behavior in this game seems to be rather irrational, results are very robust and do not markedly change with the size of the stake (Slonim and Roth, 1998; Cameron, 1999; Munier and Costin, 2002). Even demographic variables, intellectual abilities, and socio-economic status do not modulate behavior in this game (Güth et al., 2007; Nguyen et al., 2011).

There are several regions in the brain that are implicated in the representation of the subjective value of reward and punishment [for reviews see Schultz (2006); Grabenhorst and Rolls (2011)]. One of these, the anterior cingulate cortex (ACC), and in particular its dorsal part, might be of particular importance in the comparative processes discussed above. In comparison with other areas associated with the representation of reward, the ACC integrates various aspects of a decision, e.g., probability, payoff, and effort (Kennerley et al., 2009, 2011). Furthermore, the ACC evaluates not only values of alternatives during choice but also the consequences of choices made. For this, the ACC receives input from different neuronal sources associated with certain qualities of a reward and has strong connections to motor areas (e.g., Vogt et al., 1992). All of these are requirements needed to synthesize these various aspects of a given situation and to adapt preferences in the light of the current goal and the effort that has to be taken. However, this region is not necessarily related to actual decision behavior (Seo and Lee, 2007; Luk and Wallis, 2009).

Hence, it is not surprising that activation in the dorsal part of the ACC (dACC) is consistently reported in neuroimaging studies investigating decision processes in the UG (Sanfey et al., 2003; Gospic et al., 2011; Kirk et al., 2011); irrespective of participants' age (Guroglu et al., 2011). Further evidence for the involvement of the dACC in the context of the UG is provided by electrophysiological studies.

The medial frontal negativity (MFN), an event related potential which is supposed to be generated in the dACC (Gehring and Willoughby, 2002; Luu et al., 2003; Wessel et al., 2012), can be observed after the receipt of negative compared to positive feedback (Miltner et al., 1997; Luu et al., 2003; Nieuwenhuis et al., 2004b), after events that deviate from what we expect (Potts et al., 2006; Hajcak et al., 2007; Pfabigan et al., 2011), and in response to losses compared to gains (Gehring and Willoughby, 2002), irrespective of whether an action or choice preceded (Donkers et al., 2005; Martin et al., 2009). Furthermore, a similar negative deflection can be reported when we observe someone else receiving negative feedback or losing money (Fukushima and Hiraki, 2009). Generally, it is assumed that the MFN discriminates events on an abstract good-bad dimension (Nieuwenhuis et al., 2004a; Hajcak et al., 2006) or whether a goal has been achieved or not (Holroyd et al., 2006). Given that the MFN can be observed already 250 ms after the onset of an event, it serves as an index for early evaluation processes in economic decision making.

Having in mind that for some individuals the subjective value assigned to a certain reward highly depends on what others receive, the MFN should as well be modulated by social preferences like inequality aversion, altruism, or reciprocity. This has been confirmed in parts by studies investigating the UG. Fair offers elicited more positive MFN amplitudes than did unfair offers and are therefore preferred in view of the assumptions on the MFN (e.g., Boksem and De Cremer, 2010; Hewig et al., 2011). However, though results show that differences in MFN amplitude are related to concerns for fairness and rejection rate (Boksem and De Cremer, 2010; Hewig et al., 2011), it is unclear to what extent MFN amplitude differences between fair and unfair offers are affected by the proposer himself as a reference agent, or whether the MFN just differentiates between high and low amounts of money.

Findings of a recent study support the notion that the proposer accounts for alterations in the MFN. As outlined earlier one would expect a negative-going MFN after receiving an unfair offer. In their study they could show that social closeness between the proposer and the responder alters the polarity of the MFN amplitude. Offers made by a friend caused an inversion of the MFN (Campanha et al., 2011). However, a recent electrophysiological study investigated the influence of social comparison on behavior in the UG and MFN amplitudes by adding a social reference point, i.e., average proposals in other proposer-responder dyads were also presented to the responders (Wu et al., 2011); yet, no influence on the MFN amplitude could be reported. In a previous study, we added a human agent as a reference point by employing a three-person UG (Alexopoulos et al., 2012). This third player, a dummy-player so to speak, had no bearing in the game itself. Money had to be split up between all three players, and the responder, whose EEG was recorded during the game, had to accept or reject the allocation as otherwise customary in the standard UG. Results, as indicated by the MFN amplitudes, showed that responders only differentiated between fair and unfair offers toward themselves disregarding the share assigned to the dummy-player. However, offers that denoted a low share for the responder and a high share for the dummyplayer elicited more pronounced MFN amplitudes than did offers with a low share for both players. This dissociation between the two kinds of unfair offers toward the responder might indicate that the third person had an impact on the responders' MFNs, and that he/she acts as the relevant reference agent responders care about. But though several studies suggested that empathic concerns are reflected in the MFN, the MFN observed in the responders seemed to be associated with negative social preferences. Nevertheless, it must be considered that participants were usually acquainted with each other whereas, in our study the dummy-players were unacquainted and in fact their presence was simulated. Therefore, one could assume that the actual presence of the dummy-player could have changed the direction of social preferences.

In the current study we therefore changed the setting and recorded EEG simultaneously from both recipients—the responder and the dummy-player—while they were playing the threeperson UG using the same setting as reported in Alexopoulos et al. (2012). In doing so, we are able to clarify how the outcomes of the respective other are evaluated by participants who were treated fairly or unfairly themselves and to what extent agency influences concerns for fairness. Furthermore, we supposed that the actual presence of the third player changes the pattern of MFN amplitudes. Since several studies have shown that pre-play communication facilitates cooperation in social dilemma or bargaining games, respectively [for a survey see Crawford (1998)], we expected a similar effect on the early neural processes. More precisely, we expected a more negative MFN difference wave for unfair compared to fair offers assigned to the third player and an interaction of unfairness toward oneself with unfairness toward the other for unsubtracted, non-difference, ERP amplitudes. However the EEG of the dummy-player was recorded for two further reasons: First, we wanted them to be in the very same situation. Recording only the EEG of the responder could give rise to the feeling of being disadvantaged from the outset. Second, given that the dummy-players had no impact on the game, i.e., they could not punish unfair treatment, they acted as a yoked control group to clarify the impact of agency.

In addition to the ERP data individual concerns for fairness were collected, as previous studies reported that fairness concerns are related to MFN amplitude differences (Boksem and De Cremer, 2010). To this end we applied a justice sensitivity scale (Schmitt et al., 2004, 2010), which measures the degree to which individuals are concerned about injustice toward oneself and others.

# **MATERIALS AND METHODS**

#### **PARTICIPANTS**

Thirty-six undergraduate students (16 males; mean age = 23.3 ± 2.69 years) from the University of Vienna participated in the experiment. All subjects were healthy, right handed, and naïve to the paradigm applied. Handedness was assessed using the Edinburgh Handedness Inventory (Oldfield, 1971). Subjects were paid between 15 and 20 Euros on average; actual earnings depended on their performance in the game.

The study was conducted in accordance with the *Declaration of Helsinki* (1973, revised in 1983) and local guidelines and regulations of the University of Vienna and the Faculty of Psychology. Written informed consent was obtained prior to the experiment.

## **JUSTICE SENSITIVITY**

Individual differences in the perception of justice were measured using the Justice Sensitivity Inventory (Schmitt et al., 2004, 2010). This 40-item questionnaire encompasses justice sensitivity from four different perspectives: the victim, the observer, the perpetrator, and the beneficiary. Each of the four subscales is covered by 10 questions that participants have to answer on a six-point Likert scale ranging from 0 to 5. Correlations between socially desirable and undesirable traits (Schmitt et al., 2004) as well as results from social bargaining games suggest that observer and beneficiary sensitivity reflect the degree to which a person is concerned about injustice toward others (Fetchenhauer and Huang, 2004). High scores on the domain victim sensitivity reflect concerns for justice toward oneself and are related to rather selfish behavior (Gollwitzer et al., 2009).

## **STIMULUS MATERIAL**

Altogether 324 proposals representing different divisions of the amount of 12, 15, or 18 Euros between the three players were presented. Half of these proposals were generated by the computer; the other half was provided by human proposers collected pre-experimentally [for details see Alexopoulos et al. (2012)]. In each of the two conditions (computer/human proposer) subjects received 27 fair offers (1/3 of the total amount for each player, hereinafter referred to as *fair/fair* offers) and 27 offers with an unfair share (less than 15%) for both receivers (referred to as *unfair/unfair* offers). 54 offers with an unfair share for one player only (receiving less than 15%, whereas the other one received 1/3), half of them with an unfair share for the responder (referred to as *unfair/fair*) and the other half with an unfair share for the dummy-player (referred to as *fair/unfair*). In addition, 54 offers were presented that did not meet any of the previous criteria and were therefore excluded from further analysis. In all conditions, the proposers allocated at least one-third of the total amount to themselves (see **Figure 1** for examples of the different categories).

In accordance with our previous study (Alexopoulos et al., 2012) the presentation of these proposals, written in German (light gray background, black font color), consisted of three lines: the first line contained the amount the proposer (e.g., "John gets 4C") or the computer (e.g., "The computer gets 4C") wanted to keep, the second indicated the amount the responder, i.e.,

the participant, would receive (e.g., "Player 1 gets 4C"), and the third line indicated the amount the third player would get (e.g., "Player 2 gets 4C"). Offers were presented in six blocks with rest periods of varying duration in between. During these breaks both players were presented with the photographs of the proposers of the subsequent trials. Stimulus presentation was controlled by a Pentium IV 3.00 GHz computer using Eprime software (E-prime 2.0, Psychology Software Tools, Inc., Sharpsburg, Maryland).

#### **PARADIGM AND PROCEDURE**

Participants were invited in gender-matched pairs. Upon arrival we ensured that these pairs were not acquainted with one another in any way. This was a precondition for the experiment to take place. Then they were informed about the further procedure, received written instructions concerning the nature of the threeperson UG and were prepared for EEG recordings. Participants were allowed to introduce themselves to each other; however conversation was restricted to things unrelated to the experiment. In order to increase the feasibility of this setup and to emphasize that half of the proposals were made by human agents, both were shown the completed questionnaires of the proposers and were informed that they themselves, as well as the other players, would receive the amount of money they earned on four randomly chosen trials in their respective roles in this game. The roles (i.e., dummy-player or responder) were randomly assigned.

Throughout the experiment, the two sat opposite each other without eye contact in a sound-attenuated and dimly lit room. Both participants were seated in front of a 19-inch cathode ray tube monitor and were about 1.2 m apart from each other.

Each block of trials started with the introduction of the proposers, followed by 54 offers which had to be accepted or rejected by the subjects in the role of the responder (**Figure 1**). Trials were pseudo-randomized, hence each block contained the same number of human and computer offers. Offers were presented for 4000 ms followed by two squares apparent below the offer, each either containing the word "accept" or "reject." These two alternatives changed their position randomly among the trials. Responders were instructed to press the corresponding button of a response pad (PST Serial Response Box by Psychology Software Tools, Inc.) with their right hand to indicate the chosen alternative. Subsequently feedback was given for the duration of 2000 ms. The format of the feedback was similar to the offer and indicated the actual allocation. Trials were separated by a variable intertrial interval with a duration of 2300–2700 ms during which a black fixation cross was presented. At the end of each block, participants were informed about the amount of money they had gained so far followed by the introduction of the subsequent proposers. To maintain the attention of the other participant, i.e., the third player, 12 randomly chosen trials were followed by questions concerning the current offer (e.g.: Was the proposer male or female?). Below these questions two squares appeared, each of which either contained the word "yes" or "no" and subjects in the role of the third player had to press the corresponding button to answer. Subjects knew that for every correct answer both will receive 0.50 Euros additionally to the outcome of four randomly chosen trials.

#### **ELECTROPHYSIOLOGICAL RECORDINGS**

EEG data from both subjects were recorded via 61 Ag/AgCl equidistantly located scalp electrodes embedded in an elastic cap (EASYCAP GmbH, Herrsching, Germany; montage M10), referenced to non-cephalic balanced sterno-vertebral electrodes (Stephenson and Gibbs, 1951). For eye movement artifact correction, vertical and horizontal electro-oculograms (VEOG, HEOG) were recorded bipolarly from above and below the left eye (VEOG), and from right and left outer canthi (HEOG). The subjects' skin was slightly scratched with a sterile needle at all recording sites in order to minimize skin potential artifacts and to ascertain homogeneous electrode impedances below 2 k-. Simultaneously recorded signals were amplified using two separate DC-amplifiers with high baseline stability and an input impedance of 100 G- (Ing. Kurt Zickler GmbH, Pfaffstätten, Austria). Signals were digitized with a 1 kHz sampling rate and recorded within a frequency range from DC (0 Hz) to 250 Hz. Synchronization of data collection was achieved using an external signal generator synchronizing the two DC-amplifiers.

#### **DATA PREPROCESSING**

Eye movement and blink artifacts were first eliminated using a linear regression approach on the basis of parameters obtained in pre-experimental calibration trials (Bauer and Lauber, 1979). Using a template matching procedure blink coefficients were identified. Blink correction was then performed by subtracting vertical and horizontal EOG signals weighted this way from each EEG channel. Epochs of 1000 ms, 800 ms following stimulus (offer) onset and 200 ms preceding the onset, were extracted for the conditions fair/fair, unfair/unfair, fair/unfair, and unfair/fair (see **Figure 1**). For further data processing EEGLAB 6.03b was used (Delorme and Makeig, 2004). The 800 ms epochs were aligned to the 200 ms baseline preceding the presentation of the offer. Subsequently, data were down-sampled to 250 smp/s, low pass filtered (6 dB/octave slope) at 30 Hz cutoff, and linear trends were removed. To further improve data quality, e.g., correcting for artifacts occurring repeatedly, we followed the approach suggested by Delorme et al. (2007) which we already used and described in detail in Alexopoulos et al. (2012). According to Marco-Pallares and colleagues (2011), 10–20 trials are enough for measuring a reliable component, thus, subjects with less than 15 trials were excluded from further analysis. Thus, two pairs of subjects had to be excluded from further analysis since the remaining number of trials after artifact correction was too low. The remaining participants had on average 22.56 (*SD* = 2.2) trials per condition remained for each of the responders and 21.17 (2.3) for the dummy-players.

#### **DATA ANALYSIS**

Based on visual inspection of grand-averaged waveforms, scalp potential topography of difference waves, and in accordance with previous literature, the MFN was quantified as the average baseline corrected mean amplitude value in the time range between 220 and 320 ms after stimulus onset at electrode Fcz, Cz, and Pz (Alexopoulos et al., 2012; Boksem et al., 2012). Though statistical analyses revealed similar results for all electrodes; reported results are based on Cz since this electrode gave the highest effect sizes. Amplitude values of the MFN for the condition *human* and *computer* were submitted to 2 × 2 repeated measures ANOVAs with the factors *Self* (levels: fair and unfair share for oneself) and *Other* (levels: fair and unfair share for the other player) separately for both groups of subjects (responders and dummy-players). All factors were defined as within-subject factors.

Furthermore, to reduce confounding effects of other ERP components on the amplitude of the MFN and to scrutinize potential differences in processing of the outcome for the other recipient, we created difference waves. These difference waves were constructed by subtracting ERPs following fair offers from unfair offers toward the respective other, while the level of fairness toward oneself was kept constant. In this way we obtained two difference waves for each player: (1) *Self* fair, *Other* unfair minus fair, and (2) *Self* unfair, *Other* unfair minus fair. To test whether difference waves were statistically different from zero a one-sample *t*-test was applied.

To assess the relation between early neuronal processes and individual differences in justice sensitivity, MFN amplitudes, respectively the associated difference waves (unfair minus fair) at channel Cz were correlated with justice sensitivity scores (using Pearson correlation and two-tailed significance levels). Due to the low variability in acceptance rates we refrained from correlation analyzes of MFN amplitudes and decision behavior. For all analyses the significance threshold was set to *p* = 0.05. All statistical analyses were carried out with IBM SPSS Statistics 19 software (IBM Corp., Armonk, NY, USA).

### **RESULTS**

#### **PERFORMANCE**

On average responders accepted 53% (*SD* = 43.15) of the offers made by the computer, compared to 52% (*SD* = 43.35) of offers made by human proposers. There was a statistically significant difference in acceptance rates depending on which type of offer was received, <sup>χ</sup>2(3) <sup>=</sup> <sup>34</sup>.193, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>.000. *Post-hoc* analysis with Wilcoxon Signed-Rank Tests was conducted with Bonferroni correction applied, resulting in a significance level set at *p* < 0.008. Rejection rates were significantly higher for inequitable offers compared to equitable offers (for all comparisons *p* < 0.001). Offers with an unfair share for both players were rejected significantly more often than those which represented an unfair share to the dummy-player only (for details see **Table 1**).

### **ERP DATA**

#### *Responders*

For the responders, mean MFN amplitudes in the time window 220–280 ms after a proposal made by a human agent revealed

**Table 1 | Median (interquartile range) acceptance rates for human and computer proposers.**


no significant main effect for the factor *Self*, [*F*(1, <sup>15</sup>) = 1.394, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.256, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.085] and the factor *Other* [*F*(1, <sup>15</sup>) <sup>=</sup> <sup>1</sup>.396, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.256, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.085]. However, the interaction (*Self* × *Other*) was statistically significant [*F*(1, <sup>15</sup>) = 19.170, *p* = <sup>0</sup>.001, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.561]. Grand-average waveforms depicted in **Figure 2** clearly show an increased MFN following offers with a low share for both recipients (unfair/unfair). Further analyses revealed that MFN amplitudes following this kind of offers were statistically significant compared to all other possible offers (for all *p* < 0.04). Likewise, only in cases where the responder received an unfair share, the amplitudes of difference waves (unfair/unfair minus unfair/fair) were significantly different from zero (mean = −2.352μV, *t*(15) = −4.452, *p* = 0.000) (see **Figure 3**). In case the responder received a fair share, however, no effect for high and low offers assigned to the dummy-player could be found (mean = 1.152μV, *t*(15) = 1.544, *p* = 0.144). The correlation analyzes of MFN difference waves and individual differences in justice sensitivity revealed a statistical relationship given by perpetrator sensitivity being negatively related to MFN amplitudes following proposals comprising unfair amounts toward the dummy-player (*r* = −0.553, *p* = 0.033). Thus, responders who are concerned about injustice toward others exhibit larger, more negative going, MFN amplitudes following advantageous inequality (see **Figure 4** and **Table 2**).

#### *Dummy-player*

Analysis of the mean MFN amplitudes for the dummy-players revealed a marginal non-significant interaction effect for *Self* × *Other*, [*F*(1, <sup>15</sup>) <sup>=</sup> <sup>4</sup>.301, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.056, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.223]. The factor *Self* [*F*(1, <sup>15</sup>) <sup>=</sup> <sup>0</sup>.001, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.970, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.000] and factor *Other* [*F*(1, <sup>15</sup>) <sup>=</sup> <sup>3</sup>.507, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.081, partial <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.189] again did not reach significance. Grand-averaged waveforms (see **Figure 2**) of the dummy-players indicate that compared to all other possible offers, those offers with a low share for only the responders (unfair/fair) are associated with a diminished negative going component. Consequently, only in case the dummy-player received a fair share, statistically significant differences between unfair and fair offers toward the responder could be observed [mean = 1.846, *t*(15) = 3.672, *p* = 0.002]. In case the dummy-player received an unfair amount, no difference in MFN amplitudes associated with unfair compared to fair offer toward the responder could be observed [mean = −0.091μV, *t*(15) = −0.116, *p* = 0.909]. The relation between justice sensitivity and MFN difference wave was analyzed similar to the responders' data. Victim sensitivity was positively related to MFN difference waves following offers with an unfair share for the responder, regardless of whether the dummy-player received a fair share (*r* = 0.591, *p* = 0.008) or an unfair share (*r* = 0.458, *p* = 0.037; see **Figure 4** and **Table 2**). Accordingly, dummy-players who were more concerned about injustice toward themselves exhibited larger positive going MFN amplitudes following unfair offers for the responder.

None of the statistical analyses applied to the ERP data associated with proposals made by the computer reached significance neither for the responders (*p* > 0.242) nor for the dummy players (*p* > 0.328). Furthermore, we found no differences between the responders and the dummy-players with regard to justice sensitivity (*p* > 0.296 for all four scales).

**proposer at Cz for the offers: fair (***R***)/fair (***D***) (blue line), unfair (***R***)/unfair (***D***) (red line), unfair (***R***)/fair (***D***) (black line), or fair**

# **DISCUSSION**

In the current study in contrast to previous studies two participants were recorded simultaneously while playing the threeperson UG. Both participants played the part of the receivers with one of them in the role of the dummy-player having no say. The responder, on the other side, had veto power

Negative is plotted up; Zeros on the timeline indicate the onset of the offer.

and thus, was able to harm the payoff of all other players. These differences in power became apparent already about 250 ms after the onset of the different offers. For both participants a difference in MFN amplitude depending on the share assigned to the respective other can be reported. In line with previous literature, MFN amplitudes elicited by unfair offers were more negative going than those elicited by fair offers, but this only applied for the responders. The dummy-players showed to some extend the opposite pattern; unfair offers compared to fair offers toward the responder were followed by positive-going amplitudes within the time range of the MFN. Although, we found differences between MFN amplitudes when the offer is made by a human proposer, no difference in MFN amplitudes could be observed following proposals made by the computer, neither for the responder nor for the dummy-player. This might be surprising at first since acceptance rates did not differ substantially between these two conditions. However, considering that we have to differentiate between at least two different processes this might become more comprehensible. The MFN is associated with the subjective value assigned to a certain situation (Holroyd and Coles, 2008; Rigoni et al., 2010); whereas, value is derived by comparative processes. Thus, expectations or

**difference in MFN amplitude between fair and unfair offers toward the respective other each with fair shares for oneself. (A)** MFN difference wave for fair and unfair offers toward the dummy-player [fair

responders **(B)** MFN difference wave for fair and unfair offers toward the responder [unfair (*R*)/fair (*D*) – unfair (*R*)/fair (*D*)] and victim sensitivity of the dummy-player.



*Note: \*p* < *0.05, \*\*p* < *0.01.*

prior experience and available options change the absolute value of a given reward and the associated MFN amplitude. Several studies have shown that social processes are also reflected in the amplitude of the MFN [for a review see Thoma and Bellebaum (2012)], since experiences in social interactions drive the expectations we have regarding the behavior of other individuals. Therefore, it might not be too surprising that no substantial differences in the MFN amplitude can be observed between conditions when the computer acts as a proposer, especially since offers are randomized and, regarding the offer size, evenly distributed. This might suggest that the intentions of the proposers indeed influence the initial evaluation process, however do not necessarily determine whether an unequal offer is accepted or not. After all it is still not a pure computer condition, since the dummy-player still has to be considered in the current decision process. Similar results were obtained in a study in which a random number generator decided how to split the money between two players. This study was also able to show that the ACC and the medial prefrontal cortex, both regions that have been associated with the MFN component especially in the context of the UG (Campanha et al., 2011; Billeke et al., 2012), are involved in the processing of unequal offers only when the participants themselves were affected. Moreover, no activation increase could be observed in this cluster when decisions were made for someone else without the participant being directly affected, although unequal offers were still rejected (Civai et al., 2012).

Regarding the results of responders; a recent attempt to investigate MFN amplitude changes in the context of the three-person UG found that responders did not differentiate between fair and unfair offers assigned to the dummy-player (Alexopoulos et al., 2012). Nevertheless, offers that clearly favored the dummyplayer opposed to the subjects themselves were followed by the most pronounced MFN amplitudes. In contrast, offers that were equally unfair for both—the dummy-player and the responder did not reveal distinct MFN amplitudes. Being speculative, anger toward the proposer and envy toward the dummy-player may have led to the increase in amplitude. In contrast to the present study these two recipients were anonymous to each other. We assume that the change in experimental setup has led to the observed differences in the ERP patterns of the responders. In the present study offers with an equally low share for the two recipients elicited the most pronounced, negative going, amplitude at the time a MFN is usually observed. This suggests that offers comprising a fair share for at least one of the two recipients are evaluated nearly as satisfying as offers with an equally high share for all three players. Furthermore, responders clearly differentiated between high and low offers assigned to the dummyplayer, with low offers leading to a more negative going MFN, at least when they themselves received an unfair share as well.

It is well known that pre-play communication enforces cooperation in social dilemma games or bargaining games, respectively [for a survey see Crawford (1998) or Greiner et al. (2005)] investigating pre-play communication in the three-person UG. In line with this finding there are at least two explanations for the changes in MFN amplitudes: Strategic issues, since the reputation of the responder is at risk, or changes in utility, since group identity enhances "we" feelings among group members, commonly summarized as emphatic concerns (Greiner et al., 2010). Recent efforts in the field of social neuroscience provide evidence that empathy is modulated by perceived group membership (Hein et al., 2010) and that empathy-related processes are expressed in the appearance of the MFN. Receiving negative feedback is associated with an increase in MFN amplitude. Observing someone else receiving negative feedback similarly elicits a MFN. Whereas, the magnitude depends on the perceived similarity with the other (Carp et al., 2009), the closeness (Kang et al., 2010), self-reported levels of empathy (Fukushima and Hiraki, 2009), and the degree to which participants include others in their self-concept (Kang et al., 2010). Since the MFN is supposed to be generated in the ACC, the fact that the ACC is a key structure implicated in the empathic response to physical and social pain of others (Singer et al., 2004; Masten et al., 2011), further suggests that empathic concerns over strategic issues have influenced the appearance of the MFN. This view is further supported by the relation between justice sensitivity and MFN amplitudes found in the present study.

Even though MFN amplitudes did not differentiate between high and low offers assigned to the dummy-player in cases were responders received a high share, the mean amplitude of MFN difference waves varied with the degree to which subjects reported to be concerned about injustice toward others. Boksem and De Cremer (2010) already reported that MFN amplitudes following unfair offers in the standard UG varied with self-reported concerns for fairness and honesty.

In the present study the degree to which responders included the share for the dummy-player when they themselves received a fair share in the evaluation process, similarly varied with their concerns for fairness. Responders scoring high on perpetrator sensitivity exhibited larger MFN amplitudes following advantageously unequal offers. Perpetrator sensitivity is highly related to socially desirable traits as well as to cooperative behavior in social dilemma games (Schmitt et al., 2004; Gollwitzer et al., 2005). Since perpetrator sensitivity focuses on situations where people actively take advantage of another party, it is assumed to be linked to feelings of guilt (Thomas et al., 2010). For instance, one example for perpetrator sensitivity would be "I feel guilty when I am better off than others for no reason." Hence, this kind of discomfort might be reflected in higher, more negative-going, MFN amplitudes in response to unfair offers toward the dummy-player. In other words, feelings of guilt might reduce the value of the relatively high share assigned to the responder in the light of a low, unfair share toward the dummy-players.

Regarding the results of the dummy-players; also MFN amplitudes of the participants playing in the role of the dummy-player were related to justice sensitivity, though, a somewhat different picture is emerging. First of all, the dummies' MFN amplitudes, though differing with respect to the outcome of the responder, were more pronounced for fair than unfair offers toward the responder. This is in contrast to what one would expect considering the data of the responders. Nevertheless, Marco-Pallares and colleagues (2010) showed that in a competitive setting observing someone else receiving a gain led to higher, more negative-going MFN amplitudes, whereas in neutral conditions MFN amplitudes were higher following losses as compared to gains of the performer. Second, offers with low shares for the responder and high shares for the dummy-player elicited a MFN difference wave significantly different from zero, but again with positive polarity. Furthermore, the higher the scores of the dummy-players were on the victim sensitivity scale, the more positive amplitudes following low offers for the responders could be observed. Victim sensitivity covers situations associated with injustice toward oneself and is related to socially undesirable traits like vengeance, jealousy and distrust. In bargaining games victim sensitive individuals tend to be less cooperative, i.e., they offer less in the UG or dictator game (Gollwitzer et al., 2005). The dummy-players are at a disadvantage from the outset, because they have no influence on the proposed allocation. This might have led to the finding that advantageous, unequal offers are more favorable than any other possible offer and even more so in subjects who are generally more concerned about fairness toward themselves. In contrast to our previous study where the responders reacted merely selfish, the presence of the dummy-player seemed to enforce "we" feelings and empathic concerns. However, this time likewise the responders in the anonymous setting, the powerless players experienced negative social preferences. While there are parallels between those two, there are also substantial differences: Responders in the anonymous setting preferred all other offers over those that assigned a low share to themselves and a high share toward the other. Therefore, we assume that envy might play a crucial role. In contrast, the powerless players preferred offers with low shares toward the responder and high shares toward themselves, which might be more closely related to spite.

However, there is another possible explanation regarding the MFN of the dummy-players: It is assumed that the MFN distinguishes events on an abstract good-bad dimension (Nieuwenhuis et al., 2004a) or in other words indicates whether a goal has been achieved or not (Hajcak et al., 2006). This is achieved by taking into account prior knowledge or available alternatives to adapt to a changing environment and facilitate future behavior. Positive and negative reward prediction errors determine the amplitude of the MFN, unpredicted positive events decrease the amplitude and negative events increase the amplitude. In light of the assumptions concerning the appearance of the MFN this might suggest that in the social context rather the expectations regarding other people's behavior and not merely reward and punishment itself influence the amplitude of the MFN. Thus, one can argue that dummy-players might have anticipated receiving lower offers than the responders, therefore high offers for the dummy-player were an unexpected reward leading to a reduction in MFN amplitude.

To conclude, in the present study we showed that the influence of agency and physical distance on social preferences can already be observed at an early level of neural processing. As participants

# **REFERENCES**


were unfamiliar to each other prior to the experiment and we did not control for sympathy, future research has to show how the level of familiarity or sympathy will further enforce this "we" feelings.

## **ACKNOWLEDGMENTS**

Johanna Alexopoulos is a recipient of a DOC-fFORTE-fellowship of the Austrian Academy of Science.

to pro- and antisocial behavior. *J. Res. Pers.* 43, 999–1005. doi: 10.1016/j.jrp.2009.07.003


*J. Cogn. Neurosci.* 9, 788–798. doi: 10.1162/jocn.1997.9.6.788


*Psychol.* 62, 23–48. doi: 10.1146/ annurev.psych.121208.131647


of performance monitoring. *Front. Hum. Neurosci.* 6:135. doi: 10.3389/fnhum.2012.00135


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 23 April 2013; accepted: 10 June 2013; published online: 27 June 2013.*

*Citation: Alexopoulos J, Pfabigan DM, Göschl F, Bauer H and Fischmeister FPhS (2013) Agency matters! Social preferences in the three-person ultimatum game. Front. Hum. Neurosci. 7:312. doi: 10.3389/fnhum.2013.00312*

*Copyright © 2013 Alexopoulos, Pfabigan, Göschl, Bauer and Fischmeister. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Peer influence: neural mechanisms underlying in-group conformity

# *Mirre Stallen1,2\*, Ale Smidts <sup>1</sup> and Alan G. Sanfey2*

*<sup>1</sup> Department of Marketing Management, Rotterdam School of Management, Erasmus University, Rotterdam, Netherlands*

*<sup>2</sup> Centre for Cognitive Neuroimaging, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Nijmegen, Netherlands*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Vasily Klucharev, University of Basel, Switzerland Grit Hein, University of Zurich, Switzerland*

#### *\*Correspondence:*

*Mirre Stallen, Centre for Cognitive Neuroimaging, Donders Institute for Brain, Cognition and Behaviour, Radboud University, Room 1.40, PO Box 9101, 6500 HB Nijmegen, Netherlands. e-mail: m.stallen@donders.ru.nl*

People often conform to the behavior of others with whom they identify. However, it is unclear what fundamental mechanisms underlie this type of conformity. Here, we investigate the processes mediating in-group conformity by using functional magnetic resonance imaging (fMRI). Participants completed a perceptual decision-making task while undergoing fMRI, during which they were exposed to the judgments of both in-group and out-group members. Our data suggest that conformity to the in-group is mediated by both positive affect as well as the cognitive capacity of perspective taking. Examining the processes that drive in-group conformity by utilizing a basic decision-making paradigm combined with neuroimaging methods provides important insights into the potential mechanisms of conformity. These results may provide an integral step in developing more effective campaigns using group conformity as a tool for behavioral change.

**Keywords: conformity, in-group bias, MRI imaging, judgment and decision-making, social influence**

# **INTRODUCTION**

People are often influenced by others with whom they identify. They buy clothes similar to those of their peers, visit restaurants because their colleagues go there, and download music their friends listen to. By adopting the tastes of others, people show they belong to a specific group. This social factor, whereby people follow the behavior or advice of others they associate with, has been labeled *in-group influence*. It is not limited to product choice (Bearden and Etzel, 1982; Berger and Heath, 2007, 2008), but influences behavior even when identity signaling is not an issue. For instance, a field experiment on household energy conservation showed that informing people about their neighborhood's average home energy usage resulted in a change in household energy consumption, specifically toward the mean of their neighborhood (Schultz et al., 2007). Similarly, a study on conservation behavior found that hotel guests were more likely to reuse towels when informed that guests who had stayed in that same room had reused towels than if they were informed about the behavior of guests in general (Goldstein et al., 2008).

Given the powerful influence of the in-group, it therefore comes as no surprise that there has been an increase in the use of group conformity as a tool for behavioral change. People often overestimate both the degree of approval and the prevalence of negative behavior among peers, such as drinking, drug use, violence, littering, or cigarette smoking (Baer et al., 1991; Donaldson et al., 1994; Schultz, 1999; Neighbors et al., 2004; Berkowitz, 2010). Social influence-programs seek to correct these misperceptions by exposing their target groups to the actual attitudes of their peers and the real frequency of the undesirable behaviors. However, despite the initial popularity of these programs, the evidence for their success in establishing behavioral change has been mixed. Over time, many programs failed to change behavior substantially (Peeler et al., 2000; Clapp et al., 2003), and some social influence-programs even showed effects of increasing the undesirable behavior they tried to reduce (Granfield, 2005; Wechsler et al., 2003). The mixed findings on the effectiveness of social influence-programs demonstrate that it is still unclear exactly what psychological processes may mediate in-group conformity. In order to understand why and when people conform to their in-group, we need to understand the mechanisms that drive in-group conformity.

The aim of the present study was to gain greater insight into the processes underlying in-group conformity. To examine the mechanism of in-group conformity, we used functional magnetic resonance imaging (fMRI), a modern neuroscientific method that provides a non-invasive measure of neural activity by assessing regional changes in blood oxygenation [blood oxygen level dependent (BOLD) response]. Using fMRI enables us to make inferences about the processes that underlie in-group conformity, which is difficult to assess using behavioral measures alone. In addition, to investigate the basic underlying processes, we measured in-group conformity using an artificial group manipulation and using a domain that was neither relevant for identity signaling nor related to actual choice. Examining the neural processes driving in-group conformity under these minimal conditions provides fundamental insights into the basic brain mechanisms, and may help in designing more effective social norm campaigns.

Although the application of neuroimaging methods in decision-making research has increased in popularity during the last decade, only recently have neuroscientists started to identify the brain networks implicated in social influence, for example examining the influence of experts (Klucharev et al., 2008; Campbell-Meiklejohn et al., 2010), the persuasiveness of celebrities (Stallen et al., 2010), the mechanisms of racial bias (Beer et al., 2008; Van Bavel et al., 2008; Gonsalkorale et al., 2011), the influence of majority behavior on individual decision-making (Berns et al., 2005, 2010; Klucharev et al., 2009, 2011; Mason et al., 2009), and, most relevant to the current investigation, the influence of in-group membership on both money allocation (Volz et al., 2009) and helping behavior (Hein et al., 2010). Volz and colleagues (2009) investigated the neural implementation of social identity theory, which assumes that each individual has both a personal and a social identity, and that the way information is processed depends on which identity of the individual is salient at the time of decision-making (Tajfel and Turner, 1986). The results of Volz and colleagues (2009) support social identity theory by demonstrating that the social self is derived from the same cognitive processes as the individual self, as activation of both types of identities resulted in similar neural patterns in the prefrontal and parietal network. A second study on in-group influence by Hein and colleagues (2010) investigated the neurobiological basis of the decision to help either an in-group or out-group member in pain. Their results showed that seeing an in-group member in pain evoked more empathy-related responses in the brain than seeing an out-group member in pain, and demonstrated that the degree of this empathy-related response predicted in-group favoritism in actual helping behavior at a later point in time. Importantly however, none of these studies on social influence in the brain examined the processes that underlie conformity to the in-group.

# **MATERIALS AND METHODS**

#### **PARTICIPANTS AND DESIGN**

Twenty-eight healthy right-handed participants (mean age 20.7 years) took part in the experiment. All were free of neurological or psychiatric illness, head trauma or drug abuse, and none were taking medication. Written informed consent was obtained according to the local medical ethics committee, and participants were compensated financially. Data from three participants were discarded due to technical problems, and one participant was excluded because he guessed the study aim. This resulted in 24 subjects for final analyses (12 males). We used a repeated measures design with the identity of the group member (in-group or out-group member) as a single within-subject factor.

## **MATERIALS AND METHODS**

Subjects arrived alone to the experiment. Upon arrival, participants' group membership was manipulated using a minimal group paradigm approach (Tajfel et al., 1971). In the task, adapted from Volz et al. (2009), five perceptual illusions, such as the young girl-old woman illusion, were shown for 2 s each, After each illusion, two possible answers were displayed on the screen and participants were asked to choose between them. Then, participants were informed that they had been categorized as people who either focus on the foreground of visual illusions, people who focus on the background, or as people who could not be classified into either of these two categories. Unbeknownst to participants, everyone was classified as a foreground perceiver (in-group). The other two groups (background and unclassified) will be referred to as the out-group. We manipulated group membership artificially instead of using real, existing groups, as this allowed us to control the (minimal) information participants had about their in-group and out-group members, and hence ensured that the hypothesized in-group conformity effect could not be explained by factors other than group membership, such as for example perceived differences in expertise in perceptual decision-making.

## **DECISION-MAKING TASK**

After the perceptual illusion task, participants completed the decision-making task while undergoing MRI (**Figure 1**). First, participants were instructed to look at a dot pattern on a computer screen for 1.5 s. The number of dots on display ranged from 5 to 30 (*M* = 15, *SD* = 7*.*5), and the participants' goal was to estimate the number of dots as accurately as possible. The number of dots used was based on pre-tests conducted with a different set of participants (*N* = 42). Pre-tests showed that, on average, participants were able to estimate about 11 dots (*SD* = 2*.*2 dots) correctly within 1.5 s. Because we required our experiment to include both easy and difficult trials (easy trials were included to ensure motivation), we varied the number of dots from 5 to 30 across trials. After the brief presentation of the dots, participants were instructed to think about their estimate (duration jittered between 2.5 and 6 s). Next, the estimate of a previous participant was displayed. This estimate came from either a member of the same group as the participant, that is, a foreground perceiver (in-group member) or from a member of a different group (outgroup member). Group membership of the other participant (foreground perceiver, background perceiver or unclassified perceiver) was indicated by a colored cartoon of either yellow, purple or blue. Colors were counterbalanced to ensure no confound between the color of the cartoon and group membership.

After presentation of the estimates, a response screen appeared. This screen was identical to the previous screen on which the estimate was presented, except for a response bar displayed at the bottom of the screen. This bar consisted of a row of numbers from 5 to 30, on which participants were instructed to indicate their estimate. Responses were indicated by scrolling to the number of their choice and pressing a confirmation button. The estimates provided by in-group and out-group members were predetermined by a computer script and, unbeknownst to the participant, were always correct. Finally, to enhance motivation, participants were told that the group who performed best would win an (unspecified) prize, with the winning group notified at the end of the study.

The presentation of a fixation screen (duration jittered between 3.5 to 5 s) signaled the start of a new trial. Participants performed 214 trials. To maintain attention, 6 self-paced breaks were included. The total scanning session took approx. 55 min.

#### **MRI ACQUISITION PARAMETERS**

Functional images were acquired with a 1.5T Siemens Sonata scanner, using an ascending slice acquisition and a T2∗-weighted echo-planar imaging (EPI) sequence (TR 2.29 s, TE 30 ms, flip angle 70◦, slice matrix 64 × 64 mm, slice thickness 3.0 mm, slice gap 0.5 mm, FOV 224 mm). Anatomical scans were acquired with a T1-weighted MP-RAGE sequence (176 sagittal slices, TR 2.25 s, TE 3.93 ms, flip angle 15◦, slice matrix 256 × 256, slice thickness 1.0 mm, no gap, FOV 256 mm).

#### **DEPENDENT VARIABLES**

#### *Behavioral questionnaires*

To test the group manipulation, participants answered a questionnaire at the end of the experiment. This measured the level of identification ("I feel connected to the blue/yellow/purple team"), trust ("I trust people from the blue/yellow/purple team"), and the degree of positive associations ("I have positive associations with the blue/yellow/purple team") toward other participants. Responses ranged from 1 (not true at all) to 7 (very true).

#### *Conformity*

Conformity was assessed by calculating the percentage of trials on which participants gave the same judgments as the in-group or out-group member.

#### *Brain imaging analyses*

Data were preprocessed and analyzed using a standard software package (SPM8, Wellcome Department of Cognitive Neurology London). The first 5 images of each participant's EPI sequence were discarded to allow for longitudinal relaxation time. The remaining images were realigned to the first imaging volume. Functional images were corrected for motion and differences in slice time acquisition. Next, images were normalized to the Montreal Neurological Institute (MNI) template using parameters defined from the normalization of the anatomical scan to the MNI template, and images were smoothed with a Gaussian kernel of 8 mm full-width at half-maximum to reduce noise. Motion parameters were stored and used as nuisance variables

in the general linear model (GLM) analysis. A random-effects analysis within the framework of the GLM was applied to model event-related responses (Friston et al., 1995).

Four regressors were defined for each participant based on the onsets of the relevant trials: "*Conformity to In-group,*" "*Conformity to Out-group,*" "*Non-Conformity to In-group,*" and "*Non Conformity to Out-group.*" Brain responses were timelocked to the presentation of the estimate of either the ingroup or out-group member. Regressors were modeled with a canonical hemodynamic response function and a GLM analysis was used to create contrast images summarizing differences in brain activity across the *Conformity to In-group* and *Non-Conformity to In-group* trials, as well as differences in brain activity across the *Conformity to Out-group* and *Non-Conformity to Out-group* trials. To test hypotheses regarding brain areas that were uniquely involved in conformity to an in-group member, we masked the brain activity present in the In-group contrast map (*Conformity to In-group > Non-Conformity to In-group*) with the Out-group contrast map (*Conformity to Out-group > Non-Conformity to Out-group*) (*p <* 0*.*05 uncorrected) (e.g., Pochon et al., 2002; Uncapher et al., 2006; Enzi et al., 2009). This exclusive masking procedure revealed activity in the In-group contrast map that did not overlap with the brain areas involved in the Out-group contrast map (*p <* 0*.*001, uncorrected, 10-voxel minimum). To assess whether there was a relationship between brain activity underlying conformity and the self-report measures assessed, we correlated individual beta values of the reported brain activity with participants' scores on the scales measuring identification, positive associations, and trust toward in-group and out-group members.

## **RESULTS**

#### **MANIPULATION CHECK**

In line with the group manipulation, participants identified more strongly with in-group members (*M* = 4*.*7, *SD* = 1*.*0) than with out-group members (*M* = 3*.*2, *SD* = 1*.*0), *t(*23*)* = 5.4, *p <* 0*.*001 (paired *t*-test). There were no differences in identification between the two out-groups, that is, participants identified equally with out-group members that were classified as background perceivers (*M* = 3*.*3, *SD* = 1*.*1) or that were not classified (*M* = 3*.*2, *SD* = 1*.*2), *t(*23*)* = 0.6, *ns*. Consistent with an ingroup preference, participants had more positive associations with in-group members (*M* = 5*.*8, *SD* = 0*.*6) than with outgroup members (*M* = 4*.*9, *SD* = 0*.*9), *t(*23*)* = 4*.*3, *p <* 0*.*001, and participants reported greater trust in in-group members (*M* = 4*.*9, *SD* = 1*.*1) than in out-group members (*M* = 4*.*1, *SD* = 1*.*0), *t(*23*)* = 3*.*3, *p <* 0*.*005.

#### **BEHAVIORAL CONFORMITY**

Participants conformed more often to in-group judgments than to out-group judgments. The percentage of trials in which participants' judgment matched the estimate of the group member was higher when seeing the estimate of an in-group member (*M* = 67*.*8%, *SD* = 9*.*4%) than an out-group member (*M* = 65*.*4%, *SD* = 9*.*2%), *t(*23*)* = 2*.*8, *p <* 0*.*01. In-group conformity did not differ between easy (≤11 dots), and difficult trials, *t(*23*)* = 0*.*5, *n.s*.

#### **NEURAL CORRELATES OF IN-GROUP CONFORMITY (TABLE 1)**

When examining brain areas exclusively involved in conformity to the in-group, we found a significant increase in activity in right caudate, subgenual anterior cingulate cortex (subACC), right hippocampus, and in the intersection of the right posterior insula and the posterior superior temporal sulcus (pSTS) (**Figure 2**). Analyses of the In-group contrast (*Conformity to In-group > Non-Conformity to In-group*) and Out-group contrast (*Conformity to Out-group > Non-Conformity to Out-group*) directly did not reveal any significant activation patterns. Next, we calculated whether there were any correlations between the neural activity underlying in-group conformity and participants' self-reports on in-group trust and associations. Correlation analyses were conducted for each brain region found to be involved in in-group conformity, and corrected for multiple comparisons accordingly (Bonferroni-corrected *p <* 0*.*0125). We found that the activity in the posterior insula/pSTS positively correlated with participants' scores on the trustworthiness of in-group members (*r* = 0*.*53, *p <* 0*.*01). Thus, the more trustworthy participants' judged their in-group, the higher the activity in this region. No other significant correlations were found.

## **DISCUSSION**

To examine the basic processes that mediate in-group conformity, we explored the neural mechanisms underlying this effect. Activity in the caudate was selectively enhanced when participants conformed to the in-group, supporting the hypothesis that the striatum plays an important role in social influence (Klucharev et al., 2009; Campbell-Meiklejohn et al., 2010; Zaki et al., 2011). The striatum, located in the center of the brain, is a major input station for midbrain dopamine neurons and plays a primary role in the processing of rewards, including primary rewards such as liquids, foods, and sexual stimuli (Redouté et al., 2000; Berns et al., 2001; O'Doherty et al., 2003), as well as to money (Knutson et al., 2000) and more abstract rewards such as reputation or status (Izuma et al., 2008; Zink et al., 2008). The finding that the striatum is involved in in-group conformity, in conjunction with conformity-related activations in other low-level processing areas such as the subACC, an area implicated in the experience of affective states (Drevets et al., 1997), and the hippocampus, an area important for the retrieval of spatial memories (such as the dot display) (e.g., Eldridge et al., 2000), suggests that in-group conformity is mediated by fundamental value signals in the brain. Importantly, involvement of the subACC suggests that affective signals may be related to the positive experience of social inclusion in particular, as this brain region has been implicated in social acceptance (Somerville et al., 2006), and also shown to be more active for individuals low in rejection sensitivity (Burklund et al., 2007). Taken together, these findings suggest that people conform more to in-group members than to out-group members because the behavior of in-group members is more strongly associated with the experience of positive affect and reward.

Greater activity for in-group conformity was also found in a region bordering the pSTS and the posterior insula, with peak activity in the posterior insula but extending further into pSTS. The pSTS is an area often implicated in the cognitive capacity of perspective taking, typically termed Theory of Mind (Frith and Frith, 2006). The concept of Theory of Mind is defined as the


*MNI coordinates of peak activity. HEM, hemisphere; BA, Brodmann area; pSTS, posterior superior temporal sulcus; SubACC, subgenual anterior cingulate cortex.*

**FIGURE 2 | Brain regions involved in in-group conformity,** *p <* **0***.***001 uncorrected. (A)** subgenual ACC, *x* = 0; **(B)** pSTS/insula (circled in red) and hippocampus (circled in yellow), *x* = 40; **(C)** pSTS/insula (circled in red) and caudate (circled in green), *z* = 4.

understanding that others have their own individual perspective on the world, which may differ from your own. Finding that the BOLD response in the pSTS is selectively enhanced for in-group conformity is interesting, as this could imply that participants took the perspective of the other more when the other was an in-group member than an out-group member. This hypothesis supports previous work suggesting that people mentalize more about in-group than out-group members (Harris and Fiske, 2006; Freeman et al., 2010; Heatherton, 2011). Moreover, activity in the pSTS correlated with participants' self-report measures on the perceived trustworthiness of the in-group, again indicating that those who reported strong feelings of trust toward their ingroup were more in-tune with the mental state of their in-group member.

The present findings complement behavioral studies (e.g., Asch, 1951; Deutsch and Gerard, 1955; Cialdini and Goldstein, 2004; Jetten et al., 2004) and recent pharmacological work (Stallen et al., 2012) on group influence, and expand on investigations of the neural bases of both conformity (Berns et al., 2001, 2010; Klucharev et al., 2009, 2011; Mason et al., 2009; Campbell-Meiklejohn et al., 2010; Zaki et al., 2011; Berns and Moore, 2012), and in-group influence (Volz et al., 2009; Hein et al., 2010). Furthermore, our results provide potential relevant insights for the design of social influence programs. We show that conformity to the in-group is presumably mediated by both positive affect as well as perspective taking. This suggests that social influence programs may benefit by emphasizing the positive aspects associated with in-group membership rather than, for instance, stressing the negative feelings associated with social exclusion. Additionally, our data suggest that social influence-programs will work more effectively when the target is stimulated to imagine the state of mind of the in-group and "puts himself in the others" shoes', thereby facilitating perspective-taking processes which may result in more trust directed toward in-group information.

Future research could productively test these hypotheses, as the present effects are small and the interpretations here are based on previous research linking activity in specific brain regions to basic cognitive functioning. In general, the ability to assess with certainty the cognitive processing reflected by specific brain activity is challenging due to the multiple functions brain regions typically engage in (Poldrack, 2006). Follow-up behavioral and neuroimaging studies can reveal how the basic mechanisms of in-group conformity reported here are modulated by different contexts, in particular participant population and decision-making domain. For instance, the conformity effect reported here is quite small, likely due to the minimal conditions under which in-group conformity was tested. However, when using natural groups, such as friends or sports teams, and when measuring conformity in a decision-making domain more closely related to identity formation, such as consumption choice for clothing, music, hairstyle, or food (Bearden and Etzel, 1982; Berger and Heath, 2007, 2008), the present in-group conformity bias would likely be stronger. In-group conformity in contexts more relevant to identity formation may not only be mediated by mechanisms of positive affect and perspective taking as reported here, but by the activation of social identity processes as well. A candidate brain region for these processes is the dorsal medial prefrontal cortex, as previous research has found this area to be implicated in the activation of self and group identity and to correlate with a behavioral in-group bias (Volz et al., 2009). In addition, our findings encourage the study of in-group conformity across different age ranges. We found in-group conformity to be mediated by increased activity in the subACC, an area known to be involved in social inclusion (Somerville et al., 2006; Burklund et al., 2007) and positive affect (Kim et al., 2003; Sharot et al., 2007) in adults. However, in adolescents the sub-ACC has been found to correlate with social exclusion (Masten et al., 2009). This may predict that while in-group conformity in adults is primarily driven by the positive affect associated with social inclusion, in-group conformity in adolescents might be driven more by the negative affect associated with social exclusion. Social influence campaigns targeted at adolescents may therefore be more effective when emphasizing the negative aspects of social exclusion than the positive affect associated with social inclusion.

# **CONCLUSION**

The present findings complement recent work on the physiological bases of both conformity (Breiter et al., 2001; Klucharev et al., 2009, 2011; Mason et al., 2009; Berns et al., 2010; Campbell-Meiklejohn et al., 2010; Zaki et al., 2011), and in-group influence (Volz et al., 2009; Hein et al., 2010; Stallen et al., 2012). The current study is a first step toward understanding the nature of actual in-group conformity behavior, and provides a first insight into what mechanisms may drive this effect. Our data indicate that both positive associations linked to in-group members, as well as the ability to take the perspective of the ingroup, likely play an important role in in-group conformity. Understanding why group membership has such a profound influence on decision-making provides a window into one of the basic motivations underlying people's behavior, and may help in developing effective campaigns based on a social influence approach.

#### **REFERENCES**


college-students. *J. Stud. Alcohol* 52, 580–586.


The Quadruple Process model approach to examining the neural underpinnings of prejudice. *Neuroimage* 43, 775–783.


*Sexual Violence: A Practitioner's Sourcebook,* ed K. Kaufman (Holyoke, MA: NEARI Press), 147–172.


mood disorders. *Nature* 386, 824–827.


humans: an fMRI study. *Proc. Natl. Acad. Sci. U.S.A.* 99, 5669–5674.


is mediated by medial prefrontal cortex activation. *Soc. Neurosci.* 4, 244–260.


modulates the neural computation of value. *Psychol. Sci.* 22, 894–900.

Zink, C. F., Tong, Y., Chen, Q., Bassett, D. S., Stein, J. L., and Meyer-Lindenberg, A. (2008). Know your place: neural processing of social hierarchy in humans. *Neuron* 58, 273–283.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 30 September 2012; paper pending published: 06 November 2012; accepted: 06 February 2013; published online: 08 March 2013.*

*Citation: Stallen M, Smidts A and Sanfey AG (2013) Peer influence: neural* *mechanisms underlying in-group conformity. Front. Hum. Neurosci. 7:50. doi: 10.3389/fnhum.2013.00050*

*Copyright © 2013 Stallen, Smidts and Sanfey. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# Framing the ultimatum game: the contribution of simulation

#### *Barbara Tomasino1, Lorella Lotto2, Michela Sarlo3, Claudia Civai 4, Rino Rumiati <sup>2</sup> and Raffaella I. Rumiati <sup>4</sup> \**

*<sup>1</sup> IRCCS E. Medea-Associazione La Nostra Famiglia, San Vito al Tagliamento, Italy*

*<sup>2</sup> Dipartimento di Psicologia dello Sviluppo e della Socializzazione, Università di Padova, Padova, Italy*

*<sup>3</sup> Dipartimento di Psicologia Generale, Università di Padova, Padova, Italy*

*<sup>4</sup> Cognitive Neuroscience Sector, SISSA, Trieste, Italy*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Urs Fischbacher, University of Konstanz, Germany Nicola Canessa, Vita-Salute San Raffaele University, Italy Mascha Van 'T Wout, Brown University, USA*

#### *\*Correspondence:*

*Raffaella I. Rumiati, Area of Neuroscience, SISSA, Via Bonomea 265, 34136 Trieste, Italy e-mail: rumiati@sissa.it*

It has now become widely accepted that economic decisions are influenced by cognitive and emotional processes. In the present study, we aimed at disentangling the neural mechanisms associated with the way in which the information is formulated, i.e., framing effect, in terms of gain or loss, which influences people's decisions. Participants played a fMRI version of the Ultimatum Game (UG) where we manipulated bids through two different frames: the expression "I give you" (*gain*) focusing on money the respondent would receive if she/he agreed with the proponent, and the expression "I take" (*loss*) focusing on the money that would be removed from the respondent in the event that she/he accepted the offer. Neuroimaging data revealed a frame by response interaction, showing an increase of neural activity in the right rolandic operculum/insular cortex, the anterior cingulate, among other regions, for accepting the frame "I take" vs. rejecting, as compared to accepting the frame "I give you" vs. rejecting. In addition, the left occipito-temporal junction was activated for "I take" vs. "I give you" for offer 5, corresponding to the equal offer made unpleasant by the presence of the frame "I take," where is the proposer that takes the money. Our data extend the current understanding of the neural substrates of social decision making, by disentangling the structures sensitive to the way in which the information is formulated (i.e., framing effect), in terms of gain or loss.

**Keywords: ultimatum game, framing effect, anterior insula**

### **INTRODUCTION**

Cross-field research in experimental economics and cognitive psychology has clearly demonstrated how both the cognitive and emotional processes may influence economical decision-making (Bechara et al., 1997; Sanfey et al., 2003; Naqvi et al., 2006). More critically, these studies unveiled the limits of the theory of rationality proposed by von Neumann and Morgenstern (1947).

In the Ultimatum Game (UG), two players are asked to divide a given amount of money: the proponent must decide how this money should be divided, while the responder may accept or reject the offer. If the responder accepts the offer, both players receive the agreed amount, but if the responder rejects, neither of them gets anything. What has been observed is that when participants play as responders, they tend to reject about 5 of bids below the 2–3 of the total (Henrich et al., 2001), behaving against the predictions of classical economic theories of monetary maximization of utility (von Neumann and Morgenstern, 1947), as even the acceptance of an offer would constitute a minimal gain, and, as such, worth being accepted. The UG violates the classic assumption of the *homo economicus*, in that people prefer to reject a sure amount of money rather than accepting an unfair division. In order to explain this behavior, behavioral economists developed the concept of social preference, defined as a concern for the payoffs of other relevant agents in addition to the concern for one's own payoff (Carpenter, 2008), and proposed different accounts. According to one account, focused on the distribution of the outcomes, the individuals reject unequal offers because they have a preference for equal outcome (e.g., Bolton, 1991; Fehr and Schmidt, 1999). The second account focuses on intentions, and claims that people reject unfair offers in order to punish the socially unacceptable behavior of the proposer. From a psychological point of view, negative emotions, such as frustration, have been proposed as being the ultimate cause of the rejections (Pillutla and Murnighan, 1996), and psychophysiological, imaging and neuropsychological evidence supports this interpretation.

Van't Wout and colleagues, for instance, found that increased skin conductance response, a measure of emotional arousal, was associated with the rejection, compared to acceptance, of unfair offers (van't Wout et al., 2006). Sanfey et al. (2003), on the other hand, interpreted the stronger activation of anterior insula associated with rejection of unfairness as a sign of emotional arousal, as this area had traditionally been linked to negative emotions such as disgust (Sanfey et al., 2003). Koenigs and Tranel (2007) found that patients with a lesion to the vmPFC increased their rate of rejections for unfair offers, interpreting this result as a sign of a deficit in controlling frustration (Koenigs and Tranel, 2007). Interestingly, Moretti et al. (2009) confirmed that a lesion in the vmPFC led to an increased rate of rejection of unfair offers, but only when they were presented as future abstract outcomes; instead, when money was physically present during the interaction, their rate were no different from the control group (Moretti et al., 2009). This latter finding suggests that the vmPFC is involved in representing the value of future abstract rewards rather than in controlling negative emotions elicited by unfairness. Further evidence in support to the involvement of mechanisms other than negative emotional arousal is provided by Civai et al. (2010), who found that participants were more aroused when they rejected, as opposed to when they accepted, unfair offers for themselves: when asked to decide on behalf of an unknown third party, subjects rejected the same amount of unfair offers, but they showed no difference in the emotional arousal between rejection and acceptances (Civai et al., 2010). It could be possible that the increased arousal in myself condition is driven by the fact that subjects incur in the cost of rejecting, whereas in the third-party condition they do not; however, more recent behavioral data showed that, in other situations, the unfairness of the division does not prevent acceptance when the offer is advantageous for the responder (Civai et al., 2012, 2013), suggesting that other mechanisms besides pure perception of unfairness may drive the behavior in myself condition (e.g., willing to be better off the other player). By applying this same self-other manipulation, Corradi-Dell'Acqua et al. (2012) found that the medial prefrontal cortex was associated to rejections in the self condition, whereas the anterior insula was associated with rejection in third party condition (Corradi-Dell'Acqua et al., 2012), and suggested that the activation in anterior insula is triggered by social norm violation (e.g., King-Casas et al., 2008) and not just by emotional arousal [for a discussion on this issue, see Civai (2013), in this special issue].

In the domain of economics, a number of studies demonstrated that decision-making is strongly affected by gain and loss contexts (Tversky and Kahneman, 1981; De Martino et al., 2006). Specifically, in a modified version of the UG, participants' responses were compared in gain and loss sharing (Zhou and Wu, 2011). In the gain condition, the standard rules of the UG were applied. In the loss condition, accepting the offer led to the proposed division of the loss between players, whereas rejecting the offer led both players to lose the total amount of money. Results showed that the rejection rates to unfair offers were higher in the loss than in the gain condition. Other studies have demonstrated that the perception of ownership of property affects the way proposers make offers in the UG. In particular, proposers allocated more chips to the responder in the taking (i.e., the property is located at the responder, and the proposer decides how many chips to take from the responder) than in the giving condition (i.e., the property is located at the proposer, and the responders decides how many chips to give to the responder) (Leliveld et al., 2008). Indeed, the way in which the information is formulated, in terms of gain or loss, has been found to influence people's decisions. This effect is known in literature as the framing effect, which is one of the psychological phenomena explained within the prospect theory framework (Kahneman and Tversky, 1979), whereby people: (a) perceive the different options in terms of potential gains or potential losses compared with a neutral reference point, (b) consider the losses most salient than the corresponding gains (the unpleasantness of losing Euro 1000 is a stronger feeling than the pleasantness of winning the same amount), and (c) are more inclined to make risky choices in the domain of losses.

In the classic UG, offers are typically formulated so as to provide the respondent with all the information on how the money will be distributed between the two players: for example, if the sum to be divided is 10 euros, the offer is worded as "I take 8 Euro/You take 2 Euro" (Sanfey et al., 2003; van't Wout et al., 2006; Moretti et al., 2009). However, the two pieces of information are complementary. Therefore, on the one hand, one piece of information might be sufficient to make a decision, on the other, the way the offer is framed might well affect decision-making.

In a previous psychophysiological study (Sarlo et al., 2012), some of us used a modified version of the UG in which bids were manipulated through two different frames: the expression "I give you" was considered as a *gain* frame, since it focuses on money the respondent would receive if she agreed with the proponent; on the contrary, the expression "I take" was considered to frame the *losses*, since it is focused on the money that would be removed from the respondent in the event that she accepted the offer. Heart rate and skin conductance were also recorded in response to offers as indices of physiological activation. The results indicated that manipulating the frame had an effect both at the behavioral and physiological levels in males only. They showed a psychophysiological pattern suggesting a defense response (increased heart rate and skin conductance) when the offer was framed as a loss rather than as a gain, and a higher rate of rejection under the loss than the gain frame with mid-value offers (3 out of 10C). Accordingly, in the present study we hypothesized that the frame "I take" might elicit stronger bodily responses because it might be interpreted more negatively.

The framing effect has been investigated in two previous neuroimaging studies using a financial decision-making task (De Martino et al., 2006; Roiser et al., 2009). Consistent with the prospect theory assumptions, participants preferred the sure over the gamble options in the gain frame condition, and chose the gamble over the sure options in the loss frame condition. fMRI data showed that choices consistent with such framing effect were associated with amygdala activity, likely reflecting automatic emotional reactions (but see also Talmi et al., 2010). Other studies have indicated that risk aversion may also be mediated by activation of the anterior insula (Kuhnen and Knutson, 2005; Liu et al., 2007), suggesting that enhanced sensitivity to loss-framed information is associated with negative emotions and reward-related processing (Phan et al., 2002).

This is the first study to date that has investigated the neural mechanisms underlying framing effect in the UG. Based on these extant literature, first we expected to replicate the findings that correlate the activation of areas previously associated with unfairness, such as medial prefrontal cortex and anterior insula, to the type of response; moreover, we predicted a significant effect of loss ("I take") vs. gain ("I give you") frame in emotional areas such as the amygdala and anterior insula.

# **METHODS**

## **PARTICIPANTS**

A total of 17 males right-handed [mean ± SD: 93*.*5 ± 9*.*9, Edinburgh Inventory test, (Oldfield, 1971)] healthy subjects (mean age ± SD: 27*.*35 ± 3*.*88 years; age range: 22–36) were included. Male subjects were preferred to female subjects because in Sarlo et al. (2012) they more consistently showed the effect of frame. All subjects were native speakers of Italian with comparable levels of education. All subjects had normal or correctedto-normal vision and reported no history of neurological illness, psychiatric disease, or drug abuse according to their responses on self-report measures. None had any previous knowledge of the UG. All participants gave informed consent to participate in the study. The study was approved by the local ethics committee.

### **TASK AND STIMULI**

Task, stimuli, and experimental set-up were similar to those employed in a previous psychophysiological study (for details see Sarlo et al., 2012). Participants underwent a session of 16 min. The experimental instructions (see Appendix for an English-translated instruction sheet) can be subsumed as follows: participants played as *responders*. They were told that previous participants played as *proposers* and made offers by deciding how to split the amount of 10 euro at each trial (*N* = 62) that had been available by the experimenter. The participant had to decide either to accept or to reject the offer, by pressing one of two response keys. If participants accepted the offer, both (*proposer* and *responder*) will get the money as suggested, whereas if they rejected the offer, none of the players would get any of the money.

Although participants were told that they were interacting with human proposers, they were actually presented with offers defined a priori by the experimenter. There were three possible offers (factor OFFER): unfair [1C], middle [3C], and fair [5C] (in "1C out of 10" the *responder* is offered only 1 of the money at stake). Each offer was framed in two different ways (factor FRAME: "I give you/I take"). Each participant received the full range of offers, which were presented in different orders across subjects, with the constraints that (a) all the three offers should be presented first, and then repeated, (b) no more than two offers formulated with the same frame should appear consecutively, (c) the same amount of money (in the two different frames) should not be offered consecutively.

Participants were told that the proposers would receive feedback only at the end of the experiment (i.e., "covered" UG, which prevents strategic use of rejections; see Oldfield, 1971; Zamir, 2001; Civai et al., 2012; Corradi-Dell'Acqua et al., 2012). At the beginning of the experiment, participants were also told that their compensation for participating in the experiment would be proportional to the amount of money gained during the UG. Instead, irrespective of the task performance, they received the same amount of money as compensation after completion of the experiment. The subjects were not informed at the end of the experiment that we used a flat rate. An informal debriefing was carried out to assess whether participants believed whether offers were genuinely human.

## **EXPERIMENTAL SET-UP**

Participants lay supine in the MR scanner with their head fixated by firm foam pads. Presentation of the stimuli and their synchronization with the MR scanner were realized by the software Presentation® (version 9.9, Neurobehavioral Systems Inc., CA, USA). Stimuli were projected through a VisuaStim Goggles system (Resonance Technology). Subjects responded by pressing the corresponding keys of an MRI-compatible response device (Lumitouch, Lightwave Medical Industries, Coldswitch Technologies, Richmond, CA).

For each experimental trial a fixation point (500 ms) was presented, followed by the offer (e.g., "I give you/I take 5C", 6000 ms), after which a 2 s display indicated that the response ("accept"/"reject") could be made ("decision slide"). Trials were intermixed by inter-trial intervals ranging randomly from 3060 to 6720 ms with an incremental step of 60 ms. Instructions emphasized that the participants should press the selected key when the decision slide appeared on the screen.

Each experimental session included 62 randomized trials, including 54 experimental trials [3 (GAIN: 1C, 3C, 5C) × 2 (FRAME: "I take," "I give you") × 9 repetitions], yielding a total of 27 offers for each frame condition and 8 trials of no interest (2 offers with gain 2C and 2 offers with gain 4C), were included for each frame condition in order to represent the full range of offers the hypothetical proposers would make, while keeping reasonable the total number of trials (cf. Sarlo et al., 2012). Therefore, we focused on the trials representing the very unfair, the mid-value, and the very fair offers, according to previous studies (Polezzi et al., 2008; Civai et al., 2010). Eight null events (i.e., blank screens), perceived as a prolongation of the inter-trial period, were randomly interspersed among the event trials to increase the power of estimating the BOLD response (Dale and Buckner, 1997) and 30 s. of low level baseline (i.e., fixating a cross placed at the centre of the screen for 15 s at the beginning and 15 s at the end of the experiment). We therefore investigated the effect of two factors, GAIN and FRAME. Prior to the experiment, subjects practised the task outside the scanner (*N* = 20 trials): subjects were told that, in order to acquire familiarity with the structure of the task, they had to play some fake trials on a computer outside the scanner, being informed that the offers were not real, and the subject was told that they wouldn't have been calculated in the final payoff. The offers could take any amount.

#### **fMRI DATA ACQUISITION**

A 3.0-T Philips Achieva (Philips Medical System, Netherlands) whole-body scanner was used to acquire T1-weighted anatomical images and functional images using a SENSE-Head-8 channel head coil and a custom-built head restrainer to minimize head movements. Functional images were obtained using a T2∗-weighted echo-planar image (EPI) sequence of the whole brain. EPI volumes for the main experiment (*N* = 455, lasting 14.2 min) contained 30 transverse axial slices (repetition time, *TR* = 2000 ms; echo time, *TE* = 35 ms, field of view, *FOV* = 23 cm, acquisition matrix: 128 × 128, slice thickness: 3 mm with no gaps, 90◦ flip angle, voxel size: 1*.*79 × 1*.*79 × 3 mm) and were preceded by 5 dummy scans that allowed the MR signal to reach a steady state. After functional neuroimaging, high-resolution anatomical images were acquired using a T1-weighted 3-D magnetization-prepared, rapid acquisition gradient fast filed echo (T1W 3D TFE SENSE) pulse sequence (*TR* = 8*.*2 ms, *TE* = 3*.*76 ms, *FOV* = 24 cm, 190 transverse axial slices of 1 mm thickness, 8◦ flip angle, voxel size: 1 × 1 × 1 mm) lasting 8.8 min.

## **fMRI DATA PROCESSING AND WHOLE BRAIN ANALYSIS**

fMRI data pre-processing and statistical analysis were performed on UNIX workstations (Ubuntu 8.04 LTS, i386, http://www*.* ubuntu*.*com/) using MATLAB r2007b (The Mathworks Inc., Natick, MA/USA) and SPM5 (Statistical Parametric Mapping software, SPM; Wellcome Department of Imaging Neuroscience, London, UK). Dummy scans were discarded before further image processing. Preprocessing included spatial realignment of the images to the reference volume of the time series, segmentation producing the parameter file used for normalization of functional data to a standard EPI template of the Montreal Neurological Institute template provided by SPM5, re-sampling to a voxel size of 2 × 2 × 2 mm, and spatial smoothing with a 6-mm FWHM Gaussian kernel to meet the statistical requirements of the General Linear Model and to compensate for residual macro-anatomical variations across subjects. We performed a whole brain random effects analysis closely following the model previously used by some of us (Civai et al., 2012), with FRAME and the RESPONSE TYPE (i.e., reject or accept) as factors to account for neural activations related to accepting or making a rejection for offers proposed as different frames. This analysis counted 13/17 subjects, because it was necessary that all the subjects considered had rejections in both the frames in order to perform an ANOVA without empty cells. We calculated the number of cells for each condition. There was a mean of 16*,* 56 ± 9*,* 15, 12*,* 17 ± 8*,* 67, 15*,* 06 ± 9*,* 82, and 13*,* 28 ± 9*,* 39 cells for each of the four conditions: I TAKE\_ACCEPT; I TAKE\_REJECT; I GIVE YOU\_ACCEPT; I GIVE YOU\_REJECT, respectively. Importantly, the mean number of cells did not differ significantly across experimental conditions [frame, *F(*1*,* <sup>12</sup>*)* = 3*.*37, *p* = 0*.*089, n.s.; resp type, *F(*1*,* <sup>12</sup>*)* = 1*.*012, *p* = 0*.*33, n.s.; frame × resp type interaction, *F(*1*,* <sup>12</sup>*)* = 0*.*01, *p* = 0*.*90, n.s.], thus cells were comparable between conditions. On the first-level analysis, we modeled as the regressors of main interest the response types (accept/reject) and the frames "I take" and "I give you" (I\_take/accept, I\_take/reject, I\_give\_you/accept, and I\_give\_you/reject) and their temporal derivative. We also included the motor response as a further regressor of no interest In addition, to correct for motion artifacts, subject-specific realignment parameters were modeled as covariates of no interest. Low-frequency signal drifts were filtered using a cut-off period of 128 s. At the single subject level, specific effects were assessed by applying appropriate linear contrasts to the parameter estimates of the experimental conditions resulting in *t*-statistics for each voxel. For the second-level random effects analyses, contrast images obtained from individual participants were entered into a one-sample *t*-test to create a SPM{T}, indicative of significant activations specific for this contrast at the group level. We used a threshold of *p <* 0*.*05, corrected for multiple comparisons at the cluster level [using family-wise error (FWE)], with a height threshold at the voxel level of *p <* 0*.*001, uncorrected. The anatomical localization of the functional imaging results was performed using the SPM Anatomy toolbox (Eickhoff et al., 2005). To reveal the nature of the interactions, beta-values were extracted using the rfxplot toolbox (Glascher, 2009) implemented in SPM5. *t*-tests were performed over the extracted percentage signal change values to further investigate the functional properties of the areas of activation. Statistical analysis was performed with SPSS 14.0 software.

Finally, data were analyzed by using a second design matrix accounting for effects of the FRAME and the GAIN as factors, following the model previously used by some of us (Civai et al., 2012). We modeled the offers as a categorical factor with 3 levels fair (5C), middle (3C), and unfair (1C) yielding to a 2 × 3 factorial design with six conditions and their temporal derivative. The rest of the analysis was carried out as in the first model. With respect to the effect of frame, in this analysis we were interested at the contrast gain 5: "I take" vs. "I give you." We reasoned that "gain 5C" could represent a good testing condition for investigating the frame effect, since it is an equal fair offer and the effect of frame on the perception of unfairness should be null; finding an effect of "I take" vs. "I give you" frames for gain 5 would strengthen the idea that the way offers are framed influence how people perform the task.

## **STATISTICAL ANALYSES OF BEHAVIORAL DATA**

SPSS for Windows (version 14.0) was used for performing a repeated measure ANOVA with within-subject factors type of "frame" ("I take," "I give you") and "gain" (1C, 3C, 5C) on the subjects' rejection rates and response times (RTs) data. All *posthoc* comparisons between single factors were carried out using LSD Fisher's test (α ≤ 0*.*05).

# **RESULTS**

## **BEHAVIORAL RESULTS**

At the debriefing, all participants reported that they believe the offers came from genuine humans.

#### *Rejection rates*

We found a significant main effect of gain, *F(*2*,* <sup>32</sup>*)* = 23*.*91, *p <* 0*.*001, with significantly less rejections for gain 5C vs. 1C (mean ± sem, 7*.*18 ± 4*.*43 vs. 75*.*8 ± 10*.*53, *p <* 0*.*001), and for 5C vs. 3C (7*.*18 ± 4*.*43 vs. 52*.*2 ± 11*.*65, *p <* 0*.*002) compared with 3C vs. 1C (52*.*2 ± 11*.*65 vs. 75*.*8 ± 10*.*53, *p* = 0*.*076, n.s.) (See **Figure 1**). The main effect of frame [*F(*1*,* <sup>16</sup>*)* = 0*.*35, *p* = 0*.*56] and the frame × gain interaction [*F(*2*,* <sup>32</sup>*)* = 0*.*8, *p* = 0*.*45] were not significant.

Identical effects were found also when we removed from the analysis four participants who never rejected an offer, because it was necessary that all the subjects considered had rejections in both the frames in order to perform an ANOVA without

empty cells [frame, *<sup>F</sup>(*1*,* <sup>12</sup>*)* <sup>=</sup> <sup>0</sup>*.*93, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*76, n.s., <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*0022; gain, *<sup>F</sup>(*2*,* <sup>24</sup>*)* <sup>=</sup> <sup>41</sup>*.*<sup>5</sup> *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*<sup>001</sup> <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*679, with significantly less rejections for gain 5C vs. 1C (5*.*98 ± 16*.*12 vs. 95*.*29 ± 15*.*36, *p <* 0*.*001), and for 5C vs. 3C (5*.*98 ± 16*.*12 vs. 64*.*42 ± 45*.*08, *p <* 0*.*002) compared with 3C vs. 1C (64*.*42 ± 45*.*08 vs. 95*.*29 ± 15*.*36, *p* = 0*.*070, n.s.); frame × gain, *F(*2*,* <sup>24</sup>*)* = 0*.*94, *p* = 0*.*40, n.s. <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*0201].

## **fMRI RESULTS**

#### *Task-related network*

The extensive network of areas recruited by the task (task *>* implicit baseline contrast) involved clusters of activity in: (i) the cerebellum bilaterally, extending to the inferior and to the superior temporal gyrus, the amygdala, the insula and the superior parietal lobe; (ii) the left middle temporo-occipital gyrus; (iii) the middle cingulate cortex, extending to the left supplementary motor area (SMA); (iv) the superior frontal gyrus bilaterally, extending to the right SMA; and (v) the middle frontal gyrus bilaterally (see **Figure 2A**, **Table 1**).

#### *Effects of the FRAME and RESPONSE TYPE (i.e., reject or accept)*

**Figure 2** show the fMRI results for the response-related effects: network of areas differentially recruited by response "reject" (relative to "accept") for frame "I give you" **(B)** and for frame "I take." **(C)** For the main effect of frame: "*I take" vs. "I give you" (and vice versa),* no differential activation was found at the predefined statistical threshold. Based on previous neuroimaging studies on the framing effect described in the introduction (De Martino et al., 2006), we hypothesized that emotional areas such as anterior insula and amygdala, were involved in processing the frame; moreover, further imaging studies reported activations in the operculo/insular cortex associated with pain processing (e.g., Lötsch et al., 2012) and in interoceptive awareness and the representation of visceral responses associated with emotional situations (Lamm et al., 2007). Thus, a hypothesis-driven region of interest (ROI) analysis (Friston, 1997) was performed in which we tested significant increases of neural activity in the operculo/insular cortex [Anatomy Toolbox (Eickhoff et al., 2005)], for the main effect of frame. We found significant activation within the operculo/insular cortex bilaterally associated with the frame "I take" vs. "I give you." No differential activation was found for the reverse comparisons (see **Figure 2D**, **Table 1**). In the ROI analysis performed on the amygdala [Anatomy Toolbox (Eickhoff et al., 2005)] for the main effect of frame we did not found significant activation within this area.

## *Frame* **×** *response type interaction: [(FRAME\_I take: Accept> Reject)>(FRAME\_I give you: Accept > Reject)] (and vice versa)*

The frame "I take" for accepted trials (vs. rejected) controlled for the frame "I give you," differentially activated (i) the right precuneus, extending to the right superior parietal lobe (Area 7a), (ii) the right rolandic operculum/insular cortex, (iii) the right calcarine gyrus, extending to the right cuneus, (iv) the right superior temporal gyrus, (v) the left superior parietal lobe (Area 7a) extending to the left Area 2, and (vi) the left anterior cingulate (see **Figure 3**, **Table 1**). This interaction was due to an increase of neural activity for accepting frame "I take" vs. rejecting (*p* = 0*.*008, *p* = 0*.*022, *p* = 0*.*001, *p* = 0*.*023, *p* = 0*.*011, *p* = 0*.*04, respectively), which was significantly higher than that associated with accepting frame "I give you" vs. rejecting (*p* = 0*.*09, *p* = 0*.*08, *p* = 0*.*09, *p* = 0*.*1, *p* = 0*.*1, *p* = 0*.*08, n.s., respectively). No differential activation was found at the predefined statistical threshold for the reverse comparisons.

### *Gain-related effects*

This model accounted for effects of the FRAME (i.e., "I take," "I give you") and the GAIN [i.e., fair (5C) middle (3C) and unfair (1C)] factors. Irrespective of frame, unfair (1C) gain [vs. middle gain (3C)], differentially activated the anterior cingulate cortex (ACC). Furthermore, fair (5C) gain [vs. middle gain (3C)], differentially activated (i) the left superior parietal lobe (Area 7a), extending to the precuneus, (ii) the middle cingulate cortex.

For the fair (5C) gain, the frame "I take" (vs. "I give you") differentially activated the left occipito-temporal junction and the middle temporal gyrus (see **Figure 4**, **Table 2**). No differential activation was found at the predefined statistical threshold for the other comparisons (see **Table 2**).

We performed also an analysis by including a parametric modulator for the factor "gain." The parametric approach did not yield significant results at the predefined threshold.

**FIGURE 2 | (A)** Common neural networks associated with the UG task; network of areas differentially recruited by response "reject" in green (relative to "accept," in red) for frame "I give you" **(B)** and for frame "I take" **(C)**. Activations are displayed on a

rendered template brain provided by spm5. **(D)** Insula/rolandic opercular areas differential recruitment by the frame "I take" (relative to frame "I give you") displayed on a single subject template brain provided by spm5.


**Table 1 | Whole brain analysis for the model accounting for** *frame* **and** *type of response* **related effects: brain regions showing significant relative increases of BOLD response associated with the experimental conditions.**

*For each region of activation, the coordinates in MNI space are given in reference to the maximally activated voxel within an area of activation, as indicated by the highest Z-value (p < 0.05, corrected for multiple comparisons at the cluster level, height threshold p < 0.001, uncorrected). \*pSVC < 0.05, corrected. L/R, left/right hemisphere.*

## **DISCUSSION**

In this fMRI study we have investigated the neural mechanisms underlying the economical decisions people make when the information on which they rely is formulated in terms of gain or loss. Participants played a modified fMRI version of the UG with different bids preceded by two different frames: the expression "I give you" (*gain*) focusing on money the respondent would receive if she/he agreed with the proponent, and the expression "I take" (*loss*) focusing on the money that would be removed from the respondent in the event that she/he accepted the offer. Behaviorally, unfair offers were equally often rejected in both conditions. This is different from what was found in the study by Sarlo et al. (2012) in which participants rejected more when the offer was framed as a loss rather than as a gain. Our failure to confirm this result may be due to the smaller sample of subjects employed in the present study.

#### **TASK-RELATED NETWORK**

Since the task we used was a modified version (Sarlo et al., 2012) of the classical UG, we first describe the task-related network. Overall, the network that was associated with the task included

**give you"\_Accept** *>* **"I give you"\_Reject)].** Group mean

clusters of activation in key areas which have been classically found in previous studies investigating the neural underpinning of the UG (e.g., Sanfey et al., 2003; van't Wout et al., 2005; Knoch et al., 2006, 2008; Koenigs and Tranel, 2007; Tabibnia et al., 2008;

2009)].

Moretti et al., 2009; Güroðlu et al., 2010, 2011; Civai et al., 2012; Corradi-Dell'Acqua et al., 2012), such as the insular cortex bilaterally, the middle cingulate cortex, the superior frontal and the middle frontal gyri bilaterally, the inferior and superior temporal lobe. Interestingly, the task-related network included also the amygdala bilaterally, known to be related to the mediation of aggressive responses (Nelson and Trainor, 2007) and of biasing decision-making (Bechara et al., 2003; De Martino et al., 2006), and it has been found activated also in a previous fMRI study on the UG (Gospic et al., 2011).

Structures normally involved in mental calculation (Rickard et al., 2000; Zago et al., 2001; Hanakawa et al., 2003), such as the right parietal and the right precuneus, were significantly activated in the frame-by-decision interaction, in which the responder takes but, in this condition, the player accepts. When one accepts in the loss frame, one deviates more from her "expected" response. This deviation from the behavior we expect from her, could be accompanied by an increase of mental calculation resources and processing.

Importantly, all these activations have been always related to processes such as emotional processing (Sanfey et al., 2003), theory of mind (Gallagher and Frith, 2003; Amodio and Frith, 2006), cognitive processing (Sanfey et al., 2003) such as executive control, goal maintenance, and the monitoring/control of one's emotional responses (van't Wout et al., 2005; Knoch et al., 2006, 2008; Koenigs and Tranel, 2007; Moretti et al., 2009; Güroðlu et al., 2010, 2011; Baumgartner et al., 2011) triggered by the task.

## **FRAME BY DECISION INTERACTION IN THE OPERCULO-INSULAR CORTEX AND THE ANTERIOR CINGULATE CORTEX**

Despite the lack of significant interaction at the behavioral level, at the neural level we observed a frame-by-response interaction, revealing an increase of neural activity in the right rolandic operculum/insular cortex. This interaction was due to an increase of neural activity for accepting frame "I take" vs. rejecting, which was significantly higher than that associated with accepting frame "I give you" vs. rejecting. Interestingly a hypothesis-driven ROI analysis performed for testing significant increases of neural activity in the operculo/insular cortex, showed a significant activation within the operculo/insular cortex bilaterally associated with the frame "I take" vs. "I give you." Sanfey et al. (2003) found that a stronger activation in the anterior part of the insula when evaluating an unfair offer was associated to rejections. In contrast, in our study we found a stronger activation in the posterior part of the insula for accepting (compared with rejecting) the frame "I take" vs. "I give you." We interpreted the acceptance effect we found as related to a discrepancy between expected response and my decision [see also Güroðlu et al. (2010), for a similar interpretation]. Our results extend the interpretation of the role of the insula put forward by Sanfey et al. (2003) and suggest that this region may be characterized by two different functions: the anterior part of the insula might evaluate the outcome, while the posterior part of the insula might evaluate the response to the outcome. Based on our findings we can add that the posterior insula is also sensitive to the frame in which offers have been formulated.

Our cluster of activation in the operculo-insular cortex is localized more posterior than the usual one found in the anterior insula in UG fMRI studies, e.g., Sanfey et al. (2003) and also the one found in previous studies performed by some of us (Civai et al., 2012; Corradi-Dell'Acqua et al., 2012). Nevertheless, it has been shown that among other regions, such as the thalamus, the insular, anterior cingulate, primary and secondary somatosensory, premotor and supplementary motor cortices, the operculo-insular cortex is a crucial part of the pain matrix (Treede et al., 1999; Peyron et al., 2000; Apkarian et al., 2005; Bingel and Tracey, 2008). This is particularly relevant since the activation found in the anterior insula during the UG have been classically interpreted as unfair offers triggering negative emotions, given that many studies have found a crucial involvement of this area in processing emotional states, pain and distress (Damasio et al., 2000; Calder et al., 2001; Wicker et al., 2003; Corradi-Dell'Acqua et al., 2011). Evidence for the operculo-insular cortex involvement in pain processing came from studies using PET, evoked potentials or fMRI techniques (Peyron et al., 2002; Frot et al., 2007; Baumgartner et al., 2010; Oertel et al., 2012), and from studies involving direct stimulation of this area (Mazzola et al., 2009) **Table 2 | Whole brain analysis for the model accounting for** *frame* **and for** *gain* **effects: brain regions showing significant relative increases of BOLD response associated with the experimental conditions.**


*For each region of activation, the coordinates in MNI space are given in reference to the maximally activated voxel within an area of activation, as indicated by the highest Z-value (p < 0.05, corrected for multiple comparisons at the cluster level, height threshold p < 0.001, uncorrected). L/R, left/right hemisphere. As all the values are value p < 0.05, corrected for multiple comparisons at the cluster level, height threshold p < 0.001.*

and of the insular cortex (Mazzola et al., 2006). Other authors have previously found activation in a cluster localized more posterior than the anterior insula. In one of those studies, the authors (Wright et al., 2011) used a modified version of the UG and varied the social context, by inducing thus a bias in participants acceptance of objectively identical offers. They found that the objective social inequality was integrated with social context in posterior and mid-insula. Consistently, in another study (Hsu et al., 2008) it has been shown that posterior insula activity negatively correlated with inequality.

The frame × response type interaction contrast included also the fair C5 offers. We ruled out the possibility that the fair C5 offers drove the effect, since the rejection rates for fair C5 offers in the "I take" frame did not significantly differ from the frame "I give." We also would like to argue that gain 5C, being the most equal gain, is the condition that more than the others shows the frame effect: precisely because it is an equal and fair offer, the effect of frame should be null. Instead, we found that the activation in the OT junction was significantly modulated by the effect of frame for fair, C5 offers.

In our study the activation of the operculo-insular cortex was significantly increased when participants accepted (vs. rejected) the offers presented in the frame "I take," as compared to the frame "I give you." We reasoned that in the loss frame one should be more prone to reject with respect to the gain frame; it follows that when participants accept in the loss frame they deviate more from their "expected" response, even though this interpretation is speculative, as we cannot provide behavioral evidence to support the expectancy hypothesis. Accordingly, the operculoinsular cortex might signal this deviation from participants' own expected behavior. It has been proposed that the equal treatment is a default social norm, and its violation is signaled by the anterior insula (Civai et al., 2012). Further evidence supporting the view that the anterior insula signals the level of inequity aversion, and, more broadly, norm violations came also from another fMRI study (Hsu et al., 2008) in agreement with the idea that anterior insula plays a critical role in detecting social norm violations (Spitzer et al., 2007; King-Casas et al., 2008; Strobel et al., 2011), thus extending its role beyond emotional involvement (Sanfey et al., 2003). Importantly, it has been shown that the frame "I take," by acting as a loss frame, elicited the characteristic defensive response pattern that is evoked by aversive stimulation, in which increases in skin conductance are coupled with increases in heart rate (Güth et al., 1982; Sarlo et al., 2012). To sum up, the role of the anterior insula in the UG in the studies reviewed above is comparable with the one we found in the operculo-insular cortex. In addition, we add that the operculoinsular cortex is modulated by the frame in which the offers are formulated.

A further interpretation might be that operculo-insular cortex activation could be somewhat related to processes of agencyattribution and/or adoption of an egocentric vs. allocentric reference frame, and the effect may arise from the "linguistic" context involving the proposer alone ("I take") or the proposer along with the responder ("I give you"), thus modulating the activity of mechanisms of self-other distinction that are associated with posterior insula and rolandic operculum (see Vogeley and Fink, 2003; Sperduti et al., 2011). In our study, the activation of the operculoinsular cortex was significantly increased when participants were processing the frame "I take," as compared to the frame "I give you" and accepted (vs. rejected) the offers. Here agency has to be attributed to the person to whom the proposal of how to split the money is made, independent of the frame "give" or "take." This might appear in contrast with the role played by the insula in first person perspective attribution (see Vogeley and Fink, 2003; Sperduti et al., 2011) unless the player imagine changing his own perspective in to the proposer's one.

Consistently with previous studies in which the ACC showed an increased activation for unfair compared with fair offers, e.g., Sanfey et al. (2003), we found an increased activation in the ACC for unfair 1C as compared to mid-value 3C gains. Some authors, e.g., Sanfey et al. (2003) argued that the ACC has been implicated in detection of cognitive conflict (Botvinick et al., 1999; MacDonald et al., 2000), and the activation of the ACC in the context of the UG is related to the conflict between cognitive and emotional motivations. As a new feature, we also found that the ACC activation in the frame by response interaction, with increased activation for accepting (vs. rejecting) gains presented in the frame "I take," as compared to those presented in the frame "I give you," which corresponds to the most unfair condition, albeit participants accept that the proposer takes money. This condition might trigger a conflict between cognitive and emotional motivations, which in turns activates the ACC. It has been suggested that together with the insula, the ACC activation might be related to behavior that deviates from participants' personal standards (Güroðlu et al., 2011). It has been shown that, by varying degrees of intentionality, the ACC activation was increased for accepting unfair offers in the no-alternative context and for rejecting an unfair offer in fairand hyperfair-alternative contexts (Güroðlu et al., 2011). Taken together, these results indicate that accepting that the proposer takes the money, independent of the gain, is indeed a deviant choice with respect to what one normally does (Güroðlu et al., 2011).

#### **IMPLICIT MENTAL SIMULATION MECHANISMS TRIGGERED BY THE UG**

That the UG could trigger mechanisms related to mental simulation has never been proposed. With the term simulation we refer to the mental process by which people mentally visualize, or move or feel and experience situations, which occurs in the absence of the appropriate external stimuli or sensory input (mental imagery is sometimes colloquially referred to as "visualizing," "seeing in the mind's eye," "hearing in the head," "imagining the feel of," etc.) (Kosslyn et al., 1995a, 2001). It has been largely accepted that people use mental imagery, for instance, during memory retrieval, problem solving, producing descriptions, mental practice, and motivational states (Kosslyn, 1980). Thus, a mental process involving a first or third person perspective could well be carried out through imagery (Vogeley et al., 2004). Importantly, mental imagery can occur after explicit instructions (Jeannerod, 1999) but it can also be implicitly triggered (Jeannerod and Frak, 1999); implicit mental imagery occurs when subjects, even if they receive no instruction to imagine, unconsciously imagine the scene or the action while performing another task, e.g., during mental rotation of body parts (e.g., Zacks et al., 1999; Kosslyn et al., 2001), handedness recognition of a visually presented hand (e.g., Parsons and Fox, 1998), judgment as to whether an action would be easy, difficult or impossible (Johnson et al., 2002), or recognizing and understanding actions of other individuals (e.g., Johnson et al., 2002). In performing the UG, although subjects received no instruction to do so, they could represent in their mind of the imagine the action associated with the task. That individuals imagine

the situations as if they were real and feel pain when the most disadvantageous conditions are encountered could well explain why regions found associated in processing pain such as the opercular/insular cortex were found activated when subjects performed the UG. It is conceivable that while performing the UG the participants (implicitly) simulate sensations, actions, emotions, anticipating the action consequences, switch between first and third person perspective, although not instructed to do so. Accordingly, in our study we found significant clusters of activation in areas involved in mental imagery, strongly suggesting that one of the mechanisms supporting the UG performance could well be mental imagery. Indeed, at variance with the results previously found in fMRI studies on the UG, interestingly the task-related network included also a significant activation in the left superior parietal cortex, which was localized in the primary somatosensory area (Area 1). This activation is typically found in studies in which subjects actually experience the sensation or the movement, or when they imaging them (Tomasino et al., 2010). This finding may be interpreted as if the subjects implicitly simulated the gain/loss. Somatoperception corresponds to the process of perceiving the body itself, and particularly of ensuring somatic perceptual constancy (Longo et al., 2010). The somatosensory cortex is reported in studies requiring mapping of subjective feeling states arising from bodily responses (Critchley et al., 2004). It is relevant here the role of somatosensory cortices in sensory imagery of affectively-significant states. Somatosensory-based memories can be reactivated by the anterior emotion network (Damasio, 1994). It has been shown that repetitive transcranial magnetic stimulation (rTMS) to the face S1 representation impaired recognition of facial emotional expressions (Pitcher et al., 2008) and that the observation of erotic images or mutilated bodies as compared to neutral items activated the right SI and SII (Rudrauf et al., 2009). The S1 activation during the UG task thus might be related both to an increased attention to one's bodily states as if the neural representation of the experiencing subject's body is a vehicle of their emotional experience (Longo et al., 2010). The UG is a self-centered task, thus it is reasonable that the left S1 and area 2 activations might reflect mental imagery of the sensations they would physically experienced during the UG.

With respect to the parietal lobe, we found that the left superior parietal lobe (Area 7a) was significantly activated independent of the frame in which offers were formulated, i.e., "I give you" or "I take" by gain 5C as compared to 3C, thus for equal offers. In addition, the left area 7a was significantly activated by the frame by decision interaction, in which the responder takes but this time the player accepts. When you accept in the loss frame you deviate more from your "expected" response. It is not only the insula signaling this deviation from your own expected behavior, but also area 7, which has been related to egocentric (body- and body part-centered) coordinates coding (Makin et al., 2007), to the processing of multimodal integrated spatial representations in body-centered coordinates (Felician et al., 2004) and to updating postural representations of the upper limb (Pellijeff et al., 2006). In a previous study, some of us found that the left posterior IPS codes an allocentric, but not egocentric, visual model of

the body (the body structural description) (Corradi-Dell'Acqua et al., 2009). Taken together, these studies suggest that a left area 7a activation may reflect a continuous updating of egocentric and allocentric coordinates while playing the UG. Another subregion of the left parietal lobe we found activated in the frame by decision interaction was area 2, which is a somatosensory area, e.g., Grefkes and Fink (2005). The cluster of activation in area 7a extended including also the left precuneus. This region has been found activated in studies addressing episodic memory and the creation of imaginary or future personal scenarios (Cavanna and Trimble, 2006; Buckner and Carroll, 2007). It is possible that participants implicitly imagine the offers in terms of past or hypothetical future scenarios and fictive losses, see Kirk et al. (2011), or they could implicitly simulated situations in which the responder takes and the player accepts shifting from a 1st or a 3rd person perspective imagery (Ruby and Decety, 2001).

We also found that the left occipito-temporal junction was activated for the equal offer ("I take" vs. "I give you" for gain 5C). Our coordinates of the left occipito-temporal junction cluster are in the proximity to previously reported locations of extrastriate body area (EBA) in human brain (Arzy et al., 2006; Corradi-Dell'Acqua et al., 2009). Previous studies implicated the EBA by many body-part related processes including self-generated (Astafiev et al., 2004) and goal-directed (Takahashi et al., 2008) movements, as well as reaching to kinesthetically defined targets (Darling et al., 2007) and during imagery of the tool-use in near and far space (Tomasino et al., 2012). In this vein, the occipito-temporal activation clusters are modulated by the equal offer 5C, which can be hypothesized as being more unpleasant by the frame "I take" where is the proposer taking the money. Accordingly, the activation could be related to the generation of an action that might be considered a social confrontation, such a rejection. In the same line, it has been shown that activation of the middle occipito-temporal cortex was modulated by emotional and social information while participants viewed and categorized affective pictures that varied on two dimensions: emotional content (i.e., neutral, emotional) and social content (i.e., faces/people, objects/scenes) (Norris et al., 2004).

We found that the right calcarine gyrus and the right cuneus were significantly activated by the frame by decision interaction,

## **REFERENCES**


basis of embodiment: distinct contributions of temporoparietal junction and extrastriate body area. *J. Neurosci.* 26, 8074–8081. doi: 10.1523/JNEUROSCI.0745- 06.2006


in which the responder takes but this time the player accepts. We can exclude that this activation is related to visual processing of the stimuli since it is a product of the interaction term. Rather, we suggest that in this condition there is an increase of implicit visual imaginary processes, as described above, which triggers an increase of activation in areas related to visual imagery of scenes and characters (Kosslyn et al., 1995a,b, 2001; Kosslyn and Thompson, 2003), and an increase of activation in areas related to episodic memory retrieval during imagery such as the cuneus (Fletcher et al., 1995). It has been shown that V1 can be activated whenever images are formed, even if they are not necessarily used to perform a task (Klein et al., 2000). In that study, authors (Klein et al., 2000) used event-related fMRI to detect and characterize the activity in the calcarine sulcus during mental imagery. The results revealed reproducible transient activity in this area whenever participants generated or evaluated a mental image. This transient activity was strongly enhanced when participants evaluated characteristics of objects, whether or not details actually needed to be extracted from the image to perform the task (Klein et al., 2000). Interestingly, it has been shown that functional activation was larger in the right than the left hemisphere and larger in the occipital than in the occipitoparietal regions during processing of a series of pleasant, neutral, or unpleasant pictures (Lang et al., 1998). Importantly, both emotional and neutral pictures produced activity centered on the calcarine fissure, only emotional pictures also produced sizable clusters bilaterally in the occipital gyrus and in the right fusiform gyrus (Lang et al., 1998), and we argue probably related to implicit imagery processes.

To conclude, we argue that areas involved in imaginary task were found activated by our version of the UG, mainly in the frame by decision interaction, in which the responder takes and the player accepts. This condition corresponds to a loss frame, and here participants deviate from their "expected" response. Thus, deviation from participants' own expected behavior are signaled not only by the posterior insula/rolandic operculum but also trigger an increase of activation in areas related to mental imagery. Our findings extend the current understanding of the neural substrate of social decision making, by disentangling the structures sensitive to the way in which the information is formulated, i.e., framing effect, in terms of gain or loss, which influences people's decisions.

(2011). Dorsolateral and ventromedial prefrontal cortex orchestrate normative choice. *Nat. Neurosci.* 14, 1468–1474. doi: 10.1038/nn.2933


of the amygdala in decisionmaking. *Ann. N.Y. Acad. Sci.* 985, 356–369. doi: 10.1111/j.1749- 6632.2003.tb07094.x


*Neurosci*. 8, 424–431. doi: 10.1093/ scan/nss014


M., et al. (2004). The role of human left superior parietal lobule in body part localization. *Ann. Neurol.* 55, 749–751. doi: 10.1002/ana.20109


experts: a functional magnetic resonance imaging study. *Neuroimage* 19, 296–307. doi: 10.1016/S1053- 8119(03)00050-8


disrupting the right prefrontal cortex. *Science* 314, 829–832. doi: 10.1126/science.1129156


Joseph, J. E. (2007). Functional dissociation in frontal and striatal areas for processing of positive and negative reward information. *J. Neurosci.* 27, 4587–4597. doi: 10.1523/JNEUROSCI.5227- 06.2007


et al. (2008). Mentalizing in economic decision-making. *Behav. Brain Res.* 190, 218–223. doi: 10.1016/j.bbr.2008.03.003


sunny side of fairness: preference for fairness activates reward circuitry (and disregarding unfairness activates self-control circuitry). *Psychol. Sci.* 19, 339–347. doi: 10.1111/j.1467-9280.2008.02091.x


cortical representation of pain. *Pain* 79, 105–111. doi: 10.1016/S0304- 3959(98)00184-5


Zhou, X., and Wu, Y. (2011). Sharing losses and sharing gains: increased demand for fairness under adversity. *J. Exp. Soc. Psychol.* 47, 582–588. doi: 10.1016/j.jesp.2010. 12.017

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 10 February 2013; accepted: 14 June 2013; published online: 09 July 2013.*

*Citation: Tomasino B, Lotto L, Sarlo M, Civai C, Rumiati R and Rumiati RI (2013) Framing the ultimatum game: the contribution of simulation. Front. Hum. Neurosci. 7:337. doi: 10.3389/ fnhum.2013.00337*

*Copyright © 2013 Tomasino, Lotto, Sarlo, Civai, Rumiati and Rumiati. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# **APPENDIX**

## **INSTRUCTIONS**

During the experiment you will be asked to take part to the present scenario.

Another participant has been given the amount of 10 euro and that he/she has to decide how to split the amount of 10 euro with you. You cannot negotiate the proposal. You have the possibility to accept or reject the proposal, considering that:


Our study aim at observing the behavior of both the proposers, i.e., who proposes how to split the amount of money, and of the responder, who are asked to decide whether accepting or rejecting the offers.

In the past month we have involved (and scanned) a number of participants, asked to decide how they would split the amount of 10 euros, considering the rules mentioned above. The offers that you will be presented with, are therefore the proposals made by these participants.

All the participants, both the proposers and those who participated as responders like you, will receive a fixed compensation for participating in the experiment corresponding to 15 euro, plus a compensation equal to 10% of the amount of money gained.

You will not be able to identify the offers made by the single previous participants since the different proposals will be presented in a random order. The program will allow the experimenter only to load later on the offers made by the single proposers so to assign them the respective compensation.

# Segregation of the human medial prefrontal cortex in social cognition

# *Danilo Bzdok1,2, Robert Langner 1,2, Leonhard Schilbach3,4, Denis A. Engemann4,5, Angela R. Laird6, Peter T. Fox7,8 and Simon B. Eickhoff 1,2\**

*<sup>1</sup> Institute of Neuroscience and Medicine (INM-1), Research Center Jülich, Jülich, Germany*

*<sup>2</sup> Institute of Clinical Neuroscience and Medical Psychology, Heinrich Heine University, Düsseldorf, Germany*

*<sup>3</sup> Max-Planck-Institute for Neurological Research, Cologne, Germany*


*<sup>8</sup> South Texas Veterans Administration Medical Center, San Antonio, Texas*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Derek E. Nee, Indiana University, USA Mathieu Roy, University of Colorado Boulder, USA*

#### *\*Correspondence:*

*Simon B. Eickhoff, Institut für Neurowissenschaften und Medizin (INM-1), Forschungszentrum Jülich GmbH, Building 23.02, Universitätsstr 1, D-52425 Jülich, Germany e-mail: s.eickhoff@fz-juelich.de*

While the human medial prefrontal cortex (mPFC) is widely believed to be a key node of neural networks relevant for socio-emotional processing, its functional subspecialization is still poorly understood. We thus revisited the often assumed differentiation of the mPFC in social cognition along its ventral-dorsal axis. Our neuroinformatic analysis was based on a neuroimaging meta-analysis of perspective-taking that yielded two separate clusters in the ventral and dorsal mPFC, respectively. We determined each seed region's brain-wide interaction pattern by two complementary measures of functional connectivity: co-activation across a wide range of neuroimaging studies archived in the BrainMap database and correlated signal fluctuations during unconstrained ("resting") cognition. Furthermore, we characterized the functions associated with these two regions using the BrainMap database. Across methods, the ventral mPFC was more strongly connected with the nucleus accumbens, hippocampus, posterior cingulate cortex, and retrosplenial cortex, while the dorsal mPFC was more strongly connected with the inferior frontal gyrus, temporo-parietal junction, and middle temporal gyrus. Further, the ventral mPFC was selectively associated with reward related tasks, while the dorsal mPFC was selectively associated with perspective-taking and episodic memory retrieval. The ventral mPFC is therefore predominantly involved in bottom-up-driven, approach/avoidance-modulating, and evaluation-related processing, whereas the dorsal mPFC is predominantly involved in top–down-driven, probabilistic-scene-informed, and metacognition-related processing in social cognition.

#### **Keywords: social cognition, medial prefrontal cortex, meta-analytic connectivity modeling, resting state connectivity, functional decoding, data-mining**

## **INTRODUCTION**

Functional specialization in the human prefrontal cortex has been investigated since the middle of the nineteenth century primarily by lesion reports (Harlow, 1848, 1868; Broca, 1865). However, hard evidence derivable from functional double dissociations by prefrontal brain lesions is rare in humans (cf. Gaffan, 2002; Wilson et al., 2010). Nevertheless, the parts of the prefrontal cortex are known to be involved in many high-level cognitive functions, including executive control, action selection, multitasking, social cognition, or general intelligence. These disparate roles have been parsimoniously explained by different concepts, including the conjoint consideration of internal subtasks, branching and reallocation of attention, or balancing between selfgenerated and environmental information. Yet, there may be no common denominator for all functional involvements of the PFC (Wood and Grafman, 2003; Ramnani and Owen, 2004; Amodio and Frith, 2006; Burgess et al., 2006; Koechlin and Hyafil, 2007; Forbes and Grafman, 2010; O'Reilly, 2010).

In contrast, activity changes in medial aspects of the prefrontal cortex (mPFC) were frequently related to social cognition, defined as information processing related to human individuals as opposed to the physical world. Examples of such functional involvements include processing affective information (Phan et al., 2002), forming social judgments (Freeman et al., 2010; Bzdok et al., 2012b), attributing beliefs (den Ouden et al., 2005), retrieving social semantic knowledge (Contreras et al., 2012), and encountering unstable social hierarchies (Zink et al., 2008). In fact, Mitchell (2009) noted that the core domains of social psychology converge exclusively in the mPFC, rendering this scientific field naturally coherent rather than an arbitrary outcome of historical evolution. In social neuroscience, most propositions for functional specialization of the mPFC relied on the distinction between a ventral and a dorsal functional compartment. More specifically, ventral versus dorsal mPFC regions (vmPFC/dmPFC) have been variously proposed to be functionally dissociable according to emotional versus cognitive, automatic versus controlled, implicit versus explicit, outcomeoriented versus goal-oriented, or self-relevant versus otherrelevant social cognition (Amodio and Frith, 2006; Mitchell et al., 2006; Shamay-Tsoory et al., 2006; Lieberman, 2007; Olsson and Ochsner, 2008; Van Overwalle, 2009; Forbes and Grafman, 2010). The diversity of proposed functional dissociations between the vmPFC and dmPFC illustrates the current lack of consensus.

In the current study, we therefore quantitatively examined the functional organization of the mPFC along its ventrodorsal axis. First, the analysis was based on two seed regions in the vmPFC and dmPFC, respectively. These regions corresponded to locations showing significant convergence of perspective-taking tasks in a recent coordinate-based meta-analysis (Bzdok et al., 2012c). As perspective-taking is probably a uniquely human capacity (Premack and Woodruff, 1978; Tomasello et al., 2003), these two clusters of underlying convergent activity are an excellent proxy for the different functional compartments of the mPFC in human social cognition in general. Second, we delineated brain-wide connectivity of each seed according to two complementary measures of functional connectivity, task-dependent meta-analytic connectivity modeling (MACM, Eickhoff et al., 2011) and taskindependent resting state correlations (RS, Biswal et al., 1995). MACM analysis is based on co-activation patterns across a large number of databased neuroimaging experiments (i.e., brain activity under task constraints). RS analysis, in turn, is based on correlations of slow (*<*0.1 Hz) fluctuations of fMRI signals during rest (i.e., unconstrained brain activity in the absence of an externally purported task). Third, we determined a functional profile for each seed using BrainMap meta-data (Laird et al., 2011) by complementary forward and reverse functional decoding. This approach allowed for a cross-validated connectional and functional segregation of the ventral and dorsal mPFC segregation as involved in social cognition.

# **METHODS**

### **DEFINITION OF THE SEED REGIONS**

We conducted connectivity analyses and functional profiling of two seed regions in the mPFC that were derived from a recent coordinate-based meta-analysis (Bzdok et al., 2012c) using the activation-likelihood estimation (ALE) algorithm (Eickhoff et al., 2009, 2012; Eickhoff and Bzdok, 2012). This meta-analysis quantitatively summarized all neuroimaging experiments related to perspective-taking published until 2010, in all, 68 experiments reporting 724 activation foci (Bzdok et al., 2012c). It included neuroimaging experiments [fMRI and positron emission tomography (PET)] in which participants were required to adopt an intentional stance towards others, that is, predict their thoughts, intentions, and future actions. It excluded neuroimaging experiments using non-whole-brain analyses, pharmacological manipulations, or psychiatrically/neurologically diagnosed individuals. More specifically, the two chosen seed regions represent regions of converging brain activity revealed by the (cluster-level corrected) quantitative meta-analysis of neuroimaging results from various paradigms that prompt perspective-taking. Please note that the meta-analyses on empathy and morality, also reported in that meta-analytic study, did not contribute to our seeds. The previously published meta-analysis on perspective-taking thus yielded two continuous, non-overlapping clusters of convergent brain activity that served as neuroanatomical constraints for the differential localization of higher social processes in the mPFC. Put differently, those seeds reflect, first, two topographically constrained brains areas closely related to social processes and, second, the widely assumed functional segregation in this area in the neuroimaging literature on social cognition (e.g., Mitchell et al., 2006; Shamay-Tsoory et al., 2006; Van Overwalle, 2009). Each cluster's whole-brain connectivity pattern was subsequently delineated by task-dependent meta-analytic connectivity modeling and task-independent resting-state analyses. As the employed meta-analytic seeds naturally have asymmetrical shapes we repeated all analyses after fusion of the original seeds with the sagitally mirrored seeds, which yielded virtually identical results.

## **TASK-DEPENDENT FUNCTIONAL CONNECTIVITY: MACM**

The delineation of whole-brain co-activation maps for each seed was performed based on the BrainMap database (www*.*brainmap*.* org; Fox and Lancaster, 2002; Laird et al., 2011). We constrained our analysis to "normal" fMRI and PET experiments (i.e., no pharmacological interventions, no group comparisons) in healthy participants, which report whole-brain results as coordinates in a standard stereotaxic space. These inclusion criteria yielded ∼6500 eligible experiments at the time of analysis. Note that we considered all eligible BrainMap experiments because any pre-selection based on taxonomic categories would have constituted a strong *a priori* hypothesis about how different tasks etc. involve different brain networks. Yet, it remains elusive how well psychological constructs, such as emotion and cognition, map on regional brain responses (Mesulam, 1998; Poldrack, 2006; Laird et al., 2009a). To reliably determine the co-activation patterns of a given seed, we identified the set of experiments in BrainMap that reported at least one activation focus within that seed. The brain-wide co-activation pattern for each seed was then computed by ALE meta-analysis over (all foci reported in) the experiments that were associated with that particular seed (Turkeltaub et al., 2002; Eickhoff et al., 2009; Laird et al., 2009a). The key idea behind ALE is to treat the foci reported in the associated experiments not as single points, but as centers for 3D Gaussian probability distributions that reflect the spatial uncertainty associated with neuroimaging results. Using the latest ALE implementation (Eickhoff et al., 2009, 2012; Turkeltaub et al., 2012), the spatial extent of those Gaussian probability distributions was based on empirical estimates of between-subject and between-template variance of neuroimaging foci (Eickhoff et al., 2009). For each experiment, the probability distributions of all reported foci were then combined into a modeled activation (MA) map by the recently introduced "non-additive" approach that prevents local summation effects (Turkeltaub et al., 2012). The voxel-wise union across the MA maps of all experiments associated with a particular seed voxel then yielded an ALE score for each voxel of the brain that describes the coactivation probability of that particular location with the current seed voxel.

To establish which regions were significantly co-activated with a particular seed, ALE scores for the MACM analysis of this seed were compared to a null-distribution that reflects a random spatial association between experiments, but regards the within-experiment distribution of foci as fixed (Eickhoff et al., 2009). This random-effects inference assesses above-chance convergence between experiments. The observed ALE scores from the actual meta-analysis of experiments activating within a particular seed were then tested against the ALE scores obtained under this null-distribution yielding a p-value based on the proportion of equal or higher random values (Eickhoff et al., 2012). The resulting p-values were then thresholded at *p <* 0*.*05 with cluster-level family-wise error correction for multiple comparisons (clusterforming threshold at voxel-level: *p <* 0*.*001).

Differences in co-activation patterns between the seeds were assessed by first performing MACM separately on the experiments associated with either seed and computing the voxel-wise difference between the ensuing ALE maps (Eickhoff et al., 2011). All experiments contributing to either analysis were then pooled and randomly divided into two groups of the same size as the two original sets of experiments. That is, if 100 experiments in BrainMap featured activation in seed A and 75 featured activation in seed B, the resulting pool of (175) experiments would be randomly divided into a group of 100 and a group of 75 experiments. ALE-scores for these two randomly assembled groups were calculated and the difference between these ALE-scores was recorded for each voxel in the brain. Repeating this process 10,000 times yielded an empirical null-distribution for the differences in ALE-scores between the MACM analyses of the two seeds. The observed difference in ALE scores was then tested against this null-distribution yielding a p-value for the difference at each voxel based on the proportion of equal or higher random differences. The resulting non-parametric p-values were thresholded at *p >* 0*.*95 and inclusively masked by the respective main effects, i.e., the already thresholded effects from the MACM analysis of the particular seed, to focus inference on regions reliably co-activating with that seed.

#### **TASK-INDEPENDENT FUNCTIONAL CONNECTIVITY: RS CORRELATIONS**

Next, seed-wise whole-brain connectivity was assessed using resting-state correlations as an independent modality of functional connectivity. This analysis was based on RS fMRI data from 139 healthy volunteers (56 female, mean age 42.3 years) without any record of neurological or psychiatric disorders. This dataset was obtained through the 1000 Functional Connectomes Project as part of the NKI/Rockland sample (http://fcon\_1000.projects. nitrc.org/indi/pro/nki.html). Participants were instructed to keep their eyes closed and just let their mind wander without thinking of anything in particular but not to fall asleep. For each participant, 260 RS echo-planar imaging (EPI) volumes were acquired on a Siemens TimTrio 3T scanner using blood-oxygen-leveldependent (BOLD) contrast [gradient-echo EPI pulse sequence, TR = 2.5 s, TE = 30 ms, flip angle = 80◦, in-plane resolution = <sup>3</sup>*.*<sup>0</sup> <sup>×</sup> <sup>3</sup>*.*0 mm2, 38 axial slices (3.0 mm thickness) covering the entire brain]. The first four scans served as dummy images allowing for magnetic field saturation and were discarded prior to further processing using SPM8 (www*.*fil*.*ion*.*ucl*.*ac*.*uk/spm). The EPI images were first corrected for head movement by affine registration using a two-pass procedure. The mean EPI image for each participant was then spatially normalized to the MNI single-subject template (Holmes et al., 1998) using the 'unified segmentation' approach (Ashburner and Friston, 2005) and the ensuing deformation was applied to the individual EPI volumes. Finally, images were smoothed by a 5-mm FWHM Gaussian kernel to improve signal-to-noise ratio and compensate for residual anatomical variations.

The time-series data of each individual seed voxel were processed as follows (zu Eulenburg et al., 2012; Satterthwaite et al., 2013): In order to reduce spurious correlations, variance that could be explained by the following nuisance variables was removed: (1) The six motion parameters derived from the image realignment, (2) the first derivative of the realignment parameters, and (3) mean gray-matter, white-matter, and cerebrospinal fluid signal per time-point as obtained by averaging across voxels attributed to the respective tissue class in the SPM eight segmentation. All of these nuisance variables entered the model as firstand second-order terms (Jakobs et al., 2012; Reetz et al., 2012; Satterthwaite et al., 2013). Data were then band-pass filtered preserving frequencies between 0.01 and 0.08 Hz since meaningful resting-state correlations will predominantly be found in these frequencies given that the BOLD response acts as a low-pass filter (Biswal et al., 1995; Fox and Raichle, 2007).

According to this procedure, time courses were extracted for all voxels of a given seed of the individual participant and the time course of the entire seed was then expressed as the first eigenvariate of its voxels' time courses. Pearson correlation coefficients between the time series of the seeds and all other gray-matter voxels in the brain were computed to quantify RS connectivity. These voxel-wise correlation coefficients were then transformed into Fisher's Z-scores and tested for consistent deviation from zero across participants in a random-effects analysis. In particular, the Fisher's Z transformed whole-brain connectivity maps of all seeds were included in an ANOVA accounting for non-sphericity in the data originating from the fact that the different seeds represented correlated measures within each subject with unequal variance between seeds and subjects. Appropriate linear contrasts were then applied to test for regions significantly connected to the seed in the ventral and dorsal mPFC, respectively. The results of this random-effects difference analysis were cluster-level thresholded at *p <* 0*.*05 (cluster-forming threshold at voxel-level: *p <* 0*.*001), analogous to the MACM-based difference analysis.

### **CONJUNCTION AND DIFFERENCE ANALYSES ACROSS BOTH CONNECTIVITY MODALITIES**

To identify brain areas showing convergent task-dependent and task-independent functional connectivity with an individual seed, we performed a conjunction analysis across the MACM- and RS-derived (cluster-level corrected) connectivity maps using the strict minimum statistics (Nichols et al., 2005; Jakobs et al., 2012). Thus, surviving voxels were functionally associated with a given seed in both task-constrained ("focused") and taskunconstrained ("resting") brain states.

The main focus was, however, on connectivity differences between the vmPFC and dmPFC seeds. To this aim, we identified regions with significantly stronger coupling with either seed across task-dependent and task-independent functional connectivity. That is, we computed the conjunction (across both connectivity modalities) of the contrasts (between seeds) to determine regions that were more strongly connected to the ventral or dorsal seed across two disparate brain states (Cieslik et al., 2012; Reetz et al., 2012; Rottschy et al., 2012).

## **FUNCTIONAL PROFILING OF THE SEEDS**

The functional characterization of the two mPFC seeds was based on the BrainMap meta-data that describe each neuroimaging experiment included in the database. Behavioral domains code the mental processes isolated by the statistical contrasts (Fox et al., 2005) and comprise the main categories cognition, action, perception, emotion, and interoception, as well as their related subcategories. Paradigm classes categorize the specific task employed (Turner and Laird, 2012; for the complete BrainMap taxonomy, see http://brainmap*.*org/scribe/).

*Forward inference* on the functional characterization then tests the probability of observing activity in a brain region given knowledge of the psychological process, whereas *reverse inference* tests the probability of a psychological process being present given knowledge of activation in a particular brain region (Poldrack, 2006; Yarkoni et al., 2011). In the forward inference approach, a cluster's functional profile was determined by identifying taxonomic labels for which the probability of finding activation in the respective cluster was significantly higher than the overall chance (across the entire database) of finding activation in that particular cluster. Significance was established using a binomial test (*p <* 0*.*001; Eickhoff et al., 2011). In the reverse inference approach, a cluster's functional profile was determined by identifying the most likely behavioral domains and paradigm classes given activation in a particular cluster. Significance was then assessed by means of a chi-square test (*p <* 0*.*001). Base rates for activations in the respective clusters as well as base rates for tasks were taken into account using the Bayesian formulation for deriving P(Task|Activation) based on P(Activation|Task) as well as P(Task) and P(Activation). In sum, forward inference assesses the probability of activation given a psychological term, while reverse inference assesses the probability of a psychological term given activation (Cieslik et al., 2012; Reetz et al., 2012; Rottschy et al., 2012; Kellermann et al., 2013).

The contrast analyses between the two seeds' functional profiles, in turn, were constrained to those experiments in BrainMap activating either seed. That is, the task associations of experiments in this composite pool were quantified in comparison between the respective seeds and thresholded at *p <* 0*.*05 (false-discovery-rate corrected for multiple comparisons). Forward inference here compared the activation probabilities between the two seeds given a particular psychological term, while reverse inference compared the probabilities of a particular psychological term being present given activation in one or the other seed. Please note that the contrast analysis results were masked with the respective individual functional decoding results of either seed. Put differently, a psychological term can only be significantly more associated with a seeds, if it was also determined significant in the main effect of functional decoding of that seed. Finally, conjunction analyses across the two seeds' functional profiles tested for significant associations of each particular psychological term with both seeds.

Notably, this approach aims at relating defined psychological tasks to the examined brain regions instead of claiming "a unique role" of a brain region for any psychological task (Mesulam, 1998; Poldrack, 2006; Yarkoni et al., 2011). Put differently, an association of task X to brain region Y obtained in these analyses does not necessarily imply that neural activity in region Y *is limited to* task X.

## **RESULTS**

#### **FUNCTIONAL CONNECTIVITY: INDIVIDUAL ANALYSES OF SEEDS**

We first determined each seed's (**Figure 1**) functional connectivity separately by means of both task-dependent MACM and task-independent RS analyses (**Figure 2** and **Tables 1**, **2**). MACM analysis of the vmPFC seed yielded the bilateral vmPFC and dmPFC extending into the anterior cingulate cortex (ACC), amygdala/hippocampus (AM/HC), posterior cingulate cortex/retrosplenial cortex (PCC/RSC), as well as the left nucleus accumbens (NAc), temporo-parietal junction (TPJ), superior frontal gyrus, and posterior operculum (pOP). RS analysis of the vmPFC seed yielded the bilateral vmPFC and dmPFC extending into the ACC, AM, HC, NAc, posterior mid-cingulate cortex (pMCC), RSC/PCC, precuneus (Prec), TPJ, middle temporal gyrus (MTG), temporal pole (TP), precentral gyrus (PreG), pOP, and cerebellum (Cer, not depicted) as well as the right postcentral gyrus (PoG). MACM analysis of the dmPFC seed, in turn,

**FIGURE 1 | Location of the seed regions.** Seeds were drawn from an earlier coordinate-based neuroimaging meta-analysis on perspective-taking, which yielded two clusters of convergent brain activity in the ventral (beige) and dorsal (green) medial prefrontal cortex (Bzdok et al., 2012c). The centers of mass of the vmPFC and dmPFC seed are −4/52/−2 and −6/56/30, respectively. These two seeds represent a functional-structural segregation in the medial prefrontal cortex related to higher social-cognitive processing and provided the basis for the present quantitative analyses. The seeds were rendered into a T1-weighted MNI single subject template using mango (multi-image analysis GUI; http://ric.uthscsa.edu/mango/).

**FIGURE 2 | Functional connectivity of the vmPFC and dmPFC seeds.** Connectivity patterns of each seed as individually determined using meta-analytic connectivity modeling (MACM) and resting-state (RS) analyses. The color bars on the bottom represent *Z*-values. All results survived a cluster-corrected threshold of *p <* 0*.*05. Please refer to **Tables 1**, **2** for peak coordinates. All images were rendered using Caret (computer assisted reconstruction and editing toolkit; http://brainvis.wustl.edu/wiki/index.php/Caret: About). Cortical sheet inflation enhances visual intuitiveness and alleviates activation burying in sulci.

yielded the bilateral vmPFC and dmPFC extending into the ACC, AM/HC, inferior frontal gyrus (IFG), PCC/RSC, TPJ, and TP, as well as the left anterior insula (AI) and MTG. RS analysis of the dmPFC seed yielded the bilateral vmPFC and dmPFC extending into the ACC, AM, HC, IFG, pMCC, PCC/RSC, Prec, TPJ, MTG, TP, PreG, PoG, pOP, and Cer (not depicted).

## **FUNCTIONAL CONNECTIVITY: DIFFERENCE ANALYSES BETWEEN SEEDS**

To subsequently determine which brain areas are more strongly coupled with one seed than the other seed, we computed MACM and RS connectivity differences between both seeds (**Figure 3**). In MACM analyses, the brain areas more strongly coupled with the vmPFC than dmPFC comprised the bilateral vmPFC extending into the ACC, HC (extending into the AM on the right), PCC, and RSC, as well as the left NAc and pOP. In RS analyses, the brain areas more strongly coupled with the vmPFC than dmPFC comprised the bilateral vmPFC, HC, ACC, pMCC, PCC, RSC, Prec, NAc, AI, midbrain/pons, thalamus, visual cortex, posterior lateral parietal cortex, and Cer (not depicted). In MACM analyses, the brain areas more strongly coupled with the dmPFC than vmPFC, in turn, comprised the bilateral PCC, IFG, TPJ, and TP, as well as the left AM and MTG. In RS analyses, the brain areas more strongly coupled with the dmPFC than vmPFC comprised the bilateral orbitofrontal cortex, IFG, MTG, TPJ, TP, PreG, PoG, and Cer (not depicted).

## **FUNCTIONAL CONNECTIVITY: CROSS-VALIDATION BY CONJUNCTION ANALYSES**

The main goal of our study was the functional connectivity of each seed that is consistent across both types of connectivity analysis (i.e., MACM and RS). Convergence of both approaches should reveal connectivity that is consistently observed across two different states of brain function, that is, during specific task performance (MACM) and in the absence of an externally structured task (RS). To thus test for brain areas congruently connected to either seed across both types of connectivity, we computed the conjunction across the respective MACM and RS analyses (**Figure 2** and **Tables 1**, **2**). These conjunction analyses of each seed revealed the same set of brain areas as the respective MACM analysis, except for absent vmPFC connectivity to the operculum.

To test for brain areas more strongly coupled with either seed across MACM and RS analyses, we computed the conjunction across the respective MACM- and RS-based difference analyses (**Figure 4**, **Table 3**). Across MACM and RS, brain areas congruently more strongly coupled with the vmPFC than dmPFC comprised the bilateral vmPFC extending into the ACC, HC, PCC, and RSC, as well as the left NAc. Across MACM and RS, brain areas congruently more strongly coupled with the dmPFC

#### **Table 1 | Functional connectivity of the vmPFC seed.**

#### **Table 2 | Functional connectivity of the dmPFC seed.**



*Table shows coordinates derived from respective cluster peaks (x, y, z) and Z-scores (Z).*

*(Continued)*

#### Bzdok et al. Segregating medial prefrontal social processing

#### **Table 2 | Continued**


*Table shows coordinates derived from respective cluster peaks (x, y, z) and Z-scores (Z).*

than vmPFC comprised the bilateral dmPFC, IFG, and TPJ, as well as the left MTG.

Finally, the brain areas congruently coupled with the vmPFC and dmPFC across both MACM and RS analyses comprised the bilateral vmPFC, frontal pole, AM/HC, and PCC/RSC, as well as the left dmPFC and TPJ.

### **FUNCTIONAL PROFILING OF THE SEEDS**

After the characterization using connectivity analyses, we also conducted a functional characterization of the vmPFC and dmPFC seeds by determining their significant associations with BrainMap taxonomic categories (**Figure 5**). For robustness, we focused on taxonomic associations that are significant in both the forward and reverse inference analysis. Forward inference derives brain activity from a psychological term, whereas reverse inference derives a psychological term from brain activity (see Methods section). Accordingly, activity increases in the vmPFC were consistently associated with tasks related to general cognition, social cognition, as well as emotion and reward processing. Note that BrainMap experiments are labeled as related to general cognition mostly if they do not fit into any of the more specific categories. Activity increases in the dmPFC were consistently associated with tasks related to social cognition, theory of mind (i.e., perspective-taking), episodic memory retrieval, as well as processing emotion, also when derived from faces. Note that BrainMap experiments labeled as related to "Episodic Recall" are very likely to be also labeled as "Cognition.Memory.Explicit" rendering these two taxonomic subcategories highly inter-related. When quantifying the taxonomic associations of the seeds relative to each other, the vmPFC (versus dmPFC) was more consistently associated with reward processing and general cognition, while the dmPFC (versus vmPFC) was more consistently associated with (episodic) memory retrieval and theory-of-mind processing. Finally, the taxonomic associations consistent across both vmPFC and dmPFC comprised tasks related to social, emotional, and facial (i.e., "Subjective Emotional Picture Discrimination") processing.

## **DISCUSSION**

We examined the widely assumed but not directly tested ventrodorsal differentiation of the mPFC in social cognition. This test of segregation was based on a ventral and dorsal mPFC region that are both consistently related to perspective-taking as a prototypical instance of social cognition. The seeds were analyzed using two ways of functional connectivity analyses by independently delineating task-related meta-analytic connectivity modeling (MACM, Eickhoff et al., 2011) and task-unrelated resting-state correlations (RS, Biswal et al., 1995). Additionally, it was tested whether the seeds were differentially associated with psychological terms from BrainMap meta-data using forward and reverse inference. In both MACM and RS analyses, the vmPFC was more strongly connected with the nucleus accumbens (NAc), hippocampus (HC), posterior cingulate cortex (PCC), and retrosplenial cortex (RSC), while the dmPFC was more strongly connected with the inferior frontal gyrus (IFG), temporo-parietal junction (TPJ), and middle temporal gyrus (MTG). In both functional decoding analyses, the vmPFC was selectively associated with reward related tasks, while the dmPFC was selectively associated with perspective-taking and episodic memory retrieval tasks. Importantly, both vmPFC and dmPFC were functionally associated with social, emotional, and facial processing. In sum, the vmPFC was thus more closely connected to limbic and reward-related medial brain areas as well as functionally associated with processing approach- and avoidance-relevant stimuli. In contrast, the dmPFC was more connected to higher associative cortical areas as well as functionally associated with processing mental states and episodic memory.

## **CONNECTIONAL EVIDENCE FOR THE SEGREGATION BETWEEN THE vmPFC AND dmPFC**

Our convergent connectivity results across MACM and RS analyses derived from the vmPFC and dmPFC seeds agree well with many earlier findings in humans and monkeys. Importantly, the vmPFC and dmPFC have been found to be extensively interconnected in axonal tracing studies in monkeys (Barbas et al., 1999; Saleem et al., 2008), consistent with our results. In the following, we will compare the present connectivity differences between the vmPFC and dmPFC with earlier findings using other connectivity measures in humans and monkeys.

The vmPFC, on the one hand, was more strongly connected to the NAc, HC, PCC, and RSC across two different types of functional connectivity analysis in the present study. Indeed, the vmPFC, but not dmPFC, has been observed to have monosynaptical connections with the ventral striatum (VS, which anatomically includes the NAc) in axonal tracing studies in monkeys (Haber et al., 1995; Ferry et al., 2000). This is consistent with our results and probabilistic diffusion tensor imaging (DTI) tractography in humans and monkeys (Croxson et al., 2005) that quantified the VS to be substantially more likely connected to the vmPFC than dmPFC in both species. This DTI study further estimated the vmPFC to be only slightly more connected to the amygdala (AM) than the dmPFC in monkeys and humans (cf. Bzdok et al., 2012a), in line with the present AM connectivity to both vmPFC and dmPFC. Importantly, roughly balanced connectivity to the AM challenges the frequently proposed vmPFC-dmPFC distinction as emotional versus cognitive. Although monkey tracing studies indicated that the entire medial wall of the prefrontal cortex has amygdalar and cingulate connections, the most ventral part of the mPFC received strongest connections from most

limbic areas, including the HC (Carmichael and Price, 1995). This concurs with our results and a RS connectivity analysis of the human HC showing more correlation with the vmPFC than dmPFC (Vincent et al., 2006). Additionally, fibers from the medial temporal lobe (including the AM and HC) entered the mostly ventral medial partial cortex, including the RSC, as observed using DTI tractography in humans (Greicius et al., 2009). Our results are in line with monkey tracing studies showing that mostly the vmPFC but also dmPFC are directly connected to the PCC (Carmichael and Price, 1995) and RSC (Vann et al., 2009). Conversely, the PCC and RSC (but not the more dorsocaudal precuneus) were mostly connected to limbic regions and the vmPFC in a comparative RS study in monkeys and humans (Margulies et al., 2009). Concluding from previous and present connectivity findings, *the vmPFC is preferentially connected with limbic and reward-related medial brain areas.*

The dmPFC, on the other hand, was more strongly connected to the TPJ, MTG, and IFG across two different types of functional connectivity analysis in the present study. Using DTI tractography in humans the vmPFC and dmPFC have been observed to be connected to the TPJ, which in turn was connected to the MTG (Caspers et al., 2011). Although we also found convergent functional connectivity of the vmPFC and especially dmPFC to the TPJ, monosynaptical connections from the anterior prefrontal cortex to the TPJ *might* be absent in monkeys (for discussion, see Caspers et al., 2011). Existence of mPFC-TPJ connectivity in humans is supported by the present results, while our methodological approach cannot distinguish mono- and polysynaptical connections. Our results therefore cannot contribute to the more general question whether direct mPFC-TPJ connections exist in humans but not monkeys. The TPJ and IFG, both relatively more connected to the dmPFC in our study, were also reported to be connected in an axonal tracing study in monkeys (Petrides and Pandya, 2009) and in a DTI study in humans (Frey et al., 2008). Both the vmPFC and dmPFC are further known to have direct connections with the IFG and MTG based on monkey tracing data (Yeterian et al., 2012). In contrast, those two target areas were more strongly connected to the dmPFC in our functional connectivity analyses. Thus, axonal connections between the vmPFC and the IFG and MTG presumably existing in humans

**FIGURE 4 | Difference and conjunction analyses based on congruent functional connectivity of the vmPFC and dmPFC seeds.** Depicts sagittal and coronal brain slices of areas consistently more strongly coupled (left and middle column) with either seed or congruently coupled with both seeds (right column) across meta-analytic connectivity modeling (MACM) and

resting-state (RS) analyses. Please refer to **Table 3** for activation coordinates. All slices were created using mango (multi-image analysis GUI; http://ric.uthscsa.edu/mango/) on a T1-weighted MNI single subject template. Coordinates in MNI space. *<*/*>*, difference analysis; &, conjunction analysis; R, right; L, left.



*Table shows coordinates derived from respective cluster peaks (x, y, z) and Z-scores (Z). < and > denote difference analyses, while & denotes conjunction analysis.*

might be less important for social-cognitive processing than those of the dmPFC. Similarly, although DTI tractography in humans (Greicius et al., 2009) and axonal tracing in monkeys (Cavada and Goldman-Rakic, 1989) have identified fiber bundles connecting the dmPFC with the more dorsal and posterior medial parietal cortex (precuneus), this was not reflected by our functional connectivity results. Concluding from previous and present connectivity findings, *the dmPFC is preferentially connected with high association and heteromodal cortical areas of the lateral frontal, temporal, and parietal lobe.* More globally, most of the present *functional connectivity* findings of the human vmPFC and dmPFC concur very well with knowledge describing *structural connectivity* in the monkey and human brain. However, our results also show that known axonal connections between the mPFC and other parts of the brain are not always reflected in functional connectivity analyses.

#### **INTEGRATIVE SEGREGATION BETWEEN THE vmPFC AND dmPFC**

After discussing the connectivity differences between the vmPFC and dmPFC, we will now discuss the previously proposed functional properties of their respective connectivity targets (cf. Fuster, 2001). The vmPFC was more connected to the NAc, HC, PCC, and RSC. The NAc is thought to be linked to reward mechanisms that may not only modulate motivated behavior towards basic survival needs, such as food and sex, but also towards salient social cues (cf. Kampe et al., 2001; Cardinal et al., 2002; Walter et al., 2005; Schilbach et al., 2010). Neuroimaging research indeed ascribed complex reward functions to the NAc, such as the evaluation of reward expectancy in social, monetary, or drug rewards (Schultz et al., 1997; Kampe et al., 2001; Rademacher et al., 2010; Bzdok et al., 2011). The HC, in turn, is well known to be involved in memory and spatial navigation in animals and humans (von Bechterew, 1900; Scoville and Milner, 1957; O'Keefe and Dostrovsky, 1971; Maguire et al., 2000). As to the PCC and RSC, electrophysiological research in animals implicated the PCC in strategic selection (Pearson et al., 2009), risk assessment (McCoy and Platt, 2005), and outcome-contingent behavioral modulation (Hayden et al., 2008), while the RSC was implicated in navigation and approach-avoidance behavior (Vann et al., 2009). Considering only the previously reported functional properties of the here more strongly connected nodes, the vmPFC can be assumed to integrate a subnetwork (i.e., the brain areas relatively more connect to the vmPFC, excluding the vmPFC seed itself) modulating online approach-avoidance behavior by memory-informed reward and risk estimation of self-relevant environmental stimuli.

In contrast, the dmPFC was more connected to the IFG, TPJ, and MTG. As these subnetwork nodes (i.e., the brain areas relatively more connected to the dmPFC, excluding the dmPFC seed itself) are highly associative and heteromodal, there is less clarity and agreement about their discrete functional contributions. As a side note, the mere difference in the association level between the vmPFC's and dmPFC's subnetworks already indicates functional segregation (Mesulam, 1998). Moreover, the entire set of dmPFC-linked regions is well known to concomitantly increase and decrease metabolic activity as a cohesive unit, as lateral components of the so-called "default mode network" (Gusnard et al., 2001; Laird et al., 2009b; Spreng et al., 2009; Mar, 2011; Bzdok et al., 2012c; Schilbach et al., 2012). In fact, it is interesting to note that the vmPFC is more strongly connected to medial components of the default mode network (i.e., HC, PCC, RSC), whereas the dmPFC is more strongly connected to its lateral components (i.e., IFG, TPJ, and MTG). This dmPFC subnetwork was repeatedly related to self-focused reflection (Andrews-Hanna et al., 2010), contemplation of others' (Mar, 2011) and one's own (Lombardo et al., 2009) mental states, mental navigation of the body in space (Maguire et al., 1997), semantic processing (Binder et al., 2009), as well as scene construction processes when envisioning past, fictitious, and future events (Hassabis et al., 2007; Spreng et al., 2009; Bzdok et al., 2013). Interestingly, the neuroimaging studies related to processing semantic information (Binder et al., 2009), autobiographical (Spreng et al., 2009) and fictitious (Hassabis et al., 2007) events observed neural activity increases in both the vmPFC and dmPFC, although the respective neural networks resemble much more the dmPFC (rather than vmPFC) subnetwork. The conjunction of previous and present findings suggests that the dmPFC integrates a network involved in self- or otherrelated, largely sensory-independent, highly abstract (hence, less

tangible) processes across time, space, and content domains. Importantly, the previously proposed vmPFC-dmPFC distinction as outcome-oriented versus goal-oriented is challenged by our results that support outcome-oriented vmPFC processing but not specifically goal-oriented dmPFC processing. It is also important to note that both the vmPFC and dmPFC are closely related to memory retrieval as indicated by converging functional connectivity (across MACM and RS) to the HC. However, the memory-retrieved information appears to be bound with less complex neural processes in the vmPFC versus dmPFC as indicated by functional association with, for instance, less complex reward processes versus more complex perspective-taking processes.

Additionally, the here identified subnetworks belonging to the vmPFC and dmPFC corroborate an earlier hierarchical clustering analysis based on an fMRI study (Andrews-Hanna et al., 2010). In particular, seed regions were derived from comparing future versus present self-related thinking in bidirectional fMRI contrasts. Subsequent resting-state analyses of these seed regions allowed clustering into a vmPFC-associated subnetwork, including the HC and PCC/RSC, and a dmPFCassociated subnetwork, including the TPJ and MTG. The fMRI data then related, respectively the vmPFC and dmPFC subnetworks to thinking about present and future self, in line with our functional decoding results. Put differently, the vmPFC might be more closely associated with orchestrating adapted behavior by bottom-up-driven processing of "what matters now", which might be top-down modulated by more dmPFC subserved higher reflective and hypothetical processing.

#### **MORPHOLOGICAL EVIDENCE FOR THE SEGREGATION BETWEEN THE vmPFC AND dmPFC**

It may be instructive to acknowledge the relationship between the present findings on social cognition in mPFC subregions and the recently increasing evidence for the "social brain" that might have coevolved with the complexity of social relationships (Jolly, 1966; Humphrey, 1978; Byrne and Whiten, 1988; Dunbar, 1998; Dunbar and Shultz, 2007). Most importantly, independent whole-brain analyses from *structural* neuroimaging studies related the gray-matter volume (GMV) of the vmPFC to indices of social competence and social network complexity in both humans and monkeys (Lebreton et al., 2009; Powell et al., 2010; Lewis et al., 2011; Sallet et al., 2011). To our knowledge, none of these four correlations have been found yet for the dmPFC. Consequently, vmPFC, rather than dmPFC, anatomy appears to predict an individual's social behavioral dispositions and social network properties, although we found both regions to be congruently associated with social, emotional, and facial processing.

Such brain-behavior correlations in humans were also shown for the brain areas preferentially connected to the vmPFC or dmPFC in the present analysis. As to the vmPFC subnetwork, the GMV of the vmPFC and VS correlated with indices of social reward attitudes and behavior (Lebreton et al., 2009), concurring with vmPFC's relation to the NAc and reward-related tasks. Additionally, the GMV of the entorhinal cortex (connectionally and functionally closely coupled with the HC) correlated with social network size (Kanai et al., 2012), concurring with vmPFC's connectivity to the HC. Further, vmPFC and PCC/Prec GMV correlated with social network size (Lewis et al., 2011), concurring with vmPFC's stronger connectivity to the PCC. As to the dmPFC subnetwork, the GMV of the TPJ and MTG correlated with social network size (Kanai et al., 2012), while the GMV of the TPJ and IFG correlated with perspective-taking competence (Lewis et al., 2011). Moreover, the GMV of the amygdala, connected to both vmPFC and dmPFC, correlated negatively with social phobia (Irle et al., 2010) and positively with social network size (Bickart et al., 2010).

The conjunction of these recent brain-behavior correlations and the present results allow several conclusions. With respect to our seeds, inter-individual differences in social skills or social networks were most often related to morphological differences in the human and monkey vmPFC, in stark contrast to the dmPFC. With respect to the seeds' subnetworks, the reported brain-behavior correlations were roughly equally related to the more vmPFC or dmPFC connected brain areas. With respect to the type of social variable, morphological differences related to either social skills or networks do not seem to be preferentially associated with the more vmPFC or dmPFC connected brain areas.

The conclusions prompt the hypothesis that the dmPFC subserves a domain-independent neural process important for, but not specific to, social cognition. Indeed, the present results support the dmPFC's possible involvement in domain-overarching computational mechanisms given its connections to highly associative brain areas and functionally relation to different complex psychological processes. Although vmPFC and dmPFC were associated with social, emotional, and facial processing, the dmPFC probably processes these types of information on a higher level of abstraction.

## **NEUROPSYCHOLOGICAL EVIDENCE FOR THE SEGREGATION BETWEEN THE vmPFC AND dmPFC**

The conclusions derived from our findings are corroborated by brain lesion data. Consistent with the functional association of the vmPFC with reward processing as well as with a role in predominantly *self-related* behavior guided by stimulus evaluation and reward-learning, a voxel-based lesion-symptom mapping (VLSM) study in 344 neurological patients demonstrated functional-anatomical specificity of the vmPFC for value-based decision-making (Gläscher et al., 2012). However, vmPFC damage in humans also impairs an array of predominantly *otherrelated* socio-emotional processes. More specifically, consistent with vmPFC's connectivity to both the limbic system and the dmPFC, vmPFC lesions appear to impair the *integration* of (other-related) higher social, basic emotional, and facial processes, rather than any of these three classes of neural processes per se (Bzdok et al., 2012b). This is indicated by (1) disrupted emotion recognition from faces (Hornak et al., 1996) despite intact face recognition (Shamay-Tsoory et al., 2005; Monte et al., 2012), (2) sociopathic behavior in every-day life (Blair and Cipolotti, 2000) despite intact abstract reflection of social phenomena (Saver and Damasio, 1991; Damasio, 1996; Young et al., 2010), (3) disrupted affective but not cognitive perspective-taking (Stone et al., 1998; Stuss et al., 2001; Shamay-Tsoory et al., 2006; Shamay-Tsoory and Aharon-Peretz, 2007), (4) disrupted perspective-taking-based empathy despite intact simpler affective empathy (Shamay-Tsoory et al., 2009), and (5) reduced emotional impact on moral judgments (Koenigs et al., 2007; Young et al., 2010).

Put differently, vmPFC lesion might alter the subset of abstract social processes that require vmPFC-mediated relay of emotional limbic information to the dmPFC, consistent with our connectional and functional results. Indeed, faux detection (i.e., abstract social processing involving emotion processing) is impaired after damage to either the amygdalae (Stone et al., 2003) or the vmPFC (Gregory et al., 2002). The conjunction of previous lesion reports and present results therefore suggests that the vmPFC interweaves more emotional processes (mainly subserved by the limbic system) and more *ambiguous* social thought (probably subserved by the dmPFC) to shape self- and otherrelated behavioral responses to sensory events in social cognition (Shamay-Tsoory and Aharon-Peretz, 2007; Bzdok et al., 2012b).

Juxtaposing the effects of vmPFC and dmPFC lesions in humans is impeded by the scarcity of circumscribed dmPFC lesions (cf. Mochizuki and Saito, 1990; Duffy and Campbell, 1994; Wilson et al., 2010). Although quite heterogeneous, the few available dmPFC-linked lesion findings consolidate the here derived segregation within the mPFC as a function of reliance on bottom-up versus top-down processing pathways. First, the dmPFC subnetwork was normally recruited in congenitally blind individuals engaged in perspective-taking (Bedny et al., 2009). Therefore, complete lack of visual input does not appear to alter functioning of this high-level area, contrarily to low-level visual cortices. Second, a VLSM study on disturbed sleep (i.e., a state of mind independent of sensory stimulation but dependent on internally generated information) *exclusively* identified the dmPFC (Koenigs et al., 2010). Third, another VLSM study exclusively related the IFG and TPJ, both more strongly connected to the dmPFC in our study, to inner speech (Geva et al., 2011). Taken together, in individuals with an intact central nervous system, the vmPFC versus dmPFC are probably involved in predominantly bottom-up versus top-down mediated processing of social information.

## **NEUROIMAGING EVIDENCE FOR THE SEGREGATION BETWEEN THE vmPFC AND dmPFC**

Following the observed functional associations with fear and reward, the vmPFC is likely to process not only external but also visceral stimuli. Indeed, measurements of task-induced brain activity changes in humans confirm our functional decoding results by relating the vmPFC to monitoring others' (Lotze et al., 2007) and one's own (Lane et al., 1997; Phan et al., 2004) emotional responses, that is, other's (external) emotional reactions and one's own (visceral) arousal. Such real or imagined bodily states, believed to be represented in the vmPFC, probably operate as a bioregulatory disposition governing cognition and decision making (Damasio, 1996; Nauta, 1971), in line with the vmPFC's functional association with general cognition and reward processing. An fMRI study, for instance, reported specific vmPFC activity increases during other-initiated joint attention, suggesting representation of the motivational significance of social cues (Schilbach et al., 2010). Consistent with our line of interpretation, vmPFC versus dmPFC activity was moreover shown to reflect actually choice-relevant versus modeled, choice-irrelevant value in a computational fMRI study (Nicolle et al., 2012). The conjunction of previous functional neuroimaging findings and our functional profiling data consolidate the vmPFC's role in processing self- and other-related visceroaffective and motivational information as a guide in ongoing social behavior.

Moreover, the vmPFC and dmPFC were both significantly associated with social, emotional, and facial processing in the present study. This indicates that the vmPFC and dmPFC are not functionally dissociable by selective involvement in social, emotional, or facial processing, although this is frequently proposed. However, the dmPFC, but not vmPFC, was congruently associated with more complex social-cognitive tasks across forward and reverse functional decoding, including perspectivetaking and episodic memory retrieval. While the former imposes an other-focused mind set, the latter inherently entails a selffocused mind set (obviously, one can only recall scenes from one's own personal experience). Quantitative functional profiling of the dmPFC therefore indicates that the dmPFC is involved in both self- and other-oriented processing, analogous to the vmPFC. Importantly, the frequently proposed vmPFCdmPFC distinction as self versus other is challenged by our conclusions.

In particular, consistent with present functional decoding, neural activity in the dmPFC, rather than vmPFC, has been consistently interpreted to underlie inference, representation, and assessment of one's own and others' mental states in functional neuroimaging research (Gusnard et al., 2001; Gallagher and Frith, 2003; Amodio and Frith, 2006; Gilbert et al., 2006; Ochsner, 2008; Van Overwalle, 2009; Bzdok et al., 2012b; Moran et al., 2012). For instance, dmPFC (but not vmPFC) activity was related to the proficiency decline of mental state inference in elderly (Moran et al., 2012), *cognitive regulation* of one's own emotional states (Ochsner et al., 2004b) and *inference* of another person's emotional states (Ochsner et al., 2004a), as well as self-reported (Wagner et al., 2011) and experimentally measured (Zaki et al., 2009) proficiency in emotional state inference. Notably, such selfand other-related conceptualizations cannot be made based on sensory information or general knowledge about the physical world (cf. Premack and Woodruff, 1978; Leslie, 1987; Carruthers, 2009). Thus, mental state inference necessarily relies on the generation of probabilistic internal information. Supported by dmPFC's functional association with episodic memory retrieval, such prima vista non-mnemonic construction processes are likely to be subserved by the neural network underlying retrieval of past and imagination of future scenes as indicated by recent neuroimaging experiments and meta-analyses (Schacter et al., 2007; Spreng et al., 2009; Andrews-Hanna et al., 2010; Rabin et al., 2010; Bzdok et al., 2012c). Constructing such probabilistic scenes is further believed to necessarily drawn on semantic knowledge retrieval (Binder et al., 1999; Bar, 2007; Suddendorf and Corballis, 2007; Carruthers, 2009; Bzdok et al., 2012c). This would be in line with left lateralization of the dmPFC subnetwork typical of semantic processing (Binder et al., 2009). The conjunction of previous functional neuroimaging findings and present neuroinformatic findings congruently characterizes the dmPFC as a "mental sketchpad" (Goldman-Rakic, 1996) potentially implicated in modeling and binding plausible self- and other-related scenarios instructed by semantic concepts in social cognition. Again, such sensory-independent *de novo* generation of meaning representations can only be expected from highly associative, integrative brain areas such as those of the dmPFC subnetwork (Mesulam, 1998), as opposed to the vmPFC subnetwork.

# **CONCLUSION**

Although the human mPFC is neither uniquely nor solely devoted to social cognition, its central role in navigating the interpersonal space is probably one of the most often replicated findings in functional neuroimaging research. However, the strength of cognitive neuroscience comes from investigating an identical phenomenon from various conceptual and methodological perspectives (cf. Feyerabend, 1975). We therefore re-examined the widely assumed ventrodorsal functional segregation of the mPFC in social cognition in a bottom-up approach and integrated the ensuing results with different literatures. As a result of this, we comprehensively characterized *both* the vmPFC and dmPFC as relevant for self- and other-focused as well as social, emotional, and facial processing. More specifically, the vmPFC subserves predominantly non-ambiguous subjective-value-related evaluative processes driven by bottom-up pathways, whereas the dmPFC subserves predominantly ambiguous amodal metacognitive processes driven by top-down pathways. These conclusions amend a number of earlier accounts on the division of labor between ventral and dorsal aspects of the mPFC in social cognition. Ultimately, the integration of external stimulation and internal generation driven processes in the mPFC is a part of what determines social behavior.

## **REFERENCES**


brain using echo-planar MRI. *Magn. Reson. Med.* 34, 537–541.


# **ACKNOWLEDGMENTS**

This study was supported by the National Institute of Mental Health (R01-MH074457, Peter T. Fox, Angela R. Laird, Simon B. Eickhoff), the Helmholtz Initiative on Systems-Biology "The Human Brain Model" (Simon B. Eickhoff), and the German National Academic Foundation (Danilo Bzdok). The authors declare no conflict of interest.

complementary connectivity analyses, and functional decoding. *Neuroimage* (in press).


networks. *J. Comp. Neurol.* 425, 447–470.


averaging. *J. Comput. Assist. Tomogr.* 22, 324–333.


prefrontal cortex increases utilitarian moral judgements. *Nature* 446, 908–911. doi: 10.1038/nature05631


*Neuroimage* 16, 331–348. doi: 10.1006/nimg.2002.1087


Loughead, J., Calkins, M. E., et al. (2013). An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of restingstate functional connectivity data. *Neuroimage* 64, 240–256. doi: 10.1016/j.neuroimage.2012.08.052


Bzdok et al. Segregating medial prefrontal social processing

theory of mind stories. *Soc. Neurosci.* 1(3–4), 149–166. doi: 10.1080/17470910600985589


Meta-analysis of the functional neuroanatomy of single-word reading: method and validation. *Neuroimage* 16, 765–780. doi: 10.1006/nimg.2002.1131


when viewing natural social scenes. *Cereb. Cortex* 21, 2788–2796. doi: 10.1093/cercor/bhr074


*U.S.A.* 106, 11382–11387. doi: 10.1073/pnas.0902666106


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 03 March 2013; paper pending published: 03 April 2013; accepted: 14 May 2013; published online: 29 May 2013.*

*Citation: Bzdok D, Langner R, Schilbach L, Engemann DA, Laird AR, Fox PT and Eickhoff SB (2013) Segregation of the human medial prefrontal cortex in social cognition. Front. Hum. Neurosci. 7:232. doi: 10.3389/fnhum.2013.00232*

*Copyright © 2013 Bzdok, Langner, Schilbach, Engemann, Laird, Fox and Eickhoff. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# The role of medial prefrontal cortex in early social cognition

# *Tobias Grossmann\**

*Early Social Development Group, Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany*

#### *Edited by:*

*Leonie Koban, University of Colorado Boulder, USA*

#### *Reviewed by:*

*Marlene Meyer, Radboud University, Netherlands Danilo Bzdok, Research Center Jülich, Germany Haruhiro Higashida, Kanazawa University Research Center for Child Mental Development, Japan*

#### *\*Correspondence:*

*Tobias Grossmann, Early Social Development Group, Max Planck Institute for Human Cognitive and Brain Sciences, Stephanstrasse 1a, 04103 Leipzig, Germany e-mail: grossman@cbs.mpg.de*

One major function of our brain is to enable us to behave with respect to socially relevant information. Much research on how the adult human brain processes the social world has shown that there is a network of specific brain areas, also called the social brain, preferentially involved during social cognition. Among the specific brain areas involved in the adult social brain, functional activity in prefrontal cortex (PFC), particularly the medial prefrontal cortex (mPFC), is of special importance for human social cognition and behavior. However, from a developmental perspective, it has long been thought that PFC is functionally silent during infancy (first year of life), and until recently, little was known about the role of PFC in the early development of social cognition. I shall present an emerging body of recent neuroimaging studies with infants that provide evidence that mPFC exhibits functional activation much earlier than previously thought, suggesting that the mPFC is involved in social information processing from early in life. This review will highlight work examining infant mPFC function across a range of social contexts. The reviewed findings will illustrate that the human brain is fundamentally adapted to develop within a social context.

**Keywords: infancy, development, prefrontal cortex, social cognition, fNIRS**

# **INTRODUCTION**

Humans possess a number of higher cognitive skills vital for language, reasoning, planning, and complex social behavior. The prefrontal cortex (PFC) can be seen as the neural substrate that underpins much of this higher cognition (Wood and Grafman, 2003). PFC refers to the regions of the cerebral cortex that are anterior to premotor cortex and the supplementary motor area (Zelazo and Müller, 2002). Based on its neuroanatomical connections, the PFC can be broadly divided into two sections: (a) the medial PFC (mPFC) and (b) the lateral PFC (lPFC) (Wood and Grafman, 2003; Fuster, 2008). The mPFC includes the medial portions of Brodmann areas (BA) 9–12, and BA 25, and has reciprocal connections with brain regions that are implicated in emotional processing (amygdala), memory (hippocampus) and higher-order sensory regions (within temporal cortex) (for more detailed information see, Wood and Grafman, 2003; Fuster, 2008). The lPFC includes the lateral portions of Brodmann areas (BA) 9–12, BA 44, 45 and BA 46, and has reciprocal connections with brain regions that are implicated in motor control (basal ganglia, premotor cortex, supplementary motor area), performance monitoring (cingulate cortex) and higherorder sensory processing (within temporal and parietal cortex) (for more detailed information see, Wood and Grafman, 2003; Fuster, 2008).

Critically, the distinction between lPFC and mPFC in neuroanatomical terms maps onto general differences in brain function. Namely, while mPFC is thought to be mainly involved in processing, representing and integrating social and affective information, lPFC is thought to support cognitive control process (Wood and Grafman, 2003; Fuster, 2008). This general functional distinction between mPFC and lPFC can already be seen early in development during infancy (Grossmann, 2013), thus representing a developmentally continuous organization principle of PFC function. As far as brain function is concerned, mPFC has been shown to play a fundamental role in a wide range of social cognitive abilities such as self-reflection, person perception, and theory of mind/mentalizing (Amodio and Frith, 2006). This involvement of mPFC in social cognition and interaction has lead to the notion that mPFC serves as a key region in understanding self and others (Frith and Frith, 2006). Although this is not the focus of this review, it should be noted that apart from its implication in social cognitive functions in adults, mPFC has been shown to be more generally involved in a number of processes related to decision making in adults (e.g., Heekeren et al., 2008). In particular, most recently, a unifying model has been proposed that views mPFC as a region concerned with learning and predicting the likely outcomes of actions (Alexander and Brown, 2011).

Only very little is known concerning the role of the mPFC in the development of social cognition. This is particularly true for the earliest steps of postnatal development, namely during infancy (the first year of life). Addressing the question of whether mPFC plays a role in infant social cognition and if it does, to theorize about what role this might be is the goal of this review. Such a look at early social cognition during infancy through the lenses of social neuroscience is critical because it allows us (a) to understand the nature and developmental origins of mPFC function, and (b) to close a gap between the extensive behavioral work showing rather sophisticated infant social cognitive skills (Spelke and Kinzler, 2007; Woodward, 2009; Baillargeon et al., 2010) and the social neuroscience work with adults studying mature mPFC functioning (Amodio and Frith, 2006; Lieberman, 2006).

That mPFC plays an important role in the development of social cognition is evident in work examining mPFC lesions. For example, there is work comparing early onset (during infancy) and adult onset lesions to mPFC (Anderson et al., 1999). This work shows that, despite typical basic cognitive abilities, patients with mPFC lesions had severely impaired social behavior. More specifically, regardless of when the mPFC lesion had occurred, there are symptoms shared across patients with mPFC damage, including an insensitivity to future consequences of actions, defective autonomic responses to punishment contingencies, and failure to respond to interventions that would change behavior (Anderson et al., 1999). Critically, this study revealed that over and above the shared symptomatology, acquired damage to mPFC during infancy had a much more severe impact on social functioning signified by striking defects concerning social and moral reasoning, leading to a syndrome that closely resembled psychopathy. In this study, it was found that early onset damage to mPFC was related to antisocial behaviors such as stealing, violence against persons and property, severe impairment of social-moral reasoning and verbal generation of responses to social situations. Specifically, in adults with early onset lesions to mPFC, moral reasoning was conducted at a much lower level than expected by their age, such that moral dilemmas were mainly approached from an egocentric perspective characterized by avoiding punishment. Furthermore, early onset damage of mPFC was related to a limited consideration of the emotional implications of one owns behavior for others and much fewer responses generated to resolve interpersonal conflict. This suggests that mPFC plays a critical role in the acquisition of social and moral behaviors already early during ontogeny. It further suggests that in contrast to many other brain regions where damage and especially damage early in ontogeny can be compensated (Thomas and Johnson, 2008), mPFC appears to be less plastic or more vulnerable. This in turn indicates that there might be a sensitive period in development during which mPFC is required to develop and learn socially and morally appropriate behaviors. Even though the study of patients with lesions to the mPFC is of great importance in illuminating mPFC function, patients with circumscribed mPFC lesions acquired during infancy, as reported by Anderson and colleagues (1999), are extremely rare and can hence only provide limited insights into these early stages of developing mPFC function. It is therefore all the more important to employ functional neuroimaging to shed light on the development of mPFC function during infancy if we wish to better understand its role in early social cognition.

Recent advances in applying functional imaging technology to infants, specifically, the advent of using functional near-infrared spectroscopy (fNIRS) has made it possible to study the infant brain at work. fNIRS is an optical imaging method that measures hemodynamic responses from cortical regions, permitting for the localization of brain activation (Lloyd-Fox et al., 2010). Other neuroimaging techniques that are well established in adults are limited in their use with infants because of methodological concerns. For example, functional magnetic resonance imaging (fMRI) requires the participant to remain very still and exposes them to a noisy environment. Although fMRI has been used with infants, this work is restricted to the study of sleeping, sedated or very young infants. The method of fNIRS is better suited for infant research because it can accommodate a good degree of movement from the infants, enabling them to sit upright on their parent's lap and behave relatively freely while watching or listening to certain stimuli. In addition, unlike fMRI, fNIRS systems are portable. Finally, despite its inferior spatial resolution also in terms of obtaining responses from deeper (subcortical) brain structures, fNIRS, like fMRI, measures localized patterns of hemodynamic responses in cortical regions, thus allowing for a comparison of infant fNIRS data with adult fMRI data. In the last decade, there has been a surge of fNIRS studies with infants, including a number of studies that have looked at PFC activation during a wide range of experimental tasks (for review, see Grossmann, 2013). In the following sections, I shall review the available experimental evidence that implicate mPFC in infant social cognition. This review is aimed at providing an overview of the range of social contexts during which infants employ the mPFC. The review of the empirical work is organized according to the two main sensory modalities (audition and vision) in which social stimuli were presented to infants. Following the presentation of the experimental evidence, I will discuss a number of issues that arise from these studies. Finally, based on these findings, I will outline an account of what role mPFC plays in the early development of social cognition during infancy.

## **OVERVIEW OF STUDIES REPORTING mPFC ACTIVATION IN INFANTS**

Newborns enter the world with a number of behavioral biases that allow them to preferentially attend and respond to certain stimuli such as faces and voices (Grossmann and Johnson, 2007), suggesting that infants enter the world endowed with biases that allow them to preferentially engage with the social world. However, while these biases found in newborns may be a vital foundation for the emergence for the development of social cognitive skills, we are only beginning to understand what role prefrontal brain regions play in these early attempts of the developing infant to respond to her environment and organize her perceptual experiences.

## **AUDITION**

The human voice, apart from having obvious functions in linguistic communication, also carries a wealth of socially relevant information such as age, gender, and emotional state (Belin et al., 2004). Newborns have been shown to show significantly increased responses in mPFC to their own mother's voice reading a story in infant-directed speech (IDS) compared to their mothers reading the same story in adult-directed speech (ADS) (Saito et al., 2007). This indicates that newborn infants discriminate between these two forms of speech and dedicate increased mPFC processing resources to IDS, which is of high socio-affective relevance to the infant. In another study, Saito et al. (2007) also showed that mPFC activation can be obtained in response to non-maternal emotional speech. This finding suggests that it is the emotional tone of voice that characterizes positive affect in speech that drives this effect on mPFC in newborns.

Older infants (4–13 months of age) were presented with IDS and ADS sentences spoken by their own mother or a female stranger and prefrontal and temporal cortex responses were measured using fNIRS (Naoi et al., 2012). This study showed that while infants' temporal cortex discriminated between IDS and ADS regardless of speaker, PFC (including mPFC in the left hemisphere) was engaged only when the mother spoke with IDS. Together with the data from newborns presented above, this suggests that mPFC responses undergo change during infancy and become more finely tuned to the primary caregiver's voice. Indeed, in agreement with behavioral work showing that at the age of 7–9 months infants show the strongest preferences for their primary caregivers, prefrontal responses change during infancy such that at 7–9 months of age infants' prefrontal brain activity is most sensitive to their mothers' IDS.

#### **VISION**

Another important area of investigation is the work on the perception of visual social stimuli. The human face provides the infant with a wealth of socially relevant information such as age, gender and emotional state. From birth, human infants preferentially attend to faces (Johnson and Morton, 1991). For example, Tzourio-Mazoyer et al. (2002) presented 2-month-old infants with a face or a control stimulus, while measuring brain activity using positron emission tomography (PET) (note that although PET is not commonly used with infants due to the fact that it exposes infants to small amounts of radiation, the infants scanned in this study were tested in an intensive care unit as part of a clinical follow-up). In this study, when viewing faces infants not only activated regions in temporal cortex involved in distinguishing faces from other visual stimuli but also showed activation within the mPFC in the right hemisphere. This suggests that already at this young age infants recruit parts of the so-called extended face processing network that are considered to be crucial in assigning social and affective significance to faces (Haxby et al., 2000).

An important communicative signal conveyed by faces is eye gaze. The monitoring of eye gaze direction is essential for effective social learning and communication among humans (Csibra and Gergely, 2009), with eye contact being one of the most powerful modes of establishing a communicative link between humans (Kampe et al., 2003). In an fNIRS study, 4-month-old infants watched two kinds of dynamic scenarios in which a face either established eye contact or averted its gaze followed by a smile (Grossmann et al., 2008). The results revealed that, similar to what is known from adults (Kampe et al., 2003; Pelphrey et al., 2004), processing eye contact activates not only superior temporal cortex implicated in processing information from biological motion cues but also the mPFC important for social and affective communication. Moreover, in the same study, measuring electrical brain responses over PFC in another group of 4-monthold infants showed that only a smile that was preceded by eye contact evoked increased PFC responses in 4-month-old infants (Grossmann et al., 2008), supporting the notion that already in infancy mPFC plays a role in interpreting social and affective information directed at the self.

That smiling at an infant while making eye contact is a powerful cue triggering mPFC activation has also been demonstrated in another fNIRS study (Minagawa-Kawai et al., 2009), in which 9- to 12-month-old infants were presented with videos of either their own mother or a female stranger smiling at them or looking neutrally at them. Smiling at the infants evoked greater activity in mPFC regardless of the familiarity with the face, suggesting that mPFC is flexibly employed during positive social interactions. Nevertheless, mPFC activity was significantly greater in response to the own mother smiling when compared to the female stranger smiling, suggesting that infants' mPFC responses are particularly sensitive to affective cues from the primary caregiver. Interestingly, in this study it was shown that mothers exhibited a very similar mPFC response when looking at their own infants' smiling, thus pointing to a shared neural mechanism engaged during social interaction between caregivers and infants.

Eye gaze also plays an important role in coordinating attention during triadic interactions between self, other, and the environment. During a typical triadic interaction, a person may establish eye contact with another person and then direct that person's gaze to an object or event. In a recent study, fNIRS was used to localize infant prefrontal brain responses during triadic social interactions (Grossmann and Johnson, 2010). The results showed that by the age of 5 months, infants are sensitive to triadic interactions and, like adults, they recruit a specific prefrontal region localized in a dorsal part of the PFC (at the border between mPFC and lPFC) in the left hemisphere only when engaged in triadic interaction with another person but not during the conditions that controlled for certain aspects of the social interaction but were not triadic in nature (Schilbach et al., 2010). Very recently, it was shown that mPFC is not only involved when an adult guides infant attention to an object through gaze behavior, but it is also implicated in infants' detection of when a social partner followed their own gaze to an object, suggesting that infants flexibly use this brain region to coordinate attention with others regardless of whether the interaction is initiated by others or by themselves (Grossmann et al., under review).

The finding that specific parts of the mPFC play a role in triadic interactions receives more support from recent work examining the perception of human action. In this study (Lloyd-Fox et al., 2011), when 5-month-old were presented with actions (hand movements) while being addressed through eye contact and thereby creating a triadic interaction, they showed increased activation within the mPFC. The same regions of the mPFC were not active when human actions that were purely dyadic in nature such as mouth movement or eye gaze shifts.

#### **AUDITION AND VISION**

In adults, initiating a social interaction by eye contact and calling a person's name results in overlapping activity in the mPFC (Kampe et al., 2003), suggesting that, regardless of modality, the intention to make contact is detected by the same brain region. In a recent fNIRS study (Grossmann et al., 2010), 5-month-old infants watched faces that either signaled eye contact or directed their gaze away from the infant, and they also listened to voices that addressed them with their own name or another name, in order to examine the neural basis of detecting social interactive signals across modalities. The results of this study revealed that infants recruit adjacent regions in the mPFC when they process eye contact and their own name. Moreover, 5-month-old infants that responded sensitively to eye contact in the one mPFC region were also more likely to respond sensitively to their own name in the adjacent mPFC region as revealed in a correlation analysis, suggesting that responding to communicative signals in these two regions is functionally related. These fNIRS results suggest that infants at the age of 5 months selectively process and flexibly attend to social interactive signals across modalities.

## **DISCUSSION**

This review presented an overview of the experimental evidence on infants' mPFC involvement during the processing of auditory and visual social information. We have seen that infants employ mPFC during a wide range of contexts, including the perception of emotional and infant-directed speech cues in the auditory domain and the perception of faces and eye gaze cues in the visual domain. These findings support the central thesis that mPFC is important from early in ontogeny, playing a vital role in the emergence of social cognitive abilities during infancy. This notion stands in contrast to the idea that mPFC matures late and only plays a role later in ontogeny when a more explicit understanding of the social world is achieved (Singer, 2006; Blakemore, 2008).

On the basis of the evidence presented above, it could even be argued that mPFC is more important earlier in development than later in development because it is critically involved in the acquisition of social cognitive abilities from birth and becomes less important once social cognitive and interactive abilities have been robustly acquired. That this might indeed be the case is evident in the mPFC lesion work presented earlier where it was shown that early onset compared to adult onset lesions to mPFC resulted in more severe outcomes in terms of social and moral impairments (Anderson et al., 1999). More support for this view of mPFC playing a greater role earlier in development comes from neuroimaging work on social cognition with adolescents, which shows that while the engagement of posterior regions of cortex increases with age, mPFC involvement in social cognitive tasks decreases with age during adolescence (for a review, see Johnson et al., 2009). This can be seen as evidence for a reduction of the involvement of mPFC in social cognition during development, which also concurs with another line of work demonstrating that prefrontal regions play a greater role during the acquisition of a new perceptual skill (Gilbert and Sigman, 2007). Therefore, one implication of the work presented here is that mPFC plays a role in the acquisition of social cognitive skills from early in ontogeny. In general, this notion is in line with views that conceive of infants as competent social learners, entering the world readily prepared for social interaction and social thinking (Meltzoff, 2007; Spelke and Kinzler, 2007; Csibra and Gergely, 2009).

But what is the functional role that mPFC takes on in the early development of social cognition during infancy? I would like to put forward the proposal that mPFC involvement in infancy (and beyond) is likely to be important for the detection of selfrelevant information. This proposal is based on (a) the observed pattern of mPFC involvement in the studies reviewed above, and (b) an extensive body of evidence from prior work with adults, implicating mPFC in assessing and representing information with reference to the self (for a review, see Amodio and Frith, 2006). This proposal can thus be seen as a developmental extension of prior accounts of adult mPFC function into infancy. More specifically, as shown above, mPFC is involved in infants' responding to social interactive cues, which index that information is relevant to the self such as during the listening to infant-directed speech or their own name, perceiving eye contact, or experiencing a triadic interaction. This increased sensitivity to self-relevant information might serve critical learning functions because it highlights potentially useful information that others present to the infant (Sperber and Wilson, 1995; Csibra and Gergely, 2009). In support of this view, it has been shown that infants' learning is influenced and improved when they are addressed by infant-directed speech and eye contact (Singh et al., 2004; Senju and Csibra, 2008; Yoon et al., 2008). The mPFC might thus be involved in learning from others by detecting the relevance of others' actions with reference to the self. Obviously, this sensitivity to self-relevant information in infancy does not imply that infants have an explicit (conceptual) understanding of the self (Rochat, 2003, 2011). However, one may argue that the sensitivity to self-relevant information serves as a powerful foundation for developing a sense of self because it provides infants with the opportunity to experience when the self is being addressed in an interaction. In fact, it has been argued that early social interactions during infancy and the experiences gained therein can be considered the cradle of self development (Reddy, 2003).

One intriguing implication of this proposal is that by measuring mPFC involvement in a given context, one might be able to examine the extent to which an infant perceives information as self-relevant. For example, on a trial- by-trial basis one could look at infants' mPFC response to eye contact and then see whether or not infants are more likely to show gaze following in response to an eye gaze shift of a social partner. The prediction based on the proposal presented above is that on trials during which infants show mPFC involvement when seeing eye contact they should be more likely to gaze follow. In behavioral work, it has already been shown that infants are more likely to gaze follow when they had previously been presented with eye contact or heard infantdirected speech (Senju and Csibra, 2008), however, it is unclear what the underlying neural processes are that correlate with this behavioral phenomenon. Moreover, this proposed approach might also be useful in assessing inter-individual differences in the perception of relevance to the self in response to identical stimuli. In such a scenario, we might be able to identify infants that tend to show little sensitivity to perceptual social signals indicating self-relevance but also infants that are overly sensitive to social information even if it is not directed at them. The potential existence of extreme biases in either direction in early development might have serious detrimental effects on social development in the long term. For example, a strongly reduced sensitivity to self-relevant information might be linked to neurodevelopmental disorder such as autism, where it has been shown that lacking behavioral sensitivity to self-relevant signals such as eye contact and name cues are some of the earliest detectable warning signs for the later development of autism (Zwaigenbaum et al., 2005; Elsabbagh and Johnson, 2007; Elsabbagh et al., 2012). The development of biomarkers such as brain-based measures to guide an early identification of developmental disorders is still in its infancy but has been shown to be of great promise, especially when relying on measures that assess infants' responses to eye contact (Elsabbagh et al., 2012).

Despite the progress that has been made in elucidating the role of mPFC in early development, in order to gain a better and more complete picture of mPFC function in infancy, it is vital to address a number of outstanding issues. First, more work is needed to precisely map and compare activation within the mPFC across social tasks during infancy. Specifically, as far as the infant fNIRS data presented in the review is concerned, no standardized anatomical mapping of the functional activation in PFC has been employed that would allow us to compare and integrate the information about mPFC across studies and tasks in a meta-analysis. This issue becomes particularly important when one considers the fact that in adults there appear to be considerable functional divisions within mPFC (Amodio and Frith, 2006; Bzdok et al., 2013). First strives have been made at standardizing the analysis of infant fNIRS data that promise to provide a better basis for carrying out such comparisons (Cristia et al., 2013). Nonetheless, a remaining issue is the limited depth resolution of fNIRS, as commonly used in infant studies, that obtains most of the signal from superficial cortical structures but is virtually blind to deeper cortical sources (Lloyd-Fox et al., 2010). Second, there is very little work comparing mPFC activation across ages during infancy because most work is focused on one particular age group. The few studies that have looked at various age groups during infancy revealed intriguing insights into how mPFC function changes and becomes more finely tuned to social signals from the caregiver (Naoi et al., 2012). A systematic examination of mPFC function across infancy will provide important information concerning the functional specialization of this brain region. Third, another

#### **REFERENCES**


important aspect to consider is that while we have observed activation of individual mPFC regions during infancy, we do not know whether the activity of the mPFC and other brain regions is coordinated into functional networks as seen in adults. There is work using resting-state fMRI with infants indicating that some of the functional connections between certain parts of mPFC and posterior cortical regions known in adults are not yet developed in infants (Fransson et al., 2007). Furthermore, resting-state studies testing infants across various ages show that this long-range integration of cortical activity emerges throughout the first few years of life (Gao et al., 2009; Homae et al., 2010; Fransson et al., 2011). The relevance that these changes in resting-state activity have for infants' brain function while actively involved in one of the experimental tasks reviewed here is unclear, and requires attention in future work.

Taken together, the findings from the studies presented here provide evidence that mPFC plays an important role in social cognition from very early in development. Based on the reviewed experimental data, I put forward the proposal that mPFC involvement in social information processing in infancy is related to the detection of self-relevant information. This look at early social cognition through the lenses of social neuroscience allowed us to better understand the nature and developmental origins of mPFC function by closing a gap between the extensive behavioral work showing sophisticated social cognitive skills in infants and work with adults concerning the pertinent role of mPFC played in social cognition. It is my hope, that this review will further stimulate work illuminating the neural basis of social cognition in infancy and foster the crosstalk between developmental psychologists and social neuroscientists.


2-week-old to 2-year-old healthy pediatric subjects. *Proc. Natl. Acad. Sci. U.S.A.* 106, 6790–6795. doi: 10.1073/pnas.0811221106


*Proc. Biol. Sci.* 275, 2803–2811. doi: 10.1098/rspb.2008.0986


*Biobehav. Rev.* 34, 269–284. doi: 10.1016/j.neubiorev.2009.07.008


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 15 April 2013; paper pending published: 06 May 2013; accepted: 17 June 2013; published online: 05 July 2013.*

*Citation: Grossmann T (2013) The role of medial prefrontal cortex in early social cognition. Front. Hum. Neurosci. 7:340. doi: 10.3389/fnhum.2013.00340*

*Copyright © 2013 Grossmann. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any thirdparty graphics etc.*

# Dorsomedial prefrontal cortex activity predicts the accuracy in estimating others' preferences

# *Pyungwon Kang1,2†, Jongbin Lee1,2†, Sunhae Sul 1,3 and Hackjin Kim1,3\**

*<sup>1</sup> Laboratory of Social and Decision Neuroscience, Korea University, Seoul, South Korea*

*<sup>2</sup> Department of Brain and Cognitive Engineering, Korea University, Seoul, South Korea*

*<sup>3</sup> Department of Psychology, Korea University, Seoul, South Korea*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Nicholas O. Rule, University of Toronto, Canada Peter E. Mende-Siedlecki, Princeton University, USA*

#### *\*Correspondence:*

*Hackjin Kim, Department of Psychology, Korea University, 145 Anam-ro, Sungbuk-Ku, Seoul 136-701, South Korea e-mail: hackjinkim@korea.ac.kr*

*†These authors have contributed equally to this work.*

The ability to accurately estimate another person's preferences is crucial for a successful social life. In daily interactions, we often do this on the basis of minimal information. The aims of the present study were (a) to examine whether people can accurately judge others based only on a brief exposure to their appearances, and (b) to reveal the underlying neural mechanisms with functional magnetic resonance imaging (fMRI). Participants were asked to make guesses about unfamiliar target individuals' preferences for various items after looking at their faces for 3 s. The behavioral results showed that participants estimated others' preferences above chance level. The fMRI data revealed that higher accuracy in preference estimation was associated with greater activity in the dorsomedial prefrontal cortex (DMPFC) when participants were guessing the targets' preferences relative to thinking about their own preferences. These findings suggest that accurate estimations of others' preferences may require increased activity in the DMPFC. A functional connectivity analysis revealed that higher accuracy in preference estimation was related to increased functional connectivity between the DMPFC and the brain regions that are known to be involved in theory of mind processing, such as the temporoparietal junction (TPJ) and the posterior cingulate cortex (PCC)/precuneus, during correct vs. incorrect guessing trials. On the contrary, the tendency to refer to self-preferences when estimating others' preference was related to greater activity in the ventromedial prefrontal cortex. These findings imply that the DMPFC may be a core region in estimating the preferences of others and that higher accuracy may require stronger communication between the DMPFC and the TPJ and PCC/precuneus, part of a neural network known to be engaged in mentalizing.

#### **Keywords: preference estimation, dorsomedial prefrontal cortex, temporoparietal junction, posterior cingulate cortex/precuneus, thin-slice judgment, theory of mind**

# **INTRODUCTION**

We often need to infer another person's preferences based on very limited information in daily life. For example, we choose a restaurant for dinner with an invited speaker whom we have never met before, make a plan for a first date, prepare a small gift for a new business partner, or rely on our intuitive feelings about customers to see through their preferences. Although estimating the preferences of others frequently occurs without prior knowledge, most studies on this topic have focused on how people utilize known information to estimate preferences (Hoch, 1988; West, 1996; Lerouge and Warlop, 2006). Only recently, North and colleagues have shown that people can estimate the preferences of others based on shortly presented subtle and non-communicative facial expressions (North et al., 2010). The present study centered on the ability to accurately estimate another person's preferences on the basis of minimal information.

The ability to infer about others quickly and act accordingly is important for leading a successful social life, and this kind of intuitive social inference has been well documented in social psychology literature (Funder and Harris, 1986; Ambady and Rosenthal, 1992; Zaki and Ochsner, 2011). It is known that people can infer various types of information, such as personality (Berry, 1991; Gosling et al., 2002), trustworthiness (Engell et al., 2007; Van't Wout and Sanfey, 2008), competence (Todorov et al., 2005), altruism (Fetchenhauer et al., 2010), socioeconomic status (Kraus and Keltner, 2009), sexual orientation (Rule et al., 2009; Freeman et al., 2010a), violence of sexual offenders (Stillman et al., 2010), as well as preferences (North et al., 2010), on the basis of a brief (usually ranging from 2 s to 5 min) exposure to facial appearance or to an excerpt of behavior. Ambady and colleagues have emphasized the adaptive function of accurately judging others based on minimal information (Ambady et al., 1995) and have suggested that this ability reflects the interpersonal sensitivity of an individual (Ambady et al., 2001). Despite a large body of behavioral evidence, only a few neuroimaging studies have investigated the neural mechanisms underlying the accuracy of personal traits that are inferred from facial appearances (Spezio et al., 2008;

**Abbreviations:** SP, self-preference; eTP, estimated target preference; aTP, actual target preference; GP, general preference; TPJ, temporoparietal junction; ToM, theory of mind; DMPFC, dorsomedial prefrontal cortex; VMPFC, ventromedial prefrontal cortex; MPFC, medial prefrontal cortex

Rule et al., 2010, 2011), and, most importantly, no studies have been conducted on the accuracy of estimating the preferences of others.

Of most relevance to the current work are recent studies on the role of the dorsomedial prefrontal cortex (DMPFC) in interpersonal judgment (Mitchell et al., 2005b; Jenkins and Mitchell, 2010; Cooper et al., 2012). For example, Mitchell et al. (2005b) have compared neural correlates for forming impressions of other people vs. inanimate non-human objects and have found that the DMPFC is specifically engaged in processing information about other people. Another study on rapid evaluations of potential romantic partners has found that the neural activity of the DMPFC predicts the outcome of the subsequent romantic interactions (Cooper et al., 2012). Although these studies did not focus in particular on the accuracy of the preference estimation, they provide a hint that the DMPFC may play a major part in this process.

In addition, estimating the preferences of others based on intuition involves the theory of mind (ToM) that enables mentalizing (Gore and Sadler-Smith, 2011) and cognitive control, which allows the inhibition of the self-projection of one's own state (Hoch, 1988; West, 1996). For instance, if a *perceiver* (one who is required to infer the tastes of another person) is trying to guess whether a *target* (one whose tastes are predicted by the perceiver) would like to watch a Harry Potter movie, the perceiver needs to inhibit his/her own opinion from influencing the prediction (cognitive control) and to put him/herself into the target's shoes (mentalizing). Given that these processes engage DMPFC activity (Amodio and Frith, 2006; Lieberman, 2007), it is reasonable to expect that the DMPFC plays an important role in estimating preferences.

Other brain areas, such as the temporoparietal junction (TPJ) and the precuneus, have been strongly implicated in the ability to infer the mental states of others (Saxe and Kanwisher, 2003; Amodio and Frith, 2006; Mitchell, 2006; Van Overwalle and Baetens, 2009; Freeman et al., 2010b; Denny et al., 2012). The development of the ability to infer another person's mind coincides with the maturation of these structures (Sabbagh et al., 2009; Gweon et al., 2012) and, more importantly, activities in the ToM network appear to be critical for forming impressions upon seeing strangers' faces (Zaki et al., 2009; Rule et al., 2011). Taken together, these findings further imply that this network of neural structures involved in the ToM may influence the accuracy in estimating others' mental states and, therefore, may also take part in estimating the preferences of others.

The aims of the present study were to examine whether people can estimate the preferences of others based on a briefly presented facial appearance and to investigate the neural correlates of this ability. Prior to the main experiment, we ran separate sessions to select the items and targets and conducted a preliminary behavioral experiment (pretest) to confirm whether people are capable of inferring the preferences of others from facial appearances. In the main experiment, we investigated the underlying neural mechanisms. Participants were asked to estimate the preferences of targets for various items after they saw each target's facial photograph for 3 s (preference estimation task) while their brain activity was measured with functional magnetic resonance imaging (fMRI). We hypothesized that the activation of the DMPFC and other brain regions of the ToM and mentalizing network would be associated with the accuracy of the preference estimation.

# **MATERIALS AND METHODS ITEM SELECTION**

Eighteen raters were asked to evaluate the photographs of 280 items from five categories (i.e., movies, books, bags for men and women, shoes for men and women, and foods) on preference rating scales ranging from −4 (strongly hate) to 4 (strongly like). Ten among the initial 40 items from each category were selected based on the mean and standard deviation of their preference ratings. More specifically, with the aim to minimize the overlap between the preference of the general population and the preference of a target person for a given item, we avoided the items that earned a high consensus by selecting items with large variances and intermediate levels of mean preference ratings. For movies and books whose contents were not readily recognizable from the presented photographs (i.e., movie posters and book covers), the raters were asked to answer how well they knew about each item on a 4-point scale (1, never known before; 2, know the name; 3, have not seen/read it but know the contents; 4, have seen/read it). The items rated below 3 in the knowledge score by more than half of the raters were excluded. As a result, 10 items were selected for each category and used as stimuli for the pretest. For the fMRI experiment, only two categories (i.e., movies and foods) were chosen based on the results from the pretest (see *Pretest* for details).

#### **TARGET SELECTION**

We recruited 56 undergraduate students (27 males; 22*.*78 ± 1*.*95 years) through online advertisements as targets, whose preferences were to be estimated by perceivers in the pretest and in the main experiment. In order to minimize the possibility that the perceivers had met the targets before, we ensured that the targets and the participants for the fMRI experiment had been recruited from different institutions. We took facial photographs of all 56 target candidates and filmed short self-introducing video clips starting with "Hello" in Korean. For the facial photographs, the candidates were asked to make a neutral face with a slight smile. After taking the photographs and making the films, the candidates were presented a list of items that were selected as described above and asked to evaluate them on 4-point preference scales that ranged from 1 (strongly dislike) to 4 (strongly like). All candidates were informed and agreed that the photographs and video clips would be shown to other participants in another experiment. The photos and video clips were edited into an identical frame (700 × 400 pixels); the video clips were edited to a 3-s length to contain only the part in which they say "Hello" in Korean. Four male and four female targets (eight targets in total) were chosen as the final stimuli for the pretest based on the following two criteria. First, we sorted the participants according to their similarity of appearance and selected targets who were dissimilar to each other in order to maximize the between-target variability of appearances. Second, we excluded targets who showed indistinct preferences to increase the within-target variability of the preferences. The same targets were used in the video-clip and photo conditions. For the fMRI experiment, we only included female targets in order to eliminate potential gender effects, and nine female targets were selected with the same criteria.

#### **PRETEST**

We ran a pretest before the main fMRI experiment to ensure that the participants were capable of accurately estimating another person's preference in our experimental setting. Nineteen undergraduate students (eight males, 22*.*79 ± 1*.*72 years) were recruited for the pretest. Ten participants (four males) were assigned to the photo condition in which the targets were presented in photographs and nine participants (four males) were assigned to the video clip condition in which the targets were shown in the video clips. One participant who rated all items indiscriminately as highly preferred was excluded from the analyses.

The participants were asked to make guesses about the preferences of the eight targets (target-trials) or to indicate their own preference (self-trials) for various items. In the target-trials, after a 1-s fixation, a photograph (or a video clip) of a target was shown for 3 s, and this was followed by a photograph of an item. Participants were asked to guess the target's preference for the given item within 5 s. In the self-trial, the initial letters of the participant's own name were presented instead of the facial photo. The trials were presented in a pseudorandom order. There were 90 trials for each of the five item categories: eight target-trials and one self-trial with 10 items per category. For the bags and shoes categories, 10 additional trials were added to the self-trials so that the participants could report their own preferences for the items for the opposite sex as well. As a result, the participants performed 470 trials in total. At the completion of the main task, the participants were asked to estimate the preference of the general population for each of the items.

To measure the accuracy of the preference estimations, we counted the number of trials in which the participants correctly estimated the valence of the targets' preference and then calculated the proportion of these correct trials for each category. For example, if a target's preference for a given item was 4 (strongly like) and a participant estimated it as 3 (like), then this trial was regarded as correct. In contrast, if a target's preference for a given item was 2 (dislike) and a participant estimated it as 3 (like), then this trial was considered incorrect because the participant failed to match the valence of the target's preference (i.e., in the preference ratings, 1 and 2 indicate dislike, whereas 3 and 4 indicate like, see Target Selection for details). The average accuracy scores across all of the categories were significantly above chance level (50%) [*t(*17*)* = 8*.*52, *p <* 0*.*01, *d* = 4*.*13]. When we looked into each category separately, the preferences for movies, shoes, and foods were correctly estimated (all *p*s *<* 0*.*01), while the preferences for books and bags were not (all *p*s *>* 0*.*1; See **Table 1**).

These results indicated that, at least in some domains, the participants could accurately estimate the preferences of others, even with very brief exposure to limited information, such as a video clip or facial appearance. However, the possibility remained that the participants might have referred to their own preferences [e.g., self-projection, as Hoch (1988) has suggested] **Table 1 | The descriptive statistics of all conditions in the pretest and the results of the one-sample** *t***-tests.**


*SD, standard deviation. t, t-scores from one-sample t-tests against chance level on the average accuracy scores across photo and video clip conditions. \*p < 0.05.*

or to the preferences of the general population instead of considering target-specific information. To test this possibility, we analyzed the correlations between the participants' preference estimations in the target-trials (estimated target preference, eTP) and their own preferences in the self-trials (self preference, SP), as well as their estimation about the preferences of the general population (general preference, GP), for each item. The correlation coefficients were converted into z-scores using Fisher's r-to-z transformation for statistical tests. The *z*-scores that were averaged across the participants were back-transformed into the r scores reported below (Michela, 1990). The average correlation between eTP and SP was *r* = 0*.*43, *t(*21*)* = 9*.*30, *p <* 0*.*01, *d* = 4*.*05, and the average correlation between eTP and GP was *r* = 0*.*47, *t(*21*)* = 11*.*03, *p <* 0*.*01, *d* = 4*.*81, indicating that eTP was partly influenced by SP and GP. These correlations were controlled for when we analyzed the behavioral and fMRI data in the main experiment.

Some previous studies have reported that people can make better judgments with dynamic cues rather than static cues because they contain richer information (Valenti and Costall, 1997; Balas et al., 2012). However, in our study, we did not find any significant differences between the cue type (video clip vs. photo) in the estimation accuracy, except for movie items [*t(*17*)* = 3*.*13, *p <* 0*.*01, *d* = 1*.*51]. This might have been due to the relatively simple features of our video clips in which targets said a very simple word ("Hello") and rarely made facial or body movements. In addition, the estimation accuracy did not differ when the perceiver's and target's genders were the same and when they were the opposite [*t(*17*)* = 0*.*97, *ns*], but the perceivers generally estimated the preferences of the male targets better than the female targets [*t(*17*)* = 3*.*66, *p <* 0*.*01, *d* = 1*.*77]. There was no significant perceiver's gender difference [*t(*17*)* = 1*.*08, *ns, d* = 0*.*52].

### **fMRI EXPERIMENT PARTICIPANTS**

Twenty-two college students (all females, 22*.*5 ± 2*.*28 years) participated in the fMRI experiment. We recruited only female participants to rule out potential gender effects because previous studies have reported that females are better than males at thin-slice judgments about others (Vogt and Colvin, 2003; Carney et al., 2007). All participants were right-handed and screened for a history of psychiatric or neurological diseases. This study was approved by the institutional review board of Korea University.

#### **PREFERENCE ESTIMATION TASK**

The preference estimation task for the fMRI experiment was similar to that of the pretest, except for the following details (see **Figure 1**). Because we found no significant difference between the photo and video clip conditions in the pretest, we used only photo cues for the fMRI experiment. A fixation phase with 1– 3-s jittered fixation was added between the face phase and the item phase in order to better separate the two events in the eventrelated design. A 0.5-s response-display phase was added to the 3-s item phase so that the participants could see whether they pressed a response button as they intended. Unlike the pretest, the participants' own facial photographs were taken and presented in the self-trials during the face phase in order to make the visual stimuli in the self-trials comparable to those in the target-trials. Among the three categories in which the participants could estimate the preferences of others above chance level in the pretest, two categories (movies and foods) were chosen as the stimuli for the fMRI experiment. The participants performed the preference estimation task for each category separately in two scanning sessions, and the order of the two sessions was counterbalanced. Unlike the pretest, nine (all female) targets were used in the fMRI experiment. As a result, each session consisted of 10 self and 90 target-trials in total, which rendered approximately a 20-min scanning time per session. The order of the items and the targets was pseudo-randomized in order to avoid the same item or target being shown consecutively.

#### **PROCEDURE**

When the participants arrived at the experiment room, they were instructed about the preference estimation task and were told that the targets' actual preferences were measured in a separate session a few weeks before. To prevent the participants from responding randomly, we told the participants that they would receive additional monetary incentives depending on their performance if the accuracy level was above chance level. The participants' facial photographs were taken before they entered the scanning room. These photos were edited to the same size and resolution as those of the targets. After completing the preference estimation task inside the MRI scanner, the participants were asked to guess the preferences of the general population for every item that was shown in the scanner. The average payment for participation was approximately 30,000 KRW (≈ \$30).

#### **fMRI DATA ACQUISITION**

The brain images were collected on a 3-T Siemens Trio MRI scanner (MAGNETOM Trio, A Tim System; Siemens AG, Erlangen, Germany) with a 12-channel birdcage head coil at the Korea University Brain Imaging Center. We acquired high-resolution anatomical images (*TR* = 1900 ms; *TE* = 2*.*52 ms; flip angle, 9 degrees; 1 × 1× 1 mm in-plane resolution; and 256 × 256 matrix size), and then obtained functional images through gradient echo planar images (EPI) with Blood Oxygenation Level-Dependent contrast (*TR* = 2000 ms; *TE* = 30 ms; flip angle = 90 degrees; 3 × 3× 4 mm in-plane resolution; 64 × 64 matrix size; and 33 slices with no gap).

#### **fMRI DATA ANALYSES**

The fMRI data were preprocessed and analyzed with SPM8 (Wellcome Department of Imaging Neuroscience, London, UK). The images were realigned to correct for head motion, spatially normalized to the standard Montreal Neurological Institute EPI

screen for 0.5 s. In the self-trials, the perceivers reported their own

face photo (perceiver's face photo) was shown for 3 s in the target-trials (self-trials), and a photo of the item was displayed after

preference for the item.

template, and smoothed with a Gaussian kernel (6 mm full-width at half-maximum).

We constructed a general linear model for each participant including the following regressors: (1) the face phase of the selftrial, (2) the face phase of the target-trial, (3) the item phase of the self-trial along with (4) the SP rating as a parametric regressor, (5) the item phase of the target-trial along with (6) the eTP as a parametric regressor, (7) the response-display phase of the self-trial, and (8) the response-display phase of the targettrial. Additionally, six head motion regressors were included as covariates of no interest.

In order to identify the brain regions that showed significant correlations with the participants' performance on the preference estimation, we performed a regression analysis of the contrast images of the target-trials vs. the self-trials in the item phase with individual accuracy score as a covariate. Additionally, we performed a similar multiple regression analysis while controlling for the effects of SP and GP by adding the individual average correlation coefficients of the eTP with SP and GP as covariates.

Subsequently, we performed a psychophysiological interaction (PPI) analysis (Friston et al., 1997) with the peak voxel (*x* = 18, *y* = 50, *z* = 40) from the DMPFC cluster found in the multiple regression analysis as a seed region and the contrast for the main effect of the correct vs. incorrect target-trials as a psychological variable. This allowed us to identify the brain regions that showed increased functional connectivity with the DMPFC when the participants made correct estimations of the targets' preferences as compared to when they made incorrect estimations. For the PPI analysis, a design matrix was constructed to include the following three regressors: (1) the time series data from the DMPFC, (2) the psychological variable contrasting the correct and incorrect target-trials, and (3) the interaction between (1) and (2). In addition, the individual accuracy scores were regressed to the PPI between the DMPFC and other brain regions during the correct vs. incorrect target-trials. This analysis allowed us to identify the brain regions that showed stronger functional connectivity with the DMPFC in the correct than in the incorrect target-trials among the participants with higher accuracy scores.

We applied statistical significance parameters based on a peak threshold and a spatial extent threshold to correct for multiple comparisons at a level of *p <* 0*.*05. Using AlphaSim implemented in Analysis of Functional NeuroImages (AFNI), 1,000 Monte Carlo simulations were conducted to determine the spatial extent threshold [Parameters for AlphaSim: voxel threshold, *p <* 0*.*005 (uncorrected); smoothness, 8.67, 8.62, and 8.57 mm (determined by 3dFWHMx); voxel size, 2 × 2× 2 mm]. For the multiple regression analysis with the target vs. self contrasts at the item phase regressed onto the accuracy scores, we restricted the search volumes to the brain regions involved in mentalization (80,837 mm3), such as the DMPFC, the TPJ, the posterior cingulate cortex (PCC)/precuneus, and the medial prefrontal cortex (MPFC), which were defined anatomically according to the Anatomical Automatic Labeling (AAL) atlas. Those brain regions were combined to create a mask of mentalization region of interest (ROI). For the multiple regression analysis with the target vs. self contrasts at the item phase that was regressed on the correlation coefficients between eTP and SP, we restricted the search volume to the VMPFC (53,330 mm3) that was anatomically determined based on the AAL atlas. For all of the other analyses, the whole brain volume (672,900 mm3) was used to determine the spatial extent threshold.

## **RESULTS**

#### **BEHAVIORAL RESULTS**

As expected, the mean accuracy of the estimated preferences of the others for the two categories was significantly above chance level [*t(*21*)* = 11*.*00, *p <* 0*.*001, *d* = 4*.*80; see **Figure 2**]. The mean accuracy level remained above chance level even when tested separately for movie [*t(*21*)* = 6*.*68 *p <* 0*.*05, *d* = 2*.*91] and food [*t(*21*)* = 9*.*61 *p <* 0*.*05, *d* = 4*.*19] items. The accuracy scores for the two categories were not significantly different from each other [*F(*1*,* <sup>19</sup>*)* <sup>=</sup> <sup>0</sup>*.*35, *ns,* <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*02]. Thus, we combined the two categories in the subsequent analyses.

As in the pretest, we examined how the eTP was distinguished from the SP as well as the GP. The average correlation coefficient between the SP and the average of the eTP for each item was *r* = 0*.*52, *t(*21*)* = 7*.*93, *p <* 0*.*01, *d* = 3*.*46, and the average of the correlation coefficient between the GP and the average of the eTP for each item was *r* = 0*.*65, *t(*21*)* = 8*.*40, *p <* 0*.*01, *d* = 3*.*66. SP and GP seemed to be significantly correlated with eTP. We took this into account by statistically controlling for the effect of SP and GP in the fMRI analysis.

In addition, to verify the accuracy of the eTP even after controlling for its correlations with SP and GP, we performed a linear regression analysis on the actual target-preference (aTP) with the perceivers' eTP, SP, and GP ratings. Then, we examined the degrees to which the accuracy scores correlated with the beta coefficients of the eTP, SP, and GP ratings from the regression analysis. The accuracy scores correlated significantly with the beta coefficients of the eTP for movie items (*r* = 0*.*51, *p <* 0*.*05) but only marginally for food items (*r* = 0*.*40, *p* = 0*.*06). Beta coefficients

[*M* = 62*.*18 ± 5*.*19, *t(*21*)* = 11*.*00, *p <* 0*.*001].

for neither the SP nor the GP ratings showed significant correlations with accuracy scores (all *p >* 0*.*1). In addition, we regressed the eTP on aTP, SP, and GP. The beta coefficients of aTP were correlated significantly with the accuracy scores for food items (*r* = 0*.*43, *p <* 0*.*05) and marginally with those for movie items (*r* = 0*.*35, *p* = 0*.*11). Neither SP nor GP correlated significantly with the accuracy scores (all *p >* 0*.*2). In summary, although the SP and GP ratings contributed partly to the accuracy scores, the perceiver's estimations of the target's preferences seemed to be the most significant factor that accounted for the variation in the accuracy scores among the perceivers.

In addition, we examined a potential learning effect, that is, whether time or repetition had any influences on the accuracy of the estimations of the target preferences. We calculated the performances separately for each block of targets grouped by the presentation order in each perceiver and conducted a repeated measure ANOVA. This analysis yielded no significant repetition effect [*F(*9*,* <sup>387</sup>*)* = 0*.*61, *p* = 0*.*78]. We also examined if there was any potential order effect between the two separate fMRI scanning sessions in terms of estimation accuracy and again found no order effect [*F(*1*,* <sup>21</sup>*)* = 1*.*18, *p* = 0*.*29].

Finally, to examine the potential variability in terms of the readability among targets, we computed a readability score for each target by averaging the ratio of correct trials for the specific target separately for each item category (i.e., movies and foods), which indicated the degree of estimation difficulty. For example, if all perceivers correctly estimated a target's preference toward five movie items, the target's readability score for the movie category would be 5. The readability scores varied from 4.49 to 7.27 (the highest possible score was 10) for the movie category and from 4.77 to 7.90 for the food category, indicating that some targets were easier to estimate than others. Given that the correlation of the readability scores between the two categories was not significant (*r* = −0*.*02, *p >* 0*.*1), however, the variability in the readability of the targets seemed to be largely dependent on the item category rather than on the target *per se*.

#### **fMRI RESULTS**

Our primary goal was to investigate which brain regions were involved in the process of *accurately* estimating another person's preferences with minimal information. Before addressing this question, we first explored the brain regions that engaged more when estimating the preferences of others compared to oneself. We conducted a whole brain analysis by contrasting target- vs. self-trials during the item phase. No brain region survived even at a lenient statistical threshold (*p <* 0*.*1, uncorrected). From the whole brain analysis contrasting the selfvs. target-trials during the item phase, we found brain regions that are known to be involved in self-reference processing, such as the MPFC (*x* = 0, *y* = 50, *z* = 6, *Z* = 5*.*31, corrected, *p <* 0*.*05), the PCC/precuneus (*x* = −10, *y* = −50, *z* = 18, *Z* = 3*.*58, corrected, *p <* 0.05), and the left inferior parietal cortex (*x* = −48, *y* = −46, *z* = 54, *Z* = 4*.*17, corrected, *p <* 0.05) (Kelley et al., 2002; Northoff et al., 2006; Sul et al., 2012), and other brain regions (See Table S1). No significant cluster was found when we contrasted the correct vs. incorrect target-trials and the incorrect vs. correct target-trials during the item phase.

#### **NEURAL CORRELATES OF THE INDIVIDUAL DIFFERENCES IN THE ACCURACY OF ESTIMATING THE PREFERENCES OF OTHERS**

As shown in **Figure 2**, the individual accuracy scores varied significantly across the participants, and this might have been the reason why no significant cluster was observed in the main contrasts of the previous analyses. Thus, we aimed to examine the neural correlates of the individual differences in the accuracy of estimating targets' preferences. We performed a regression analysis in which the individual contrast maps of the target- vs. self-trials during the item phase were regressed against the individual accuracy scores as a covariate. This analysis revealed that individuals with higher accuracy scores showed greater activity in the DMPFC (*x* = 18, *y* = 50, *z* = 40, *Z* = 3*.*42, corrected, *p <* 0.05, **Figure 3**, **Table 2**) during the evaluation of the items for the targets compared to oneself. This cluster survived even when we controlled for the effects of SP and GP by adding the correlation coefficients between eTP and SP, and eTP and GP as covariates to the same multiple regression model (*x* = 16, *y* = 52, *z* = 42, *Z* = 3*.*42, corrected, *p <* 0*.*05). We found no significant brain regions other than the DMPFC when we expanded the search volume to the whole brain. In addition, the test for the negative association between the individual accuracy scores and the target vs. self contrast did not yield any significant result.

#### **FUNCTIONAL CONNECTIVITY BETWEEN THE DMPFC AND OTHER BRAIN REGIONS**

Considering that the DMPFC is part of the neural network of mentalization along with the other ToM regions, such as the TPJ and the PCC/precuneus (Saxe and Kanwisher, 2003; Frith and Frith, 2006), we expected that the DMPFC would communicate with other structures in the network during the estimations of the targets' preferences. Specifically, we hypothesized that the communication between the DMPFC and the ToM regions would be stronger when the estimations were correct than when they were incorrect. To address this question, we performed a PPI analysis. We defined the DMPFC as a seed region and sought the brain regions that showed stronger functional connectivities with the DMPFC during the correct than the incorrect target-trials. This analysis revealed that the DMPFC (*x* = 18, *y* = 48, *z* = 42, *Z* = 4*.*37, corrected, *p <* 0.05), the MPFC (*x* = −4, *y* = 60, *z* = 2, *Z* = 3.83, corrected, *p <* 0.05), and the PCC/precuneus (*x* = 22, *y* = −56, *z* = 40, *Z* = 3*.*78, corrected, *p <* 0.05) showed stronger functional connectivity with the DMPFC when the participant's estimations for a target's preferences were correct than when they were incorrect. The other brain regions that showed significantly stronger functional connectivity with the DMPFC during the correct vs. the incorrect trials are reported in **Table 2**. No brain regions showed a stronger connectivity with the DMPFC during the incorrect vs. the correct target-trials.

More importantly, we also examined how the individual variations in the accuracy of the preference estimation interact with the functional connectivity between the DMPFC and the ToM regions. When the individual accuracy scores were regressed to the PPI map that was obtained from the procedure



*MNI, Montreal Neurological Institute; R, right; L, left; PPI, psychophysiological interaction; SP, self-preference; aTP, actual target preference; GP, general preference.*

above, significant clusters were found in the PCC (*x* = −2, *y* = −44, *z* = 40, *Z* = 3*.*73, corrected, *p <* 0.05, **Figures 4A,C**) and the right TPJ (*x* = 48, *y* = −60, *z* = 28, *Z* = 3*.*55, corrected, *p <* 0.05, **Figures 4B,D**). In other words, the functional connectivity between these regions and the DMPFC became stronger among the participants with higher accuracy when they guessed correctly than when they guessed incorrectly during the target-trials. No brain region showed a negative association

**FIGURE 3 | Dorsomedial Prefrontal Cortex (DMPFC) activity predicts the accuracy of the estimations of the targets' preferences.** The DMPFC (*x* = 18, *y* = 50, *z* = 40, *Z* = 3.42, corrected, *p <* 0.05) activity that occurred in response to the target- vs. self-trials during the item phase predicted the individual variability in the accuracy of estimating the preferences of the targets. **(A)** Coronal, **(B)** Sagittal, and **(C)** Axial views of the DMPFC. **(D)** Scatter plot of the beta estimates of the DMPFC in the contrast of target- vs. self-trials as a function of accuracy scores.

between the accuracy scores and its connectivity with the DMPFC.

#### **NEURAL CORRELATES OF THE INDIVIDUAL DIFFERENCES IN UTILIZING OTHER SOURCES FOR ESTIMATION**

In order to investigate the brain regions related to estimations based on SP or GP, we performed a multiple regression analysis on the target- vs. self-contrast maps during the item phase with the beta coefficients attained from the regression analysis that regressed the eTP on the aTP, SP, and GP in the behavioral results. This analysis allowed us to find the brain regions that are associated with the extent of the influence of SP and GP on the eTP. We found that the VMPFC (*x* = 20, *y* = 34, *z* = −16, *Z* = 3.77, corrected, *p <* 0.05) and the ventral tegmental area (*x* = −6, *y* = −18, *z* = −22, *Z* = 3*.*76, corrected, *p* = 0*.*06) activities showed significant correlations with the beta coefficients of SP (**Figure 5**). Consistent with the findings on the accuracy scores, the beta coefficients of aTP correlated with the activities in the DMPFC (*x* = 16, *y* = 52, *z* = 40, *Z* = 3.76, corrected, *p <* 0.05), the right inferior frontal gyrus (*x* = 50, *y* = 14, *z* = 18, *Z* = 3.57, corrected, *p <* 0.05), and the left temporal pole (*x* = 50, *y* = 8, *z* = −6, *Z* = 4*.*38, corrected, *p <* 0.05). There were no significant clusters correlated with the beta coefficients of GP.

#### **DISCUSSION**

Humans are highly social animals; the ability to estimate others' preferences in an accurate and reliable manner may be essential for successful social adaptation. In daily interactions, we often do this on the basis of minimal information. The present study demonstrated that people can estimate the preferences of others based on briefly presented subtle and non-communicative facial appearances. Participants in the present study were asked to make guesses about unfamiliar target individuals' preferences for various items after looking at their faces for 3 s. The overall accuracy of the estimations was significantly above chance level. Importantly, this remained significant even after controlling for the participants' own preferences and their beliefs about the preferences of the general population. The fMRI data revealed

**FIGURE 4 | The strength of the functional connectivity with the DMPFC associated with higher accuracy in estimating the preferences of the targets.** The psychophysiological interaction with DMPFC activity during correct vs. incorrect estimation trials that increased as a function of the individual estimation accuracy scores were observed in posterior cingulate cortex [PCC; **(A)**: *x* = −2, *y* = −44, *z* = 40, *Z* = 3*.*73, corrected, *p <* 0.05] and the right temporoparietal junction [TPJ; **(B)**: *x* = 48, *y* = −60, *z* = 28, *Z* = 3*.*55, corrected, *p <* 0.05]. Scatter plots of the connectivity strength between the DMPFC and **(C)** the PCC, and **(D)** the right TPJ as a function of the individual estimation accuracy scores.

that higher accuracy in the preference estimations was associated with greater activity in the DMPFC when the participants guessed the targets' preferences relative to their own. This result indicate that the accurate estimation of others' preferences may require increased activity in the DMPFC. In addition, those with higher accuracy in estimating the preferences of others showed increased functional connectivity between the DMPFC and a network of ToM regions, such as the TPJ and PCC/precuneus, particularly when their estimations were correct rather than incorrect. In summary, the present study provided the first evidence that DMPFC activity may be critically related to success in estimating others' preferences and that higher accuracy may require stronger communication between the DMPFC and a network of neural structures, including the TPJ and the PCC/precuneus, which are now widely known to be involved in taking another person's perspective during mentalizing.

#### **ROLE OF THE DMPFC IN ESTIMATING OTHERS' PREFERENCES**

From both evolutionary and ontogenetic perspectives, social environments must have forced humans to develop a neural system that is specialized for estimating others' preferences. Such a system seems to require a change in mental mode, or perspective, which may critically determine the successful and accurate estimation of others' preferences. Yet, only a few neuroimaging studies have investigated the neural mechanisms underlying the accuracy of inferring personal traits from facial appearances, which has been reported to involve the amygdala (Rule et al., 2010, 2011) and the insula (Spezio et al., 2008). Unlike these studies, however, we did not find any association between these structures and the accuracy of estimating others' preferences. One possible explanation for the gap between the findings of the previous studies and the present study might be that, compared to the inference of personal traits, estimating another person's preferences may require higher-level cognitive processes, such as perspective taking and mentalization, which involve activity in the DMPFC rather than other subcortical regions. In addition, functional and anatomical evidence seem to indicate strong reciprocal connections between the DMPFC and the structures listed above (Amaral and Price, 1984; Augustine, 1996; Kim et al., 2011), suggesting that the DMPFC might be a key center in integrating signals that carry information from those subcortical structures.

**FIGURE 5 | Ventromedial Prefrontal Cortex (VMPFC) and Ventral Tegmental Area (VTA) activity associated with the impact of self-preference on the estimated target preference. (A)** The VMPFC (*x* = 20, *y* = 34, *z* = −16, *Z* = 3.77, corrected, *p <* 0.05) and **(B)** the VTA (*x* = −6, *y* = −18, *z* = −22, *Z* = 3.76, corrected, *p* = 0*.*06) activities during the target vs. self conditions showed positive correlations with the individual variabilities in the degree of the impact of self-preference (SP) on the estimated target preference (eTP). **(C)** Scatter plot of the beta estimates of the VMPFC in the contrast of the self- vs. target-trials as a function of the impact of SP on eTP.

The DMPFC has been considered a component of a global mentalization network (Mitchell et al., 2005a; Amodio and Frith, 2006; Frith and Frith, 2006; Mitchell, 2006; Lieberman, 2007; Schiller et al., 2009; Jenkins and Mitchell, 2010; Muscatell et al., 2012). Given that inferring information about another person involves the ToM and mentalizing ability (Gore and Sadler-Smith, 2011), it may be natural to reason that the DMPFC would play a key role in preference estimation. The DMPFC has also been implicated in various aspects of social behavior, particularly interpersonal judgments, such as forming impressions of other people or predicting the outcomes of future relationships (Walter et al., 2004; Mitchell et al., 2005b; Jenkins and Mitchell, 2010; Cooper et al., 2012), as mentioned in the Introduction, and this is more relevant to the present study. In addition, the DMPFC seems to be important for assessing the value of risky choices, especially for another person (Jung et al., 2013). Similar to the present study, Jung et al. (2013) have observed that risky decisions for others vs. oneself are related to stronger functional connectivity between the DMPFC and the structures known to be involved in mentalization, such as the TPJ and the PCC. These findings suggest that the DMPFC may be more sensitive to social evaluations rather than to one's own value assessments, as has been proposed by Cooper et al. (2010, 2012), and that the role of the DMPFC in estimating the values of another person's choices may be dependent upon integrated signals that come from a network of neural structures specialized for mentalization.

## **THE DMPFC AS A CORE COGNITIVE SYSTEM**

The successful estimation of others' preferences often requires cognitive control that inhibits the self-projection of one's own state (Hoch, 1988; West, 1996). That is, in order to correctly guess the preferences of others, one needs to inhibit his/her own opinion that may influence the estimation. According to a recent theoretical framework about the neural mechanisms in an attentional cognitive task, the DMPFC can be considered a core system for monitoring and modulating other attentional submodules, such as the TPJ, which are involved in stimulusdriven shifts in attention (Dosenbach et al., 2006; Corbetta et al., 2008). This theory suggests that the TPJ, a core part of the ventral attention system, acts like an efficient steering system that reorients attention from a current focus to information that is more relevant to the goal. Perhaps, the self-projection of one's own preference is a highly automatic process, and it may often be difficult to override this process, even during the estimation of others' preferences. Thus, the ventral attention system needs to be engaged to reallocate attention to more relevant external sources of information such as the appearances of others. Consistent with this hypothesis, it has been recently observed that the activation level of the mentalization network, including the right TPJ, reflects the accuracy of interpersonal inferences, based on visual information from faces, in estimating leadership competency (Rule et al., 2011) and the affective states of others (Zaki et al., 2009).

It is still debated whether modules for the ToM and for attention are colocalized or are segregated within the TPJ (Mitchell, 2008; Scholz et al., 2009). Yet, it is tempting to speculate that, when evaluating an item for another person, momentary fluctuations in the TPJ activity signaling are required to shift attention and communicate with the DMPFC in order to consider the person's facial appearance, which is the information more relevant to the task goal. As can be shown by the present findings, communication between the TPJ and the DMPFC may be critical for successful value estimations from the perspective of others. This argument is further corroborated by the fact that the strength of the functional connectivity between the DMPFC and the TPJ was stronger among participants with a higher preference estimation accuracy for correct vs. incorrect trials in the present study. It remains to be answered in future studies whether a similar network can be recruited, even when the perspective-taking aspect of the present task is substituted by a purely non-social component.

## **THE DMPFC AND MODELED CHOICES vs. CHOICES FOR OTHERS**

The view of the DMPFC functioning in making choices for others has been challenged by a recent study in which DMPFC activity reflects modeled vs. executed choices rather than other vs. self choices (Nicolle et al., 2012). Despite the significance of this finding in expanding our view of the role of DMPFC in mentalization, it is important to note that the participants in the study had prior knowledge about the choices of the partners through extensive practice and, thus, the task used in the study did not require active inferences of the partners' preferences. Given that uncertainty is an inevitable key component of estimating the choices of others, the DMPFC appears to have a privileged role in inferring the preferences of others (Jenkins and Mitchell, 2010; Cooper et al., 2012; Jung et al., 2013), at least before we become fully familiar with the preferences of others. It is important to examine whether the role of the DMPFC in modeling choices changes as a function of learning the preferences of others.

## **THE ROLE OF THE VMPFC IN THE ESTIMATION OF OTHERS' PREFERENCE**

One possible way that one's estimation goes awry from the actual preferences of others may be the application of one's own preferences to the estimation process. This type of self-projection of one's own preferences appears to be a highly automatic process that is often difficult to override during the estimation of others' preferences. Interestingly, we found a large range of individual variability in the degree of self-projection during the estimation of other's preference (i.e., the beta coefficients of the SP accounting for eTP in the multiple regression analysis), and this type of individual variability was significantly predicted by the activation level of the VMPFC during the target vs. self conditions. In other words, those whose VMPFC activity increased during the estimation of others' preferences tended to project their own preferences onto the others, which could have then resulted in inaccurate estimations. A large body of literature now supports the primary role of the VMPFC in encoding subjective values critical for one's own decisions (Kim et al., 2006; Kable and Glimcher, 2007; Kim et al., 2007; Chib et al., 2009). Combining these findings with our previous account for the role of the DMPFC, it can be concluded that, for accurate and successful estimations of others' preferences, the DMPFC and TPJ need to work together and be engaged in order to control the activity of VMPFC and inhibit the intrusion of one's own preference and to reallocate the attention to more relevant external sources of information such as the appearances of others.

## **PREFERENCE ESTIMATION AND IMPLICIT INFERENCE ABOUT PERSONALITY**

What particular information from faces do people utilize for the successful estimation of others' preference? One may easily come up with a hypothesis that the preference estimation task used in the current study might be considered an applied version of the inference of the target's personality. Although the present study might require some degree of inferences about personality, the estimations about the target's preferences might not be based solely on explicit and effortful inferences about personality, especially given that the time for estimation was not long enough (∼3 s) for any conscious and deliberate inferences about the target's personality. Consistent with this argument, during the debriefing, no participants reported that they tried to apply the target's inferred personality to the estimation of the target's preferences. Therefore, although it is not clear at this point what particular information from the faces the perceivers used for target-preference estimation, this type of estimation process might have been influenced by personality traits that were inferred, perhaps at an implicit level, just as some personality traits, such as extroversion and conscientiousness, can be quickly and accurately read from faces (Carney et al., 2007). These issues will need to be resolved further in future studies.

## **CONCLUSIONS**

The present study examined the role of the DMPFC in estimating the preferences of strangers. Consistent with the existing literature on thin slicing, the participants were able to estimate the preferences of strangers significantly above chance level, even with brief presentations of non-communicative facial appearances. Importantly, the activity in the DMPFC and close communication with the ToM and the mentalization network in the brain was found to be associated with the accuracy of the estimation. The present findings add to the literature in the rapidly growing field of decision neuroscience by providing unequivocal neural evidence for thin-slice judgments and social perception. Future studies that focus on the mechanisms underlying the individual differences in the accuracy of estimating others' preferences will also lead to fruitful outcomes in both industrial and clinical applications.

## **ACKNOWLEDGMENTS**

This work was supported by the National Research Foundation of Korea Grant funded by the Korean Government (NRF-2012- S1A3-A2033375). We thank Hee Jung Jung, Min-Woo Lee, and M. Justin Kim for their comments on the early version of the manuscript.

#### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at: http://www.frontiersin.org/journal/10.3389/ fnhum.2013.00686/abstract

### **REFERENCES**


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 01 May 2013; accepted: 29 October 2013; published online: 26 November 2013.*

*Citation: Kang P, Lee J, Sul S and Kim H (2013) Dorsomedial prefrontal cortex activity predicts the accuracy in estimating others' preferences. Front. Hum. Neurosci. 7:686. doi: 10.3389/fnhum.2013.00686*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Kang, Lee, Sul and Kim. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Investment and repayment in a trust game after ventromedial prefrontal damage

# *Giovanna Moretto, Manuela Sellitto and Giuseppe di Pellegrino\**

*Department of Psychology, Centre for Studies and Research in Cognitive Neuroscience, University of Bologna, Cesena, FC, Italy*

#### *Edited by:*

*Corrado Corradi-Dell'Acqua, University of Geneva, Switzerland*

#### *Reviewed by:*

*Jack Van Honk, Utrecht University, Netherlands Claudia Civai, University of Minnesota, USA*

#### *\*Correspondence:*

*Giuseppe di Pellegrino, Centre for Studies and Research in Cognitive Neuroscience, University of Bologna, Viale Europa 980, 47521 Cesena, FC, Italy*

*e-mail: g.dipellegrino@unibo.it*

Although trust and reciprocity are ubiquitous in social exchange, their neurobiological substrate remains largely unknown. Here, we investigated the effect of damage to the ventromedial prefrontal cortex (vmPFC)—a brain region critical for valuing social information—on individuals' decisions in a trust game and in a risk game. In the trust game, one player, the investor, is endowed with a sum of money, which she can keep or invest. The amount she decides to invest is tripled and sent to the other player, the trustee, who then decides what fraction to return to the investor. In separate runs, ten patients with focal bilateral damage to the vmPFC and control participants made decision while playing in the role of either investor or trustee with different anonymous counterparts in each run. A risk game was also included in which the investor faced exactly the same decisions as in the trust game, but a random device (i.e., a computer), not another player, determined the final payoffs. Results showed that vmPFC patients' investments were not modulated by the type of opponent player (e.g., human vs. computer) present in the environment. Thus, vmPFC patients showed comparable risktaking preferences both in social (trust game) and nonsocial (risk game) contexts. In stark contrast, control participants were less willing to take risk and invest when they believed that they were interacting with people than a computer. Furthermore, when acted as trustee, vmPFC patients made lower back transfers toward investors, thereby showing less reciprocity behavior. Taken together, these results indicate that social valuation and emotion subserved by vmPFC have a critical role in trusting and reciprocity decisions. The present findings support the hypothesis that vmPFC damage may impair affective systems specifically designed for mediating social transaction with other individuals.

**Keywords: trust, risk, reciprocity, social valuation, vmPFC, lesion studies**

## **INTRODUCTION**

Trust is an essential ingredient of human exchange (Arrow, 1974); it lubricates social and economic transactions, and has been long recognized as a critical antecedent of cooperative behavior (Ostrom and Walker, 2003). Trust can be defined as one's willingness to place resources at disposal of another party in situations in which there is uncertainty regarding the other party's motive, intentions and actions (Mayer et al., 1995; Rousseau et al., 1998). An action that is trusting of another is one that creates the possibility of mutual benefit, if the other person is cooperative. Yet trusting behaviors also imply the risk of injury or loss to oneself if the other person defects. Overriding aversion to such risks is required for trust to emerge (Kosfeld et al., 2005).

Although theoretical work has identified a number of factors likely to influence trust (Mayer et al., 1995; Lewicki and Wiethoff, 2000), fundamental questions remain about how trust actually operates. For instance, while a commonly held view suggests that trust is a result of rational calculation and higher cognitive processes (Coleman, 1990), in some accounts trust is held to be founded in emotional processes (Hardin, 2002; Butler et al., 2003). Consistent with this latter account, behavioral studies suggest that incidental emotions significantly influence social exchange and trust (Dunn and Schweitzer, 2005). Moreover, several neuroimaging studies have shown that tasks that require social valuation (Winston et al., 2002; Somerville et al., 2006), or cooperation with another individual (McCabe et al., 2001; Gallagher et al., 2002; Rilling et al., 2002, 2004; Tomlin et al., 2006) activate brain regions known to process emotions, including the anterior cingulate cortex and adjacent medial frontal cortex. Importantly, when subjects interact with partners they know to be just computers, these activations are not seen, suggesting that they reflect the interpersonal nature of the task (McCabe et al., 2001; Rilling et al., 2004; Tomlin et al., 2006; van den Bos et al., 2007). Neuroimaging studies, however, do not settle whether a given brain region is necessary for a particular behavior. This issue could be addressed by studying human subjects with focal brain damage. Remarkably, however, only few studies provided causal evidence linking brain areas integral to emotional processes to trusting behavior (van Honk et al., 2012).

Here, we examined whether emotions, specifically social emotions subserved by the ventromedial prefrontal cortex (vmPFC), affect people's willingness to trust others. Several evidence suggests this possibility. First, the vmPFC is densely interconnected with basolateral amygdala, ventral striatum, and subcortical structures that control autonomic and visceral responses (Carmichael and Price, 1995; Haber et al., 2006), and is therefore ideally located for generating emotional responses, and guiding social interactions. Second, neuroimaging studies in humans have implicated the vmPFC in guiding behavioral choice under uncertainty (Hsu et al., 2005; De Martino et al., 2006), and have argued that this region is critical for balancing potential gains against losses to ensure optimal decision-making in social context (De Quervain et al., 2004). Finally, damage to the vmPFC in humans can be associated with strikingly poor judgment and decision-making (Eslinger and Damasio, 1985; Bechara et al., 1994, 1997; Koenigs et al., 2007), due to markedly reduced (Ciaramelli et al., 2007; Koenigs et al., 2007; Krajbich et al., 2009; Moretto et al., 2009), or poorly regulated (Koenigs and Tranel, 2007) emotions.

To address whether the vmPFC plays a necessary role in the decision to trust a stranger, a sample of patients with adult-onset vmPFC lesions, as well as healthy control subjects (HC) and patients with lesions outside the frontal lobe (non-FC patients), played in the role of investor in a one-round trust/investment game (Berg et al., 1995). This game involves real monetary exchanges between two anonymous individuals, the investor and the trustee, who receive each a sum of money from the experimenter. The investor can keep all the money or decide to invest some amount, which is tripled by the experimenter and sent to the trustee. Next, the trustee decides how much of the tripled amount to return. Money sent by the investor is used to measure her trust, while money returned by the trustee is used to measure her trustworthiness.

Clearly, the decision to trust entails a risk (Rousseau et al., 1998). Uncertainty regarding whether the trustee intends to and will honor the investor's trust is the source of risk. This raises the important concern over whether a person's attitude toward general risk influences trust (Eckel and Wilson, 2004; Karlan, 2005; Schechter, 2007). To control for between-group differences in risk attitudes, we therefore also implemented a risk game offering the same options and payoffs as the trust game, but in which a random device (e.g., a computer), not a human partner, determined the investor's risk. The risk game constitutes a critical control condition because recent behavioral (Bohnet and Zeckhauser, 2004; Hong and Bohnet, 2007; Bohnet et al., 2008; Houser et al., 2009) and neurobiological (Kosfeld et al., 2005; Baumgartner et al., 2008) evidence strongly indicates that the decision to trust is not only determined by risk aversion (i.e., the negative emotion associated with the possibility of losing objects or money) but also by betrayal aversion, that is, the fear to be betrayed by another in social exchange. Betrayal aversion plays no role in the risk game, since random devices are incapable of intentionality or awareness, and they cannot really betray our trust. Therefore, the contrast between trust game and risk game is ideal to assess whether vmPFC damage specifically affects trusting behavior in social exchanges (rather than risk-taking behavior in general), because—except for the type of opponent partner (human vs. computerized partner)—everything else remains constant across these two games. Based on previous findings showing that regions in the vmPFC may be critical for valuing social information (Amodio and Frith, 2006), particularly when the implications of another individual's intentions must be taken into account before acting (Rudebeck et al., 2008; Behrens et al., 2009; Moretti et al., 2009; Ciaramelli et al., 2013), we hypothesized that investors in the vmPFC-lesioned group would show higher money transfers than those in the control groups, especially in the trust game in which both social and non-social risks operate to inhibit trust.

Several researchers (Andreoni and Miller, 2002; Cox, 2004) have argued that measures of trust taken from the trust game do not discriminate between actions motivated by trust and actions motivated by altruism or generosity. To address this question, we measured the amount of money participants returned when they played the role of trustee in a separate session. If lesion to the vmPFC increases generosity rather than trusting behavior, then one might hypothesize that a player will send more as investor and return more as trustee, thus appearing both more trusting and trustworthy.

Finally, we included a measure of the investor's subjective expectation about the trustee's back transfer at different investment levels. This in order to control whether vmPFC patients apparently trust more because they are more optimistic about the trustee's trustworthiness (e.g., they have higher expected back transfers).

#### **MATERIALS AND METHODS**

#### **PARTICIPANTS**

Three groups of subjects participated in the study: (a) a group of patients with focal lesions involving the vmPFC (the vmPFC group, *n* = 10), (b) a control group of patients with damage sparing the frontal cortex (the non-FC group, *n* = 10), and (c) a control group of healthy subjects (the HC group, *n* = 10), who were matched on age, education and sex with the vmPFC group. Brain-damaged patients were recruited from the Centre for Studies and Research in Cognitive Neuroscience in Cesena. They were selected on the basis of the location of their lesion evident on computerized tomography (CT) or magnetic resonance imaging (MRI) scans.

**Table 1** shows demographic and clinical data, as well as the Mini-Mental Status Examination score (MMSE; Folstein et al., 1983). There were no significant differences between vmPFC


*MMSE* = *Mini-Mental State Examination.*

patients and comparison groups with regard to age, education, and clinical variables (*p* > .05 in all cases).

In the vmPFC group, lesions principally involved the vmPFC, which is defined as the medial one-third of the orbital surface and the ventral one-third of the medial surface of the frontal lobe, following the boundaries laid out by Stuss and Levine (2002). Lesion etiology was hemorrhage due to ruptured aneurysm of the anterior communicating artery in 9 out of 10 vmPFC patients, and to traumatic brain injury in one. The vmPFC damage was bilateral (although often asymmetrically so) in six cases, right unilateral in two cases, and left unilateral in two cases. All vmPFC patients presented with clinical evidence of a decline in social interpersonal conduct, impaired decision-making and emotional functioning, but had generally intact intellectual abilities (see **Table 2**).

The non-FC patients were selected on the basis of having damage that did not involve the mesial orbital/vmPFC and frontal pole, and also spared the amygdala in both hemispheres. In this group, lesions were unilateral in nine patients (in the left hemisphere in five cases, and in the right hemisphere in four cases) and bilateral in one patient, and were caused by ischemic or hemorrhage stroke in nine cases, and by traumatic brain injury in one case. In the non-FC group, lesion sites included the lateral aspect of the temporal lobe in six patients, the lateral occipital area in two patients, and the occipito-parietal junction in the remaining two patients.

Normal participants were healthy volunteers who were not taking psychoactive medication, and were free of current or past psychiatric or neurological illness as determined by history.

All subject groups were administered a short neuropsychological battery including tests with potential sensitivity to frontal damage, as well as intelligence and memory tests (results are provided in **Table 2**). The groups differed significantly only in their performance on the Stroop task, with vmPFC subjects making more errors than both non-FC patients and HCs (Mann– Whitney U-test, *p* < .05). Patients were not receiving psychoactive drugs at the time of testing, and had no other diagnosis likely to affect cognition or interfere with participation in the study (e.g., significant psychiatric disease, alcohol misuse, history of cerebrovascular disease, focal neurological examination). Neuropsychological and experimental studies were all conducted in the chronic phase of recovery, more than a year post-onset. All lesions were acquired in adulthood. Patients gave informed consent to participate in the study according to the Declaration of Helsinki (International Committee of Medical Journal Editors, 1991) and the Ethical Committee of the Department of Psychology, University of Bologna.

#### **LESION ANALYSIS**

Lesion analysis was based on the most recent clinical CT or MRI. The location and extent of each lesion were mapped by using MRIcro software (Rorden and Brett, 2000). The lesions were manually drawn by a neurologist with experience in image analysis onto standard brain template from the Montreal Neurological Institute (MNI), which is based on T1-weighted MRI scans, normalized to Talairach space. This scan is distributed with SPM99 and has become a popular template for normalization in functional brain imaging. For superimposing of the individual brain lesions, the same MRIcro software was used. **Figure 1** shows the extent and overlap of the brain lesions in the brain-damaged patients. Brodmann's areas (BA) affected in vmPFC group were areas 10, 11, 12, 32 (subgenual portion), and 24, with region of maximal overlap occurring in BA 10 and 11.

## **EXPERIMENTAL DESIGN AND PROCEDURES**

Every participant in the experiment played the role of investor in two treatment conditions: a trust game and a risk game. In the trust game, the subject played a standard trust game and she knew her counterpart was human; we call this the human interaction treatment. In the risk game, the subject knew her counterpart was a computer making random decisions; we call this the computer interaction treatment. Trust and risk games were played in separate sessions with an interval of at least 1 week between them. Half of the participants in each group played the trust game in the first session, and half the risk game in the first session.

All experiments took place in a quiet room in which an opaque, removable partition wall was used to create two separate settings. On either side of the wall, we placed a desk with a computer. Participants sat at one desk in front of the computer, while at the other desk sat either an actor who played in the role of the trustee (trust game), or no one (risk game). As a result, playing partners could be separated visually, thereby providing betweensubject anonymity, without separating them audibly, thus lending our set-up credibility. Before each session, instructions about the nature and rules of the game were presented on the computer, and the experimenter verbalized them to ensure that participants understood them. In the instructions, it was emphasized that participants in the trust game would play the game anonymously and only once with each opponent player, and that they would receive the money earned in the game. Differently, in the risk game it was emphasized that participants would play with a computer counterpart. After reading the instructions, subjects were required to complete a quiz that required them to state the amount of money that each player would receive under various



*SRM* = *Standard Raven Matrices (scores in percentile values); ITS* = *Interpersonal Trust Scale; PNR* = *Personal norm of Reciprocity scale.*

**FIGURE 1 | Location and overlap of brain lesions.** The panel shows the lesions of the 10 patients with vmPFC damage superimposed on the same seven axial slices and on the mesial view of the standard MNI brain. The level of the axial slices has been marked by white

horizontal lines on the mesial view of the brain. z-coordinates of each axial slice are given. The color bar indicates the number of overlapping lesions. In each axial slice, the left hemisphere is on the left side.

hypothetical circumstances. The game started once the subject successfully finished the quiz.

Subjects in the role of the investor received no feedback about their partner's decision between the different interactions. At the end of each session, the experimenter put the cash payoff earned by subject during the game into an opaque envelope that was sealed and signed by the participant. Earnings envelops were kept by the experimenter between games. Subjects did not receive feedback about the outcome of any game until the end of the experiment in order to avoid income effects and the possibility that current decisions were influenced by an opponent's previous decisions. All games were paid out at the end.

#### *Human interaction treatment*

Participants acted as investor in a series of nine rounds of a trust game against nine different anonymous human partners via a computer interface. At the beginning of each round, the actor that played the role of the trustee entered the room and sat at her position. When both investor and trustee were ready, the interaction started. Each round was presented as text through a series of five screens. A 6-s initial screen depicted a silhouette of a human figure and indicated the endowment (E) available for both players in the current round. There were three equiprobable initial E, e6, e9 and e12, presented in random order during the game. The second screen posed the question "How many Euros between 0 and E do you transfer to Participant B?" and remained visible until a response was given. Participants were given the opportunity to send any integer amount from zero to their entire endowment available, and were instructed to indicate their decision by pressing the numeric keys of the computer keyboard. Following the response, a screen indicating the investor's transfer and the amount received by the trustee (three times the amount invested) was presented for 4 s. Then, a variable 5- to 15-s waiting screen informed that the trustee (Participant B) was deciding how much of the tripled amount to send back. Subjects were informed that Participant B could choose the amount from any integer between zero and the tripled amount they have transferred to her/him. Finally, a screen signaled the end of the round. The trustee went out of the room and after a short break was replaced by another actor to begin the next round. When the trustee was out of the room, the investor was asked about her expectation about the trustee's back transfer.

#### *Computer interaction treatment*

Participants were instructed that they would play nine rounds of a risk game in which a random mechanism determined the outcome of the game. In the risk game, everything was identical to the trust game, except that subjects played against a computerized partner. A silhouette of a computer was displayed in the initial screen to indicate the computer interaction. Participants were informed that, in each round, the computer would randomly choose the amount to transfer back from any number between zero and the tripled amount they have transferred to it.

In a separate session, participants played five rounds of a trust game in the role of trustee against five different anonymous investors via a computer interface. The experimental setup was as before, except that participants were assigned the role of trustee (Participant B), and an endowment of e9 was available for both players in every round. Each new round began with a 6-s initial screen that depicted a silhouette of a human figure and indicated that e9 were available for both players in the current round. Then, a variable 5- to 15-s waiting screen informed that the investor (Participant A) was deciding how much between e0 and e9 to transfer to the trustee (Participant B). Next, a screen indicating the investor's transfer and the amount received by the trustee was presented for 4 s. The investor's transfers, X, were predetermined and presented randomly, and included one transfer of each e0, e3, e5, e7 and e9, so that the trustee received e0, e9, e15, e21 and e27, respectively. Then, the question "How many Euros between 0 and 3X do you transfer back to Participant A?" appeared on the screen and remained visible until a response was given. Participants were given the opportunity to send back any integer amount from zero to the tripled amount received, and were instructed to indicate their decision by pressing the numeric keys of the computer keyboard. Following the response, a screen signaled the end of the round. The trustee went out of the room and after a short break was replaced by another actor to begin the next round. Note that participants in all groups faced exactly the same set of investors' transfers. Thus, behavioral differences across these three groups cannot be attributed to differences in the distribution of investors' transfers.

## *Questionnaires*

Approximately 2 weeks after the experiment, participants also completed three self-report questionnaires that assessed selected personality traits. The Personal Norm of Reciprocity (PNR) scale is a 27-item questionnaire measuring three dimensions (nine items each) of reciprocity (i.e., the propensity to reward those who have behaved nicely and punish those who behaved badly): positive reciprocity, negative reciprocity, and beliefs in reciprocity (Perugini et al., 2003); the Interpersonal Trust Scale (ITS) includes 25 component questions requiring subjects to express their trust expectations across a variety of social situations and with diverse social agents (Rotter, 1967).

## **RESULTS**

**Figure 2** illustrates investors' average transfer as a function of initial endowment, separately for the trust and risk game. We performed a mixed design ANOVA on transfer amounts with Group (vmPFC, non-FC, and HC) as a between-subjects factor, and Treatment (human, and computer), and Endowment (e6, e9, and e12) as within-subjects factors. When necessary, pairwise comparisons were conducted using the Fisher LSD test, which is considered the most powerful technique for post hoc tests involving three groups (Cardinal and Aitken, 2006). Analysis showed a

significant main effect of Group, *F*(2, 27) = 9.62, *p* < .001, η<sup>2</sup> *p* = .42, revealing that investors in the vmPFC group had overall significantly higher transfer levels (e5.7 out of a mean endowment of e9) than had investors in the HC (e4.3) and non-FC group (e4.2; both *ps* < .001).There was also a significant main effect of Treatment, *<sup>F</sup>*(1, 27) <sup>=</sup> 7.56, *<sup>p</sup>* < .01, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = .22, indicating slightly higher transfers in the computer (e5) than in the human (e4.5) interaction, and a significant main effect of Endowment, *F*(2, 54) = 100.14, *p* < .001, η<sup>2</sup> *<sup>p</sup>* = .79, demonstrating that investors' transfer was modulated by initial endowment available.

More critically, analysis showed a significant Treatment by Group interaction, *<sup>F</sup>*(2, 27) <sup>=</sup> 4.92, *<sup>p</sup>* < .02, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = .27, indicating that the between-group differences in amount sent depended on the human *vs.* computer interaction. Pairwise comparisons showed that when participants played against a human partner, average transfer was significantly higher in the vmPFC group (e5.8) than in both non-FC (e3.7) and HC group (e3.9; both

*ps* < .05), while transfers of the control groups did not differ (*p* > .05). By contrast, when participants played against a computerized partner, there was no significant difference between investors' transfer across groups (vmPFC: e5.5; non-FC: e4.6, HC: e4.7; all *ps* > .05). The identical pattern of results was found when the data were analyzed using nonparametric methods. The Kruskal-Wallis test showed a significant difference amongst the three groups in the trust game (*H* = 12.8, *df* = 2, *p* < .002), but no difference in the risk game (*H* = 4.78, *df* = 2, *p* = .09). Indeed, out of 10 subjects in each group, eight vmPFC patients showed mean transfer levels higher than 50% of initial endowment in the trust game, whereas only three non-FC patients, and four HC displayed such transfers in the trust game. Conversely, in the risk game, nine vmPFC patients, seven non-FC patients, and seven HC displayed mean transfers higher than 50% of initial amount.

The above results suggest that, while control participants decreased their trust level when playing against a human partner as compared to a non-human partner, vmPFC patients failed to modulate their trust based on the recipient of their choices. Thus, damage to vmPFC would lead to an apparent increase in transfer levels in the trust experiment but not in the risk experiment. Accordingly, investors' transfers in the vmPFC group were not modulated at all by the type of opponent player present in the environment (e5.82 and e5.53, for the trust and risk game, respectively, *p* > .05). In sharp contrast**,** both control participants were more reluctant to invest in the trust game (e3.71 and e3.88, for non-FC and HC group, respectively), in which interpersonal interactions determines the risk, than in the risk game (e4.69 and e4.74; *p* < .05, and *p* = .01, for non-FC and HC group, respectively), in which a non-social, random mechanism constitutes the risk. This latter result is highly consistent with previous literature in healthy subjects (see De Quervain et al., 2004; Bohnet et al., 2008; Aimone and Houser, 2009; Houser et al., 2009) suggesting that the prospect for betrayal plays a role in trusting decisions well beyond aversion towards monetary loss.

Next, we performed an analysis to explore whether vmPFC patients differed from control groups in their subjective expectations about trustee back transfers in the trust game. To this end, a mixed design ANOVA, with Group (vmPFC, non-FC, and HC) as a between-subjects factor, and Endowment (e6, e9, and e12) as a within-subjects factor, was conducted on expected back transfers divided by the amount sent (a value > 1 indicates expected gain, whereas a value < 1 indicates expected loss from the exchange). Results revealed a significant main effect of Endowment, *<sup>F</sup>*(2, 54) <sup>=</sup> 6.70, *<sup>p</sup>* < .003, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = .20. More importantly, however, there was no main effect of Group (*F* = 1.42, *p* = .26), nor any interaction between Group and Endowment (*F* = 1.25, *p* = .30), revealing that the three groups of participants believed to obtain on average the same return for their money transferred as investor. Thus, results suggest that the apparent increase in trusting behavior in vmPFC-damaged participants does not depend on subjects' beliefs about others' trustworthiness, which was not significantly altered.

We next tested whether trustees' repayments to their investor in the trust game differed across the three groups of participants (**Figure 3**). A one-way ANOVA on trustees' average back transfers showed a marginally significant effect of Group, *F*(2, 27) = 3.20, *<sup>p</sup>* <sup>=</sup> .06, <sup>η</sup><sup>2</sup> *<sup>p</sup>* = .20. Pairwise comparisons revealed that vmPFC trustees made significantly lower back transfers than HC trustees (mean back transfer: e4.10 and e5.72, for the vmPFC and HC group, respectively, *p* = .02). The non-FC group (mean back transfer: e4.97) was not significantly different from the vmPFC or HC groups (both *ps* > .05), possibly due to higher variance in performance observed in this group. Thus, results indicate that individuals with vmPFC damage do not show more trustworthy or altruistic behavior than control groups.

#### **PERSONALITY QUESTIONNAIRES**

**Table 2** shows self-report measures of impulsivity, trust, and reciprocity for all three groups of subjects. There were no statistical differences across the three experimental groups on ITS scores (Kruskal-Wallis test, *H* = 4.07, *df* = 2, *p* = .09). Likewise, we found no significant difference amongst the three groups in positive reciprocity scores (Kruskal-Wallis test, *H* = 2.09, *df* = 2, *p* = .35), negative reciprocity scores (Kruskal-Wallis test, *H* = .83, *df* = 2, *p* = .65), and beliefs of reciprocity scores (Kruskal-Wallis test, *H* = .75, *df* = 2, *p* = .69) of the PNR scale.

## **DISCUSSION**

We show that, following vmPFC damage, economic investments are not modulated by the type of opponent player (e.g., human vs. random mechanism) present in the environment. That is, patients with lesions in the vmPFC showed comparable risk-taking preferences both in social (trust game) and private (risk game) contexts. In stark contrast, control participants were less willing to take risk and invest when they believed that they were interacting with people than a computer (Bohnet and Zeckhauser, 2004). Thus, vmPFC patients invested significantly more than control subjects in the trust game, whilst no difference was observed in the risk game.

These abnormal economic investments were not a general effect of brain damage, because control patients' behavior was comparable to that of healthy individuals in the trust game, but rather were caused by lesion in a specific prefrontal region vmPFC. Furthermore, the investment of vmPFC-damaged patients in trustees cannot be simply attributed to difference in intellectual, executive or memory abilities, because performance at several neuropsychological tests was similar for vmPFC patients and control participants.

Several mechanisms involved in trusting behavior might be disrupted following damage of vmPFC. One possibility is that vmPFC damage causes a general increase in altruism and prosocial inclinations. On this account, vmPFC damage should affect not only the prosocial behavior of the investors but also that of the trustees. However, the data concerning the trustees' repayments to their investors in the trust game failed to show more trustworthy or altruistic behavior in the vmPFC group than control groups. On the contrary, data showed reduced generosity in the trustees' repayment in the vmPFC than in the control groups, thereby indicating that effect of vmPFC damage on trust is not caused by increased generosity or inclination to behave prosocially. This finding is completely consistent with a recent neuropsychological study (Krajbich et al., 2009) demonstrating that vmPFC damage significantly reduces trustworthiness, possibly due to impaired sense of guilt, a sociomoral emotion that plays a critical role also in moral decisions (Ciaramelli et al., 2007; Koenigs et al., 2007; Moretto et al., 2009).

Another possible mechanism behind the effect of vmPFC on trust is that damage to this region alters patients' subjective expectations or beliefs about others' trustworthiness or positive reciprocity. In other words, lesion to the vmPFC may render patients more optimistic about the probability of a good return from the investment. However, results showed that these expectations do not differ significantly between vmPFC and control groups, therefore ruling out the possibility that vmPFC patients show more trusting behavior because of unusual or rose-colored beliefs about other players' repayments. Furthermore, also self-report measures of trust (Rotter, 1967), and reciprocity (PNR, Perugini et al., 2003), indicate that vmPFC patients and control groups hold similar beliefs about others' trustworthiness and reciprocity. That is, when vmPFC subjects are involved in abstract questions concerning their level of trust or reciprocity they are able to answer not differently from controls groups. This finding is perfectly coherent with results from several other studies (Koenigs et al., 2007; Krajbich et al., 2009; Moretti et al., 2009) showing that an explicit knowledge of social rules, as well as expectations and beliefs are intact and normally accessible following vmPFC damage. Despite this retained knowledge, however, vmPFC patients fail in valuing social information in social interaction and decision-making (Damasio, 1994).

As indicated at the outset, a critical finding of this study emerges when comparing mean investors' transfer in the trust and the risk games across the three groups of participants. We found, that following vmPFC damage, patients showed higher and similar investments in both games. That is, vmPFC patients did not distinguish between interactions with an intentional agent and those with a computer program that randomly generated outcomes. In striking contrast, control participants were less likely to invest when they believed that they were interacting with people than a computer opponent (Bohnet and Zeckhauser, 2004; Houser et al., 2009), revealing that normal economic decisions are driven by factors beyond mere probability, and that "people care not only about the payoff outcome but also about how the outcome came to be" (Bohnet and Zeckhauser, 2004). Accordingly, trust decisions, relative to risk decisions, entail additional costs, costs shown to be above and beyond mere monetary losses, which diverse authors (Bohnet and Zeckhauser, 2004; Bohnet et al., 2008; Fehr, 2009; Houser et al., 2009) have explained as due to betrayal aversion, namely, the fear to be exploited by others in social interactions. Here, we suggest that, after vmPFC damage, people lack such exploitation aversion, due to impaired social valuation (Amodio and Frith, 2006; Mitchell et al., 2006; Rudebeck et al., 2006; Hare et al., 2010; Tricomi et al., 2010; Zaki et al., 2013), which makes them more willing to take risk arising from interpersonal exchanges. Concerns about "others" do not matter for vmPFC patients, so that they perceive the decision of whether or not to trust basically as a risky choice and decide based on their expectations of trustworthiness and their propensity to risk. That is, it does not matter whether the risk is constituted through the uncertain behavior by the trustee, or through a random mechanism. In this sense, vmPFC patients behave more "rationally" than control participants in our trust game: they only care about their own payoffs and are hardly betrayal averse, as predicted by the standard economic model.

Thus, the seemingly greater level of trust observed in vmPFC patients could be related to their incapacity to value social information and consider negative anticipatory emotional responses related to trusting behavior, specifically they could fail to anticipate in their decision process the value of negative emotional responses associated with the risk of betrayal. Obviously, vmPFC patients' neglect of potential betrayal and increased willingness to take social risk may invite exploitation and attract selfish actors, which may explain, in part, why their social and financial investments are bound to fail.

A previous study of trust behavior in humans with vmPFC damage failed to find significant difference in economic investment between vmPFC patients and control groups (Krajbich et al., 2009). Several methodological differences may account for the contrasting results between these studies. First, in our trust game choices were continuous and quantitative (e.g., the investor decides how much of her endowment to transfer to the trustee), whereas, in Krajbich et al.'s (2009) study, investor had only binary choices (e.g., trust *vs*. no trust). The binary-choice trust game is easy to implement, but it is less sensitive and likely captures less variation in investor's trusting behavior. Second, economical exchange with interacting partners was more realistic and salient in the present than in previous study (e.g., their subjects were told that their partners were in another city and were in contact with the experimenter over the phone), which may have also enabled us to find the reported effect. Third, our study involved a larger vmPFC patient sample, which allowed us to reveal a significant difference in trusting behavior after vmPFC damage.

Furthermore, our findings are completely in line with recent evidence of increased rate of investment during a trust game, but not during a risk game, in participants with selective basolateral amygdala damage (van Honk et al., 2012), a region heavily interconnected with the vmPFC (Koolhaas et al., 1990; Bachevalier and Loveland, 2006). The amygdala and vmPFC are thought to act closely together as a part of the neural circuitry regulating many goal-directed behaviors (Murray and Izquierdo, 2007), thereby allowing the selection of advantageous actions in the face of various competing behavioral options. Interestingly, Bos et al. (2010) found decreased trustworthiness in women after being administered testosterone, a hormone targeting on the amygdala. As suggested by Johnson and Breedlove (2010), testosterone might reduce interpersonal trust by acting on neurons in the amygdala to increase communication to systems enabling fearful responses, while reducing communication to orbitofrontal cortex, whereas oxytocin might boost interpersonal trust (see Kosfeld et al., 2005), acting on the same systems with opposite effects.

Thus, previous and current findings suggest that (basolateral) amygdala and vmPFC are critically involved in social economic decisions. Note, however, that several findings from animal studies (see Murray and Izquierdo, 2007, for a review) suggest that, although amygdala and vmPFC functionally interact in mediating some types of adaptive choices, they make distinct contributions to emotional responses and reward processing. For example, while the greater level of trust after basolateral amygdala damage has been interpreted in terms of pathological altruism and generosity (van Honk et al., 2012), the reduced trustworthiness observed in current and previous study (Krajbich et al., 2009) shows such a view to be untenable for vmPFC-lesioned patients. Further research will be necessary to specify the nature of the interaction between the vmPFC and amygdala and how dysfunctions in this circuit differentially contribute to economic decisions in a social context.

Altogether, the above evidence suggests that vmPFC patients, as well as amygdala-lesioned patients, might lack of a mechanism of social vigilance, that is, they could be impaired in the recruitment of social emotions that need to be anticipated correctly in order for decisions to be made optimally. vmPFC, deemed as tuned to the evaluation of social information (Amodio and Frith, 2006), might fail in the recollection of past emotions related to a specific decision by upregulating the value/consequences of future options based on the resulting affective states (Bechara, 2005). However, another mechanism that might be impaired in vmPFC patients is prospection. Prospection refers to the ability to self-project in time (also referred to as mental time travel) to pre-experience future events (Buckner and Carroll, 2007). An impaired prospection might result in myopic, impulsive behaviors. Shortsighted decision-making is indeed a peculiar outcome of vmPFC disruption, resulting in increased impulsive behavior during intertemporal choice (Sellitto et al., 2010, 2011) in increased willingness to judge as acceptable personal violations (Ciaramelli et al., 2007; Moretto et al., 2009; Ciaramelli and di Pellegrino, 2011); in reduced acceptance rate of unfair offers from a human partner (when monetary gains were presented as abstract amounts to be received later) (Moretti et al., 2009); in reduced interpersonal disgust (Ciaramelli et al., 2013). Indeed, the large investments of vmPFC patients in the trust game can be considered shortsighted, impulsive decisions (see also van Honk et al., 2012, for a similar argument).

Taken together, the reported findings allow us to suggest that a lesion in the vmPFC might impair the strategic planning and anticipation of consequences of future events, by both disrupting the correct anticipation of emotions (social emotion, in the current case) to assign them a value, and preventing the optimal construction of possible scenarios following the choice.

In conclusion, these data showed that vmPFC has a critical role in trusting decisions and, in general, is essential for the normal valuation of social stimuli during an economic exchange with another person. These findings are highly compatible with current theories maintaining that vmPFC is a critical neural substrate for forecasting the (positive and negative) emotional consequences of available options in order to guide future behavior, both in personal and societal decision-making (Bechara and Damasio, 2005). Finally, the reported findings provide evidence for theoretical approaches to social cognition and decision-making that emphasize the pivotal role of medial prefrontal cortex in the integration of multiple signals to generate adaptive behavior (Montague and Berns, 2002).

## **ACKNOWLEDGMENTS**

This work was supported by grants from the University of Bologna (Ricerca Fondamentale Orientata), and the Ministero Istruzione Università e Ricerca (PRIN, 2010, protocol number: 2010XPMFW4\_009) to GdP.

### **REFERENCES**


Arrow, K. J. (1974). *The Limits of Organization.* NY, USA: Norton.


neurocognitive prospective. *Nat. Neurosci.* 8, 1458–1463. doi: 10. 1038/nn1584


*Science* 275, 1293–1295. doi: 10. 1126/science.275.5304.1293


ral basis of altruistic punishment. *Science* 305, 1254–1258. doi: 10. 1126/science.1100735


ral systems responding to degrees of uncertainty in human decisionmaking. *Science* 310, 1680–1683. doi: 10.1126/science.1115327


of cooperation in two-person reciprocal exchange. *Proc. Natl. Acad. Sci. USA* 98, 11832–11835. doi: 10. 1073/pnas.211415698


special topic forum not so different after all: a cross- discipline view of trust. *Acad. Manage. Rev.* 23, 393– 404.


Sellitto, M., Ciaramelli, E., and di Pellegrino, G. (2011). The neurobiology of intertemporal choice: insight from imaging and lesion studies. *Rev. Neurosci.* 22, 565–574. doi: 10. 1515/rns.2011.046

Somerville, L. H., Heatherton, T. F., and Kelley, W. M. (2006). Anterior cingulate cortex responds differentially to expectancy violation and social rejection. *Nat. Neurosci.* 9, 1007– 1008. doi: 10.1038/nn1728


Cohen, J. D. (2007). Dissociating affective evaluation and social cognitive processes in the ventral medial prefrontal cortex. *Cogn. Affect. Behav. Neurosci.* 7, 337–346. doi: 10.3758/cabn.7.4.337


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 07 June 2013; accepted: 03 September 2013; published online: 25 September 2013.*

*Citation: Moretto G, Sellitto M and di Pellegrino G (2013) Investment and repayment in a trust game after ventromedial prefrontal damage. Front. Hum. Neurosci. 7:593. doi: 10.3389/fnhum.2013.00593*

*This article was submitted to the journal Frontiers in Human Neuroscience.*

*Copyright © 2013 Moretto, Sellitto and di Pellegrino. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Neural correlates of the behavioral-autonomic interaction response to potentially threatening stimuli

#### *Tom F. D. Farrow1 \*, Naomi K. Johnson1, Michael D. Hunter 1, Anthony T. Barker 2, Iain D. Wilkinson3 and Peter W. R. Woodruff <sup>1</sup>*

*<sup>1</sup> Sheffield Cognition and Neuroimaging Laboratory, Academic Clinical Psychiatry, University of Sheffield, Sheffield, UK*

*<sup>2</sup> Department of Medical Physics, Sheffield Teaching Hospitals NHS Trust, Sheffield, UK*

*<sup>3</sup> Academic Unit of Radiology, University of Sheffield, Sheffield, UK*

#### *Edited by:*

*Susanne Leiberg, University of Zurich, Switzerland*

#### *Reviewed by:*

*Annette B. Bruehl, University Hospital of Psychiatry Zurich, Switzerland Clas Linnman, Boston Children's Hospital, USA*

#### *\*Correspondence:*

*Tom F. D. Farrow, Sheffield Cognition and Neuroimaging Laboratory, Department of Neuroscience, Academic Clinical Psychiatry, The Longley Centre, Northern General Hospital, University of Sheffield, Norwood Grange Drive, Sheffield, S5 7JT, UK. e-mail: t.f.farrow@sheffield.ac.uk*

Subjective assessment of emotional valence is typically associated with both brain activity and autonomic arousal. Accurately assessing emotional salience is particularly important when perceiving threat. We sought to characterize the neural correlates of the interaction between behavioral and autonomic responses to potentially threatening visual and auditory stimuli. Twenty-five healthy male subjects underwent fMRI scanning whilst skin conductance responses (SCR) were recorded. One hundred and eighty pictures, sentences, and sounds were assessed as "harmless" or "threatening." Individuals' stimulus-locked, phasic SCRs and trial-by-trial behavioral assessments were entered as regressors into a flexible factorial design to establish their separate autonomic and behavioral neural correlates, and convolved to examine psycho-autonomic interaction (PAI) effects. Across all stimuli, "threatening," compared with "harmless" behavioral assessments were associated with mainly frontal and precuneus activation with specific within-modality activations including bilateral parahippocampal gyri (pictures), bilateral anterior cingulate cortex (ACC) and frontal pole (sentences), and right Heschl's gyrus and bilateral temporal gyri (sounds). Across stimulus modalities SCRs were associated with activation of parieto-occipito-thalamic regions, an activation pattern which was largely replicated within-modality. In contrast, PAI analyses revealed modality-specific activations including right fusiform/parahippocampal gyrus (pictures), right insula (sentences), and mid-cingulate gyrus (sounds). Phasic SCR activity was positively correlated with an individual's propensity to assess stimuli as "threatening." SCRs may modulate cognitive assessments on a "harmless–threatening" dimension, thereby modulating affective tone and hence behavior.

**Keywords: functional magnetic resonance imaging (fMRI), skin conductance response (SCR), emotional salience, autonomic arousal, threat, affective tone**

# **INTRODUCTION**

Constantly changing environmental stimuli are rapidly processed by the brain to allow reorienting of cognitive resources such as attention toward possible threats (Öhman et al., 2001a,b). Such potential threats are identified by their emotional salience—a stimulus's state or quality of standing out relative to neighboring stimuli. One output of this stimulus-relevance cognitive processing is via the autonomic nervous system (ANS), controlling visceral functions such as perspiration, heart rate, respiration, and pupil diameter. However, due to positive and negative feedback between the cognitive and autonomic systems (Hugdahl, 1996), cognitive processing may be modulated by state or trait ANS activity, thereby subtly influencing how we attend to our environment, which in turn affects our behavior. Previous functional MRI studies have examined the BOLD response to threat processing, but mostly without measuring the ANS component.

Affective tone, an "emotional coloring" of the mental state accompanying every act or thought, arises from a dynamic interaction between cognitive assessment and ANS activity (Ross, 1997). Disturbance of this dynamic interaction, for example, in schizophrenia, may manifest as "sinister attribution bias" in which patients attribute negative connotations to apparently benign situations (Peer et al., 2004; Premkumar et al., 2008; Cohen and Minor, 2010). Physiological parameters such as ANS arousal which underpin affective tone are likely to vary along continua, both within healthy individuals and within pathological states (Wout et al., 2004; Horan et al., 2008; van Os et al., 2009). Hence, individuals within the "healthy" continuum may demonstrate varying levels of ANS and BOLD activity in response to potentially threatening stimuli which could influence the way in which they perceive stimuli and thereby interpret the world (Martin and Penn, 2001; Allen et al., 2007).

Emotionally salient pictures have been reported to activate amygdala, ventromedial prefrontal cortex (vmPFC), posterior hippocampus, and visual cortex (Kesler-West et al., 2001; Öhman et al., 2001b; Vuilleumier et al., 2001; Anders et al., 2004; Lee et al., 2004; Northoff et al., 2004; Heinzel et al., 2005; Garrett and Maddock, 2006; Grimm et al., 2006; Bryant et al., 2008; Premkumar et al., 2008; Kemp et al., 2009). However, many of these studies have used threatening (i.e., angry or fearful) faces. Faces, irrespective of emotion displayed, have specialized brain regions associated with their perception (Kesler-West et al., 2001; Narumoto et al., 2001; Vuilleumier et al., 2001; Britton et al., 2006; Tsao et al., 2006; Tsao and Livingston, 2008) and are restricted in how the displayed emotion is interpreted by healthy individuals (Calder et al., 2001) and should therefore be viewed as a "special case" of threat perception rather than a general exemplar (Britton et al., 2006). In contrast, many nonface stimuli could be described as "threat-ambiguous" in that they are open to subjective interpretation, based on previous experience, personality traits (Gard and Kring, 2009), and state levels of cognitive and autonomic arousal (VaezMousavi et al., 2007; Coccaro et al., 2009). Hence, for the "picture" condition in the current study, we used non-face stimuli. Furthermore, all stimuli (pictures, sentences, and sounds) were piloted to ensure that many were not at the extreme ends of a "harmlessthreatening" continuum. This allowed us to analyze behavioral responses on an individual basis (rather than pre-categorizing stimuli at the beginning of the study as either "harmless" or "threatening").

Studies of visually presented threat-related words have reported activation of the left inferior frontal gyrus (IFG) (Blackwood et al., 2000) and amygdala (together with left lingual gyrus and posterior parahippocampal gyrus; Isenberg et al., 1999; Compton et al., 2003). Previous research into the neural bases of pleasant and unpleasant sounds has mainly concerned music (Blood et al., 1999; Koelsch et al., 2005; Pallesen et al., 2005; Eldar et al., 2007). In their PET study, Blood et al. (1999) reported rCBF changes in paralimbic and neocortical areas when musical consonance and dissonance were varied (synonymous with a pleasant to unpleasant range). Notably, these neocortical areas were distinct from areas of primary auditory cortex (involved in pitch and loudness discrimination) or secondary auditory cortex (involved in harmonic, melodic, and rhythmic pattern detection). The relative lack of neuroimaging research into auditory compared with visual stimuli probably has much to do with the difficulties of presenting sounds in a noisy MRI scanner (Di Salle et al., 2003). In the present study, we minimized the difficulties associated with auditory interference by using a "sparse" EPI protocol which allows stimuli to be presented during silent gaps in the scanner sequence.

Previous research on threat perception has also examined response times (RTs) to threatening or negatively valenced stimuli (Cloitre et al., 1992; Estes and Verges, 2008), and counter-intuitively reported *increased* RTs to threatening compared with neutral stimuli. One possible explanation for this finding is that salient stimuli produce opposing effects on attention and behavior such that salience facilitates the identification of threat but slows or inhibits responses to it (Estes and Verges, 2008).

In summary, a number of previous researchers have investigated neural responses to emotionally salient visual and auditory stimuli, though these studies have often involved the "special case" of faces or unambiguous stimuli which were pre-categorized as positive or negative (or "harmless" or "threatening" or "pleasant" or "unpleasant"). By recording SCRs and fMRI BOLD signal to individually rated stimuli we sought to investigate the modulating effect of ANS arousal on brain activation. This concurrent collection of fMRI and SCR data allowed us to examine what we term a psycho-autonomic interaction effect [PAI; comparable with the more often reported psychophysiological interaction effects (PPI)] to "threat-ambiguous" stimuli. Specifically, this convolution methodology allowed examination of BOLD responses attributable to an interaction between autonomic and behavioral responses above and beyond those activations attributable to autonomic and behavioral responses separately. We chose to use pictures, sentences and sounds to allow investigation of stimulus-modality-dependent and -independent factors.

We hypothesized that stimuli subjectively assessed as "threatening" compared with those assessed as "harmless" would be associated with increased RTs (Cloitre et al., 1992; Estes and Verges, 2008) and SCR amplitudes (Hugdahl, 1996). We also hypothesized that stimuli subjectively assessed as "threatening," irrespective of modality or accompanying phasic SCR, would be associated with increased amygdala activity compared with stimuli assessed as "harmless" (Bishop et al., 2004; Bertolino et al., 2005). We furthermore hypothesized modality-specific activations to "threatening" compared with "harmless" stimuli, specifically, (1) vmPFC, posterior hippocampus, and visual cortex to pictures (Lee et al., 2004; Northoff et al., 2004; Heinzel et al., 2005; Garrett and Maddock, 2006; Grimm et al., 2006); (2) IFG to sentences (Isenberg et al., 1999; Blackwood et al., 2000; Compton et al., 2003); and (3) auditory cortex to sounds (Blood et al., 1999; Koelsch et al., 2005; Pallesen et al., 2005; Eldar et al., 2007). Finally, we hypothesized that phasic SCR activity would be associated with activation of dorso-posterior brain regions (Fredrikson et al., 1998; Patterson et al., 2002) and that PAIs would show dissociable, between-modality activations. In light of the continuum of neuropsychological profiles in healthy volunteer cohorts (Martin and Penn, 2001; Wout et al., 2004; Horan et al., 2008; van Os et al., 2009), we also sought to investigate the influence of schizotypal personality traits on the recorded autonomic and behavioral responses.

## **MATERIALS AND METHODS**

## **ETHICS STATEMENT**

All subjects gave written informed consent. The study was approved by the North Sheffield Research Ethics Committee.

#### **STIMULUS DEVELOPMENT AND PILOTING**

Sixty picture stimuli from the International Affective Picture System (IAPS; Lang et al., 1997) and sixty sentence and sound stimuli developed within our laboratory were piloted on large cohorts (*>*65) of healthy subject as to whether they were "harmless" or "threatening." Individual stimuli varied considerably as to the percentage of raters subjectively assessing them as threatening thereby confirming their subjective threat-ambiguous nature. Experimental stimuli used are listed in Appendix **Table A1**.

#### **SUBJECTS AND NEUROPSYCHOLOGICAL ASSESSMENT**

Twenty-five healthy right-handed males (22 ± 2 years old; estimated IQ—National Adult Reading Test, NART; Nelson, 1982 113 ± 6; range 97–123; 16 ± 1 years of education) participated in the study. Study recruitment inclusion criteria comprised being aged 20–35, male, right handed, no current or previous significant neurological or psychiatric disorder, normal or corrected-tonormal vision, no hearing impairment and no general contraindication to MR imaging. Personality-based neuropsychological data were collected from all subjects. Oxford-Liverpool Inventory of Feelings and Experiences sub-scale scores (O-LIFE; Mason et al., 1995; Mason and Claridge, 2006) were: "Unusual Experiences" 3 ± 4, range 0–17 (mean ± SD); "Cognitive Disorganization" 6 ± 5, range 0–17; "Introvertive Anhedonia" 3 ± 2, range 0–6; and "Impulsive Nonconformity" 8 ± 4, range 2–19. Empathy Quotient scale scores (EQ; Baron-Cohen and Wheelwright, 2004) were 46 ± 10; range 32–70; and Paranoia and Suspiciousness Questionnaire scores (PSQ; Rawlings and Freeman, 1996) were 9 ± 6; range 2–27. These tests were chosen to measure individual personality traits, which may be associated with a vulnerability to schizoptypal behavior (psychosis-proneness) and hence a tendency to over-attribute threat (Braunstein-Bercovitz, 2000).

## **INTRA-SCANNER SCR RECORDING**

ANS activity was measured via skin-conductance response (SCR) recording. A typical phasic SCR is temporally very similar to the BOLD hemodynamic response and is therefore a suitable measure with which to sub-average or convolve fMRI data. MRcompatible SCR equipment was based on a battery powered, electrically isolated, same electrode configuration implementation of a previously published method (Shastri et al., 2001). SCRs sampled at 20 Hz from the medial phalange of the left index and middle fingers, using 8 mm diameter Ag/AgCl electrodes were recorded concurrently with fMRI and behavioral response data.

## **fMRI IMAGING**

Subjects underwent three 12 min fMRI scans (EPI "sparse" sequence; 60 time points; TR = 12 s; TA = 3 s; TE = 40 ms; SENSE factor = 1.5; FOV = 240 mm; matrix size = 128 × 128, 32 × 4 mm thick contiguous axial slices) at 1.5 Tesla (Eclipse, Philips Medical Systems, Ohio, USA). This data acquisition sequence setup yielded a voxel size of 1.8 × 1*.*8 × 4 mm. The sparse sequence allows stimuli to be delivered during scanner silent periods (apart from the noise of the helium compressor pump), and for data acquisition to be targeted at a period immediately after task completion, utilizing the physiological delay and dispersion between neuronal activity and its resulting hemodynamic response (Eden et al., 1999). In an order-counterbalanced design, subjects viewed pictures or sentences via a head-coil mounted mirror or listened to sounds via electrostatic headphones. All 180 stimuli (60 pictures, sentences, and sounds; Appendix **Table A1**) were presented for 4 s each during scanner silence, immediately followed by 3 s of fMRI signal acquisition and a further 5 s of scanner silence (**Figure 1**). Hence a new stimulus was presented every 12 s (**Figure 1**). Between presentation of individual pictures and sentences, and continuously during the presentation of sounds, a centrally located fixation cross was displayed. Throughout all scans, the words "Harmless" and "Threatening" were displayed at the bottom of the screen, in a laterality-balanced design (i.e., for half the subjects "Harmless" was displayed on the left of the screen and on the right for the other half of subjects). In a forced-choice

design, subjects behaviorally assessed each stimulus as subjectively "harmless" or "threatening" via an intra-scanner button box using their right index and middle fingers.

#### **SCR DATA ANALYSES**

SCR traces (14,400 data points per 12-min scan) were analyzed in Ledalab v.3.2.9 (www*.*ledalab*.*de/; Benedek and Kaernbach, 2010a) using the Continuous Decomposition Analysis method to distinguish the phasic (driver) information from the underlying tonic sudomotor nerve activity. Raw SCR data were smoothed via convolution with a Hann window to reduce error noise and fitted to a bi-exponential Bateman function. Data were optimized by a conjugated gradient descent algorithm to reduce the error between them and the inbuilt SCR model. These processing steps allowed computation of a stimulus-locked "integrated skin conductance response" (ISCR), a time-integration of the continuous phasic activity for each stimulus. The ISCR thus represents an unbiased and time-sensitive measure of sympathetic activity in response to each stimulus (Benedek and Kaernbach, 2010b). For investigating whether SCRs may modulate RTs, each stimulus epoch was also classified via Mindware EDA 2.40 (Mindware Technologies Ltd., OH, USA) as having a significant phasic SCR "present" or "absent" ("a 'typical' SCR comprising trough, peak and half-return components, identified within 12 s of stimulus onset; trough-to-peak amplitude = 0.15 µS"). Custom MATLAB scripts (v. R2007b; The MathWorks, Inc., Sherborn, MA, USA) extracted stimulus-locked peak amplitude data for group-averaging of SCRs within and across subjects.

#### **fMRI DATA ANALYSES**

Functional MRI data were analyzed in SPM8 (Wellcome Department of Imaging Neuroscience, London; www*.*fil*.*ion*.*ucl*.* ac*.*uk/spm/) implemented in MATLAB v. R2007b on a PC. The EPI images for each run were corrected for head movement by affine registration using a two-pass procedure by which images were initially realigned to the first image and subsequently to the mean of the realigned images. After realignment, the mean EPI image for each run was spatially normalized to the Montreal Neurological Institute (MNI; Mazziotta et al., 2001) single subject template using the unified segmentation approach (Ashburner and Friston, 2005). The resulting parameters of a discrete cosine transform, which define the deformation field necessary to move the data into the space of the MNI tissue probability maps, were then combined with the deformation field transforming between the latter and the MNI single subject template. The ensuing deformation was applied to the individual EPI volumes, which were thereby transformed into the MNI single-subject space and resampled at 2 × 2 × 2 mm voxel size. The normalized images were smoothed using a 6 mm fullwidth at half-maximum Gaussian kernel to meet the statistical requirements of the General Linear Model and to compensate for residual macro-anatomical variations. For each scan (three per subject), ISCR data (one data point per stimulus epoch) and each individual's harmless/threatening behavioral data were used for regression analysis. At this first level of analysis the BOLD responses were convolved with a canonical hemodynamic response function (HRF), and its temporal derivative. The silent periods of the EPI sequence were modeled in the design matrix by separately specifying the TR (12 s) and TA (3 s). Given the significant differences in reaction times between "harmless" and "threatening" assessments (see "Results" section), an additional reaction time regressor was also added to the model. Hence, for each of the 75 scans, three regression matrices were created: (1) an 8-column regression matrix comprising 1 column of ISCR data, 1 column of reaction time data and 6 columns of subject's movement parameters (obtained from the preprocessing realignment stage); (2) an 8-column regression matrix comprising 1 column of individual behavioral data (harmless = −1; threatening = 1), 1 column of reaction time data and 6 columns of subject's movement parameters; and (3) a 10-column regression matrix comprising 1 column of the convolution between ISCR and behavioral response, 2 columns of separate ISCR and individual behavioral data, 1 column of reaction time data and 6 columns of subject's movement parameters. This final 10-column matrix allowed examination of BOLD responses attributable to an PAI; i.e., brain activity above and beyond those activations separately attributable to the ISCR and behavioral data. These first-level regression analyses were group-averaged at the second-level using a fully flexible factorial design, with factors of subject and modality (picture, sentence, or sound). In this random-effects model, we allowed for violations of sphericity by modeling non-independence across images from the same subject and unequal variances between conditions and subjects as implemented in SPM8. In line with recent guidelines (Lieberman and Cunningham, 2009), analysis of our novel and exploratory complex social neuroscience paradigm was conducted at a significance threshold of *p <* 0*.*001 uncorrected for multiple comparisons with a minimum extent threshold of 10 voxels. Analysis of the neural correlates of electrodermal activity (i.e., SCR) which has previously been shown to be associated with robust functional activations (Fredrikson et al., 1998; Patterson et al., 2002) was conducted at a significance threshold of *p <* 0*.*05 corrected for family wise error (FWE). MNI coordinates of all supra-threshold voxels were transformed into Talairach coordinates (Talairach and Tournoux, 1988) using the "mni2tal.m" Matlab script (http://imaging*.*mrc-cbu*.*cam*.*ac*.*uk/ imaging/MniTalairach).

## **RESULTS**

#### **BEHAVIORAL, PHYSIOLOGICAL, AND NEUROPSYCHOLOGICAL**

For behavioral RTs, in a 3 × 2 × 2 within-subject, repeatedmeasures ANOVA (picture or sentence or sound × "harmless" or "threatening" × presence or absence of an SCR), there was a main effect of subjective assessment ["threatening" longer RTs than "harmless"; *F(*1*,* <sup>24</sup>*)* = 14*.*51, *p* = 0*.*001; **Figure 2**], a main effect of modality [sounds longer RTs than sentences; sentences longer RTs than pictures; *F(*2*,* <sup>48</sup>*)* = 98*.*05, *p <* 0*.*001; **Figure 2**], but no main effect of the presence or absence of an SCR [*F(*1*,* <sup>24</sup>*)* = 0*.*26, *p* = 0*.*614]. There were no significant differences in the number of SCRs to stimuli assessed as "threatening" compared with those assessed as "harmless" (percentage figures in chart bars; **Figure 2**). However, for SCR amplitudes, a 3 × 2 within-subject, repeated-measures ANOVA (picture or sentence or sound ×

dotted columns; "harmless" responses are shown as plain columns. Error bars are 95% confidence intervals. There was a main effect of modality on RTs [sounds longer RTs than sentences, which had longer RTs than pictures; *F(*2*,* <sup>48</sup>*)* = 98*.*05, *p <* 0*.*001], a main effect of subjective assessment on RTs

no main effect of presence or absence of an SCR on RTs [*F(*1*,* <sup>24</sup>*)* = 0*.*26, *p* = 0*.*614; data not shown; repeated measures ANOVA]. There were no significant differences in the percentage of SCRs to stimuli assessed as "threatening" compared with those subjectively assessed as "harmless" (% figures in chart bars).

"harmless" or "threatening"), revealed a main effect of assessment ["threatening" greater SCR amplitude than "harmless"; *F(*1*,* <sup>24</sup>*)* = 8*.*32, *p* = 0*.*008; **Figure 3**] and a trend toward a main effect of modality [sounds greater SCR amplitudes than pictures; pictures greater SCR amplitudes than sentences; *F(*2*,* <sup>48</sup>*)* = 3*.*0, *p* = 0*.*059;

**Figure 4**], but no interaction [*F(*2*,* <sup>48</sup>*)* = 2*.*22, *p* = 0*.*12]. *Post-hoc* pair-wise comparison (Tukey's HSD test) showed that "threatening" sounds and pictures were associated with significantly greater SCR amplitudes than sounds and pictures assessed as "harmless" (*t* = 2*.*65, *p* = 0*.*006 and *t* = 1*.*89, *p* = 0*.*033, respectively;

**Figure 4**), but that there was no significant difference for sentences (*t* = 0*.*33, *p* = 0*.*372; **Figure 4**).

compared with those assessed as "harmless" (dotted line) evoked significantly

There was a significant positive correlation between an individual's average ISCR and number of stimuli assessed as "threatening" for sentences (*r* = 0*.*431, *p* = 0*.*016) and sounds (*r* = 0*.*385, *p* = 0*.*032), but not for pictures (*p >* 0*.*1). There were no significant correlations between ISCR or number of stimuli assessed as "threatening" and O-LIFE, EQ or PSQ scale scores (*p >* 0*.*1).

#### **fMRI—AUTONOMIC (ISCR) REGRESSOR**

Across all stimuli (i.e., without differentiating between modalities), ISCR was associated with activations including bilateral precentral gyrus/supplementary motor area [SMA; Brodmann's Area (BA) 4/6], medial prefrontal cortex (mPFC; BA 8), precuneus/cuneus (BA 7/19), thalamus [dorso-medial (DM) nucleus], bilateral lingual gyrus (BA 18) and cerebellum (**Table 1**; **Figure 5**; *p <* 0*.*05 FWE). Separately, for picture, sentence and sound stimuli, this dorsal (precentral gyrus/SMA) and posterior (lingual gyrus/cerebellum) activation was replicated, though the DM-thalamic activation was only present for picture and sentence stimuli (i.e., not sounds). However, sound-stimuli ISCR data were associated with activation of left amygdala.

#### **fMRI—BEHAVIORAL RESPONSE REGRESSOR**

Across all stimuli, "threatening" compared with "harmless" behavioral assessments were associated with activation of bilateral middle frontal gyrus (MidFG; BA 10/46), mPFC/frontal pole (BA 10), anterior cingulate cortex (ACC; BA 24/32), precuneus (BA 7), and lingual gyrus (BA 18; **Table 2**; **Figure 6**). Threatening pictures were associated with activation including bilateral parahippocampal gyrus/lingual gyrus (BA 30/19), bilateral angular gyrus/temporo-parietal junction (BA 39), mPFC/ACC (BA 10/32), and posterior cingulate/precuneus (BA 31/7; **Table 3A** **Table 1 | Picture, sentence, and sound stimuli. Brain activations associated with integrated skin conductance response (ISCR) activity (see Figure 5).**

sentence stimuli subjectively rated as "threatening" or "harmless" (*p >* 0*.*1).


*Co-ordinates are shown in standardized neuroanatomical space (Talairach and Tournoux, 1988). R, right; L, left; BA, Brodmann's area; SMA, supplementary motor area; mPFC, medial prefrontal cortex; MidFG, middle frontal gyrus; DM, dorso-medial. Co-ordinates without a corresponding extent threshold are shown in italics and refer to sub-clusters of the preceding activation. P < 0.05 corrected for family wise error (FWE).*

and **Figure 7**). This bilateral parahippocampal gyrus activation survived FWE correction at *p <* 0*.*05. Threatening sentences were associated with activation including bilateral MidFG/frontal pole (BA 10), bilateral ACC (BA 24), posterior cingulate and

## **Table 2 | Pictures, sentences, and sounds. Brain activations associated with "Threatening" compared with "Harmless" behavioral judgments (see Figure 6).**


*Co-ordinates are shown in standardized neuroanatomical space (Talairach and Tournoux, 1988). R, right; L, left; BA, Brodmann's area; mPFC, medial prefrontal cortex. Co-ordinates without a corresponding extent threshold are shown in italics and refer to sub-clusters of the preceding activation. P < 0.001 uncorrected for multiple comparisons; extent threshold = 10.*

precuneus (BA 30/7; **Table 4A** and **Figure 9**). This left ACC activation survived FWE correction at *p <* 0*.*05. Threatening sounds were associated with activation including right transverse temporal gyrus (also known as Heschl's gyrus; BA 41) and bilateral middle/superior temporal gyrus (BA 21/22/42; **Table 5A** and **Figure 11**).

**FIGURE 6 | "Threatening"** *>* **"harmless" regressor across picture, sentence, and sound stimuli.** Main effect of stimuli subjectively assessed as "threatening" across modalities. Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 2** for anatomical descriptions and co-ordinates.

## **fMRI—AUTONOMIC (ISCR)-BEHAVIORAL RESPONSE CONVOLVED REGRESSOR**

Across all stimuli, the interaction between autonomic (ISCR) and "threatening" or "harmless" assessment responses—our PAI was associated with activation of right MidFG (BA 10; T&T co-ordinates 24 42 −7) and left mid-cingulate gyrus (BA 24; −6 −23 36). Threatening picture-ISCR interactions were associated with activation of right fusiform gyrus/parahippocampal gyrus (BA 37; **Table 3B** and **Figure 8**). Threatening sentence-ISCR interactions were associated with activation of right insula and MidFG (BA 10), left thalamus [ventral posterolateral (VPL) nucleus], left superior temporal gyrus (BA 22) and left cerebellum (**Table 4B** and **Figure 10**). Threatening sound-ISCR interactions were associated with activations including left mid-cingulate gyrus, bilateral postcentral gyrus (BA 1/2/3), bilateral IFG (BA 44/47) and right inferior parietal lobule (BA 40; **Table 5B** and **Figure 12**). This left mid-cingulate gyrus activation survived FWE correction at *p <* 0*.*05.

## **DISCUSSION**

In agreement with our first and second hypotheses, picture, sentence, and sound stimuli subjectively assessed as "threatening" compared with those assessed as "harmless" had significantly longer RTs and increased SCR amplitudes (except non-significantly for sentence SCR amplitudes). Parieto-occipitothalamic brain regions were associated with autonomic arousal (ISCR) across stimulus modalities, in broad agreement with previous research (Fredrikson et al., 1998; Patterson et al., 2002). Across stimulus modalities, stimuli assessed as "threatening"


**Table 3A | Pictures. Brain activations associated with "Threatening" compared with "Harmless" behavioral judgments (see Figure 7).**

**Table 3B | Pictures. Brain activations associated with psycho-autonomic interaction (PAI) of integrated skin conductance response (ISCR) and behavioral response (see Figure 8).**


*Co-ordinates are shown in standardized neuroanatomical space (Talairach and Tournoux, 1988). R, right; L, left; BA, Brodmann's area; paraH, parahippocampal; post., posterior; TTG, transverse temporal gyrus; MidFG, middle frontal gyrus; mPFC, medial prefrontal cortex; ACC, anterior cingulate cortex; IPL, inferior parietal lobule; TPJ, temporo-parietal junction; g., gyrus. Co-ordinates without a corresponding extent threshold are shown in italics and refer to sub-clusters of the preceding activation. P < 0.001 uncorrected for multiple comparisons; extent threshold* = *10.*

activated prefrontal and precuneus regions, but in contrast to the ISCR findings, there were very clear modality-specific activations. The ISCR-behavioral response convolution (PAI) analysis revealed modality-specific activations which were distinct from

**FIGURE 7 | Pictures.** Activations associated with "threatening" compared with "harmless" behavioral responses. Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 3A** for anatomical descriptions and co-ordinates.

those seen in the separate ISCR and behavioral-response analyses. Subjects' average ISCRs were positively correlated with the number of sentence and sound stimuli assessed as "threatening." Contrary to our remaining hypotheses we did not find that stimuli assessed as "threatening" were routinely associated with supra-threshold amygdala activity or a relationship between schizotypal personality traits and autonomic or behavioral responses.

The brain areas associated with autonomic arousal, which function in parallel with cognitive assessment of environmental stimuli, included left amygdala (sounds), dorsomedial thalamic nucleus (pictures and sentences), precuneus, lingual gyrus, and motor cortex (bilateral precentral gyrus/SMA). The amygdala, thalamic, precuneus, and SMA activations are likely directly related to autonomic arousal (Critchley et al., 2003; Napadow et al., 2008; Zhang et al., 2012). The lingual gyrus has previously been associated with the generation and representation of SCRs (Critchley et al., 2000). The precuneus has also been associated with emotional self-regulation (Johnston et al., 2010) whilst the motor cortex has been associated with internal attributions of events whether or not the "self " was viewed as an active intentional agent (Blackwood et al., 2000). Activation of the motor cortices may also prepare the body to move away from threat, though some research has actually reported a decreased activity in primary motor cortex during anticipation of an aversive event (cognitively induced fear; Butler et al., 2007). However, it has also been reported that different aspects of the emotional response, namely arousal and valence, may be mediated by different brain circuits (Anders et al., 2004). Anders and colleagues, using human and animal pictures from the IAPS,

**FIGURE 8 | Pictures.** Psycho-autonomic interaction (PAI) between integrated skin conductance response (ISCR) and behavioral response ("threatening" *>* "harmless"). Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 3B** anatomical descriptions and co-ordinates.

**FIGURE 9 | Sentences.** Activations associated with "threatening" compared with "harmless" behavioral responses. Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 4A** for anatomical descriptions and co-ordinates.

studied the trial-by-trial correlation of brain activation with SCR, startle response and subjective ratings of valence and arousal. Post-scan arousal reports to individual pictures were positively correlated with SCR. Using a region-of-interest approach Anders

#### **Table 4A | Sentences. Brain activations associated with "Threatening" compared with "Harmless" behavioral judgments (see Figure 9).**


**Table 4B | Sentences. Brain activations associated with psycho-autonomic interaction (PAI) of integrated skin conductance response (ISCR) and behavioral response (see Figure 10).**


*Co-ordinates are shown in standardized neuroanatomical space (Talairach and Tournoux, 1988). R, right; L, left; BA, Brodmann's area; ACC, anterior cingulate cortex; MidFG, middle frontal gyrus; STG, superior temporal gyrus; MTG, middle temporal gyrus; VPL n., ventro-postero-lateral nucleus. Co-ordinates without a corresponding extent threshold are shown in italics and refer to sub-clusters of the preceding activation. P < 0.001 uncorrected for multiple comparisons; extent threshold* = *10.*

and colleagues showed that activation of the amygdala and insula positively correlated with valence ratings, whilst arousal ratings were correlated with thalamic and frontomedial cortex activity. Peripheral physiologic responses (SCR and startle response) were localized to regions of anterior parietal cortex, primarily somatosensory association areas. Furthermore, Anders and colleagues report a functional segregation of brain structures differentiating SCR and startle responses from verbal responses. Specifically, whilst SCRs were associated with frontomedial cortex activity and startle responses with amygdala activity, verbal ratings of valence and arousal were associated with activation of insula and thalamus respectively.

**FIGURE 10 | Sentences.** Psycho-autonomic interaction (PAI) between integrated skin conductance response (ISCR) and behavioral response ("threatening" *>* "harmless"). Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 4B** for anatomical descriptions and co-ordinates.

**FIGURE 11 | Sounds.** Activations associated with "threatening" compared with "harmless" behavioral responses. Flexible factorial design *p <* 0*.*001. Extent threshold = 10. See **Table 5A** for anatomical descriptions and co-ordinates.

Contrary to our third hypothesis, we did not find amygdala activation to be routinely associated with all stimuli assessed to be "threatening." One possible explanation for this lack of robust amygdala activation is that rather than being necessary **Table 5A | Sounds. Brain activations associated with "Threatening" compared with "Harmless" behavioral judgments (see Figure 11).**


#### **Table 5B | Sounds. Brain activations associated with psycho-autonomic interaction (PAI) of integrated skin conductance response (ISCR) and behavioral response (see Figure 12).**


*Co-ordinates are shown in standardized neuroanatomical space (Talairach and Tournoux, 1988). R, right; L, left; BA, Brodmann's area; MTG, middle temporal gyrus; STG, superior temporal gyrus; TTG, transverse temporal gyrus (also known as Heschl's gyrus); IPL, inferior parietal lobule; TPJ, temporo-parietal junction; paraH, parahippocampal; ACC, anterior cingulate cortex; IFG, inferior frontal gyrus; MidFG, middle frontal gyrus; OFC, orbitofrontal cortex; SMA, supplementary motor area. Co-ordinates without a corresponding extent threshold are shown in italics and refer to sub-clusters of the preceding activation. P < 0.001 uncorrected for multiple comparisons; extent threshold* = *10.*

for fear perception *per se*, the amygdala is active when the rest of the brain cannot easily predict (1) what sensations mean? (2) what to do about them? or (3) what value they hold in that context? (Lindquist et al., 2012). Hence, the subjectively variable level of threat conferred by our stimuli may have lessened the difference in amygdala activation between "harmless" and "threatening" subjective assessments. An alternative explanation is that if a proportion of stimuli subjectively assessed as "harmless" were actually experienced as pleasant and thus led to positive arousal and hence amygdala activation, that our main contrast of interest (i.e., threatening *>* harmless), would not have shown a significant difference in relative amygdala activation.

threshold = 10. See **Table 5B** for anatomical descriptions and co-ordinates.

Our modality-specific hypotheses of areas more activated by "threatening" than "harmless" assessments were in the main confirmed for pictures (lingual gyrus, parahippocampal gyrus, and mPFC) and sounds (right Heschl's gyrus and bilateral superior temporal gyrus), but less so for sentences, where the bilateral dorsal (cognitive) ACC and MidFG activations were predominant (as opposed to the left IFG which we hypothesized). One possible explanation for the lack of predicted activations to threatening sentences is that our hypothesis was based on previous research into threatening versus non-threatening words (Isenberg et al., 1999; Blackwood et al., 2000; Compton et al., 2003) which may require less cognitive processing and deliberation than full sentences. Sensory facilitation of auditory cortex by emotional cues as we have shown was recently reported (Plichta et al., 2011) in a functional near-infrared spectroscopy (fNIRS) study using pleasant, unpleasant and neutral sounds from the International Affective Digitized Sound System (IADS; Bradley and Lang, 1999) database. However, Plichta and colleagues report that both pleasant and unpleasant sounds led to significantly greater auditory cortex activation than neutral sounds, with no significant difference between pleasant and unpleasant. As our present study involved subjects making assessments on a "harmless"–"threatening" binary dimension, it is likely that our "harmless" category contained stimuli which could be described as both "pleasant" and "neutral".

The dorsal (cognitive) division of ACC which was activated by threatening pictures and sentences, is classically associated with error detection and monitoring as opposed to the ventral (affective) ACC which is classically associated with assessing the salience of emotional information (Bush et al., 2000). Though by this "classical model," activation of ventral ACC would better fit with the task demands, recent research (Shackman et al., 2011) has argued for a more general role for the anterior midcingulate cortex (aMCC), specifically in generating aversively motivated behavior across affect, pain and cognition. This "adaptive control hypothesis" by which the aMCC activates when the most adaptive course of action is uncertain and outputs to motor centers executing goal-directed behavior fits neatly with making subjective assessments of potentially threatening environmental stimuli. An alternative explanation for the brain activations seen for the threatening-harmless contrasts is that they reflect the fronto-parietal networks implicated in top-down attention (Corbetta and Shulman, 2002) and that threatening stimuli elicited more attention than harmless ones. This latter explanation and the "adaptive control hypothesis" are of course not mutually exclusive.

Our ISCR-behavioral response convolution (PAI) analyses were designed to reveal brain regions above and beyond those BOLD activations attributable to autonomic and behavioral responses separately. Results included right parahippocampal gyrus for pictures, right insula and ACC for sentences and left mid-cingulate gyrus/bilateral IFG for sounds. It is noteworthy from these modality-specific findings that there was greater interaction between SCR and behavior in high order visual cortex (Malach et al., 2002) for pictures and that the role of the insula in the detection and awareness of bodily changes ("interoception") has been the subject of much recent research (Craig, 2003, 2009; Critchley et al., 2004; Simmons et al., 2006; Singer et al., 2009) as these bodily changes may modulate cognitive interpretation and hence behavior. As regards the activations obtained by the PAI for sounds, previous research into the SCR orienting response (Williams et al., 2000) reported that "significant" compared with "familiar" stimuli activated brain regions including ventral ACC and ventral mPFC.

As we had no implicit baseline, our main contrast of interest compared how subjects assessed the subjective valence of stimuli. Hence our power to detect significant differences between conditions was restricted by the relatively subtle difference between the "active" and "baseline" conditions (stimuli assessed as being "threatening" and "harmless," respectively) and relevant analyses are reported at an uncorrected statistical threshold. Such a liberal threshold is in line with recent guidelines for analysis of complex social neuroscience paradigms (Lieberman and Cunningham, 2009). Similarly, the minimum extent threshold chosen (10 voxels) for the novel imaging contrasts was justified in our original ethics and research protocol as appropriate due to the exploratory nature of study.

Whilst a large proportion of the reported results are in line with our original *a priori* hypotheses, they are also occasionally at odds with more recently reported results of the neural and autonomic correlates of affective processing (i.e., those published after the present study was begun). Critchley (2009) in a review of the extant literature highlights the role of the anterior cingulate and insula in the response and representation of bodily states in specific behavioral contexts. Though we reported activation of right insula associated with autonomic arousal, our activation of anterior cingulate was primarily associated with "threatening" behavioral assessments. However, another recent study (Zhang et al., 2012), using a stop signal task to examine the neural correlates of SCRs reported activation of the SMA, middle cingulate gyrus and precuneus, findings which are much more in agreement with the current findings. Another recent study (Henderson et al., 2012) measured the neural correlates of spontaneous fluctuations in skin sympathetic nerve activity (SSNA) via direct recording from the common fibular nerve (as opposed to inferring SSNA from SCR). Using positively and negatively charged emotional images from the IAPS dataset to evoke autonomic arousal, SSNA was associated with more frontal regions (including orbital, dorsolateral, and vmPFC) than has generally been previously reported. Henderson and colleagues did however also report robust activation of right precuneus as we have done in the present study. Finally, two recent studies have examined the role of personality in modulating neural responses to anticipating threat in the form of electric shocks (Drabant et al., 2011) and neural and autonomic responses to threatening facial expressions and body postures (Kret et al., 2011). In Drabant and colleagues' study, shock anticipation was associated with increased SCRs and corresponding activation of brain areas, many of which overlap with those reported in the present study, including precentral gyrus, thalamus, insula, and mid-cingulate cortex (ACC). Individual neuroticism scores in Drabant and colleagues' study were negatively correlated with activation of left IFG and insula. Kret and colleagues meanwhile examined the influence of negative affectivity and social inhibition on neural responses to videos of fearful and angry actors. While individuals with increased negative affectivity showed reduced activation of core emotion systems (including cortical and sub-cortical regions such as amygdala) socially inhibited individuals over-activated a broader, though exclusively cortical, network (including temporo-parietal junction, superior temporal gyrus, and orbitofrontal cortex).

Contrary to our final hypothesis we did not find a relationship between personality traits (as measured by the O-LIFE, EQ, and PSQ) and behavioral or autonomic responses. Previous studies have suggested a relationship between the main personality dimensions (the so-called "big five"; Digman, 1990) and SCR latency, but not magnitude (Mardaga et al., 2006). Our finding of a positive correlation between an individual subject's average ISCR and the number of "threatening" assessments they made suggests that such behavioral-autonomic modulations may be present over state- or mood-length periods, but are not related to measures of sub-clinical psychosis-proneness. This correlation between ISCR and "threatening" responses may also be related to an individual's underlying neuroticism (Drabant et al., 2011), a personality trait which we did not directly measure.

# **LIMITATIONS**

We have utilized a relatively liberal height and extent threshold for our fMRI results, which may have led to reporting of some Type I errors (i.e., false positives). However, use of a mapwide false discovery rate (FDR) and family-wise error (FWE) of *p <* 0*.*05 has been reported to be unduly conservative for novel complex cognitive and affective social neuroscience processes as were examined in this study (Lieberman and Cunningham, 2009). Use of the "sparse" fMRI sequence was required for delivery of sound stimuli and was kept for picture and sentence stimuli to facilitate inter-modality comparison. However, this necessarily restricted the time sampling window, though the data collection period was targeted at a period immediately after task completion, utilizing the physiological delay and dispersion between neuronal activity and its resulting hemodynamic response (Eden et al., 1999). The lack of an implicit baseline condition was considered a worthwhile trade-off to obtain greater statistical power for the relatively subtle main contrast of interest (i.e., "threatening" *>* "harmless"). However, this prevented us from examining the main effect of "harmless" + "threatening" assessments to offer evidence regarding current speculations on the amygdala being a novelty detector, rather than a threat detector (Blackford et al., 2010). Our use of only male volunteers also means that we are unable to comment on the possible gender-specific nature of any activations or behavioral response characteristics.

Finally, our hypotheses in this initial study were restricted to greater brain activations to "threatening" compared with "harmless" stimuli, and brain activations positively correlated with ISCR. Consequently we had no specific *a priori* hypotheses about, and so insufficient power to confidently interpret, activations related to the reverse contrasts ("harmless" greater than "threatening" or negative correlations with ISCR).

# **FUTURE STUDIES**

Whilst the SCR may provide a purer measure of sympathetic activity than heart rate or pupil diameter (Wallin, 1981; Öhman et al., 2000), a future study may benefit from examining more than one of these, as there is evidence that SCR and heart rate may separately code the arousal and valence aspects of affective experience, respectively (Bradley et al., 2001). Future studies may also benefit from a measure of an individual's sensitivity to visceral cues such as heartbeat-detection (Katkin et al., 2001). Katkin and colleagues used backward-masked images of fear-relevant stimuli to show that subjects who could detect their heartbeats performed better than chance at predicting a forthcoming electric shock associated with the conditioned stimuli. Hence, a measure of interoceptive sensitivity to sympathetic arousal could index an underlying trait-bias toward negative interpretations of "ambiguous" stimuli (Richards et al., 2003). These hunches or "gut feelings" may be another important modulator of cognitive evaluation of emotionally salient stimuli (Dalton et al., 2005), and hence important in our understanding of the role of relevant structures such as the insula. Investigation of the strength of this negativity-bias may also benefit from a continuous rating scale of "threat" as opposed to a binary forced-choice metric. Such a continuous rating scale may also be beneficial in separating genuinely "threatening" stimuli from more generally "negative" stimuli, which may have been classified as "threatening" when given a binary choice and have therefore contributed little signal, but potentially problematic noise to the relevant fMRI contrasts.

# **CONCLUSIONS**

In summary, convolving concurrently acquired SCR and fMRI measurements during assessment of potentially threatening

# **REFERENCES**


attentional and attributional bias: an fMRI approach to paranoid delusions. *Psychol. Med.* 30, 873–883.


stimuli allows more sophisticated assessment of the component processes which comprise an "emotional response." Our data are broadly, but not fully in line with previous studies. Hence, further studies are likely required to provide a baseline against which to test future hypotheses about cognitive and autonomic system interaction abnormalities which may underlie various neuropsychiatric disorders.

# **ACKNOWLEDGMENTS**

We thank colleagues from the Academic Unit of Radiology, University of Sheffield and the participants in this study. We also thank Dr. Kwang-Hyuk Lee for statistical advice.

Neuropsychology of fear and loathing. *Nat. Rev. Neurosci.* 2, 352–363.


representation of galvanic skin conductance responses: a functional magnetic resonance imaging study. *J. Neurosci.* 20, 3033–3040.


related content. *Cereb. Cortex* 17, 2828–2840.


activity correlation with skin conductance changes: an fMRI study. *Neuroimage* 17, 1797–1806.


evidence for a psychosis pronenesspersistence-impairment model of psychotic disorder. *Psychol. Med.* 39, 179–195.


ductance orienting. *Neuroreport* 11, 3011–3015.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 21 September 2012; accepted: 17 December 2012; published online: 1 January 2013. 6*

*Citation: Farrow TFD, Johnson NK, Hunter MD, Barker AT, Wilkinson ID and Woodruff PWR (2013) Neural correlates of the behavioral-autonomic interaction response to potentially threatening stimuli. Front. Hum. Neurosci. 6:349. doi: 10.3389/fnhum.2012.00349*

*Copyright © 2013 Farrow, Johnson, Hunter, Barker, Wilkinson and Woodruff. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in other forums, provided the original authors and source are credited and subject to any copyright notices concerning any third-party graphics etc.*

# **APPENDIX**

#### **Table A1 | Stimuli used (Available as online supplementary material).**


*(Continued)*

#### **Table A1 | Continued**


*aLang et al., 1997.*