# **OLFACTORY MEMORY NETWORKS: FROM EMOTIONAL LEARNING TO SOCIAL BEHAVIORS**

**Topic Editors Regina M. Sullivan, Donald A. Wilson, Nadine Ravel and Anne-Marie Mouly**

BEHAVIOURAL NEUROSCIENCE

#### *FRONTIERS COPYRIGHT STATEMENT*

© Copyright 2007-2015 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.

Cover image provided by Ibbl sarl, Lausanne CH

**ISSN** 1664-8714 **ISBN** 978-2-88919-486-5 **DOI** 10.3389/978-2-88919-486-5

### *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

## **OLFACTORY MEMORY NETWORKS: FROM EMOTIONAL LEARNING TO SOCIAL BEHAVIORS**

#### Topic Editors:

**Regina M. Sullivan,** Emotional Brain Institute, Nathan Kline Institute, Orangeburg, NY 10962, USA **Donald A. Wilson,** Emotional Brain Institute, Nathan Kline Institute, Orangeburg, NY 10962, USA **Nadine Ravel,** Lyon Neuroscience Research Center, INSERM U1028, CNRS UMR5292, University Lyon1, France

**Anne-Marie Mouly,** Lyon Neuroscience Research Center, INSERM U1028, CNRS UMR5292, University Lyon1, France

Odors are powerful stimuli that can evoke emotional states, and support learning and memory. Decades of research have indicated that the neural basis for this strong "odor-emotional memory" connection is due to the uniqueness of the anatomy of the olfactory pathways. Indeed, unlike the other sensory systems, the sense of smell does not pass through the thalamus to be routed to the cortex. Rather, odor information is relayed directly to the limbic system, a brain region typically associated with memory and emotional processes. This provides olfaction with a unique and potent power to influence mood, acquisition of new information, and use of information in many different contexts including social interactions. Indeed, olfaction is crucially involved in behaviors essential for survival of the individual and species, including identification of predators, recognition of individuals for procreation or social hierarchy, location of food, as well as attachment between mating pairs and infant-caretaker dyads. Importantly, odors are sampled through sniffing behavior. This active sensing plays an important role in exploratory behaviors observed in the different contexts mentioned above. Odors are also critical for learning and memory about events and places and constitute efficient retrieval cues for the recall of emotional episodic memories.

This broad role for odors appears highly preserved across species. In addition, the consistent early developmental emergence of olfactory function across diverse species also provides a unique window of opportunity for analysis of myriad behavioral systems from rodents to nonhuman primates and humans. This, when combined with the relatively conserved organization of the olfactory system in mammals, provides a powerful framework to explore how complex behaviors can be modulated by odors to produce adaptive responses, and to investigate the underlying neural networks.

The present research topic brings together cutting edge research on diverse species and developmental stages, highlighting convergence and divergence between humans and animals to facilitate translational research.

# Table of Contents


Lionel Pazart, Alexandre Comte, Eloi Magnin, Jean-Louis Millot and Thierry Moulin

*24 A Review on the Neural Bases of Episodic Odor Memory: From Laboratory-Based to Autobiographical Approaches*

Anne-Lise Saive, Jean-Pierre Royet and Jane Plailly


and David Sander

*64 Olfactory Preference Conditioning Changes the Reward Value of Reinforced and Non-Reinforced Odors*

Nicolas Torquet, Pascaline Aimé, Belkacem Messaoudi, Samuel Garcia, Elodie Ey, Rémi Gervais, A. Karyn Julliard and Nadine Ravel

*73 Modulation of Olfactory Sensitivity and Glucose-Sensing by the Feeding State in Obese Zucker Rats*

Pascaline Aimé, Brigitte Palouzier-Paulignan, Rita Salem, Dolly Al Koborssy, Samuel Garcia, Claude Duchamp, Caroline Romestaing and A. Karyn Julliard


*109 Interactions with the Young Down-Regulate Adult Olfactory Neurogenesis and Enhance the Maturation of Olfactory Neuroblasts in Sheep Mothers*

Maïna Brus, Maryse Meurisse, Matthieu Keller and Frédéric Lévy

*120 Differential Memory Persistence of Odor Mixture and Components in Newborn Rabbits: Competition between the Whole and Its Parts*

Gérard Coureaud, Thierry Thomas-Danguin, Frédérique Datiche, Donald A. Wilson and Guillaume Ferreira


Ying Huo, Qi Fang, Yao-Long Shi, Yao-Hua Zhang and Jian-Xu Zhang

*166 Mouse Grueneberg Ganglion Neurons Share Molecular and Functional Features with C. Elegans Amphid Neurons*

Julien Brechbühl, Fabian Moine and Marie-Christine Broillet

*178 Sniff Adjustment in an Odor Discrimination Task in the Rat: Analytical or Synthetic Strategy?*

Emmanuelle Courtiol, Laura Lefèvre, Samuel Garcia, Marc Thévenet, Belkacem Messaoudi and Nathalie Buonviso

*188 The Olfactory Bulb Theta Rhythm Follows All Frequencies of Diaphragmatic Respiration in the Freely Behaving Rat*

Daniel Rojas-Líbano, Donald E. Frederick, José I. Egaña and Leslie M. Kay


Camille Ferdenzi, Johan Poncelet,Catherine Rouby and Moustafa Bensafi

*234 Properties and Mechanisms of Olfactory Learning and Memory* Michelle T. Tong, Shane T. Peace and Thomas A. Cleland


## Olfactory memory networks: from emotional learning to social behaviors

*Regina M. Sullivan1,2, Donald A. Wilson1,2,3, Nadine Ravel <sup>4</sup> and Anne-Marie Mouly4 \**

*<sup>1</sup> Emotional Brain Institute, Nathan Kline Institute, Orangeburg, NY, USA*

*<sup>2</sup> Child and Adolescent Psychiatry, The Child Study Center, New York University Langone Medical Center, New York, NY, USA*

*<sup>3</sup> Neuroscience and Physiology, Sackler Institute, New York University School of Medicine, New York, NY, USA*

*<sup>4</sup> Lyon Neuroscience Research Center, INSERM U1028, CNRS UMR5292, University Lyon1, Lyon, France*

*\*Correspondence: ammouly@olfac.univ-lyon1.fr*

#### *Edited and reviewed by:*

*Nuno Sousa, University of Minho, Portugal*

**Keywords: olfaction, olfactory memory, social odors, sniffing behavior, odor preference, odor aversion**

Odors are powerful stimuli that can evoke emotional states, and support learning and memory. Decades of research have indicated that the neural basis for this strong "odor-emotional memory" connection is due to the uniqueness of the anatomy of the olfactory pathways. Indeed, unlike the other sensory systems, the sense of smell does not pass through the thalamus to be routed to the cortex. Rather, odor information is relayed directly to the limbic system, a brain region typically associated with memory and emotional processes. This provides olfaction with a unique and potent power to influence mood, acquisition of new information, and use of information in many different contexts including social interactions. Indeed, olfaction is crucially involved in behaviors essential for survival of the individual and species, including identification of predators, recognition of individuals for procreation or social hierarchy, location of food, as well as attachment between mating pairs and infant-caretaker dyads. Importantly, odors are sampled through sniffing behavior. This active sensing plays an important role in exploratory behaviors observed in the different contexts mentioned above. Odors are also critical for learning and memory about events and places and constitute efficient retrieval cues for the recall of emotional episodic memories.

This broad role for odors appears highly preserved across species. In addition, the consistent early developmental emergence of olfactory function across diverse species also provides a unique window of opportunity for analysis of myriad behavioral systems from rodents to nonhuman primates and humans. This, when combined with the relatively conserved organization of the olfactory system in mammals, provides a powerful framework to explore how complex behaviors can be modulated by odors to produce adaptive responses, and to investigate the underlying neural networks.

The present research topic brings together cutting edge research on diverse species and developmental stages, highlighting convergence and divergence between humans and animals to facilitate translational research. It is composed of 25 articles and encompasses 5 sections: human olfaction, odor preferences and aversions, odors and social behavior, olfaction and sniffing, and olfactory memory.

### **HUMAN ODOR MEMORY AND PERCEPTION**

Pazart et al. (2014) designed an fMRI study to describe the differences in brain activation in wine experts compared to control subjects. They observed specific areas activated in the experts' brains during all phases of wine tasting and reported that wine experts showed a more immediate and targeted sensory reaction to wine stimulation than control subjects. Saive et al. (2014a) reviewed and discussed the different ways of investigating episodic memory in human, ranging from autobiographical approaches to laboratory-based approaches including more ecological paradigms. In an accompanying article, Saive et al. (2014b) presented a novel laboratory-based approach to study episodic memory, which led to the formation and subsequent retrieval of an integrated and multimodal memory of episodes comprising odors (What) localized spatially (Where) within a visual context (Which context). Li (2014) then reviewed rodent and human research in olfactory aversive conditioning, promoting a sensory cortical model of threat perception. By elucidating threat encoding in the sensory cortex, this proposed model may provide new insights into the pathophysiology of emotional disorders, pointing to a concrete clinical intervention target in the sensory cortex. Pool et al. (2014) investigated whether odor discrimination abilities could be enhanced in humans through learning. They used an appetitive Pavlovian conditioning paradigm, during which one of two perceptually non-distinguishable odors was associated with a rewarding taste. They reported a dissociation between explicit perception and physiological reactions: although participants were not able to explicitly perceive a difference, they reacted faster, inhaled more and had higher skin conductance responses when confronted with the reward-associated odor compared to its similar version that was never associated with the reward.

#### **ODOR PREFERENCES AND AVERSION**

Torquet et al. (2014) designed a new paradigm to study conditioned olfactory preference (odor-sugar) in rats and investigate resulting changes in the hedonic valence of the odors. They showed that discriminative appetitive conditioning induces not only a preference for the reinforced odor, but also a stable devaluation of the non-reinforced stimulus. Aimé et al. (2014) investigated the modulation of olfactory sensitivity and glucose-sensing by the feeding state in obese Zucker rats using a conditioned odor aversion learning paradigm. The data show that obese rats displayed a higher olfactory sensitivity and a 2 fold higher concentration of glucose in the olfactory bulb and cortex than in control lean rats, thus providing strong arguments toward establishing the OB glucose-sensing as a key factor for sensory olfactory processing. Using the same odor aversion paradigm, Ferry (2014) further showed that the orexinergic system which triggers food intake, acts like fasting to increase not only olfactory sensitivity, but also the strength of the odor-malaise association. Finally, Fitzgerald et al. (2014) explored the role of the olfactory tubercle, a structure with known anatomical connectivity with both brain reward and olfactory structures, in regulating odor-motivated behaviors. They confirmed that electrical stimulation of the olfactory tubercle was rewarding and possessed the capacity to alter odor-directed preference behaviors, supporting the notion that the olfactory tubercle is integral to motivated behaviors and likely involved in odor hedonics.

### **ODORS AND SOCIAL BEHAVIOR**

Brus et al. (2014) investigated the influence of parturition and learning of the lamb odor on neurogenesis in sheep mothers. They reported that learning of the olfactory signature of the lamb, which occurs during the first mother/young interactions, induces a down-regulation in olfactory neurogenesis associated with an enhancement of olfactory neuroblasts maturation. The authors made the assumption that fewer new neurons decrease cell competition in the olfactory bulb and enhance maturation of those new neurons selected to participate in the learning of the young odor. Using another animal model of mother-young interaction in rabbits, Coureaud et al. (2014) addressed the question of memory persistence of odor mixture and components in newborn rabbits. Interacting with the mother during the daily nursing, newborn rabbits experience her body odor cues including the mammary pheromone which promotes the very rapid appetitive learning of simple or complex stimuli (odorants or mixtures) through associative conditioning. The data showed that newborn rabbits have access to both elemental and configural information in certain odor mixtures, and competition between these distinct representations of the mixture influences the persistence of their memories. Social transmission of food preference is another example of social learning involving the olfactory modality. Nicol et al. (2014) showed that this learning can occur under anesthesia and promotes extensive, but spatially distinct, changes in mitral cell networks to both cued and uncued odors as well as in evoked glutamate and GABA release. These data suggest that olfactory learning under anesthesia in mice is supported by broadly similar changes in olfactory bulb encoding as shown previously in conscious paradigms. Another kind of social odor is provided by predator odors. Takahashi (2014) reviewed current research on the olfactory and neural systems that activate fear elicited by predator odors. Interconnections from olfactory systems to brain circuits activated by predator odor are discussed in relation to autonomic, endocrine, and fear-related unconditioned and conditioned fear. Using chronic exposure to a predator or its scent as a social stress, Huo et al. (2014) investigated the role of serotonin as a modulator of the effects of this stress on male-male aggression and sexual attractiveness of urine odor. They found that exposure to a predator or its scent causes down-regulation of male–male aggression and sexual attractiveness of urine odor, and that such inhibitory effects were reduced or abolished in mice lacking brain serotonin. Finally in mice the Grueneberg ganglion located at the tip of the nose close to the entry of the naris, has been identified as a chemodetector of alarm pheromones. Brechbühl et al. (2013) investigated the conserved multisensory modalities of mouse Grueneberg ganglion neurons. They found striking similarities between mouse Grueneberg ganglion neurons and nematode amphid neurons, suggesting that the ability of an organism to detect cues from similar origin occurs in a cluster of specialized olfactory neurons that has been conserved throughout evolution.

#### **OLFACTION AND SNIFFING**

Olfaction and respiration are tightly linked since odors are sampled through sniffing behavior. Courtiol et al. (2014) reported that sniffing behavior is not only a matter of odorant sorption properties. Indeed a given odorant was sniffed differently depending on the odor pair in which it was presented. These data suggest that sniffing is adjusted in a synthetic manner that is dependent on the context in which the odorant is presented. Rojas-Líbano et al. (2014) monitored sniffing behavior, simultaneously with the local field potential of the olfactory bulb in rats moving freely in a familiar environment. They found that the local field potential in the olfactory bulb represents the sniff cycle with high reliability at every sniff frequency and can therefore be used to study the neural representation of motor drive in a sensory cortex. Under some circumstances, sniffing behavior can also be used as an index of cognitive processes providing a useful tool at very young ages. Indeed, Boulanger Bertolus et al. (2014) monitored freezing and sniffing behavior during odor fear conditioning at different ages of the rat's life, in order to investigate the learning of interval duration between two events in this paradigm. They observed temporal patterns for freezing and/or respiration curves in pups as young as 12 days post-natal, suggesting that infant rats are able to encode time durations as well as, and as quickly as, adults despite their immature brain. In active rodents, sniffing is often phase locked with other orofacial sensorimotor behaviors like whisking, and head movements. Sirotin et al. (2014) further examined the relationship between sniffing and ultrasonic vocal output of rats in a social environment. They found that ultrasonic vocalization of the 50 kHz family is largely restricted to periods of active sniffing (5–10 Hz), with the calls produced exclusively during exhalations and causing an instantaneous reduction in sniff rate. These results show that ultrasonic vocalizations are an integral part of the rhythmic orofacial behavioral ensemble. In humans, there is psychophysiological evidence that sniffing is modulated by subjective pleasantness of an odor: sniff duration and sniff volume increase when pleasant odors are sampled compared to unpleasant ones. Ferdenzi et al. (2014) investigated how repeated exposure to odors can affect their pleasantness. For this, sniff duration and volume were recorded and paired with ratings of odor pleasantness and intensity. The data showed that affective habituation occurs with repeated exposure, which can be observed both at the self-reported level and on sniffing behavior.

### **OLFACTORY MEMORY**

Tong et al. (2014)reviewed some established molecular and structural mechanisms of memory with a focus on the time courses of post-conditioning molecular processes. They described the properties of odor learning intrinsic to the olfactory bulb and reviewed the utility of the olfactory system of adult rodents as a memory system in which to study the cellular mechanisms of cumulative learning. Recent work has begun exploring the role of sleep in olfactory memory. Barnes and Wilson (2014) reviewed recent evidence of sleep-dependent changes in olfactory system structure and function which contribute to odor memory and perception. During slow-wave sleep, the piriform cortex becomes hypo-responsive to odor stimulation and instead displays sharp-wave activity similar to that observed within the hippocampal formation. Furthermore, the functional connectivity between the piriform cortex and other cortical and limbic regions is enhanced during slow-wave sleep compared to waking. This combination of conditions may allow odor memory consolidation to occur during a state of reduced external interference and facilitate association of odor memories with stored hedonic and contextual cues. Olfactory structures have been early reported to exhibit oscillatory population activities which have been proposed to subserve memory processes such as encoding, consolidation and retrieval. Martin and Ravel (2014) reviewed our current knowledge about the conditions in which two main oscillatory rhythms linked to odor processing, namely beta (15– 40 Hz) and gamma (60–100 Hz) are observed at the first stages of olfactory processing, the olfactory bulb and the piriform cortex. Based on a critical reexamination of those data, the authors propose hypotheses on the functional involvement of beta and gamma oscillations for odor perception and memory. It is well known from the literature that sensory neural activity is highly context dependent and shaped by experience and expectation. Mandairon et al. (2014) further showed that association of an odorant with a visual context in mice allowed the visual context alone to elicit a behavioral response similar to that elicited by the olfactory stimulus when it was initially presented, and a neural activation pattern in the olfactory bulb that was highly correlated with that elicited by the associated odorant, but not other odorants. The authors concluded that in rodents, neural representation of an odorant in primary sensory areas can be elicited in its absence by exposure to the context to which the odorant was previously associated, further suggesting that rodents can build internal representation of the olfactory stimulus. Finally, Gelperin (2014) underlined that chemosensory processing, and particularly olfactory information processing, is a particularly attractive modality within which to seek comparative insights into cognitive processes underlying learning and memory. He concluded that the study of olfactory information processing may be the most general and fruitful approach to the study of comparative cognition, including consciousness, in the majority of vertebrate animal species.

**REFERENCES**

Aimé, P., Palouzier-Paulignan, B., Salem, R., Al Koborssy, D., Garcia, S., Duchamp, C., et al. (2014). Modulation of olfactory sensitivity and glucose-sensing by the feeding state in obese Zucker rats. *Front. Behav. Neurosci.* 8:326. doi: 10.3389/fnbeh.2014.00326


**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 January 2015; accepted: 01 February 2015; published online: 17 February 2015.*

*Citation: Sullivan RM, Wilson DA, Ravel N and Mouly A-M (2015) Olfactory memory networks: from emotional learning to social behaviors. Front. Behav. Neurosci. 9:36. doi: 10.3389/fnbeh.2015.00036*

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

*Copyright © 2015 Sullivan, Wilson, Ravel and Mouly. 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.*

## An fMRI study on the influence of sommeliers' expertise on the integration of flavor

#### *Lionel Pazart <sup>1</sup> \*†, Alexandre Comte1,2,3†, Eloi Magnin2,3, Jean-Louis Millot <sup>2</sup> and Thierry Moulin1,2,3*

*<sup>1</sup> Inserm Clinical Investigation Centre 1431, Clinical Investigation Centre, Besançon University Hospital, Besancon, France*

*<sup>2</sup> Laboratoire de Neurosciences, (EA-481), University of Franche-Comté, Besancon, France*

*<sup>3</sup> Département de Recherche en Imagerie Fonctionnelle, Besançon University Hospital, Besancon, France*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Wen Li, University of Wisconsin-Madison, USA*

#### *\*Correspondence:*

*Lionel Pazart, Inserm Clinical Investigation Centre 1431, Besançon University Hospital, 2 Place St Jacques, 25030 Besancon, France e-mail: lpazart@chu-besancon.fr*

*†These authors have contributed equally to this work.*

Flavors guide consumers' choice of foodstuffs, preferring those that they like and meet their needs, and dismissing those for which they have a conditioned aversion. Flavor affects the learning and consumption of foods and drinks; what is already well-known is favored and what is new is apprehended. The flavor of foodstuffs is also crucial in explaining some eating behaviors such as overconsumption. The "blind" taste test of wine is a good model for assessing the ability of people to convert mouth feelings into flavor. To determine the relative importance of memory and sensory capabilities, we present the results of an fMRI neuro-imaging study involving 10 experts and 10 matched control subjects using wine as a stimulus in a blind taste test, focusing primarily on the assessment of flavor integration. The results revealed activations in the brain areas involved in sensory integration, both in experts and control subjects (insula, frontal operculum, orbitofrontal cortex, amygdala). However, experts were mainly characterized by a more immediate and targeted sensory reaction to wine stimulation with an economic mechanism reducing effort than control subjects. Wine experts showed brainstem and left-hemispheric activations in the hippocampal and parahippocampal formations and the temporal pole, whereas control subjects showed activations in different associative cortices, predominantly in the right hemisphere. These results also confirm that wine experts work simultaneously on sensory quality assessment and on label recognition of wine.

#### **Keywords: fMRI, flavor, expertise, wine, olfaction pathways, taste**

### **INTRODUCTION**

Flavor is a crucial subject of study for understanding eating behavior, for the prevention of obesity, overdrinking and other eating disorders, and for the foods and drinks industry. The cortical integration of olfactory and gustatory information could modulate mechanisms involved in food selection and emotional reactions relating to the chemical senses (Fu et al., 2004). Flavor of foodstuffs refers to this combination of sensations perceived inside the mouth, combining taste (savor) and smells (aromas), as well as trigeminal somatosensory perception and thermal perception (Auvray and Spence, 2008; Prescott, 2012). Several previous neuroimaging studies analyzed brain regions activated by intrinsic cues of flavor (Cerf-Ducastel and Murphy, 2001; O'Doherty et al., 2001; Kobayashi et al., 2004; Kikuchi et al., 2005; Boyle et al., 2007a,b) and the convergence of taste and retronasal olfaction was mainly elicited the anterior part of the orbitofrontal cortex (de Araujo et al., 2003; Small et al., 2004; Small and Prescott, 2005). More precisely, flavor integration following retronasal stimulation may involve brain structures like the insula, the frontal operculum and the caudal orbitofrontal cortex (OFC) (Cerf-Ducastel and Murphy, 2001; de Araujo et al., 2003; Small et al., 2004; Small and Prescott, 2005) but also the amygdala, and cerebellum (Cerf-Ducastel and Murphy, 2001).

Other neuroimaging studies confirmed the high effects on taste and flavor perception by different extrinsic cues such as the appearance of the foodstuff, packaging design, brand name, geographical origin, price, subjective flavor preferences (McClure et al., 2004; Plassmann et al., 2008; Kühn and Gallinat, 2013; Okamoto and Dan, 2013; Van den Bosch et al., 2014) and also the simple evocation of the odor or product name (Royet et al., 2013a; Bensafi et al., 2014).

One main factor that might influence the effect of either intrinsic or extrinsic cues could be the strength of the taste/flavor memory associated with the cue (Okamoto and Dan, 2013). Indeed, the extraordinary performance of experts in many matters (chess, bridge, music, wine etc.) raises the question of the origin of their faculties. It is often found that experts and novices use different criteria to categorize domain-specific problems, in that novices use simplistic surface features whereas experts use underlying principles (Vicente and Wang, 1998).

Wine expertise provides an interesting field in which to test theories of skill acquisition since it is generally believed to be based mainly on advanced perceptual skills rather than cognitive ones, such as categorical knowledge or episodic memory (Hughson and Boakes, 2002; Saive et al., 2014). Perceived quality of a wine is dependent on consumers' level of expertise (Sáenz-Navajas et al., 2013). Experience tends to generate idiotypic recollections, to which new wines are compared (Hughson and Boakes, 2002). Accordingly, odor experts who are trained daily can acquire better olfactory sensitivity, and thus olfactory mental imagery capacities develop with practice and do not result from innate skill (Plailly et al., 2012). During the creation of mental images of odors, expertise influences not only the primary olfactory area (piriform cortex) but also the OFC and the hippocampus, regions that are involved in memory and the formation of complex sensory associations, respectively (Royet et al., 2013b; Saive et al., 2014). In these areas, the magnitude of activation was negatively correlated with experience: the greater the level of expertise, the lower the activation of these key regions (Royet et al., 2013b). Nevertheless, in wine expertise several behavioral studies (Brochet and Dubourdieu, 2001; Morrot et al., 2001; Hughson and Boakes, 2002; Parr et al., 2002; Ballester et al., 2008; Brand and Brisson, 2012) have compared wine experts and novices, and surprisingly no difference in olfactory sensitivity was revealed between them. This discrepancy between odor experts and wine experts could be explained by the needed integration of several sensory modalities for wine: sight, orthonasal olfaction, somesthesia, and chemical senses including trigeminal sensitivity, taste, and retronasal olfaction (Brochet and Dubourdieu, 2001; Morrot et al., 2001). Parr et al. (2002) demonstrated that wine experts have similar sensitivity for wine-related components such as tannin or alcohol, and similar odorant naming abilities compared to those of wine novices but superior explicit identification and memory recognition for wine-relevant odorants. For some authors, this greater ability of wine experts to recognize and identify odors is probably due to better semantic knowledge (Hughson and Boakes, 2002), however this remains an open question (Parr et al., 2004; Ballester et al., 2008). In this way, in addition to the structures involved in taste and retro-olfaction, those mechanisms might involve in wine experts a predominance of the left hemisphere, involved in analytic and linguistic treatment, and probably the temporal lobes and hippocampus, involved in episodic and semantic memory, related to previous experiences. In contrast, Parr et al. demonstrated the importance of perceptual skill, namely sensory memory for odorant and trigeminal perception, rather than semantic memory ability, in wine-relevant olfactory expertise (Parr et al., 2004).

To determine the relative importance of different types of memory and sensory capabilities in wine expertise, few studies have been performed with neuroimaging techniques and until present Castriota-Scanderbeg's study might be the only study involving wine experts and novices (Castriota-Scanderbeg et al., 2005).

Castriota-Scanderbeg's study compared via fMRI the brain activations of two groups of 7 wine experts (sommeliers) and 7 novices who received randomly 2 ml bolus of 3 Italian wine or glucose solution via a multi-channel device inserted in the mouth. Sommeliers showed greater activity in the left insula and orbitofrontal cortex than the novices. The principal areas activated in the novices were the primary gustatory cortex and the regions associated with emotional processing (Castriota-Scanderbeg et al., 2005). This study confirmed the involvement of olfactory memory in wine assessment by wine professionals. However, the primary olfactory area (piriform cortex) was not activated either in wine experts or novices, and surprisingly no activation occurred between the wine and glucose solution in mouth, i.e., during "taste phase." Significant differences were observed only after swallowing a bolus, i.e., during the after-taste phase. Thus, the integration of flavor seems to be delayed contrary to what is usually observed by sommeliers and reserved to expert unlikely results of behavioral experiments.

Our hypothesis is that a specific effect of expertise may be observed early during the taste phase anticipating the after-taste phase, and requiring flavor memory and episodic memory rather than semantic memory. Therefore, we would like to replicate Castriota-Scanderbeg's study with a tasteless comparator to avoid the possible overshadowed effect of glucose solution on the taste part of brain activation. Thus, our study objective was to evaluate brain activity with fMRI during the taste and after-taste phases of wine tasting vs. a tasteless water comparator directly delivered into the mouth of a sample of matched-pair experts and novices tasters in order to determine the relative importance of memory and sensory capacities in wine flavor integration.

### **MATERIALS AND METHODS SUBJECTS**

We recruited 20 subjects, 10 famous sommeliers (seven men and three women) from France and Switzerland and 10 matched controls. All sommeliers had been active professionals for at least 5 years. Most had received awards such as "Best sommelier in the world," "Best sommelier in Europe," "Best sommelier in France," or "Best sommelier in Switzerland" and were working in prestigious restaurants in either France or Switzerland. Each sommelier was matched with a control subject of the same sex and same age (±5 years) because chemosensory abilities can vary with gender and age (Doty, 1989; Yousem et al., 1999; Brand and Millot, 2001; Wang et al., 2005; Lundstrom and Hummel, 2006). Subjects were all aged between 24 and 67 years. In addition, the control subjects were from the same region as their matched expert in order to limit bias induced by the difference of regional flavor habits in each pair. All subjects were right-handed and nonsmokers, as smoking habits can also influence sensory abilities (Katotomichelakis et al., 2007). All subjects underwent a medical examination to screen for MRI contra-indication and for any possible gustatory or olfactory dysfunctions before the study. The protocol was approved by the local ethics committee (Comité de Protection des Personnes CPP Est II) and declared to the national authority (N◦ UF: 1013; DGS 2006/0494). Written informed consent was obtained from all participants.

#### **CHOICE OF STIMULI AND STIMULUS DELIVERY**

Two wines (one white chardonnay variety, Arbois 2004, and one red "black pinot" variety, Côte du Jura 2006) were chosen for their good sensorial qualities from an expert point of view by an experienced sommelier (C. Menozzi) who was not participating in the study. He also tasted different types of water (including distilled water and physiologic serum) and deemed the local water to be the most suitable for control (in terms of salinity, tastelessness, and low minerality) and rinsing. A multi-channel custom-built gustometer was used to deliver the wine and water to the subjects (Andrieu et al., 2014). This device comprises three reservoirs, one for each type of liquid, and a computer-controlled pneumatic distributor, which dispatches air toward several exits. Each reservoir has its own opaque polythene tube, which transports the liquid toward the subject's mouth. The bolus was delivered in two stages: (1) the pneumatic distributor injected air into the appropriate reservoir in order to push the required volume of liquid into the tube; (2) air was injected into the connected tube, pushing the liquid into the subject's mouth. The bolus (2 ± 0.11 ml) was delivered in 0.34 ± 0.06 s. The three tubes were contained within a larger silicon tube (10 mm exterior diameter). Consequently, the subject only felt one tube in their mouth and was unable to see which liquid was being delivered.

Before undergoing MRI, each participant tested the device used for the administration of wine samples with the reference solution (2 ml samples of water) to familiarize him or herself with the experimental tasting procedure. This involved lying in a horizontal position and limiting any lip or jaw movements. Subjects were asked to swish water in their mouths, just as they were required to do with wine or water during the scan session. The same paradigm was used for the acquisition of images.

#### **EXPERIMENTAL PROTOCOL**

To avoid confusion between "taste as in common usage," and "taste as a unimodal sensation," hereafter we will follow the convention in chemosensory studies where "taste" refers to a unimodal sensation.

During the first phase (taste), subjects had to swish either wine or water (bolus of 2 ml) in their mouth for 7 s. At the end of the taste phase, an auditory cue prompted the tasters to swallow the liquid, signaling the beginning of the second phase (after-taste), which lasted 13 s. Each 2 ml bolus of wine was repeated twice before rinsing. There was a rest period of 15 s after each rinsing, resulting in a block of 1 min 15 s ([7 + 13 s] × 3 + 15 s). Each subject performed five blocks. The durations of the taste and after-taste periods were the same as in a previously validated study (Castriota-Scanderbeg et al., 2005). During the scan, subjects were asked to keep their eyes closed. Participants received no information on the number of wines chosen or their characteristics. In addition, the order in which the red and white wines were delivered was randomized according to test at least once each variety of wine and was therefore unknown to the subjects.

#### **MRI DATA ACQUISITION**

The functional MRI study was performed on a 3-Tesla (GE Healthcare Signa HDxt, Milwaukee, WI) MR system with a standard 40 mT/m gradient using blood–oxygen level-dependent (BOLD) fMRI. Foam cushions were used to minimize head movements within the coil. The experiment began with the acquisition of a high-resolution, T1-weighted, 3-dimensional anatomical scan (BRAVO sequence). This scan was acquired in 134 slices with 0*.*47 × 0*.*47 × 1*.*2 mm resolution. Functional images were then obtained parallel to the anterior-posterior commissure line, covering the entire cerebrum (30 slices) using an echo planar imaging (EPI) sequence (slice thickness = 4.5 mm; *TR* = 2500 ms; *TE* = 35 ms, matrix = 128 × 128; FoV = 256 mm; Flip Angle = 90◦; phase acceleration factor = 2; auto-shimming).

#### **fMRI DATA ANALYSIS**

Image time-series analysis was performed using BrainVoyager QX 2.1 (Brain Innovation, Maastricht, The Netherlands). The time-series were corrected for slice acquisition time, realigned with their corresponding T1 volumes, warped into standard space (Talairach and Tournoux, 1988), re-sampled into 2 mm isotropic voxels, motion-corrected using Levenberg-Marquarts's least square fit for six spatial parameters, highpass-filtered for removal of low frequency drifts, corrected voxel-wise for linear drifts, and spatially smoothed using a 5-mm full-width at half-maximum Gaussian kernel.

The general linear model (GLM) was computed from the 20 z-normalized volume time courses. For all stimuli of interest, i.e., rest period, taste period and after-taste period, box-car time courses with a value of 1 for the stimuli of interest and values of 0 for the remaining time points were convolved with a theoretical hemodynamic response function (Boynton et al., 1996) and were entered as predictors into the design matrix of the study. Contrast analyses were based on random effects GLMs of the z-normalized volume time courses.

Analyses of the taste and after-taste periods were firstly performed for the entire group (20 subjects) using a statistical threshold of *q*(FDR) *<* 0.01 corrected for multiple comparisons. A minimum cluster size of 48 mm<sup>3</sup> was set. As suggested by Zald and Pardo (2000)for controlling in-mouth non-gustatory factors, we considered the water after swallowing period as a reference for the contrasts of wine vs. water for both periods.

The same analyses were then performed for each group separately (experts and controls). As the number of subjects in each sub-group is half the entire group, another statistic was chosen.

A cluster size threshold yielding the equivalent of a whole-brain corrected for a multiple comparison significance level of *P <* 0*.*05 was used after voxel-wise thresholding at *P <* 0*.*005 (uncorrected). The BrainVoyager Cluster-Level Statistical Threshold Estimator plug-in estimating the overall significance level by determining the probability of false detection through Monte Carlo simulation was used (with 10,000 Monte Carlo iterations).

Finally, for both the taste and after-taste periods, a group comparison was carried out to identify the brain areas affected by the level of expertise in wine appreciation and perception. A statistical extent threshold of *P <* 0*.*05 corrected for multiple comparisons after a voxel-wise thresholding at *P <* 0*.*005 was used.

### **RESULTS**

**Table 1** shows the cerebral activations for the whole sample of subjects (experts and controls) obtained for the contrast "wine minus water" during taste phase and after-taste phase. During the taste phase, activations were found in the insula, the frontal lobe (bilateral motor area and right superior and dorso-lateral prefrontal cortex), pallidum, left parahippocampic gyrus and left thalamus. During the after-taste phase, activations were again present in the insula and in various areas of the frontal lobe including the orbito-frontal cortex.

When considering experts and controls separately, the same contrasts revealed that certain activations were specific to one group of subjects in both the taste and the after-taste periods



*Cerebral activations for the whole sample of subjects (experts and controls) obtained for the contrast "wine minus water" during the taste and after-taste phases [q(FDR) < 0.01 corrected for multiple comparisons]. L, Left; R, Right; K, size of the cluster in number of connected voxels of 1 mm*3*; x, y, z, Talairach coordinates of the maximum peak.*

(**Table 2**). Consequently, further analyses focused on the contrasts between experts and controls during the wine taste phase and the wine after-taste phase, which was the aim of the study.

**Table 3A** shows the activated regions when contrasting experts and control subjects during the wine taste phase. When contrasting experts minus control subjects, activations were observed in the brainstem (left bulbo-pontic junction extended to left trigeminal nucleus), the cerebellum and subcortical areas (locus niger, globus pallidus). Cortical activations were present in the hippocampi, parahippocampal gyri, amygdalae, periamygdal cortex (entorhinal and perirhinal cortex), temporal and occipital lobes and in the right anterior insula. There were more widespread bilateral activations of the parietal lobes in control subjects compared to experts. During the wine taste phase, control subjects activated 18 regions vs. 9 for experts.

**Table 3B** shows the activated regions when contrasting experts and control subjects during the wine after-taste phase. Experts exhibited fewer cerebral activations (in terms of number and size of clusters) compared to controls. These activations involved the right temporal lobe and left hippocampus. In the reverse contrast (control subjects minus experts), activations involved the frontal, temporal and parietal (postcentral gyrus) cortices and the anterior insula. Subcortical activations were restricted to the caudate nucleus.


**Table 2 | Wine minus Water contrast by group (simple main effects).**

*corrected for multiple comparison*

*of the maximum peak.*

 *was used after voxel-wise thresholding*

 *at P < 0.005* 

*(uncorrected).*

 *L, Left; R, Right; K, size of the cluster in number of connected voxels of 1 mm*3*; x, y, z, Talairach coordinates*

 *<*


**Table 3 | Group** 

**comparison**

 **analysis during wine taste and wine after-taste.**

*was used after voxel-wise thresholding*

 *at P < 0.005* 

*(uncorrected).*

 *L, Left; R, Right; K, size of the cluster in number of connected voxels of 1 mm*3*; x, y, z, Talairach coordinates*

 *of the maximum peak.*

### **DISCUSSION**

This study was designed to describe the differences in brain activity in wine experts compared to control subjects, especially during wine tasting as stimuli compared to water, in order to confirm the influence of expertise on flavor integration. In addition, we would like to identify the type of memory used by experts in order to demonstrate experts requiring flavor memory and episodic memory rather than semantic memory. As expected, we observed specific areas activated in the experts' brains during all phases of wine tasting. Structures involved in the sensory and cognitive tasks of the expertise-related process were activated either during the taste phase, corresponding to gustatory and trigeminal sensations, and the after-taste phase, corresponding to retroolfactory sensation. In addition, an unexpected and interesting result during the taste phase was that specific brainstem activation was observed in the expert group, suggesting that expertise can modify sensory treatment in addition to cortical cognitive processes.

#### **ABILITY OF THE PARADIGM TO FOCUS ON FLAVOR**

#### *Choice of reference stimulus*

Direct comparison between our results and those from the only neuroimaging study on wine experts vs. novices, by Castriota-Scanderbeg et al. (2005), should be interpreted with caution because of the difference of wines and reference stimulus in both studies. Instead of using glucose as the reference stimulus, we chose neutral water as the reference stimulus used currently in neuroimaging (Zald and Pardo, 2000) and wine behavioral experiments. The authors of this previous study argued their choice in order to adequately control the sweet components of the gustatory stimulus (but only one of their three tested wines was sweet) and to avoid somatosensory and motor components of the task (Castriota-Scanderbeg et al., 2005). Accordingly, the use of a glucose drink as reference stimulus might have overshadowed the taste part of brain activation while a sweet drink might present similar taste and trigeminal characteristics as tested wines. Furthermore, there is a great consumer preference for sweet wines in many countries and wine experts seem to prefer wines with less added glucose than the novices (Blackman et al., 2010). Sweet drinks might therefore appear more "pleasant" for novices than for experts and introduce another bias since a subjectively pleasant stimulus would have preferentially activated the medial OFC whereas an unpleasant stimulus would have preferentially activated the lateral OFC (Rolls et al., 2003). Moreover, comparing activations in response to a sweet solution or a bitter solution, tasting sweet solution caused greater activations in the OFC whereas tasting a bitter solution resulted in greater activations in the cingulate cortex, operculum and precentral gyrus (Van den Bosch et al., 2014). So in order to minimize bias, we chose exclusively dry wines for stimuli and a tasteless and odorless comparator (neutral water).

#### *Isolation of in-mouth stimulation*

We replicated the design of Castriota-Scanderbeg's study (Castriota-Scanderbeg et al., 2005) using a multi-channel custom-built gustometer to deliver the wine or comparator directly into subjects' mouths, with identical volume (2 ml) and duration of bolus stimuli, the same frame of run and similar fMRI analysis. This design tries to isolate the influence of inmouth sensations on the flavor integration without orthonasal olfactory stimulation or external cues. Similar designs have been adopted in numerous neuroimaging studies on gustation (Kühn and Gallinat, 2013; Van den Bosch et al., 2014) despite wellknown bias due to the supine conditions of the experiment, the lowering perceived intensity functions for taste stimuli affected by the stimulus delivery technique in the MRI scanner (Haase et al., 2009) and by the small amount of stimulus (2 vs. 5–20 ml for sip and spit techniques of behavioral experiments).

In the case of gustatory stimulation, we cannot exclude the possible contribution of other factors, since intraoral stimuli involve several types of processing in addition to gustation: olfaction, somatosensation, and oral movements. This complex interaction makes it difficult to identify the brain regions selectively processing gustation (Kobayashi, 2006).

Rozin (1982) first suggested that olfaction is a dual sense modality because it contributes to the perception of external and internal substances (via orthonasal and retronasal olfaction, respectively). Orthonasal perception can identify objects at a distance, and retronasal perception contributes to flavor and hence food identification in the mouth. These two "olfactory senses" differ physiologically in terms of delivery of odors to the olfactory epithelium (Pierce and Halpern, 1996), but also in terms of connections between senses and cognitive impact. Small and Prescott (2005) have demonstrated that routes of delivery produced differential activations in the insula/operculum, amygdala, thalamus, hippocampus, and caudolateral orbitofrontal cortex in orthonasal *>* retronasal and in the perigenual cingulate and medial orbitofrontal cortex in retronasal *>* orthonasal in response to chocolate, but not lavender, butanol, or farnesol, so that an interaction of route and odorant may be inferred. These findings demonstrate differential neural recruitment depending upon the route of odorant administration (orthonasal or retronasal) and suggest that its effect is influenced by whether an odorant represents a food or not (Small and Prescott, 2005). Small and Prescott (2005) explain these observations by the fact that taste perception is almost always accompanied by olfactory and oral somatosensory perception in the context of eating, whereas olfaction often occurs separately outside the context of eating. Thus, it would appear logical, in the identification of food in the mouth, to combine the food's qualities (savor, palatability and aromas) into a unitary perception (Prescott, 1999). Wine tasting in the mouth typically involves simultaneous gustatory, trigeminal and retro-olfactory information (Brochet and Dubourdieu, 2001). Wine could be a very good model to test the ability of people to perform this convergence of sensorial information in flavor integration. Like Castriota-Scanderbeg, we adopt the terminology of "taste phase" and "after-taste phase" to indicate the period before and after swallowing instead of dissociating responses to taste and smell stimuli. An issue is the succession of both phases (taste and after-taste). In this experiment (and in every experiment studying these two distinct periods) it is unavoidable that both phases are consecutive. In fact, flavor compounds are progressively released from the wine during the mouth process before swallowing particularly with the impregnation by saliva (Salles et al., 2011). So, even if the main perception during the taste phase remains from the gustation and trigeminal stimulation, there is an overlapping with the beginning of retronasal olfaction stimulation. Secondly, mouth movements and swallowing play a role in the enhancement of retronasal odor perception analogous to that played by sniffing in orthonasal perception (Burdach and Doty, 1987). This may lead to a diminishing of observed effects during the after-taste phase due to retro-nasal stimulation (even weak) during the taste phase, and due to memory processes (particularly with experts). Accordingly, the after-taste phase corresponds mainly to retronasal olfactory stimulation with a residual part of the preceding taste stimuli. In addition the 7 s taste phase duration may also lead to the situation where a slow response to the first process can fall into the time window of the after-taste phase and be interpreted as reflecting this second process. However, there is little chance that areas of interest (that is, those involved in olfaction, gustation and memory), if recruited during the taste phase, start to be active only during the after-taste phase. As such, responses even time-locked to the two events are still convoluted. One has then to keep in mind this issue when interpreting results with such paradigms.

In order to secure the exclusivity of in-mouth stimulation, no information was given about the color, types or number of wine since exposure to visual or verbal semantic odor/taste information alone could activate the piriform cortex, the amygdala or the insula (Kobayashi et al., 2004; González et al., 2006; Bensafi et al., 2014).

Nevertheless, wine tasting typically involves three phases contributing to a final synthesis of flavor analysis: firstly, wine tasters normally appreciate the sight of the wine and mainly the color, activating occipital visual areas, secondly they sniff the wine and orthonasal olfaction impacts on gustation and olfactory areas, finally they absorb a small amount of wine in their mouth, trill the wine, aerating it and allowing the flavors to be perceived by retroolfaction before being spat. In our design, we shunted sight and ortho-nasal olfaction and so flavor integration is not complete.

#### **SENSORY/FLAVOR INTEGRATION**

#### *Taste phase and in-mouth sensations*

The main difference between our results and Castriota-Scanderbeg's study is the presence of taste phase activations, especially brain stem responses which suggest that expertise also impacts on basic taste processing. The modification of the paradigm by using water as a control explores this part of flavor processing that was possibly overshadowed by glucose control in Castriota-Scanderbeg's study, as previously explained.

Although no significant differences emerged between wine and glucose either in controls or in sommeliers during the taste period in Castriota-Scanderbeg's study, we found significant cerebral activations in the insula, frontal lobe (bilateral motor area and right superior and dorso-lateral prefrontal cortex), cerebellum, pallidum, left parahippocampic gyrus and left thalamus for the whole sample of subjects (**Table 1**) obtained for the contrast "wine minus water" during taste phase. These regions correspond to an area commonly activated by gustatory stimulation (Kobayashi, 2006): the superior frontal, middle frontal, inferior frontal, precentral, and postcentral gyri, insula/frontal operculum, inferior parietal lobe, and cerebellum and in addition to these regions, the thalamus and the region including the putamen.

Our study showed that the volume of activated regions in the insula/frontal operculum during gustatory stimulation was higher in the left hemisphere than in the right. Although most neuroimaging studies have shown that the right insula is more intensively activated than the left (Cerf-Ducastel and Murphy, 2001; Small and Prescott, 2005; Kobayashi, 2006) some studies have shown that activation in the left insula could be equal to or more dominant than the right insula (Kinomura et al., 1994; Francis et al., 1999).

There is no activation of the OFC during the taste phase either for experts or novices (**Table 2**), but a passive gustatory stimulus may not always be sufficient to activate the region (Kobayashi et al., 2004).

When considering the wines as stimuli and comparing experts to control subjects (experts minus control subjects, **Table 3**), there appeared to be involvement of the bulbo-pontic junction and trigeminal nuclei in the brainstem, suggesting specific chemosensory information processing in experts than controls. Motor activity induced by swallowing may have engaged brainstem activation. However, both groups had to swish and swallow the bolus of wine and no activations were found in the motor cortical area for this contrast. Although rarely mentioned, this result highlights the importance of trigeminal sensitivity in addition to gustatory perception in wine analysis. Indeed, several descriptors used in wine tasting (such as astringent, bitter, spicy, sharp or sweet) are typically trigeminal-type descriptors (Laska et al., 1997). Trigeminal activations were not mentioned by Castriota-Scanderbeg et al. (2005), which may have been due to the sweet type of reference stimuli and one of the three tested wines. It is somewhat surprising to find different sensory processing in experts compared to controls in this first stage of brainstem integration of the stimuli.

Activity was observed in the amygdala, enthorhinal and perirhinal cortices and anterior insula. Activation of the anterior insula is congruent with the taste phase as this area is involved in the integration of multimodal input such as olfactory (Sobel et al., 2003), gustatory (Small et al., 1997, 1999) and trigeminal (Lombion et al., 2009) stimuli. Its role in the hedonic evaluation of chemosensory stimulations (Fulbright et al., 1998; Small et al., 2001) and discrimination processes (Bengtsson et al., 2001) has also been suggested. Activity in the amygdala may have corresponded to the selective perception of olfactory stimuli via the retronasal pathway which started in experts before swallowing, since its activation strongly characterizes olfactory processes (Zatorre et al., 1992; Sobel et al., 2003). Moreover, psychophysical investigations in humans and behavioral work in animals have shown that the taste system plays an integral role in odor processing. While there is evidence to support the anticipation of taste-like properties by odors, there have been few reports of the acquisition of odor-like properties by taste (Prescott, 2012). In animals, some authors have demonstrated that taste input affects olfactory processing via a specialized "association" area (Desgranges et al., 2010). However, other works in conscious rats have shown that the gustatory system directly influences olfactory processing in the primary olfactory cortex (Maier et al., 2012). These results identify the posterior olfactory (piriform) cortex as a likely site for gustatory influences on olfactory processing.

We were surprised by the apparent lack of activation of the piriform cortex in the study by Castriota-Scanderbeg, in novices and experts, in neither the taste phase nor the after-taste phase (Castriota-Scanderbeg et al., 2005). Surprisingly, we found a similar absence in our study. However, as the threshold we used was maybe too strict; to enhance the piriform cortex we performed a ROI analysis centered on it. At a low threshold an activation is observed in the right piriform cortex for the contrast wine minus water for all subjects during the after-taste phase. Peak is obtained at 25, 9, −14 (Talairach coordinates), *t* = 2*.*24, *p* = 0*.*037. A cluster size of 11 voxels of 1 mm<sup>3</sup> is observed when the threshold is set at *p* = 0*.*05 (uncorrected), and a cluster size of 2 voxels of 1 mm<sup>3</sup> at *<sup>p</sup>* <sup>=</sup> <sup>0</sup>*.*04 (uncorrected). With such a small activated volume combined with such a poor statistic, we could consider that there is no activation of the piriform cortex in our study.

The piriform cortex is a small structure in humans and its proximity to the insular lobe may render identification of activations in this area difficult. This area corresponds to the retro-nasal olfactory process mainly during the after-taste phase. Although animal studies have identified the posterior olfactory (piriform) cortex as a likely site for gustatory influences on olfactory processing (Maier et al., 2012), the piriform cortex may respond preferentially to orthonasal odors, reflecting its role in olfaction, enhanced by sniffing (Zatorre et al., 1992; Sobel et al., 1998). Piriform activation by an odorant stimulus administered in solution form into the mouth (retronasal olfactory pathway) was found inconsistently in the literature (Small and Prescott, 2005). The reference study by Cerf-Ducastel and Murphy (2001) showed activation of right piriform cortex in the group analysis of the 6 involved subjects, but there was inconstancy at an individual level with activation of the left piriform in one subject, the right piriform in another, and both sides for a third subject, i.e., in less than half of the sample, and activations were found only with one stimulant (citral) among four. Some authors discuss in detail the inconstancy of the activation of this region in neuroimaging studies of olfaction (Small and Prescott, 2005). The inconstancy of activation of primary olfactory structures could be due to many reasons including the anatomical variability of the inferior frontal and lateral temporal areas, technical conditions, type of stimulus used (odorant in aqueous solution), the single retronasal pathway and the short process with adaptation and/or habituation effects. In addition, the initial amplitude of the activation decreases from block to block when using a block paradigm (Sobel et al., 2003). In this way, our results were similar with previous studies that showed that the piriform cortex would not be activated during the taste and flavor integration phase (Small et al., 1997; O'Doherty et al., 2001). This result could also be explained by the results of Small and Prescott (2005) who demonstrated differential neural recruitment depending on the route of odorant administration (ortho or retronasally) and by some behavioral experiments exploring the extent to which the aroma or non-volatile fractions are responsible for the overall flavor differences of wines perceived in-mouth (Sáenz-Navajas et al., 2012; Villamor and Ross, 2013). A study performed under three different conditions (nose-close, retronasal perception only and retro- and orthonasal perception) have clearly shown that, globally, aroma perception is not the major driver of in-mouth sensory perception of red wine, which is undoubtedly primarily driven by the perception of astringency (Sáenz-Navajas et al., 2012). So we can attribute the absence of piriform activation mostly to the technical condition of supine administration of a small bolus of odorant in solution via a tube in mouth by shunting the orthonasal pathway. Perhaps wine experts can dissociate in the brain the three classical assessments of a wine (sight, orthonasal olfaction and then in-mouth sensations) before they make a synthesis.

#### *After-taste phase and flavor integration*

Firstly, our results in all subjects (experts and controls), showed activations in the insula, the operculum and the orbito-frontal cortex (**Figure 1**), which are all involved in taste/odor integration (Small et al., 1997, 2007; Rolls, 2008; Bender et al., 2009) as the key nodes of the "flavor network" (Small and Prescott, 2005). These regions represent the primary, secondary, and tertiary gustatory areas, and secondary and tertiary olfactory areas of the brain. Nevertheless, the piriform cortex was not activated during after-taste phase in our study, nor in the study by Castriota-Scanderbeg.

Common activations during the after-taste phase that are observed in both studies are the greater involvement of the

**FIGURE 1 | Example of activations involving flavor integration (from Table 2).** Visible activations are left pars opercularis and left and right insula for controls during the taste phase, right and left anterior insula and right orbitofrontal cortex for controls during the after-taste phase, left insula for experts during the taste phase, orbitofrontal cortex and the associative occipital cortex for experts during the after-taste phase.

right side in the control group while bilateral and especially left frontal activation can be observed in the expert group. This suggests that our paradigm is effective in exploring expertise of flavor integration. Castriota-Scanderbeg et al. (2005) found in sommeliers a higher activation in the anterior insular, which is presumed to be involved in the integration of olfaction and gustation, as well as in the LPFC areas. In our study, experts were mainly characterized by recruitment of the hippocampal formation, regions of the temporal lobe and associative visual cortex. This could be explained by the persistence of memory processes. Neither the dorsolateral prefrontal cortex nor the orbito-frontal or anterior insular cortices were recruited during the after-taste phase in experts, as was the case in the study by Castriota-Scanderbeg et al. (2005). In their study, the integration of sensory processes appeared to continue during the after-taste period whereas, in our study, only memory structures persisted during the second phase. This difference between the two studies may be explained by the methodological designs, particularly the use of glucose as a reference stimulus.

Control subjects showed predominant activations in the right hemisphere. Temporal activations were more numerous than in the previous phase but no hippocampal or parahippocampal activations were observed. Frontal, parietal and occipital regions were involved to a lesser degree than previously and may have corresponded to the persistence of ineffective retrieval strategies. In control subjects, the anterior insula was only activated in the after-taste phase, whereas in experts, sensory integrationrelated regions were no longer activated during this phase. This result indicates that experts showed a more immediate and targeted sensory reaction to wine stimulation than control subjects (**Figure 2**).

Olfactory and gustatory pathways appeared to be reciprocally connected. Asymmetrical involvement of the gustatory and olfactory regions in flavor processing is also supported by perceptual experiences and neuroimaging studies of taste/odor integration. The "flavor network" model involves multisensory integration, and the system can be subsequently engaged by unimodal stimulation (Prescott, 2012; Small, 2012).

Finally, it is proposed that there is asymmetric contribution of olfaction and gustation to flavor, such that only retronasally perceived odors (via the mouth) and odors previously experienced with taste (irrespective of mode of delivery) engage the flavor system (Prescott, 2012). To improve the understanding of the expertise on wine flavor integration, further studies should take into account the three phases of wine tasting, and neuroimaging protocol design should integrate the sight of wine, then orthonasal stimulation and finally in-mouth sensations. Our protocol, as well as Castriota-Scanderbeg's study, could be too restrictive, and this could explain the lack of important activations such as the piriform cortex activations.

However, and despite limitations of our study, our results are compatible with clear evidence for the overlapping and integration of gustatory, tactile and olfactory inputs in the insular cortex (Small, 2012). The core flavor percept is then conveyed to upstream regions in the brainstem and thalamus, as well as downstream regions in the amygdala, orbitofrontal cortex and

anterior cingulate cortex to produce the rich flavorful experiences that guide our feeding behavior (Small, 2012).

#### **PARTICIPATION OF VARIOUS MEMORIES**

Although olfaction is the least easily categorizable and recognizable sensory modality (Richardson and Zucco, 1989), sommeliers have the unique ability to verbalize descriptors with all their senses. Sensory perceptions enable experts to provide an analytical description by referring to a large corpus of previously memorized and categorized data (Vedel et al., 1972), while novices cannot find the vocabulary to describe their olfactory and gustatory sensations.

Experts, but not novices, can write descriptions that they themselves or other experts can later match to the appropriate samples (Lawless, 1984; Solomon, 1990). A lot of behavioral studies tend to attribute the greater ability of wine experts to recognize and identify wine-relevant odorants to better semantic knowledge (Parr et al., 2002; D'Alessandro and Pecotich, 2013), and authors suggest that wine expertise may be more cognitive, rather than perceptual, expertise (Ballester et al., 2008). These mechanisms in wine experts might imply a predominance of the left hemisphere involved in analytic and linguistic treatment and probably the temporal lobes and hippocampus involved in episodic and semantic memory related to previous experiences. Other studies suggest the importance of perceptual skill, namely sensory memory for odorant and trigeminal perception, rather than semantic memory ability, in wine-relevant olfactory expertise (Parr et al., 2004).

In our study, numerous activations were observed in areas involved in memory processes, predominantly in the left hemisphere (**Figures 3**, **4**). Hippocampal and parahippocampal activations were observed during retrieval tasks eliciting episodic or autobiographic memory or familiarity (Stark and Squire, 2001; Spaniol et al., 2009). Activity in the anterior temporal lobe corresponds to semantic memory (Rogers et al., 2006; Visser et al., 2010a,b) and left side involvement may particularly characterize the verbal knowledge used by experts when describing and labeling the wine in the recognition task.

Bilateral activation of the occipital cortex (associative area) may also have been related to mental imagery. Activations of the occipital gyri were also noted by Royet et al. (1999) in judgments of edibility of different odorants.

In some professional activities and hobbies, tasks requiring expertise are supported by specific sensory and cognitive skills. Cerebral imaging studies have provided insight into the adaptive cerebral networks underlying these abilities. With expertise, an "economic" mechanism may result in enhanced efficiency, reduced effort and increased spontaneity (Hund-Georgiadis and von Cramon, 1999; Maguire et al., 2000; Lotze et al., 2003; Plailly et al., 2012). Accordingly, during the after-taste phase, fewer activations are observed in experts vs. novices and mainly in episodic and semantic memory network (temporal pole, hippocampus and parahippocampal gyrus) and mental imaging (occipital associative area) suggesting that experts classify and compare this sensorial stimulus with their own episodic experience and semantic knowledge. The comparison of control minus expert did not show specific activation of the olfactory or gustatory structures

**FIGURE 3 | Example of activations involving memory (from Table 3) for the contrast experts minus controls.** Activations in green are obtained during the taste phase; those in orange are obtained during the after-taste phase. Visible activations are both in the amygdala/hippocampus complex (taste phase), parahippocampal gyri (right and left for the taste phase, right for the after-taste phase), occipital associative cortex (right during the taste phase, left during the after-taste phase).

during taste phase and then only insula during the after-taste phase suggests a slower and incomplete analysis of the stimulus. During the after-taste phase diffuse activation probably suggests an ineffective retrieval strategy for the novice. For the novice, the treatment requires deeper thinking and passes very quickly to a high level in the cortical areas of the brain, using a much more diffuse and therefore less specific network. Unable to recognize or recall experts' knowledge, they quickly use their episodic memory (Tulving, 1995). They try to associate taste and perceived memories of places and people, in order to contextualize and identify it, and associate it with emotion. From an information processing perspective, this would suggest that experts could have more attention (working memory capacity) to direct to the task at hand, unencumbered by the associated semantic and affective input (Parr et al., 2004).

The hypothesis on memory mechanisms involved in experience should be explored further by time-series analysis methods such as Dynamic Causal Modeling or Granger Causality Analysis (Seth et al., 2013) to better understand the dynamic and interaction of the different kind of memory during wine flavor analysis.

#### **ROLE OF EXPERTISE**

The perceived quality of a wine is dependent on consumers' level of expertise (Sáenz-Navajas et al., 2013). Wine tasting expertise involves advanced discriminative and descriptive abilities with respect to wine. Cortical and brainstem activations showed two different and complementary mechanisms of wine expertise: a

perceptive mechanism of modulation of the afferent input of information corresponding to the activated gustatory and trigeminal brainstem structures, and specialized cognitive analysis, with focalized cortical activations especially in the left structures involved in memory, language and chemosensory analysis.

A sommelier can distinguish a subtle difference of taste in wine by training their ability to integrate information from gustatory and olfactory senses with past experience.

During the wine tasting phase, control subjects showed fewer but larger activated regions than experts. In control subjects, these activations predominantly occurred in the right hemisphere, and were widespread in the parietal, occipital and frontal cortices. Frontal activations included the frontal operculum which is, like the above mentioned adjoining anterior insula, a putative primary taste cortex (Rolls and Scott, 2003; Pritchard and Norgren, 2004). It is also considered to be a secondary cortex related to odor memory (Savic, 2002). Right occipital activation including the associative visual cortex and parieto-occipital junction was found. Once again, we can hypothesize that this corresponds to mental imaging of past wine tasting experiences and of terms used to describe the taste of wine (fruity for example) or even its color (Qureshy et al., 2000). These associative visual areas were more activated in controls than in experts (in terms of cluster size). This could have been because most of this information corresponded to the experts' semantic knowledge (especially verbal), making them less likely to need to refer to visual images.

During after-taste phase, experts were mainly characterized by recruitment of the hippocampal formation, regions of the temporal lobe and associative visual cortex (**Figure 4**). This could be explained by the persistence of memory processes. Neither the dorsolateral prefrontal cortex nor the orbito-frontal or anterior insular cortices were recruited during the after-taste phase in experts, as was the case in the study by Castriota-Scanderbeg et al. (2005). In their study, the integration of sensory processes appeared to continue during the after-taste period whereas, in our study, only memory structures were persistent during the second phase. This difference between the two studies may be explained by the methodological designs, particularly the use of glucose as a reference stimulus. Control subjects showed predominant activations in the right hemisphere. Temporal activations were more numerous than in the previous phase but no hippocampal or parahippocampal activations were observed. Frontal, parietal and occipital regions were involved to a lesser degree than previously, but again this widespread recruitment of cerebral areas might correspond to the persistence of ineffective retrieval strategies. In control subjects, the anterior insula was only activated in the after-taste phase, whereas in experts, sensory integrationrelated regions were no longer activated during this phase. This result indicates that experts showed a more immediate and targeted sensory reaction to wine stimulation than control subjects. This analysis delay in the control subjects is a logical consequence of their level of expertise. Similarly, Plailly et al. (2012) demonstrated that the right anterior insula is more activated in students than in professionals in perfumery during odor imagery tasks.

#### **EXTERNAL VALUE FOR OTHER ALCOHOLIC BEVERAGES**

The discrepancy of our results with the previous ones from Castriota-Scanderbeg raise questions about the role of expertise during the initial taste phase which should be explored by further studies using a direct comparison between water, glucose solution, salty solution, sweet wine and dry wine to improve the understanding of the expertise of flavor integration and the modification induced by those factors. Otherwise, one might think that our results are largely generalizable to different test situations of alcoholic beverages comparing experts and novices. The involvement of brain structures involved in memory is indeed expected in experts regardless of the type of beverage, as well as the participation of different associative cortical areas in novice subjects. However, the specific compounds and sensory qualities of wine can marginally change significant differences (Villamor and Ross, 2013). The technical conditions of taste (taste phase) and olfaction (after-taste phase) may have some effects (Sáenz-Navajas et al., 2012) and because the same brain areas are not equally concerned whatever the flavor (sweet, bitter, sour) or aroma (Rolls et al., 2003; Small and Prescott, 2005), special attention should focus on different positive hedonic valence such as dry or sweet wine. Slightly different locations of the orbitofrontal cortex may be suspected in the case of strong bitterness such as beer relative to sweet flavors (Small et al., 2007). Trigeminal sensations during the test phase can however vary greatly (depending on alcohol content, tannin, astringency, pH*...*) and therefore recruit more or less the secondary somatosensory cortex S II (Bensafi et al., 2008; Billot et al., 2011; King et al., 2013).

### **CONCLUSION**

This study describes the differences in brain activity of sommeliers compared to "naive" subjects during a blind taste test of wine samples as stimuli relative to water, in order to confirm the influence of expertise on the integration of flavor.

The present study was based on different stimuli compared to the only previous study with sommeliers and used a tasteless reference. Expertise impacts basic taste processing, and when comparing expert minus control, during the taste phase, early sensorial and hedonic structures (trigeminal nucleus, amygdala, insula) and familiarity structures (parahippocampal gyrus) are activated suggesting early analysis of the stimulus. Interestingly, brain stem responses have been observed in experts during this taste phase. During the after-taste phase, the experts showed greater activation in the left orbitofrontal cortex, left insular cortex and dorsolateral prefrontal cortex. Our results are consistent with previous studies, particularly in terms of the lateralization of the expertise-related process (predominantly restricted to the left hemisphere in experts). As suggested, the leftward specialization occurs at the expense of normal rightward activity.

Our study also revealed more involvement in experts of hippocampal, parahippocampal, anterior temporal regions and associative occipital area associated with different types of memory, suggesting that wine experts probably process by similitude to try to recognize all characteristics (country origin, *cepage*, "appellation," *millesime*) of the wine, and further analyses such as Dynamic Causal Modeling or Granger Causality Analysis are needed to deeply understand the chronological functioning of brain networks of wine experts.

More generally, our results indicate that wine experts showed a more immediate and targeted sensory reaction to wine stimulation than control subjects. The influence of expertise on flavor integration may mainly comprise quicker sensorial integration with an economic mechanism reducing effort and increasing efficacy. Experts seem to also activate sensory memory and episodic memory as well as working memory and semantic memory. These results confirm that wine experts work simultaneously on sensory quality assessment and on label recognition of wine. To improve the understanding of the effect of expertise on wine flavor integration, further studies should take into account the three phases of wine tasting, and neuroimaging protocol design should integrate the sight of the wine, then orthonasal stimulation and finally in-mouth sensations.

#### **ACKNOWLEDGMENTS**

We would like to thank Holly Sandu (Medical and scientific translator) for her assistance in translating and proofreading our manuscript. We would like to thank Christophe Menozzi (2001, Maître Sommelier de France UDSF) for his contribution to the establishment of the experimental scheme and the choice of wines. We would like to thank the wine domains that provided the wine: Arbois 2004, Chardonnay "Les Brulées" Domaine St Pierre, Mathenay, FR; Côte du Jura 2006, Pinot noir "Cuvée dédicacée," Benoît Badoz, Poligny, FR. We wish to give special thanks to the "Conseil Regional de Franche-Comte" for its financial support allowing the use of the MR-3T Scanner for this research.

#### **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: 30 July 2014; accepted: 26 September 2014; published online: 16 October 2014.*

*Citation: Pazart L, Comte A, Magnin E, Millot J-L and Moulin T (2014) An fMRI study on the influence of sommeliers' expertise on the integration of flavor. Front. Behav. Neurosci. 8:358. doi: 10.3389/fnbeh.2014.00358*

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

*Copyright © 2014 Pazart, Comte, Magnin, Millot and Moulin. 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.*

## A review on the neural bases of episodic odor memory: from laboratory-based to autobiographical approaches

### **Anne-Lise Saive , Jean-Pierre Royet and Jane Plailly \***

Olfaction: from Coding to Memory Team, Lyon Neuroscience Research Center, CNRS UMR 5292—INSERM U1028—University Lyon1, Lyon, France

#### **Edited by:**

Donald A. Wilson, New York University School of Medicine, USA

**Reviewed by:** Maria Larsson, Stockholm University, Sweden Rachel Herz, Brown University, USA

#### **\*Correspondence:**

Jane Plailly, Olfaction: from Coding to Memory Team, Lyon Neuroscience Research Center, CNRS UMR 5292—INSERM U1028—University Lyon1, 50 Avenue Tony Garnier, 69366 Lyon Cedex 07, France e-mail: plailly@olfac.univ-lyon1.fr

Odors are powerful cues that trigger episodic memories. However, in light of the amount of behavioral data describing the characteristics of episodic odor memory, the paucity of information available on the neural substrates of this function is startling. Furthermore, the diversity of experimental paradigms complicates the identification of a generic episodic odor memory network. We conduct a systematic review of the literature depicting the current state of the neural correlates of episodic odor memory in healthy humans by placing a focus on the experimental approaches. Functional neuroimaging data are introduced by a brief characterization of the memory processes investigated. We present and discuss laboratory-based approaches, such as odor recognition and odor associative memory, and autobiographical approaches, such as the evaluation of odor familiarity and odor-evoked autobiographical memory. We then suggest the development of new laboratory-ecological approaches allowing for the controlled encoding and retrieval of specific multidimensional events that could open up new prospects for the comprehension of episodic odor memory and its neural underpinnings. While large conceptual differences distinguish experimental approaches, the overview of the functional neuroimaging findings suggests relatively stable neural correlates of episodic odor memory.

**Keywords: episodic memory, recognition memory, autobiographical memory, olfaction, behavior, approaches, neural bases, human**

#### **INTRODUCTION**

Human episodic memory is the long-term memory process that enables one to mentally and consciously relive specific, personal events from the past (Tulving, 1972, 1983). It is associated with a feeling of mental time travel, a sense of self, and the autonoetic consciousness that allows one to be aware of the subjective time at which events happened (Tulving, 2001, 2002). Although this definition is accepted, episodic memory is experimentally studied through a large set of paradigms that differ in all dimensions of the memory. The content of the memory and the procedures for encoding and retrieval vary in complexity and ecological validity, while the retention time varies in delay. As a consequence, "episodic memory" refers to an ensemble of memory processes. To provide a general picture of episodic memory, it is thus of interest to orient this investigation by the experimental approach. Two different approaches are usually employed to investigate the explicit retrieval of past events: *laboratory-based approaches* and *autobiographical approaches* (McDermott et al., 2009). In the first case, experimenters test the memorization of artificial episodes created in the laboratory, whereas in the second case, experimenters test the retrieval of real-life memories encoded in the participants' past. McDermott et al. (2009) further emphasized that the two methods differ in time "*not only in* *that the events of interest have occurred on different timescales (weeks or years for studies in the autobiographical memory tradition compared with minutes/hours in the laboratory memory tradition): It can take people on the order of 8–12 s to construct a vivid autobiographical memory (Robinson, 1976), compared to recognition memory decisions, which often occur in a second or two*".

Episodic memory depends on the medial temporal lobe, which is composed of different interconnected subregions, including the hippocampus and adjacent parahippocampal, perirhinal and entorhinal cortices (Milner et al., 1968; Squire, 1992; Cohen and Eichenbaum, 1993). The contribution of each of the medial temporal lobe components to the memory process and their connectivity with the neocortex has been widely investigated (Suzuki and Amaral, 1994; Burwell and Amaral, 1998; Witter et al., 2000; Squire et al., 2004; Davachi, 2006; Diana et al., 2007; Eichenbaum et al., 2007). In summary, the cortical projections encompass two parallel pathways. In one pathway, sensory areas project inputs that are critically involved in object perception onto the perirhinal cortex and hence onto the lateral entorhinal cortex. In the other pathway, the parahippocampal cortex and then the medial entorhinal cortex receive visuospatial information. Both entorhinal cortices then converge onto the hippocampus and allow for the representation of the object in the visuospatial context in which it was experienced.

**Abbreviations:** fMRI, functional magnetic resonance imaging; PET, positron emission tomography.

Phenomenologically, the sense of smell demonstrates a close relationship with episodic memory. Odors are well known to be particularly powerful memory cues. Among all sensorial stimuli, odors appear to trigger the most vivid and emotional memories (e.g., Hinton and Henley, 1993; Chu and Downes, 2002; Herz and Schooler, 2002; Larsson and Willander, 2009). This property is usually explained from an anatomical point of view. The olfactory input has direct connections via the olfactory bulb and the primary olfactory (piriform) cortex onto two key structures involved in emotion and memory: the amygdala and hippocampus (**Figure 1**; Carmichael et al., 1994; Insausti et al., 1997; Haberly, 1998). In contrast with other sensory modalities, projections from the sensory input onto these two structures do not pass via the thalamus. From these areas, information is then conveyed to the secondary olfactory cortices composed of the orbitofrontal cortex (OFC) and the insular cortex.

The strong anatomical connection between olfactory and memory structures makes olfaction a privileged sense for accessing memories. However, in light of the amount of behavioral data describing the characteristics of episodic odor memory, the paucity of information available on the neural substrates of this function is startling. The purpose of this review is threefold: (1) to assess and discuss the current knowledge of the neural correlates of episodic odor memory by presenting functional data from healthy participants; (2) to describe the diversity of paradigms and therefore the diversity of cognitive processes by focusing on laboratory-based approaches, such as odor recognition memory and odor-associative memory, and on autobiographical approaches, such as the evaluation of odor familiarity and odor-evoked autobiographical memory; and (3) to point to new experimental and theoretical directions that episodic odor memory research could profitably pursue. To fulfill this triple objective, we choose to present the literature data according to experimental approaches and not to follow the chronological order of publications.

### **LABORATORY-BASED APPROACHES FOR STUDYING THE NEURAL BASES OF EPISODIC ODOR MEMORY**

In laboratory-based approaches for studying episodic odor memory, participants artificially encounter odors in laboratory settings during a first phase (named the "encoding phase"), and then, the memory trace of this odor is questioned in a second phase (named the "test phase"). We will describe in detail three types of laboratory-based approaches to test episodic odor memory, with the level of complexity increasing from the memory of a single item (i.e., the odor recognition) to the memory of an odor using its verbal label (i.e., the odor-verbal recognition memory) and finally to the memory of an association between two items of different modalities (i.e., the crossmodal odor associative memory).

#### **ODOR RECOGNITION MEMORY**

Recognition memory for odors received very little attention until the 1970s. The first study was led by Engen and Ross (1973). In this typical odor recognition paradigm, the participants were exposed to target odors in laboratory settings and, after a retention interval, were asked to decide whether the odor probe was an old stimulus (target odor) or a new one (distractor odor). This paradigm can be defined as investigating the *explicit recognition of laboratory odors*. The authors demonstrated that the memory of odors has very little long-term loss. Laboratory odors were less well recognized than laboratory pictures after a short interval of time (73% correct recognition), but they were better recognized than these laboratory pictures after 4 months (**Figure 2A**; Engen,

**FIGURE 2 | Odor recognition memory. (A)** Ability to recognize laboratory pictures and odors over a span of 1 year. The hypothetical curve of the ability to recognize episodic odors (odors associated with significant real-life experiences) is shown for comparison (adapted from Engen, 1987). **(B)** Impact of semantic processing on odor recognition memory performances. Memory scores for odors that were previously associated with no labels, chemical labels, labels generated by the participants or veridical labels. \*\* p < 0.01; \*\*\* p < 0.001 (adapted from Jehl et al., 1997).

1987). However, this specificity of odor recognition memory has been challenged more recently and significant forgetting of odors over time was observed (e.g., Murphy et al., 1991; Larsson, 1997; Olsson et al., 2009).

The robust ability to accurately recognize odors has been consistently demonstrated (e.g., Lawless and Cain, 1975; Lawless, 1978; Rabin and Cain, 1984; Goldman and Seamon, 1992). Nevertheless, as highlighted in Herz and Engen (1996), odor recognition performance strongly depends on the experimental conditions. First, the odor set size and odor similarities both affect odor recognition: a greater number of odors and a closer similarity among odors result in lower scores (Engen and Ross, 1973; Lawless and Cain, 1975; Jones et al., 1978; Schab, 1991). Second, the perceived qualities of odors influence recognition memory. For example, evidence suggests that the unpleasantness of odors and their high intensity improve the robustness of memories (Larsson et al., 2009). Third, performances in odor recognition are strongly and positively dependent on the amount of semantic information regarding the odor source, as observed in the influence of odor familiarity (**Figure 2B**) and odor-naming ability (e.g., Rabin and Cain, 1984; Lesschaeve and Issanchou, 1996; Jehl et al., 1997; Larsson and Backman, 1997; Bhalla et al., 2000; Frank et al., 2011). Fourth, recognition memory performances can also be affected by the type of procedure engaged in encoding. While no differences emerge for odors learned intentionally or incidentally (Engen and Ross, 1973; Larsson et al., 2003, 2006), the processing task used to encode odor affects the subsequent recognition of odors. Odors are better recognized after elaborative processing (verbal definition, association with a life episode) than after pure odor perceptual processing (Lyman and McDaniel, 1986, 1990). Thus, the importance of semantic processing in odor recognition must be taken into account and, as Schab (1991) previously noted, "*A more realistic assessment of the odor-recognition data reported in the literature, therefore, acknowledges that recognition performance is the joint result of memory for perceptual odor information and memory for covertly generated verbal associations to the odors*".

Two states of awareness are thought to be involved in recognition memory retrieval: *recollection*, which involves the remembering of an item along with contextual and associative details, and *familiarity*, where an item is seen as familiar but no other contextual information is remembered (Mandler, 1980). The recollective experience is experimentally approached through the *Remember/Know procedure* (Tulving, 1985) in order to determine how much recollection and familiarity contribute to different kinds of recognition. The recollective experience occurring in odor recognition memory is influenced by several factors: odor familiarity and identifiability, and gender (Larsson et al., 2003, 2006; Olsson et al., 2009). For instance, Larsson et al. (2006) showed that recognition is more based on recollection than familiarity for familiar odors, and is more based on familiarity and guessing than on recollection for unfamiliar odors.

The neural basis of odor recognition memory has been approached in four studies using standard recognition memory tests. Two positron emission tomography (PET) studies, which were among the first neuroimaging studies on olfactory cognitive processes, highlighted the brain regions specifically involved in long-term odor recognition memory in comparison with shortterm odor memory processes (Savic et al., 2000; Dade et al., 2002). These two studies noted the importance of the prefrontal and posterior-parietal regions in long-term odor memory. They also revealed the role of the PC, especially its right part, in odor recognition. This right-hemisphere superiority in odor recognition has also been reported in patients with brain lesions. Despite a few discrepancies (Hudry et al., 2003), either patients with right temporal lobe or right orbitofrontal lesions or those with right temporal lobe epilepsy perform more poorly than do patients with left-sided lesions in odor recognition tests (Rausch et al., 1977; Carroll et al., 1993; Jones-Gotman and Zatorre, 1993).

Two of our studies recently further elucidated odor recognition memory by investigating the neural basis of this process as a function of task performance using event-related functional magnetic resonance imaging (fMRI; Royet et al., 2011; Meunier et al., 2014). Recognition memory performances were assessed using parameters from signal detection theory, which has widely dominated recognition memory theory since the 1950s (Swets, 1964; Lockhart and Murdock, 1970). From the experimental conditions (target *vs*. distractor) and the participants' behavioral responses ("Yes" *vs*. "No"), four response categories were defined: Hit or Miss when the target items were accurately recognized or incorrectly rejected, respectively, and Correct Rejection (CR) or False Alarm when the distractor items were correctly rejected or incorrectly recognized, respectively. Using both standard and multivariate analyses, we observed that correct and incorrect recognition and rejection induced distinct neural signatures (Royet et al., 2011). Mainly, activity in the hippocampus and the parahippocampal gyrus was associated with the correct recognition of odors, whereas the perirhinal cortex was associated with errors in recognition and rejection. More strikingly, we observed a decreased involvement of the anterior hippocampus when memory performances increased during correct recognition and rejection (**Figure 3A**). These findings led to the hypothesis that a greater ease when performing the task results in less activation in the hippocampus. Recently, we explored the functional connectivity of the networks underpinning correct and incorrect olfactory memories using graph theory (Meunier et al., 2014). We found that among 36 regions of interest, the hippocampus, caudate nucleus, anterior cingulate and medial temporal gyrus were more frequently connected together during correct odor recognition and thus formed a specific module of this condition (**Figure 3B**). The poor odor recognition performances observed in patients with hippocampal lesions (Levy et al., 2004) agrees with the essential role of the hippocampus in odor recognition memory.

#### **ODOR RECOGNITION MEMORY FROM VERBAL LABELS**

Odor recognition memory has also been investigated through the recognition of odor verbal labels where the odors are presented during the encoding phase and the odor labels are retrieval cues (Buchanan et al., 2003; Cerf-Ducastel and Murphy, 2006; Lehn et al., 2013). This paradigm can be defined as testing the *explicit recognition of the verbal labels of laboratory odors* and addresses the label-odor association. Although no statistical comparison was performed, the behavioral results depicted by Buchanan et al. (2003) suggested that the odor-verbal recognition paradigm leads to lower memory scores than those for the odor-odor recognition paradigm. This empirical observation indicates that odor recognition is more difficult when triggered by a label than by the odor itself.

The neural substrates of odor retrieval through odor name recognition have been investigated a couple of times (Cerf-Ducastel and Murphy, 2006; Lehn et al., 2013). The two studies were consistent with regards to the ensemble of brain regions involved in this odor memory process and revealed consistent activation in the hippocampus, PC, amygdala, OFC

and cerebellum. However, comparing odor-name and objectname recognition memories, Lehn et al. (2013) further showed that the hippocampus was activated during the recognition memory of both types of cues, thus providing clear evidence for modality-independent functions of the hippocampus. In turn, a region encompassing the left anterior insula, PC and amygdala, in addition to the left OFC, the left frontal pole and the right cerebellum, were specific to the olfactory modality (**Figure 4**).

piriform cortex; Puta, putamen; Tha, thalamus (adapted from Meunier et al.,

An advantage of using verbal cues is the facilitation of crossmodal comparisons because identical sensory inputs (retrieval cues) are used for different types of stimuli (Lehn et al., 2013). However, the main drawback of this technique is the typically weak link between an odor and its verbal label (Lawless and Cain, 1975; Engen, 1987). Humans perform poorly when identifying common odors from smell alone (Engen

2014).

and Pfaffmann, 1960; Cain, 1979). This difficulty makes the recognition more complex. When a verbal label is presented during the retrieval phase, two strategies can be implemented. The participants can compare the label they were reading to all the labels explicitly or implicitly generated during the encoding phase, a task that involves semantic-based recognition memory. They can also decide whether the odor evoked by the test label matches the memory trace of the encoded odors, a task that refers to an episodic-based recognition memory. Thus, the use of a verbal label to test odor recognition obscures the nature of the memory processes involved during retrieval.

#### **CROSSMODAL ODOR ASSOCIATIVE MEMORY**

In contrast to odor recognition memory from the odor label, crossmodal odor associative memory is related to the association of an odor with a non-odor item. The capacity of healthy adult volunteers to retrieve associations between two items, including an odor, has been demonstrated through two main paradigms. The *paired-associate paradigm* tests the ability to recall the item previously associated with an odor during explicit encoding. Davis (1975, 1977) showed a disadvantage for odors as associative stimuli in comparison with abstract visual stimuli. However, they also observed that this disadvantage decreased with higher odor familiarity and with higher dissimilarity within odor sets, a result that is consistent with the observations reported above in terms of the impact of familiarity and qualitative similarity on odor recognition memory performances (see Section Odor Recognition Memory). The *odor source paradigm* tests the ability to retrieve limited contextual information associated with the odor perception during encoding. For instance, participants were asked to explicitly remember either a specific room (Takahashi, 2003) or a specific space on a board (Gilbert et al., 2008; Pirogovsky et al., 2009) in which the odors were initially presented or to remember the gender of the experimenter presenting the odors during the encoding phase (Gilbert et al., 2006; Pirogovsky et al., 2006; Hernandez et al., 2008). Overall, these studies demonstrated that odor recognition is superior to the recognition of the source, that explicit *vs.* implicit encoding improves the memory for the source but not for the odor itself, and that aging affects odor source memory than on odor recognition (Takahashi, 2003; Gilbert et al., 2006, 2008; Pirogovsky et al., 2006, 2009; Hernandez et al., 2008).

Functionally, crossmodal odor associative memory has been investigated only twice using the paired-associate paradigm. In the study led by Gottfried et al. (2004), objects were paired with odors, and the participants were instructed to imagine a link between each object and the smell (*a priori*, the objects had no explicit link with odor). The effect of "odor context" on the neural responses was then examined during retrieval when these same objects were presented among distractors. In other words, this paradigm studied the implicit recall of the odor through the explicit recognition of the object that was previously paired with the odor but not the conscious retrieval of the odor. This memory process can be defined as an *implicit crossmodal recall of laboratory odor context*. Gottfried et al. (2004) showed evidence for the reactivation of the right posterior PC during successful object recognition in the absence of olfactory stimulation, just by the specific reactivation of the association between the recognized object and its paired odor. The authors further demonstrated that the involvement of the primary olfactory cortex is independent of the odor valence and that this structure is more sensitive to the retrieval of odor than the retrieval of visual stimuli. More importantly, the authors found that odor retrieval involved the right anterior hippocampus, and hence hypothesized that this structure has an important role in the binding between both items. A recent neuropsychology study supports this hypothesis and shows that amnesic subjects with hippocampal damage have impaired odor-place memory but intact odor recognition (Goodrich-Hunsaker et al., 2009). Yeshurun et al. (2009) also suggested a specific role of the hippocampus for odor associative memory. They based their study on the finding that the first odorto-object association is stronger than subsequent associations of the same odor with other objects (Lawless and Engen, 1977). They paired object photos twice with a different odor, a different sound or a different odor-sound stimulus each time. One week later, the participants were presented with the object photos and had to explicitly recognize, among distractors, the odor associated with the object during encoding through odor labels. This task can be defined as investigating the *explicit crossmodal recognition of laboratory odor context*. Yeshurun et al. (2009) observed hippocampal activation for early olfactory but not auditory associations regardless of whether they were pleasant or unpleasant. These findings confirmed the hypothesis that the first olfactory associations enjoy a privileged brain representation that is underlined by the hippocampus.

The odor associative memory paradigms allow the examination of long-term odor memory involving more complex processes than those implicated in the memory of a single item (i.e., odor recognition memory). In these paradigms, the memory concerns the association between an item and a given context. However, the richness of the context is usually limited and materialized by a single other dimension. Therefore, the gap between odor associative memory and odor autobiographical memory is still wide. As highlighted by Schab (1991) "*the conditions under which an odor often is reported to evoke the recollection of past episode differ significantly from those of a paired-associate task. In the former, a single ambient odor triggers the remembrance of a personal episode of which the odor itself was an integral part, whereas in the latter a series of different odors is presented, typically in small bottles, and the learning task is deliberate and requires the acquisition of unrelated and personally irrelevant information*".

### **AUTOBIOGRAPHICAL APPROACHES FOR STUDYING THE NEURAL BASIS OF EPISODIC ODOR MEMORY**

In odor-evoked autobiographical approaches, the content of the memory refers to the participants' past, and its retrieval is triggered with odors. First, we will present the experiments that questioned the memory of previously encountered odors and investigated the feeling of familiarity and unfamiliarity. Then, we will present the studies that addressed the recall of real-life events and investigated odor-evoked autobiographical memories.

#### **FEELING OF FAMILIARITY OF ODORS**

Odor autobiographical memory can be investigated through the feeling of familiarity generated by odors that are presented in laboratory settings. This paradigm refers to the *explicit recognition of self-relevant odor*. As we previously described, "*The feeling of familiarity is a long-term recognition memory process referring to a subjective state of awareness based on judgments of the item's prior occurrence. It involves the recognition of the item's perceptual features and eventually of conceptual or semantic features, without the confirmatory conscious recollection of contextual information and/or without identification*" (Plailly et al., 2007). A consensus emerges from the evaluation of odor perceptual characteristics. There is consistent evidence for positive correlations between the ratings of odor familiarity and those of intensity and pleasantness (e.g., Jellinek and Köster, 1983; Ayabe-Kanamura et al., 1998; Distel et al., 1999; Royet et al., 1999). Familiar odors have also been described as more simple, in terms of ease of interpreting an odor meaningfully (Sulmont et al., 2002). Recently, Delplanque et al. (2008) argued that the relation between pleasantness and familiarity is nonlinear: pleasantness ratings were positively correlated with familiarity ratings for pleasant odors, but not for unpleasant odors, a result that has been subsequently replicated (Plailly et al., 2011; Ferdenzi et al., 2013).

Our research team was the first to address the neural basis of the familiarity process. In the first studies, we compared periods of brain activity recorded when participants rated the familiarity of a large set of familiar or unfamiliar odors to periods when they detected the presence of odors (Royet et al., 1999, 2001; Plailly et al., 2005). Participants were instructed to make familiarity judgments based on their life experiences (i.e., "*Does this odor seem familiar to you?"*). This paradigm avoided the need for an initial experimental encoding phase. Greater activation of the right OFC and the right PC was observed when the participants evaluated odor familiarity compared with when they detected odors (Royet et al., 1999, 2011; Plailly et al., 2005). The lateralization of this memory process (Royet and Plailly, 2004) was consistent with the higher familiarity of odors presented to the right nostril than those presented to the left nostril (Broman et al., 2001). This could also explain the right hemisphere lateralization of the odor process observed in the first studies when odorants were passively perceived because the odorants were familiar and could have automatically triggered recognition (e.g., Zatorre et al., 1992; Yousem et al., 1997; Sobel et al., 1998; Savic et al., 2000; Poellinger et al., 2001). Our studies on odor familiarity evaluation further emphasized the role of the left inferior frontal gyrus, a key region for semantic processing, which is most likely activated in an attempt to gather semantic information to identify the smell (Royet et al., 1999, 2011; Plailly et al., 2005). Additional activations were observed in the brain regions involved in emotion (amygdala), visual mental imagery (fusiform and occipital gyri) and memory (hippocampus and parahippocampal gyrus) processes, reflecting the large set of cognitive processes engaged during the evaluation of odor familiarity (Plailly et al., 2005).

Savic and Berglund (2004) and Plailly et al. (2007) revealed that familiar and unfamiliar odors are processed by different neural circuits. Savic and Berglund (2004) reported that the passive perception of odorants selected to be familiar *vs.* unfamiliar elicited specific activation of the right parahippocampal gyrus, right middle and inferior temporal gyri, and the left parietal cortex covering the precuneus. In addition, the familiarity ratings obtained after functional acquisitions were positively correlated with activation in the left inferior frontal gyrus and the right parahippocampal gyrus (**Figure 5A**), suggesting that the smelling of familiar, but not that of unfamiliar, odors engages neural circuits mediating semantic association and episodic retrieval functions. Our research team completed the preceding results by unveiling the existence of a bimodal neural system engaged in the feeling of familiarity *vs.* unfamiliarity (Plailly et al., 2007). The neural correlates of self-rated familiarity evoked by items of two modalities, odors and musical excerpts, overlapped within an extensive bimodal neural system that included the prefrontal, inferior frontal, parieto-occipital and medial temporal lobe brain regions in the left hemisphere (**Figure 5B**). We further concluded that because this system also overlaps with the familiarity processing of other types of stimuli (i.e., faces, voices, pictures and verbal items), a multimodal neural network might underlie the feeling of familiarity. Interestingly, we revealed the existence of neural processes specific to the feeling of unfamiliarity, which might be related to the detection of novelty, with a main bimodal activation in the right insula.

#### **ODOR-EVOKED AUTOBIOGRAPHICAL MEMORY**

Odor-evoked autobiographical memory can be investigated through the recall of the life episode associated with an odor. This paradigm refers to the *explicit recall of autobiographical memories evoked by self-relevant odor*. Odors are exceptional cues for evoking personal autobiographical memories. Behavioral evidence has demonstrated that odors are more effective triggers of emotional memories than the same cue presented in other sensory formats or even in the form of odor labels (Hinton and Henley, 1993; Chu and Downes, 2002; Herz and Schooler, 2002; Herz, 2004, 2012; Herz et al., 2004; Larsson and Willander, 2009; Arshamian et al., 2013). Another specificity of odor-evoked autobiographical memories is that they produce a unique age distribution and favor childhood memories stemming from the first decade of life rather than from young adulthood, which is the typical reminiscence bump for memories evoked by verbal and visual information (Chu and Downes, 2000; Willander and Larsson, 2006; Larsson and Willander, 2009; Miles and Berntsen, 2011). Furthermore, empirical evidence indicates that odor-evoked memories are associated with stronger feelings of being brought back in time (Herz and Schooler, 2002; Herz, 2004; Willander and Larsson, 2006, 2007; Arshamian et al., 2013) and are thought of and talked about less than memories elicited by visual or verbal variants of the same items (Rubin et al., 1984). Finally, odors may also be more likely than visual or verbal cues to elicit perceptual-based memories; visual or verbal cues in turn provide more conceptual-based memories (Herz and Cupchik, 1992; Goddard et al., 2005; Willander and Larsson, 2007).

Although the high potential of odors to generate the successful recall of autobiographical memories has been behaviorally demonstrated, the neural basis remains little explored. Only two studies have investigated the neural underpinnings of odorevoked autobiographical memories. Herz et al. (2004) explored whether the brain correlates of personal memories elicited by the smell of a perfume were different from those elicited by the sight of this perfume. Arshamian et al. (2013) compared memories evoked by either personally meaningful odors or pleasant control odors. In both studies, the authors observed activation in the parahippocampal gyrus, the amygdala, and the middle occipital gyrus. These regions play a crucial role in memory, emotion and visual mental imagery, and their engagement could explain the fact that odors are especially potent reminders of autobiographical experiences. Interestingly, Arshamian et al. (2013) raised two important issues. The first was inspired by the debate opposing *the multiple memory trace theory consolidation model* that postulates that the hippocampus and neocortex are in constant interaction (Nadel and Moscovitch, 1997, 1998) and the *standard model of memory consolidation* where the passage of time leads to a disengagement of the hippocampus and an additional recruitment of the prefrontal cortex (Marr, 1971; Squire et al., 1984). Arshamian et al. (2013) observed that hippocampal activation did not vary as a function of memory remoteness, which supports the notion of a permanent role of the hippocampus in the retrieval of odor-evoked autobiographical memories (**Figure 6**). Second, because of the early reminiscence bump in olfaction, the authors tested whether odors were differentially coded depending on the decade in which the stimulus was encoded. They observed a greater involvement of regions devoted to perceptual processes (e.g., the orbitofrontal cortex) during the recall of first-decade odor-evoked memories and a greater recruitment of regions involved in semantic processing (the left inferior frontal gyrus) during the recall of second-decade odor-evoked memories. This result suggests that the autobiographical recall is based more on perceptual processing and less on semantic processing when memories refer to early life experiences.

### **LABORATORY-ECOLOGICAL APPROACHES FOR STUDYING THE NEURAL BASIS OF EPISODIC ODOR MEMORY**

The two main approaches for studying episodic memory developed above, the laboratory-based and autobiographical approaches, each have pros and cons. In the laboratory-based approach, artificial and simple episodes are encoded and recalled in controlled conditions in the laboratory. This method enables the manipulation of the encoding conditions and the retention time and allows the oversight of recall veracity. However, the tobe-remembered materials that are developed by experimenters are poor in comparison with a real-life episode. In the autobiographical approach, the retrieval of real-life memories that are

encoded in the participants' past is triggered by an experimental cue. This approach allows for the recall of real-life events in quite ecological conditions, but the veracity of the recalled events cannot be controlled. McDermott et al. (2009) have underscored the interest in proposing a new approach to study and understand human episodic memory, one that is halfway between the two traditional approaches and retains the respective advantages of each. Fulfilling those expectations, several *laboratory-ecological approaches* have been recently devised to study episodic memory (Pause et al., 2010, 2013; Holland and Smulders, 2011; Milton et al., 2011; Easton et al., 2012; Saive et al., 2013). On the one hand, these approaches are close to Tulving's definition of episodic memory (Tulving, 1972, 1983) by allowing the conscious and controlled recollection of specific and complex events from the past. On the other hand, they are derived from contentbased approaches developed in animals proposing to define the content of episodic memory as *What happened*, *Where* and *When* (Clayton and Dickinson, 1998; Griffiths and Clayton, 2001; Babb and Crystal, 2006; Crystal, 2009). In addition to the threedimensional content of the episodic memories, Clayton et al. (2003) argued that these memories must also be integrated, flexible and trial unique. Subsequently, Easton and Eacott (2008; Eacott and Easton, 2010) enriched this operational definition of episodic memory by considering an alternative to the temporal dimension. They proposed replacing this dimension by the specific occasion or context in which the event occurred (*Which context*); this context encompasses the time when important but also the emotion, semantic knowledge, visual scene, or auditory and olfactory environments.

In the study of episodic odor memory, the laboratoryecological approaches are still rare, although the necessity to elaborate new paradigms has been raised for more than 20 years. Schab (1991) wrote that "*discrepancy between experience and past experimental research is due to less than optimal choice of procedures in the laboratory studies. One means of studying odor-cued recall in the laboratory is to 'create' a personal significant event*". This insight led Schab and Cain (1992) to suggest an example of a laboratory-based, personally significant event, which consisted of a scenario during which the participants witness a specific emotional event in the context of ambient odor and sound. This emotional event could be tested later to investigate the power of odor *vs.* sound to evoke episodic memory retrieval. The authors hypothesized that "*Such an experiment might support the popular expectation regarding odor-evoked retrieval because it may stimulate the environmentally realistic event more faithfully*". However, their reflections did not give rise to any experiment. Sometime later, Aggleton and Waskett (1999) imagined an ingenious experiment where visitors to a museum were re-exposed to the ambient smell of a previous exhibition and were questioned about their memories of this exhibition. The odor specifically acted as an effective retrieval cue and improved their memory performances. This approach allowed for the investigation of the retrieval of a real-world episode but not in its entirety. The authors only tested the content of the exhibition and not the context or the emotion associated with the event. Along the same lines, Herz and Cupchik (1995) and Herz (1998) attempted to address the power of emotion triggered by odor to induce the recall of a memorieslike association created in the laboratory. They used a pairedassociate paradigm in which emotional paintings or pictures were paired with emotional odors or a verbal, visual, musical or tactile variant of the same cue. The mean percentages of paintings or pictures correctly recalled were similar across modalities, but the odor-evoked memories were significantly more emotionally loaded than the memories cued by the other modalities. The directions toward which this experiment went were exciting, but they were not further developed. Additionally, the paradigm was never enriched to match the content-based episodic-like memory definitions (Tulving, 1972; Easton and Eacott, 2008).

To investigate odor-evoked episodic memory, we recently developed an original laboratory-ecological approach deeply inspired by episodic-like memory tasks performed by animals (Saive et al., 2013). It was as ecologically valid as possible, yet the encoding and retrieval conditions were fully controlled. The tobe-remembered episodes were trial-unique, rich, close to real-life episodes, and in agreement with the definitions of episodic memory proposed by Tulving (1972) and Easton and Eacott (2008). During the encoding phase, the participants freely explored three unique episodes, one episode per day. Each unique episode was composed of three unfamiliar odors (What) positioned at three specific locations (Where) within a visual context (i.e., a picture of a landscape; Which context). We intentionally selected unfamiliar and largely unidentifiable odors and arbitrarily linked the odors, spatial locations and visual contexts in each episode to limit associative semantic processes. On the fourth day, the odors were used to trigger the retrieval of the complex episodes in a recall test. The participants were asked to recognize odors and to correctly remember the visuospatial context in which they were encountered, ensuring the evaluation of the memory content accuracy (**Figure 7A**). The participants were highly proficient in recognizing the target odors among distractors and retrieving the spatio-contextual environment of the episode with a rather high confidence level (Saive et al., 2013). This observation suggests that when an association between odors, spatial locations and contexts is encoded, the association forms an integrated representation retrievable by the participants. More recently, using a similar procedure, we observed that memory performances were influenced by the emotional content of the odor, regardless of their valence; both pleasant and unpleasant odors generated greater recognition and episodic retrieval than did neutral odors (**Figure 7B**; Saive

et al., 2014). Our new approach is adapted to fMRI constraints and should permit further investigations of the neural basis of episodic odor memory.

### **CONCLUSION AND FUTURE DIRECTIONS**

Episodic odor memory is experimentally studied through a large set of paradigms and, as a consequence, the concept of "episodic odor memory" refers to an ensemble of memory processes which varied in complexity from the recognition of a single odor to the autobiographical memory evoked by odor. While large conceptual differences distinguish the laboratory-based and the autobiographical approaches, each approach has specificities that are complementary to the understanding of the neural underpinnings of the episodic odor memory. In laboratory-based approaches, the content of the memory is fully controlled and brain signals can be analyzed regarding the accuracy of the participants' responses, allowing for the distinction between the neural substrates related to memory success or to memory failures. For example, a module of tightly-connected brain regions (hippocampus, caudate nucleus, anterior cingulate and medial temporal gyrus) is specifically involved when odors are accurately recognized (Meunier et al., 2014), while the perirhinal cortex is specifically associated with memory errors (Royet et al., 2011). In autobiographical approaches, the access to real-life memories allows for the involvement of a wider ensemble of cognitive processes. The personal significance of the cue item generates the engagement of semantic processes, as highlighted by the role of the inferior frontal gyrus (Royet et al., 1999, 2011; Savic and Berglund, 2004; Plailly et al., 2005, 2007), and of emotional and visual imagery processes reflecting the vividness of the recalled memories (Herz et al., 2004; Plailly et al., 2005). Studying autobiographical memories also enables addressing consolidation process over time and suggests a continuous engagement of the hippocampus whatever the age of the memory (Arshamian et al., 2013).

While the two experimental approaches differ in their conception of episodic memory, the overview of the functional neuroimaging findings suggests a core of relatively stable neural correlates of episodic odor memory regardless of the approach. The major role of the PC in human episodic odor memory is consensual. This finding agrees with the associational properties of the primary olfactory cortex observed in animals (Litaudon et al., 1997; Haberly, 2001; Wilson and Stevenson, 2003) and its role in working odor memory in humans (Zelano et al., 2009). The involvement of the PC in episodic odor memory is modalityspecific (Gottfried et al., 2004; Lehn et al., 2013), it is independent of odor valence (Gottfried et al., 2004; Yeshurun et al., 2009), and it tends to be lateralized to the right (vs. left) hemisphere (Savic et al., 2000; Dade et al., 2002; Gottfried et al., 2004; Plailly et al., 2005; Cerf-Ducastel and Murphy, 2006). The hippocampus is also consistently observed in both approaches, which is consistent with a large amount of literature that stresses the importance of this brain region in episodic memory (e.g., Suzuki and Amaral, 1994; Burwell and Amaral, 1998; Witter et al., 2000; Squire et al., 2004; Davachi, 2006; Diana et al., 2007; Eichenbaum et al., 2007). The literature involving the olfactory modality further shows that hippocampal activation reflects the memory performance (Royet et al., 2011; Lehn et al., 2013), and that while the hippocampus is engaged in the episodic memory of different sensory modalities (Plailly et al., 2007; Lehn et al., 2013), it has a privileged role for the first olfactory associations (Yeshurun et al., 2009). Additionally to the PC and hippocampus, laboratory-based and autobiographical approaches are concordant in the role of prefrontal, infero-temporal, postero-parietal and medial temporal lobe brain regions in odor episodic memory. Thus, the present review agrees with previous report demonstrating that brain networks involved in classical autobiographical studies partially overlap with those found in more controlled laboratory episodic memory tasks (Cabeza et al., 2004; Burianova and Grady, 2007; Cabeza and St Jacques, 2007).

We believe that the development of laboratory-ecological approaches that control the encoding and retrieval of specific and multidimensional laboratory episodes can yield new discoveries for the comprehension of episodic memory. By controlling each aspect of the to-be-remembered event and of its retrieval, specific questions can be addressed. For example, the close relationship between olfaction, emotion and memory, commonly illustrated as the Proust phenomenon (Chu and Downes, 2000), can be further explored by manipulating the emotional strength of the episode during encoding and by manipulating the sensory modality of the cue that triggers episodic retrieval during the test phase. Furthermore, Mitchell and Johnson (2009) stressed the importance to rate amount of details of various types or vividness, emotional valence, arousal, because they provide specific information that explain the complex inter-play of cognitive processes that are characteristic when retrieving rich memories and that can be related to brain activity. Such features are relatively easy to measure and can be crucial in the understanding of the different processes underlying episodic memory. We further suggest the investigation of the brain as whole through the use of specific analysis techniques. Most cerebral imaging functional studies have used univariate statistical analyses to localize individual aspects of brain function, and have restricted investigation to specialized cognitive sub-systems. Various techniques for measuring functional connectivity are to date available and their use can represent a considerable improvement in the understanding of episodic memory. This sum of efforts will be the basis of real advances in this field and will bring substantial progress in the understanding of the behavioral specificities of episodic odor memory.

#### **ACKNOWLEDGMENTS**

This work was supported by the Centre National de la Recherche Scientifique (CNRS), the LABEX Cortex (NR-11-LABX-0042) of Université de Lyon within the program "Investissements d'Avenir" (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR) and research grants from the Région Rhône-Alpes (CIBLE 10 015 772 01). Anne-Lise Saive was funded by the Roudnitska Foundation.

#### **REFERENCES**


and olfactory knowledge. *Chem. Senses* 38, 175–186. doi: 10.1093/chemse/ bjs083


of cortical efferents. *Hippocampus* 7, 146–183. doi: 10.1002/(sici)1098- 1063(1997)7:2<146::aid-hipo4>3.0.co;2-l


distinct neural signatures. *Front. Hum. Neurosci.* 5:65. doi: 10.3389/fnhum.2011. 00065


by way of parallel parahippocampal-subicular pathways. *Hippocampus* 10, 398–410. doi: 10.1002/1098-1063(2000)10:4<398::aid-hipo6>3.3. co;2-b


**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 2014; accepted: 20 June 2014; published online: 07 July 2014*. *Citation: Saive A-L, Royet J-P and Plailly J (2014) A review on the neural bases of episodic odor memory: from laboratory-based to autobiographical approaches. Front. Behav. Neurosci. 8:240. doi: 10.3389/fnbeh.2014.00240*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience*. *Copyright © 2014 Saive, Royet and Plailly. 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.*

## A unique memory process modulated by emotion underpins successful odor recognition and episodic retrieval in humans

### *Anne-Lise Saive\*, Jean-Pierre Royet , Nadine Ravel , Marc Thévenet , Samuel Garcia and Jane Plailly*

*Lyon Neuroscience Research Center, CNRS UMR 5292 - INSERM U1028 - University Lyon1, Lyon, France*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

*Reviewed by: Max L. Fletcher, University of Tennessee Health Science Center, USA Yaara Yeshurun, Princeton*

#### *University, USA \*Correspondence:*

*Anne-Lise Saive, Lyon Neuroscience Research Center, CNRS UMR 5292 - INSERM U1028 - University Lyon1, 50 Avenue Tony Garnier, 69366 Lyon, France e-mail: anne-lise.saive@ olfac.univ-lyon1.fr*

We behaviorally explore the link between olfaction, emotion and memory by testing the hypothesis that the emotion carried by odors facilitates the memory of specific unique events. To investigate this idea, we used a novel behavioral approach inspired by a paradigm developed by our team to study episodic memory in a controlled and as ecological as possible way in humans. The participants freely explored three unique and rich laboratory episodes; each episode consisted of three unfamiliar odors (What) positioned at three specific locations (Where) within a visual context (Which context). During the retrieval test, which occurred 24–72 h after the encoding, odors were used to trigger the retrieval of the complex episodes. The participants were proficient in recognizing the target odors among distractors and retrieving the visuospatial context in which they were encountered. The episodic nature of the task generated high and stable memory performances, which were accompanied by faster responses and slower and deeper breathing. Successful odor recognition and episodic memory were not related to differences in odor investigation at encoding. However, memory performances were influenced by the emotional content of the odors, regardless of odor valence, with both pleasant and unpleasant odors generating higher recognition and episodic retrieval than neutral odors. Finally, the present study also suggested that when the binding between the odors and the spatio-contextual features of the episode was successful, the odor recognition and the episodic retrieval collapsed into a unique memory process that began as soon as the participants smelled the odors.

**Keywords: episodic memory, recognition memory, encoding, olfaction, visuospatial context, emotion, breathing, human**

### **INTRODUCTION**

Human episodic memory is the memory that permits the conscious re-experience of specific personal events from the past (Tulving, 1972, 1983) and is associated with a feeling of mental time travel (Tulving, 2001, 2002). Because the investigation of this ability in animals is controversial, content-based approaches have been developed that focus on the different types of information stored in memory: *What* happened, *Where* and *When* (Clayton and Dickinson, 1998; Griffiths and Clayton, 2001; Babb and Crystal, 2006; Crystal, 2009). Subsequently, based on human phenomenological experiences of event recall, Easton and Eacott (2008; Eacott and Easton, 2010) enriched this refined definition of episodic memory. They widened its third dimension, replacing the temporal dimension with the specific occasion or context in which the event occurred, thereby leading to a "*What*, *Where*, *Which occasion*, or *Which context*" definition. The authors considered episodic memory as a "snapshot" of an episode in which time can form a part of the context but is not the only contextual marker. Emotion, semantic knowledge, the visual scene, or auditory and olfactory environments can also define the context of the episode. For example, when you remember the last time you went to a restaurant, you can recall where and when it was, as well as the occasion for which you were there, with whom, what you ate, and if you had a good evening. Importantly, these approaches did not consider the memory in terms of autonoetic consciousness, and therefore, were referred to as episodic-like memory (Clayton and Dickinson, 1998; Clayton et al., 2003).

In humans, two approaches are usually used to study past event retrieval. In the *ecological* approach, experimenters test autobiographical memory by interrogating participants about real-life memories encoded in their past (Fink et al., 1996; Levine et al., 2004; Piolino et al., 2004; Nadel et al., 2007; Janata, 2009). This approach is quite ecological because it is close to real-life recall, but the veracity of the recalled events cannot be controlled for. In the *laboratory-based* approach, experimenters test the memorization of artificial episodes created in the laboratory using recognition tasks (Konishi et al., 2000; Daselaar et al., 2003; Donaldson et al., 2010; Royet et al., 2011; Herholz et al., 2012), thereby permitting control of the encoding conditions, the retention time and the veracity of the retrieval. However, the information to

**Abbreviations:** CR, Correct rejection; FA, False alarm; WWW, Retrieval of the three dimensions (*What*, *Where*, *Which context*) of the episode; WWhich, Retrieval of the *What* and *Which context* dimensions of the episode; WWhere, Retrieval of the *What* and *Where* dimensions of the episode; What, Retrieval of the *What* dimension of the episode

be remembered is often one-dimensional (e.g., *What*) and is therefore poor in comparison with a real-life episode. To limit the drawbacks of such methods, new *laboratory-ecological* approaches halfway between these two traditional methods have recently been devised to explore human episodic memory (Pause et al., 2010, 2013; Holland and Smulders, 2011; Milton et al., 2011; Saive et al., 2013). We proposed such an intermediate approach that was deeply inspired by tasks developed to study episodic-like memory in animals to determine the experimental conditions that best evaluate episodic memory while remaining ecologically valid (Saive et al., 2013). This approach allowed the controlled study of trial-unique free encoding, retention delay and the retrieval of rich and complex episodes composed of unnamable odors (What) located spatially (Where) within a visual context (Which context).

Phenomenologically, olfaction, memory and emotion are closely linked. Odors are particularly evocative reminders of past events. Among all sensorial stimuli, odors trigger more vivid and emotional memories (Hinton and Henley, 1993; Herz and Cupchik, 1995; Chu and Downes, 2002; Larsson et al., 2009). This phenomenon can be explained because the functions of olfaction, memory and emotion involve anatomically tight brain areas. The primary olfactory cortex includes the piriform-periamygdaloid cortex, which gives way gradually to the lateral entorhinal cortex. From these areas, the olfactory signal is respectively transmitted to the amygdala and to the CA1 of the hippocampus (Price, 1973; De Olmos et al., 1978; Shipley and Reyes, 1991) before being sent to the secondary olfactory cortices, the orbitofrontal and insular cortices. Therefore, from its birth in the olfactory epithelium, the olfactory signal is relayed through two or three neurons to the brain structures critical for emotion and memory (for review, Eichenbaum, 2000; Sergerie et al., 2008). Despite some consensus on odor pleasantness especially for very pleasant and very unpleasant odors (Moncrieff, 1966), the emotion generated by odors can greatly differ between individuals (Ferdenzi et al., 2013). The differences in emotional responses to odors can result from variations in genetic backgrounds (Keller et al., 2007) but likely mainly result from differences in personal experience (Engen, 1991; Robin et al., 1998; Herz, 2001; Herz et al., 2004). The association between an odor and the emotional content in which it occurs determines its future hedonic tone and explains why the same odor can be perceived as either pleasant or unpleasant.

The objective of the current study was first to investigate the cognitive processes of episodic memory by combining in an original way the laboratory and autobiographical approaches. Second, it was to test the still-unexplored hypothesis that the emotion carried by odors facilitates the memory of specific unique events. To investigate this idea, we adapted our episodic memory task and addressed the episodic retrieval of episodes comprising three different odors positioned at specific locations within a visual context to create rich multidimensional episodes (Saive et al., 2013). To identify the differential influence of emotion on episodic memory, we tested the effects of emotion carried by odors on the behavioral and physiological responses of the participants during encoding and retrieval.

### **MATERIALS AND METHODS**

#### **PARTICIPANTS**

Twenty-five healthy participants [13 women; age: 21.4 ± 2.1 years (mean ± standard deviation)] consented to participate in the experiment. All participants were right-handed and reported normal senses of smell and no visual impairments. They provided written informed consent as required by the local Institutional Review Board in accordance with French regulations for biomedical experiments with healthy volunteers [Ethical Committee of CPP Sud-Est IV (CPP 11/007), ID RCB: 2010-A-01529-30, January 25, 2011] and received financial compensation. The study was conducted in accordance with the Declaration of Helsinki.

### **STIMULI AND MATERIALS**

#### *Odorants*

Eighteen odorants consisting of essential oils and single or mixtures of monomolecular chemical compounds were selected *a priori* based on their distinctiveness and relatively low identifiability and familiarity. The odorants were subdivided into two sets (Sets 1 and 2) of nine odors each. Set 1 was composed of butanol, calone, carrot, cis-3-hexenyl salicylate, dihydromyrcenol, methyl octine carbonate, musk, rosemarel and stemone. Set 2 was composed of allyl amyl glycolate, basil, birch oil, citronellol, ethyl acetyl acetate, linalyl acetate, rose oxide, styrallyl acetate and tobacco.

The odorants were presented using a 20-channel computercontrolled olfactometer adapted from an olfactometer previously described by Sezille et al. (2013). Briefly, this odor diffusion system was developed to synchronize odorous stimuli with breathing. Undiluted odorants were contained in a 10-ml U-shaped Pyrex® tube (VS Technologies, France) filled with odorized microporous substances. Odorized airflows and air carrier were sent to and mixed in a homemade mixing head made of polytetrafluoroethylene and connected to the nostrils. The participant's respiratory signal was acquired using a nasal cannula and was used to trigger the odor stimulation through an airflow sensor. The airflow rate was set at 3 l/min, and the odorants were delivered over 4 s.

#### *Spatio-contextual environment*

The spatio-contextual environment was presented within the experimental setup previously described by Saive et al. (2013), but modified for the present study. Three landscape pictures presented full-screen (1280 × 1024 pixels, 72 dpi) constituted the visual contexts (a coastal cliff, a lavender field and a mountain landscape; **Figure 1A**). For each of the three contexts, circles symbolized nine spatial locations: 6 were colored in gray, and 3 were colored in orange. When the circle was orange, it was associated with an odor; otherwise, it was gray. All spatial locations of the orange circles and all odors differed between the contexts.

#### *Multidimensional episodes*

Three multidimensional episodes were created, which were each composed of three odors (What) associated with specific locations (Where) within a given visual context (Which context). Three multidimensional episodes were created, which were each composed of three odors (What) associated with specific

locations (Where) within a given visual context (Which context). To limit associative semantic processes, the odors, spatial locations and visual context were arbitrary linked.

An in-house LabView software (version 8.6 or higher) controlled the presentation of odors, pictures and circles and recorded the participants' responses and breathing throughout the experiment. The participants were requested to breathe normally and avoid sniffing behaviors (**Figure 2**). To interact with the software, the participants used a trackball (Kensington, Redwood Shores, CA, USA). When the participants clicked on a circle, the odor stimulus was delivered at the beginning of the subsequent expiration, enabling the odor to be perceived at the beginning of the next inspiration (on average 2 s later). The volume, amplitude and duration of each inspiratory cycle were recorded, and the respiratory frequency was calculated.

#### **EXPERIMENTAL PROCEDURE**

The experimental procedure consisted of four sessions performed over the course of 4 successive days. The first three sessions were used for encoding, and the retrieval occurred in the fourth session (**Figure 1B)**. A full night of sleep followed each of the encoding sessions to promote consolidation and to reduce interference (Maquet, 2001; Stickgold, 2005). Participants completed the four sessions at the same time of the day to limit the differential

influence of internal states (hunger, satiety) on olfactory and cognitive processes between sessions (Jiang et al., 2008; Plailly et al., 2011).

of odor perception, fading with time).

There were two groups of participants: G1 and G2. For G1, the Set1 odorants were defined as the targets, and the Set2 odorants were defined as the distractors. For G2, the Set2 odorants were defined as the targets, and the Set1 odorants were defined as the distractors.

#### *Encoding*

During encoding, the participants freely discovered one episode per day for 7 min (**Figure 1B**). They were asked to explore all dimensions of the episode as much as possible by paying attention to the background picture, the circles superimposed on this background, and the odors that are delivered when clicking on the orange circles. No memorization instruction was given, thereby ensuring free encoding, similar to what arises in real-life situations. The participants were only informed that they would be questioned about their perception of the episodes on the fourth day. The order of the three episodes was randomized between the participants.

### *Retrieval*

Retrieval was performed on the fourth day. The session consisted of three blocks of 24 trials, and each block corresponded to the presentation of 15 target odors and 9 distractor odors. Each target odor was presented five times, and each distractor odor was presented three times. For a given block, the target and distractor odors were presented in a pseudorandom order such that two presentations of the same odor were separated by at least two trials. The odor presentation order was counterbalanced between the participants.

Each trial began with an odor recognition task (**Figure 1B**). The participants were presented the odors and had to determine whether they recognized the smell ("*Do you recognize this smell?*") as having been previously presented during the encoding. Two situations could happen. 1) If the participants responded "*Yes,*" they were then asked to retrieve the entire episode associated with the odorant and to press on the trackball if they succeeded in less than 20 s after the odor was sent ("*Press when you remember the context*"). After this delay, they were given up to 10 s to choose both the accurate visual context and the exact location of the odor by selecting one of the three pictures, followed by one of the nine circles. A response was considered correct when the participants selected both the accurate context and the specific location previously associated with the odor during the encoding. 2) If the participants responded "*No*," they had to press on the trackball ("*Press the button*") and rest until the next trial.

Following this retrieval task, the strength of the association between the spatial location and the visual context of an event was tested. The participants had to recall the three locations (orange circles) associated with the odors in every visual context during the encoding.

#### *Rating of odor intensity, pleasantness, and familiarity*

At the end of the experiment, the participants were asked to rate the odorants in terms of intensity, pleasantness and familiarity using non-graduated scales. The pleasantness scale was divided into two equal parts by a "*neutral*" value separating the ratings of unpleasantness and pleasantness. The intensity, pleasantness and familiarity ratings were *a posteriori* transformed into scores from 0 to 10.

#### **DATA ANALYSIS**

#### *Encoding*

For each participant, the number of clicks was computed per odor. For each odor, the time periods between two consecutive clicks (delay) were measured, and the mean delay was then determined. The time window between the two clicks served as the time frame for the analyses of breathing parameters (e.g., the volume, amplitude and duration of the inspiratory cycles and the respiratory frequency). The influence of the odor characteristics (intensity, pleasantness and familiarity) on the behavioral and physiological (breathing) data was tested. The relationship between the encoding and the retrieval was investigated by analyzing the behavioral and physiological data during the encoding as a function of the subsequent memory performances.

### *Retrieval*

Recognition memory performance was assessed using parameters from the signal detection theory (Lockhart and Murdock, 1970). From the experimental conditions (target vs. distractor) and the participants' behavioral responses ("*Yes*" vs. "*No*"), four response categories were defined: Hit and Miss occurred when the target items were accurately recognized or incorrectly rejected, respectively, and correct rejection (CR) and false alarm (FA) occurred when the distractor items were correctly rejected or incorrectly recognized, respectively. In the framework of the signal-detection theory, a memory score (*d*- *<sup>L</sup>*) reflected the participant's ability to discriminate between the target and distractor items. This score was determined from the Hit and FA scores and was calculated as follows:

$$d\_L' = \ln \frac{HR(1 - FR)}{FR(1 - HR)}$$

Where *HR* represents the Hit rate [(Hit + 0.5)/(*Nt* + 1)], *FR* represents the false alarm rate [(FA + 0.5)/(*Nd* + 1)] and *Nt* and *Nd* represent the number of target and distractor odors, respectively, for which the participants provided an answer. Memory scores may be good or poor (positive or negative values, respectively).

In the episodic retrieval test, we focused the analyses on the participants' accurate responses for the target odors (Hit). Four types of responses were then defined depending on the recall accuracy. When the participants correctly recognized the target odors, they could accurately remember both the location and the context (WWW), the location only (WWhere), or the context only (WWhich) or they could be mistaken about both dimensions (What). These different scenarios were named *episodic combinations*. The theoretical proportions of these episodic combinations resulting from responses given randomly were 0.019 for WWW [1 response ("*Yes/No*") out of 2 ∗ 1 context out of 3 ∗ 1 location out of 9], 0.148 for WWhich [1 response ("*Yes/No*") out of 2 ∗ 1 context out of 3 ∗ 8 locations out of 9], 0.037 for WWhere [1 response ("*Yes/No*") out of 2 ∗ 2 contexts out of 3 ∗ 1 location out of 9] and 0.296 for What [1 response ("*Yes/No*") out of 2 ∗ 2 contexts out of 3 ∗ 8 locations out of 9].

The response times for odor recognition and episodic retrieval were considered. The response times corresponded to the durations between the first inspiration after the odor was delivered and 1) the "*Yes/No*" response for the odor recognition task and 2) the "*I remember the context*" response for the episodic retrieval task. The same time boundaries were used to record and analyze breathing parameters during the odor recognition and episodic retrieval tasks.

#### **STATISTICAL ANALYSIS**

Behavioral and physiological data were z-scored [(x − μ)/σ] at the individual level to remove bias based on inter-individual differences. The number of each response given during the odor recognition and episodic retrieval tasks was further normalized by the number of trials after removal of one odor *a posteriori* from the data ("*Odor intensity, pleasantness and familiarity*"). The statistic main effects of the factors and interactions were determined using repeated measurements ANOVAs followed by *posthoc* bilateral Student *t*-tests when main effects and/or interactions were significant. The effects were considered significant at *p <* 0*.*05. The relation between perceptual ratings of odors (intensity, pleasantness, familiarity) or memory performances with behavioral measures (number of clicks, delay between clicks) or breathing parameters was tested using Pearson tests. In these cases, to control for the Type I error rate associated to multiple comparisons, we applied the Bonferroni correction by dividing the probability alpha by the number of comparisons. Statistical analyses were performed using Statistica (StatSoft®, Tulsa, OK, USA).

### **RESULTS**

#### **ODOR INTENSITY, PLEASANTNESS, AND FAMILIARITY**

On average, the odorants were perceived as moderately intense (5.31 ± 1.44; range: 1.49–7.15), relatively neutral (4.85 ± 1.38 range: 2.22–6.92) and unfamiliar (4.54 ± 1.61; range: 1.60–7.33). The intensity of the allyl amyl glycolate was rated as weak (1.49 ± 1.93) when compared with that of the other odorants. The Grubbs test, which was used to test for outliers, indicated that this intensity value abnormally deviated from the mean (*G* = 2*.*66, *p* = 0*.*04). As a consequence, the data related to allyl amyl glycolate were excluded from further analyses.

#### **MEMORY PERFORMANCES**

The effects of the set of target odors (Set1 vs. Set2) selected for the participants of G1 and G2 and of the age of the episodes (1–3 days) on the behavioral and breathing responses observed during the encoding and retrieval sessions were evaluated. The influence of the repetition of the odors (5 times for targets and 3 times for distractors) on memory performances, response times, and breathing during retrieval was also tested. No significant main effects or interactions were found, and thus we did not take these factors into account in the subsequent analyses. Second, as the effect of context (coastal cliff, lavender field, and mountain landscape) was confounded with the nature of the three odors associated with each context, we could not specifically analyze it.

#### *Encoding*

The investigation of the odors during the encoding was analyzed as a function of the odor characteristics. The participants smelled, on average, each odor 5.5 (±2.6) times by clicking on the circles. The number of clicks for each odor for all participants was significantly negatively correlated with the odor intensity [*r* = −0*.*22, *t*(1*,*210) = 3*.*30, *p* = 0*.*001, αadjusted = 0.017] but not the odor familiarity and pleasantness (*ps >* 0*.*11). The mean delay between the two odor investigations was 29.8 (±13.5) s. These delays were not correlated with the intensity, pleasantness, or familiarity of the odors (*ps >* 0*.*05*,* αadjusted = 0*.*017). The duration, amplitude and volume of the inspirations and the respiratory frequency did not vary significantly as a function of the odor's intensity, pleasantness and familiarity (*ps >* 0*.*04*,* αadjusted = 0*.*017).

#### *Odor recognition*

The participants were presented the target and distractor odors and were asked whether they had smelled them during the encoding phase. The memory score was high (*d*- *<sup>L</sup>* = 2*.*33 ± 1*.*18), which indicated that the participants were very proficient in recognizing old odors and rejecting new ones. The proportions of the different response categories (Hit, Miss, CR, and FA) are shown in **Figure 3A**. The proportion of correct responses (Hit + CR) was significantly higher than the proportion of incorrect responses (Miss + FA) [*F*(1*,* 24) = 135*.*29, *p* = 0*.*0001]. While odor type (target vs. distractor) and response accuracy significantly interacted [*F*(1*,* 24) = 4*.*11, *p* = 0*.*045], no significant differences were observed between Hit and CR and between Miss and FA (*ps >* 0*.*06).

**Figure 3B** represents the influence of response accuracy (correct vs. incorrect) and odor type (target vs. distractor) on the response times. Response accuracy [*F*(1*,* 24) = 29*.*33, *p* = 0*.*001] but not odor type [*F*(1*,* 24) = 1*.*98, *p* = 0*.*17] significantly impacted the response times; the participants responded more rapidly when answering accurately (Hit + CR: 4.75 ± 1.71 s) than inaccurately (Miss + FA: 6.10 ± 2.44 s). Response accuracy and odor type significantly interacted [*F*(1*,* 24) = 9*.*17, *p* = 0*.*004]; the participants gave correct responses more rapidly than incorrect

responses when the target odors were presented (*p* = 0*.*001) but not when the distractor odors were presented (*p* = 0*.*19). The participants also answered more rapidly for the Hit responses than for the Miss, CR, and FA responses (*ps <* 0.001).

The breathing variations were analyzed as a function of response accuracy and odor type. No significant effects of response accuracy and odor type on the duration, amplitude and volume of the inspiration (*ps >* 0*.*23) or the respiratory frequency (*p* = 0*.*07) were found. However, a significant interaction was identified between both factors and the duration [*F*(1*,* 24) = 13*.*85, *p* = 0*.*001] and respiratory frequency [*F*(1*,* 24) = 7*.*51, *p* = 0*.*008] but not the amplitude and volume of the inspirations (*ps >* 0*.*18). As shown in **Figure 3C**, the duration of the participants' breath was shorter and their respiratory frequency was higher when they recognized the odors ("*Yes*" responses: Hit, FA) than when they rejected them ("*No*" responses: Miss, CR).

The recognition performances did not depend on the exploratory behavior of the odors during the encoding. The number of accurate odor recognitions (Hit) was not correlated with the number of clicks (*p* = 0*.*62, αadjusted = 0*.*025) and the mean delay between the clicks (*p* = 0*.*62, αadjusted = 0*.*025).

#### *Episodic retrieval*

When the participants recognized an odor as the target, they were asked to retrieve the spatio-contextual environment in which it occurred. We focused our analysis on the responses following correct odor recognition (Hit). The proportions of the episodic combinations are represented in **Figure 4A**. The proportions of WWW, WWhich and What were significantly higher than the proportion of WWhere [*F*(3*,* 66) = 20*.*55, *p* = 0*.*001; *post-hoc*, *ps <* 0.001]. The proportions of complete accurate (WWW) and partially accurate responses (WWhich, WWhere) that were given by the participants differed significantly from the random responses (*ps <* 0.017), while the proportion of inaccurate responses (What) did not differ from the proportion of random responses (*p* = 0*.*19). Thus, the participants were able to retrieve the spatio-contextual environment of the episodes using the recognition of an odor, they recalled only a part of the episode, or they did not recall anything and responded randomly. The subsequent analysis did not include the responses associated with the WWhere episodic combination because of the small amount of data.

The response times were then analyzed (**Figure 4B**). A significant main effect of the episodic combinations was found [*F*(2*,* 46) = 18*.*56, *p* = 0*.*001]. The response times of the participants were significantly faster for perfect accurate responses (WWW) than for partially inaccurate responses (WWhich: *p* = 0*.*016). The response times were even faster for WWhich than for inaccurate What responses (*p* = 0*.*001). In other words, the more incorrect the answers, the slower the participants answered. Interestingly, the time interval between the odor recognition and the episodic retrieval responses did not significantly vary with the episodic combinations [*F*(2*,* 46) = 2*.*11, *p* = 0*.*14].

The mean durations and volumes of the inspirations are given for the episodic combinations WWW, WWhich and What in **Figure 4C**. These durations and volumes significantly varied with the episodic combinations [*F*(2*,* 46) = 5*.*31, *p* = 0*.*008 and *F*(2*,* 46) = 4*.*88, *p* = 0*.*011, respectively]. The duration and volume of the inspirations were greater when the participants remembered the spatio-contextual environment associated with the odor (WWW) than when they did not remember it (What, *ps <* 0*.*001). No significant differences in the respiratory frequency and amplitude of the inspirations were observed (*ps >* 0*.*15).

The influence of the exploratory behavior of odors during encoding on the episodic performances was investigated. The number of accurate episodic retrievals (WWW) was not correlated with the number of clicks (*p* = 0*.*70), and the mean delay between clicks (*p* = 0*.*69).

Following this episodic retrieval, the strength of the association between the spatial location and the visual context of an episode was tested. On average, the participants accurately recollected 80 ± 7% of the spatial locations associated with each visual context. These performances did not significantly depend on the visual context [*F*(2*,* 46) = 1*.*76, *p* = 0*.*19], which indicated that no difference in the strength of the visuospatial associations biased the episodic performances.

#### **INFLUENCE OF EMOTION**

To investigate the influence of emotion on the memory performances, we created three odor pleasantness categories. Given that the pleasantness ratings of the odors widely varied among the participants (**Figure 5A**), we selected the two more pleasant,

normalized inspiration duration and volume for each episodic combination. The dashed horizontal lines indicate the random levels computed for the episodic combinations. Vertical bars represent the SD; ∗*p <* 0*.*05; ∗∗*p <* 0*.*01; ∗∗∗*p <* 0*.*001.

the two more neutral and the two more unpleasant odors for each participant. The odors selected for these three pleasantness categories differed significantly in terms of intensity [*F*(2*,* 46) = 15*.*14, *p* = 0*.*001] and familiarity [*F*(2*,* 46) = 20*.*37, *p* = 0*.*001]: the unpleasant odors were perceived as more intense and less familiar (6.36 ± 1.85; 3.05 ± 2.20, respectively) than the neutral odors (4.25 ± 2.01; 3.74 ± 2.38, respectively), while the pleasant odors (6.29 ± 1.45; 6.69 ± 2.09, respectively) were perceived as more intense and familiar than the neutral odors (*ps <* 0.001).

#### *On memory performances*

During the encoding, the number of clicks and the mean delay between two clicks did not differ between the pleasantness categories (*ps >* 0*.*71), indicating that the emotions carried by the odors did not influence their exploration.

The proportions of correct recognition (Hit) of odors differed significantly from the random responses whatever the emotion of odors (*ps <* 0*.*002), but it significantly varied as a function of the pleasantness category [*F*(2*,* 46) = 5*.*42, *p* = 0*.*007; **Figure 5B**]. The pleasant and unpleasant odors were recognized more accurately than the neutral odors (*p* = 0*.*024 and *p* = 0*.*003, respectively).

Considering episodic retrieval performances, the proportions of complete accurate responses (WWW) differed significantly from the random responses when triggered by pleasant and unpleasant (*ps <* 0*.*042) but not neutral odors (*p* = 0*.*72). The proportion of partial accurate responses (WWhich) significantly varied from random responses when triggered by pleasant odors only (*p* = 0*.*042; neutral and unpleasant odors, *ps >* 0*.*12), while the proportion of inaccurate responses (What) did not differ from the proportion of random responses whatever the pleasantness category of the odors (*ps >* 0*.*20). We observed a significant effect of the pleasantness category on the number of accurate episodic retrieval (WWW) responses [*F*(2*,* 46) = 3*.*27, *p* = 0*.*046, **Figure 5C**] but not on the number of partial episodic retrieval (WWhich, **Figure 5D**) or inaccurate episodic retrieval (What, **Figure 5E**) responses (*ps >* 0.56). The number of WWW was significantly higher when the odors that triggered the memory were more pleasant or more unpleasant than neutral (*p* = 0*.*047 and *p* = 0*.*024, respectively). No significant difference was found between the pleasant and unpleasant odors (*p* = 0*.*79). Thus, the emotion carried by the odors only improved the retrieval of accurate episodic memories, regardless of the positive or negative valence of the emotion. Importantly, while odor pleasantness categories differed in terms of familiarity and intensity, the accurate odor recognition (Hit) and episodic retrieval (WWW) performances were not significantly related to these ratings (*ps >* 0*.*49).

#### *On response time and breathing*

Regardless of the performances, the participants answered with similar response times regardless of the pleasantness category of the odors during odor recognition [*F*(2*,* 46) = 0*.*97, *p* = 0*.*39] and episodic retrieval [*F*(2*,* 46) = 1*.*26, *p* = 0*.*30]. Regardless of the performances, the participants answered with similar response times regardless of the odor pleasantness category during odor recognition [*F*(2*,* 46) = 0*.*97, *p* = 0*.*39] and episodic retrieval [*F*(2*,* 46) = 1*.*26, *p* = 0*.*30]. Performing two-way Session x Category ANOVAs on breathing data, we found a significant effect of pleasantness category on inspiration volume and duration [*F*(2*,* 48) = 5*.*42, *p* = 0*.*008 and *F*(2*,* 48) = 5*.*66, *p* = 0*.*006, respectively], and significant effects of pleasantness category and sessions on respiratory frequency [*F*(2*,* 48) = 3*.*34, *p* = 0*.*044 and *F*(2*,* 48) = 6*.*56, *p* = 0*.*003, respectively]. No significant effect was found for amplitude, and no significant interaction between factors was found whatever the breathing parameters. Thus, participants inspired more deeply, with longer inspirations, and less frequently for neutral and pleasant odors than unpleasant odors, whatever the session (*ps* = 0*.*017). They inspired also less frequently during episodic retrieval than during encoding (*p* = 0*.*018).

### **DISCUSSION**

The present novel laboratory-based episodic memory approach, which was adapted from a previous paradigm developed by our team (Saive et al., 2013), succeeded in the formation and subsequent retrieval of an integrated and multimodal memory of episodes comprising odors (What) localized spatially (Where) within a visual context (Which context). Successful odor recognition and episodic memory were not related to differences in the odor investigation at encoding and were paralleled by modifications in both the response time and breathing patterns. However, memory performances were influenced by the emotional content of the odor, with both pleasant and unpleasant odors generating higher recognition and episodic retrieval than neutral odors.

#### **RECOGNITION AND EPISODIC MEMORY PROCESSES**

The behavioral data revealed a high ability to recognize odors previously encountered in laboratory settings. The unfamiliar odors freely encoded during episode discovery were proficiently recognized among the new odors encountered afterwards, as indicated by a very high memory score. The good memory recognition performances were supported by the behavioral measures. The participants answered more rapidly when they successfully recognized the target odors than for all the other responses. Moreover, the duration of the participants' breath was shorter and their respiratory frequency was higher when they accurately recognized the odors than when they rejected them. These response times and breathing observations are consistent with previous reports (Jehl et al., 1997; Olsson and Cain, 2003; Royet et al., 2011) and could be evidence for a serial identity matching process between the memory trace and the actual percept (Bamber, 1969). Until a match was found between the odor cue and the odor memory traces, the participants needed to follow the memory search (which ended in higher response times for *No* than *Yes* responses) and keep the odor "in their nose," which led to expanded respiratory cycles. These results demonstrate the efficiency of our paradigm in generating the encoding of unknown odors and their later recognition.

The old odors were not only very well recognized but they also triggered the retrieval of past unique episodes at a level far above chance. From the accurate recognition of an odor, the participants were able either to retrieve the complete visuospatial context of the episodes or correctly recall only the context of the episodes. Otherwise, they did not remember any information related to the episode and answered randomly. Two scenarios are possible to explain the cognitive processes engaged in episodic retrieval: a serial recollection of the three dimensions (What, Where, and Which context) or an immediate recall of the whole episode. In the first scenario, when an odor was recognized, the participants interrogated their memories until the exact position of the odor in the exact context was recalled. In the second scenario, the episode was fully recovered from odor perception, all of its dimensions at once. The analysis of the response times revealed that the more information the participants retrieved about the episode, the faster they answered. However, the time period between odor recognition and episodic retrieval remained constant regardless of the accuracy of the episodic retrieval; this finding suggests that the content of the memory was already fully recovered from the odor recognition or that the episodic retrieval was already fairly advanced. Therefore, the response time data more strongly support the retrieval of the whole episode at once rather than a serial recall of its dimensions. The detailed analyses of the cognitive processes involved in our paradigm led us to support for the collapse of the recognition and episodic retrieval processes into a unique memory retrieval process when the binding between the odors and the spatio-contextual features of the episode is successful. The odor perception might generate the simultaneous recognition of the odor and the recall of other episodic features, such as the characteristics of the odor, the localization of the orange circle on the visual background or the mood the participants were in. These memories seem to be triggered as soon as the participants smelled the odor. Therefore, the odor recognition of the odor would be included in the episodic retrieval as one feature of the episode. Otherwise, when unsuccessful, the recognition and episodic retrieval memory process might be distinct.

Recognition and episodic performances were independent of the way the odors were investigated at encoding and the odors' intrinsic characteristics. The only exception was the odors that were less intense and were investigated more often, most likely to better characterize them. Given the amount of evidence indicating a serial position effect on recognition memory, with first and more recent items more likely to be recognized (Deese and Kaufman, 1957; Murdock, 1962), as well as on autobiographical memory, with events from late childhood or young adulthood and recent events more likely to be remembered (Crovitz and Schiffman, 1974; Crovitz and Quina-Holland, 1976), we might have expected primacy and recency effects to be observed. However, our data demonstrated that odor recognition and episodic memories were similar whether the day of encoding was the first, second or the last day, thereby confirming previous results (Saive et al., 2013). Thus, these performances were stable over time and were not dependent on the age of the retrieved episode. Furthermore, the performances were not impacted by the multiple presentations of the odors during the retrieval phase, although it has been demonstrated that repeated presentations of odors increase their familiarity (e.g., Jehl et al., 1995). These high and stable memory performances might reflect the influence of the multimodality and the episodic nature of our task. Odors are better recognized when associated with indices of other modalities or when associated with an episode of life during encoding (Lyman and McDaniel, 1986, 1990). When exploring the episodes, the participants were experiencing a new, rich and complex event, very similar to real-life encoding situations, which enhanced the strength of the whole memory trace. The full nights of sleep obtained between the encoding sessions may also have strengthened the consolidation of the memory traces and limited the interference between the episodes (Maquet, 2001; Stickgold, 2005; Alger et al., 2012; Abel and Bäuml, 2014).

Odors that triggered the retrieval of the spatio-contextual environment were associated with increased duration and volume of inspirations compared with odors that did not trigger any recall. These data are consistent with previous studies investigating breathing during autobiographical retrieval (Masaoka et al., 2012a,b). The current variation in breathing during memory construction raises interesting questions. Were the physiological responses a consequence of a successful episodic search or were they necessary for the search to be successful? In other words, were the breathing characteristics modified by the retrieval of the elements of the episodes or did they reflect an intense memory search? These questions are reminiscent of findings that showed attention and mental imagery processes are associated with larger sniffs when participants succeed in the tasks (Bensafi et al., 2003, 2005; Plailly et al., 2008). It is further possible that the reconstruction of the memory necessitated a relaxed state that was reflected in slower respiration. A previous study showed that yoga breathing specifically increased spatial memory performances (Naveen et al., 1997).

#### **IMPACT OF EMOTION GENERATED BY ODOR ON MEMORY RETRIEVAL**

Compared to neutral odors, both pleasant and unpleasant odors generated increased recognition and more complete episodic retrieval. This suggests that the intensity of the emotion, also called emotional arousal, but not the valence (pleasant vs. unpleasant) enhanced memory retrieval. Many studies have indicated an emotional arousal benefit on memory in humans (Burke et al., 1992; Cahill and McGaugh, 1995; Laney et al., 2004). For example, Cahill and McGaugh (1995) have shown that the higher the arousal content of a story, the better the long-term retention. This beneficial aspect of human memory would be highly adaptive, enabling more efficient accessibility of emotional memory, and is strongly dependent on the amygdala (Hamann, 2001). Interestingly, the effect of emotion on accurate odor recognition was in fact only observed when the complete episode was accurately recalled. Incomplete or inaccurate recalls of the episodes were not influenced by emotion. The fact that the accurate recognition of the odor and the accurate retrieval of the episodes were affected the same way by emotion is another argument favoring the idea that, in the case of an efficient episodic retrieval, these two memory processes might be collapsed into a unique memory process.

When did emotion influence episodic memory? Emotion can modulate the creation, storage and recollection phases of episode processing (Holland and Kensinger, 2013). First, arousing items are noticed quickly, and attention is preferentially directed toward them, potentially promoting their encoding (Kensinger and Corkin, 2004; MacKay et al., 2004; Leclerc and Kensinger, 2008). Furthermore, both pleasant and unpleasant odors trigger the modulation of skin conductance and heart rate measures (Alaoui-Ismaïli et al., 1997a,b; Bensafi et al., 2002; Royet et al., 2003). Thus, in the present study, the odors might have generated automatic emotional responses that might have modulated the participant's attention and induced improved encoding of all associated information. Second, emotional arousal could also influence the memory consolidation. Indeed, it has been shown that sleep not only promotes the general consolidation of new acquired memory traces (Maquet, 2001; Stickgold, 2005) but also specifically supports emotional memories (Wagner et al., 2006; Holland and Lewis, 2007; Groch et al., 2013). Finally, emotion can modulate retrieval by increasing how easily the memory comes to mind following cue perception and by increasing the amount of remembered details (Kensinger, 2009; Melcher, 2010). In the current experiment, odor pleasantness influenced the accurate retrieval of olfactory episodes. Importantly, odor pleasantness did not differentially impact the exploratory behavior (number of clicks and delays between clicks) during encoding and its influence on breathing did not differ between sessions. Therefore, in the frame of the experimental conditions of our study, we can suggest that odor pleasantness had only an impact on the consolidation or memory retrieval but not on the encoding of the episodes.

Which memory process was influenced by emotion? In our case, the emotion triggered by odors enhanced both the odor recognition itself and the retrieval of the entire episode. Emotional arousal enhances the binding of contextual details or dimensions when they are an integral part of the emotional stimulus (Mather, 2007; Mather and Nesmith, 2008; Nashiro and Mather, 2011). In our study, we suggest that the dimensions of the episodes were encoded as features of the emotional odors and were combined in an integrated unique memory trace. Taken together, remembering how the features of an event were associated together is a critical aspect of episodic memory that seems to be promoted by emotion.

In conclusion, our study represents the first laboratoryecological approach involving olfactory dimension that allows the conscious and controlled recollection of specific and complex events from the past. It combines in a very original way the advantages of the laboratory-based approaches that allow the control of encoding and recall conditions, and of autobiographical-based approaches that enable the retrieval of real life episodes (Saive et al., in revision). Furthermore, of interest to the entire neuroscientist community devoted to the study of memory, our paradigm enables the ecological and direct comparison between episodic and recognition memory processes, rather than indirect assessment based on the comparison between recollection and familiarity processes engaged in simpler memory tasks.

It demonstrates that humans are capable of encoding and remembering rich and unique laboratory episodes triggered by odors. The episodic nature of the task generates high and stable memory performances, accompanied by slower and deeper breathing. It shows for the first time that the emotion carried by odors, regardless of their valence, does not influence encoding behavior but promotes their accurate recognition and the accurate retrieval of the visuospatial context of the episodes. Importantly, this study also suggests that when the binding between the odors and the spatio-contextual features of the episode is successful, the odor recognition and episodic retrieval collapse into a unique memory process that begins as soon as the participants smell the odors. However, further investigations are needed to validate this observation. The use of cerebral imaging techniques represents the ideal tool to test it. We hypothesize that the neural signature of the successful retrieval of episodic information will be observed from the mere odor perception.

#### **ACKNOWLEDGMENTS**

This work was supported by the Centre National de la Recherche Scientifique (CNRS), the LABEX Cortex (NR-11-LABX-0042) of Université de Lyon within the program "Investissements D'Avenir" (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR) and research grants from the Région Rhône-Alpes (CIBLE 10 015 772 01). Anne-Lise Saive was funded by the Roudnitska Foundation.

### **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: 24 March 2014; accepted: 19 May 2014; published online: 06 June 2014. Citation: Saive A-L, Royet J-P, Ravel N, Thévenet M, Garcia S and Plailly J (2014) A unique memory process modulated by emotion underpins successful odor recognition and episodic retrieval in humans. Front. Behav. Neurosci. 8:203. doi: 10.3389/fnbeh. 2014.00203*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Saive, Royet, Ravel, Thévenet, Garcia and Plailly. 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.*

## Learning to smell danger: acquired associative representation of threat in the olfactory cortex

### **Wen Li 1,2\***

<sup>1</sup> Department of Psychology, University of Wisconsin-Madison, Madison, WI, USA

<sup>2</sup> Waisman Center, University of Wisconsin-Madison, Madison, WI, USA

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Jonas K. Olofsson, Stockholm University, Sweden Tyler Lorig, Washington and Lee University, USA

#### **\*Correspondence:**

Wen Li, Department of Psychology, University of Wisconsin-Madison, Brogden Hall, 1202 W. Johnson Street, Madison, WI 53706, USA e-mail: wenli@psych.wisc.edu

Neuroscience research over the past few decades has reached a strong consensus that the amygdala plays a key role in emotion processing. However, many questions remain unanswered, especially concerning emotion perception. Based on mnemonic theories of olfactory perception and in light of the highly associative nature of olfactory cortical processing, here I propose a sensory cortical model of olfactory threat perception (i.e., sensory-cortex-based threat perception): the olfactory cortex stores threat codes as acquired associative representations (AARs) formed via aversive life experiences, thereby enabling encoding of threat cues during sensory processing. Rodent and human research in olfactory aversive conditioning was reviewed, indicating learning-induced plasticity in the amygdala and the olfactory piriform cortex. In addition, as aversive learning becomes consolidated in the amygdala, the associative olfactory (piriform) cortex may undergo (long-term) plastic changes, resulting in modified neural response patterns that underpin threat AARs. This proposal thus brings forward a sensory cortical pathway to threat processing (in addition to amygdala-based processes), potentially accounting for an alternative mechanism underlying the pathophysiology of anxiety and depression.

**Keywords: threat encoding, olfactory sensory cortex, acquired associative representation, aversive conditioning, olfaction, anxiety**

### **INTRODUCTION**

Whether it is roaming on the safari or dozing by the fireplace, the ability to quickly detect threat (e.g., a tiger or a burning rug) and initiate appropriate responses can mean life or death for an organism. Decades of neuroscience research in emotion processing has reached a strong consensus: the amygdala extracts biological significance of a sensory cue and initiates and controls affective and motivational responses to the stimulus (LeDoux, 2000, 2012; Adolphs, 2013). In terms of threat perception, a widely held view is that the amygdala projects emotionally charged outputs to the sensory cortex, thereby enabling perceptual analysis of potential danger (Phelps and LeDoux, 2005; Vuilleumier and Pourtois, 2007). These theories have been incorporated into neural models of emotional disorders, shedding important light on the pathophysiology of anxiety and depression (Davis, 1992; Rauch et al., 2006; Clark and Beck, 2010).

However, striking findings have arisen recently, suggesting that threat processing can operate independently of the amygdala. Patient S.M. an individual with complete bilateral amygdala lesions, demonstrated intact early threat perception (Tsuchiya et al., 2009); she and two other patients with similar amygdala lesions also developed panic attacks and intense fear when challenged with high concentration CO<sup>2</sup> (Feinstein et al., 2013). In addition, extensive research indicates very swift (at latencies around 100 ms) threat processing in the visual cortex (Li et al., 2007, 2008b; Krusemark and Li, 2011, 2013; Forscher and Li, 2012; also cf. Vuilleumier and Pourtois, 2007), preceding the latencies of threat processing measured with depth-electrode recording in the amygdala (Oya et al., 2002; Krolak-Salmon et al., 2004). Accordingly, instead of relying on amygdala input, early threat perception may depend on inputs from other brain areas (e.g., the orbitofrontal cortex; Barrett and Bar, 2009), or simply consummate during the initial sensory feedforward sweep. Indeed, a seminal paper by Pessoa and Adolphs (2010) promotes a multi-path framework, arguing for extra-amygdala neural circuits in threat processing.

Towards that end, this article draws evidence from longstanding animal research and recent human data (reviewed below), proposing a sensory cortical model of threat perception (i.e., sensory-cortex-based threat perception): the sensory cortex stores threat codes/representations, thereby enabling perceptual encoding of threat information once an environmental input reaches the sensory cortex. This model highlights an *active*, *independent* role of the sensory cortex in threat perception, as opposed to the conventionally understood role of passively processing/integrating threat-laden inputs generated elsewhere in the brain (e.g., the amygdala; Phelps and LeDoux, 2005; Vuilleumier and Pourtois, 2007). This model makes clear evolutionary sense by permitting categorization of biological significance in the stage of sensory analysis, prompting an organism to respond with minimal delay. Importantly, by putting forward a sensory-based mechanism, in addition to limbic-based threat processing, this model would help reconcile controversial findings in the literature as discussed above, such as the very swift threat perception, even in the absence of intact amygdala.

### **A (OLFACTORY) SENSORY CORTICAL MODEL OF THREAT PERCEPTION**

To provide a mechanistic explication of this account, a few principles need to be emphasized. As James (1890) asserted, "every perception is an acquired perception," human perception is largely learned and depends on long-term memory (Goldstone, 1998; Stevenson and Boakes, 2003). This proposed model thus takes a learning perspective, building on mnemonically-based threat codes/representations acquired through life experiences. Indeed, except for a limited set of innate phobic objects (e.g., snakes; Ohman and Mineka, 2001), the copious repertoire of threat cues in humans appears to be learned and accumulated over the course of life, varying from the concrete (e.g., germs and guns) to the abstract (e.g., disease and death).

In addition, given the associative nature of memory and threat processing, this model centers on the associative (secondary) sensory cortex, characterized by dense intrinsic and extrinsic neural connections, as a primary site of threat code storage and threat encoding. In particular, extrinsic top-down connections transmitting contextual information can modulate sensory cortical activity to facilitate context-relevant behavior (Cohen and Maunsell, 2011; Harris and Mrsic-Flogel, 2013); this focus on associative sensory cortex would thus permit context- or statedependent flexibility (adaptive to affective, motivational and physiological states) in perceiving threat while ensuring sensory fidelity (Proffitt, 2006; Barsalou, 2008; Krusemark and Li, 2013; Krusemark et al., 2013).

Furthermore, I chose olfaction as a model system for this account. The olfactory cortex has served as a model system for the cortical representation of associative memory (Gluck and Granger, 1993; Haberly, 1998), owing to the fact that olfactory perception is deeply rooted in memory (Stevenson and Boakes, 2003; Wilson and Stevenson, 2003) and that olfactory cortical processing is highly associative (Wilson and Sullivan, 2011). Moreover, akin to threat encoding specifically, olfaction is uniquely related to emotion in function and anatomy (Schiffman, 1974; Carmichael et al., 1994), given their phylogenetic proximity. Studies have shown that olfactory perception shifts readily with a perceiver's affective state (Herz et al., 2004; Chen and Dalton, 2005; Herz, 2005; Pollatos et al., 2007; Krusemark et al., 2013); odor affective value (vs. odor character, "lemon" or "orange") may even represent the dominant dimension in olfactory perception (Yeshurun and Sobel, 2010). Finally, odor hedonicity is posited to be borne directly out of emotional experiences attached to an odor (Herz, 2005). Taken together, these properties of olfactory perception represent a particularly close fit to the model here.

Mechanistically, this proposed account rests on long-term plasticity (as a form of long-term memory storage) in the sensory cortex, consequent to aversive associative learning. As accruing evidence suggests that the sensory cortex contains richly interconnected neurons, whose patterns of firing as a whole encode sensory input (Harris and Mrsic-Flogel, 2013), this model highlights modified neural response patterns induced by longterm plasticity in the sensory cortex. Critically, these patterns would reflect acquired associative representations (AARs) that encode the threat meanings learned via negative experiences (in addition to the sensory features of the stimuli). As such, constituting sensory neural codes of acquired threats, these threat AARs would underpin threat perception in the sensory cortex.

Based on the fear learning literature reviewed below, the genesis of threat-relevant sensory cortical long-term plasticity and threat AARs could involve two components (**Figure 1**): (1) acquisition/consolidation of aversive associative learning in the amygdala, thereby attaching threat meanings to innocuous odors; and (2) over time, the initial amygdala-based learning gives rise to long-term sensory cortical plasticity. That is, the associative olfactory (posterior piriform) cortex (PPC) undergoes plastic changes, resulting in an updated neural response pattern (i.e., a threat AAR) to the conditioned odor. Accordingly, subsequent encounters of the conditioned odor will activate this threat AAR in the PPC, supporting olfactory cortical encoding of threat. Finally, outputs from this sensory process (i.e., threat-laden sensory impulses) can trigger fear responses via projections to a wide range of associative neural networks (especially the amygdala, prefrontal cortex and brain stem structures).

### **AMYGDALA MEDIATES THE ACQUISITION/CONSOLIDATION OF OLFACTORY AVERSIVE CONDITIONING**

It is a well-known fact that repeated paired stimulation of a stimulus (CS, e.g., a tone) and a salient stimulus (US, e.g., an electric shock or a drop of water) often result in conditioning (e.g., Pavlovian or emotional conditioning), and the CS thus acquires a new threat/reward meaning (Pavlov, 1927/1960). In terms of the neural mechanism, extensive research has ascribed a key role to the amygdala, especially the basolateral complex (comprising the lateral, basal and accessory basal nuclei), in aversive conditioning. As described in influential reviews (LeDoux, 2000, 2012; Maren and Quirk, 2004; Myers and Davis, 2007), the lateral nucleus of amygdala reliably exhibits increased spike firing and longterm potentiation during conditioning, underscoring the lateral nucleus as a primary site of conditioning acquisition and consolidation. Furthermore, pre-training damage to the lateral nucleus directly impairs fear conditioning, indicative of its causal role in this process. Via direct or indirect intra-amygdala connections, the lateral nucleus triggers activation of the central nucleus of the amygdala, which initiates and controls the expression of the acquired fear via projections to a set of midbrain and brainstem structures (e.g., hypothalamus and periaqueductal gray). Finally, it is worth noting that for auditory fear conditioning, the magnocellular medial geniculate nucleus could mediate the initial learning (Weinberger, 2011).

The aversive conditioning literature has primarily involved the auditory sense and, to a lesser extent, the visual sense. Nevertheless, olfactory conditioning research has yielded similar conclusions (cf. Mouly and Sullivan, 2010). Electrophysiological studies in rodents indicate that the basolateral amygdala exhibits potentiated responses during olfactory aversive conditioning (Rosenkranz and Grace, 2002) and shortly after (Rattiner et al.,

2004; Sevelinges et al., 2004). Highlighting its critical role in olfactory conditioning acquisition, pre-training lesions or pharmacological inactivation/inhibition of the basolateral amygdala significantly reduces conditioned fear or aversion to the CS odor (Cousens and Otto, 1998; Wallace and Rosen, 2001; Walker et al., 2005; Miranda et al., 2007; Sevelinges et al., 2009). Furthermore, post-training inactivation of the basolateral amygdala (Kilpatrick and Cahill, 2003; Sevelinges et al., 2009) would largely attenuate conditioned aversion, implicating this area in the consolidation of olfactory aversive associative learning. Using biomarkers of synaptic plasticity to reflect fear learning, research also further reveals plastic changes, during or shortly after conditioning, in the basolateral amygdala of rodents exposed to paired odor-shock stimulation, in the form of heightened expression of brain derived neurotrophic factor (BDNF; Jones et al., 2007) or increased concentrations of glutamate and GABA (Hegoburu et al., 2009).

trigger emotion responding. **(B)** Neural mechanisms. Initial association

A substantial body of human neuroimaging research in aversive conditioning has emerged, albeit largely concerning auditory and visual CS (Sehlmeyer et al., 2009). Human neuroimaging research of olfactory aversive conditioning remains scant. Functional magnetic resonance imaging (fMRI) data from our lab indicate conditioned responses evoked by the CS odor during the acquisition phase (**Figure 2A**, Li et al., 2008a). That is, conforming to an exponential decay observed in prior imaging studies of visual aversive conditioning (Büchel et al., 1998; LaBar et al., 1998), the amygdala response to the CS odor (vs. CS- odor) increases sharply in early trials and declines in later trials. In addition, pairing odors with painful (CO2) trigeminal stimulation in human subjects, a new fMRI study reveals significant response enhancement to the CS odor in the amygdala during conditioning (Moessnang et al., 2013). These extant findings thus concur with conclusions of the general literature, confirming the role of amygdala in the acquisition of human olfactory conditioning.

hippocampus; CM = corticomedial nucleus of amygdala.

Notably, the olfactory anatomy is fairly distinct from other senses; it lacks the thalamic relay critical for signal transmission in other modalities, and its inputs terminate in the corticomedial nucleus of amygdala (Carmichael et al., 1994) versus the lateral nucleus for other sensory inputs (Luskin and Price, 1983; Savander et al., 1996). Despite these disparities, fear learning in the amygdala is fairly generic, contrasting with sensory-specific plasticity in the olfactory cortex.

### **OLFACTORY CORTEX SUPPORTS ACQUIRED ASSOCIATIVE REPRESENTATIONS (AARs) VIA OLFACTORY AVERSIVE CONDITIONING**

#### **TRANSFER OF AVERSIVE ASSOCIATIVE LEARNING FROM THE AMYGDALA TO THE OLFACTORY CORTEX**

As discussed earlier, the current model requires long-term learning-based plasticity in the sensory cortex to substantiate threat codes (threat AARs). In fact, although the role of amygdala in long-term fear memory is still debatable (LeDoux, 2000; McGaugh, 2004), the sensory cortex has long been implicated as a site of storage and retrieval of remote associative memory (Mishkin, 1982; Squire, 1987; Damasio, 1989). Namely, long-standing views posit that as the memory of an object becomes fully consolidated in mediotemporal structures (e.g., the hippocampus), the object-specific sensory cortex (e.g., auditory

enantiomer counterpart improves markedly after conditioning whereas the unconditioned enantiomer pair remains indistinguishable. **(C)** In parallel, response patterns for the CS pair become divergent (relative to the non-conditioned/nCS pair). Differential odor maps (spatial configurations of response intensities in all active PPC voxels) within each pair are displayed at the top of the bar graph, with strong-colored voxels reflecting large disparities between the counterparts. Notably, the post-conditioning differential map for the CS pair contains far more voxels of strong colors. **(D)** Plasticity in the PPC—enhanced response to the target odor after prolonged mere exposure.

odors (not shown here). This enhanced amygdala-olfactory-cortex connection may facilitate the transfer of learning from the amygdala to the olfactory cortex. Yellow lines represent intrinsic connections initially significant, green lines those that become significant in anxiety and red intercepting lines modulation by odors in anxiety. OFC = orbitofrontal cortex; PPC = posterior piriform cortex; pgACC = pregenual anterior cingulate cortex; Amyg. = amygdala; olf. = olfactory; APC = anterior piriform cortex. Panels A–C are adapted from Li et al. (2008a), Panel D from Li et al. (2006) and Panel E from Krusemark et al. (2013).

cortex to a tone or olfactory cortex to an odor) gradually takes over to support long-term storage of the memory (Haberly and Bower, 1989; Gluck and Granger, 1993; Gluck and Myers, 1993; Squire and Wixted, 2011). Regarding aversive associative learning in particular, the sensory cortex could undergo plastic changes to serve as a primary storage site of long-term fear memory, following the initial fear learning (Weinberger, 2004).

Three mechanisms may mediate this transfer (**Figure 1B**). Firstly, long-range, low-frequency (theta) oscillatory activity in the amygdala is potentiated following conditioning, thereby facilitating amygdala interaction with sensory cortical storage sites to induce plasticity in these regions (Haberly and Bower, 1989; Gluck and Granger, 1993; Paré et al., 2002). Secondly, aversive conditioning potentiates amygdala efferents to the nucleus basalis, driving its acetylcholine release in the sensory cortex to mediate long-lasting sensory cortical plasticity (McGaugh et al., 2002; Weinberger, 2007). Thirdly, negative affective states induced by conditioning can intensify amygdala efferents to the sensory cortex to induce cortical plasticity, as suggested by a recent human fMRI in our lab (Krusemark et al., 2013). Combining dynamic causal connectivity analysis (Friston et al., 2013) and anxiety induction in a simple odor detection task, we demonstrate that an induced anxious state can reorganize the olfactory sensory circuitry, incorporating the amygdala as an integral step. That is, following anxiety induction, initially insignificant efferents from the anterior piriform cortex (APC) to the amygdala become important; also, efferents from the amygdala to PPC are further strengthened (**Figure 2E**). Notably, this circuitry reorganization is accompanied by a negative shift in perceived odor valence. Conceivably, by (almost invariably) inducing anxious/negative affective states, aversive conditioning can similarly enhance amygdala discharges to the PPC, promoting plastic changes in this area.

In consequence, long-lasting plasticity would arise in the sensory cortex, which then selectively updates neuronal ensemble response patterns to the CS, substantiating the long-term memory of acquired threat value in the CS. Notably, Weinberger's lab was the first to show that the primary auditory cortex is the locus of long-term plasticity due to auditory fear conditioning, supporting altered sensory encoding of auditory CS (Weinberger, 2007, 2011; Weinberger and Bieszczad, 2011). To date, there has been considerable evidence of olfactory cortical plasticity as a result of olfactory aversive conditioning.

#### **OLFACTORY CORTICAL PLASTICITY INDUCED BY OLFACTORY AVERSIVE CONDITIONING**

The olfactory cortex consists of the anterior olfactory nucleus, olfactory tubercle, cortical nucleus of the amygdala, piriform cortex and entorhinal cortex (Carmichael et al., 1994; Shipley and Ennis, 1996; Haberly, 1998). The piriform cortex, divided into anterior and posterior piriform cortices (APC and PPC), is the largest subarea of the olfactory cortex. As described in excellent recent reviews (Gottfried, 2010; Mori and Sakano, 2011; Wilson and Sullivan, 2011), the APC serves as a primary olfactory cortex influenced strongly by bulbar mitral cell afferents and thus maintains considerable fidelity to the molecular properties of an odorant, whereas the PPC anatomically and functionally resembles a higher-level association cortex, supporting higherorder olfactory perception (e.g., odor quality encoding and categorization). Therefore, the PPC is postulated as the primary locus of olfactory threat AARs and threat encoding in the current model (**Figure 1**).

Indeed, akin to the associative and malleable nature of PPC, computational modeling and *in vitro* physiological studies suggest that long-term potentiation is more readily induced in the PPC (than APC), enabling long-term memory storage, whereas the APC is more associated with sensory processing and simple forms of short-term memory (Lynch and Granger, 1989; Jung et al., 1990). Empirical data of olfactory aversive conditioning largely concur with this view. After olfactory conditioning, PPC (but not APC) in the trained animals exhibits stronger local field potentials (Litaudon et al., 1997; Sevelinges et al., 2004) or BDNF expression (Jones et al., 2007) compared to the baseline. Hegoburu et al. (2009) also demonstrate increases in GABA and glutamate concentrations in the PPC of trained rats. To note, this PPC plasticity persists up to 30 min into odor-shock conditioning, contrasting with transient concentration increases of these amino acids in the basolateral amygdala (observed in the same study). These distinct plasticity time courses may correspond to the differential functions of these two areas: the amygdala is critical for initial learning while the PPC is important for long-term memory of learning. It is also worth noting that CS-odor evoked response potentiation has been observed in the APC of awake rats using single-unit recording (Barnes et al., 2011; Chen et al., 2011). This positive finding could be ascribed to the efficacy of the methodology, but as PPC activity was not assessed in these studies, it is unclear whether PPC plasticity coexisted or even mediated the APC changes. Finally, recent evidence also suggests that fear learning can induce plasticity in even lower levels of the olfactory hierarchy (i.e., olfactory receptor neurons and the olfactory bulb; Jones et al., 2008; Kass et al., 2013).

Concerning human evidence, our aforementioned conditioning study also shows that the PPC (but not APC) exhibits significant CS-evoked response changes, paralleling the rodent findings (**Figure 2C**, Li et al., 2008a); this superior plasticity in PPC further accords with a previous fMRI study in the lab involving simple perceptual learning, where the PPC but not APC demonstrates response enhancement following prolonged odor exposure (Li et al., 2006; **Figure 2D**). Critically, our conditioning study reveals that behavioral and PPC neuronal ensemble responses to two initially indistinguishable odor enantiomers (mirror-image molecules) become distinct after one of them is paired with electric shock (**Figures 2B** and **C**, Li et al., 2008a). This divergence contrasts with convergent response augmentation in the amygdala to both the CS odor and its non-conditioned enantiomer counterpart (**Figure 2A**). These findings thus emphasize that PPC plasticity is selective to the CS odor, independent of amygdala plasticity that is generalized to similar cues. This specialized PPC plasticity could thus underpin sensory cortical representation of the acquired threat value in the CS odor, supporting discrimination perception of the CS odor versus its counterpart following conditioning.

Finally, direct evidence for *long-term storage* of aversive conditioning in the olfactory cortex has emerged from a series of experiments conducted by Sacco and Sacchetti (2010). The authors demonstrate long-term memory of shock conditioning: trained rats exhibit strong freezing responses to the CS one month posttraining. Nevertheless, lesioning CS-specific secondary sensory (visual, auditory and olfactory) cortex (including the PPC), one month post-training, largely abolishes the conditioned responses. Notably, lesions to the secondary sensory cortex do not impair new fear learning or recent (e.g., 1 day post-training) fear memories; this coincides with previous research which induced sensory cortical lesions either before or shortly after conditioning and failed to find impairment in fearing learning or recent fear memory (Romanski and LeDoux, 1992; Rosen et al., 1992; Falls and Davis, 1993; Campeau and Davis, 1995), suggesting that the sensory cortex is not critical for acquisition and consolidation of fear memory. Nonetheless, this recent study evinces that the secondary sensory cortex is essential for long-term storage and retrieval of acquired threat value in the CS. That is, as time elapses after initial learning, the CS will need to activate the threat AAR stored in the secondary (associative) sensory cortex to elicit conditioned responses.

### **CONCLUSION**

This review integrates mnemonic theories of olfaction and evidence of olfactory aversive associative learning, promoting a sensory cortical model of threat perception. The amygdala may mediate olfactory associative learning and transfer this learning to the olfactory cortex. The consequent long-term plasticity in the olfactory associative cortex (PPC) may serve to support olfactory threat AARs (representing acquired threat value in conditioned odors). These threat AARs can independently enable sensory cortical encoding of threat and trigger various responses via projections to associative neural networks. This sensory pathway may specialize in mandatory, reflexive, and sensory-specific forms of threat encoding, whereas the amygdala, especially via interaction with sensory and ventral prefrontal cortices, may chiefly be responsible for flexible, context-relevant, and amodal (abstract) threat processing (Krusemark and Li, 2013). In addition, these parallel systems may confer further ecological advantage by integrating a fine-tuned sensory module for *specific* threat identification and a broadly-tuned amygdala module for *sensitive* threat detection (Li et al., 2008a). By elucidating threat encoding in the sensory cortex, this proposed model may provide new insights into the pathophysiology of emotional disorders, pointing to a concrete clinical intervention target in the sensory cortex.

#### **LIMITATIONS AND FUTURE DIRECTIONS**

Compared to rich data in auditory fear learning (Weinberger, 2004, 2007), a limitation of the olfactory fear learning literature is the limited evidence of long-term plasticity/memory storage in the sensory cortex, which is a critical neural basis of the neurosensory account of threat perception. That is, very few studies in this literature have assessed olfactory sensory cortical plasticity after a prolonged delay (e.g., 2 weeks or 1 month after initial learning), and there is virtually no human evidence. Therefore, future research is warranted to isolate evidence of long-term memory of acquired threat in the sensory cortex, especially in human subjects.

Another notable limitation concerns the spatial resolution of the fMRI methodology, a first-line non-invasive method in probing neural activities in humans. With millions of neurons in each fMRI voxel (a volume of few cube millimeters; Logothetis, 2008), patterns of voxel-wise fMRI signal intensity would reflect large-scale configurations of neural ensemble activity as opposed to single-unit neuronal response patterns. Also, various methods have been applied in pattern-based fMRI analysis, warranting comparisons and cross-validations of those findings in the future. That said, it is also worth noting that given the considerable level of redundant coding in neuronal populations in the sensory cortex (e.g., the often correlated firing of neighboring neurons; Smith and Kohn, 2008; Luczak et al., 2009), these large-scale patterns observed in humans could still provide useful insights into sensory coding, especially when integrated with animal electrophysiological data.

#### **ACKNOWLEDGMENTS**

The author would like to thank N. Weinberger for excellent input and the members of the Li Laboratory for comments and assistance. The author is supported by R01MH093413.

#### **REFERENCES**


of the amygdala in the rat. *Neurosci. Lett.* 211, 167–170. doi: 10.1016/0304- 3940(96)12750-6


**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: 07 January 2014; accepted: 09 March 2014; published online: 07 April 2014.*

*Citation: Li W (2014) Learning to smell danger: acquired associative representation of threat in the olfactory cortex. Front. Behav. Neurosci. 8:98. doi: 10.3389/fnbeh.2014. 00098*

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

*Copyright © 2014 Li. 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*.

## Sweet reward increases implicit discrimination of similar odors

#### **Eva Pool 1,2\*, Sylvain Delplanque1,2 , Christelle Porcherot <sup>3</sup> , Tatiana Jenkins <sup>2</sup> , Isabelle Cayeux<sup>3</sup> and David Sander 1,2**

<sup>1</sup> Swiss Center for Affective Sciences, University of Geneva-CISA, Geneva, Switzerland

<sup>2</sup> Laboratory for the Study of Emotion Elicitation and Expression, Department of Psychology, FPSE, University of Geneva, Geneva, Switzerland

<sup>3</sup> Firmenich, SA, Geneva, Switzerland

#### **Edited by:**

Nadine Ravel, Center for Research in Neuroscience of LYON (CRNL), France

#### **Reviewed by:**

Emma Robinson, Bristol University, UK Tao Jiang, Lyon Neuroscience Research Center CRNL, France

#### **\*Correspondence:**

Eva Pool, Swiss Center for Affective Sciences, University of Geneva-CISA, Case Postale 60, 1211 Geneva 20, Switzerland e-mail: eva.pool@unige.ch

Stimuli associated with emotional events signal the presence of potentially relevant situations, thus learning to rapidly identify this kind of stimuli can be highly beneficial. It has been demonstrated that individuals acquire a better perceptual representation of stimuli associated with negative and threatening emotional events. Here we investigated whether the same process occurs for stimuli associated with positive and rewarding emotional events. We used an appetitive Pavlovian conditioning paradigm during which one of two perceptually non-distinguishable odors was associated with a rewarding taste (i.e., chocolate). We investigated whether appetitive conditioning could improve the recognition of the odor associated with the reward, rendering it discriminable from its similar version that was never associated with the reward. Results revealed a dissociation between explicit perception and physiological reactions. Although participants were not able to explicitly perceive a difference, they reacted faster, inhaled more and had higher skin conductance responses when confronted with the reward-associated odor compared to its similar version that was never associated with the reward. Our findings have demonstrated that positive emotional associations can improve the implicit perceptual representation of odors, by triggering different physiological responses to odors that do not seem to be otherwise distinguishable.

**Keywords: reward, odor discrimination, emotional learning, perceptual representation, incentive salience, taste-odor conditioning**

### **INTRODUCTION**

Understanding how organisms deal with their limited attentional resources in an environment composed of a virtually infinite number of stimuli has always been of main interest for cognitive sciences (Posner, 1980). For instance, if two persons are chatting and watching their children on a playground, they will not be able to represent every action their children make, because a part of their resources is invested in the conversation. Nonetheless, some particular actions such as the children crying or calling for them will be represented in spite of the conversation. Why is the processing of those particular stimuli privileged? It has been proposed that emotional stimuli may have a facilitated access to an organism's attentional resources (Vuilleumier, 2005).

A large amount of evidence has shown that emotional stimuli are difficult to ignore (e.g., Segerstrom, 2001; Compton et al., 2003); that our gaze is rapidly oriented toward them (e.g., Nummenmaa et al., 2009; Theeuwes and Belopolsky, 2012); that disengaging attention from them is hard (e.g., Fox et al., 2001; Yiend and Mathews, 2001; di Pellegrino et al., 2011); and that in a complex environment with competing stimuli, they have a prioritized access to attentional resources (e.g., Ohman et al., 2001; Anderson, 2005; Hodsoll et al., 2011). These effects of emotional stimuli on cognitive processing have been shown to be at least mediated by an enhancement of the neuronal activity linked to sensorial processing. Indeed, neuroscientific evidence strongly suggests that sensorial information is processed more efficiently if it has emotional content (see Phelps and Ledoux, 2005; Vuilleumier, 2005, for a review). More particularly, the activity linked to sensory processing at very early stages is amplified during the processing of emotional stimuli, as demonstrated with electroencephalography, (e.g., Pourtois et al., 2004; Brosch et al., 2008; Hickey et al., 2010) and functional imagery (e.g., Vuilleumier et al., 2001; Grandjean et al., 2005; Mohanty et al., 2008).

The degree of specificity of this emotional facilitation remains to be fully understood. Is the perceptual enhancement selective for a particular emotional stimulus or does it generalize to neutral stimuli with similar perceptual characteristics? If it is specific, the discrimination between an emotional stimulus and a perceptually similar but neutral stimulus should be easier compared to the discrimination between two neutral stimuli. Li et al. (2008) investigated these questions for the processing of negative threatening stimuli by using a triangular test to measure the discriminability of similar odors (i.e., enantiomers) combined with an aversive emotional learning paradigm. The triangular test is a forced choice procedure used to investigate perceptual discriminability, in which a stimulus is presented twice with a third different but similar stimulus that has to be identified (Laska and Teubner, 1999). Through this procedure Li et al. (2008) demonstrated that after aversive emotional learning, during which one of the two odors was associated with an unpleasant electric shock, two odors that were initially indistinguishable became discriminable. The enhancement of the sensorial abilities through aversive emotional learning seems to thus be highly selective, when the learning context requires high discrimination abilities. This mechanism could be highly adaptive, because it allows the organism to discriminate stimuli with an emotional meaning from other less meaningful but similar stimuli and thereby preventing the organism to overreact when it is not necessary. To the best of our knowledge, this mechanism has never been tested for positive emotional stimuli. Positive and negative emotional stimuli should have the same influence on the sensorial processing, because the privileged status of the emotional stimuli would depend on the affective relevance of the stimulus for the concerns of the individual perceiving it, rather than on a distinct valence specific mechanism (Frijda, 1988; Sander et al., 2005). If an individual learns to associate an intrinsically neutral stimulus to an emotional event, either a threat or a reward, the stimulus that was initially neutral but acquired affective relevance through the learning experience also gains a privileged access to the individual resources (for aversive learning see Alpers et al., 2005; Koster et al., 2005; for appetitive learning see Seitz et al., 2009; Austin and Duka, 2010; Hickey et al., 2010; Anderson et al., 2011; Notebaert et al., 2011; Pool et al., 2014).

Based on this assumption, we investigated whether (i) discrimination capabilities could be enhanced through an appetitive emotional learning using positive rewarding stimuli and, if any, (ii) the degree of the selectivity of this enhancement at an implicit and an explicit level. To do so, we adapted and modified the experimental procedure used by Li et al. (2008), by combining a triangular test in which participants were asked to distinguish between two similar odors with an appetitive emotional learning paradigm. We selected two pairs of indistinguishable odors as the conditioned stimuli in a Pavlovian conditioning task. For the first pair of odors, one of them was associated with a rewarding piece of chocolate (Positive Conditioned Stimulus; CS+) whereas its similar version was not (CS+ similar). For the second pair, neither odor was associated with the pleasant reward (Negative Conditioned Stimulus; CS- and CS-similar). If positive and negative stimuli possess similar reinforcing capabilities then the odor that acquired affective relevance by being associated with a reward should be better discriminated from the similar odor compared to the odors that have never been associated with the reward. During the triangular test, two different measures of odor discrimination were used. The first was an explicit and classical measure consisting in the accuracy of the behavioral choices (Laska and Teubner, 1999; Li et al., 2008), the second was an implicit measure consisting in the inspiration volume (Bensafi et al., 2003, 2007; Frank et al., 2003; Mainland and Sobel, 2006). The inspiration volume was used as an implicit measure for two reasons. First, it is modulated by the pleasantness of the odor (Bensafi et al., 2003; Frank et al., 2003); since in appetitive conditioning the value of the reward is transferred to the CS, the odor associated with the reward should become more pleasant (De Houwer et al., 2001) and consequently, the inspiration volume could be modulated. Second, the inspiration volume is not influenced by any conscious strategy (Bensafi et al., 2007). We predicted that the odor that had acquired affective relevance by being associated with the reward would be discriminated from its pair: (a) explicitly as reflected by a higher accuracy in the behavioral choices during the triangular test after conditioning and (b) implicitly as reflected by a bigger inspiration volume for the reward-associated odor than for its similar version.

### **MATERIALS AND METHODS**

### **PARTICIPANTS**

Eighteen participants who liked chocolate and were not dieting were recruited on the premises of the University of Geneva. They received 20 Swiss francs for their participation. Two participants were later excluded due to a technical problem during the physiological recordings. The 16 participants included in the analysis (5 males; 26 ± 3.22 years old) had no problems with odor perception and were not wearing any fragrance. Participants were asked to refrain from eating 4 h before the experimental session to increase their motivation to eat chocolate.

### **MATERIALS**

### **Stimuli**

Two different pairs of odors were used (Firmenich, SA, Geneva, Switzerland). Each pair was composed by an odor that was qualitatively coherent in a feeding context and its similar version. This latter was created by mixing the original odor with a small percentage of *Hedion®* (methyl dihydrojasmonate), a fresh jasmin like odor. More specifically, first pair was composed of a *lemon* odor (100%) and a *lemon-Hedion* blend odor (98% and 2% respectively), the second of a *strawberry* odor (100%) and a *strawberry-Hedion* blend odor (97% and 3% respectively). These odors were selected based on a previous pilot study (*N* = 20) that revealed that when participants were asked to distinguish an odor mixed with *Hedion* from its pure version, their performances were not significantly better than chance, *t*(19) = −0.85, *p =* 0.430. The pure *Hedion* was also used to control for its perceptibility. Odors were administered through a computer-controlled olfactometer with an air flow fixed at 1.5 L/min that delivered the olfactory stimulation rapidly, without thermal and tactile confounds (Pool et al., 2014) via a nasal cannula.

The rewarding chocolate consisted of a small 0.5 g piece of chocolate (dark or milk, according to the participants' preferences) to prevent that satiation processes modified the chocolate's rewarding value during the conditioning procedure (Small et al., 2001).

#### **Physiological recordings**

Respiration and electrodermal activities were recorded using the MP150 Biopac Systems (Santa Barbara, CA) with a 1000 Hz sampling rate. The respiratory activity was recorded through a 2.5 mm tube (interior diameter) positioned at the entrance of the participants' right nostril, on the nasal cannula used to deliver the odorants, and connected to a differential pressure transducer (TSD160A; ± 2.5 cm H2O sensitivity range) to continuously recorded variations in the nostril airflow. The signal was first low-pass filtered at 1 Hz and the duration of the inspiration was calculated as the length of the depression (in ms) between two consecutive crossing of the zero values after the stimulus onset. The integral of flow variations and its maximal value within this duration were also calculated for each trial.

Electrodermal activity was measured with Beckman Ag–AgCl electrodes (8-mm diameter active area) filled with a skin conductance paste (Biopac) attached to the palmar side of the middle phalanges of the second and third fingers of the participants' nondominant hand. Specific skin conductance responses (SCRs) were measured in microSiemens and analyzed offline (Bandpass filter: 0.05–5 Hz). They were scored as changes in conductance starting in the 1- to 5-s interval after the beginning of the stimulus. SCRs were square root transformed to normalize the data (Dawson et al., 2000).

#### **PROCEDURE**

Before the experiment started, physiological sensors were positioned on the participants. They also indicated whether they preferred dark (*N* = 9) or milk chocolate (*N* = 7) so that the reward used in the Pavlovian conditioning could be consequently adapted. Subsequently, they completed a control triangular test to measure their *Hedion* perception. They then completed the experimental procedure consisting in four triangular tests, two before and two after the Pavlovian conditioning. At the end of the experiment, they answered questions concerning the manipulation check. The entire experimental session took around 75 min.

#### **Pavlovian conditioning**

Each pair of odors was assigned to the Pavlovian role of the "CS+" or "CS−". For the CS+ pair, one of the two odors was associated with a rewarding taste of chocolate (CS+) whereas the other one was not (similar CS+). For the "CS−", neither of the two odors was associated with a reward (CS−; similar CS−). The assignment of the odors to the Pavlovian roles was counterbalanced across participants. For the sample included in the analysis, all odors were used as CS+ for at least 3 participants: the lemon (*N* = 5), the lemon-*Hedion* (*N* = 5), the strawberry (*N* = 3) and the strawberry-*Hedion* (*N* = 3).

There were 18 trials for each odor for a total of 72 trials. Every trial began with a 3 s countdown followed by an inspiration cue that request the participant to breath in evenly and trigger the release of the odor for 2.5 s. A gray patch then appeared and participants were asked to press as quickly as possible the "x" key to remove the patch and reveal a picture of chocolate for the CS+ and a red cross for all the other stimuli (i.e., the similar CS+, the CS− and the similar CS−). They were also explicitly asked to guess whether a particular odor could predict the rewarding chocolate delivery. When the chocolate picture was displayed on the screen, participants ate a small piece of chocolate and drank a sip of water (Prévost et al., 2012). To underline that the reward delivery depended on the odor and not on their action, participants were told that the key-pressing task was a measure of their sustained attention, independent of the odor-reward contingencies (Talmi et al., 2008). They were also informed that not responding during the 1-s interval after the gray patch onset would trigger the next screen anyway (Talmi et al., 2008). Every trial ended with an intertrial interval of 4–6 s for the non-rewarded trials and of 2–4 s for the rewarded trials (see **Figure 1**). The Pavlovian conditioning procedure lasted around 30 min.

After the conditioning task, participants took a small break (1–2 min) and then they evaluated on a visual analog scale presented on the screen, the subjective pleasantness (from "extremely unpleasant" to "extremely pleasant"), intensity (from "not perceived" to "extremely strong") and familiarity (from "not familiar at all" to "extremely familiar") of each odor and of the *Hedion* odor alone (e.g., Delplanque et al., 2008). The odor evaluation procedure lasted about 5 min.

#### **Triangular test**

We created a computerized version of the triangular discrimination test designed by Laska and Teubner (1999). Three pictures of

bottles associated with three different odors were presented on the screen. Participants were told that two of the odors were identical and one they had to identify was different.

Each trial began with a screen displaying the first bottle pointed by an arrow for 2–5 s, then a 3 s countdown began followed by an inspiration cue and the odor was delivered for 1 s. Afterwards, the bottle remained on the screen for 8–5 s (so that the inter-stimulus interval was 10 s in total). Then, the second and the third bottle appeared and the same procedure was repeated. Once the three odors had been smelled, participants were instructed to use the mouse to determine the bottle containing the odd odor by clicking on the corresponding picture of bottle. During the inter-trial interval participants were asked to wait for 15 s. All possible combinations of apparition of the three bottles (and odors; i.e., 6 in total) were presented to each and randomly across participants. Separate triangular tests were administered for each pair of odors (lemon and lemon-*Hedion* blend; strawberry and strawberry-*Hedion* blend) used in the conditioning. To accomplish a triangular test, participants took around 6 min.

At the beginning of the experiment, participants accomplished a control triangular test in which they were asked to discriminate *Hedion* from odorless air.

#### **Manipulation check**

Participants were asked to eat a piece of the chocolate that was used during the conditioning and to evaluate, on a visual analog scale presented on the screen, its subjective pleasantness (from "extremely unpleasant" to "extremely pleasant"), intensity (from "extremely weak" to "extremely strong") and familiarity (from "never tasted before this experiment" to "extremely familiar"). Subsequently, they answered two questions about chocolate: one about motivation (i.e., "On a scale from 1 to 10, how much would you say that you sometimes crave chocolate?") and the other about hedonic pleasure (i.e., "On a scale from 1 to 10, how much would you say that you like chocolate?"; see Pool et al., 2014). This last part of the experiment took around 5 min.

#### **RESULTS**

#### **DATA ANALYSIS**

For the analysis of the Pavlovian conditioning and the triangular tests we used a repeated measures analysis of variance (ANOVA) and planned contrasts according to the hypothesis that was tested. Effect sizes are measured as eta squared (η 2 ) for the repeated measure ANOVA and as Cohen's d (*d*) for the planned contrasts.

#### **MANIPULATION CHECK**

Participants evaluated the piece of chocolate's taste used as reward as being pleasant (*M* = 93.16, *SD =* 8.77 out of 100); intense (*M* = 76.01; *SD =* 19.11 out of 100); and familiar (*M* = 94.53, *SD =* 6.41 out of 100), showing that the chocolate used as reward was indeed perceived as a pleasurable experience. Moreover, they reported a mean of 9.41 (*SD* = 1.01) out of 10 for the likeability item and a mean of 8.54 (*SD* = 1.29) out of 10 for the craving item on the questions about chocolate, showing that they associated chocolate with hedonic pleasure and motivation (Berridge and Robinson, 2003).

#### **HEDION PERCEPTIBILITY**

In the control triangular test, participants discriminated the *Hedion* odor from the odorless air with an accuracy of 71.87% (±23.36%) which was significantly better than the chance level of 33%, *t*(15) = 7.59, *p* < 0.001, *d* = 2.6. For the odor evaluation, participants reported a mean of 44.59 (±24.46; out of 100) for the intensity scale, a mean of 59.25 (±16.81; out of 100) for the pleasantness scale and a mean of 53.87 (±19.72; out of 100) for the familiarity scale.

#### **PAVLOVIAN CONDITIONING**

To assess the success of the Pavlovian conditioning, we used the reaction time of the key-pressing task and the SCRs as implicit indexes and the likeability rating of the odors used as CSs after conditioning as an explicit index.

The reaction times of three participants could not be recorded due to technical problems. All responses that were more than three standard deviations from the mean (<1% of the trials), or absent (<3% of the trials) were removed. A planned contrast revealed a specific effect of conditioning: participants were faster when the CS+ odor (*M* = 384.08, *SD* = 25.77) was delivered than when the similar CS+, the CS− and the similar CS− odors were delivered (*M* = 405.66, *SD* = 21.53), *t*(12) = 2.44, *p* = 0.031, *d* = 0.30. To test whether this difference could be accounted for in terms of reward learning, we conducted two other planned contrasts that revealed that this difference was significant in the second part of the conditioning, *t*(12) = 2.62, *p* = 0.022, *d* = 0.32, but not in the first, *t*(12) = 0.16, *p* = 0.860.

The likeability analysis revealed a general effect of conditioning: participants did not evaluate the CS+ odor as significantly more likeable than the similar CS+, the CS− and the similar CS−, *t*(15) = 1.19, *p* = 0.252; but they globally evaluated the CS+ and the similar CS+ as being more likeable than the CS− and the similar CS−, *t*(15) = 2.45, *p* = 0.026, *d* = 0.47 (see **Figure 2A**).

Autonomic responses recorded during the same phase were specifically affected by the conditioning: a planned contrast revealed that the amplitude of the SCR was bigger for the CS+ than for the similar CS+, the CS− and the similar CS−, *t*(12) = 2.41, *p* = 0.032, *d* = 0.21<sup>1</sup> (see **Figure 2B**).

#### **TRIANGULAR TEST**

Similarly to the conditioning, results on the triangular test showed specific effects only for the implicit measure, but not for the explicit one.

To test our hypothesis for the explicit measure, we applied a 2 (Pair: CS+ or CS− pair) × 2 (Session: pre- or post-conditioning)

<sup>1</sup>Three participants were excluded from the analysis for missing data (listwise analysis). The maximal peak of their skin conductance appeared before the 1 s delay after the stimulus on set on at least one of the four conditions (e.g., CS+, CS+ similar, CS−, CS− similar). Therefore we could not consider these responses as being triggered by the stimulus. To guarantee the comparability of the results of the skin conductance and the auto-reports of the perceived pleasantness, we ran supplementary analyses. An analysis including all participants with mean substitutions for missing data gave a similar result *t*(15) = 2.42, *p* = 0.028. Moreover, if we removed the three missing participants from the analysis of the auto reports of pleasantness, the main results did not change *t*(12) = 2.75, *p* = 0.017.

repeated measures ANOVA on the accuracy of behavioral choices of the triangular tests. The analysis did not reveal any significant effect (all *p<sup>s</sup>* > 0.05), *t*-test showed that the performance of the participants were not significantly different from the chance level of 33%, neither before nor after conditioning (all *p<sup>s</sup>* > 0.05; see **Figure 3A**).

To test our hypothesis for the implicit measures, we applied a 2 (Odor: CS+ or similar CS+) × 2 (Session: pre- or post-conditioning) repeated measures ANOVA on the integral of inspiration flow during the triangular tests. The analysis revealed a significant interaction, *F*(15,1) = 5.41, *p* = 0.034, η <sup>2</sup> = 0.26 showing that after conditioning participants inhaled more for the CS+ than for the similar CS+, *t*(15) = 2.49, *p* = 0.027, *d* = 0.28, this difference was not present before conditioning, *t*(15) = 0.69, *p* = 0.490. The same analysis on the CS− and the similar CS− did not reveal any significant effect (all *p<sup>s</sup>* > 0.05; See **Figure 3B**).

triangular test in which participants were asked to discriminate two similar odors. In the positive conditioned pair (CS+), one odor was associated with the reward (CS+) whereas the other was not (similar CS+). In the negative

associated with the reward. **(B)** Means of the integral of the volume of the inspiration during the triangular tests. Error bars (±1 SEM) are adapted for within design (Cousineau, 2005).

Since the ANOVAs revealed that conditioning increased the inspiration flow for the CS+ compared with similar CS+ but not for CS− and its similar version, we used planned comparisons to compare the inspiration across the fours odors after conditioning. Participants inhaled more when the CS+ was perceived than when the similar CS+, the CS− and the similar CS− were perceived *t*(15) = 2.23, *p* = 0.041, *d* = 0.25; <sup>2</sup> This difference was not present before conditioning (*p* = 0.511).

The same analysis conducted on the SCR amplitude did not reveal any significant effect (all *p* > 0.05).

### **DISCUSSION**

The objective of this experiment was twofold: first to investigate whether perceptual discrimination of odors could be enhanced through appetitive emotional learning, and second, the degree of selectivity of this enhancement at an implicit and explicit level. To do so, an olfactory stimulus was associated with a sweet reward (i.e., chocolate) and we tested the capacity of the participants to discriminate it from another perceptually similar but nonconditioned stimulus. Before conditioning, the conditioned odor and its similar version were not discriminated in the triangular test, neither on the basis of accuracy of choices, nor for strength of the inspiration. After conditioning, we did not find any statistical evidence showing that participants were able to discriminate the odors in the triangular test, nonetheless their inspirations were more intense for the reward-associated odor compared to its similar non-conditioned version. A similar pattern was shown for the conditioning indexes: participants globally evaluated the reward-associated odor and its similar version more pleasurable compared with the evidently perceptually different nonconditioned odors, without differentiating the reward-associated odor from its similar non-conditioned version. However, participants' responses to the reward-associated odor were differentiated from its similar non-conditioned version and the evidently perceptually different non-conditioned odors based on the implicit indexes of conditioning (i.e., the SCR's amplitude after conditioning and the reaction times during conditioning). In sum, the explicit behavioral measure demonstrated that the conditioning effect of the CS+ is generalized to its similar version, whereas the physiological and implicit behavioral measure showed a specific effect of conditioning which influences selectively the CS+ but not its similar version.

Consistent with Li et al. (2008), our findings showed that emotional learning can enhance the discrimination between similar stimuli at both behavioral (reaction time) and physiological levels (inspiration and skin conductance); however by contrast Li et al. (2008) found an increase of the accuracy of the behavioral choices in the triangular test induced by aversive conditioning, while participants' explicit discrimination was not increased in our study after the appetitive conditioning. Many factors could explain this absence of effect (e.g., number of trials, number of participants, odor choice). Nonetheless, such a dissociation between the explicit and the implicit measures is congruent with several findings in the literature.

First, the dissociation between the auto-reported evaluation of the odors' pleasantness and the volume of the inspiration during the odor perception supports the idea that the inspiration volume is modulated by the affective properties of the odors (Bensafi et al., 2003; Frank et al., 2003; Mainland and Sobel, 2006), independently of the conscious processing (Bensafi et al., 2007). The way in which the odor is inspired can in turn influence different aspect of the odor perception such as its intensity or its identity (Mainland and Sobel, 2006). We did not find evidence showing that the increased inspiration volume improved the odor identity discrimination. However, this is congruent with the literature showing that the improvement of the odor identity is not proportional to the inspiration volume, but rather depends on the interaction between the intrinsic properties of the odor (e.g., sorption rate) and other parameters of the inspiration (e.g., velocity, Sobel et al., 1999; Mainland and Sobel, 2006).

Second, the fact that even though participants did not seem to be able to explicitly discriminate the reward-associated odor, they responded with more motivated reactions when they perceived is consistent with studies showing that it is possible to elicit affective reactions even when participants are not able to consciously report them (Winkielman and Berridge, 2004; Winkielman et al., 2005). Several theories (Murphy and Zajonc, 1993; Winkielman and Berridge, 2004; Winkielman and Schooler, 2011) postulate that the processing of affective stimuli can involve differential physiological responses and subjective feelings accessible through verbal reports. Affective processing may occur rapidly and automatically without accessing conscious processing, thereby triggering physiological responses without a parallel conscious and verbally reportable experience (Murphy and Zajonc, 1993; Silvestrini and Gendolla, 2011; Bornemann et al., 2012).

There might by an adaptive value in these differential effects of conditioning on the implicit and explicit levels. The generalization of the conditioning effects to similar stimuli at an explicit level, might allow the organism to flexibly stay vigilant to all potential indicators of reward, whereas the specificity of the conditioning effects at an implicit level might allow the organism to economize resources by reacting with expensive emotional reactions, even when positive, only when it is strictly necessary.

Finally, the increase of physiological discrimination of similar odors by appetitive emotional learning using positive rewarding stimuli is in line with studies showing an increased efficiency of sensorial processing for positive emotional stimuli (e.g., Brosch et al., 2008; Hickey et al., 2010; Hietanen and Nummenmaa, 2011) and with the prediction of appraisal theories (Frijda, 1988; Sander et al., 2005). Indeed, appraisal theories propose that both negative and positive events influence the perceptual system because they are underlain by an affective relevance mechanism rather than distinct valence-specific mechanisms. Whereas neutral stimuli associated with a threat acquired affective relevance because they increase the chances of avoiding a danger, neutral stimuli associated with a reward acquired affective relevance because they increase the chances of satisfying a goal.

<sup>2</sup>The inspiration before and after for the CS+ was not tested, because sniffing can vary throughout time for several factors (e.g., habituation, see Wirth et al., 1998; or familiarity, see Delplanque et al., 2008). The differential measure (CS+; CS+ similar) is the most appropriate and precise one to test our hypothesis, because it is based on a subtraction that is free of biases related to the time course.

#### **CONCLUSIONS**

The findings of this experiment suggest that emotional learning using positive rewarding stimuli can increase implicit discrimination of perceptually similar odors. The ability to learn to precisely identify stimuli that are affectively relevant not only increases the chances to indeed react to emotional events, but it also reduces the chances of reacting with expensive emotional reactions when it is not necessary. For instance, if two persons are watching their children and chatting, they will intensively react specifically to the voice of their children calling them, and not to all the voices of other children calling their parents. Furthermore, our findings suggest reward-associated odors can trigger affective reactions, even when it seems that they cannot be consciously discriminated. This deserves to be further investigated since it opens an interesting window to the influence that an odor can have on our experience.

### **ACKNOWLEDGMENTS**

This research was supported by the National Center of Competence in Research (NCCR) for the Affective Sciences, financed by a grant from the Swiss National Science Foundation (51NF40- 104897), hosted by the University of Geneva, and was also supported by a research grant from Firmenich, SA, to David Sander and Patrik Vuilleumier. The authors thank Nadine Gaudreau and all the members of the Perception and Bioresponses Department of the Research and Development Division of Firmenich, SA, for their precious advice and technical competence.

#### **REFERENCES**


Notebaert, L., Crombez, G., Van Damme, S., De Houwer, J., and Theeuwes, J. (2011). Signals of threat do not capture, but prioritize, attention: a conditioning approach. *Emotion* 11, 81–89. doi: 10.1037/a0021286


**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 January 2014; accepted: 16 April 2014; published online: 06 May 2014*. *Citation: Pool E, Delplanque S, Porcherot C, Jenkins T, Cayeux I and Sander D (2014) Sweet reward increases implicit discrimination of similar odors. Front. Behav. Neurosci. 8:158. doi: 10.3389/fnbeh.2014.00158*

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

*Copyright © 2014 Pool, Delplanque, Porcherot, Jenkins, Cayeux and Sander. 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*.

## Olfactory preference conditioning changes the reward value of reinforced and non-reinforced odors

#### *Nicolas Torquet <sup>1</sup> \*, Pascaline Aimé 2, Belkacem Messaoudi 3, Samuel Garcia3, Elodie Ey 4, Rémi Gervais 3, A. Karyn Julliard3 and Nadine Ravel <sup>3</sup> \**

*<sup>1</sup> Neurophysiology and Behavior Laboratory, Neuroscience Paris-Seine – IBPS, CNRS UMR 8246 – UPMC UM CR18 – INSERM U1130, Paris, France*

*<sup>2</sup> Department of Pathology and Cell Biology, Columbia University Medical Center, New York, USA*

*<sup>3</sup> Equipe Olfaction: du Codage à la Mémoire, Centre de Recherche en Neurosciences de Lyon, CNRS UMR 5292 - INSERM U1028, Université Lyon 1, Lyon, France*

*<sup>4</sup> Human Genetics and Cognitive Functions, Neuroscience Department, Institut Pasteur, CNRS UMR 3571 Genes, Synapses and Cognition, Paris, France*

#### *Edited by:*

*Regina M. Sullivan, Nathan Kline Institute, USA*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Anthony Sclafani, Brooklyn College of The City University of New York, USA*

#### *\*Correspondence:*

*Nicolas Torquet, Neurophysiology and Behavior Laboratory, Neuroscience Paris-Seine - IBPS, CNRS UMR 8246 - UPMC UM CR18 - INSERM U1130, Université P. et M. Curie, Boite 16, 9 quai St. Bernard, 75005 Paris, France e-mail: nicolas.torquet@upmc.fr; Nadine Ravel, Centre de Recherche en Neurosciences de Lyon, CNRS UMR 5292 - INSERM U1028 - Université Lyon1, Equipe "Olfaction : Du codage á la mémoire" Site de Gerland - 50 Avenue Tony Garnier, F-69366 Lyon Cedex 07, France e-mail: nravel@olfac.univ-lyon1.fr*

**INTRODUCTION**

Rodents are macrosmatic animals with olfactory abilities allowing them to detect and discriminate a wide range of odors and accomplish complex olfactory-mediated cognitive tasks (Slotnick, 2001). By relying on olfaction, rodents can learn to discriminate distant appetitive and edible items from aversive and potentially hazardous items. Although some odors or tastes can generate innate aversion as well as preference behaviors (Apfelbach et al., 2005; Kobayakawa et al., 2007; Ventura and Worobey, 2013), most food choices result from an associative process between food sensory characteristics and the post-ingestive consequences. This process most often leads to switch the initial neutral reward value of a specific food toward either a preference or an aversion (Slotnick, 2001; Gautam and Verhagen, 2010).

When consuming, we are exposed to two types of chemosensory stimuli: first, the odor, when items are at distance, and then the flavor (a combination of retronasally transmitted olfactory stimulation and taste) when items are in the mouth (Pierce and Halpern, 1996). These odor-taste interactions have been described as taste-odor synesthesia resulting from pairing of the two chemosensory modalities via oral and retronasal stimulation

Olfaction is determinant for the organization of rodent behavior. In a feeding context, rodents must quickly discriminate whether a nutrient can be ingested or whether it represents a potential danger to them. To understand the learning processes that support food choice, aversive olfactory learning and flavor appetitive learning have been extensively studied. In contrast, little is currently known about olfactory appetitive learning and its mechanisms. We designed a new paradigm to study conditioned olfactory preference in rats. After 8 days of exposure to a pair of odors (one paired with sucrose and the other with water), rats developed a strong and stable preference for the odor associated with the sucrose solution. A series of experiments were conducted to further analyze changes in reward value induced by this paradigm for both stimuli. As expected, the reward value of the reinforced odor changed positively. Interestingly, the reward value of the alternative odor decreased. This devaluation had an impact on further odor comparisons that the animal had to make. This result suggests that appetitive conditioning involving a comparison between two odors not only leads to a change in the reward value of the reinforced odor, but also induces a stable devaluation of the non-reinforced stimulus.

**Keywords: olfaction, odor, behavior, rat, olfactory preference, conditioning, discrimination**

during food ingestion (Verhagen and Engelen, 2006). Indeed, in humans, food-related odorants are often reported and described as tastes (vanilla for example reported as sweet) and a novel odor can acquire a taste even after a single pairing (for a review see Stevenson and Tomiczek, 2007). From experimental work in rodent, it has become clear that taste-like qualities of odors are learned and are not innate (Gautam and Verhagen, 2010).

A large number of studies propose an unraveling of the contribution of each type of chemosensory stimulus in the learning process, most of them using aversive conditioning paradigms (Palmerino et al., 1980; Batsell et al., 1999; Batsell and Blankenship, 2002; Miranda, 2012). In such protocols, the presentation of a solution characterized by a specific odor or taste can be associated with a gastric malaise induced by an intraperitoneal injection of lithium chloride and leads to a strong further avoidance of this solution (Garcia et al., 1968). Interestingly, whereas a pure taste aversion could be induced, even when the malaise is delayed from the ingestion, odor aversion alone is more difficult to obtain and less resistant to an increase of ingestion-malaise delay (Palmerino et al., 1980). However, when the odor is combined with a specific taste during the initial experience (taste potentiated odor aversion, TPOA) or ingested without any gustatory impact, a very robust aversive behavior is obtained even when the odor is presented alone during the test (Rusiniak et al., 1979; Palmerino et al., 1980; Miranda, 2012). As a consequence, further animal models using conditioned flavor aversion or preference were developed. These models showed that an odor present in a solution or an ingested aliment could become aversive (Slotnick et al., 1997; Chapuis et al., 2007) or attractive exactly as observed in taste models (Holder, 1991; Boakes et al., 2007).

The respective role of the individual components of a flavor, odor and taste, have been less explored regarding appetitive learning. To our knowledge, only two paradigms have clearly established a strict Conditioned Olfactory Preference (COP; Holder, 1991; Lucas and Sclafani, 1995). Compared to previous work in which odorants were consumed by the animal, in the Holder's paradigm, the odors used (almond or peppermint) were presented during 4 consecutive days in close proximity with the ingested solution (sucrose or saccharin) but the rats never made gustatory contact with them. A two-bottle test performed 24 h later revealed that animals preferred the odor previously paired with sucrose over saccharin, suggesting both a calorie-mediated and an appetitive taste-mediated preference (Holder, 1991). In this paradigm however, as each odor is paired with a sugarsweetened solution, it is difficult to interpret the real change of the reward value for the odor paired with non-nutritive saccharin. Indeed previous studies have demonstrated that the animals preferred odor paired with the nutritive diet (Baker and Booth, 1989; Holder, 1991) and strong odor preferences could be obtained for odors paired with the post ingestive actions of nutrients without the presence of added taste cues (Lucas and Sclafani, 1995). In the present study, we therefore propose to investigate the influence of appetitive learning on the reward value of both the odor used as the conditioned stimulus (CS+), paired with sucrose, and a second odor (CS−) that is paired with the neutral solution, pure water. Additionally, we also investigate whether COP could be obtained from an initially preferred odor as well as from a less preferred one.

Compared to previous studies (Holder, 1991; Lucas and Sclafani, 1995), the present work provides new and additional informations on Conditioned Olfactory Preference. In experiment 1, we first evaluated the spontaneous behavioral preferences of several odors presented in pairs. We then induced a conditioned olfactory preference by coupling one of the odors with a sucrose solution, and a second one with pure water, which remains a positive stimulus for water-deprived animals but less appetitive. The protocol efficacy was measured immediately and 1 month later by testing how the conditioning procedure impacts the initial observed preference for the conditioning pair of odors. A series of additional experiments was performed to evaluate how the conditioning procedure also affected the spontaneous preference for the two tested odors compared to others not involved in the conditioning paradigm (Experiments 2 and 3). This allowed us to demonstrate that, as expected, the conditioned odor was reinforced (increased positive valence). Interestingly, we also reported that the reward value of the non-conditioned odor was depreciated compared to any nontraining odor.

### **MATERIALS AND METHODS**

### **ANIMALS**

Experiments were carried out in accordance with the European Community Council Directive of November 24th, 1986 (86/609/EEC) for the care and use of laboratory animals. The experimental protocols were approved by the Lyon1 University Ethics Committee (Direction of Veterinary Service #693870202).

A total of forty male Wistar rats (275–300 g, Charles River, L'arbresle, France) were involved in this study. They were housed individually in Plexiglas chambers at constant temperature and relative humidity (22 ± 0.5◦C and 50 ± 5%) and under a 12-h light: 12-h dark cycle (light on at 6.00 AM) at least 1 week before the beginning of the experiments.

A total of forty rats were included in the 3 experiments. Sixteen of them participated in experiments 1 and 2. In experiment 1, eight rats were assigned to either group 1 or group 2. In experiment 2, animals from both groups were equally distributed between group 3 and group 4. Twenty-four naïve animals participated in experiment 3.

#### **ODORS**

Odors were purchased from Sigma-Aldrich (France). According to their chemical properties, they were diluted either in mineral oil or in water. The dilution was also adjusted as a function of their vapor pressure to be judged as moderate and equal in intensity by the experimentalist. In the following, + and − symbols with brackets [] refer to the enantiomer structure of the odors while the + and − symbols outside brackets refers to the association made during conditioning with sucrose (+suc) or water (−) respectively. Geraniol, eugenol, and two enantiomers of carvone (carvone[−] and carvone[+]) were diluted to 10% in mineral oil, limonene[−] was diluted to 5% in mineral oil and iso-amylacetate was used at a dilution of 10% in water. Geraniol and eugenol were used for experiment 1, carvone[−], carvone[+] and limonene[−] for experiment 2 and the two enantiomers of carvone, limonene [−] and iso-amylacetate in experiment 3. 150µL of each odor solution was distributed on a 3-cm cotton disc (3M™ T156 Oil Sorbent Sheets). The disc was placed into the 3.2-cm cap of a tube, face turned toward the cage and secured in place by a thin grid of metal. The metal grid, the disc and the 3-cm cap were punched with a 16 mm hole allowing them to be placed around the spout of the bottle of drinking solution. Therefore, the rats could never touch the disc, as described in the experiments of Holder (1991). Importantly, such a device also precludes the bottle solution to be polluted by the odor.

#### **EXPERIMENTAL SET UP (FIGURE 1)**

Behavioral tests were conducted in parallel for four rats in individual Plexiglas operant chambers (330 × 210 × 180 mm). The chambers were set side by side so that the animals could not see each other. Two plastic tubes were mounted on the opposite sides of the flat ceiling of each chamber. These tubes (made from 15 ml polypropylene conical centrifuge tubes, Falcon, France) were cut and fire polished to build a 0.4 mm spout which protruded approximately 5 mm into the chamber, allowing the rats to drink from the spout with ease by rising up. The amount of liquid consumed by the rat from each tube was measured by a

device called licking box. This system has already been used to measure odor detection in different states of satiety and described in details elsewhere (Aimé et al., 2007). Each tube was connected to a custom-made capacitance circuit which allowed detecting, visualizing and recording of individual licks at each bottle across time during the experimental sessions. Generated files were then transferred into a database developed in our laboratory to be analyzed offline. In parallel, the global consumption of each rat from each bottle was also measured to check the accuracy of the licking system.

#### **GENERAL PROTOCOL**

To motivate their drinking behavior, rats had been acclimated to a restricted access to water during 1 week before the beginning of habituation. During the overall course of the study, their daily access to water was restricted to 15 min in the experimental cage, supplemented by a 20 min period of free access each afternoon at 6:00 PM in their home cages.

#### *Habituation*

During 5 days, rats were placed in the experimental cages with two bottles of water for a 15 min session. This phase was devoted to train rats to drink from the device, and to avoid stress during the conditioning.

#### *Pretest*

To assess a possible spontaneous aversion or preference, each pair of odors to be used in the following experiment was presented to the animal. During this phase, for each 15 min session, two bottles of tap water were each associated with one odor of the pair and simultaneously presented. The same pair was tested on two consecutive days switching the location of the odors to rule out place preference.

#### *Conditioning*

During the 8 days of conditioning, rats were exposed during 15 min to two odors paired with two bottles containing different solutions. Odor A was paired with a bottle containing a solution of 3.4% sucrose (A+suc), while odor B was paired with a bottle of tap water (B−). The preference for specific sugar concentration varies according to the level of water deprivation imposed to the animals. We used a concentration of 3.4% of sucrose as validated by Holder (1991). In our water restriction condition, this concentration was still preferred by the rats over water. Throughout the study, the position of each bottle was switched each day to rule out place preference.

#### *Final test*

Two bottles containing tap water associated with odor A or B were presented for 15 min on two consecutive daily sessions switching their position. Results are presented as the mean intake over 2 days allowing for position switch.

The same protocol was used for the three experiments, with a variation in conditioning odors according to the experiment (see **Table 1** for the details of each group). When additional tests were performed (experiments 2 and 3), two additional conditioning sessions were conducted between each test to avoid a possible extinction of preference induced by the repetition of odor presentation without sucrose-pairing.

#### **STATISTICAL ANALYSIS**

The coefficient of correlation between the number of licks and the global water consumption measured during experiment 1 indicated a good reliability of our licking system (*r* = 0*.*985). As a consequence, we chose the number of licks as the study variable. To normalize data, we expressed consumption from each bottle as a ratio of the number of licks for this bottle on the sum of the number of licks for the two bottles. This variable was then compared between subjects according to the different experimental groups and for a given subject as a function of experimental condition (pretest, test). Within group differences were then analyzed using the non-parametric Wilcoxon test (WT). The changes in bottle position were never found to induce a significant difference in liquid consumption, we therefore pooled the average number of licks on each bottle at the two positions.

#### **Table 1 | Summary of experiments.**


*Pretest, Odors presented associated with water; Conditioning,* +*suc, odor paired with sucrose (CS*+*);* −*, odor paired with water (CS*−*); Tests,* +*, odor previously used as a CS*+*;* −*, odor previously used as a CS*−*; no sign, this odor has not been included in the conditioning.*

### **RESULTS**

#### **EXPERIMENT 1**

#### *Specific methods*

Group 1 (*n* = 8) was conditioned to geraniol [i.e., geraniol was paired with sucrose (G+suc) and eugenol with water (E−)] while group 2 (*n* = 8) was conditioned to eugenol [i.e., eugenol was paired with sucrose (E+suc) and geraniol with water (G−)]. Two tests were performed to estimate olfactory preference: one immediately after the end of the conditioning period and the other, 1 month later to investigate the long-term retention of conditioning (see **Table 1**).

#### *Results (Figure 2)*

Before conditioning, no significant preference between geraniol and eugenol was detected in any group (WT: group 1, *p* = 0*.*326; group 2, *p* = 0*.*552). However, as soon as the second odor-sucrose association was run, rats exhibited a clear preference for the odor paired with the sucrose solution. We observed that the odor was progressively used to rapidly locate the sucrose solution. After conditioning, both groups showed a preference for the bottle associated with the odor previously paired with sucrose (WT test: group 1, *n* = 8, *p* = 0*.*002; group 2, *n* = 8, *p* = 0*.*011, see **Figure 2**). When rats from both groups were tested again 1 month later, their preference for the odor previously paired with sucrose was still significant (WT: group1, *n* = 8 and group 2, *n* = 8, *p* = 0*.*001).

#### *Discussion*

Experiment 1 confirmed the efficiency of our conditioning protocol. Although no spontaneous preference was observed for geraniol or eugenol, as expected, in both groups, a clear preference for the sucrose-paired odor developed after 8 days of conditioning. This olfactory preference was independent of the odor paired with sucrose since the two groups had a similar learning curve. Interestingly, this acquired preference was stable since, even 1 month after the last odor-sucrose exposure; the rats still preferred the odor previously paired with sucrose.

#### **EXPERIMENT 2**

#### *Specific methods*

Three pretests were conducted before conditioning to assess possible spontaneous preference. For that purpose, odors were presented in pairs both associated with water, on separate sessions and in the following order: carvone[+] (C[+]) vs. carvone[−] (C[−]), carvone[+] vs. limonene[−] (L[−]).

These pretests were followed by an 8-day period of conditioning. During this phase, carvone [−] was associated with sucrose (C[−]+suc) L[−] with water (L[−]−). After conditioning, we first presented C[−] vs. L[−] both associated with tap water to confirm the acquisition of odor preference. We then separated the sixteen rats into two groups (3 and 4) mixing animals from the two groups previously used (group 1 and group 2). In group 3, we tested C[+] vs. L[−] and then C[−] vs. C[+]. In group 4, we reversed the test order and first tested C[−] vs. C[+], and then C[+] vs. L[−] (see **Table 1** for a summary of all performed tests).

#### *Results (Figure 3)*

Regarding the pretests performed before conditioning, no significant difference of intake between C[−] and C[+] (*n* = 16, WT, *p* = 0*.*940), C[+] and L[−] (*n* = 16, WT, *p* = 0*.*537) and C[−] and L[−] (*n* = 16, WT, *p* = 0*.*640) emerged from the pretests.

After pairing sucrose to C[−], as expected, rats drank more from the bottle associated with this odor (*n* = 16, WT, *p <* 0*.*001) than L[−] even in the absence of sucrose. In the following tests, rats from group 3 preferred to drink from the bottle associated with C[+] rather than from the one associated with L[−] (*n* = 8, WT, *p <* 0*.*001). However, when C[+] was simultaneously presented with C[−] the rats exhibited a preference for the bottle odorized with the sucrose-paired odorant C[−] (*n* = 8, WT *p* = 0*.*009).

Rats from group 4 submitted to the same tests in a reverse order, drank more from the bottle associated with C[−] than from the one associated with C[+] (*n* = 8, WT *p* = 0*.*011). Subsequently, they also preferred C[+] to L[−] (*n* = 8, WT, *p* = 0*.*049).

#### *Discussion*

This second experiment confirmed the efficacy of the protocol to induce an odor preference with another pair of stimuli. This experiment also addressed the question of specificity for odor preference acquisition. This was assessed by adjusting carvone enantiomers discriminability. Before conditioning, both groups of rats exhibited no preference between these two odors. This result could be interpreted in two ways: either the animal is not able to discriminate these two odorant molecules or it is indeed able to discriminate them but has no preference. After conditioning, both groups exhibited a preference for the enantiomer previously paired with sucrose (C[−]) when simultaneously presented with L[−] or with C[+]. When a preference for C[+] to L[−] was observed, one could interpretate that the rats transferred the positive value acquired by one enantiomer to the other because they are perceived as similar. However, after conditioning, both groups exhibited a preference for C[−] vs. C[+], implying that the rats have the capacity to discriminate among the two carvone enantiomers. This change, compared to the pretest situation, could be interpreted either as a learning-induced improvement of enantiomer discrimination as suggested by others (Escanilla et al., 2008) and/or by a learning-induced change in motivation for choosing C[−]. A further hypothesis that fits with the previous two ideas, is that rats drinking more from the C[+] bottle when simultaneously presented with L[−], could be due to a generalized sugar association to "carvons" or simply because they avoided the odor that has never been paired with sucrose. This issue was addressed in experiment 3.

#### **EXPERIMENT 3**

#### *Specific methods*

This experiment was carried out on a group of 24 rats. Four pretests were performed to evaluate spontaneous preference for each odor presented by pairs before conditioning in the following order: carvone[−] (C[−]) vs. geraniol (G) for all rats (*n* = 24); limonene[−] (L[−]) vs. iso-amylacetate (I) for group 1 (*n* = 12) and eugenol (E) vs. carvone[+] (C[+]) for group 2 (*n* = 12); carvone[−] vs. limonene[−] for all rats (*n* = 24). The experiment included a conditioning phase. During this 8 day period, carvone[−] was associated with sucrose (C[−]+suc), and limonene[−] with water (L[−]). The acquired preference for C[−] vs. L[−] was then tested. On following sessions, group 1 was tested with L[−] vs. I and group 2 with C[+] vs. E. Finally, all rats were tested with C[−] vs. G (see **Table 1**).

#### *Results (Figure 4)*

Before conditioning C[−] and L[−] were equally approached (*n* = 24, WT, *p* = 0*.*082). Following conditioning, both groups acquired a clear preference for C[−] (*n* = 24, WT, *p <* 0*.*001).

During the pretest, rats from group 1 exhibited a spontaneous preference for L[−] compared to I (*n* = 12, WT, *p* = 0*.*009), which was abolished during conditioning (*n* = 12, WT, *p* = 0*.*909). Rats from group 2 showed a tendency to prefer E over C[+] during the pretest but this failed to reach significance (*n* = 12, WT, *p* = 0*.*076), which was not modified after conditioning (*n* = 12, WT, *p* = 0*.*056). Pre-tests revealed a significant spontaneous preference for G vs. C[−] (*n* = 24, WT, *p <* 0*.*001). Conditioning reversed this preference (*n* = 24, WT, *p <* 0*.*001).

#### *Discussion*

The aim of experiment 3 was to elucidate the nature of change in the reward value induced by an odor preference conditioning. Thus, after preference acquisition for C[−] compared to L[−], each odor of this pair was presented with a novel one to evaluate the new preferences.

Before learning, animals from group 1 preferred to drink from the bottle associated with L[−] when this odor was simultaneously presented with I. This preference disappeared after conditioning, which led us to the conclusion that, during preference conditioning, rats also learnt that L[−] had never been associated with sucrose. As a consequence, its natural reward value decreased.

Moreover, results obtained from group 2 suggested that learning-induced changes in the reward value are specific to odorants used during the conditioning phase. Indeed, pairing C[−] with sucrose had no effect on how C[+] was perceived compared to E. We therefore interpret that in experiment 2, the generalization from C[−] to C[+] might be the consequence of avoidance of L[−] that has never been paired with sucrose.

Rats spontaneously preferred G to C[−], but after conditioning, their preference was reversed. This clearly indicates that learning increases carvone reward value. Putting together this result with those of the tests with C[+] and E, we could conclude that the present olfactory preference conditioning procedure selectively increased C[−] reward value and, as a consequence, devaluated L[−]. This experiment also confirmed the ability of rats to distinguish these two enantiomers, even when they had never been presented simultaneously for direct olfactory comparison.

#### **GENERAL DISCUSSION AND CONCLUSIONS**

These series of experiments confirm the possibility to induce a strong and reliable conditioning olfactory preference by associating the consumption of a sucrose solution and the close delivery of an odor without any ingestion of it. Indeed, the odor concentrated around the water bottle's spout was perceived as a characteristic of the solution contained inside the bottle. Experiments 1 and 2 assessed, for the first time, a display of olfactory preference for two different odor pairs (eugenol, geraniol and carvone[−], limonene[−]). Conditioning led to a specific preference for the odor associated with a sucrose solution, and as shown in experiment 1, this preference was maintained for at least 1 month. Experiment 2 and 3 addressed the question of specificity for odor preference acquisition and elucidated the nature of change in the reward value induced by such an odor preference conditioning.

As mentioned in the introduction, few studies have explored the individual contribution of odor and taste in conditioned food preferences. In the experiment performed by Holder (1991), each odor was paired with a different sweet taste solution (either sucrose or saccharin) and the animals were continuously exposed during 4 days in their home cage and tested 48 h later, after a period of access to plain water. In our protocol, only one of the two odors was associated with a sugar solution while the other was paired with plain water. This presentation was performed in an experimental cage during 8 consecutive daily sessions of 15 min. Moreover, the animals were also allowed to drink water in their home cage for 20 min each day. The advantage of using such an exposure mode was to maximally control the experimental environment and to avoid any diffusion of the odorant and subsequent contextual learning. Despite these differences, in both experiments, a clear preference was obtained for the odor paired with sucrose compared to saccharin in Holder's study and water in our case. Sucrose solution has very often been used for appetitive conditioning (Holder, 1991; Boakes et al., 2007) since its sugar taste is highly and spontaneously preferred (Shepherd, 2006), with a non-negligible energy supply (Myers

and Sclafani, 2001). Other protocols that have used saccharin to avoid energy supply (Baker and Booth, 1989; Myers and Sclafani, 2001); demonstrated that conditioned odor preferences were reinforced by the postingestive effects of caloric substances (Myers and Sclafani, 2001). Nevertheless this reinforcement has been suggested to be dependent on the feeding state of animals (Harris et al., 2000). We cannot rule out the possibility that the metabolic effect of sucrose has contributed to the success and stability of our conditioning procedure. However, the main objective of our study was not to address this specific issue but rather to increase as much as possible the rewarding value of the odor paired with sucrose.

In experiments 2 and 3, the strategy of the animal's choice was investigated. In experiment 2, the reinforced odor (C[−]+) was replaced by its enantiomer (C[+]) during the test and presented simultaneously with the non-reinforced odor (L[−]−). Rats seemed to consider the new odor as the previous reinforced one. This result could lead to two distinct interpretations, either the rats were unable to discriminate between the two enantiomers of carvone, or they were generalizing the olfactory preference to closely related odorants. However, when rats later had the choice between the two enantiomers, they showed that they could discriminate between them and correctly chose the reinforced odor (C[−]+). This result confirmed their ability to discriminate the enantiomers carvone[−]/carvone[+] when necessary and to generalize when appropriate. In this experiment, the order of the pair presentations was reversed for group 2. When the rats had to choose between the enantiomers (carvone[−]/carvone[+]) before choosing between C[+] and L[−]−. The group exhibited a good discrimination between the enantiomers but always chose C[+] vs. L[−]−. However, this preference was less marked, as if rats knew C[+] was not the previously reinforced odor. This last result triggered another question: why did the rats choose C[+] when compared with L[−]−? It could be due to a generalization of the acquired value of C[−]+ to "carvons," or to a decrease in the reward value of the non-reinforced odor (L[−]−). Experiment 3 allowed us to answer this question and to validate the second hypothesis.

The goal of experiment 3 was indeed to test whether COP also modified the reward value of the non-reinforced odor. This was assessed by comparing how the conditioning procedure affected the preference between the odor paired with water in our paradigm and different odors never included in the conditioning procedure. **Figure 5** summarizes the variations of reward value for the odors used in experiment 3, following pairing C[−] with sucrose and L[−] with water. The comparison of preference before (**Figure 5A**) and after (**Figure 5B**) the olfactory conditioning shows that, as expected, C[−] became preferred over L[−]. In addition, the initial preference for G vs. C[−] was reversed by the conditioning. This confirmed a development of preference for the odor paired with sucrose during conditioning, independently of the value of L[−]. In experiment 2, rats preferentially chose C[+] over L[−]. We first hypothesized that this could be explained by a generalization between the two enantiomers of carvone. However, the initial spontaneous preference for L[−] vs. I disappeared totally, suggesting a decrease in the reward value of L[−] through conditioning. For new pairs of odors like E vs. C[+], conditioning had no influence on the preference for one or the other odor. If there was a generalization between the two carvones, C[+] would have been preferred over E as was the case for C[−]+ compared to G. This comparison therefore weakens the hypothesis of a generalization between the two carvones and instead confirms a decrease in reward value for L[−]− after conditioning, leading to a preference for C[+] when the two odors were simultaneously proposed.

In the present protocol, changes in reward value were measured by the comparison of the consumption of two solutions, each associated with a different odor. Conditioning resulted in a preference for one of these odors, but, interestingly, the nonreinforced odor was also avoided when challenged with another odor. It remains an open question as to whether this avoidance reflects a strategy (a kind of learning by exclusion within the comparison of the two odors) or whether it reflects a real change in reward value, with a depreciation of the value of this odor (i.e., this odor acquires a negative value). Indeed, two theories could explain this reduced consumption of water associated with L[−] after conditioning. The first one is the *model-free theory*. In this case, animals learn by trial-and-error and each new trial reinforces their preference for the sucrose-associated odor (Doll et al., 2012). According to this theory, after many comparisons between C[−]+ and L[−]−, rats consider C[−]+ as rewarding and they will actively seek it. This theory interprets the learnt avoidance of

L[−]− as reflecting a negative value, in a kind of habit behavior acquired during conditioning. However, it seems unlikely that water alone decreases the value of the odor with which it is associated. The *model-based theory*, in contrast with the model-free learning, describes adaptive and dynamic value inference in learning tasks (using a "world model") (Doll et al., 2012), where the animal learns the structure of the task. For instance, it infers that the structure of the test is similar to the conditioning, that is, the test always includes a non-reinforced odor and a reinforced one. In this perspective, L[−]− would not become a negative cue, and would not be avoided, but would signal the other odor to be potentially reinforced, explaining the preference. Furthermore, decisions would arise from a goal-directed control, in contrast to the habit behavior developed in the model-free theory. An interesting insight on these processes may come from the extinction of this preference and, more precisely, from the observation of potential changes in reward values over time, for the CS+ and CS−, respectively.

The present protocol offers a framework to explore in parallel the behavior and the neurobiological processes involved in olfactory appetitive learning. Indeed, in this paradigm the odors are always perceived in an orthonasal way and odor-taste integration might have recruited circuits distinct from those of flavor perception. Using an olfactory aversion paradigm, our group described two different circuits involved according to how the odor was presented (ortho vs retronasally) to the animal during conditioning (Chapuis et al., 2009). It would be interesting to test whether if it is also the case for appetitive learning. More generally, trying to understand the importance of smell to the perception of flavor and the formation of cognitive and emotional responses to food will contribute to the biomedical knowledge required to solve today's rising obesity and diabetes rates.

### **ACKNOWLEDGMENTS**

The authors want to thank Jeremie Naudé and Lauriane Harrington for valuable comments on the manuscript. This work was funded by the French ANR "Programme national de Recherche en alimentation et nutrition humaine." AROMALIM Project: "Représentations sensorielles de l'arôme des aliments et état nutritionnel: de la réception à la cognition." Resp: Rémi Gervais (UMR 5020, Lyon) and supported by the LABEX Cortex.

#### **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: 05 April 2014; accepted: 06 June 2014; published online: 03 July 2014. Citation: Torquet N, Aimé P, Messaoudi B, Garcia S, Ey E, Gervais R, Julliard AK and Ravel N (2014) Olfactory preference conditioning changes the reward value of reinforced and non-reinforced odors. Front. Behav. Neurosci. 8:229. doi: 10.3389/fnbeh. 2014.00229*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Torquet, Aimé, Messaoudi, Garcia, Ey, Gervais, Julliard and Ravel. 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.*

## Modulation of olfactory sensitivity and glucose-sensing by the feeding state in obese Zucker rats

**Pascaline Aimé<sup>1</sup>† , Brigitte Palouzier-Paulignan<sup>1</sup> , Rita Salem<sup>1</sup> , Dolly Al Koborssy<sup>1</sup> , Samuel Garcia<sup>1</sup> , Claude Duchamp<sup>2</sup> , Caroline Romestaing<sup>2</sup> and A. Karyn Julliard<sup>1</sup>\***

<sup>1</sup> Team "Olfaction: From Coding to Memory", Lyon Neuroscience Center, INSERM U1028-CNRS 5292- Université Lyon1, Lyon, France <sup>2</sup> Laboratoire d'Ecologie des Hydrosystèmes Naturels et Anthropisés, CNRS 5023, Villeurbanne, France

#### **Edited by:**

Donald A. Wilson, New York University School of Medicine, USA

#### **Reviewed by:**

Daniel W. Wesson, Case Western Reserve University, USA Edgar Soria, Institut National de la Santé et de la Recherche Médicale, France

#### **\*Correspondence:**

A. Karyn Julliard, Team "Olfaction: From Coding to Memory", Lyon Neuroscience Center, INSERM U1028-CNRS 5292- Université Lyon1, 50 Av. Tony Garnier F-69366 Lyon Cedex 07, France e-mail: karyn.julliard@univ-lyon1.fr

#### **†Present address:**

Department of Pathology and Cell Biology, Columbia University Medical Center, New York, NY 10032, USA

The Zucker fa/fa rat has been widely used as an animal model to study obesity, since it recapitulates most of its behavioral and metabolic dysfunctions, such as hyperphagia, hyperglycemia and insulin resistance. Although it is well established that olfaction is under nutritional and hormonal influences, little is known about the impact of metabolic dysfunctions on olfactory performances and glucose-sensing in the olfactory system of the obese Zucker rat. In the present study, using a behavioral paradigm based on a conditioned olfactory aversion, we have shown that both obese and lean Zucker rats have a better olfactory sensitivity when they are fasted than when they are satiated. Interestingly, the obese Zucker rats displayed a higher olfactory sensitivity than their lean controls. By investigating the molecular mechanisms involved in glucose-sensing in the olfactory system, we demonstrated that sodium-coupled glucose transporters 1 (SGLT1) and insulin dependent glucose transporters 4 (GLUT4) are both expressed in the olfactory bulb (OB). By comparing the expression of GLUT4 and SGLT1 in OB of obese and lean Zucker rats, we found that only SGLT1 is regulated in genotype-dependent manner. Next, we used glucose oxidase biosensors to simultaneously measure in vivo the extracellular fluid glucose concentrations ([Gluc]ECF) in the OB and the cortex. Under metabolic steady state, we have determined that the OB contained twice the amount of glucose found in the cortex. In both regions, the [Gluc]ECF was 2 fold higher in obese rats compared to their lean controls. Under induced dynamic glycemia conditions, insulin injection produced a greater decrease of [Gluc]ECF in the OB than in the cortex. Glucose injection did not affect OB [Gluc]ECF in Zucker fa/fa rats. In conclusion, these results emphasize the importance of glucose for the OB network function and provide strong arguments towards establishing the OB glucose-sensing as a key factor for sensory olfactory processing.

**Keywords: olfactory sensitivity, obesity, olfactory bulb, glucose-sensing, GLUT4, SGLT1, extracellular glucose concentration, feeding state**

### **INTRODUCTION**

The Zucker "fatty" rat is a model of genetic obesity. The mutation, named fatty or *fa*, is autosomic recessive, therefore while *fa/fa* rats become obese at 3–5 weeks of age, the heterozygous *fa/*+ rats remain phenotypically normal. The *fa* mutation affects the extracellular part of the leptin receptor which demonstrates a weaker affinity for leptin and an altered signal transduction (White and Martin, 1997; Yamashita et al., 1997). Leptin receptors are present in several hypothalamic nuclei playing an important role in the central regulation of energy balance (Tartaglia et al., 1995; Mercer et al., 1996; Fei et al., 1997; Shioda et al., 1998; Burguera et al., 2000). Because leptin cannot exert its anorectic action in Zucker *fa/fa* rats, these animals develop a marked hyperphagia during the first weeks of life (Vasselli et al., 1980). Consequently, the Zucker *fa/fa* rats exhibit a severe obese phenotype, highly similar to the human obese syndrome (Guerre-Millo, 1997; Beck, 2000). Zucker *fa/fa* rats display many metabolic and hormonal defects including hyperleptinemia, hyperinsulinemia, mild to moderate hyperglycemia, glucose intolerance as well as peripheral and central insulin resistance (Zucker and Antoniades, 1972; Bray, 1977; Martin et al., 1978; Ikeda et al., 1986; Pénicaud et al., 1987; Zarjevski et al., 1992; Apweiler and Freund, 1993; Guerre-Millo, 1997; Beck, 2000). As a consequence of the leptin receptor gene mutation, in the obese Zucker rats, intracerebroventricular (ICV), injections of insulin or of non-metabolized 2-deoxy-D-glucose do not change food intake in contrast these injections induce hypo- or hyper-phagia, respectively in lean Zucker rats (Ikeda et al., 1980, 1986). The absence of behavioral response to insulin and glucose ICV injections in obese Zucker rats is partly due to an impairment of hypothalamic glucose-sensing neurons. These neurons cannot adapt their firing to fluctuations of interstitial glucose concentration and cannot take part in the regulation of feeding behavior according to the metabolic needs of the organism (Rowe et al., 1996; Spanswick et al., 1997, 2000; Colombani et al., 2009). Among the molecules involved in glucose-sensing, the mRNA expression of glucokinase, a glucose metabolizing enzyme and of GLUT2, a glucose transporter is altered in the hypothalamus of *fa/fa* Zucker rats (Bogacka et al., 2004). Moreover, the regional brain glucose utilization is modified in obese Zuckers, compared to lean controls (Tsujii et al., 1988; Marfaing-Jallat et al., 1992; Doyle et al., 1993). All together these findings converge to the conclusion that, at least in brain areas wellknown to be involved in the control of food intake, the neuronal glucose sensitivity is largely impaired in hyperphagic *fa/fa* Zucker rats.

Food consumption is regulated by sensory modalities among which olfaction takes an important part. In turn, olfactory sensitivity is modulated by feeding state (Aimé et al., 2007) because normal-weight Wistar rats demonstrate a better olfactory sensitivity when they are fasted than when they are satiated (Aimé et al., 2007). Moreover, ICV administrations of leptin or insulin, two anorectic hormones, decrease the olfactory sensitivity of fasted rats to the level of satiated ones (Aimé et al., 2007; Julliard et al., 2007). Conversely, central infusions of orexin A or ghrelin, two orexigenic peptides, increase the olfactory sensitivity of satiated rats to the level of fasted ones (Julliard et al., 2007; Tong et al., 2011). These converging evidences demonstrate that the olfactory bulb (OB), the first central structure responsible for the olfactory information processing, is targeted by signals (orexin A, ghrelin, leptin, insulin, NPY and CCK) involved in the regulation of energy balance (Palouzier-Paulignan et al., 2012). In addition to the hormones involved in food intake regulation, a large body of evidence indicates that the olfactory system is also sensitive to blood-borne nutrients. Two neuronal markers of glucose sensitivity, insulin-dependent glucose transporter type 4 (GLUT4) and sodium glucose co-transporter type 1 (SGLT1), are found in the olfactory system (El Messari et al., 1998). Recent patch clamp studies have shown that mitral cells change their firing rate in response to changes in glucose concentration, identifying these neurons as glucose sensors (Tucker et al., 2010, 2013) and also to changes in insulin concentration (Fadool et al., 2000; Kuczewski et al., 2014).

Given the extreme metabolic features of the obese Zucker *fa/fa* rats described above and the recent studies demonstrating a strong effect of the peripheral and central signals involved in the regulation of energy balance on olfactory processing and glucose sensitivity in the olfactory system, the present study was conducted to (i) test whether the obese Zucker *fa/fa* rats display a normal regulation of the olfactory sensitivity by the feeding states; (ii) compare the olfactory sensitivity of lean Zucker *fa/*+ rats and obese Zucker *fa/fa* rats; and (iii) investigate whether the glucose concentration, GLUT4 and SGLT1 expression in the OB might be altered in the genetically obese and moderately diabetic Zucker *fa/fa* rat.

### **MATERIALS AND METHODS**

#### **ANIMALS PREPARATION**

Male lean (*fa/*+; *n* = 24) and obese (*fa/fa*; *n* = 24) Zucker rats, were purchased from Charles River Laboratories. The animals were handled (5 min/day) and weighed daily. On arrival, they were 7–8 weeks old and weighed 299 ± 7.2 g (*fa/+)* and 344 ± 11.8 g (*fa/fa*). Experiments were carried out in accordance with the European Community Council Directive of November 24th, 1986 (86/609/EEC) for the care and use of laboratory animals. The Animals were housed in groups (4 rats) in plexiglas chambers at constant ambient temperature and relative humidity (22 ± 0.5◦C and 50 ± 5%). They were acclimated to a 12 h light/12 h dark inverted cycle, with the light turned on at Zeitgeber time zero (ZT0). Upon arrival, the rats were given *ad libitum* access to food (chow pellets, Harlan, France) and water. 2–3 weeks prior to experimental procedure, the rats were gradually habituated to a 20 h/day food restriction (FR) schedule in which they had access to food during dark phase only, from ZT16 to ZT20. Since daily fluctuations in glycemia and insulin levels are cued by food intake (Kaul and Berdanier, 1975; Sitren and Stevenson, 1978), a single daily meal was imposed to synchronize the circadian variations of glycemia and insulin secretion among the animal cohort. Animals were handled (5 min/day) and weighed daily to assess their adaptation to FR. The daily amount of food consumed was measured for lean and obese rats in order to avoid weight loss and allow stabilization or a small gain of body weight. The daily food distribution schedule was performed by an automatic food distribution system (homemade) piloted by a computer using Matlab software.

#### **BEHAVIORAL PROCEDURE**

The behavioral test was performed as previously described (Aimé et al., 2007, 2012; Julliard et al., 2007; Tong et al., 2011) with minor modifications, as follow. 12 Zucker *fa/*+ and 12 Zucker *fa/fa* rats were submitted to the behavioral procedure. The conditioned odor aversion (COA) protocol consisted in pairing the ingestion of an odorized drink with an intraperitoneal (i.p.) injection of lithium chloride (LiCl, Sigma-Aldrich). This procedure induces a robust aversion to the odor diluted in the drinking solution. The olfactory sensitivity is then tested by measuring the aversive behavioral response to a range of lower odor concentrations. In order to induce drink intake during behavioral test, in addition to the FR, rats were placed on a 23 h water restriction schedule that started 1 week prior behavioral procedure and was maintained throughout the behavioral protocol. Animals had access to water for 1 h, during the daily food distribution. Experiments consisted of two 5-min daily sessions occurring at 9 h intervals. The first session (S1) started at the beginning of the dark phase (ZT13; after 17 h of fasting) and the second (S2) started during the dark phase, during the postprandial phase (ZT22, 2 h after the end of the meal; **Figure 1A**). This paradigm allowed each animal to be successively tested in two steady physiological states, fasted (S1) and satiated (S2) each day of the behavioral test. In a first period, corresponding to the habituation, rats were trained for 2 days to lick pure water at the two drinking tubes of the experimental cage (not shown). The experimental set-up allowed the recording of licking behavior using a two-tube device described elsewhere (Aimé et al., 2007, 2012; Julliard et al., 2007). During the next 2 days (**Figure 1B**: D1 to D2, Aversion acquisition), rats only had access to water odorized with isoamyl-acetate (ISO, Sigma-Aldrich). An ISO consumption of more than 0.5 mL was paired 15 min later, with an i.p. injection of LiCl (10 mL/kg at 0.15 M) to induce gastric malaise and establish a COA to ISO. ISO was used

an i.p. injection of LiCl to induce a COA. On D3, corresponding to aversion test (Avers. test), the COA efficiency was tested by giving the animals the choice between water odorized with ISO 10−<sup>5</sup> and pure water (10−<sup>5</sup> /W). During the olfactory detection test period (D4–D8), rats were offered a choice between pure water and water odorized with ISO at lower dilutions ranging from 10−<sup>11</sup> to 10−<sup>7</sup> . On D9, (Avers. re-test) the COA stability was assessed by giving the rats the choice between ISO 10−<sup>5</sup> and pure water (10−<sup>5</sup> /W).

at the dilution of 10−<sup>5</sup> , for which the solution has a strong odor but has been shown to be tasteless (Slotnick et al., 1997). Thus at this dilution, the rats can only identify ISO via the olfactory pathways. Depending on individuals, 1–4 conditioning sessions (S1 and S2 for each 2 days) were required for the animal to develop a strong aversion to ISO 10−<sup>5</sup> . On D3 (**Figure 1B**: Aversion test), the COA efficiency was tested by giving the animals the choice between pure water and water odorized with ISO 10−<sup>5</sup> . Then the animals were submitted to the olfactory detection test (**Figure 1B**: D4 to D8). For this, rats were offered a choice between pure water and water odorized with ISO at lower dilutions. At the beginning of each experimental session of the olfactory detection test, the animals were placed under the odorized water tube. Rat olfactory detection for ISO was thus assessed using a forced-choice task, and not by using a simple choice task, since the thirsty rats were forced to smell the odorized tube first. This procedure was chosen to avoid the possibility that the rats, highly motivated by thirst, would go by chance to the pure water tube first, drink only water, and not sample the ISO tube. The conditioned aversion was then tested using drinking solution with ISO dilutions ranging from 10−<sup>11</sup> to 10−<sup>7</sup> . Each given odor concentration was presented to the same rat in fasted (ZT13) and later in satiated (ZT22: *i.e.,* 2 h after the food intake end) status. On D9 (**Figure 1B**: Aversion re-test), COA stability was assessed by giving the rats the choice between the standard ISO solution (10−<sup>5</sup> ) and pure water.

For each concentration, an olfactory detection index was calculated. It corresponds to the proportion of the number of licks at the pure water tube normalized to the total number of licks (odorized + pure water) in the experimental device. When rats perceived ISO in the drinking solution (and consequently avoided it), they licked more pure water during the experimental session, resulting in a higher olfactory detection index. In addition, the number of licks during the first ISO consumption (first licking burst) in the experimental device was recorded. This data allowed us to accurately measure the delay necessary for the animals to be able to detect the aversive ISO diluted in the drinking solution. Data from rats starting the experimental session by drinking at the pure water tube (maximum 2 rats out of 12 per ISO dilution) was excluded from this analysis. In order to exclude a potential bias induced by the number of conditioning sessions on the strength of the COA, a Pearson correlation test was performed between the number of LiCl injections and the olfactory detection index for each animal, for each nutritional status and for each ISO concentration tested. No correlation was observed (*P* > 0.05) for both *fa/fa* and *fa/*+ rats. The number of lick during the first burst, the end of the licking burst and the total number of licks at each tube during the experimental sessions were assessed and analyzed using SciPy and MySql database software (Open Source Licenses). A licking burst was defined by a train of high frequency licks (7–10 Hz) and the end of licking burst was determined when the last lick was followed by a period of inactivity lasting at least 1 s.

#### **PHYSIOLOGICAL MEASUREMENTS**

To characterize the two metabolic steady states of the fasted and satiated rats, the concentrations of plasma glucose, plasma insulin, and OB insulin were measured. Peripheral blood glucose level was determined by sampling 5 µL of tail blood 1 h before and 2 h after food intake for fasted and satiated rats respectively. Glucose levels were monitored with a glucose meter (Accu-Chek® Roche, Mannheim, Allemagne/Performa). Plasma and OB insulin levels were measured in fasted (ZT12; *n* = 6 for each genotype) or in satiated (ZT22; *n* = 6 for each genotype) state. At the end of the behavioral procedure, rats were deeply anesthetized using an i.p. injection of ketamine (Imalgene, 80 mg/kg) and xylazine (Rompun, 10 mg/kg), then rats were euthanized, and their OBs immediately frozen in liquid nitrogen. Trunk blood was collected in heparinized tubes, and the plasma fraction was separated by centrifugation at 2000 g for 5 min. Insulin was extracted from one OB per rat according to the procedure of Baskin et al. (1983) and the other OB was saved for Western blotting. To determine the influence of the yield of the insulin extraction procedure, samples with known amounts of insulin, were submitted to the same protocol. The mean extraction output was found to be ∼40%. Plasma and OB insulin levels were determined using a solid-phase, two-site enzyme immunoassay following the manufacturer's protocol (Mercodia Ultrasensitive Rat Insulin ELISA).

#### **WESTERN BLOT ANALYSIS**

One of two OB removed from rats previously used for behavioral procedure was homogenized in ice-cold homogenization buffer (Tris 5 mM, EGTA 2 mM, pH 7.4) supplemented with an antiprotease cocktail (SIGMA P8340, 10 µl per mg tissue). Western Blot analysis was performed with OB homogenates from rats euthanized either in fasted (ZT12; *n* = 4 for each genotype) or in satiated (ZT22; *n* = 4 for each genotype) state. Protein concentrations were determined by BCA assay according to the manufacturer's recommendation (PIERCE #23225). 100 µg of proteins were loaded into 7.5% SDS-polyacrylamide gels (SDS-PAGE) and electrophoresed for 1.5 h at 110 V. Proteins were then electro-transferred for 1 h at 300 mA onto a PVDF membrane. Membranes were blocked with 5% milk in TBS containing 0.05% of Tween 20 and incubated overnight with polyclonal antibodies directed against GLUT4 (Millipore 07-1404; dilution 1:250), or SGLT1 (Santa Cruz sc-20584; dilution 1:1000). Equal loading was verified using Ponceau red stain, and by detection of the control protein β-actin (Sigma; diultion 1:8000). Membranes were washed in 0.05% Tween-PBS buffer and incubated with horseradish peroxidase-conjugated secondary antibody (dilution 1:10000). Signals were detected using the enhanced chemiluminescence detection system (Pierce #32106). Immunoblots were scanned using a desktop scanner (Epson Perfection V350) and Adobe Photoshop. Band intensities were determined using Scion Image (Scion Corporation, USA).

#### **IMMUNOSTAINING**

Animals were anesthetized (using the same protocol as described previously for ELISA and Western Blot analysis) and euthanized in the fasted (ZT12; *n* = 4 for each genotype) or in the satiated (ZT22; *n* = 4 for each genotype) state. Immunofluorescence was performed on fresh frozen brain samples by using a modification of a published method (Julliard and Hartmann, 1998). Brain cryosections were pre-incubated for 15 min with a blocking buffer containing 0.1 M phosphate buffer saline (PBS, pH = 7.4), 3% bovine serum albumin (BSA, Sigma-Aldrich) and 5% normal serum from the host species of the antibodies. The sections were then incubated overnight at 4◦C with primary antibodies for SGLT-1 (R-16) (1:100; Santa Cruz Biochemicals, Santa Cruz, CA). A GLUT4 mouse antibody was used which recognizes an epitope in the cytoplasmic portion of GLUT4 [1F8] (1:100 abCam). The sections were washed with 0.1 M PBS/3% BSA and incubated for 1 h at room temperature with anti-rabbit IgG, anti-mouse IgG, or anti-goat secondary antisera coupled to Alexa 488 (1:100), Cy3 (1:200; Jackson Immunoresearch), or Cy5 (1/100; Jackson Immunoresearch) respectively. After the final wash with PBS, slides were mounted with Vectashield mounting medium containing DAPI for nuclear staining (Vector Laboratories). Images were acquired using a Zeiss Apotome epifluorescence microscope equipped with a digital camera and Axiovision software.

#### **BIOPROBE MEASUREMENTS OF EXTRACELLULAR GLUCOSE Surgical procedure**

Eight Zucker *fa/*+ and eight Zucker *fa/fa* rats were used during measurements of extracellular glucose in the brain. Rats in the fasted (ZT12, *n* = 4 for each genotype) or satiated (ZT22, *n* = 4 for each genotype) state were anesthetized with urethane (1.5 mg/kg, i.p.). The anesthetized animals were placed in a stereotaxic apparatus and kept on a heating pad, and additional doses of urethane were supplied as needed. The surgical procedure consisted of drilling three burr holes into the skull to expose the lateral region of each OB and the somatosensory cortex as a control brain area. The first glucose-oxidase biosensor was implanted into the lateral part of one OB, within or close to the glomerular layer (GL; coordinates: AP +6.5 mm from Bregma, M/L −2.3 mm and D/V −1.0 mm from dura, **Figure 2**). The second glucoseoxidase biosensor was inserted in the controlateral somatosensory cortex (coordinates: AP −3.5 mm from Bregma, M/L +2.3 and D/V −2.3 mm from dura). A control BSA sensor was implanted in the other OB to measure nonspecific variations in oxidation current (coordinates similar to glucose biosensors implanted in OB, **Figure 2**). A reference electrode (Ag/AgCl) was placed into neck muscles during *in vivo* recordings.

#### **Preparation of electrochemical sensors**

The glucose biosensor uses glucose oxidase and amperometric detection of hydrogen peroxide (Vasylieva et al., 2011). The tip of the probe was coated with glucose oxidase only on 100–150 µm in length (Vasylieva et al., 2011) to target mainly the GL, which thickness is ranging from 100 to 300 µm. Glucose oxidase metabolizes glucose in the extracellular fluid. An oxidation current is thus generated and measured using Neurolabscope software. Each biosensor was connected to a potentiostat, which sent readings of the current generated by glucose in extracellular fluid to a computer. Glucose biosensors have been shown previously to have a range of 0–10 mM with *in vitro* sensitivity of 1.6 ± 0.4 nA/mM (mean ± SEM). To confirm the accuracy of the biosensors, prior to implantation and immediately following testing, biosensor probe were placed in 0.1 M PBS, connected to the potentiostat, and readings were allowed to stabilize (generally stable within 15–30 min).

#### **Selectivity and calibration of biosensors**

Before the experiments all biosensors were tested for detection of serotonin (5-HT, 20 µM in PBS; Sigma) and H2O<sup>2</sup> (1 µM in PBS). Only electrodes exhibiting less than 1.2 µA.mM−<sup>1</sup> cm−<sup>2</sup> response for 5-HT were included in the study. *In vitro* calibrations were performed in standard PBS (0.01 M, pH 7.4) and solutions were maintained at a temperature of 36.5◦C, comparable to the brain of an anesthetized rat (Zhu et al., 2004). The reference electrode (Ag/AgCl) was placed directly in the solution. After a stable baseline reading, the glucose sensors were calibrated using glucose solutions at different concentrations (0; 0.5; 1; 1.25 and 1.5 mM) to establish the nA/mM ratio. The applied voltage for amperometric studies was +500 mV. Biosensors were calibrated before and after each real-time *in vivo* experiment to ensure the sensitivity remained stable. Quantitative assessments of brain glucose concentrations were obtained by subtracting the nonspecific current of the control biosensor (BSA) from the output of the glucose biosensors.

#### **Real-time measurements in vivo**

To compare glucose level variations in the OB and the cortex to the peripheral glucose concentrations, 5 µl blood samples were collected from the femoral artery at the beginning of the surgery and then every 10 min thereafter. Glucose readings were performed with a glucose meter (Accu-Chek® Roche, Mannheim, Allemagne/Performa).

Measurements of extracellular glucose in rat brains started 1 h after the implantation of the electrodes, to allow restoration of the blood-brain barrier. Fasted rats (*n* = 4 for each genotype) were anesthetized before food intake, and satiated rats (*n* = 4 for each genotype) after food intake.

and a control BSA sensor (BSA, green line) in the other bulb. The latter was set

Recordings started once the electrodes were implanted, and lasted for approximately 3 h. Currents obtained after the signal stabilization corresponded to the initial steady state of the animal, *i.e.*, fasted or satiated. To evaluate possible fluctuations of central glucose level during dynamic glycemia conditions, an i.p. injection of glucose (Lavoisier, Paris, France 30%; i.p. 3 g/kg) to the fasted rats, or a subcutaneous injection of insulin (Sigma, Saint Quentin-Fallavier, France; 7.5 U/mL; subcutaneous, 25 U/kg) to the satiated rats, were given to measure the effects of acute modifications in peripheral glucose levels on central structures (OB and cortex). 1 h later, rats in the induced hyperglycemic state were injected with insulin and rats in the induced hypoglycemic state were injected with glucose. In 1 of 9 satiated rats glucose was monitored in the OB alone, and three rats received only insulin injection. In 1 of 8 fasted rats, glucose was monitored only in the OB, and three rats received only glucose injection. When recordings ended, rats were euthanized using sodium pentobarbital (i.p. 3 g/kg).

#### **STATISTICAL ANALYSIS**

Data are shown as mean values ± SEM. For olfactory detection all percentage measures were transformed using the arcsine square root transformation to normalize the data and stabilize variance (Sokal and Rohlf, 1981). For behavioral data, Western Blot data, and extracellular glucose measurements, statistical analysis were performed by either a Student's *t*-test or one- or two-way repeated-measures ANOVA depending on the data set. A Student-Newman-Keuls (SNK) *post hoc* test was used to complete the analysis when appropriate (Statview software). Physiological parameters (glycemia, plasma and OB insulin level) were analyzed using a non-parametric Mann-Whitney test, or Wilcoxon test when data were paired (Statview Software).

granular cell layer, GL: glomerular layer, MCL: mitral cell layer).

#### **RESULTS**

#### **EFFECT OF FOOD RESTRICTION ON BODY WEIGHT OF LEAN AND OBESE ZUCKER RATS**

Before any other experiment, *fa/*+ and *fa/fa* Zucker rats were gradually habituated to a 20 h/day FR schedule (**Figure 1A**). In order to determine the effect of FR schedule on weight gain, body weight of the rats was measured, on their arrival, just before the FR beginning and after 3–4 weeks of FR corresponding to the end of the experimental procedures (**Figure 3**). On their arrival, the rats were 7–8 weeks old and the body weight of the two genotypes was already significantly different (299 ± 7.2 g *vs*. 344 ± 11.8 g in *fa/*+ *vs*. *fa/fa* respectively; *P* < 0.001 Mann Whitney test). At the end of the experimental procedures, the difference in the body weight between *fa/*+ and *fa/fa* and rats was larger (312 ± 7.1 g *vs*. 412 ± 7.1 g in *fa/+ vs*. *fa/fa* respectively; *P* < 0.0001 Mann Whitney test). While the *fa/*+ rats' body weight remained stable during the FR, (306 ± 5.7 g *vs.* 312.3 ± 4.4 g before FR *vs.* after the experimental procedure, *P* = 0.1, paired *t*-test) the *fa/fa* rats' body weight had continued to increase (372.9 ± 7.4 g *vs.* 412 ± 7.1 g before FR *vs.* after the experimental procedure, *P* < 0.0001 paired *t*-test).

#### **EFFECT OF FEEDING STATE ON GLYCEMIA, INSULINEMIA, AND OB INSULIN LEVELS IN LEAN AND OBESE ZUCKER RATS**

In order to characterize the physiological hallmarks of the two genotypes of Zucker rats and to evaluate the physiological effects of the feeding state, glycemia, as well as plasma and OB insulin levels were measured in both 20 h fasted and satiated *fa/*+ and *fa/fa* Zucker rats (**Table 1**). A two-way ANOVA

with feeding state and genotype as factors revealed a significant effect of these two factors on all three measurements; glycemia (*F*(1,20) = 8.4, *P* < 0.01 and *F*(1,20) = 5.0, *P* < 0.05 respectively), plasma (*F*(1,20) = 14.4, *P* < 0.005 and *F*(1,20) = 247.3, *P* < 0.0001 respectively) and OB insulin levels (*F*(1,20) = 38.7, *P* < 0.0001 and *F*(1,20) = 320.7, *P* < 0.0001 respectively). For each genotype, the three measurements were significantly higher in satiated than in fasted rats (**Table 1**: Mann-Whitney tests).

By comparing the two genotypes, obese *fa/fa* rats were moderately hyperglycemic (only in satiated state), compared to lean *fa/*+ rats and their insulin levels (in plasma and in OB) were higher than in *fa/*+ regardless of the feeding state (**Table 1**: Mann-Whitney tests). The difference in insulin concentration between the two rat strains was smaller in the OB (3–3.87 fold more concentrated in satiated and fasted states, respectively) than in the plasma (5.5–10 fold more concentrated in satiated and fasted states, respectively).

#### **EFFECT OF FEEDING STATE AND OF GENOTYPE ON OLFACTORY DETECTION**

In order to compare the olfactory detection abilities of *fa/*+ and *fa/fa* rats in fasted and satiated states, a behavioral test based on COA was performed and olfactory detection indexes were measured for ISO dilutions ranging from 10−<sup>11</sup> to 10−<sup>7</sup> (**Figures 4A,B**). During the olfactory detection test (**Figure 1B**, from day 4 to day 8, D4 to D8), a two-way ANOVA with feeding state and odor as factors revealed significant effects of feeding state (*fa/*+ *F*(1,88) = 15.8, *P* < 0.01; *fa/fa F*(1,88) = 30.4, *P* < 0.0001) and odor (*fa/*+ *F*(4,88) = 27.1, *P* < 0.0001; *fa/fa F*(4,88) = 15.8, *P* < 0.0001) on olfactory detection indexes. *Post hoc* analyses (paired *t*-test) revealed that *fa/*+ rats have higher olfactory detection indexes in the fasted state compared to the satiated state, for ISO 10−<sup>8</sup> (fasted: 62.6 ± 6.8%, satiated: 19.5 ± 3.9%, *P* < 0.0001) and ISO 10−<sup>7</sup> (fasted: 95.3 ± 1.7%, satiated: 55.6 ± 9.0%, *P* < 0.0005). This effect was even more pronounced in *fa/fa* rats, which have higher olfactory detection indexes in the fasted state compared to the satiated state for a concentration as low as ISO 10−<sup>9</sup> (fasted: 51.3 ± 9.0%; satiated: 17.0 ± 4.0%, *P* < 0.01) but also for ISO 10−<sup>8</sup> (fasted: 58.2 ± 7.4%, satiated: 12.9 ± 3.2%, *P* < 0.0001) and ISO 10−<sup>7</sup> (fasted: 89.8 ± 5.6%, satiated: 61.2 ± 9.2%, *P* < 0.005).

To ensure that the COA was robust and maintained throughout the olfactory detection test, the olfactory detection indexes were measured at ISO 10−<sup>5</sup> for all the animals, before (**Figure 1B**, Aversion test, on D3) and after (Aversion re-test, on D9) the olfactory detection test. During the aversion test (D3), carried out after the conditioning, the animals licked almost exclusively from the pure-water tube, as demonstrated by olfactory indexes close to 100% (**Figure 4C**, fasted *fa/*+: 99.97 ± 0.03%, satiated *fa/*+: 99.98 ± 0.02%, fasted *fa/fa*: 99.85 ± 0.15%, satiated *fa/fa*: 99.80 ± 0.10%) showing that the COA was well established at the beginning of the olfactory detection test. On the last day of the behavioral experiment (Aversion re-test, D9), all the animals drank, again, almost exclusively at the pure-water tube (fasted *fa/*+: 99.90 ± 0.06%, satiated *fa/*+: 99.38 ± 0.23%, fasted *fa/fa*: 98.01 ± 1.80%, satiated *fa/fa*: 98.05 ± 0.90%),


For each genotype, glycemia, OB and plasma insulin levels were statistically higher in satiated compared to fasted animals. For each feeding state (fasted; Fast. and satiated; Sat.) levels of these three types of measure were significantly higher in fa/fa rats compared to their fa/+ counterparts, except for glycemia of fasted rats (Mann-Whitney test, n = 6/feeding state/genotype Zucker, ns: P > 0.05).

indicating that COA was maintained throughout the behavioral experiment.

In order to further compare the olfactory detection abilities of *fa/*+ and *fa/fa* rats, the number of licks were measured during the first consumption of each given ISO concentration in the experimental device (**Figures 4D–F**). This measurement shows the number of licks that were necessary for the animal to detect the aversive ISO odor diluted in the drinking solution. One should expect that the better the animal can detect the ISO, the lower the number of licks during the first consumption would be. A three-way ANOVA with genotype, feeding state and odor as factors revealed significant effects of these three factors on the number of licks during the first ISO consumption (genotype: *F*(1,132) = 4.3, *P* < 0.05, feeding state: *F*(1,132) = 15.5, *P* < 0.0005, odor: *F*(4,132) = 7.1, *P* < 0.0001). Fasted animals required fewer licks than satiated animals to detect and leave the aversive solution (*post hoc* analysis SNK), further demonstrating that animals have better olfactory sensitivity when fasted than satiated. In fasted states (**Figure 4D**), *fa/fa* rats required fewer licks than *fa/*+ rats to detect and leave the aversive solution at ISO 10−<sup>11</sup> (*fa/fa*: 33.89 ± 7.6, *fa/*+: 220.3 ± 78.0, *P* < 0.05 *t*-test), and at ISO 10−<sup>8</sup> (*fa/fa* 15.67.5 ± 7.3, *fa/*+: 80.17 ± 29.2, *P* < 0.05 *t*-test). In satiated state (**Figure 4E**) *fa/fa* leaved the aversive solution faster than *fa/*+ rats at ISO 10−<sup>10</sup> (*fa/fa*: 138.25 ± 21.2, *fa/*+: 247.67 ± 44.3, *P* < 0.05 *t*-test) and also at ISO 10−<sup>9</sup> *(fa/fa*: 142.41 ± 35.2, *fa/*+: 275.33 ± 40.0, *P* < 0.05 *t*-test). In addition when data obtained in fasted and in satiated states were pooled (**Figure 4F**), the lick number during the first burst was significantly lower for *fa/fa* rats compared to their lean counterparts at ISO 10−11, ISO 10−<sup>8</sup> and ISO 10−<sup>9</sup> (*P* < 0.001, *P* < 0.05, *P* < 0.05 respectively, *t*-test). Neither genotype (*fa/+ vs. fa/fa* for each feeding state: *t*-test) nor feeding state (fasted vs. satiated for each genotype: paired *t*-test) effect was observed on the lick number of the first burst at the pure water tube during the habituation period observed (*P* > 0.05, fasted rats: *fa/*+ 76.16 ± 21.5, *fa/fa* 55.16 ± 6.11; satiated rats: *fa/*+ 110.8 ± 17.1, *fa/fa* 78.25 ± 11.1). Altogether, these results indicated that *fa/fa* rats have a better olfactory sensitivity than *fa/*+ rats.

#### **EFFECTS OF FEEDING STATE ON EXPRESSION OF GLUT4 AND SGLT1 IN OB OF THE TWO GENOTYPES**

In order to analyze the localization of SGLT1 and GLUT4 within the different OB layers in Zucker *fa/fa* and *fa/*+ rats, an immunofluorescence experiment was performed. SGLT1 and GLUT4 were found in different layers of the OB. In both lean and obese Zucker rats, immunostaining of SGLT1, a transporter of non-metabolized glucose, revealed a layer-specific pattern (**Figures 5A,B**). The highest staining was observed in the inner part of EPL (iEPL) with a stronger expression in *fa/fa* rats (**Figure 5A**). In the two genotypes, the outer part of the EPL (oEPL) was unlabeled (**Figures 5A,B**). SGLT1 immunostaining was also detected in some glomeruli more particularly in *fa/fa* Zucker rats. In both lean and obese Zucker rats, the highest level of GLUT4 immunostaining was detected in the glomerular layer (GL) and in the mitral cell layer (MCL; **Figures 5C,D**) while the nerve layer appeared unstained. Within the glomerular neuropil, GLUT4 staining varied from strongly to unlabeled glomeruli. The pictures shown in the manuscript are representative of SGLT1 and GLUT4 immunostaining in the OB of both rat genotypes and were consistent across rats within each group. We could not see any difference in the localization of GLUT4 and SGLT1 in fasted rats compared to satiated rats, for both genotypes (data not shown).

The effects of feeding state and genotype on GLUT4 and SGLT1 protein levels in the OB were analyzed by Western blotting (**Figure 6**). Total tissue GLUT4 and SGLT1 were analyzed using the total protein recovery from the OBs of four satiated and four fasted rats of each genotype (**Figure 6A**). A two-way ANOVA on SGLT1 protein levels (**Figure 6B**), with feeding state and genotype as factors, revealed no effect of feeding state, but a significant effect of genotype (*F*(1,12) = 14.87, *P* < 0.005). *Post hoc* analyses (SNK) of these data confirmed a higher expression of SGLT1 in *fa/fa* Zucker rats compared to *fa/*+ rats. This higher expression is observed in the satiated state only, as revealed by Mann-Whitney test (*P* < 0.05). Neither feeding state nor genotype had an effect on GLUT4 levels (**Figure 6C**) (*F*(1,12) = 0.54, *P* = 0.47 and *F*(1,12) = 1.146, *P* = 0.3, respectively). Together these results indicated that SGLT1 protein levels were significantly increased in the OB of Zucker *fa/fa* rats compared to *fa/*+ rats.

#### **REAL TIME IN VIVO MONITORING OF EXTRACELLULAR FLUID GLUCOSE IN OB AND CORTEX OF LEAN AND OBESE ZUCKER RATS During steady states**

Extracellular fluid glucose concentration ([Gluc]ECF) was measured simultaneously in the OB glomerular layer and the

olfactory detection indexes were significantly higher for fasted fa/+ rats compared to satiated fa/+ rats for ISO 10−<sup>8</sup> and ISO 10−<sup>7</sup> . The mean olfactory detection indexes were significantly higher for fasted fa/fa rats compared to satiated fa/fa rats for ISO 10−<sup>9</sup> , ISO 10−<sup>8</sup> and ISO 10−<sup>7</sup> . **(C)** Aversion test and aversion re-test. Bar graphs showing consumption was significantly lower in fa/fa rats compared to fa/+rats for ISO 10−<sup>11</sup> and ISO 10−<sup>9</sup> . **(F)** The mean number of licks during the first ISO consumption was significantly lower for fa/fa rats compared to fa/+ rats for ISO 10−11, ISO 10−<sup>10</sup> and ISO 10−<sup>9</sup> regardless to feeding state.

somatosensory cortex of lean or obese Zucker rats, in fasted or satiated state. 2 weeks prior to the experimental study, animals were habituated to a 2 h feeding/22 h starvation schedule. At the beginning of the experiment, the metabolic status of the rats was considered to be steady, since satiety or hunger had been maintained for several hours. Glycemia was monitored to check each metabolic (fasted or satiated) steady state. As shown in **Table 1**, glycemia was significantly higher in satiated than in fasted state and in *fa/fa* than in *fa/*+ rats. The effects of feeding state and genotype on [Gluc]ECF in OB and cortex were analyzed by using a three-way ANOVA with brain areas (OB and cortex), feeding state and genotype as factors. A significant effect on [Gluc]ECF was observed during the initial steady state for factors "brain areas" and "genotype" (*F*(1,24) = 13.6, *P* < 0.005; *F*(1,24) = 21.6, *P* < 0.0001 respectively, **Figure 7**). [Gluc]ECF was significantly higher in the OB than in the cortex, and in *fa/fa* than *fa/*+ rats (SNK *post hoc* tests). No significant effect of feeding state was observed on [Gluc]ECF despite differences in plasma glucose levels between satiated and fasted rats (**Table 1**). For each genotype, [Gluc]ECF was significantly higher in OB compared to cortex (Wilcoxon test, \* *P* < 0.05 for each rat genotype; in mM: Fasted: *fa/*+ in the OB, 1.48 ± 0.35, in the cortex, 0.92 ± 0.33; *fa/fa* OB, 3.05 ± 0.5, cortex, 1.67 ± 0.23; Satiated: *fa/*+ OB, 1.48 ± 0.25; cortex, 0.55 ± 0.09; *fa/fa* OB, 2.35 ± 0.5; cortex 1.5 ± 0.15, **Figure 7**). In both OB and cortex [Gluc]ECF were higher in *fa/fa* compared

to *fa/*+ (Mann Whitney test, <sup>e</sup> *P* < 0.05 for each area brain, **Figure 7**).

### **During dynamic states following insulin and glucose injections**

Real time *in vivo* measurements of [Gluc]ECF were recorded simultaneously from the OB and the somatosensory cortex of *fa/*+ and *fa/fa* Zucker rats either fasted (**Figures 8A,C** respectively) or satiated (**Figures 8B,D** respectively). For each genotype, the two groups of rats (fasted and satiated) received two injections: one of glucose (G-Inj) and one of insulin (I-Ins). Fasted rats (**Figures 8A,C**) first received a glucose injection (G-Inj1) followed by an insulin injection (I-Inj2). This order

was reversed for satiated rats (**Figures 8B,D**, I-Inj1, G-Inj2). Glucose and insulin injections induced substantial fluctuations in extracellular glucose levels, especially in the OB of *fa/*+ rats. In *fa/fa* compared to *fa/*+ rats, glucose injection (G-Inj) induced less pronounced variations of [Gluc]ECF. and insulin injection (I-Inj) induced a slower decrease of [Gluc]ECF. For each feeding state (satiated and fasted) a mixed ANOVA was performed with an intra-subject factor, corresponding to the three conditions (steady-state, Inj1 and Inj2) and two intersubject factors: genotype (lean and obese) and brain area (OB and Cortex). Both in fasted (**Figure 8E**) and satiated

**FIGURE 7 | Genotype effect on the [Gluc]**ECF **at steady state in OB and cortex**. [Gluc]ECF in OB (full bars) and in cortex (Ctx, hatched bars) of lean fa/+ (left) and obese fa/fa (right) in fasted (Fast., blue) and satiated (Sat., red) sates. A three-way ANOVA with brain areas (OB and cortex), feeding state and genotype as factors showed a significant effect of brain areas (not shown) and genotype (not shown). Both in OB and Ctx, [Gluc]ECF were higher in fa/fa compared to fa/+ (Mann Whitney tests, <sup>e</sup> P < 0.05 for each area brain). For each genotype, [Gluc]ECF was significantly higher in OB compared to Ctx (not shown). Values are expressed as the mean ± SEM, n = 4/feeding state/genotype.

(**Figure 8F**) rats, the statistical analysis showed a significant difference between conditions (*F*(2,24) = 26.66, *P* < 0.0001; *F*(2,24) = 29.28, *P* < 0.0001, respectively), genotype (*F*(1,24) = 8.35, # *P* < 0.01 **Figure 8E**; *F*(1,24) = 22.48, ### *P* < 0.0005, respectively **Figure 8F**) and brain areas (*F*(1,24) = 11.91, *P* < 0.005; *F*(1,24) = 5.56, *P* < 0.05, respectively). In each feeding state (satiated and fasted), [Gluc]ECF was higher in *fa/fa* than in *fa/*+ rats, and in the OB than in the cortex (SNK *post hoc* tests). Moreover G-Inj increased [Gluc]ECF significantly above steady-state and I-Inj. levels, and I-Inj decreased [Gluc]ECF significantly below steady-state and G-Inj levels (SNK *post hoc* tests).

Next, we calculated the difference of [Gluc]ECF measurements between steady state and after the first injection (G-Inj1 or I-Inj1). This measurement corresponds to the fluctuations of [Gluc]ECF shown in **Figure 9**. In each group (satiated and fasted rats) a two way ANOVA was performed with brain areas (OB and cortex) and genotype (*fa/*+ and *fa/fa*) as factors. The ANOVA demonstrated a significant effect of brain areas in fasted (*F*(1,12) = 7.03, *P* < 0.01) and satiated (*F*(1,12) = 9.81, *P* < 0.01) rats. Greatest [Gluc]ECF fluctuations were observed in the OB compared to the cortex (SNK *post hoc* tests). The ANOVA also demonstrated a significant difference on [Gluc]ECF fluctuation between *fa/*+ and *fa/fa* rats, but only in fasted rats (**Figure 9**, *F*(1,12) = 5.1, # *P* < 0.05), the fluctuations being greatest in *fa/*+ than in *fa/fa* rats. Further analysis demonstrated that this was due to fluctuations in the OB, because glucose injections had induced smallest fluctuations of [Gluc]ECF in the OB of *fa/fa* rats than of *fa/*+ (Mann Whitney test <sup>φ</sup> *P* < 0.05). In satiated *fa/*+ rats only, insulin injection, had

Composite figures compiled from in vivo real time recordings of [Gluc]ECF, performed simultaneously in OB (full line) and cortex (Ctx, dotted line) of lean fa/+ **(A, B)** and obese fa/fa rats **(C, D)** in fasted **(A, C: Fast., blue)** and satiated **(B, D Sat., red)** states. In fasted rats, fasted **(E)** and satiated **(F)** rats, mixed ANOVA revealed a significant effect of genotype (in fasted rat # P < 0.01; in satiated rats ### P < 0.0005), conditions (not shown) and brain areas (not shown). SS: steady State; G: glucose; I: insulin.

**first injection of either Glucose or Insulin in OB and cortex**. Bars represent the difference calculated between [Gluc]ECF measured at steady state and after the first injection of glucose **(blue)** or insulin **(red)**. [Gluc]ECF fluctuations were observed in OB **(full bar)** and cortex (Ctx, **dotted bar)** of lean fa/+ **(left)** and obese fa/fa rats **(right)**. When glucose was injected in first in fasted rats (Fast.), a significant effect of the genotype on [Gluc]ECF fluctuation was observed (ANOVA test, # P < 0.05; NS P > 0.05), [Gluc]ECF fluctuation being smaller in the OB of fa/fa compared to fa/+ rats (Mann Whitney, <sup>φ</sup> P < 0.05). No difference was observed in the Ctx. When insulin was injected in first in satiated rats (Sat.), only in fa/+ rats [Gluc]ECF fluctuation was significantly higher in OB than in cortex (Mann Whitney, \* P < 0.05; ns P > 0.05).

induced a greater [Gluc]ECF fluctuation in the OB than in the cortex (**Figure 9**, Mann Whitney tests <sup>∗</sup> *P* < 0.05).

#### **DISCUSSION**

The present study provides new information on the impact of a metabolic disorder on olfactory abilities and glucosesensing in the olfactory system. Indeed, obese *fa/fa* Zucker rats, displayed a higher olfactory sensitivity than lean *fa/*+ rats, regardless of their feeding status (fasted or satiated). In addition, the expression of glucose-sensing markers and glucose fluctuations in the OB were strikingly different between the two rat genotypes. First, SGLT1 expression was higher in the OB of obese rats compared with lean rats; while GLUT4 levels were similar for the two rat genotypes. Second, [Gluc]ECF measured at steady states (fasted or satiated) was higher in *fa/fa* rats than in *fa/*+ rats, with a consistently higher concentration measured in the OB compared to the cortex. Third, OB [Gluc]ECF was differentially affected in *fa/fa* and *fa/*+ rats when glycemia was dynamically modified by peripheral injections of glucose or insulin. Indeed, after glucose injection (acute hyperglycemia), obese rats showed a slight increase of OB [Gluc]ECF (≈0.2 mM), whereas lean rats demonstrated a much higher increase (0.9 mM). Insulin injection (acute hypoglycemia) decreased [Gluc]ECF similarly in the OB of the two rat strains. Differences observed in olfactory detection between obese and lean rats will be discussed in link with the modulation of glucose-sensing and of food intake.

#### **OBESE fa/fa ZUCKER RATS HAVE HIGHER OB INSULIN CONCENTRATION AND HIGHER [GLUC]**ECF **THAN LEAN fa/+ RATS**

In addition to peripheral hyperinsulinemia and hyperglycemia, obese Zucker *fa/fa* rats displayed higher OB insulin levels than their lean counterparts. This is consistent with previous studies that reported that insulin level is highly increased in the cerebrospinal fluid of *fa/fa* compared to *fa/*+ Zucker rats (Stein et al., 1983; Figlewicz et al., 1985). These results provide evidence that brain insulin is derived from circulating insulin and suggest that obese Zucker rats do not have any defect of insulin transport through the blood brain barrier (BBB). Although some early contradictory studies suggested that insulin binding was altered in the brain of obese *fa/fa* Zucker rats (Melnyk, 1987; Wilcox et al., 1989), subsequent reports convincingly demonstrated that insulin receptors (IRs) had similar expression, number, distribution, binding affinity and tyrosine kinase activity in the brain of lean and obese Zucker rats (Livingston et al., 1993; Amessou et al., 2010). Interestingly, rats carrying at least one copy of the *fa* allele (*fa/fa* and *fa/*+) present lower insulin concentration (Baskin et al., 1985) and lower insulin binding (Figlewicz et al., 1985) than nonmutated +/+ rats. Therefore, these deficits are independent of a severe obese phenotype because they are found in obese *fa/fa* and lean *fa/*+ rats. Altogether, these data indicate that changes in insulin signaling are unlikely to account for the differences observed in the present study.

In both OB and cortex, [Gluc]ECF was found to be higher in *fa/fa* than in *fa/*+ Zucker rats. This is consistent with the higher brain glucose uptake observed in obese rats compared to lean controls (Liistro et al., 2010). It is interesting to note that a higher central glucose availability is associated with a lower local cerebral glucose utilization (Doyle et al., 1993), especially in brain areas implicated in the neuroendocrine regulation of food intake and in odor-taste perception, such as the hypothalamus and the OB (Marfaing-Jallat et al., 1992). Moreover, in both rat genotypes, [Gluc]ECF at steady states was found to be higher in the OB than in the cortex and this is consistent with the compartmentalization of [Gluc]ECF according to the brain area studied and to the level of neural activity (McNay and Gold, 1999; McNay et al., 2001). Work from the group of Magistretti has demonstrated that neuronal activity and glucose metabolism are tied together (Magistretti et al., 1999). Indeed, OB presents a higher functional activity than the somatosensory cortex (Yang et al., 1998). Moreover, the OB glomerular layer not only has remarkably high glucose consumption rates (Nawroth et al., 2007) but also possesses very high capillary network density (Chaigneau et al., 2007) combined with a distinct microvasculature (Yang et al., 1998) and a high permeability of the BBB (Ueno et al., 1996). Together these hallmarks suggest that OB not only elicits high-energy demands but could also be a glucose-sensing brain area.

#### **OBESE fa/fa ZUCKER RATS HAVE A HIGHER EXPRESSION OF GLUCOSE-SENSING MOLECULAR MARKERS IN THE OB THAN LEAN fa/+ RATS**

In the OB of both rat genotypes, the molecular markers of glucose-sensing GLUT4 and SGLT1 were detected in different layers of the OB. GLUT4 immunostaining was observed in glomeruli and in some mitral cells, while SGLT1 was mainly observed in the iEPL. This distribution is similar to that shown by *in situ* hybridization in Allen Brain Atlas (Allen Institute for Brain Science, 2010). This distinct regional location of GLUT4 and SGLT1 support the idea that these two families of glucose transporters play different roles in neuronal network processing as it has been previously reported in other brain areas (Yu et al., 2010, 2013).

Obesity did not modulate GLUT4 expression in the OB, since obese *fa/fa* and lean *fa/*+ rats presented similar GLUT4 localization and level of expression. However, obesity changed SGLT1 expression, which was higher in satiated *fa/fa* than in satiated *fa/*+. Concerning GLUT4, our results are in agreement with previous work performed on hyperinsulinemic-hyperglycemic db/db mice, which showed that GLUT4 protein levels were unchanged in the cortex and the OB (Vannucci et al., 1998). Accordingly, GLUT4 mRNA levels did not change in the cortex of hyperinsulinemic *fa/fa* Zucker rats (Alquier et al., 2006). Winocur and collaborators suggested that the stability of GLUT4 expression is related to the insulin resistance of *fa/fa* Zucker rats (Winocur et al., 2005), because GLUT4 translocation to the plasma membrane can no longer be triggered by insulin signaling as it is in a normal metabolic context (McEwen and Reagan, 2004). Indeed, in the hippocampus of Zucker rats, GLUT4 association to the plasma membrane was significantly reduced although total GLUT4 protein expression was not affected (Winocur et al., 2005). In order to confirm this hypothesis, it will be interesting to further study and compare GLUT4 subcellular localization in the OB of the two genotypes of Zucker rats. Concerning SGLT1, the up-regulation observed in the OB of *fa/fa* rats is consistent with previous studies performed on peripheral tissues in obese, hyperinsulinemic and diabetic rodent models as well in human (Morton and Hanson, 1984; Ferraris and Vinnakota, 1995; Dyer et al., 2002; Osswald et al., 2005; Tabatabai et al., 2009). In central nervous system, during pathological conditions such as ischemia or epileptic seizure which induce an over-consumption of glucose (Poppe et al., 1997; Elfeber et al., 2004) SGLT1 was up-regulated (Yu et al., 2010, 2013). Up-regulation of SGLT1 is proposed to compensate for impairment in GLUTs function (Yu et al., 2013). Thus, in the context of insulin resistance, the deficiency in GLUT4 translocation could explain the SGLT1 up-regulation.

#### **FASTED RATS HAVE A BETTER OLFACTORY SENSITIVITY THAN SATIATED RATS, REGARDLESS OF THE GENOTYPE**

Olfactory sensitivity of both *fa/*+ and *fa/fa* Zucker rats is modulated by the feeding states. They have lower olfactory detection and detect the aversive odorant more quickly in the fasted state than in the satiated state, as we previously observed with normalweight Wistar rats (Aimé et al., 2007). Olfactory sensitivity is modulated by a number of peripheral and central signals involved in the regulation of energy balance such as leptin, orexin A, ghrelin and insulin (Julliard et al., 2007; Tong et al., 2011; Aimé et al., 2012; Palouzier-Paulignan et al., 2012). The OB is the target of orexinergic fibers originating from the lateral hypothalamic nucleus (Peyron et al., 1998). In addition, the olfactory system expresses, among others, receptors for orexin A, ghrelin, leptin, insulin, NPY and CCK (for review, see Palouzier-Paulignan et al., 2012). Our results suggest that although many of these signals are altered in the obese Zucker *fa/fa* rats, the redundancy of orexigenic and anorectic molecules acting on the olfactory system can ultimately maintain the modulation of olfactory sensitivity by the feeding state. For instance, we found that the OB insulin content in both *fa/fa* and *fa/*+ Zucker was modulated by the feeding states, with satiated animals showing a higher OB insulin level than fasted animals. Interestingly, we have previously demonstrated in Wistar rats that such fluctuations of OB insulin levels are sufficient to decrease the olfactory sensitivity of fasted animals to the level of satiated ones (Aimé et al., 2012). Changes of insulin levels in OB and in plasma were correlated confirming that the OB is highly sensitive to fluctuations in circulating insulin levels. This result is consistent with the general agreement that OB is the brain region containing the highest level of IRs (Hill et al., 1986; Unger et al., 1989; Marks et al., 1990). This receptor allows pancreatic insulin to enter the brain across brain capillaries (Schwartz et al., 1990; Banks et al., 1997; Woods et al., 2003; Banks, 2004). At the entire brain level, the rate of insulin entrance is regulated by several physiological factors, including the feeding state (Woods et al., 2003; Banks, 2004). When animals are fasted, the ability of insulin to cross the BBB is reduced, leading to a positive correlation between blood and cerebrospinal fluid insulin levels (Strubbe et al., 1988). Although one report suggested that OB insulin binding is modulated by the feeding state and reduced by chronic fasting (Marks and Eastman, 1989), we have recently shown that the OB IR expression is not modulated by the feeding state (Aimé et al., 2012). Consistent with the latter, nutrient availability does not modulate IR levels in brain regions involved in energy homeostasis regulation (Bowlby et al., 1997). Together, these data indicate that insulin signaling in the OB is dependent on the fluctuations of peripheral insulin levels, and participate in the modulation of synaptic transmission of odor-related stimuli by the feeding states.

Interestingly, both genotypes showed similar OB [Gluc]ECF in steady fasted or satiated states although glycemia was much higher in the satiated state. This is consistent with our previous observation in Wistar rats (data not shown) and could be due to adjustments of the glucose transport capacity at the BBB in response to brain metabolic rate and glucose availability (for review see Bradbury, 1993; Leybaert, 2005; Banks, 2006). Three mechanisms have been proposed to explain the modulation of glucose transport across the BBB (see for review Leybaert, 2005) an increase in: (i) the number of GLUT1; (ii) the intrinsic activity of the GLUT1; and/or (iii) the glucose concentration gradient over the barrier, thereby stimulating the driving force for glucose movement. It is possible that in metabolic disorders, these three mechanisms are not altered all together or at the same time. In the present study, the OB [Gluc]ECF stability observed in *fa/fa* rats could be the result of the modulation of either GLUT1 expression or intrinsic activity at the BBB. In future experiments, it will be interesting to study more precisely these different mechanisms of glucose transport through the BBB during the different steps of feeding state in the OB of *fa/fa* and *fa*/+ Zucker rats.

#### **OBESE fa/fa ZUCKER RATS HAVE A HIGHER OLFACTORY SENSITIVITY THAN LEAN fa/+ RATS**

In the present study, Zucker *fa/fa* rats demonstrated a better olfactory sensitivity than Zucker *fa/*+ rats. This finding is supported by several recent studies. Thanos and collaborators demonstrated that *fa/fa* Zucker rats displayed altered brain metabolic responses to food olfactory stimuli in several brain regions (Thanos et al., 2008). The same group also demonstrated that olfactory cues for a high-fat food stimulus elicit heightened behavioral responses in the obese Zucker *fa/fa* rat compared to lean Zucker *fa/*+ rats (Thanos et al., 2013). Interestingly, obese ob/ob mice (unable to produce leptin) and db/db mice (carrying a random mutation in the leptin receptor gene) are able to locate a hidden food reward based on olfactory cues, much faster than control mice. Several metabolic dysfunctions could explain the heightened olfactory perception of obese rats. As previously proposed, leptin is likely to mediate these responses (Thanos et al., 2013). Indeed, *fa/fa* rats are insensitive to leptin, an anorectic signal known to decrease olfactory sensitivity (Julliard et al., 2007; Aimé et al., 2012). The *fa* mutation of the Zucker rat (*fa/fa*) affects the leptin receptor gene and prevents the long form of the receptor (Ob-Rb) from being expressed (Chua et al., 1996). The Ob-Rb form of the receptor is expressed in the brain (Mercer et al., 1996; Fei et al., 1997) and is present at different levels of the olfactory system, including the OB (Shioda et al., 1998; Getchell et al., 2006; Baly et al., 2007; Prud'homme et al., 2009). Previous reports have demonstrated that leptin is able to modulate olfactory behavior as well as neuronal activity (Getchell et al., 2006; Julliard et al., 2007; Prud'homme et al., 2009; Savigner et al., 2009). Indeed, ICV administration of leptin decreases the expression of the activation marker c-fos in mitral and granular OB neurons (Prud'homme et al., 2009). Consistent with these studies, in a previous report, we found that ICV administration of leptin decreases the olfactory sensitivity of fasted animals to the level of satiated ones (Julliard et al., 2007). Here, we demonstrated that Zucker *fa/fa* rat, expressing a non-functional form of the leptin receptor, display a better olfactory sensitivity than the Zucker *fa/*+ rats, consistent with the aforementioned reports showing that leptin is responsible for a decrease in olfactory abilities.

Other metabolic dysfunctions could contribute to increase the olfactory perception of obese rats. The level of orexigenic peptides such as neuropeptide Y, ghrelin and orexin A is upregulated in the hypothalamus of obese Zucker rats (Beck et al., 1990, 2003; Sanacora et al., 1990; Mondal et al., 1999) and ghrelin and orexin A are known to increase olfactory sensitivity (Julliard et al., 2007; Tong et al., 2011). Zucker *fa/fa* rats are also insensitive to the anorectic action of CCK (McLaughlin and Baile, 1980); and the expression level of proopiomelanocortin (POMC), the precursor of the anorectic neuroptide αMSH, is decreased in the arcuate nucleus of *fa/fa* rats (Yamamoto et al., 2002). Although peripheral insulin injection induced similar fluctuations of OB [Gluc]ECF in *fa/fa* and *fa/*+ Zucker rats, peripheral glucose injection affected OB [Gluc]ECF differentially in lean and obese Zucker rats. A

peripheral glucose injection induced a greater increase of OB [Gluc]ECF in the OB of Zucker *fa/*+ rats, compared to Zucker *fa/fa* rats. This result is supported by a recent report that found no significant difference of brain glucose levels in obese Zucker rats after glucose injection compared to fasting brain glucose levels (Liistro et al., 2010). The brain of obese animals is chronically overexposed to glucose (even during fasting), as a consequence, it seems no more able to respond to physiologic glucose fluctuations. Importantly, the alteration of the molecular markers of glucose-sensing (up-regulation of SGLT1 expression and a probable alteration of GLUT4 trafficking) could also participate to the differences of olfactory detection observed between the two genotypes. Indeed, both SGLT1 and GLUT4 have been shown to be involved in neuronal glucose-sensing (McEwen and Reagan, 2004; O'Malley et al., 2006; Yu et al., 2013). This sensitivity for glucose is observed in specialized neurons located in numerous brain regions involved directly or indirectly in homeostasis control (for review see Routh, 2010; Levin et al., 2011), including the OB (Tucker et al., 2010, 2013). In the OB, the mitral cells have recently been demonstrated to be glucose sensitive (Tucker et al., 2010, 2013) therefore glucose signaling could participate to the modulation of the olfactory sensitivity. Glucose-sensing neurons adapt their mean firing rate to the local fluctuations of extracellular glucose levels (González et al., 2008; McCrimmon, 2008). It is well-known that glucose-sensing is altered in the hypothalamus of obese animals (Rowe et al., 1996; Spanswick et al., 1997, 2000), and alteration of glucose-sensing in the olfactory system could play a role in the modulation of olfactory sensitivity observed in *fa/fa* rats. Altogether, Zucker *fa/fa* rats overexpress orexigenic peptides known to increase olfactory sensitivity (Julliard et al., 2007; Tong et al., 2011) and are insensitive to anorectic signals known to decrease olfactory sensitivity (Julliard et al., 2007; Aimé et al., 2012) and they present an altered glucose-sensing in the OB. This imbalance in nutrient sensing, in orexigenic and anorectic signals is likely to account for the increase of olfactory sensitivity demonstrated here by the Zucker *fa/fa* rats. However in the present study, it remains difficult to decipher the respective role of obesity and of glucose-sensing alteration related to hyperglycemia on olfactory sensitivity. In order to go further, it will be intriguing to study olfactory sensitivity and expression of glucose-sensing genes on female Zucker rats because numerous gender-related metabolic differences exist in peripheral (Clark et al., 1983; Corsetti et al., 2000; Gustavsson et al., 2011) as well as in central tissues (Bogacka et al., 2004). Indeed, females develop obesity and insulin resistance but remain normoglycemic (Clark et al., 1983; Corsetti et al., 2000) and express differently glucosesensing genes in hypothalamus compared to males (Bogacka et al., 2004).

### **CONCLUSION**

The whole of these data indicate that obese Zucker *fa/fa* rats demonstrate marked glucose intolerance and an alteration in the expression of glucose-sensing markers in the OB. These newly found impairments, along with the well-described multiple metabolic dysfunctions of obese Zucker *fa/fa* rats modulate the processing of olfactory information and contribute to increase the olfactory sensitivity. Ultimately, changes in olfactory sensitivity participate in the hyperphagia and food-related disorder characteristic of the Zucker *fa/fa* rat.

### **AUTHORS CONTRIBUTIONS**

All experiments were performed in the Centre de Recherche en Neurosciences de Lyon (CRNL). Inserm U1028-CNRS 5292- UCBL1, Team—Olfaction: From Coding to Memory, Lyon, France. Pascaline Aimé and A. Karyn Julliard were responsible for the conception and design of the experiments; Pascaline Aimé, Rita Salem, Caroline Romestaing, Dolly Al Koborssy and Claude Duchamp were responsible for the collection of the data; Pascaline Aimé, Brigitte Palouzier-Paulignan, Rita Salem and A. Karyn Julliard were responsible for data analysis and interpretation; and Samuel Garcia designed the software used in analysis. Pascaline Aimé, Brigitte Palouzier-Paulignan, Claude Duchamp and A. Karyn Julliard drafted the article and revised it critically for important intellectual content. All authors approved the final version of the manuscript.

#### **ACKNOWLEDGMENTS**

This work was supported by Agence Nationale de la Recherche Grant ANR-05-PNRA 1E07 Aromalim, by the Ministère de la recherche et des nouvelles technologies, by the Centre National de la Recherche (CNRS) and the Claude Bernard University of Lyon (University Lyon1). We would like to thank the staff of the Neurochem facility (Université Claude-Bernard-Lyon I, 69373 Lyon Cedex 08, France), Anne Meiller for the engineering of the glucose biosensors.

#### **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: 15 April 2014; accepted: 01 September 2014; published online: 17 September 2014*.

*Citation: Aimé P, Palouzier-Paulignan B, Salem R, Al Koborssy D, Garcia S, Duchamp C, Romestaing C and Julliard AK (2014) Modulation of olfactory sensitivity and glucose-sensing by the feeding state in obese Zucker rats. Front. Behav. Neurosci. 8:326. doi: 10.3389/fnbeh.2014.00326*

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

*Copyright © 2014 Aimé, Palouzier-Paulignan, Salem, Al Koborssy, Garcia, Duchamp, Romestaing and Julliard. 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 orexinergic system influences conditioned odor aversion learning in the rat: a theory on the processes and hypothesis on the circuit involved

### **Barbara Ferry\***

Centre of Research in Neuroscience Lyon, CNRS UMR 5292 - INSERM U1028 UCBL1, Lyon, France

#### **Edited by:**

Regina M. Sullivan, Nathan Kline Institute and NYU School of Medicine, USA

#### **Reviewed by:**

Guillaume Ferreira, Institut National de la Recherche Agronomique (INRA), France Gina Lorena Quirarte, Universidad Nacional Autonoma de Mexico, Mexico Markus Fendt, Otto-von-Guericke University Magdeburg, Germany

#### **\*Correspondence:**

Barbara Ferry, Centre of Research in Neuroscience Lyon, CNRS UMR 5292 - INSERM U1028 UCBL1, 50 Avenue Tony Garnier, 69366 Lyon, France e-mail: bferry@olfac.univ-lyon1.fr

A large variety of behaviors that are essential for animal survival depend on the perception and processing of surrounding smells present in the natural environment. In particular, food-search behavior, which is conditioned by hunger, is directly driven by the perception of odors associated with food, and feeding status modulates olfactory sensitivity. The orexinergic hypothalamic peptide orexin A (OXA), one of the central and peripheral hormones that triggers food intake, has been shown to increase olfactory sensitivity in various experimental conditions including the conditioned odor aversion learning paradigm (COA). COA is an associative task that corresponds to the association between an olfactory conditioned stimulus (CS) and a delayed gastric malaise. Previous studies have shown that this association is formed only if the delay separating the CS presentation from the malaise is short, suggesting that the memory trace of the odor is relatively unstable. To test the selectivity of the OXA system in olfactory sensitivity, a recent study compared the effects of fasting and of central infusion of OXA during the acquisition of COA. Results showed that the increased olfactory sensitivity induced by fasting and by OXA infusion was accompanied by enhanced COA learning performances. In reference to the duration of action of OXA, the present work details the results obtained during the successive COA extinction tests and suggests a hypothesis concerning the role of the OXA component of fasting on the memory processes underlying CS-malaise association during COA. Moreover, referring to previous data in the literature we suggest a functional circuit model where fasting modulates olfactory memory processes through direct and/or indirect activation of particular OXA brain targets including the olfactory bulb, the locus coeruleus (LC) and the amygdala.

**Keywords: associative learning, olfactory memory, orexin, fasting, rat**

### **INTRODUCTION**

A large variety of behaviors that are essential for animal survival depend on the sensory perception and processing of odors present in the natural environment. In particular, food-search behavior, which is conditioned by hunger, is directly driven by the perception of odors associated with food (Le Magnen, 1959) and several studies have demonstrated that odor processing is influenced by the nutritional status of the animal. For example, olfactory system activity was shown to be directly modulated according to hunger and satiation status (Pager et al., 1972; Pager, 1974, 1978; Royet et al., 1983; Apelbaum et al., 2005). Moreover, fasting enhanced odor detection in rats whereas satiety reduced detection of odors in general (Aimé et al., 2007), and more precisely of one odorant specifically associated with the food type involved in the satiation (O'Doherty et al., 2000; Mulligan et al., 2002). Some data suggest that the central nervous system (CNS) regulates food-search behavior by modulating the detection threshold of the food odorant itself through centrifugal innervations (Doucette et al., 2007; Doucette and Restrepo, 2008; Fletcher and Chen, 2010). A large body of data indicates that the hypothalamus plays an important role in this process. Firstly, anatomic characterization of the lateral hypothalamus (LH) has shown the existence of a functional loop between structures involved in the first level of odor detection and the hypothalamus (Peyron et al., 1998; Sakurai, 2005; Swanson et al., 2005; Hahn and Swanson, 2010). Secondly, the crucial role of the basal hypothalamus, and in particular of orexigenic (appetite-stimulating) and anorexigenic (appetiteinhibiting) neurochemicals, in appetite regulation and energy balance has long been established (see Rodgers et al., 2002 for review). Thirdly, among the multitude of neurochemicals found in the hypothalamus, the most recently discovered orexigenic peptides (orexin A, OXA and orexin B, OXB) have been shown to be strongly involved in the regulation of feeding and energy metabolism (see Willie et al., 2001 for review) and the OXA has been involved in olfactory sensitivity. In particular, our team showed that intracerebroventricular (icv) infusion of OXA in rat increased olfactory detection performance in the same way as physiologically induced fasting (Aimé et al., 2007; Julliard et al., 2007). Fourthly, lateral hypothalamic orexin neurons project directly to the olfactory bulb (OB; Peyron et al., 1998; Nambu et al., 1999; Caillol et al., 2003; Shibata et al., 2008) and the two classes of receptors for OXA and for OXB have been found in OB neurons (Caillol et al., 2003; Hardy et al., 2005). In addition, central infusion of OXA increased OB Fos responses to food odor in both fasted and satiated animals (Prud'homme et al., 2009). All these results indicate that the orexin system is involved in the control of feeding behavior by modulating olfactory sensitivity.

It is, however, very unlikely that olfactory sensitivity is completely dissociated from olfactory memory. Rusiniak et al. (1982) and Slotnick et al. (1997) showed that the more intense the odor, the stronger the memory of its association with a reinforcement. Interestingly, other than its projection on the primary olfactory centers, the hypothalamic OXA neurons project to various structures involved in olfactory associative learning (see Rodgers et al., 2002 for review). Moreover, the OXA system was shown to be involved in the memory processes underlying various kinds of learning (Jaeger et al., 2002; Telegdy and Adamik, 2002; Mair and Hembrook, 2008; Di Sebastiano et al., 2011; Sears et al., 2013; Soya et al., 2013). In addition, Touzani and Sclafani (2002) have shown that the lesion of the LH induced a deficit in conditioned flavor aversion paradigms. Therefore, it can be suggested that the OXA system, may influence odor memory formation, directly, by modulating olfactory sensitivity, and indirectly, by activating particular hypothalamus target regions through the modulation of olfactory sensitivity.

In a natural environment, the relevance of the odor coming from a food source encountered by an animal during foodsearch is a crucial key, determining approach and ingestion of the food. Whether the odor of the food is new to the animal or has previously acquired a hedonic valence during a first intake experience will condition either approach or avoidance. Acquisition of hedonic valence by a food item has been shown to result from conditioned learning during which the sensory stimuli characterizing a particular food (odor and taste) become associated with the positive (energy input) or negative (gastric malaise, poisoning) consequences of the ingestion of the food, so that processing the odor and taste stimuli will cue the appropriate approach or avoidance responses (Rescorla, 1988; Holland, 1990; Mackintosh, 1991). These kinds of association have been experimentally studied for years (Slotnick and Katz, 1974; Nigrosh et al., 1975; Slotnick, 1984) and conditioned food aversion paradigms, such as conditioned taste or odor/taste-potentiated odor aversion learning, have provided fundamental insights into the mechanisms and CNS structures involved in food-reward/food-poisoning associations (see Miranda, 2012 for review).

One such paradigm, conditioned odor aversion (COA), is the avoidance of an odorized-tasteless solution (conditioned stimulus, CS) the ingestion of which precedes toxicosis (unconditioned stimulus, US). During COA acquisition, the presentation of the CS is separated from the US administration by a temporal gap that can be of various amplitudes depending on whether the CS is mixed (proximal presentation; Slotnick et al., 1997; Chapuis et al., 2007) or presented close to the solution (Palmerino et al., 1980; Rusiniak et al., 1982; Ferry et al., 1996, 2006). Thus COA is a trace conditioning that has been suggested to result from the association of the memory trace of the CS and the delayed US (see Bures and Buresova, 1990; Roldan and Bures, 1994).

Taking these data together with the fact that (i) the amygdala has been widely involved in the processes underlying the association between the CS and the delayed US during COA (see Miranda, 2012 for review), (ii) OXA neurons project to the amygdala and (iii) hypothalamic OXA neurons are activated by cues associated with consummatory rewards such as food (Harris et al., 2005), it is suggested that the OXA system may play a role in the learning and memory processes linked to higher cognitive aspects of feeding.

In order to test this hypothesis, a recent study has aimed to describe the role of the central OXA system in COA learning (Ferry and Duchamp-Viret, 2014). The results showed that fasting and icv infusion of OXA before the acquisition significantly enhanced COA performances. Moreover in that study, data obtained during the elevated plus maze task showed enhanced anxiety in Fasted but not in OXA infused animals suggesting that the enhancing effect of fasting on COA performances is likely mediated, at least in part, by the OXA component of fasting. In order to further precise which process the OXA system is involved in, the present experiment extends our previous one (Ferry and Duchamp-Viret, 2014) and includes a comparison between the effect of icv OXA infusion and fasting on the extinction of COA learning.

#### **MATERIAL AND METHODS**

The description of the material and methods has been simplified and subjects, surgery and microinfusion procedure are detailed in a previous article (Ferry and Duchamp-Viret, 2014).

Briefly, three groups of animals were acclimated for 7 days to a 23 h 45 min water deprivation schedule. Groups OXA and artificial CSF (aCSF) were constituted by animals implanted in the lateral ventricle that were microinfused with OXA (10 µg dissolved in 3 µl of sterile CSF, Sigma) or aCSF (3 µl, Harvard Apparatus) 20 min before the acquisition of the COA task (on Day 8). A third experimental group (Fasted) was constituted by animals placed in a 24-h food-deprivation schedule before acquisition of COA. Acquisition of COA consisted by the presentation of an olfactory CS (scented water corresponding to isoamylacetate, Sigma-Aldrich France, mixed with tap water at a final concentration of 10−<sup>6</sup> ) followed 20 min after by an injection of 0.15 M lithium chloride-inducing gastric malaise (LiCl, 10 ml/kg; i.p., US). Conditioned aversion to the odor was tested 48 h later (on Day 11) and COA extinction learning took place from Days 12 to 15. All the testing and extinction sessions were conducted under a food-satiated condition in order to prevent any effect of fasting on the olfactory sensitivity between groups (Aimé et al., 2007; Julliard et al., 2007) and consisted of 15-min presentation of the CS (one bottle test). This procedure is based on previous studies (Julliard et al., 2007 for the OXA infusion and fasting Ferry et al., 2007 for the COA procedure). In order to focus our study on the role of OXA system in memory processes underlying COA learning, it is important to note that animals

in our procedure received one OXA infusion before the CS-US

Day 15. The mean water intake measured on the last day of habituation (Day

#### **RESULTS**

pairing and not during the test.

**Figure 1** illustrates the mean scented water intake measured during the acquisition (Day 8), the test (Day 11) and the four successive COA extinction sessions (Days 11 to 15) for the various groups. For comparison purpose, the last water-drinking habituation session (Day 7) has also been represented. As shown by this figure, performances obtained during the test and the number of extinction sessions differed between groups, depending on the treatment.

A two-factor ANOVA with repeated measures with Treatment (aCSF, OXA and Fasted) as between-subjects variable and Session of extinction (Day 11 to Day 15) as within-subject variable revealed significant Treatment (*F*(2,25) = 8.96, *P* < 0.001) and Session effects (*F*(4,100) = 103.85, *P* < 0.001) and a significant Treatment × Session interaction (*F*(8,100) = 5.40, *P* < 0.001).

Between-group analysis confirmed that the OXA and Fasted groups developed a stronger COA that extinguished more slowly than the aCSF group. One-way ANOVA on the mean scented water intake confirmed these observations, with inter-group differences at Day 11, Day 12, Day 13 and Day 14 (*F*(2,25) = 12.86, *P* < 0.001; (*F*(2,25) = 8.51, *P* < 0.01; *F*(2,25) = 9.95, *P* < 0.001 and *F*(2,25) = 6.83, *P* < 0.01 respectively). *Post-hoc* Bonferroni pairwise comparisons revealed significant differences between aCSF and both OXA and Fasted groups at Day 11, Day 12 and Day 13 (from *P* < 0.001 to *P* < 0.05), while the OXA group differed from the aCSF and Fasted groups at Day 14 (*P* < 0.01 and *P* < 0.05 respectively). One-way ANOVA within-group comparisons showed a significant effect of factor Session on COA performances in each of the groups (*F*(4,40) = 9.99; *P* < 0.001; *F*(4.45) = 12.54; *P* < 0.001 and *F*(4,40) = 15.85; *P* < 0.001 for aCSF, OXA and fasted groups respectively). In the aCSF group the extinction curve increased rapidly between D11 and D15. *Post-hoc* Bonferroni's tests confirmed this observation and showed only a significant difference in the variables obtained in D11 and D12 compared to those obtained in D14 and D15 (*P* < 0.05 to 0.001). In the OXA and Fasted group, extinction curves reached a ceiling more slowly than the aCSF group. *Post-hoc* Bonferroni tests confirmed this observation and showed a significant difference in the means solution intake in D11 and D12 compared to D14 and D15 (*P* < 0.05 to 0.001) and a significant difference in the same variable between D13 and D15 (*P* < 0.001 and *P* < 0.01) in the OXA and Fasted groups respectively).

between Fasted and aCSF group. • P < 0.05 between OXA and Fasted group.

These statistical analyses show that scented water (CS) intake increased significantly in each group throughout the sessions, but it has to be borne in mind that the curves combine two distinct if related phenomena: extinction of COA learning and extinction of CS neophobia.

Concerning the neophobia, the figure shows a decrease in the mean solution intake between D7 and D8 in the three experimental groups and the decrease appears to be stronger in the Fasted group. A two-factor ANOVA with repeated measures with Treatment (aCSF, OXA and Fasted) as between-subjects variable and Session (Day 7 versus Day 8) as within-subject variable confirmed this description and revealed significant Treatment (*F*(2,25) = 6.24, *P* < 0.01) and Session effects (*F*(1,25) = 395.3, *P* < 0.001) and a significant Treatment × Session interaction (*F*(2,25) = 27.1, *P* < 0.001). Pairwise intergroup comparisons (oneway ANOVA) indicated no difference between groups on D7 and a significant decrease in mean scented water intake in Fasted group compared to aCSF and OXA groups (one-way ANOVA (2,25) = 28.92; *P* < 0.001, *post-hoc* Bonferroni *P* < 0.001).

Experimental COA extinction reflects not loss of the original memory trace but rather new learning whereby the CS comes to predict no US. It has been suggested that this learning inhibits the previously acquired conditioned response to the olfactory CS (Rescorla and Heth, 1975; Robbins, 1990).

Therefore, and in order to establish the real number of sessions needed to extinguish COA in each group taking the neophobia effect into account, additional analyses were performed considering the mean scented water intake measured during the acquisition in the variables.

One-way ANOVAs performed on these variables revealed a significant effect of factor Session in the aCSF and OXA groups (*F*(5,48) = 8.17; *P* < 0.001; *F*(5,54) = 14.1; *P* < 0.001 respectively). In the aCSF group, *post-hoc* Bonferroni analyses revealed that only mean scented water intake at Day 11 significantly differed from Day 8 (*P* < 0.01), suggesting that COA extinguished in a single session in this group. In the OXA group, *post-hoc* Bonferroni analyses revealed significant differences in the data obtained at Day 8 versus Day 11, Day 12 and Day 13 (*P* < 0.001, *P* < 0.001 and *P* < 0.05 respectively), suggesting that COA extinguished over three sessions in this group. Concerning the Fasted group, the fact that animals were conditioned and tested under two different conditions (food-deprived versus foodsatiated), rendered the comparison between D8 and the following extinction sessions irrelevant due to the fact that the high level of CS neophobia displayed by the Fasted group on D8 may be attributed to stress induced by the fasting condition (see Section Discussion). Therefore, the data obtained in the Fasted group from Days 11 to 15 were compared to those of the aCSF group on Day 8. Here, one-way ANOVA revealed an effect of the session (*F*(5,48) = 16.2; *P* < 0.001) and *post-hoc* Bonferroni tests revealed a significant difference between data measured at D8 versus D11, D12 and D13 (*P* < 0.001, *P* < 0.001 and *P* < 0.05 respectively), suggesting that COA extinguished over three sessions in this group.

Taken together, these data show that COA extinction was significantly faster in the aCSF than in the OXA and Fasted groups. Moreover, the fact that the OXA group differed from the aCSF and Fasted groups at Day 14 might suggest that the process of COA extinction differed between OXA and fasted groups.

### **DISCUSSION**

One of the main results represented on **Figure 1** was that fasting and central OXA infusion induced similar COA enhancement compared to the Control group. The fact that all animals were satiated during the test rules out the possibility that factors such as olfactory hypersensitivity or stress induced by fooddeprivation may have influenced the process of COA extinction and/or retrieval.

As shown in **Figure 1**, the Fasted group displayed a decrease in mean scented solution intake between the last water-drinking habituation session (Day 7) and the acquisition (Day 8), suggesting that, despite the use of a very low concentration of isoamylacetate solution (ISO, 10−<sup>6</sup> ), the acute 24-h food and water deprivation schedule enhanced neophobia for a novel olfactory stimulus.

Neural processing of olfactory information is closely linked to the physiological and nutritional status of the organism, and fasting has been shown to increase OB reactivity (Pager et al., 1972; Apelbaum et al., 2005) and olfactory detection in rats (Aimé et al., 2007). Therefore, the strong neophobia toward the scented water observed in the Fasted group during acquisition may have resulted from the enhanced olfactory sensitivity induced by starvation. Although this starvation effect on olfactory sensitivity has been suggested to be mediated by activation of the central OXA system (Julliard et al., 2007), the very slight difference observed at acquisition (Day 8) between the OXA and aCSF groups suggests that the neophobia toward the CS displayed by the Fasted group at acquisition cannot have been simply and exclusively due to OXA-induced enhancement of olfactory detection under fasting. In this respect, data have shown that short-term (24-h) food deprivation induced a significant increase in serum corticosterone (Das et al., 2005; Johansson et al., 2008; Nowland et al., 2011) and anxiety levels (Ferry and Duchamp-Viret, 2014). Therefore, it can be assumed that the strong neophobia observed in the Fasted group at COA acquisition may have resulted, at least in part, from the combination of enhanced olfactory detection and increased anxiety induced by fasting. Moreover, the fact that OXA infusion did not induce any anxiety (Ferry and Duchamp-Viret, 2014) suggests the strong neophobia observed in the Fasted group at Day 8 was unlikely to have been mediated by central release of OXA.

The results presented in **Figure 1** show that COA extinction differed between groups: while the aCSF group needed one session to extinguish COA, the Fasted and OXA groups needed three. The extinction phenomenon reflects the inhibition of the conditioned response by new learning of CS-no US (Rescorla and Heth, 1975) and it has been suggested that resistance to extinction of a CS-US association is directly dependent on the strength of the CS-US memory trace (Eisenberg et al., 2003). The present results suggest that the fasting and OXA infusion conditions both enhanced CS-US association strength, and the similarity between the Fasted and OXA groups in terms of extinction suggests that the enhancing effect of fasting on CS-US association strength is mediated, at least in part, by a central release of OXA.

Now COA is a trace conditioning that results from several processes that follow one another over time: during acquisition, CS and US processing is followed by association of the two stimuli; then, the CS-US association is consolidated and finally retrieved during the test when the CS is presented for the second time. Some studies have shown that behavioral effects of icv OXA infusion, such as feeding and drinking behavior or olfactory hypersensitivity, persist for at least 3 h (Sakurai et al., 1998; Edwards et al., 1999; Kunii et al., 1999; Julliard et al., 2007). Therefore, the effects of fasting and OXA infusion on COA observed in the present study may have resulted from changes in the processes of acquisition and/or consolidation taking place on Day 8.

### **HYPOTHESIS ON THE PROCESSES BY WHICH FASTING AND OXA MAY HAVE INFLUENCED COA LEARNING**

#### **CS-US Acquisition**

*Olfactory hypersensitivity and olfactory memory trace duration.* As discussed in the Introduction, the acquisition of COA reflects the association between the memory trace of the olfactory CS and the delayed visceral US (see Bures and Buresova, 1990; Roldan and Bures, 1994). Several studies have shown that, when ingested, a tasteless olfactory stimulus (such as ISO) acquires a strong aversive value, even with CS-US intervals equivalent to those generally used for tastes (Rusiniak et al., 1982; Bouton et al., 1986; Slotnick et al., 1997; Chapuis et al., 2007). At a concentration of 10−<sup>4</sup> in tap water, ISO, used as the CS for COA, has been shown to be resistant to a relatively long interval (up to 30 min) before delivery of the US (Chapuis et al., 2007; Ferry et al., 2007; Miranda et al., 2007). Interestingly, results obtained in aCSF group showed that ISO mixed in tap water at a concentration of 10−<sup>6</sup> was able to induce a mild COA that extinguished in one session when the time interval (ISI) separating the CS from the US was about 20 min. Therefore, it is suggested that, in our conditions, the CS trace decayed over the 20-min ISI, at the end of which it is weakly associated to the US.

The effectiveness of an olfactory CS in inducing strong COA when paired with a delayed illness has been shown to be directly related to the intensity of the CS used during acquisition (Rusiniak et al., 1982; Slotnick et al., 1997). Thus, in the light of the increased olfactory sensitivity previously reported with 24-h fasting and icv OXA infusion (Julliard et al., 2007), it is possible that the strong COA obtained in the Fasted and OXA groups was directly linked to an enhanced memory for the CS resulting from the change in olfactory perception. Interestingly, a similar cause-effect relationship was described in a previous experiment by our team, in which entorhinal cortex lesion enhanced COA learning performances, an effect that was accompanied by olfactory hypersensitivity (Ferry et al., 1996). Therefore, by enhancing olfactory sensitivity, fasting and OXA infusion may have enhanced the duration of the olfactory trace by enhancing CS salience, thus rendering possible its association to the delayed toxicosis.

*Higher US processing.* In the same vein, Garcia and Koelling (1967) demonstrated that the strength of a conditioned response to an ingested CS previously paired with a gastric malaise US varies directly with the intensity of the US. Concerning the COA, similar results have been described and the strength of COA varied directly with the intensity of the US used (Rusiniak et al., 1982; Bouton et al., 1986; Slotnick et al., 1997; Chapuis et al., 2007; Ferry et al., 2007). It could therefore be argued that the enhanced COA observed in the fasted group resulted from the visceral discomfort induced by injecting the LiCl used in the COA protocol, being more intense in a 24-h fasted animal, enhancing the strength of the CS-US association. However, some studies have shown that the orexinergic system is independent of the system that deals with visceral processing of the LiCl-induced intoxication in conditioned aversive learning (Touzani and Sclafani, 2002; Di Sebastiano et al., 2011). Therefore, any indirect effect of fasting or OXA infusion on COA strength via a change in US processing may be discounted.

Regardless of this, Winsky-Sommerer et al. (2004) have shown that OXA system is activated by the corticotropin-releasing factor (CRF) that is released in condition of acute stress. Considering these data, it could be suggested that the stress induced by the US administration may have influenced the release of OXA through the activation of the CRF system in both groups of animals. If the fasting condition also enhanced the OXA level by the same route, it could be argued that the similarity in the COA performances obtained in Fasted and OXA groups was due to similar enhanced level of OXA with the different experimental conditions. If so, this would suggest that the stress induced by the US LiCl may have influenced the CS-US association or the processes underlying the CS-US memory. Even though this hypothesis could explain the similarity in the COA performances between Fasted and satiated OXA groups, however future studies will aim at verifying whether an acute restraint stress can be physiologically compared to this induced by a LiCl-induced intoxication and also whether the administration of the US in our conditions can induce OXA release.

#### **CS-US consolidation**

Finally, the temporal evolution of CS-US consolidation remains largely unknown; however, Dudai (1996) and Dudai and Morris (2000) proposed that consolidation involves two types of processes: synaptic consolidation, accomplished within the first minutes to hours after the CS-US association has been acquired; and system consolidation, involving reorganization of the brain circuits encoding the memory, which takes weeks, months or even years to be accomplished. Considering the duration of OXA action (at least 3 h, Sakurai et al., 1998; Edwards et al., 1999; Kunii et al., 1999; Julliard et al., 2007), it may be suggested that the enhanced COA observed in the OXA group was mediated by enhanced synaptic consolidation of the memory processes. According to this view, recent data have shown that central OXA system is involved in the acquisition and in the consolidation of fear conditioned learning (Soya et al., 2013). In order to test the involvement of the OXA system in COA consolidation process, it could be of interest to test the effect of selective blockade of OXA receptors at various times during COA learning. In this way, we assume that if the OXA system is selectively involved in the CS-US acquisition process, infusion of the antagonist before the CS presentation would disrupt both short (3–4 h after acquisition) and long term (24 h) memories. In contrast, if the OXA system is involved in consolidation, the pharmacological blockade of OXA receptors would impair selectively long-term memory leaving short-term memory intact (Sears et al., 2013).

As a first conclusion, the present results show that physiological or OXA-induced fasting affected COA through changes in memory processes occurring during the acquisition of a CS-US association and/or during the synaptic consolidation of this association. The olfactory hypersensitivity induced by fasting may influence acquisition of the CS-US association by enhancing the formation and maintenance of the CS memory trace.

#### **HYPOTHESIS ON THE NEUROBIOLOGICAL SUBSTRATE INVOLVED IN THE EFFECTS OF FASTING AND OXA ON COA**

The neurobiological substrate through which OXA affects the memory processes underlying COA learning remains to be elucidated. However, some reports open up a number of possible hypotheses according to which OXA release during fasting may enhance learning performance through a direct or indirect influence on particular hypothalamic projection targets.

#### **Olfactory bulb (OB) and locus coeruleus (LC)**

As previously mentioned, olfactory sensitivity cannot be dissociated from olfactory memory, and some data suggest that the indirect effects of fasting and OXA on the olfactory memory trace formation through increased olfactory sensitivity may involve the olfactory bulb (OB). As previously mentioned, olfactory sensitivity cannot be dissociated from olfactory memory, and some data suggest that the indirect effects of fasting and OXA on the olfactory memory trace formation through increased olfactory sensitivity may involve the OB. Firstly, in addition to its well documented role in detection and discrimination, the OB is involved in memory processes underlying various kinds of olfactory learning in adult rats (see Mandairon and Linster, 2009 for review). Moreover, the OB receives direct OXA innervation from the LH (de Lecea et al., 1998; Peyron et al., 1998; Sakurai et al., 1998; Shibata et al., 2008) and an increase in OB electrophysiological response induced by fasting and OXA has been described (Pager et al., 1972; Gervais and Pager, 1982; Apelbaum and Chaput, 2003; Apelbaum et al., 2005; Hardy et al., 2005). In addition, Prud'homme et al. (2009) found that OXA antagonist treatment blocked the enhancement of OB Fos responses to a food odor. Secondly, some evidence suggests that the OXA neurons terminating in the locus coeruleus (LC) may provide a second indirect pathway for orexinergic modulation of olfactory processing: direct OXA fibers innervate the LC (Horvath et al., 1999) and activation of OXA receptors in the LC increases cell firing of intrinsic noradrenergic (NA) neurons (Trivedi et al., 1998; Hagan et al., 1999). In addition, the LC projects over 40% of its neurons directly into the OB (McLean et al., 1989) and this large NA input has been shown to modulate OB excitability, olfactory perception and olfactory learning and memory abilities (see Devore and Linster, 2012 for review).

These data suggest that the fasting-induced increase in olfactory sensitivity observed in the present and other studies probably involved the OXA system in the OB. Moreover, and in the light of the work by Escanilla et al. (2012), it may be suggested that this increased olfactory sensitivity resulted directly from OXA system activation (Hardy et al., 2005) in the OB and/or indirectly through the effect of LC-OXA system activation on NA release in the OB. Finally, it is possible that activation of both systems may simultaneously influence the strength of olfactory memory through enhanced olfactory processing. In order to test this hypothesis, future studies will aim at verifying whether the presentation of a new olfactory stimulus in fasted animals can be correlated to changes in NA release in the OB.

#### **Amygdala and locus coeruleus (LC)**

Orexinergic innervation of the extended amygdala (including the basolateral amygdala, BLA) was clearly described by Schmitt et al. (2012). OXA administered into the LH significantly elevated cFosimmunoreactivity in the amygdala (Mullett et al., 2000) and fasting increased OXA mRNA levels in the amygdala (Lu et al., 2000). OXA applied in acute rat brain slices activated neurons in the amygdala (Bisetti et al., 2006). Moreover, the amygdala receives OB and visceral inputs (Saper and Loewy, 1980; Inui et al., 2006), and may be a nodal point at which olfactory and neuroendocrine stimuli are integrated to modulate feeding behavior (King, 2006).

Otherwise, a large amount of data indicates that the amygdala, and more precisely the BLA, is involved in the acquisition of COA (Sevelinges et al., 2009) and more precisely in the processes underlying the formation of the olfactory memory trace and its maintenance across the ISI during COA (Ferry et al., 1995; Ferry and Di Scala, 1997, 2000). Although a direct effect of starvation-induced OXA release in the amygdala on COA cannot be excluded, to our knowledge, the involvement of the OXA system in the amygdala has never been demonstrated in learning and memory.

On the other hand, some data indicate that the enhancing effect of starvation on COA memory processes could be indirectly mediated by activation of the NA system in the amygdala. The LC projects strongly onto the amygdala (Fallon et al., 1978) and the BLA β-adrenergic system is involved in the memory processes underlying the association between odor and delayed US during COA (Miranda et al., 2007). Given the direct action of LH orexinergic neurons on the LC (Horvath et al., 1999; van den Pol et al., 2002), activation of the LC-amygdala NA system during processing of the new odor CS may be potentiated by fastinginduced OXA release. Possibly, the strength of the olfactory memory trace, and/or its association to the US, was influenced by activation of this pathway in the Fasted and OXA groups.

Although the results shown on **Figure 1** do not identify particular structures receiving OXA projections as being involved in the memory processes underlying COA acquisition, all the above-mentioned data lead us to propose a model according to which OXA release during fasting enhances learning performance through a direct or indirect influence on the circuit involved in COA. Our model shown on **Figure 2** comprises some of the structures involved in this circuit including the LC, amygdala and OB, which receive LH-OXA projections. According to this model, the activation of the LH-OXA system induced by fasting reinforces the role of each structure in the circuit by enhancing the neural processes underlying attention and olfactory memory through direct and/or indirect influences.

This model is consistent with the idea of Cleland and Linster (2005) according to which the multiple feedback and feed-forward interactions between olfactory and non-olfactory areas contribute to complex processes, such as filtering and

constructing olfactory representations, and compare these representations to those previously acquired in order to cue an appropriate response to relevant stimuli. The present model also supports the hypothesis that centrifugal modulatory inputs influence olfactory processing and learning mechanisms within the OB (Sullivan et al., 2000; Linster and Cleland, 2002; Yuan et al., 2003), leading us to extend these influences to other key structures involved in attention and olfactory learning and memory (Aston-Jones and Cohen, 2005; Miranda, 2012).

Of course the list of structures included in this model is not exhaustive; involvement of other feedback and feed-forward interactions between these structures and others (e.g., piriform cortex, entorhinal cortex, orbitofrontal cortex, hippocampus, etc.) will have to be considered in order to achieve a more realistic idea of the circuit actually involved in food conditioned learning (see Ferry et al., 2006, 2007; Chapuis et al., 2009; Sahay et al., 2011; Wilson and Sullivan, 2011; Chapuis et al., 2013).

#### **CONCLUSION**

Feeding behavior is part of a complex integrated adaptive system, governed by the brain, in which the processing of metabolic signals reflecting the animal's nutritional state (gastrointestinal distention, blood glucose, feeding peptides such as OXA, etc.) and of olfactory signals indicative of food determines the appropriate response to a food source. However, the differentiation between palatable and unpalatable items that conditions ingestive behavior often depends on previous experience during which the odor of the food acquired (or did not, in the case of a new odor) a hedonic valence after feeding, through CS-US associative learning. By showing that OXA system influenced the processes underlying the CS-US association, or/and consolidation of this association during COA, the present study introduces a new mechanism by which the LH-OXA system may influence the processes that enable animals to learn to select food available in the environment and to adapt their behavior to previous experience through a modulation of complex neural circuit activity. Finally, the OXA system represents a critical link between peripheral energy balance and CNS mechanisms that coordinate olfactory processing and memory, especially in the physiological state of fasting.

#### **ACKNOWLEDGMENTS**

This work was supported by the LABEX CORTEX (ANR-11-LABX-0042) of Université de Lyon, within the program "Investissements d'Avenir" (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR).

#### **REFERENCES**


responses to isoamyl acetate or food odour in rats: roles of orexins and leptin. *Neuroscience* 162, 1287–1298. doi: 10.1016/j.neuroscience.2009.05.043


Yuan, Q., Harley, C. W., and McLean, J. H. (2003). Mitral cell beta1 and 5-HT2A receptor colocalization and cAMP coregulation: a new model of norepinephrine-induced learning in the olfactory bulb. *Learn. Mem.* 10, 5–15. doi: 10.1101/lm.54803

**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: 05 February 2014; accepted: 18 April 2014; published online: 06 May 2014*.

*Citation: Ferry B (2014) The orexinergic system influences conditioned odor aversion learning in the rat: a theory on the processes and hypothesis on the circuit involved. Front. Behav. Neurosci. 8:164. doi: 10.3389/fnbeh.2014.00164*

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

*Copyright © 2014 Ferry. 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*.

## Olfactory tubercle stimulation alters odor preference behavior and recruits forebrain reward and motivational centers

#### **Brynn J. FitzGerald<sup>1</sup> , Kara Richardson<sup>1</sup> and Daniel W. Wesson1,2\***

<sup>1</sup> Department of Neurosciences, Case Western Reserve University, Cleveland, OH, USA

<sup>2</sup> Department of Biology, Case Western Reserve University, Cleveland, OH, USA

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Carmen Agustín-Pavón, Imperial College London, UK Satoshi Ikemoto, National Institutes of Health, USA

#### **\*Correspondence:**

Daniel W. Wesson, Department of Neurosciences, Case Western Reserve University, 2109 Adelbert Avenue, Cleveland, OH 44106, USA e-mail: dww53@case.edu

Rodents show robust behavioral responses to odors, including strong preferences or aversions for certain odors. The neural mechanisms underlying the effects of odors on these behaviors in animals are not well understood. Here, we provide an initial proof-of-concept study into the role of the olfactory tubercle (OT), a structure with known anatomical connectivity with both brain reward and olfactory structures, in regulating odor-motivated behaviors. We implanted c57bl/6 male mice with an ipsilateral bipolar electrode into the OT to administer electric current and thereby yield gross activation of the OT. We confirmed that electrical stimulation of the OT was rewarding, with mice frequently self-administering stimulation on a fixed ratio schedule. In a separate experiment, mice were presented with either fox urine or peanut odors in a three-chamber preference test. In absence of OT stimulation, significant preference for the peanut odor chamber was observed which was abolished in the presence of OT stimulation. Perhaps providing a foundation for this modulation in behavior, we found that OT stimulation significantly increased the number of c-Fos positive neurons in not only the OT, but also in forebrain structures essential to motivated behaviors, including the nucleus accumbens and lateral septum. The present results support the notion that the OT is integral to the display of motivated behavior and possesses the capacity to modulate odor hedonics either by directly altering odor processing or perhaps by indirect actions on brain reward and motivation structures.

**Keywords: olfaction, olfactory cortex, self-stimulation, ventral striatum**

### **INTRODUCTION**

Odors have long been known to possess degrees of attractiveness or aversion (Locke and Grimm, 1949). These hedonics can either be innate from birth or in other cases, conditioned through learning. In both cases, perception of odors on the ends of the hedonic spectrum may elicit robust behavioral reactions. For instance, the bold odor of decaying meat elicits quite a repulsive reaction in humans whereas contrastingly, the sweet smell of freshly baked bread is in most cases pleasant. The neural mechanisms underlying these hedonic-driven behavioral responses are becoming increasingly known (e.g., Sullivan and Leon, 1987; Mennella and Garcia, 2000; Rolls et al., 2003; Stevenson and Repacholi, 2003; Sullivan, 2003; Bensafi et al., 2007; Grabenhorst et al., 2007; Baum, 2009; Doucette et al., 2011; Ferrero et al., 2011; Bensafi et al., 2012; Kass et al., 2013), yet major questions still remain.

Rodents are an excellent model for studying the neurobiological mechanisms of odor hedonics. From birth, and during early postnatal life, rats and mice display robust behavioral responses to odors, especially maternal odors to aid in maternal localization and feeding (Blass and Teicher, 1980; Sullivan, 2003; Logan et al., 2012). Most commonly studied in adult rodents, fearful responses (aversion, freezing/immobility, threat assessment) are reliably observed in response to predator odors (Blanchard et al., 2001; Wallace and Rosen, 2001; Takahashi et al., 2005; Ferrero et al., 2011). Thus rodents must possess a highly sophisticated system for the detection and response to odors.

The control of odor hedonic-driven behaviors likely requires not only a fully functional olfactory system to detect and discriminate the stimulus over background stimuli (for review see Wilson and Mainen, 2006; Wilson and Sullivan, 2011), but also the relay of this information into emotional and reward-related brain structures. The olfactory tubercle (OT) is an olfactory structure residing in the ventral striatum with large amounts of known anatomical connectivity into brain reward structures (Ikemoto, 2007; Wesson and Wilson, 2011). Due to this, we previously predicted that the OT serves a major role in regulating odor hedonics (Wesson and Wilson, 2011). Possible evidence for the regulation of rodent behavioral responses to odors by the OT was provided in a recent study by Agustín-Pavón et al. (2014). In the mentioned study, the authors created lesions containing a portion of the OT and observed that female mice with lesions displayed less attraction to male odors (Agustín-Pavón et al., 2014). This finding raises the interesting possibility that the OT regulates odor-hedonic behaviors either by means of its intrinsic rewarding properties (e.g., Prado-Alcalá and Wise, 1984; Ikemoto, 2003) and/or connectivity with reward and motivated behavior centers.

In the present study, we sought to further explore the role of the OT in odor-guided behaviors and the brain reward system in mice. To manipulate OT activity, we employed focal micro-stimulation of the OT using bipolar electrodes—a wellestablished method to probe principles of both olfactory and reward system activity (Freeman, 1960; Phillips and Mogenson, 1969; Prado-Alcalá and Wise, 1984; Mouly et al., 1985; Mouly and Holley, 1986; Wilson and Sullivan, 1990; Carlezon and Chartoff, 2007). As predicted based upon previous results in rats (Prado-Alcalá and Wise, 1984; Ikemoto, 2003), we found that mice selfadministered current stimulation into the OT. Further, persistent automatic stimulation of the OT altered behavior in a threechamber preference test. Finally, in separate groups of mice, we explored the recruitment of brain reward centers using the immediate early gene *c-Fos*. We found that OT stimulation not only recruited OT neurons focally, but also those of structures known connected to the OT—providing initial mechanistic insights into the likely importance of the OT to odor hedonics.

## **MATERIALS AND METHODS**

### **EXPERIMENTAL SUBJECTS**

Adult male c57bl/6 mice (2–4 months of age), bred and maintained within the Case Western Reserve University School of Medicine animal facility were used. Food and water were available *ad libitum* except during behavioral testing. All experiments were conducted in accordance with the guidelines of the National Institutes of Health and were approved by the Case Western Reserve University's Institutional Animal Care Committee.

#### **CHRONIC STIMULATING ELECTRODE IMPLANTATION SURGERY**

A first cohort of mice (*n* = 8) were initially anesthetized with Isoflurane anesthesia (3.5–3% in 1 L/min O2) before being transferred and mounted into a stereotaxic frame where Isoflurane was further provided (3–1%). Core body temperature was maintained at 38◦C with a hot water-filled heating pad. Upon confirmation of anesthesia depth, the head was shaved, cleaned with betadine and 70% EtOH, and a single injection of lidocaine (0.1 ml of 1% in H2O, S.C.) was administered within the future wound margin. A single incision was made from ∼3 mm posterior of the nose along the midline to lambda and the skull surface cleaned with 3% H2O2. A single craniotomy (1 mm diameter) was created on the skull overlying the site of the OT for implantation of the stimulating electrode. The electrode consisted of 240 µm diameter stainless steel insulated wires (A-M Systems, Carlsborg, WA, USA) twisted together and connected by silver epoxy onto an Omnetics micro PS1 connector (Minneapolis, MN, USA). A micromanipulator was used to lower the bipolar electrode into the craniotomy and further into the site of the OT. The electrodes and plug were then cemented onto the skull by means of dental cement and the wound closed with Vetbond (3 M; St. Paul, MN, USA). Rimadyl (Carprofen, Pfizer animal health, 5 mg/kg, S.C.) was administered immediately following surgery and animals allowed to recover on the heating pad for >4 h. Rimadyl was administered daily for 5 days post-op. Food and water were available *ad libitum* except during behavioral procedures. All animals were singly-housed starting the day of implantation on a 12:12 h (light:dark) schedule with all behavioral procedures occurring during the light phase of the cycle (12:00:18:00 h). At least 5 days of recovery from surgery was allowed prior to any behavioral procedures. Following all behavioral procedures, mice were overdosed with urethane (3 mg/kg, I.P.) and transcardially perfused with 10 ml of 0.9% NaCl followed by 15 ml of 10% formalin and brains removed for histological verification of electrode sites.

#### **ACUTE OLFACTORY TUBERCLE (OT) STIMULATION**

A separate cohort of 21 mice were anesthetized via urethane injection (1.0 mg/kg, I.P.) and mounted on a stereotaxic frame upon a water-filled heating pad (38◦C) for acute OT stimulation. The basic surgical methods follow as described above for the chronic stimulating electrode implantation surgery, but with a few notable differences described herein. A stimulating bipolar electrode (same as described above) was lowered into the site of the OT. The stimulating electrode connector was then connected by a headstage via a motorized commutator to a Cygnus Technology SIU-91 isolated current source (Delaware Water Gap, PA, USA). Stimulated mice received 5 trials of current delivery (200 s train of bimodal, rectangular pulses, 50 ms in pulse width (i.e., 10 Hz), 100 µA in amplitude) at a 1 min inter-stimulus interval. Sham mice simply remained on the heating pad with the electrode in their OT for the same duration of time as the stimulated mice. Following which, the electrode was gently raised out of the brain and the animals transferred onto a heating pad for 90 min prior to transcardial perfusion as described above.

### **SELF-ADMINISTRATION BEHAVIOR TESTING**

The self-stimulation chamber was made of acrylonitrile butadiene styrene (ABS) plastic and consisted of a 150 × 150 mm (W × L) floor bordered by 225 mm tall walls and an open celling. One wall was removable to allow insertion and extraction of the mouse from the testing chamber. Above the chamber was a video camera for recording behavioral events as well as a motorized commutator (Tucker Davis Technologies, Alachua, FL, USA) to allow the mice to be freely mobile but still connected to the stimulation tether. In the center of the chamber floor was a 10 × 25 mm piezo electric foil ("touch pad"; Parallax, Inc., Rocklin, CA, USA) for reception of paw presses. All self-administration testing was performed in a dark room with illumination enough to see the subject's behavior provided by a single dim red light.

Digitization of paw presses and triggering of stimulation via the paw presses occurred by means of a Tucker Davis Technologies recording amplifier (RZ5) running custom code. A threshold was set for triggering of stimulation based upon an average touch pad voltage while mice freely explored the self-stimulation chamber during acclimation. During self-stimulation testing therefore, touch pad contact that crossed the threshold triggered the delivery of current stimulation (0.5 s train of bimodal, rectangular pulses, 50 ms in pulse width (i.e., 10 Hz), 100 µA in amplitude).

On 2 consecutive days mice were connected to the stimulation tether and allowed to freely explore the self-stimulation chamber (with stimulation triggering off) for 30 min of acclimation to the testing apparatus. Following, over the course of the next days, mice were again connected to the stimulation tether and allowed to explore the self-stimulation chamber for behavioral testing wherein contact with the touch pad triggered stimulation. Stimulation occurred on a fixed ratio 1 schedule. Importantly, touch pad contact must have been released in order for the mouse to receive the next stimulation upon contact. Mice were allowed access to stimulation in the self-stimulation chamber for up to 60 min each day throughout which the number of presses and the time of each press event were recorded.

#### **THREE-CHAMBER ODOR PREFERENCE BEHAVIOR TESTING**

For a test of odor preferences, a 600 × 300 × 300 mm (length × width × height) clear acrylic chamber was divided into three equal zones with black markings. Above the chamber was a video camera for recording behavioral events as well as a motorized commutator (Tucker Davis Technologies) to allow the mice to be freely mobile but still connected to the stimulation tether. All testing was performed in a dark room with illumination enough to see the subject's behavior provided by a single dim red light.

A single perforated dark plastic stimulus container (20 mm diameter × 20 mm tall) was placed on each end of the preference chamber for all testing. These stimulus containers were designed to allowing olfactory inspection of their contents but no distinct visual, somatosensory, or gustatory cues. On the first day, the mice were connected to the stimulation tether and allowed to freely explore the preference chamber (with stimulation off) for 30 min of acclimation to the testing chamber and clean odor-less stimulus containers. On the second and third days, the mice were again connected to the stimulation tether but this time the stimulus containers contained either a 1 × 10−<sup>3</sup> dilution of fox urine<sup>1</sup> placed on a cotton ball (100 µl fluid) or 3 g of crushed peanut. We predicted that these two different stimuli would elicit unique investigation behaviors (time spent/investigation) related to their emotional values (Takahashi et al., 2005) and thereby would provide a test as to whether or not stimulation of the OT would impact odor hedonic-related behaviors. The side of the preference chamber containing each stimulus was counterbalanced across all mice. On 1 day per mouse (counterbalanced across mice), current stimulation was provided throughout the entire duration of a daily session (continuous bimodal train, rectangular pulses, 50 ms in pulse width (i.e., 10 Hz), 100 µA in amplitude). On each day the testing lasted 500 s, throughout which the amount of time spent in each zone of the preference chamber and the number of zone crosses were recorded onto video. Videos were scored offline by a single experimenter (K.R.) manually tallying zone crosses (defined by contact of all four paws across the divider line) and cumulative time based upon the video stopwatch.

#### **c-Fos IMMUNOHISTOCHEMISTRY AND QUANTIFICATION**

Alternate 40 µm coronal brain sections were acquired from mice which received the acute OT stimulation paradigm or sham

<sup>1</sup>http://Predatorpee.com

controls using a sliding microtome. ≥5 sections/mouse spanning regions ∼0.8–0.4 mm anterior to bregma were collected and left floating in 0.03% sodium azide in Tris-buffered saline (TBS, pH 7.4) until staining. *c-Fos* immunohistochemistry followed the methods of Kang et al. (2011b). First, the brain slices were rinsed in TBS and then quenched in a solution of 0.3% hydrogen peroxide in methanol. They were rinsed again in TBS, and then in 0.1% TX-100 in TBS. Following the rinses, the sections were blocked in 5.0% NDS in 0.1% TX-100 (Jackson ImmunoResearch, West Grove, PA, USA) for an hour and then incubated overnight at 4◦C in the anti-*c-Fos* primary antibody (1:1000, Calbiochem, EMD Millipore, Billerica, MA, USA). A subset of slices in each run were used as a primary antibody control to ensure specificity of staining. The next day, the sections were rinsed in a diluting buffer, incubated in a secondary antibody (1:600, Jackson ImmunoResearch), rinsed again in 0.1% TX-100 in TBS, and incubated in an Avidin/Biotinylated enzyme Complex kit (Vector Laboratories, Inc., Burlingame, CA, USA). Finally, the sections were rinsed with 0.1% TX-100 in TBS, and incubated in a peroxidase substrate kit with diaminobenzidine (Vector Laboratories) and rinsed with ddH2O. Sections were then transferred onto slides and, after drying, cover slipped with Permount (Fisher Scientific, Pittsburgh, PA, USA).

*a priori* regions for analysis included the OT, ventral pallidum (VP), nucleus accumbens (NAc), lateral septum (LS), and caudate putamen (CPu). These regions were identified using known cytoarchitectural features (Paxinos and Franklin, 2000) and imaged at 20x magnification using a Leica microscope and a 3MP camera. Equal size (200 µm<sup>2</sup> ) bounding boxes were overlaid upon the digital images for cell counting. The location of the bounding boxes were held constant across mice. Any cell bodies which touched the bounding box were excluded from counts. *c-Fos*+ cell bodies were manually identified and counted by a single observed based upon density of the 3,3<sup>0</sup> -diaminobenzidine (DAB) reaction versus background. Sections containing significant damage within the bounding box from the stimulating electrode were excluded from analysis and replaced with non-damaged sections. All steps including sectioning, staining, imaging, and quantification were completed in a group-counterbalanced order by a single experimenter blind to the experimental group of the tissue (B.F.).

#### **ELECTRODE PLACEMENT VERIFICATION**

All stimulation sites were verified by post-mortem histological examinations of slide-mounted 40 µm coronal brain sections stained with a 0.1% cresyl violet solution or in other cases, DAPI (40 ,6-diamidino-2-phenylindole, Life Technologies, City State, USA). We defined an OT stimulation site as successful when the wires terminated within either layers i, ii, and/or iii of the OT (**Figure 1**). Electrode tip locations were verified by multipleobservers (B.F. and D.W.) with reference to a mouse brain atlas (Paxinos and Franklin, 2000). Data associated with sites outside of the OT were entirely excluded from this study.

#### **DATA ANALYSIS**

Behavioral data (time spent in preference zones, # touch pad presses, # zone crossings) were compared between conditions

(stimulation on vs. stimulation off) with an ANOVA. The total number of *c-Fos*+ neurons was compared between conditions (stimulated vs. sham) and hemispheres (ipsilateral vs. contralateral) within brain regions specified also using an ANOVA. Data were analyzed in Origin 8.5 (Northampton, MA) with a significance level of *p* < 0.05. Values are reported as mean ± SEM unless otherwise noted.

### **RESULTS**

We first asked whether OT electrical stimulation is rewarding in mice. To address this, we allowed a cohort of eight mice chronically implanted with bipolar stimulating electrodes into the OT to freely explore the self-stimulation chamber for 2 days wherein contact with a touch pad triggered OT stimulation (see Section Materials and Methods). Following, on the third day, mice were again placed into the self-stimulation chamber and numbers of touch pad presses recorded. Stimulation was allowed *ad libitum* over two blocks of 15 min, separated by a single block of 15 min wherein touch pad contact did not trigger stimulation. During the first 15 min block, mice readily pressed the touch pad (**Figure 2A**). The touch pad contacts were not resultant from random contact by the mice since turning off the stimulus in the middle 15 min block entailed a significant decrease in touch pad presses in all but one mouse (7/8, 87.5%) (**Figures 2A, B**) (*F*(1,12) = 19.576, *p* = 0.0008). Reinstatement of the *ad libitum* reward delivery significantly restored touch pad presses in the final 15 min block (**Figures 2A, B**) (*F*(1,12) = 13.786, *p* = 0.003). These data demonstrate that electrical stimulation of the OT is rewarding in mice.

We next investigated the influence of OT stimulation on odor preference behavior in the three-chamber preference test in the same mice. Notably, one mouse (**Figure 2A**, dashed line) displayed highly aberrant behavior from the group in the self-stimulation testing and was thus excluded from this preference experiment. A separate mouse did not explore the preference chamber and instead stayed immobile in a single zone—qualifying exclusion. Across the remaining six mice, all spent statistically similar time in both end zones in the presence of blank odor vials (*F*(1,10) = 0.035, *p* = 0.855) (data not shown). This baseline data verifies that mice did not have a preference towards simply being on one side of the chamber vs. the other. Next, over 2 days, mice were tested for preference behavior among odorized zones, with OT stimulation being provided consecutively on one of those days. Mice received the stimulation on counterbalanced days (1/2 mice received it on day 1 and 1/2 on day 2), and thus in this design effects of OT stimulation on preference behavior at the population level can be considered independent of learning. No differences between these two groups were observed and thus their data were pooled together (*p* = 0.446, Kolmogorov-Smirnov test). We found that in the absence of OT stimulation, mice spent significantly greater time in the peanut zone of the chamber vs. either the neutral (*F*(1,10) = 11.901, *p* = 0.0062) or the fox urine zones (*F*(1,10) = 11.882, *p* = 0.0063) (**Figure 3**). OT stimulation strikingly abolished this difference, with statistically indistinguishable time spent in both the fox and peanut zones (*F*(1,10) = 0.004, *p* = 0.949) (**Figure 3**). No effect of OT stimulation was observed on the number of side crossings in the preference apparatus (*F*(1,10) = 0.235, *p* = 0.639) (data not shown), suggesting that the effect of OT stimulation on the display of behavior in the odor preference task was independent of stimulation influencing gross locomotor activity.

Based upon the above behavioral findings demonstrating that OT stimulation influences odor-driven behavioral responses, we next sought to test the mechanisms whereby OT stimulation may alter hedonic-related behaviors. We predicted based upon known anatomical connectivity between the OT and brain reward structures (Ikemoto, 2007), that OT stimulation recruits forebrain structures necessary for reward (Koob and Le Moal, 2001; Berridge, 2003; Ikemoto, 2007). Therefore, a separate cohort of anesthetized mice received OT stimulation (*n* = 13) or sham OT stimulation (*n* = 8) in a paradigm mimicking that received while awake (see Section Materials and Methods) and later, their brains probed by means of immunohistochemistry for levels of *c-Fos* expression (Sagar et al., 1988; **Figure 4A**). The use of anesthetized

mice for this analysis was advantageous to ensure changes in *c-Fos* expression were directly due to OT stimulation, vs. extraneous influences of OT stimulation upon behavior.

Confirming the physiological potency of the stimulation paradigm, we found that OT stimulated mice had a significantly greater number of *c-Fos*+ neurons in their ipsilateral OT than sham treated mice (*F*(1,19) = 4.679, *p* = 0.0435) (**Figure 4B**). No group effect was observed when comparing between contralateral OT hemispheres (*F*(1,19) = 1.157, *p* = 0.296) (**Figure 4B**). Looking beyond the OT, we found that mice receiving OT stimulation had a significantly greater number of *c-Fos*+ neurons compared to sham mice in their LS (*F*(1,19) = 4.852, *p* = 0.040) and NAc (*F*(1,19) = 4.828, *p* = 0.0406), but not their CPu (*F*(1,19) = 0.032, *p* = 0.861) nor VP (*F*(1,19) = 1.793, *p* = 0.196) (ipsilateral vs. ipsilateral) (**Figure 4C**). Across all of these structures, only in the CPu did the number of *c-Fos*+ neurons in the contralateral hemisphere significantly differ between stimulated and sham groups (*F*(1,19) = 6.09, *p* = 0.023) (**Figure 4C**). No group effect of stimulation was observed between the contralateral hemispheres in the LS (*F*(1,19) = 0.943, *p* = 0.344), NAc (*F*(1,19) = 0.931, *p* = 0.347), or the VP (*F*(1,19) = 2.179, *p* = 0.156). In all structures analyzed, the number of *c-Fos*+ neurons was statistically similar comparing between the ipsilateral to contralateral hemispheres (*p* > 0.05). For a reason we are unaware of, perhaps related to damage of near-by electrode insertion, sham mice displayed significant increases in the number of *c-Fos*+ neurons between their contralateral and ipsilateral hemispheres in both the CPu (*F*(1,14) = 15.067, *p* = 0.0017) and NAc (*F*(1,14) = 12.684, *p* = 0.0031). Taken together, these results suggest that the influence of OT stimulation on motivated and odor-driven behaviors may occur via connectivity between the OT and the NAc, LS, and possibly CPu.

#### **DISCUSSION**

In this study we provide an initial proof-of-concept exploration into the role of the OT, a structure with known anatomical connectivity with both brain reward and olfactory structures, in regulating odor-motivated behaviors and their possible mechanisms. We confirm that electrical stimulation of the OT was rewarding (Prado-Alcalá and Wise, 1984), in this case in mice, and possessed the capacity to alter odor-directed preference behaviors. In separate experiments we also found that OT stimulation significantly increased the number of *c-Fos* positive neurons in not only the OT, but also in forebrain structures essential to motivated behaviors, including the NAc and LS. The present results support the notion that the OT is integral to motivated behaviors and likely involved in odor hedonics.

**FIGURE 3 | OT stimulation alters odor-guided behaviors. (A)** Mice were allowed to explore a three-chamber odor preference chamber for 10 min with one side containing crushed peanuts (3 g) and the other fox urine (1:10 dilution) both contained within a perforated dark plastic container. On counterbalanced days, current stimulation was delivered throughout the entire duration of the session. On the day with stimulation off, mice spent significantly greater time in the peanut zone, whereas this was abolished with OT stimulation on. n = 6 mice. n.s. = not significant. p-values = ANOVA followed by Fisher's PLSD.

#### **INFLUENCE OF THE OLFACTORY TUBERCLE (OT) ON REWARD AND ODOR-GUIDED BEHAVIORS**

In the present study we found that mice self-administered electrical current into the OT. This finding is analogous and complimentary to a much earlier finding by Prado-Alcalá and Wise (1984) also one by Ikemoto demonstrating that rats readily selfadminister cocaine into the OT and further that cocaine infusions into the OT are sufficient for the development of a conditioned place preference (Ikemoto, 2003). Our self-administration experiments in the present study were not designed to be interpreted as novel in theory by any means but instead to reinforce the concept that the OT is instrumental in driving reward-related behaviors (Koob et al., 1978; Prado-Alcalá and Wise, 1984; Alheid and Heimer, 1988; Heimer, 2003; Ikemoto, 2003). Reward system projections into the OT include the rostral linear nucleus of the ventral tegmental area (Del-Fava et al., 2007), the medial forebrain bundle (Gaykema et al., 1990), the NAc (Zahm and Heimer, 1993), and the substantia nigra (Fallon et al., 1978). Thus, possibly due to both intrinsic (DAergic receptor expression (Li and Kuzhikandathil, 2012)) and extrinsic factors (inter-network connectivity), the OT appears capable of eliciting motivated behaviors. Exploring possible mechanisms of connectivity between the OT and the LS and CPu, as suggested herein, will be important in furthering our understanding of the OTs' rewarding properties.

In addition to being connected with reward-related structures, the OT also receives dense innervation from secondary neurons in the olfactory bulb (e.g., White, 1965; Scott et al., 1980; Schwob and Price, 1984; Nagayama et al., 2010; Kang et al., 2011a) and consequently represents and processes odors in a manner similar to the primary (piriform) olfactory cortex (Payton et al., 2012). A recent study reported that lesions of the anteriomedial ventral striato-pallidum complex (including the OT) altered preferences of female mice for male sociosexual odors (Agustín-Pavón et al., 2014). This finding posed the important question as to whether or not manipulation of OT activity alone might be sufficient to alter odor guided behavior? Our present results support and extend the previous finding by Agustín-Pavón et al. (2014) by demonstrating that electrical manipulation of OT activity, specifically, possesses the capacity to modulate odor-guided behaviors. The use of electrical stimulation to modulate and explore principles of olfactory system function and olfactory behaviors is well established (Mouly et al., 1985; Mouly and Holley, 1986; Wilson and Sullivan, 1990) and thus we employed that method herein for bulk modulation of OT activity. Using this, we found that stimulation of the OT abolished preference for peanut odor vs. fox urine odor. Notably, using this gross odor preference behavior to test the OTs involvement in olfaction is a considerably insensitive assay and thus additional studies will be needed employing more precise olfactory read-outs to explore unique and specific aspects of olfactory behavior under control by the OT. Indeed, as they stand, our results demonstrate the role for the OT in modulating odor-guided behaviors, but not specifically modulating olfactory perception. That said, our results from the odor preference assay do support the idea that OT local and/or inter-regional processing modulates odor-guided behaviors if not at least in an indirect manner. This indirect mechanism could be by shaping motivation and/or the reward features of the odors (e.g., via modulation of structures identified in **Figure 4**). Directly, OT stimulation might perturb basic aspects of odor processing known to occur in the OT (Wesson and Wilson, 2010; Payton et al., 2012; Carlson et al., 2014) and thus alter odor perception more specifically. We predict that the effects observed in the present paper stem form a combination of both direct and indirect impacts of OT activity on behavior.

#### **POSSIBLE NEURAL SUBSTRATES WHEREBY OLFACTORY TUBERCLE (OT) AFFECTS BEHAVIOR**

In the present study, we found that moderate electrical stimulation of the OT resulted in significant recruitment of neurons within brain structures believed essential to motivated behaviors (Koob and Le Moal, 2001; Berridge, 2003; Ikemoto, 2007), including the NAc and LS. Notably, this list is not exhaustive in that we did not analyze every possible motivated behavior center, but instead approached this study with an *a priori* set of forebrain regions. Likely analyzing additional brain structures, especially the ventral tegmental area and orbitofrontal cortex, would provide additional evidence for the functional interconnectedness of the OT (Barbas, 1993; Illig, 2005; Del-Fava et al., 2007). Additionally, the medial vs. lateral aspects of the OT are hypothesized differentially involved in olfactory and reward-related behaviors (Josephson et al., 1997; Ikemoto, 2007) and thus, it is possible that our grouping of OT stimulation sites spanning the entire OT overlooked more subtle aspects of neuronal output which otherwise might be observed if one were to employ regionallyrestricted stimulation.

It is interesting to speculate upon why OT stimulation did not more strongly activate structures like the VP which also hold strong interconnectedness with the OT (Millhouse, 1986). While the number of *c-Fos*+ neurons was increased

bar = 20 µm. **(B)** OT stimulation elicits a significant increase in c-Fos+ neurons in the ipsilateral OT. c-Fos+ neuron counts were quantified within a

quantified within a 200 µm<sup>2</sup> bounding box. p-values = ANOVA followed by Fisher's PLSD.

in this structure, this was not significant. It is possible that in the majority of animals bulk OT stimulation was insufficient to recruit distinct sub-populations of neurons innervating the VP and thus VP activation was minimal. Indeed, populations of VP projecting neurons from the OT are GABAergic (Meyer et al., 1989; Hsieh and Puche, 2013) and thus the effects of OT stimulation on the VP would be inhibitory. While the present study yields novel evidence upon the effects of gross activation of the OT, employing more precise stimulation methods, including genetically guided cell-specific methods, will be needed to more closely address mechanisms of OT connectivity.

### **ACKNOWLEDGMENTS**

This work was supported by NSF grant IOS-1121471 to Daniel W. Wesson and a Battelle Science award from Denison University to Brynn J. FitzGerald.

#### **REFERENCES**


Zahm, D. S., and Heimer, L. (1993). Specificity in the efferent projections of the nucleus accumbens in the rat: comparison of the rostral pole projection patterns with those of the core and shell. *J. Comp. Neurol.* 327, 220–232. doi: 10.1002/cne. 903270205

**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 January 2014; accepted: 24 February 2014; published online: 14 March 2014.*

*Citation: FitzGerald BJ, Richardson K and Wesson DW (2014) Olfactory tubercle stimulation alters odor preference behavior and recruits forebrain reward and motivational centers. Front. Behav. Neurosci. 8:81. doi: 10.3389/fnbeh.2014.00081*

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

*Copyright © 2014 FitzGerald, Richardson and Wesson. 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.*

## Interactions with the young down-regulate adult olfactory neurogenesis and enhance the maturation of olfactory neuroblasts in sheep mothers

#### **Maïna Brus 1,2,3,4 , Maryse Meurisse1,2,3,4 , Matthieu Keller 1,2,3,4 and Frédéric Lévy1,2,3,4\***

1 INRA, UMR 85, Physiologie de la Reproduction et des Comportements, Nouzilly, France

<sup>2</sup> CNRS, UMR 7247, Nouzilly, France

<sup>3</sup> Université François Rabelais, Tours, France

4 IFCE, Nouzilly, France

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Muriel Koehl, French Institute of Health and Medical Research, France Grace Schenatto Pereira, Universidade Federal de Minas Gerais, Brazil

#### **\*Correspondence:**

Frédéric Lévy, INRA, UMR 85, Physiologie de la Reproduction et des Comportements, F-37380 Nouzilly, France e-mail: frederic.levy@tours.inra.fr

New neurons are continuously added in the dentate gyrus (DG) and the olfactory bulb of mammalian brain. While numerous environmental factors controlling survival of newborn neurons have been extensively studied, regulation by social interactions is less documented. We addressed this question by investigating the influence of parturition and interactions with the young on neurogenesis in sheep mothers. Using Bromodeoxyuridine, a marker of cell division, in combination with markers of neuronal maturation, the percentage of neuroblasts and new mature neurons in the olfactory bulb and the DG was compared between groups of parturient ewes which could interact or not with their lamb, and virgins. In addition, a morphological analysis was performed by measuring the dendritic arbor of neuroblasts in both structures. We showed that the postpartum period was associated with a decrease in olfactory and hippocampal adult neurogenesis. In the olfactory bulb, the suppressive effect on neuroblasts was dependent on interactions with the young whereas in the DG the decrease in new mature neurons was associated with parturition. In addition, dendritic length and number of nodes of neuroblasts were significantly enhanced by interactions with the lamb in the olfactory bulb but not in the DG. Because interactions with the young involved learning of the olfactory signature of the lamb, we hypothesize that this learning is associated with a down-regulation in olfactory neurogenesis and an enhancement of olfactory neuroblast maturation. Our assumption is that fewer new neurons decrease cell competition in the olfactory bulb and enhance maturation of those new neurons selected to participate in the learning of the young odor.

**Keywords: olfactory bulb, hippocampus, maternal behavior, olfactory learning, brain plasticity**

### **INTRODUCTION**

In most mammals, newborn neurons are continuously provided in two main structures of the brain throughout life, the dentate gyrus (DG) of the hippocampus and the olfactory bulb (Ming and Song, 2005; Curtis et al., 2007; Brus et al., 2013). In the DG, stem cells reside in the subgranular zone (SGZ) and give rise to neuroblasts which become new granule neurons in the overlying granule cell layer (GCL; Ming and Song, 2005). In the olfactory system, neural stem cells function as primary precursors in the subventricular zone (SVZ) located on the wall of the lateral ventricles. These cells produce transient amplifying cells which rapidly divide to produce neuroblasts. The neuroblasts migrate toward the olfactory bulb along the rostral migratory stream. After reaching the olfactory bulb, the new cells migrate radially and mature into granular interneurons for the majority of them (Rochefort and Lledo, 2005).

In rodents, a growing body of literature aiming to block neurogenesis by using various methodologies shows the role of hippocampal neurogenesis in some forms of hippocampus– dependent learning tasks despite some inconsistent results (Deng et al., 2010; Arruda-Carvalho et al., 2011; Gu et al., 2012). Similarly, olfactory neurogenesis is involved in the memory processing of olfactory cues (Lazarini and Lledo, 2011). Blocking neurogenesis impaired olfactory perceptual learning (Moreno et al., 2009), short-term and long-term memory (Breton-Provencher et al., 2009; Lazarini et al., 2009; Sultan et al., 2010). In addition it has been recently demonstrated that immediate activation of newborn olfactory neurons, by using an optogenetic approach, enhances discrimination learning and memory when the task is difficult (Alonso et al., 2012). A common role in pattern separation has been proposed for newborn hippocampal and olfactory neurons and adult neurogenesis could constitute an adaptive mechanism to optimally encode contextual or olfactory information (for review, see Sahay et al., 2011).

While the implication of adult neurogenesis has been demonstrated in spatial and olfactory learning, a better understanding of the role of adult born neurons in an ethological context has begun to emerge. Social environment can modify hippocampal neurogenesis. Decrease in cell proliferation induced by social isolation rearing could be reversed by subsequent group rearing (Lu et al., 2003). Exposure to chronic social stress dramatically decreases cell proliferation in the DG of rats, mice and tree shrews (Gould et al., 1997; Czeh et al., 2002; Mitra et al., 2006). In a socio-sexual context, exposure to the male urine compounds that are involved in mate recognition increased the survival of granule cells in the accessory olfactory bulb, the main olfactory bulb (MOB), and the DG of female mice (Mak et al., 2007; Oboti et al., 2011). Moreover, suppression of neurogenesis by an anti-mitotic agent prevented mate recognition (Oboti et al., 2011) and the display of preference for dominant male in female mice (Mak et al., 2007). Another approach to assess the role of adult neurogenesis consists in evaluating whether newly-generated neurons might functionally integrate the olfactory network which process olfactory information. In male hamsters, double immunohistochemistry labeling for Fos, a marker of cell activation, and neuronal nuclei (NeuN), a marker of post-mitotic neurons, shows that olfactory bulb cells born in adulthood are activated by socio-sexual stimuli such as estrous female or aggressive male (Huang and Bittman, 2002).

A link between adult neurogenesis and parenting has not been clearly established yet (Lévy et al., 2011). Some studies indicate a regulation of neurogenesis by parturition and the onset of motherhood. In all of rodent species studied so far parturition and the early *postpartum* period are accompanied by a significant decrease in cell proliferation in the hippocampus. In primiparous mother rats, this was reported at *postpartum* day 1, 2 and 8 (Darnaudery et al., 2007; Leuner et al., 2007; Pawluski and Galea, 2007) although no effect was observed later, at *postpartum* day 28 and after weaning (Leuner et al., 2007). Parturition and early *postpartum* period do not stimulate cell proliferation in the SVZ of mice but an increase is observed at 7 days *postpartum* (Shingo et al., 2003). Surprisingly, whether cell survival in the DG or in the MOB is altered during parturition and early *postpartum* period at the onset of maternal behavior is not known in rodents. Rather, cell survival in the DG of rats was assessed either at *postpartum* day 14 (Darnaudery et al., 2007) or 21 (Pawluski and Galea, 2007) and both studies report a significant decrease when compared to virgins. A few studies have investigated the importance of stimuli provided by neonatal pups but outside the context of parturition. Nulliparous rats exposed to pups show increased cell proliferation in the DG when compared to nulliparous females regardless of their parental response (Pawluski and Galea, 2007). Likewise, virgin female prairie voles exposed to pups exhibit increased hippocampal cell proliferation (Ruscio et al., 2008). Although an increase in cell survival in the DG was reported in virgin females 21 days after pup-exposure (Pawluski and Galea, 2007), the influence on survival of the newborn neurons either in the DG or in the MOB at the time of pup-exposure is not known. The consequences of neurogenesis ablation on the onset of maternal behavior have been investigated in mice. Irradiation of the SVZ induces minor disturbances of maternal behavior (Feierstein et al., 2010b). However, infusion of an anti-mitotic agent which transiently impairs both hippocampal and olfactory neurogenesis has been shown to affect maternal behavior but only when animals are tested in an anxiogenic environment (Larsen and Grattan, 2010), whereas genetic manipulations inducing profound and long-term alterations of neurogenesis impair nursing behavior in the home cage (Sakamoto et al., 2011).

In sheep, a down-regulation of cell proliferation has been observed in mothers in contact with their lambs for 2 days both in the DG and the SVZ (Brus et al., 2010). However there has been no report examining a change in survival of newly-born neurons in the DG or the MOB that could occur during the early *postpartum* period. In addition, no study has disentangled the influence of parturition and the first interactions with the young on cell survival and this could improve our understanding of the contribution of neurogenesis to maternal behavior. In this context, maternal behavior in sheep constitutes an interesting model in which endocrine changes occurring at parturition and olfaction play a central role (Lévy et al., 1995; Lévy and Keller, 2008). In addition, not only infantile odors become very potent stimuli allowing the development of maternal care but they also provide a basis for individual recognition of the offspring. Ewes develop discriminative maternal care, called maternal selectivity, favoring their own young at suckling while rejecting any alien young. This recognition is based on the learning of olfactory characteristics of the lamb and takes place within the first hours after parturition (Lévy et al., 2004; Lévy and Keller, 2009). Some of the learning mechanisms reside in extensive neurochemical changes occurring in the MOB at parturition (Lévy et al., 1993; Lévy and Keller, 2009). In addition to these neurochemical changes, olfactory neurogenesis could provide another mechanism through which olfaction can contribute to the onset of maternal behavior and associated learning.

The aim of the present study was to evaluate the influence of parturition and learning of the lamb odor on the survival of newborn neurons. Bromodeoxyuridine (BrdU), a marker of cell division, was used in combination with two markers of neuronal maturation (doublecortin (DCX), an early maturation marker and NeuN), to compare both hippocampal and olfactory neurogenesis between virgins and parturient ewes which could interact or not with their lamb. In addition, because learning accelerates the maturation of the dendritic trees of newborn neurons in the DG (Tronel et al., 2010; Lemaire et al., 2012), and motherhood is accompanied by changes in the morphology of newborn neurons in the MOB (Kopel et al., 2012), we assessed the influence of lamb olfactory learning on this maturation by measuring the dendritic length and the number of nodes of new neuroblasts.

### **MATERIALS AND METHODS**

#### **ANIMALS**

Experiment was conducted on 17 Ile de France ewes, of 1.5–2 years of age, from the INRA research center in Nouzilly (Indre et Loire, France) approved by local authority (agreement number E37-175-2). Animals were permanently housed indoors, with free access to water and were fed with lucerne, maize, straw and a supplement of vitamins and minerals. Animal care and experimental treatments complied with the guidelines of the French Ministry of Agriculture for animal experimentation and European regulations on animal experimentation (86/609/EEC) and were performed in accordance with the local animal regulation (authorization No. 006352 of the French Ministry of Agriculture in accordance with EEC directive). Ewes were sacrificed by a licensed butcher in an official slaughterhouse (authorization No. A37801 E37-175-2 agreement UEPAO). All efforts were made to minimize the number of animals (5–6 animals per group).

#### **BROMODEOXYURIDINE INJECTIONS AND TISSUE PREPARATION**

Four months before sacrifice, ewes were housed in an individual pen (2 × 1 m) and received four intravenous injections of BrdU, (1 injection/day, 20 mg/Kg in 0.9% saline; Sigma-Aldrich, France), a thymidine analogue incorporated into the DNA during the S-phase of the mitotic division. Doses of BrdU and timing between injections and sacrifice were based on a previous study reporting that maturation of adult-born cells both is much longer in sheep than that of rodents and that the highest proportion of new mature neurons is found at 4 months after BrdU injections (Brus et al., 2013).

### **GROUPS**

Three groups of ewes were constituted (**Figure 1**). In parturient groups, mating was synchronized by the use of vaginal sponges containing 45 mg of fluorogestone acetate for 14 days followed by an intra-muscular injection of pregnant-mare-stimulating gonadotropins to induce ovulation. Just after parturition, mothers were either left 48 h with their lambs in their individual pen in the same barn ("With Lamb" group, *n* = 5), or were separated from them for 48 h ("No Lamb" group, *n* = 6). After being separated from their lamb immediately after birth, ewes of the "No Lamb" group were placed in a different barn to avoid any contact with lambs and ewes were housed together in a large pen to avoid stress induced by separation from the young. All the ewes had never given birth before the study. Lambing occurred within a period of gestation of 149 ± 4 days. Against all expectations, at birth the lambs displayed low vigor preventing them from feeding normally, probably due to BrdU injections in early pregnancy. Thus, in the "With lamb" group, adoptions have been performed with newborn lambs provided by the flock of the research center. It has been well established in previous studies that adoption, when performed at birth, are without any consequences on the quality of the mother-young relationship in comparison to normal mother-young lambing (Keverne et al., 1983; Kendrick et al., 1991; Lévy et al., 2010). In this group, maternal behavior was observed for 10 min at 0, 6 and 24 h after parturition to completely ensure that maternal care was normally provided to lambs. At 2 days *postpartum* just before sacrifice, selectivity was tested by presenting an alien lamb to the mother and rejection and acceptance behaviors were recorded for 3 min. The alien lamb was then taken away and the ewe was observed

with her own lamb for an additional 3 min (Keller et al., 2004, 2005). These tests indicated that all the ewes of the "With Lamb" group were maternal and selective. The "Virgin" group (*n* = 6) was composed of nulliparous anoestrus ewes of similar age than the two parturient groups and housed together.

#### **BRAIN PERFUSION AND IMMUNOHISTOCHEMISTRY**

Two days after lambing, ewes were anesthetized with thiopental and decapitated by a licensed butcher in an official slaughterhouse. Brains were immediately perfused via carotid arteries with 2 L of 1% sodium nitrite in phosphate buffer saline, followed by 4 L of ice-cold 4% paraformaldehyde solution in 0.1M phosphate buffer (pH 7.4). The brain was dissected out, cut into blocks and post-fixed in the same fixative for 48 h. The tissues were then stored in 30% sucrose for 2 days until sectioned. Frontal sections were cut at a thickness of 30 µm using microtome or cryostat and stored at −20◦C in cryoprotectant.

To reveal BrdU positive cells in the MOB and the DG, a peroxydase single-immunolabeling was used as described previously (Brus et al., 2013). To characterize the populations of BrdU positive cells which could be affected by parturition and interaction with the young, double immunolabeling was performed against the BrdU and two markers of neuronal maturation, NeuN for mature neurons (Valley et al., 2009) and the DCX for neuroblasts (Gleeson et al., 1999; Brown et al., 2003). Sections of the MOB and the hippocampus were treated with a solution of Tris Buffer Saline 0.025M (TBS, pH=7.4)-Triton 0.3%-Azide 0.1%-Bovine Serum Albumin (BSA) 0.1% (TBSTA-BSA) for 1 h. After one rinse in TBS, sections were treated with 2N HCl in TBS for 30 min at room temperature. After three rinses in TBS, sections were incubated overnight in primary rat anti-BrdU (1:300; AbCys AbC117-7513, Paris, France) and primary mouse anti-NeuN (1:1000, Chemicon MAB377, Millipore, St. Quentin-en-Yvelines, France) or primary goat anti-DCX (1:100, Santa Cruz Biotechnology, Tebu-Bio, Le Perray en Yvelines, France) in TBSTA-BSA. The following day, sections were rinsed three times in TBS, and were incubated in two secondary antibodies simultaneously for 1 h 30 min in TBS 0.025M, pH7.4—Saponine 0.3%—BSA 0.1% (TBS-Saponine-BSA), except for BrdU/DCX for which secondary antibodies were incubated in TBS-rabbit serum 1%-saponine 0.3%. Secondary antibodies used for BrdU/NeuN immunolabeling were a donkey anti-rat CY3 (1:300, Immunotech, Jackson ImmunoResearch, United Kingdom), and a goat anti-mouse 488 (1:300, AlexaFluor A11029, Molecular Probes, Eugene, Oregon, USA); for BrdU/DCX, we used a rabbit anti-rat 488 (1:300, AlexaFluor, Molecular Probes, Eugene, Oregon, USA) and a Donkey anti-goat CY3 (1:300, Immunotech, Jackson ImmunoResearch, UK). After four rinses in TBS, sections were immersed in a Hoechst bath for 2 min (Hoechst 33258, 2 µg/ml in water, Invitrogen, USA), rinsed in two baths of water and one bath of TBS (5 min each), then cover-slipped under fluoromount-G (SouthernBiotech, Birmingham, AL, USA) and stored at 4◦C in dark.

Because in a previous study we observed cell proliferation within the MOB (Brus et al., 2010), we evaluated the influence of parturition and interactions with the young on cell proliferation in this olfactory structure as well as in the DG. To this end, single-immunolabeling was performed against the Ki67 marker, an endogenous marker of cell division which is expressed at all the phases of the cellular cycle (Kee et al., 2007). Sections were treated with TBSTA-BSA for 1 h and were incubated overnight in primary antibody rabbit anti-Ki67 (1:500, Abcam ab15580- 25, Cambridge, UK) in TBSTA-BSA at room temperature. The following day, sections were rinsed four times in TBS and were incubated in secondary antibody sheep anti-rabbit (1:400, produced by the INRA center of Nouzilly) in TBS-BSA 0.1% during 3 h at 4◦C. After four rinses in TBS 0.1%, sections were incubated with rabbit peroxydase anti peroxydase (PAP, 1:80000, Dako Z0113, Trappes, France) overnight at 4◦C. The last day, after two rinses in TBS and two rinses in Tris-HCl (0.05M, pH 7.6) sections were reacted for peroxydase detection in a solution of 3,3<sup>0</sup> -Diaminobenzidine tetrahydrochloride (DAB, 0.15 mg/mL; Sigma) containing 0.001% H2O<sup>2</sup> and 0.018% nickel ammonium sulfate for 7 min.

### **QUANTIFICATION**

The number of BrdU+ (10 sections/animal, 800 µm between sections) and Ki67+ cells (6 sections/animal, 950 µm between sections) was assessed by counting peroxydase/DAB-stained frontal sections of the MOB and the DG through different levels along the rostrocaudal axis, using a light microscope (Axioskope 2, Zeiss, Germany) on a magnification of x20 and cell count analysis software (computerized image analysis Mercator, Explora Nova, La Rochelle, France). The counter was blind to the experimental group. Areas of the MOB (granular and periventricular layers) and the DG (GCL and SGZ) were measured with this system through an x2.5 objective (MOB) and an x10 objective (DG). Cell densities were then calculated by dividing the numbers of BrdU+ or Ki67+ cells by the layer area. The Ki67+ cell density corresponds to the mean of cell density measured in the granular and the periventricular layers for the MOB and in the GCL and the SGZ for the DG. Densities of BrdU+ cells and proportions of new neurons and neuroblasts were counted only in the target structures of newly-born cells integration, the granular layer of the MOB and the GCL of the DG.

To determine the percentage of BrdU+/NeuN+ cells and BrdU+/DCX+ cells in the MOB and in the DG, approximately 100 cells per ewe were observed for the three groups (around 500–600 cells per group). Each BrdU+ cell was analyzed in its entire *z*-axis, with 0.5 µm step intervals, through an x40 oil immersion objective, using a confocal laser-scanning microscope (LSM700, Zeiss, Germany) equipped of excitation wavelengths 488 and 555. Cells rotated in orthogonal planes to verify double labeling with NeuN or DCX. For each selected cell that showed colocalization of BrdU with NeuN or DCX, an image was collected with the software Zen (Carl Zeiss, Germany). All images shown correspond to one focal plane (0.5 µm) and were imported into Gimp Pack Mode 2.6 software to adjust brightness and contrasts.

To determine the development of the newborn neuroblasts in the different groups, the number of nodes and the length of the dendritic arbor were measured in 16–18 BrdU+/DCX+ cells per ewe in the GCL of the MOB and of the DG. Each BrdU+/DCX+ cell was analyzed with confocal laser scanning microscope (LSM 700, Zeiss, Germany), in its entire *Z*-axis with 0.5 µm step interval, using x63 oil immersion objective to measure the length between the cell body to the end of the longest process. Total length of the dendritic tree was obtained by summing the length of all processes of each BrdU+/DCX+ cell. The number of nodes was obtained by counting the number of occurrences of branch points in the dendritic arbor. Interestingly, these newborn cells seemed to be at an early stage of maturation as most of them displayed only one process. Thus, the population of BrdU+/DCX+ was separated in two categories depending on the number of nodes and the percentage of cells with no nodes (one process, less mature) or with one or more nodes (two or more processes, more mature) was calculated (**Figures 4B–E**). Ambiguous cases were further analyzed using a semiautomatic neuron tracing system Imaris (Bitplane, USA).

#### **STATISTICAL ANALYSIS**

As densities of BrdU+ and Ki67+ cells, proportions of BrdU+/NeuN+ and BrdU+/DCX+ cells, and dendritic lengths of BrdU+/DCX+ cells were not normally distributed, the data were analyzed with nonparametric tests (Siegel, 1956). Intergroup comparisons were analyzed using two-tailed Kruskal-Wallis and Mann-Whitney tests. As the percentage of BrdU+/DCX+ cells which displayed one or more nodes was normally distributed, the data were analyzed with a one-way analysis of variance (ANOVA), and significance was probed by the Newman-Keuls test. Statistical analyses were performed using the statistical package SPSS 10 (Chicago, IL, USA) and the level of statistical significance was set at *p* ≤ 0.05. All data were represented as median and interquartile ranges except the number of nodes which were represented as mean ± SEM. Because we found that cell proliferation is down-regulated in *postpartum* ewes (Brus et al., 2010), we predicted that density of Ki67+ cells will be lower in parturient groups and therefore we used one-tailed Kruskal-Wallis and Mann-Whitney tests for this variable.

#### **RESULTS**

#### **SURVIVAL OF NEUROBLASTS AND NEW MATURE NEURONS IN THE MAIN OLFACTORY BULB (MOB) AND THE DENTATE GYRUS (DG)**

In the granular layer of the MOB and in the GCL of the DG, the densities of BrdU+ cells did not significantly differ between groups (MOB: *H* = 2.27, *p* = 0.3; DG: *H* = 4.48, *p* = 0.1; **Figures 2A, B**).

The proportion of newborn neurons was measured in each group by using a double immunofluorescent labeling for BrdU and NeuN, a marker of post-mitotic neurons, and DCX, a marker of neuroblasts (**Figures 3E–H**). In the granular layer of the MOB, only the proportion of BrdU+/DCX+ cells significantly differed between groups (*H* = 8.27, *p* = 0.02; **Figure 3A**). This proportion was significantly lower in the "With Lamb" group compared to the "Virgin" or the "No Lamb" groups ("With Lamb" vs. "Virgin" or "No Lamb" groups: *U* = 2.47, *p* = 0.01; **Figure 3A**). The proportion of new post-mitotic neurons (BrdU/NeuN+ cells) did not differ between groups (*H* = 1.23, *p* = 0.5; **Figure 3B**).

Contrary to the MOB, only the proportion of BrdU+/NeuN+ cells significantly differed between groups in the GCL of the DG (*H* = 11.07, *p* = 0.004; **Figure 3D**). The proportion of new mature neurons was significantly lower in the two parturient groups

interquartile ranges) in the granular layer of the MOB **(A)** and in the DG **(B)**. There is no significant difference between groups for both regions.

compared to the Virgin group ("With Lamb" vs. "Virgin" groups: *U* = 0, *p* = 0.004; "No Lamb" vs. "Virgin" groups: *U* = 0, *p* = 0.002). However, the proportion of BrdU+/DCX+ cells did not significantly differ between groups (*H* = 0.73, *p* = 0.7; **Figure 3C**).

#### **DENDRITIC LENGTHS OF NEWBORN NEURONS IN THE MAIN OLFACTORY BULB (MOB) AND THE DENTATE GYRUS (DG)**

To assess the influence of interactions with the young on the development of the dendritic arbor of new neurons, we measured dendritic lengths and number of nodes of BrdU+/DCX+ cells in the granular layer of the MOB and in the GCL of the DG (**Figure 4**).

In the MOB, dendritic lengths and number of nodes significantly differed between groups (dendritic length: *H* = 6.01, *p* =

0.049; nodes: *F* = 6.58, *p* = 0.01). The highest dendritic lengths were found in the "With Lamb" group and significantly differed from the "No Lamb" (*U* = 3, *p* = 0.03; **Figure 4A**) and showed a tendency to differ from the "Virgin" group due to an animal showing an extreme value (*U* = 5, *p* = 0.08; **Figure 4A**). The percentage of BrdU+/DCX+ cells possessing nodes was significantly greater in the "With Lamb" group than in the "Virgin" or "No Lamb" groups ("With Lamb" vs. "Virgin" groups: *p* = 0.04; "With Lamb" vs. "No Lamb" groups: *p* = 0.007; **Figure 4B**).

NeuN **(B–D)** and DCX **(A–C)**. In the MOB, only the proportion of BrdU+/DCX+ cells in the "With Lamb" group significantly differed from the two other

By contrast in the DG, no difference in dendritic lengths and in the number of nodes were found between groups (dendritic length: *H* = 1.35, *p* = 0.5; nodes; *F* = 1.04, *p* = 0.38; **Figures 4D, E**). Finally, the diameter of BrdU+/DCX+ cell bodies did not differ between groups in both structures (MOB; *F* = 2.33, *p* = 0.13; DG: *F* = 1.56, *p* = 0.24).

#### **CELL PROLIFERATION IN THE MAIN OLFACTORY BULB (MOB) AND THE DENTATE GYRUS (DG)**

To evaluate the influence of parturition and interactions with the young on cell proliferation, a single immunolabeling was performed against the Ki67 protein, an endogen marker of cell division, in the MOB and the DG (**Figure 5**). In both the MOB and the DG, the density of Ki67+ cells significantly differed between groups (MOB: *H* = 6.13, *p* = 0.025, **Figure 5A**; DG: *H* = 7.59, *p* = 0.01, **Figure 5B**). For both structures, in the two parturient groups, the density of Ki67+ cells in the MOB was significantly lower compared to the "Virgin" group (MOB: "With Lamb" vs. "Virgin" groups: *U* = 2, *p* = 0.01; "No Lamb" vs. "Virgin" groups: *U* = 7, *p* = 0.04; DG "With Lamb" and the "Virgin" groups: *U* = 1, *p* = 0.01; "No Lamb" group vs. "Virgin" groups: *U* = 8, *p* = 0.05).

dentate gyrus, BrdU: bromodeoxyuridine, DCX: doublecortin, NeuN: neuronal

#### **DISCUSSION**

nuclei. \* p ≤ 0.05, \*\* p < 0.01.

Two main results have been obtained in this study. First, interactions with the young and associated learning, but not parturition, reduce the survival of neuroblasts in the MOB, whereas in the DG, parturition induces a decrease in the number of new neurons. These results suggest that learning of the olfactory signature of the lamb, which occurred during the first mother/young interactions, is specifically associated with a down-regulation in olfactory neurogenesis. Secondly, a positive effect on the development of the dendritic length of new neuroblasts has been specifically observed in the MOB of mothers interacting with their lamb suggesting that olfactory learning accelerates the maturation of adult-born neurons in the MOB but not in the DG.

In the MOB, while the number of BrdU+ cells was similar between parturient groups and the "Virgin" group, the population of neuroblasts was decreased by interactions with the young. This reduction is probably not a consequence of a decrease in new cell production during gestation since a positive, rather than a negative, influence of pregnancy on cell proliferation has been described in the SVZ of rodents (Shingo et al., 2003; Furuta and Bridges, 2005; Larsen and Grattan, 2010). In addition, parturient ewes separated from their lambs did not show any change of cell survival in comparison to virgins indicating that parturition could not account for this down-regulation. However, the suppressive effect of interactions with the lamb for 2 days on the survival of neuroblasts could be a consequence of

various behaviors involved in maternal care. For instance licking the newborn lamb depends on olfactory attraction to amniotic fluids (Lévy et al., 1983). However since licking behavior occurs at parturition and only last for 2 h it is unlikely involved in the decrease in neuroblasts population observed after 2 days of mother-young contact. Somatosensory stimulation associated with suckling could also account for this suppressive effect. However there is so far no evidence in the literature that suckling

can have a direct effect on olfactory neurogenesis. In this respect, it would be of interest to assess olfactory neurogenesis in mothers whose udders are covered to prevent lambs from suckling. In rodents, restriction of suckling and tactile stimulation in lactating mothers in the presence of pups modulate neurochemical activity of the MOB, suggesting the importance of olfactory interactions with the young (Munaro, 1990). In sheep, during the early *postpartum* period various neurochemical and electrophysiological changes in the functional circuitry of the MOB underlie the formation of olfactory memory for lamb (Sanchez-Andrade et al., 2005; Lévy and Keller, 2008). Learning of the lamb odor could also be accompanied by changes in olfactory neurogenesis. Numerous studies report that bulbar neurogenesis varies following olfactory learning and that these changes could affect or not mature new neurons (see review: Feierstein et al., 2010a). However, whether these changes concern neuroblasts is not known. Here we show that neuroblasts but not more mature neurons are sensitive to social interactions. The decrease in the number of BrdU+/DCX+ cells cannot be explained by an enhancement of neuronal maturation since we did not observe an increase in BrdU+/NeuN+ cells in ewes interacting with their lambs. This decrease rather suggests that neuroblasts die during the 2 days of lamb exposure. Studies of adult-born neurons showed that olfactory learning has a complex effect on neuronal turnover, increasing or decreasing the survival of newborn olfactory cells as a function of their age (Mandairon et al., 2006; Mouret et al., 2008). Further studies using different timing of BrdU injections are needed to examine whether newborn cells of different age and maturation could have been differently affected.

Together with a reduced number of neuroblasts, a decrease in cell proliferation in the MOB is observed in both groups of parturient ewes in comparison to virgins, as previously shown (Brus et al., 2010). The functional significance of this reduced proliferation and maturation processes in the MOB remains unknown. It may accentuate a survival-promoting effect by eliminating newborn neurons that are not functional and would favor the maturation and integration of new neurons which participate in learning by reducing cell competition. Computational studies support this hypothesis by showing that the rate of cell proliferation and cell survival contribute to the stability of the neural activity in the network (Lehmann et al., 2005; Butz et al., 2006). This assumption is supported by our results showing that dendritic maturation of neuroblasts is only enhanced in mothers interacting with their lamb in the MOB.

Two main non-exclusive mechanisms could be considered to account for this accelerated maturation. Firstly, nursing and suckling could influence the morphology of new neurons as several studies report an effect of lactation on dendritic arbor in the mother brain. For instance, enhanced spine density has been reported in the hippocampus of lactating rats (Kinsley et al., 2006). Similarly lactating females have longer basal dendritic length compared to diestrous females in the medial preoptic area, a key region for maternal behavior (Keyser-Marcus et al., 2001). However, a negative effect has been reported on oxytocin neurons having less dendritic branches and a fewer total dendritic length in lactating rats compared to virgin rats (Armstrong and Stern, 1998; Stern and Armstrong, 1998). Secondly and more interestingly, learning the familiar lamb odor could induce this increased maturation as a positive influence of olfactory stimulation. The effect of odor learning on dendritic trees of newborn neurons has already been reported. Sensory deprivation drastically decreases

"Virgin" group. \* p ≤ 0.05, \*\* p < 0.01.

the number, the dendritic length and the spine density of newborn granule cells in the MOB (Rochefort and Lledo, 2005) and odor enrichment induces structural synaptic plasticity of adult-born granule neurons (Livneh and Mizrahi, 2011). Since evidence of an influence of learning but not of suckling on the maturation of new adult neurons has been reported in the literature, we hypothesized that learning of the olfactory signature of the lamb is involved in this maturation process. Further studies need to be done to establish whether suckling and/or learning of the lamb odor are responsible for this increased maturation of neuroblasts and whether this could also affect new mature neurons.

In the DG, while no change in the number of BrdU+ cells was found between groups, survival of new mature neurons was reduced by parturition itself, since parturient ewes isolated from their lamb showed a decrease in BrdU+/NeuN+ cells in comparison to virgins. The effect of pregnancy on the decrease in cell production is unlikely since no changes in hippocampal neurogenesis have been reported either during early or late pregnancy in rodents (see review: Lévy et al., 2011). However we cannot exclude a species dependent regulation of neurogenesis during pregnancy. These effects could also be due to the stress induced by removal of the lamb. This is however unlikely since we observed a similar decrease in the survival of mature neurons in mothers staying with their lambs. The inhibition of neurogenesis could be the consequence of hormonal changes, mainly steroids and glucocorticoids occurring at parturition. Oestradiol induced a decrease in survival of newborn neurons (Barker and Galea, 2008) and applications of corticosterone suppress cell survival in the DG of the hippocampus (Wong and Herbert, 2004, 2006). Interestingly, oestradiol and corticosterone were found to affect neurogenesis and some interactions exist between both hormones. For example, adrenal steroids mediate the suppression of cell proliferation induced by oestradiol in ovariectomized females (Ormerod and Galea, 2001). The interplay between estradiol and cortisol at parturition may also mediate levels of cell survival in the DG and further studies are needed to better determine this complex influence on neurogenesis. In addition, our results suggest a hormonal influence depending on cell type since no change in the proportion of neuroblasts was found between groups in the DG. Recent studies report that high corticosterone differently affects population of new cells according to their maturity stage (Gonzalez-Perez et al., 2011; Lussier et al., 2013). Different effects on different cell lineages may represent adaptative actions of glucocorticoids, which provide a compensatory mechanism to protect some types of cells from death in the hippocampus. Such a mechanism would prevent from a dysfunction of the hippocampus during the *postpartum* period and would allow a better spatial memory in lactating females (Kinsley et al., 1999).

In conclusion, this study supports the hypothesis that the maternal brain undergoes neuronal changes, such as adult neurogenesis, which could constitute an adaptive response to motherhood by favoring learning ability. More specifically, interactions with the young are associated with a down-regulation in olfactory neurogenesis and an enhancement of neuroblast maturation. Our hypothesis is that fewer new neurons decrease cell competition in the olfactory bulb and enhance maturation of those new neurons selected to participate in the learning of the lamb odor. Further experiments will aim at understanding how these adult born neurons could be integrated to the neural network involved in maternal behavior.

### **AUTHOR CONTRIBUTIONS**

Maïna Brus, Maryse Meurisse, Matthieu Keller and Frédéric Lévy designed and performed research, analyzed data and wrote the paper.

### **ACKNOWLEDGMENTS**

The authors would like to acknowledge the financial support of the "Programme transversal INRA/Institut Pasteur No. 319", the ANR programme blanc (2012-2015) PLASTMATBEHAV and the INRA/Région for the grant to M. Brus. We particularly thank (1) N. Jouaneau and M. Chauvet for histological preparation; (2) J. Cognié for veterinary assistance; (3) C. Moussu, E. Archer and F. Cornilleau for their assistance at lambing; (4) J.-P. Dubois and A. Arnould for sacrifice of animals; (5) D. Capo and its staff for animal breeding; (6) G. Le Pape for statistical analysis; (7) L. Szymanski for English corrections; and (8) the cellular imaging platform (PIC) of UMR PRC, Nouzilly.

### **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: 29 November 2013; accepted: 03 February 2014; published online: 18 February 2014.*

*Citation: Brus M, Meurisse M, Keller M and Lévy F (2014) Interactions with the young down-regulate adult olfactory neurogenesis and enhance the maturation of olfactory neuroblasts in sheep mothers. Front. Behav. Neurosci. 8:53. doi: 10.3389/fnbeh.2014. 00053*

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

*Copyright © 2014 Brus, Meurisse, Keller and Lévy. 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.*

## Differential memory persistence of odor mixture and components in newborn rabbits: competition between the whole and its parts

#### **Gérard Coureaud<sup>1</sup>\*, Thierry Thomas-Danguin<sup>1</sup> , Frédérique Datiche<sup>1</sup> , Donald A. Wilson<sup>2</sup> and Guillaume Ferreira3,4**

<sup>1</sup> Centre des Sciences du Goût et de l'Alimentation (CSGA), UMR 6265 CNRS, UMR 1324 INRA, Université de Bourgogne, Dijon, France

<sup>2</sup> Department of Child and Adolescent Psychiatry, New York University Langone School of Medicine, New York, NY, USA

<sup>3</sup> Nutrition and Integrative Neurobiology Group, INRA UMR 1286, Bordeaux, France

<sup>4</sup> Université de Bordeaux, Bordeaux, France

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Christiane Linster, Cornell University, USA Mouna Maroun, University of Haifa, Israel

#### **\*Correspondence:**

Gérard Coureaud, Centre des Sciences du Goût et de l'Alimentation (CSGA), UMR 6265 CNRS, UMR 1324 INRA, Université de Bourgogne, 9E Boulevard Jeanne d'Arc, 21000 Dijon, France e-mail: gerard.coureaud@ u-bourgogne.fr

Interacting with the mother during the daily nursing, newborn rabbits experience her body odor cues. In particular, the mammary pheromone (MP) contained in rabbit milk triggers the typical behavior which helps to localize and seize the nipples. It also promotes the very rapid appetitive learning of simple or complex stimuli (odorants or mixtures) through associative conditioning. We previously showed that 24 h after MP-induced conditioning to odorants A (ethyl isobutyrate) or B (ethyl maltol), newborn rabbits perceive the AB mixture in a weak configural way, i.e., they perceive the odor of the AB configuration in addition to the odors of the elements. Moreover, after conditioning to the mixture, elimination of the memories of A and B does not affect the memory of AB, suggesting independent elemental and configural memories of the mixture. Here, we evaluated whether configural memory persistence differs from elemental one. First, whereas 1 or 3-day-old pups conditioned to A or B maintained their responsiveness to the conditioned odorant for 4 days, those conditioned to AB did not respond to the mixture after the same retention period. Second, the pups conditioned to AB still responded to A and B 4 days after conditioning, which indicates stronger retention of the elements than of the configuration when all information are learned together. Third, we determined whether the memory of the elements competes with the memory of the configuration: after conditioning to AB, when the memories of A and B were erased using pharmacological treatment, the memory of the mixture was extended to day 5. Thus, newborn rabbits have access to both elemental and configural information in certain odor mixtures, and competition between these distinct representations of the mixture influences the persistence of their memories. Such effects certainly occur in the natural context of mother-pup interactions and may contribute to early acquisition of knowledge about the surroundings.

**Keywords: Oryctolagus cuniculus, newborn, odor mixture, configural perception, stimulus representation, retention, memory persistence**

### **INTRODUCTION**

In some cases, mixtures of volatile molecules are perceived as a collection of independent, identifiable elements; the perception is then elemental (e.g., Laing and Francis, 1989; Laska and Hudson, 1993; Linster and Cleland, 2004). However, some mixtures induce a configural processing. Then, the mixture gives rise to either a unique and novel odor quality, different from the odor qualities of the elements (robust configural perception; e.g., Smith, 1996; Jinks and Laing, 1999; Kay et al., 2005), or to a novel quality perceived in addition to the qualities of the odorants (weak configural perception; Rescorla, 1973; Kay et al., 2005). Configural odor processing has been described with different mixtures and different approaches in a variety of species from invertebrates to vertebrates, including humans (e.g., Derby et al., 1996; Linster and Smith, 1999; Valentincic et al., 2000; Wise and Cain, 2000; Deisig et al., 2001; Wiltrout et al., 2003; Mandairon et al., 2006; Riffell et al., 2009; Gottfried, 2010; Wilson and Sullivan, 2011). For instance, data in human adults revealed that a mixture of two odorants (AB), one smelling like strawberry (A: ethyl isobutyrate) and the other like caramel (B: ethyl maltol), generates the configural perception of a pineapple odor at a specific ratio of A/B (30/70 v/v; Le Berre et al., 2008, 2010; Barkat et al., 2012). Interestingly, recent results in a newborn mammal, the newborn rabbit, showed similar configural processing abilities with the same AB mixture, at the same ratio. Indeed, after single appetitive conditioning to odorant A or to odorant B by pairing with the mammary pheromone (MP) (naturally contained in rabbit milk and experimentally used as unconditioned stimulus in associative conditioning procedure; Coureaud et al., 2006, 2010), rabbit pups respond to the conditioned element and respectively to the AC or BC mixtures (C: guaïacol). However, they do not respond to the AB mixture. These results suggest that they perceive AB differently from the odors of A and of B, while they perceive AC or BC as the sum of their component odors (Coureaud et al., 2008, 2009a, 2011a). Furthermore, after single conditioning to the AB mixture, rabbit pups respond to the components A and B in addition to AB, indicating that the AB mixture is perceived in a weak configural way (Coureaud et al., 2008). Very recently, combining behavioral approaches with pharmacological tools, we provided another confirmation of this neonatal weak configural perception of AB and demonstrated that rabbit pups memorize the AB configuration independently (at least in part) from the representations of each element. After conditioning to AB, amnesia of A and B did not propagate to AB: pups that did not respond to either A or B still responded to AB (Coureaud et al., in press). Thus, after AB conditioning, distinct elemental and configural memories of the mixture are created.

In the previous studies in newborn rabbits, memories of the elements A and B or of the AB configuration were evaluated 24 or 48 h after conditioning, which are the optimal delays for responsiveness to a stimulus conditioned by single pairing with the MP (Coureaud et al., 2006). However, to date, whether memory persistence of the configural odor of the AB mixture differs from that of the odorants A and B (also detected in the mixture) had never been evaluated; here, we hypothesized that they could be distinguished. This issue constituted the first and main goal of our study. Besides understanding early memory and perception in the rabbit, this evaluation could more broadly help provide additional and original information on the general topic of odor object perception in mammals. Indeed, as said above, a perceptual match has been evidenced for the AB mixture between rabbit pups and human adults, which suggests a relative conservation in the processing of certain odor mixtures across species.

Our second goal aimed to assess whether memory retention of the configural AB odor and of the elemental odors of A and B depends on the age of the pups at conditioning. Indeed, it is known in newborn mammals (especially in altricial mammals, which develop very rapidly) that the meaning acquired by a conditioned odorant can change from one day to another (e.g., Barr et al., 2009; Sullivan and Holman, 2010). In previous experiments, conditioning was mainly performed in 2-day-old rabbit neonates. In the present study, conditioning was therefore conducted either 1 or 3 days after birth to determine the influence of development on the perception and memory of a complex odor stimulus.

These two issues were evaluated by taking advantage of the configural nature of the previously described AB mixture. To date, this is the only mixture which has been extensively characterized regarding its perceptual properties (weak configural perception) in the rabbit both in terms of behavior (Coureaud et al., 2008, 2009a, 2011a; Sinding et al., 2011), memory (Coureaud et al., in press) and brain processing (Schneider et al., in preparation). To that goal, we conditioned either 1- or 3-day-old pups to simple odorants A or B, or to their binary mixture, and assessed the retention of the conditioned stimuli from 24 to 96 h after conditioning using independent groups for each stimulus and each retention interval (Experiment 1). Pups conditioned to the AB mixture were also tested for their responsiveness to A and to B with the aim to evaluate, by means of a within-subject analysis, whether responsiveness to AB differed over time from that of the single odorants (Experiment 2). Finally, as differences appeared in the responsiveness to the mixture and to its odorants in the first two experiments, we determined whether competition occurred between the memory of AB and the respective memories of A and B created during conditioning to the whole mixture (Experiment 3). Indeed, interference and competition between different associative memories have been classically reported (e.g., Eisenberg et al., 2003; Suzuki et al., 2004; Bradfield and Balleine, 2013) and might contribute to the present results. Therefore, we either prevented the formation of the configural memory by successive conditioning to the single odorants only, or erased the elemental memory after initial conditioning to the AB mixture followed by separate reactivations of odorants A and B and amnesiainducing anisomycin (AN) injection (Coureaud et al., 2009b, 2011b, 2013). We then evaluated whether these two procedures influenced the responsiveness to the AB mixture after a long retention interval.

### **MATERIALS AND METHODS**

#### **ANIMALS AND HOUSING CONDITIONS**

Males and females New-Zealand rabbits *Oryctolagus cuniculus* (Charles River strain, L'Arbresle, France) from the Centre de Zootechnie of the University of Burgundy (Dijon) were kept in individual cages. A nest box (0.39 × 0.25 × 0.32 m) was added on the outside of the pregnant females' cages 2 days before delivery (day of delivery was day 0; d0). To equalize pups' nursing experience, all females had access to their nest between 11:30– 11:45 a.m. This procedure allowed females to follow the brief (3–4 min) daily nursing of the species (Zarrow et al., 1965). Animals were kept under a constant 12:12 light:dark cycle (light on at 7:00 a.m.) with ambient air temperature maintained at 21– 22◦C. Water and pelleted food (Lapin Elevage 110, Safe, France) were provided *ad libitum*. In the study, 524 newborns (from 106 litters) were used.

The study was carried out under the local, institutional and national rules (French Ministries of Agriculture, and of Research and Technology) regarding the care and experimental use of the animals. All experiments were conducted in accordance with ethical rules enforced by French law, and were approved by the Ethical Committee for Animal Experimentation (no. 2406).

#### **ODORANTS**

The stimuli consisted of 2-methylbut-2-enal (the Mammary Pheromone, MP, CAS# 497-03-0), ethyl isobutyrate (odorant A, CAS# 97-62-1), ethyl maltol (odorant B, CAS# 4940-11-8) for pure components, and of the AB mixture. This mixture included 0.3 × 10−<sup>5</sup> and 0.7 × 10−<sup>5</sup> g/ml of components A/B; the 30/70 v/v ratio elicits the configural perception of a pineapple odor in human adults due to blending properties (Le Berre et al., 2008, 2010; Barkat et al., 2012), and weak configural perception in newborn rabbits (Coureaud et al., 2008, 2009a, 2011a; Sinding et al., 2011).

The MP allowed inducing the learning by the pups of odorant A, odorant B, or the AB mixture through associative conditioning (see below). It was used at 10−<sup>5</sup> g/ml, a concentration known to be highly efficient to promote conditioning (Coureaud et al., 2006). Thus, the A-MP and B-MP blends were prepared at a final concentration of 10−<sup>5</sup> g/ml of each constituent. The AB-MP blend included 1 × 10−<sup>5</sup> g/ml of MP and 0.3 and 0.7 × 10−<sup>5</sup> g/ml of odorants A and B.

Single odorants A and B were also used in the reactivation procedure (Experiment 3b) at a concentration of 10−<sup>5</sup> g/ml.

In all experiments, we deliberately kept constant the overall concentration of the different stimuli (single odorants or AB mixture and blends; 10−<sup>5</sup> g/ml) and maintained this constancy between conditioning and testing to avoid direct influence of the concentration on our results and to focus on the influence of complexity (single odorant vs. mixture).

All the odorants were purchased from Sigma-Aldrich (Saint-Quentin Fallavier, France) and all the final solutions were prepared in a solvent composed of 0.1% of ethanol (anhydrous, Carlo Erba, Val de Reuil, France) and 99.9% of MilliQ water (Millipore, Molsheim, France).

#### **ODOR CONDITIONING, REACTIVATION AND PHARMACOLOGICAL TREATMENT**

Conditioning sessions were run on day 1 or 3 after birth in an experimental room close to the breeding room. The pups from an individual litter were transferred by groups of 5 (usual case) or 4 (Experiment 3b) into a box maintained at room temperature. The MP-induced conditioning was run following a procedure previously described (e.g., Coureaud et al., 2006, 2008, 2009a,b, 2013; Sinding et al., 2011; Charra et al., 2013). This procedure offered the advantage of being extremely rapid (single trial) and appetitive (thus avoiding the possible modulation of responsiveness due to the negative emotional state that may occurred after aversive conditioning).

In most cases, just before the conditioning session, 4 ml of the A-MP (Experiment 1), B-MP (Experiment 1) or AB-MP (Experiments 1, 2 and 3b) blends were pipetted on a pad (19 × 14 cm, 100% cotton) then held 2 cm above the pups for 5 min. The conditioning occurred 1 h before the daily nursing (10:30 a.m.) to equalize the pups' motivational state and limit the impact of satiation on behavioral responses (Montigny et al., 2006). Two minutes after the end of the conditioning, the pups were individually marked with ink and returned to their nest. The box containing the pups was rinsed with alcohol and with distilled water after each conditioning session.

In one group (Experiment 3a), the procedure was the same except that pups were conditioned successively to odorants A and B: they were exposed to the A-MP blend for 2.5 min (for the half of the pups, randomly chosen, and B-MP for the other half), then transferred to a second box in which they remained non

stimulated for 1 min before being exposed to the B-MP (or A-MP) blend for 2.5 min.

Finally, in two other groups (Experiment 3b), 24 h after the conditioning to the AB mixture, the memory of pups was reactivated by exposure to odorant A then odorant B in half of the pups (or conversely in the other half) following the same procedure than above (2.5 min per odorant, delay inter-stimulation: 1 min). Immediately after reactivation, anisomycin (AN; Sigma-Aldrich) was injected to the half of the pups (42 mg/kg, i.p.) after dilution in 0.9% NaCl solution and adjustment to pH 7.2 with 1N HCl. The AN was used after reactivation performed at 24 h as we have previously demonstrated the effectiveness of this procedure in erasing memory of odor element(s) in newborn rabbits (Coureaud et al., 2009b, 2011b, 2013, in press). Control for the effect of AN injection was carried out with the other half of animals, which received saline 0.9%. As in other studies with newborn or adult mammals (e.g., Davis and Squire, 1984; Gruest et al., 2004; Desgranges et al., 2008) we considered that AN in newborn rabbits may induce a real amnesia and not a perturbation in responsiveness due to an aversive effect. Pups were returned to the nest just after AN or saline injection.

### **BEHAVIORAL ASSAY**

The behavioral assay occurred 24, 48, 72 or 96 h after the conditioning, i.e., on days 2, 3, 4 or 5 when the conditioning occurred on day 1, and on days 4, 5, 6 or 7 when it happened on day 3. The assay was run in the experimental room previously used for conditioning and reactivation, and happened also 1 h before the daily nursing to limit the impact of satiation on motivation and behavioral responsiveness (Montigny et al., 2006). It consisted of an oral activation test during which a pup was immobilized in one gloved hand of the experimenter, its head being left free. The odor stimulus was presented for 10 s with a glass rod 0.5 cm in front of the nares (e.g., Coureaud et al., 2006, 2008, 2011a, 2013). A test was positive when the stimulus elicited headsearching movements (vigorous, low amplitude horizontal and vertical scanning movements displayed after stretching towards the rod) usually followed by grasping movements (labial seizing of the rod extremity). Non-responding pups displayed no response except sniffing. Pups were tested in groups of 4 or 5 (same groups than during the conditioning), and only once, on a given day (i.e., different groups were tested on different days).

In Experiment 1, the pups were tested for their responsiveness to one stimulus only, except those which were conditioned to the AB mixture. The latter were indeed tested to AB but also (these results are the results of Experiment 2) to odorants A and B. In Experiment 3, the pups were tested to A, B and AB. Successive testing involved the presentation of a first stimulus to a pup, then a second stimulus to another pup, and so on with an inter-trial interval of 60 s. The order of stimuli presentation was systematically counterbalanced from one to another pup. If a pup responded to a stimulus, its nose was softly dried before the next stimulation. The pups were immediately reintroduced in their nest after testing.

#### **STATISTICS**

Each group was composed of 18–20 pups except the groups of Experiment 3b (with saline and AN injection; *n* = 11–12 pups). The proportions of pups responding during the behavioral assay were compared using either the χ 2 -test of Pearson (with Yates correction when necessary) when the groups were independent (i.e., distinct groups tested for their response to a same stimulus on different days, or to different stimuli on a given day), or the Cochran's *Q*-test when the groups were dependent (i.e., pups from a same group tested for their response to the three stimuli). When the Cochran's *Q*-test was significant, proportions of responding pups were compared 2 × 2 by the χ 2 -test of McNemar. Degrees of freedom are indicated when >1. Data were considered as significant when the two-tailed test ended with *p* < 0.05.

### **RESULTS**

#### **EXPERIMENT 1. RETENTION OF A, B AND AB AFTER THEIR RESPECTIVE LEARNING ON DAY 1 OR DAY 3**

To compare the retention of the odors of odorant A, odorant B and of the AB configuration, four groups of pups were conditioned to A on day 1, four other groups to B and four other groups to AB (*n* = 20 pups/group, one pup died in one of the groups; all groups were independent). Each group was tested for its behavioral responsiveness to the conditioned stimulus at one single time point only: 24, 48, 72 or 96 h after the conditioning. The same experiment was conducted with 16 other independent groups of pups conditioned on day 3 instead of day 1 (*n* = 20/group, one pup died in two groups and two pups in another group).

After conditioning on day 1 (**Figure 1**, left column), the responsiveness of pups to the conditioned stimulus decreased over time whatever the nature of the stimulus (i.e., in pups conditioned to A, or B or AB; χ <sup>2</sup> > 19.37, *df* = 3, *p* < 0.001 in the three situations). In pups conditioned to A, the responsiveness to the odorant was maximal and similar 24 and 48 h after the conditioning (>85%; χ <sup>2</sup> < 1, *p* > 0.05), lower at 72 h (55%; χ <sup>2</sup> > 4.3, *p* < 0.05 compared to 24 and 48 h) and 96 h (35%; χ <sup>2</sup> > 8.4, *p* < 0.01 compared to 24 and 48 h), and not different between 72 and 96 h (χ <sup>2</sup> = 1.6, *p* > 0.05). Regarding the pups conditioned to B, the pattern of responsiveness was nearly the same. The responsiveness was maximal and equivalent at 24 and 48 h (>85%; χ <sup>2</sup> = 1.4, *p* > 0.05), lower at 72 h (60%; comparison with 24 h: χ <sup>2</sup> = 7.6, *p* < 0.01, with 48 h: χ <sup>2</sup> = 3.1, *p* = 0.07) and 96 h (42.1%; comparison with 24 and 48 h: χ <sup>2</sup> > 6.03, *p* < 0.01), and similar between 72 and 96 h (χ <sup>2</sup> = 1.2, *p* > 0.05). After conditioning to the AB mixture, the pups highly and similarly responded to AB 24 and 48 h later (95%) and less at 72 h (65%; comparison with 24 and 48 h: χ <sup>2</sup> = 3.9, *p* < 0.05). At 96 h, only 5% of the pups responded, a level which was lower compared to 24, 48 and 72 h (χ <sup>2</sup> > 18.6, *p* < 0.001). Interestingly, whereas the responsiveness of the three categories of pups (conditioned to A, B and AB) did not differ at 24, 48 and 72 h (between categories comparisons: χ <sup>2</sup> < 1.29, *df* = 2, *p* > 0.05), at 96 h the pups conditioned to AB responded less to this stimulus than the pups conditioned to A or B (χ <sup>2</sup> = 7.74, *df* = 2, *p* < 0.05; comparisons AB vs. A or B: χ <sup>2</sup> > 3.91, *p* < 0.05; comparison A vs. B: χ <sup>2</sup> < 0.5, *p* > 0.05).

After conditioning on day 3 (**Figure 1**, right column), the retention of the acquired stimulus changed over time independently of its nature (χ <sup>2</sup> > 8.28, *df* = 3, *p* < 0.05 in the three situations). In pups conditioned to A or B, the responsiveness was maximal and close at 24 and 48 h (>75%; χ <sup>2</sup> < 0.9, *p* > 0.05 for A or B comparisons), lower at 72 h (around 60%) compared to 24 h (χ <sup>2</sup> > 4.8, *p* < 0.05 for A or B), and lower at 96 h compared to 24 and 48 h for A (45%; χ <sup>2</sup> > 5.2, *p* < 0.05) or compared to 24 h for B (50%; χ <sup>2</sup> = 7.4, *p* < 0.001). The responsiveness was similar between 96 and 72 h both in pups conditioned to A and in pups conditioned to B (χ <sup>2</sup> < 0.6, *p* > 0.05). In pups conditioned to AB, the responsiveness to AB was strong and equivalent at 24 and 48 h (>70%; χ <sup>2</sup> = 1.4, *p* < 0.05), but it became extremely weak as soon as 72 h (<10.5%; comparisons between 72 or 96 vs. 24 or 48 h: χ <sup>2</sup> > 11.8, *p* < 0.001). While the responsiveness to the conditioned stimulus was equivalent between pups conditioned to A, B or AB at 24 and 48 h (χ <sup>2</sup> < 0.5, *df* = 2, *p* > 0.05), it was lower in pups conditioned to AB than to A or to B at 72 and 96 h (χ <sup>2</sup> > 8.91, *df* = 2, *p* > 0.05; A or B vs. AB: χ <sup>2</sup> > 4.88, *p* < 0.05; A vs. B: χ <sup>2</sup> < 0.5, *p* > 0.05).

Thus, whatever the information that was learned (A, B, or AB) on either day 1 or 3, its retention by the pups decreased over the 4 post-conditioning days. However, the retention of the AB mixture or of its odorants was not the same: all the pups stopped responding to the mixture 72 and/or 96 h after conditioning while a significant proportion of them (>35%) were still responding to the odorant they have previously learned. Besides the time of memory testing, the animal's age at conditioning also influenced the retention of AB memory: pups conditioned to AB at day 3 stopped responding earlier than those conditioned at day 1 (72 h vs. 96 h). This effect of age was not observed for A and B memories when comparing conditioning at day 1 and 3.

#### **EXPERIMENT 2. RETENTION OF A, B AND AB AFTER LEARNING OF AB ON DAY 1 OR 3**

Results of Experiment 1 showed clear differences in the retention of the odors of A, B and AB when each odorant or the mixture were learned separately by different pups, suggesting that the memory retention of the binary mixture is weaker compared to the retention of its components. Here, we assessed whether similar results could be obtained at the individual level (withinsubject design), namely after conditioning to AB (during which acquisition of the A and B elements and of the AB configuration happens; Coureaud et al., 2008, 2009a, 2011a; Sinding et al., 2011) and during successive testing to the three stimuli. To that goal, the pups conditioned to the AB mixture on day 1 or 3 in Experiment 1 (4 independent groups/day) were tested for their response to A and to B, in addition to AB, at one single time point only, i.e., either at 24, 48, 72 or 96 h. According to the results of Experiment 1, it was expected that the retention of the configuration would be shorter than the retention of each element that compose the mixture.

After conditioning to AB on day 1 (**Figure 2**, upper part), the pups strongly and similarly responded to AB, A and B at 24 and 48 h (>95%; *Q* < 1.1, *df* = 2, *p* > 0.05). At 72 h, the responsiveness decreased for AB (Exp. 1) but also for A and for B (responsiveness to A or B at 72 vs. 24 or 48 h: χ <sup>2</sup> > 3.9, *p* < 0.05); it remained however similar between AB, A and B (65–75%; *Q* = 1.2, *df* = 2, *p* > 0.05). At 96 h, the responsiveness to A and to B was still around 60 and 75% (thus lower than at 24 and 48 h for A: χ <sup>2</sup> = 7.6, *p* < 0.01, equivalent for B: χ <sup>2</sup> = 3.1, *p* = 0.07, and similar for A and for B compared to 72 h: χ <sup>2</sup> < 1.02, *p* > 0.05). Strikingly, in the same animals, while responsiveness to A and B was relatively maintained, the responsiveness to AB was dramatically reduced (5%; *Q* = 23.3, *df* = 2, *p* < 0.001; AB vs. A or B: McNemar χ <sup>2</sup> > 9.09, *p* < 0.01; A vs. B: McNemar χ <sup>2</sup> = 1.3, *p* > 0.05).

Similar results were observed after conditioning on day 3 (**Figure 2**, lower part), though with a slightly different time course. After conditioning to AB on day 3, the responsiveness of pups was high and similar to AB, A and B 24 h later (>90%; *Q* = 3, *df* = 2, *p* > 0.05) and 48 h later (70–85%; *Q* = 3, *df* = 2, *p* > 0.05). At 72 and 96 h, the responsiveness to A and B remained around 50 and 60% (lower than at 24 h, χ <sup>2</sup> > 5.6, *p* < 0.01, but not different than at 48 h, χ <sup>2</sup> < 2.1, *p* > 0.05), but the responsiveness to AB was weaker (<10.5%; for the two periods: *Q* > 14.88, *df* = 2, *p* < 0.001; AB vs. A or B: McNemar χ <sup>2</sup> > 5.14, *p* < 0.05; A vs. B: McNemar χ <sup>2</sup> < 1.3, *p* > 0.05).

Thus, in line with the data from Experiment 1 obtained with independent groups of pups, the responsiveness of newborn rabbits to the mixture and to its components decreased over time after conditioning to the mixture and individual testing to the three A, B and AB stimuli: it disappeared earlier for the mixture than for the odorants, and earlier after conditioning on day 3 than on 1.

Moreover, on postnatal day 5, the proportion of pups still responding to AB was higher in pups conditioned on day 3 than in pups conditioned on day 1 (70 vs. 5%; χ <sup>2</sup> = 15.3, *p* < 0.001) while the responsiveness to A or to B was similar (60–85%; χ <sup>2</sup> < 0.5, *p* > 0.05). This suggests that the difference observed for AB was the consequence of memory processes and not of pure developmental effects.

#### **EXPERIMENT 3. COMPETITION BETWEEN THE MEMORIES OF THE ELEMENTS AND OF THE CONFIGURATION**

According to the results of Experiments 1 and 2, newborn rabbits have a distinct memory and retention of the AB mixture and of its components. After acquisition of the AB mixture, this difference in terms of retention could be due to a competition between the memory of each odorant and the memory of the AB configuration (since the pups perceived both the odorants and the configuration in the weak configurally processed AB mixture). To evaluate this hypothesis, two procedures were followed.

The first procedure (Experiment 3a) attempted to prevent the creation of a memory for the configuration. In a previous paper (Coureaud et al., 2008), we showed that after successive conditioning to odorant A then odorant B on day 1, the pups responded 24 h later to the AB mixture (they did not display such a response after conditioning to a single odorant). Here, this paradigm was repeated but with a test of responsiveness 96 h later. This group (*n* = 20 pups, two of them died, results concerned 18 pups) was compared to the group conditioned to the whole mixture on day 1 and tested on day 5 (+96 h) in Experiment 2 (i.e., a group that learned the AB configuration). Whereas the conditioning to the AB mixture was followed by an absence of responsiveness to AB but not to A and B at 96 h (**Figure 3A**, same results as in **Figure 2**), the successive learning of A and B induced a high and similar level of responsiveness to both the mixture and the individual odorants (66.6–72.2%; *Q* = 1, *df* = 2, *p* > 0.05) (**Figure 3B**). The responsiveness to AB was therefore higher in the situation of successive learning of the elements than after learning of the mixture (72.2 vs. 5%; χ <sup>2</sup> = 15.6, *p* < 0.001). The responsiveness to the odorants was similar in the two experimental conditions (60– 75%; χ <sup>2</sup> < 0.5, *p* > 0.05). Therefore, an absence of perception of the AB configuration during conditioning, due to separate and successive conditioning to odorants A and B, was followed by the maintenance of responsiveness to the AB mixture 4 days later.

The second procedure (Experiment 3b) attempted to promote the creation of memories to both the configuration and the elements after AB conditioning, and to make the pups rapidly amnesic of the individual odorants' odor. We hypothesized that

in this situation, the responsiveness to AB could be maintained 4 days after the conditioning because of the absence of competition between the configural and elemental memories. Thus, two independent groups of newborns (*n* = 12/group, 4 pups/litter, 2 pups/litter/group) were conditioned on day 1 to the AB mixture, and reactivated the day after (day 2) by separated exposure to each individual odorant (A then B for half of the pups in each group, B then A for the other half). Immediately after memory reactivation, the first group was injected with saline, while the other group received an AN injection inducing amnesia. The pups were all tested for their responses to AB, A and B on day 5, 96 h after the conditioning (72 h after the reactivation). Blocking memory for the individual odorants induced major differences in mixture memory between the groups. The pattern of response in the saline-treated group (**Figure 3C**) was the same as in pups from previous experiments which were not manipulated on day 2 (**Figure 3A**): they did not respond to AB but still displayed a

high and similar level of responsiveness to A and to B (8.3 vs. 75.0 vs. 66.6%, respectively; *Q* = 12.6, *df* = 2, *p* < 0.01; AB vs. A or B: McNemar χ <sup>2</sup> > 5.1, *p* < 0.05; A vs. B: McNemar χ <sup>2</sup> < 0.5, *p* > 0.05). In contrast, newborns treated with AN (1 pup died, *n* = 11 for this group in the results) showed a completely reversed pattern of responsiveness to the mixture compared to the elements (*Q* = 10.3, *df* = 2, *p* < 0.01) (**Figure 3D**): they responded strongly to the mixture but very weakly to the individual odorants (63.6 vs. 9.1 vs. 9.1%; AB vs. A or B: McNemar χ <sup>2</sup> = 4.16, *p* < 0.05). Comparison of the responsiveness between the two groups clearly showed that AN-treated pups responded more to AB and less to A and to B than saline-treated neonates (χ <sup>2</sup> > 5.49, *p* < 0.05).

Thus, when rabbit pups forgot the odors of the individual elements initially acquired during the conditioning to the AB mixture, their responsiveness to the mixture was extended over time. This result strongly supports our hypothesis that the learning

of the mixture induced a competition between the memory of the AB configuration and the memory of the elements A and B.

### **DISCUSSION**

The present results, obtained in rabbit pups using the AB weak configural mixture, demonstrate that the memory of odor elements is more robust and lasts longer than the configural memory of the mixture, and in fact can interfere with the maintenance of the configural memory. In the absence of elemental odor memory, the duration of the configural memory of the AB mixture is significantly enhanced. Although further work is required to determine whether these results generalize to other mixtures (weak configurally, configurally or, in contrast, elementally perceived mixtures), the present findings have important implications for understanding the mechanisms of odor mixture (and odor object) perception and the organization of odor mixture memory.

Advantage of elemental over configural memory of the AB mixture was first observed in the duration of memory following conditioning to the elements alone or to the mixture (Experiment 1). For example, conditioning to either element (A or B) alone induced a memory of that element that extended for at least 96 h. In contrast, conditioning to the binary mixture induced memory that extended no more than 72 h. Interestingly, the duration of memory for the mixture was age-dependent, while the duration of elemental memory was not. Animals conditioned to the mixture on day 1 reactivated memory for the mixture for 72 h, while animals conditioned on day 3 displayed mixture memory for no more than 48 h. Additional work is required to determine the mechanisms of this age-dependent variation in the duration of configural memory. Regarding general memory mechanisms, one may hypothesize that the information is more easily processed when the animal is exposed to a single odorant compared to a mixture of odorants. To be encoded correctly, a weak configural mixture could require a higher arousal level than individual odorants perceived separately, since the animals might have to share attentional level between the different stimuli of the mixture to engage associative learning for all stimuli (Sharot and Phelps, 2004). As a result, after only one conditioning session, the memory trace of a complex odor stimulus might be rather fragile and its decay might be faster over time in comparison with more simple stimuli. Here, the present conditioning was appetitive. One may therefore wonder whether similar results would be obtained with aversive conditioning, using odor-malaise association for instance (see Gruest et al., 2004), since a stronger arousal is then supposed to occur during acquisition. It would be of interest to evaluate in future experiments whether this other kind of conditioning induces longer lasting configural memory of the AB mixture.

Advantage of elemental over configural memory of odor mixture was also found after associative conditioning with the AB mixture, and within-subject testing to both the mixture and its elements (Experiments 2 and 3). After conditioning to this mixture known to be perceived configurally in humans (Le Berre et al., 2008; Barkat et al., 2012), rabbit pups expressed a weak configural memory of the mixture, meaning that they expressed both a memory for the mixture's configuration and for the individual elements. In particular, they did not respond to AB at 96 h or 72 h (after conditioning on day 1 or 3, respectively) but still responded to A and to B at these retention times (Experiment 2). In terms of odor mixture processing, these results may seem surprising. Indeed, if the mixture is perceived in a weak configural way during the retention test, rabbit pups should perceive in the mixture both the elements they have previously learned and the AB configuration. Moreover, when pups have learned independently enough elements forming a mixture, they can respond (generalize) to the mixture even if the mixture is perceived in part configurally (Experiment 3a showing responsiveness to AB after successive learning of A and B; see also Coureaud et al., 2008). Therefore, after conditioning to AB on day 1 and because the pups responded to A and to B at 96 h, a response to the mixture could be expected. To explain the lack of response to AB at 96 h, we hypothesize that the AB mixture is, initially, weakly configurally perceived during the conditioning with the MP, but that a particular associative strength is then given to the configuration compared to the elements. As a consequence, the pups would perceive the mixture more as the configuration when they are exposed again to the mixture several days after conditioning. This hypothesis is supported by theoretical considerations suggesting that during conditioning to complex stimulus, a specific value is assigned to the configural representation ("unique cue") independently of that given to each element (Rescorla, 1973; Rescorla et al., 1985). Alternatively, one may assume that during the conditioning, the presence of the MP leads rabbit neonates to perceive the AB mixture less configurally (i.e., weak configurally) compared to its processing in the absence of unconditioned stimulus (robust configural perception).

Importantly, the present results suggest that the long-term memory of the elemental and configural representations of the AB mixture compete. That is, with intact elemental memory, configural memory of the mixture degrades significantly faster than memory of the elements (Experiment 2). In contrast, if elemental memory is disrupted by selective reconsolidation blockade, configural memory of the mixture is maintained significantly longer (Experiment 3b). This indicates that removing a potential source of interference facilitates memory performance for the AB configural information. The retrieved memory of the AB mixture can thus be regarded as the sum of conflicting processes involving, in our case, the elemental memory and the configural memory. The outcome seems dependent on the dominance of one of the memories at the time of retrieval, here the elemental memory, a result that contributes to unravel aspects of memory organization (e.g., Eisenberg et al., 2003; Suzuki et al., 2004; Bradfield and Balleine, 2013). The competition would rely on the retention time (Experiments 1 and 2) and the simultaneous (but not successive; Experiment 3a) exposure to odorants forming a weak configurally perceived mixture. This interference/competition may happen theoretically and first during consolidation of configural and elemental memories of the mixture. However, this assumption has certainly to be ruled out, since we have previously shown in newborn rabbits that consolidation of odor stimuli learned by association with the MP happened in the first 4 h after conditioning (Coureaud et al., 2009b); the consolidation phase is thus terminated well before the erasure of A and B memories performed here (i.e., 24 h after conditioning; Experiment 3b). Therefore, in the present paradigm, interference/competition between configural and elemental memories of the mixture certainly occurs during maintenance and/or recall of the memories rather than during consolidation phase, with elemental memory the stronger of the two. Even if the most probable explanation for the absence of response to AB at 96 h is the absence of recognition of the AB configuration, one can not exclude the possibility that rabbit pups still find familiarity in AB but without any motivational significance.

The present results do not allow direct determination of the mechanisms of elemental and configural memory interaction, although hypotheses based on the known neurobiology of odor mixture processing can be developed. In rodents, odor mixtures are processed differently by the olfactory bulb and primary olfactory (piriform) cortex (Wilson and Sullivan, 2011). For example, the activity of olfactory bulb mitral/tufted cell neural ensembles is strongly consistent with pattern separation processing– responding uniquely to even minor changes in mixture elements (Barnes et al., 2008; Chapuis and Wilson, 2011; Sahay et al., 2011). Thus, it has been hypothesized that olfactory bulb neural ensembles respond to odor mixture features or components, rather than to the mixture configuration. In strong contrast, piriform cortical neural ensembles respond in a manner consistent with configural processing. In fact, even at the single-unit level of the piriform cortex, individual cells can distinguish between mixtures and their components in cross-adaptation procedures (Wilson, 2003). Mitral/tufted cells, on the other hand, show strong cross-adaptation between mixtures and their components, again consistent with an elemental or feature-dependent process within the olfactory bulb (Wilson, 2003). After having reported in newborn rabbits mapping of activations induced by the MP or by a MP-learned odorant in the olfactory bulb and central regions, including the piriform cortex (Charra et al., 2012, 2013) we are currently assessing the brain activity mapping of the AB mixture (Schneider et al., in preparation).

Odor memory is dependent on plasticity within both the olfactory bulb and piriform cortex (Wilson et al., 2004), even in neonates (Moriceau and Sullivan, 2005; Charra et al., 2013; Fontaine et al., 2013). In the context of the present results, and waiting for further results obtained with the AB mixture as with other odor mixtures, we hypothesize that the more stable elemental memory of an odor mixture may mainly depend on olfactory bulb plasticity, while configural memory of the same mixture may mainly depend on piriform cortical processes (in accordance with the presumed role of piriform cortex as an associative cortex; Johnson et al., 2000). These cortical processes may be less robust during early development than memorydependent events in the olfactory bulb, resulting in the distinction between these two memory forms in the present study. The less robust configural memory might depend on differences in the developmental program of both the olfactory bulb and piriform cortex: functional odor maps in the glomerular layer seems rather well defined after birth (Guthrie and Gall, 1995) whereas the maturation of inhibitory processes and the intracortical associational fibers throughout the 3-layered piriform cortex (Schwob and Price, 1984) might follow a slightly different time curve (Garske et al., 2013). One may also suggest that this distinction does not rely on robustness of cortical processing only, but on changes during animal development in the dialogue between sensory/mnesic regions of the brain (olfactory bulb/piriform cortex/amygdala/hippocampus), in accordance with the behavioral and adaptive needs of the animal (Schacher and Hu, 2014). Future work could combine the powerful reconsolidation-mediated dissection of memory utilized here, with neurophysiological techniques to explore neurobiological underpinnings of these basic memory and perceptual processes.

Finally, the present findings are further evidence that while odor mixtures are perceived sometimes configurally, and conscious analysis of odor mixtures is notoriously difficult (e.g., Jinks and Laing, 1999), information about the underlying individual components can remain intact in both neonatal rabbits (shown here) and adult humans (Grabenhorst et al., 2007). In the case of newborn rabbits, this may be especially important given the life and death importance of odor recognition to interact with the mother, attach to the nipples, survive and grow up. Having access to both the elements and the configuration may help ensure successful recognition, improved discrimination between odorous substrates or conspecifics in the surroundings, adaptation to the actual environment or anticipation of social and feeding changes that will come later in development.

#### **ACKNOWLEDGMENTS**

We sincerely thank Valérie Saint-Giorgio, Nicolas Malaty, Florent Costilhes, Jérôme Antoine and all the Centre de Zootechnie from the University of Burgundy for their cooperation. The work was supported by French ANR-2010-JCJC-1410-1 MEMOLAP to Gérard Coureaud, Thierry Thomas-Danguin, Frédérique Datiche and Guillaume Ferreira.

#### **REFERENCES**

Barkat, S., Le Berre, E., Coureaud, G., Sicard, G., and Thomas-Danguin, T. (2012). Perceptual blending in odor mixtures depends on the nature of odorants and human olfactory expertise. *Chem. Senses* 37, 159–166. doi: 10. 1093/chemse/bjr086


**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 April 2014; accepted: 26 May 2014; published online: 16 June 2014*. *Citation: Coureaud G, Thomas-Danguin T, Datiche F, Wilson DA and Ferreira G (2014) Differential memory persistence of odor mixture and components in newborn rabbits: competition between the whole and its parts. Front. Behav. Neurosci. 8:211. doi: 10.3389/fnbeh.2014.00211*

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

*Copyright © 2014 Coureaud, Thomas-Danguin, Datiche, Wilson and Ferreira. 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*.

## Olfactory bulb encoding during learning under anesthesia

**Alister U. Nicol <sup>1</sup>\* † , Gabriela Sanchez-Andrade <sup>2</sup>† , Paloma Collado<sup>3</sup> , Anne Segonds-Pichon<sup>4</sup> and Keith M. Kendrick<sup>5</sup>\* †**

<sup>1</sup> Sub-department of Animal Behaviour, University of Cambridge, Cambridge, UK

<sup>3</sup> Department of Psychobiology, Universidad Nacional Educación a Distancia (UNED), Madrid, Spain

<sup>4</sup> Bioinformatics Group, The Babraham Institute, Cambridge, UK

<sup>5</sup> Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China, Chengdu, China

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Donald A. Wilson, New York University School of Medicine, USA Matthieu Keller, Centre National de la Recherche Scientifique, France

#### **\*Correspondence:**

Alister U. Nicol, Sub-department of Animal Behaviour, University of Cambridge, Madingley, Cambridge CB23 8AA, UK e-mail: aun10@cam.ac.uk Keith M. Kendrick, Key Laboratory for Neuroinformation, University of Electronic Science and Technology of China, 4, North Jianshe Road, Chengdu, Sichuan 610054, China e-mail: k.kendrick.uestc@gmail.com

†These authors have contributed equally to this work (i.e., joint first authors).

Neural plasticity changes within the olfactory bulb are important for olfactory learning, although how neural encoding changes support new associations with specific odors and whether they can be investigated under anesthesia, remain unclear. Using the social transmission of food preference olfactory learning paradigm in mice in conjunction with in vivo microdialysis sampling we have shown firstly that a learned preference for a scented food odor smelled on the breath of a demonstrator animal occurs under isofluorane anesthesia. Furthermore, subsequent exposure to this cued odor under anesthesia promotes the same pattern of increased release of glutamate and gamma-aminobutyric acid (GABA) in the olfactory bulb as previously found in conscious animals following olfactory learning, and evoked GABA release was positively correlated with the amount of scented food eaten. In a second experiment, multiarray (24 electrodes) electrophysiological recordings were made from olfactory bulb mitral cells under isofluorane anesthesia before, during and after a novel scented food odor was paired with carbon disulfide. Results showed significant increases in overall firing frequency to the cued-odor during and after learning and decreases in response to an uncued odor. Analysis of patterns of changes in individual neurons revealed that a substantial proportion (>50%) of them significantly changed their response profiles during and after learning with most of those previously inhibited becoming excited. A large number of cells exhibiting no response to the odors prior to learning were either excited or inhibited afterwards. With the uncued odor many previously responsive cells became unresponsive or inhibited. Learning associated changes only occurred in the posterior part of the olfactory bulb. Thus olfactory learning under anesthesia promotes extensive, but spatially distinct, changes in mitral cell networks to both cued and uncued odors as well as in evoked glutamate and GABA release.

**Keywords: anesthesia, microdialysis, mitral cells, multiarray electrophysiology, neurotransmitters, olfactory bulb, olfactory learning, social transmission of food preference**

#### **INTRODUCTION**

A number of studies have provided evidence for the occurrence of associative learning under general anesthesia. Initial findings using auditory fear conditioning paradigms suggested that learning under anesthesia only occurred if epinephrine was given during the conditioning procedure (Weinberger et al., 1984; Gold et al., 1985). However, subsequent studies using both auditory and olfactory conditioning paradigms have provided evidence for learning under anesthesia without the necessity for epinephrine treatment. For both auditory and olfactory conditioning under anesthesia evidence for altered responses of neurons in the medial prefrontal cortex (Rosenkranz et al., 2003; Laviolette et al., 2005) and lateral amygdala has been reported (Rosenkranz and Grace, 2002; Rosenkranz et al., 2003; Fenton et al., 2013). All these studies have used paradigms where a CS+ is associated with foot-shock and effects of associations with positive reinforcers have not been investigated. Additionally there is evidence from *in vivo* neurotransmitter release, and localized pharmacological intervention and electrophysiological recording studies in both sheep (Kendrick et al., 1992, 1997) and mice (Wilson et al., 1987; Brennan et al., 1998), that plasticity changes occurring within primary sensory cortex, notably the olfactory bulb, are important for learning. However, whether similar changes occur in the olfactory bulb during learning under anesthesia is unknown.

Odor learning in mammals, under various paradigms, has been shown to be supported, to a considerable extent, by biochemical and physiological changes occurring in the mitral cell layer of the olfactory bulb. Learning-related elevations in extracellular levels of glutamate and gamma-aminobutyric acid (GABA), and an increase in the ratio of GABA relative to glutamate have been found in both sheep (Kendrick et al., 1992) and mice

<sup>2</sup> Wellcome Trust Sanger Institute, Hinxton, Cambridge, UK

(Brennan et al., 1998) following olfactory learning. This suggests a mechanism involving reciprocal increases in both excitation and inhibition, where the relative impact on inhibitory activity is greater. Other reported extracellular changes include increased noradrenaline, nitric oxide and aspartate (Kendrick et al., 1997; Brennan et al., 1998). In the accessory olfactory bulb (AOB), an area associated with pheromonal perception and learning, neurochemical (Brennan et al., 1995) and electrophysiological (Binns and Brennan, 2005) effects consistent with such a mechanism have also been reported in relation to the Bruce effect in mice, whereby exposure to the odor of an unfamiliar male causes termination of pregnancy. Local field potential recordings in the AOB suggest selective inhibition of the familiar pheromone in the underlying recognition system. In the main olfactory system however odor learning can be associated with both increased and decreased responses of mitral cells (Wilson et al., 1987; Kendrick et al., 1992).

One of the most robust models of olfactory learning in rodents is the social transmission of food preference. Rodents such as mice and rats are generally neophobic with regard to novel foods, preferring to eat food items which are familiar to them. However, following social interaction with a conspecific "demonstrator" which has previously consumed a novel food, otherwise naïve "observer" animals subsequently show a preference for the same novel food by consuming more of that food than an alternative novel one (Galef and Wigmore, 1983; Valsecchi and Galef, 1989). Indeed, the acquired attractiveness of the novel food may be such that the consumption of this food initially exceeds the preceding consumption of the animals' normal daily diet (Galef and Whiskin, 2000). The social transfer of food preference does not require direct physical contact between the demonstrator and observer animals since it is mediated by carbon disulfide (CS2; Galef et al., 1988), a metabolic by-product carried in the exhaled breath of rodents. Effective training of the observer may be accomplished using an anesthetized demonstrator (Galef and Wigmore, 1983; Valsecchi and Galef, 1989), or even replacing the demonstrator with an artificial surrogate, such as a wad of cotton wool carrying a novel food odor and a few drops of CS<sup>2</sup> (Galef et al., 1988).

We have previously provided behavioral evidence that social transmission of food preference can occur under anesthesia in mice, using anesthetized demonstrators (Burne et al., 2010), although whether this involves neurochemical and electrophysiological changes in the olfactory bulb similar to those seen following learning in awake animals is unknown. In another study on responses of single neurons in the olfactory bulb of anesthetized mice test odors were mixed with the anesthetic gas and produced reliable mitral cell responses (Lin et al., 2005). In the present study we therefore used this same approach to investigate firstly *in vivo* neurochemical changes occurring post-learning in anesthetized mice in response to odors smelled on the breath of a demonstrator mouse. During the learning phase of the experiment both observer and demonstrator mice were anesthetized. In a second experiment multielectrode array (24 electrode) electrophysiological recordings were used to monitor the olfactory responses of neurons in the mitral cell layer of the olfactory bulb in isofluorane anesthetized mice both during and shortly after learning using a paradigm where animals were exposed to a novel (CS+) odor with CS2.

### **EXPERIMENTAL PROCEDURES**

#### **ANIMALS**

All procedures were conducted under licence in accordance with UK Home Office regulations (Animals (Scientific Procedures) Act, 1986).<sup>1</sup>

The subject animals were adult male mice (129svC57B6) which were bred in-house (Babraham Institute) and maintained to an age of 3–6 months before any experimental procedures. Mice were housed in same-sex groups of 2–5 animals in standard M3 cages, under temperature-controlled conditions in a 12 h light—12 h dark cycle (lights on at 07:00 h). All animals were handled and weighed on each of the 2 days before the start of experimentation.

#### **PREPARATION OF FOOD ODORS**

Scented foods were prepared by introducing an additive to the normal diet in powdered form. The additives used were cumin (0.2%), ginger (1%), coriander (1.5%) and cocoa (2.0%), each obtained from a local retailer and stored in airtight containers. Odor samples were prepared by inserting a non-airtight capsule containing 2–3 g of powdered food into a polyvinyl fluoride gas sampling bag (Adtech Polymer Engineering, UK). All air was removed from the bag, which was then refilled with 1 L of nitrogen gas (odorless). Preparation of scented foods and food odors was done in a room separate from areas where experimental procedures were performed. Odor bags containing CS<sup>2</sup> (∼800 ppm) were prepared by placing a 1 cm square of filter paper into an evacuated odor bag, adding 0.2 µL of CS<sup>2</sup> using a 10 µl precision syringe (Hamilton Bonaduz AG, Switzerland) through a sealed injection site, then refilling the bag with 1 L of nitrogen gas. Samples of N<sup>2</sup> gas were also prepared.

#### **BEVAVIOR**

#### **Restricted feeding**

Prior to training, mice were submitted to a restricted feeding regime (see **Figure 1** and legend). During the restricted feeding regime, animals exhibiting weight loss exceeding 15% of initial weight were returned to normal feeding and excluded from further procedures.

#### **Training procedure 1—social transfer of food preference**

On day 4 of restricted feeding, demonstrator mice were transferred individually to a clean cage which was the same as the home-cage but with no sawdust, and allowed to feed for 2 h *ad libitum* from a dish (6 cm diameter × 2 cm high) containing 2–3 g of food. The food presented in this loading period was either the normal powdered diet, or food scented with either cumin or cocoa. Mice eating less than 0.1 g in this period were not used as demonstrators and were excluded from further experimental procedures. After loading, the demonstrator was

<sup>1</sup>http://www.legislation.gov.uk/ukpga/1986/14/enacted

anesthetized (25% Hypnorm, 25% Hypnovel, i.p., 0.1 ml/g body weight). Observer mice were anesthetized (1.5% isofluorane in O<sup>2</sup> delivered at 150 ml/min via a mask over the nose) and, when fully anesthetized, placed individually in close nose to nose proximity, but without actual direct physical contact, with the demonstrator mouse for 1 min. There were two groups of trained observers: for one group the training food (i.e., the food used to load the demonstrator) was cocoa-scented (cocoa-trained, *n* = 10 mice), for another group the training food was cumin-scented (cumintrained, *n* = 7). For a third group the demonstrator had been fed the normal powdered diet (control, *n* = 11). Observers remained anesthetized for 1–2 min after removal from isofluorane. During this recovery period they were placed in a standard cage and on a heated plate to keep them warm. They were then returned to their home-cage where they were allowed 2 h access to plain powdered food (∼2 g per mouse). On day 5, observers in each of the three groups were tested for their preference between cumin-scented and cocoa-scented food.

#### **Training procedure 2—simulated social transfer of food preference**

Preliminary behavioral experiments were conducted in order to establish the naïve preference in this paradigm between a number of different combinations of odor pairs. From these tests ginger was chosen as the training odor and coriander as an alternative (untrained) one because there was a stable apparently innate preference for coriander over ginger. For the task, observer mice (*n* = 37) were anesthetized (as in Training Procedure 1) and then transferred from the mask carrying the isofluorane and O<sup>2</sup> anesthetic gas to one carrying the anesthetic gas with an odor introduced. Odors were added to the anesthetic gas by first drawing off a sample of the odor from an odor bag into a 50 ml syringe, then delivering this into the anesthetic gas using a syringe pump (Harvard Apparatus, UK—model 22). Three groups of mice were used: group 1 (Ginger + CS2, *n* = 14) received 30 ml/min ginger food odor with 15 ml/min CS2, group 2 (Ginger, *n* = 10) received 30 ml/min ginger food odor, and group 3 (CS2, *N* = 13) received 15 ml/min CS2. These odors were mixed with 120 ml/min of the anesthetic gas, and any shortfall beneath 150 ml/min was made up with N<sup>2</sup> gas. Odor exposure was for 1 min. After this, they were placed in a heated cage to recover from anesthesia and then returned to their home-cage where they were allowed to feed for 2 h (as in Training Procedure 1). On day 5, observers in each of the three groups were tested for their preference between gingerscented and coriander-scented food.

#### **Food preference testing**

Observer mice were tested individually under red light for food preference. On day 5 of restricted feeding, each mouse was placed in the center of the test arena in which there were two food cups in diagonally opposite corners (**Figure 1**), each containing a different scented food. Mice were left undisturbed for 30 min, monitored and recorded via an overhead video camera. Tests were conducted during the 2 h period when the mice had been fed on the preceding days of restricted feeding. At the end of the test the food remaining in each cup, and the amount displaced into the outer dish, was weighed to determine the amount of each food consumed.

For each food type the amount consumed in the test was expressed as a percentage of the total food consumption during the test. Training Procedure 1: Data were divided in to three groups depending on the food the demonstrator had eaten, cocoa-trained, cumin-trained and plain-trained. To establish whether there was a training related food preference, a Kruskall Wallis test followed by Dunn's multiple comparisons test was used to compare the cocoa-scented food consumption.

### **IN VIVO NEUROCHEMISTRY**

#### **Preparation for in vivo microdialysis sampling**

Animals included in the microdialysis study were those used in the behavioral study under Training Procedure 1. Immediately after completion of the food preference test (**Figure 2A**), mice were anesthetized with an i.p. injection (0.1 ml/g body weight) of Avertin (5 g 1,2,2,2 -tribromoethanol, 3 ml 2-methylbutan-2-ol, 20 ml ethanol, 222 ml saline). Animals were then placed in a stereotaxic frame (Kopf Instruments, California) where they were fixed using ear bars and a bite bar. Anesthesia was maintained with 1.5% isofluorane in oxygen (150–200 ml/min) delivered through a diffuser (Univentor Ltd, Malta) through a mask over the nose. Body temperature was maintained using a homeothermic heated blanket system and a rectal probe (Harvard Instruments, UK). A microdialysis probe (MAB 4.15.2 Cu, Microbiotech, Sweden) was inserted unilaterally into the external plexiform layer of the left olfactory bulb through an incision in the scalp and a small craniotomy (AP 3.9 mm, LR 5.25 mm, DV 5.25; **Figure 2B**). Coordinates were calculated from the bregma and dural surface (dorso-ventral coordinates) according the mouse brain atlas of Paxinos and Franklin (2001) and were sufficiently posterior to avoid any possibility of sampling from the AOB.

### **Sampling protocol and analysis**

Dialysis probes were perfused with a Krebs-Ringer solution (pH 7.4, 138 mM NaCl, 1.5 mM CaCl2, 11 mM NaHCO3, 5 mM KCl, 1 mM MgCl2, 1 mM NaH2CO3) at 1.5 µl/min, using a syringe pump (CMA-10; CMA Microdialysis, Sweden), throughout the course of the experiment. Sampling commenced 90 min after probe implantation. Samples (25 µl) were collected at 15 min intervals (**Figure 2B**) into tubes containing 2 µl acetic acid, and frozen (−20◦C) immediately for further analysis by high performance liquid chromatography (HPLC). During the 5th and 10th of these intervals, the normal air supply to the mask was switched from the normal gas supply (isofluorane in O<sup>2</sup> and N2) to a balanced supply carrying a food odor introduced from an odor bag (see above), cumin in one interval, cocoa in the other. The order of presentation of the two food odors was randomized between animals, and independent of the food consumed by the demonstrator animal. The odor, carried in N2, was diffused through a hypodermic needle (25 G, 0.5 × 16 mm) into a gas supply carrying isofluorane in O<sup>2</sup> using a syringe pump (Harvard Apparatus, UK—Model 33 Twin Syringe Pump).

At the end of the experiment, mice were killed by cervical dislocation, brains were removed and probe placement confirmed. Concentrations of the amino acids glutamate, gamma-aminobutyric acid (GABA) and aspartate were quantified by HPLC with fluorescence detection following derivatization with ophthaldialdehyde (Sigma) performed with an autosampler (Gilson 231), as previously described (Kendrick et al., 1996).

All data were collected using Chromeleon 6.5 software (Dionex, Sunnyvale, USA). Stable, baseline concentrations of extracellular amino acids were taken from the average of the two samples preceding the odor challenge. Changes from baseline (100%) in response to the odor challenges were analyzed using a Wilcoxon matched-pairs signed rank test. Correlation between learning and changes in neurotransmitter release was assessed using a Spearman correlation test. The amount of cued food eaten was correlated with the percentage change in glutamate and GABA release in response to the cued food odor.

#### **ELECTROPHYSIOLOGICAL RECORDING Preparation for recording**

After initial induction of anesthesia (25% hypnorm, 25% hypnovel, i.p., 0.01 ml/g body weight), mice were placed on a heated blanket (33.4◦C) in a stereotaxic frame, and their heads fixed with ear bars and a bite bar. Anesthesia was maintained with 1.5% isofluorane in air (30% Nitrogen, 70% Oxygen) delivered through a mask over the nose. Breathing was monitored using a thermistor located in the mask (see **Figure 3A**).

The left olfactory bulb was exposed through a dorsal craniotomy large enough to accommodate the electrode array (approx. 2 mm AP × 1.5 mm ML), and the dura retracted. Exposed tissue was bathed in sterile saline delivered at 33.4◦C. A 6 × 4 electrode array (sharpened tungsten microelectrodes, 300–500 kΩ, tip separation 350 µ) was positioned at the surface of the bulb, and advanced until spontaneous extracellular neuronal activity (action potentials, "spikes") were detected across a substantial portion of the array, ranging from 3–15 electrodes per mouse (average = 8 electrodes). The objective was to make recordings

**Frontiers in Behavioral Neuroscience www.frontiersin.org** June 2014 | Volume 8 | Article 193 |

intake). Relative consumption of the two foods was significantly different across the groups (p = 0.0007). While plain food-trained mice (controls, whose demonstrator had been given normal plain food) showed no significant preference for either food, cocoa-trained mice ate almost exclusively cocoa-scented food and cumin-trained animals ate nearly none (p-values given in the Figure are relative to cumin-trained mice). **(B)** Immediately after being tested for food preference, mice were used for microdialysis experiment. A microdialysis probe was placed in the posteromedial region of the

uncued odor. In the control (plain-trained) mice on the other hand there were no differences in concentrations of either transmitter during exposure to the two odors. **(D)** For test mice, the amount of cued food consumed during the preference test was positively correlated with concentrations of GABA collected during exposure to the cued but not uncued odor during microdialysis sampling (right panel). In the control mice, there was no such relationship between release of either transmitter during exposure to a food odor, and the quantity of that

food consumed in the preference test (left panel).

from the dorsal mitral cell layer across the majority of the main olfactory bulb. Once this condition was satisfied, the preparation was left to stabilize for a period of no less than 2 h. The placement of recording arrays was sufficiently posterior to avoid any possibility of recordings being made from the AOB.

#### **Recording protocols and analysis**

Neuronal activity was sampled using a 64 channel Plexon Multichannel Acquisition Processor (MAP, Plexon Inc., USA). The signal from each electrode was sampled at 30 kHz. Extracellularly recorded action potentials (spikes) were isolated from the continuous signal when this signal exceeded a negative triggering threshold set at an estimated 5× background (∼−25 µV). Experimental markers and ventilation were recorded using a Power1401 laboratory interface and Spike2 software (Cambridge Electronic Design Ltd., UK). This system was also used for experimental control.

An experimental trial comprised a pre-stimulus period of at least 10 s, during which ventilatory and neuronal activity were visually monitored for stability, followed by a 10 s odor presentation (see **Figure 3B**). Using the animals' monitored ventilation, the onset of odor presentation was automated to a point midway through exhalation, odor onset occurring at the first such point after the 10 s stable pre-stimulus period. Odor delivery was as described above in neurochemistry methods (sampling), with the exception that the odors used here were ginger and coriander, or ginger paired with CS<sup>2</sup> during the training phase. Neuronal activity sampled during inhalation and exhalation was considered separately in subsequent analyses.

Recordings were made during three experimental phases: (1) pre-training; (2) training; and (3) post-training. In the pretraining phase, mice were presented with the three food odors: normal, ginger and coriander. These were presented in blocks of 5 trials, in balanced order, a total of 10 trials per odor. Training comprised 10 trials in which the training odor (ginger) was presented in combination with CS2. The post-training phase replicated the pre-training phase, using the same presentation order as used in pre-training recordings. Throughout the recordings, an inter-trial interval of ≥5 min was used. For some of the animals, a final block of trials was conducted in which CS<sup>2</sup> alone was presented. On completion of the electrophysiological recordings, animals received a lethal dose of pentobarbital (Dolethal, 0.5 ml i.p.).

Initial processing of the data to extract the times and waveforms of spikes in relation to stimulus and ventilatory information was performed using Spike2 software. For the majority of electrodes where spikes were detected, these represented the activities of multiple neurons. The activities of individual neurons were sorted from multineuronal activity using a machine learning algorithm applied to principal components extracted from the spike waveforms using principal components analysis, and waveform properties detected by curve fitting (Horton et al., 2007). This spike-sorting technique was realized in custom-written software running in MatLab. For each trial, the activity of each neuron was partitioned into data sampled during inhalation or during exhalation, and data sampled pre-stimulus presentation, or during stimulus presentation.

On the basis of data sampled during inhalation in the prestimulus period, the activity of each neuron was categorized as parametric or non-parametric. In each trial, the ratio of mean to standard deviation was calculated for the firing rate of each neuron. The average of this ratio was then calculated across the full set of trials for each stimulus condition. If this average was ≥2, the neuron was defined as "parametric" under the given condition, meaning that the distribution of the spikes across the time windows was considered normal. If the average was <2, the neuron was defined as "non-parametric". This property was taken into account in subsequent analyses.

Other studies mainly report weaker neuronal responses to sensory stimuli in anesthetized animals when compared to awake animals (see Brown and Horn, 1994), and rely on cumulative change in activity across multiple recordings (e.g., McLennan and Horn, 1992), although one study on the mouse olfactory bulb has reported stronger responses to odor stimuli under ketamine anesthesia (Rinberg et al., 2006). Here we have adopted a similar approach by using change in activity across multiple neurons, but also address the issue of weak single neuron responses by employing Cohen's D as a measure of response strength. Using this approach, single neurons were classified as weakly, moderately or strongly responsive.

Neuronal responses were quantified using the squared log ratio of the firing rate during stimulus presentation relative to the firing rate before. This was calculated for each neuron across the 10 trials (considered as technical replicates). Only the neurons for which data were obtained under the full series of seven conditions (plain, ginger and coriander before and after training, and training with ginger and CS2) were considered in analyses. Initially focusing on the neurons identified as parametric (see above), two-way repeated measures ANOVA was applied, with odor as the within group factor and subject as the between group factor.

### **RESULTS**

#### **NEUROTRANSMITTER CHANGES IN RESPONSE TO LEARNED ODORS**

Prior to preparation for neurochemical analysis by *in vivo* microdialysis, mice were anesthetized and trained by social transfer of food preference from an anesthetized conspecific (Training Procedure 1, **Figure 1**). Twenty-four hours after recovery from anesthetic, they were tested for food preference (**Figure 2A**). The total amount of food consumed in the preference test by observer mice in each group was not significantly influenced by the food used to load the demonstrator (0.47 sg plain-trained, 0.27 g cocoa-trained and 0.25 g cumin-trained—*p* > 0.05). However, the relative consumption of each food by the observers in the preference test varied significantly according to the odor presented in the demonstrator's exhaled breath when both animals were anesthetized (Kruskal-Wallis, χ 2 <sup>2</sup> = 142.67, *p* = 0.0007). The trained observers consumed more of the training food than the alternative food, cocoa-trained groups ate significantly more of the cocoa-scented food than the cumin-trained group (Dunn's Multiple Comparison Test, *p* < 0.001). Mice in the control group (exposed to the breath of demonstrators fed with plain food) showed no significant preference for either scented food, although they ate significantly more cocoa-scented food than cumin-trained animals (Dunn's Multiple Comparison Test, *p* < 0.05). On completion of these procedures, animals were again anesthetized and prepared for *in vivo* microdialysis (**Figure 2B**).

The full microdialysis sampling protocol was successfully completed in 13 trained and 9 control mice. Four trained and two control animals were either excluded either because the full sampling protocol could not be completed due to technical problems or, in the trained group, did not show clear evidence of a learned preference. **Figure 2B** shows the typical medial posterior olfactory bulb placements of microdialysis probes. Glutamate and GABA concentrations were measured in 15 min microdialysis samples collected during both baseline conditions and exposure to the different odors (i.e., cocoa-scented and cumin-scented food odors; see protocol in **Figure 2B**). Changes in transmitter concentrations in samples taken during odor presentations were calculated relative to the two preceding 15 min baseline sample periods.

In mice trained by exposure to demonstrators that had eaten a scented food (cumin or cocoa), glutamate and GABA concentrations were only significantly increased in microdialysis samples taken during presentation of these same odors (Wilcoxon test: Glutamate *p* = 0.0237; GABA *p* = 0.0266—**Figure 2C**). There was no significant change in the ratio of glutamate to GABA during presentation of the cued odor (*p* > 0.1) although the magnitude of evoked GABA release tended to be higher than that of glutamate. There were no significant changes in the concentrations of either glutamate or GABA, or in the glutamate/GABA ratio in samples collected during exposure to the alternative control odor not present on the demonstrator's breath. There were no changes in glutamate or GABA concentrations during presentation of either odor to control mice which had been exposed to demonstrator mice that had eaten only plain unscented food (**Figure 2C**).

The increased concentrations of GABA during presentation of the cued food were significantly positively correlated with the amount of that food consumed in the preference test (Spearman rank test: *r* = 0.57, *p* = 0.042; **Figure 2D**). There was no significant correlation between either glutamate or GABA concentrations and the total amount of food consumed in the preference test. Furthermore, in the untrained controls the combined consumption of both foods in the preference test was not significantly correlated with release of either transmitter (**Figure 2D**).

### **MULTIARRAY ELECTROPHYSIOLOGICAL CHANGES IN RESPONSE TO LEARNED ODORS**

#### **Behavior**

In the electrophysiological studies ginger was used as the training odor and coriander as the alternative test odor. These were delivered from odor bags rather than anesthetized demonstrators (see Training Procedure 2). Mice in three groups were each exposed to different odors while anesthetized: (1) ginger + food odor; (2) CS2; and (3) ginger + food odor paired with CS2. While the total amount of food consumed in the preference test did not vary across the three groups the relative consumption of coriander-scented food and ginger-scented food in the preference test reflected odor exposure under anesthesia. In both control groups, mice exposed under anesthesia to either ginger or CS2, food consumption indicated a clear preference for coriander over ginger (**Figure 3D**; paired *t*-tests, Ginger: *t*<sup>9</sup> = 3.46, *p* = 0.007 and CS2: *t*<sup>12</sup> = 3.61, *p* = 0.004). Mice which had been exposed under anesthesia to Ginger and CS<sup>2</sup> combined showed no such preference since they consumed similar quantities of both foods (**Figure 3D**). A comparison of the Ginger and CS<sup>2</sup> alone groups combined with the Ginger + CS<sup>2</sup> group also showed a significant difference in terms of the relative consumption of ginger-scented food (*t*<sup>33</sup> = 3.295, *p* = 0.002). Thus, in these mice the innate preference for coriander over ginger seen in controls was overcome.

#### **Electrophysiological recordings**

Electrophysiological recordings were made from a separate cohort of mice to those used in the behavioral tests described above. The mice were anesthetized throughout all electrophysiological recording procedures and their breathing pattern was not influenced differentially by the odors used. While the first breathing cycle (inhale and exhale) after odor presentation was significantly faster than the average pre-stimulus breathing rate, this did not vary across the different odors and returned to normal in the second and subsequent breathing cycles during stimulus presentation.

In seven mice, neuronal activity was detected at a total of 50 electrodes (3–15 active channels per animal). From these spike trains, the activity of 260 individual neurons was discriminated by spike sorting (Horton et al., 2007).

For each experimental condition, neurons were identified as parametric or non-parametric on the basis of their background activity (prestimulus activity) during inhalation. Neurons classified in this way did not necessarily remain in the same category across all conditions.

Analyses of odor-evoked changes in neuronal activity were initially performed on activity sampled during inhalation amongst neurons classified as parametric. Activity recorded during odor presentation was compared to that in the pre-stimulus period across presentations of each odor.

#### **Changes in multiunit activity**

In the pre-training period (i.e., before presentation of the training odor combined with CS2), a similar increase in multiunit activity was evoked by presentation of each food odor in the parametric neurons (mean ± sem respectively plain 5.2% ± 2.4, ginger 3.5% ± 2.1; coriander 5.4% ± 2.7—**Figure 3C**). During training, the response to the ginger food odor combined with CS<sup>2</sup> showed a trend towards being greater than that evoked by presentation of the ginger food odor alone (paired *t*-test, *t*<sup>84</sup> = 1.81, *p* = 0.074, 10.4% ± 2.7—**Figure 3C**). After training, the response to the ginger food odor alone (13.03% ± 2.7—**Figure 3C**) was significantly greater (*t*-test vs. ginger pretraining, *t*<sup>84</sup> = 2.58, *p* = 0.011) than that evoked by presentation of this odor combined with CS<sup>2</sup> during training, and significantly greater than the responses to the coriander food odor (paired *t*-test, *t*<sup>84</sup> = 3.47, *p* = 0.001), which was not the case before training (paired *t*-test *p* = 0.577). For all but one of the mice, there was a consistent increase in the size of the response to the ginger odor in multiunit activity amongst parametric neurons; in the remaining mouse the size of this response did not increase. Significant variability between mice was related to the size of the training-related increase in the response to the ginger food odor. After completion of the posttraining tests for responsiveness to the three food odors, some animals (*n* = 3) were also tested with CS<sup>2</sup> alone. The multiunit response to this odor was similar to the pre-training responses to food odors confirming that CS<sup>2</sup> odor alone did not evoke a differential response compared to the other odors used.

In the multiunit activity sampled during exhalation from parametric neurons, there was a significantly larger increase in activity in response to presentation of the ginger food odor combined with CS<sup>2</sup> during training than in response to any of the three food odors in the pre-training recordings. However, none of these responses to the food odors was altered after training.

The general profile of responsiveness in the multiunit activity amongst the non-parametric neurons was similar to that seen in the parametric ones (data not shown).

#### **Changes in single neuron activity**

The responses of single neurons in these recordings were generally weak, and few reached significance using conventional methods (e.g., *t*-tests). Using Cohen's D as a measure of response strength yielded a larger sample of responsive neurons, thereby improving the power of subsequent analyses investigating the changing profiles of neuronal responsiveness to the different odors.

The proportion of single neurons showing increased activity in response to the ginger food odor after training showed a trend towards significance (29/85, 34% before and 39/85, 46%— McNemar test, *p* = 0.123) as well as during the training period when ginger was presented with CS<sup>2</sup> (40/85, 47%, *p* = 0.1). This increase was largely confined to neurons producing excitatory responses (increase in firing rate) with the proportion of neurons producing inhibitory responses remaining stable. The proportion of neurons responding to plain food odor did increase significantly from pre-training (33/85, 39%) to posttraining (47/85, 55%, *p* = 0.005). Again this increase was confined to the neurons producing excitatory responses, but here was also accompanied by a decrease in the proportion of neurons producing inhibitory responses. The proportion OF neurons responding to the coriander food odor increased slightly, but not significantly, between the pre-training (34/85, 40%) and post-training (41/85, 48%, *p* = 0.349) periods, although that of neurons with an excitatory response to coriander declined, but not significantly, from pre-training (25/85, 29%) to posttraining (19/85, 22%, *p* = 0.391). Among the 85 recorded individual neurons, 6 changed their response to coriander from excitatory to inhibitory after training, whereas 13 changed their response to ginger from inhibitory to excitatory. This difference in the pre vs. post-training pattern of response to the two odors was significant (McNemar test, *df* = 84, *p* = 0.016).

#### **Localization of learning-evoked electrophysiological changes**

Preliminary analyses of the spatial distribution of responsiveness and learning-related effects were made by comparing effects across the X (anteroposterior, 6 columns of electrodes) and Y (mediolateral, 4 rows) dimensions of the electrode array (see **Figure 4**).

In the pre-training period, there were no significant differences in either dimension in the change in multiunit activity evoked by presentation of any of the three odors (ginger, coriander, or plain food). In the post-training period, there was also no significant difference in the distribution of learning-related change in responsiveness in the Y dimension of the array (i.e., medial vs. lateral). However, the increased multiunit response to ginger following training exposure to ginger combined with CS<sup>2</sup> (see **Figure 3**) was restricted to the posterior part of the array (**Figure 4**). Only in this part of the array were the changes in multiunit activity evoked by presentation of ginger and CS<sup>2</sup> during training, and by ginger alone after training, significantly greater than the multiunit response to ginger before training (paired *t*-tests, *t*<sup>30</sup> = 2.5, *p* = 0.018 and *t*<sup>30</sup> = 2.89, *p* = 0.007 respectively). Also, only in the posterior part of the array was the multiunit response to ginger significantly greater than that to coriander after training (*t*<sup>30</sup> = 2.62, *p* = 0.014). Furthermore, after training there were significantly more excitatory than inhibitory responses to ginger in the posterior part of the array (23% before vs. 37% after, *p* < 0.05). There was no change in responsiveness to either of the other stimuli. In the anterior part of the array, there was no training-related change in the size of the response to any stimulus, and no significant difference between excitatory and inhibitory responses either before or after training.

#### **DISCUSSION**

Overall our results have provided both neurochemical and electrophysiological evidence *in vivo* that similar plasticity changes

occur in mouse olfactory bulb encoding associated with learning under anesthesia as previously reported in olfactory learning contexts in conscious animals (Wilson et al., 1987; Kendrick et al., 1992, 1997; Brennan et al., 1998). This provides yet further evidence that the mammalian brain can undergo plasticity changes and learning in unconscious as in conscious states and also demonstrates similarities in altered olfactory bulb encoding associated with olfactory learning in both cases. Our findings also establish this anesthetized learning model as potentially useful for more detailed investigation of the precise neural encoding changes associated with learning which can be more difficult to achieve using awake, behaving animals.

ginger was restricted to the posterior half of the array.

Our present results provide support for our previous findings that learning in the social transfer of food preference paradigm can occur under anesthesia (Burne et al., 2010). Our previous study showed however that food preference transferred in this way is influenced by different types of anesthetic agents which may depend upon which transmitter receptors are targeted by them (Burne et al., 2010). Specifically, those acting by blocking NMDA receptor activity (e.g., Vetalar—Ketamine hydrochloride) appear to prevent acquisition of preference for a novel food, whereas a GABA receptor agonist (Sagatal), produce an aversion rather than a preference for the odor presented during anesthesia. Indeed, a previous study has reported increased responsivity to odors in olfactory bulb mitral cells under ketamine anesthesia in mice, compared to conscious recordings (Rinberg et al., 2006), possibly suggesting reduced selectivity and increased noise in mitral cell networks might contribute to its blockade of learning effects. In the present work, we therefore used isofluorane which is a membrane stabilizer that does not act specifically on receptor mechanisms, and leaves cortical neurons responsive to synaptic input (Orth et al., 2006).

Our observed *in vivo* increases in release of both glutamate and GABA in the olfactory bulb in response to learned odors in anesthetized mice is similar to that observed in conscious sheep (Kendrick et al., 1992, 1997) and mice (Brennan et al., 1995) in other olfactory learning paradigms. Importantly, in the current study strength of learning, as indexed by the amount of cued odor food eaten, was positively correlated with the magnitude of GABA concentration changes. Overall this provides further support for a common mechanism whereby learned odors evoke both a greater excitatory mitral cell (glutamate) and greater inhibitory granule cell (GABA) responses in the olfactory bulb. Unlike studies in conscious mice (Brennan et al., 1995) and sheep (Kendrick et al., 1992) there was no significant increase in the ratio of glutamate to GABA associated with olfactory learning in the current study, although there was a trend in this direction with the change in learned odor-evoked GABA release being slightly greater than for glutamate. This might reflect an impact of anesthesia in weakening learning-induced facilitation of mitral to granule cell communication via enhanced dendrodendritic synaptic sensitivity and/or the different learning paradigm used compared with other studies.

The findings from multi-array electrophysiological recordings provide further evidence that learned associations with complex odorants result in an altered response profile across an extensive population of mitral cell neurons. Before training by exposure to ginger and CS<sup>2</sup> together, responses (both in multiunit and single neuron activity) to each food odor (plain, ginger, coriander) were similar, and also similar to those for CS<sup>2</sup> presented alone, and no differential effects of odors were seen on breathing rate. Thus, CS<sup>2</sup> itself did not evoke greater changes than other odors and its presentation was only accompanied by an elevated response when it was paired with the ginger food odor. After training, increased responsiveness was maintained only for the ginger food odor and this elevated response to the training odor was rapid, occurring shortly after its combination with CS2. In agreement with single neuron recordings from the olfactory bulb of the conscious maternal sheep which had learned to recognize the odors of their lamb (Kendrick et al., 1992), the cued odor primarily evoked an increase in the proportion of neurons exhibiting excitatory responses. While we cannot rule out the possibility that some of the neurons recorded by the array were inhibitory granule cells, the electrodes are likely to have been primarily mitral cells. This assumption is based on (a) the depth of penetration of the microelectrode array; and (b) the large signal to noise ratio of the spikes detected which is characteristic of the mitral cell rather than the granule cell layer; and (c) since granule cells do not have an axon they are less likely to generate "spikes" which can be recorded extracellularly. As such the activity detected represents the output from the granule cell layer, where the learning processes are thought to take place in the network between inhibitory granule cells and mitral cell dendrites (Brennan and Keverne, 1997; Brennan and Kendrick, 2006; Sanchez-Andrade and Kendrick, 2010). Thus it would appear likely that learning primarily increases responses of glutamatergic mitral cell neurons to the learned odor, which is consistent with increased *in vivo* glutamate release.

Despite evidence for enhanced release of GABA in the olfactory bulb during exposure to the learned odor this does not appear to be associated with extensive reductions in mitral cell responses to learned odors in the olfactory bulb of the anesthetized mouse, in the current study, or in the conscious sheep (Kendrick et al., 1992). This contrasts with conscious recordings from the mouse AOB in the pregnancy block model where mating produced a reduction in mitral cell responses to the learned odors from the urine of a male (Binns and Brennan, 2005). However, this may reflect the fact that in the latter model it is proposed that pregnancy block does not occur in response to the familiar male because his odors fail to induce increases in mitral cell responses which would promote changes in the hypothalamo-pituitary axis and subsequent pregnancy termination (see Brennan and Kendrick, 2006). For the main olfactory system on the other hand it is assumed that learned odors with positive associations can more effectively promote behavioral and endocrine changes through an enhanced response in mitral cells having a greater impact on the relevant downstream projection region (see Brennan and Kendrick, 2006; Sanchez-Andrade and Kendrick, 2010). Possibly the parallel enhancement of inhibitory GABA release in this case is more important for helping to minimize interference from other, potentially competing odors, by reducing responses of other adjacent mitral cells. Indeed, the finding in the current study of a trend towards an increased proportion of mitral cells exhibiting reduced responses to the uncued odor supports such a possibility, and similar evidence for reduced mitral cell responses to other odors has also been observed in the sheep olfactory bulb (Kendrick et al., 1992). In such a system, increased lateral inhibition through elevated GABA release could also contribute to facilitated responses to the cued odor. Interestingly, similar evidence for increased inhibitory responses to unlearned stimuli has been reported in the chick brain following visual imprinting (Nicol et al., 1995; also Nicol et al., submitted). Enhanced inhibition of mitral cells via granule cells is associated with improved discrimination between odors (Abraham et al., 2010), and thus learning-associated changes in odor-evoked GABA release may reflect an improved ability to discriminate learned odors from other potentially competing ones.

In the context of the current study the dual influence of training in enhancing responses to the cued (ginger) odor but reducing them to an uncued (coriander) one suggests that the changes in responsiveness to ginger food odor were not simply due to sensitization following exposure, as similar exposure to each food odor induced contrasting results. Further, short-term changes in responsiveness were not apparent in pre-training exposures to any food odor and no significant changes were observed in responses across the 10 trials with each odor.

The complex odors used in our study influenced the activity of a large proportion of mitral cells recorded by the recording arrays in line with a recent study showing dense representation of natural odorants in the mouse olfactory bulb (Vincis et al., 2012). On the other hand learning associated changes were localized to mitral cell populations in the dorsal posterior rather than anterior part of the olfactory bulb suggesting a degree of spatial localization, at least in terms of neural plasticity changes. *In vivo* neurochemical changes associated with learning were also found in posterior regions of the olfactory bulb although we did not investigate whether they were absent in more anterior regions. Possibly our electrophysiological findings may reflect differential localization of processing for innate compared with learned odors reported previously in the mouse olfactory bulb (Kobayakawa et al., 2007). This is in accordance with findings that the necklace glomeruli receiving input from the specialized guanylyl cyclase GC-D sensory neurons in the posterior region of the olfactory bulb responds to social signals like CS<sup>2</sup> and mediate social transmission of food preference in mice (Munger et al., 2010). Perhaps more anterior populations of dorsal mitral cells are innately tuned to other classes of key biological odorants which exhibit less plasticity in response to new learned associations. Clearly such a possibility requires further investigation.

A limitation of our study is that the large differences in scale for olfactory bulb activity changes made using microdialysis sampling and recording microelectrodes make it difficult to draw anything other than tentative conclusions linking neurotransmitter changes with altered mitral cell responses. It is also important to emphasize the temporal differences in the two different sampling approaches as well as the fact that in the microdialysis experiment changes were sampled in the olfactory bulb 24 h after learning whereas for the recording experiments changes were restricted to a relatively short period before, during and immediately after learning. Thus the pattern of neurochemical changes measured would therefore have reflected post-consolidation learning whereas the electrophysiological changes are associated with initial acquisition of learning and may have been further modified post- consolidation. However, other studies of consolidation in visual recognition memory suggest that neurons recruited during in acquisition are likely to be those maintained post-consolidation (Horn et al., 2001; Jackson et al., 2008). Recordings were also restricted to the dorsal mitral cell layer and other patterns of response might have occurred in lateral and ventral layers.

Our findings provide further evidence that the mammalian brain is capable of supporting some forms of learning under anesthesia, although to the best of our knowledge this is the first study to show detailed neurochemical and neurophysiological plasticity changes occurring in a primary sensory processing region. A converging body of evidence has also implicated sleep as being crucial in memory formation (Ellenbogen et al., 2006; Arzi et al., 2012), and specifically in memory consolidation (Walker and Stickgold, 2004; Rasch et al., 2007; Jackson et al., 2008), and thus learning-related changes during unconscious states would appear to be both important and complimentary to those occurring during conscious ones. While we have observed broadly similar patterns of neural plasticity changes in the olfactory bulb under anesthesia to those known to occur under conditions of conscious learning more detailed comparisons of learning using the same paradigm under conscious and unconscious conditions are required before we can establish whether certain differences exist. In particular factors such as the duration and specificity of memory and neural plasticity changes need further investigation. A further interesting question is whether the established role of neurogenesis in some learning paradigms involving both the hippocampus and olfactory bulb (Lazarini and Lledo, 2011) can also occur under anesthesia or other unconscious states.

### **CONCLUSIONS**

Overall we have provided the first in vivo neurochemical and neurophysiological evidence for olfactory learning under anesthesia in mice being supported by broadly similar changes in olfactory bulb encoding as shown previously in conscious paradigms and involving potentiated glutamate and GABA release from mitral and granule cell synapses. Observed changes support the presence of dense, but also spatially restricted representation of changes with the olfactory bulb associated with learned associations with complex odorants. The pattern of changes found reveal enhanced excitatory responses in mitral cell populations but also corresponding reductions in responses in those tuned to other, potentially interfering odors.

#### **AUTHOR CONTRIBUTIONS**

Alister U. Nicol, Gabriela Sanchez-Andrade, and Keith M. Kendrick conceived and designed the study; Alister U. Nicol, Gabriela Sanchez-Andrade and Paloma Collado carried out the experimental work; Alister U. Nicol, Gabriela Sanchez-Andrade, Anne Segonds-Pichon and Keith M. Kendrick analyzed and interpreted the data; Alister U. Nicol, Gabriela Sanchez-Andrade, Anne Segonds-Pichon, Paloma Collado and Keith M. Kendrick drafted and approved the final version of the paper.

### **ACKNOWLEDGMENTS**

This work was supported by the Biotechnology and Biological Sciences Research Council (BBSRC) UK. Paloma Collado was supported by grant: MEC: PR2005-0081.

#### **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: 30 March 2014; accepted: 09 May 2014; published online: 05 June 2014*. *Citation: Nicol AU, Sanchez-Andrade G, Collado P, Segonds-Pichon A and Kendrick KM (2014) Olfactory bulb encoding during learning under anesthesia. Front. Behav. Neurosci. 8:193. doi: 10.3389/fnbeh.2014.00193*

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

*Copyright © 2014 Nicol, Sanchez-Andrade, Collado, Segonds-Pichon and Kendrick. 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*.

## Olfactory systems and neural circuits that modulate predator odor fear

### *Lorey K. Takahashi\**

*Department of Psychology, University of Hawaii at Manoa, Honolulu, HI, USA*

#### *Edited by:*

*Regina M. Sullivan, NYU School of Medicine, USA*

#### *Reviewed by:*

*Jeffrey B. Rosen, University of Delaware, USA Gordon A. Barr, Children's Hospital of Philadelphia, USA*

#### *\*Correspondence:*

*Lorey K. Takahashi, Department of Psychology, University of Hawaii at Manoa, 2530 Dole St., Sakamaki C400, Honolulu, HI 96822, USA e-mail: lkt@hawaii.edu*

When prey animals detect the odor of a predator a constellation of fear-related autonomic, endocrine, and behavioral responses rapidly occur to facilitate survival. How olfactory sensory systems process predator odor and channel that information to specific brain circuits is a fundamental issue that is not clearly understood. However, research in the last 15 years has begun to identify some of the essential features of the sensory detection systems and brain structures that underlie predator odor fear. For instance, the main (MOS) and accessory olfactory systems (AOS) detect predator odors and different types of predator odors are sensed by specific receptors located in either the MOS or AOS. However, complex predator chemosignals may be processed by both the MOS and AOS, which complicate our understanding of the specific neural circuits connected directly and indirectly from the MOS and AOS to activate the physiological and behavioral components of unconditioned and conditioned fear. Studies indicate that brain structures including the dorsal periaqueductal gray (DPAG), paraventricular nucleus (PVN) of the hypothalamus, and the medial amygdala (MeA) appear to be broadly involved in predator odor induced autonomic activity and hypothalamic-pituitary-adrenal (HPA) stress hormone secretion. The MeA also plays a key role in predator odor unconditioned fear behavior and retrieval of contextual fear memory associated with prior predator odor experiences. Other neural structures including the bed nucleus of the stria terminalis and the ventral hippocampus (VHC) appear prominently involved in predator odor fear behavior. The basolateral amygdala (BLA), medial hypothalamic nuclei, and medial prefrontal cortex (mPFC) are also activated by some but not all predator odors. Future research that characterizes how distinct predator odors are uniquely processed in olfactory systems and neural circuits will provide significant insights into the differences of how diverse predator odors activate fear.

**Keywords: predator odor, fear, main and accessory olfactory systems, amygdala, hippocampus, medial hypothalamus, medial prefrontal cortex**

### **BACKGROUND**

Fear and anxiety are activated by threat and the ability to regulate their responses is essential to adaptation and survival. Moreover, an extensive body of work indicates that abnormalities in the detection of threat may lead to pathological fear and anxiety (Lang et al., 2000; Charney, 2004; Green and Phillips, 2004; Blanchard et al., 2011; Britton et al., 2011). Thus, the biology of fear has attracted considerable attention in relation to the causal and modulatory factors linked to normal and exaggerated fear and anxiety states.

Animal models of fear that involve exposing prey to predator odor offer fundamental insights into the biology of threat detection and behavioral expression. Many species depend on olfactory sensory systems to detect a predator and engage in antipredator behavior. Chemosensory cues or predator odors may at times be the only information available to prey that are hiding, under cover, or unable to visually detect the source of threat. Wide-ranging field and laboratory studies have discussed the diverse behavioral repertoire prey animals display when predator chemosensory cues are detected (Kats and Dill, 1998; Apfelbach et al., 2005) and the decision-making consequences prey exhibit under of risk of predation (Lima, 1998; Bytheway et al., 2013). In addition, burgeoning research is focusing on the neurobiology of predator-induced stress. In particular, exposing mice and rats to predator odor stimuli, which induces long-lasting behavioral and physiological effects, are increasingly becoming a useful, animal model that may offer insights into the pathophysiology of humans undergoing uncontrollable stress and anxiety as in posttraumatic stress disorder (Mackenzie et al., 2010; Clinchy et al., 2011; Corley et al., 2012; Matar et al., 2013).

This review highlights current research on the olfactory and neural systems that activate fear elicited by predator odors. The review begins with an overview of the olfactory systems that detect odors derived especially from predator fur, urine, and synthesized chemosignals from predators. Interconnections from olfactory systems to brain circuits activated by predator odor will then be discussed in relation to autonomic, endocrine, and fear-related unconditioned and conditioned fear.

#### **DIVERSE OLFACTORY SYSTEMS DETECT PREDATOR ODORS**

The olfactory system consists of several subsystems (Breer et al., 2006; Munger et al., 2009) that include the main olfactory system (MOS), the accessory olfactory system (AOS), the septal organ of Masera, the Grueneberg ganglion (GG), and the trigeminal system. Research on predator odor detection has focused almost entirely on the two well-known MOS and AOS. A long-standing view is the MOS serves as a broadly tuned odor sensor that responds to a multitude of volatile, airborne chemicals that convey information such as the location of food and the whereabouts of predators and prey (Firestein, 2001). In contrast, the AOS mediates innate responses to non-volatile, fluid-phase chemical cues or pheromones (Meredith, 1991; Breer et al., 2006), which are chemicals released from one organism that influence the behavior or physiology of another organism of the same species (Karlson and Luscher, 1959). Pheromones likely evolved to facilitate intraspecific communication such as the reproductive and social status of a conspecific (Dulac and Torello, 2003). However, the role of the AOS has now broadened to include not only intraspecific communication but also interspecific signaling. That is, the AOS also detects the odor of non-conspecifics such as predators (Ben-Shaul et al., 2010; Papes et al., 2010) and these interspecific signaling odors are referred to as kairomones (Dicke and Grostal, 2001). Thus, both the MOS and AOS have the potential to detect predator odor and activate unconditioned fear behavior. This review section discusses our expanding knowledge of the different olfactory subsystems, especially in laboratory mice and rats, in processing predator odors.

### **THE MAIN OLFACTORY SYSTEM**

The MOS consists of olfactory sensory neurons in the main olfactory epithelium (MOE) that express odorant receptors (ORs) and trace amine associated receptors (TAARs) (Buck and Axel, 1991; Liberles and Buck, 2006) and project to the MOB for further olfactory information processing. The MOB then projects directly or indirectly to brain regions that modulate physiological and behavioral functions.

### **THE MOS OF THE RAT AND MOUSE APPEARS ESPECIALLY SENSITIVE IN PROCESSING THE PREDATOR ODOR TRIMETHYLTHIAZOLINE**

Although only a few studies have investigated the role of the rat MOS in detecting predator odors, an important observation is the MOS does not process all types of predator odors. For example, exposure to cat odor obtained from the body or fur induces only a modest increase in Fos expression in the glomerular cell layer in the MOB (McGregor et al., 2004). In contrast, 2,4,5 dihydro 2,5 trimethylthiazoline (TMT), a highly volatile synthesize compound from red fox anal secretions induces a pronounced increase in Fos expression in both the granular layer and dorsal lateral portion of the MOB glomerular layer (Illig and Haberly, 2003; Day et al., 2004; Staples et al., 2008). An early study also reported that rats exposed to the odor of fox feces exhibited elevations in olfactory bulb mitral cell multiunit responses that were accompanied by increased vigilance, as indicated by heightened EEG and neck muscle EMG recordings and freezing behavior (Catterelli and Chanel, 1979). A role of the MOB in processing TMT is further supported in a study showing that olfactory bulb ablation effectively reduces freezing to TMT (Ayers et al., 2013).

Consistent with the limited work in rats, a number of studies in mice have further supported a role of the MOS in detecting TMT. Studies showed that intranasal perfusion of zinc sulfate, which induces transient anosmia of the MOS (McBride et al., 2003), reduces freezing and avoidance to TMT (Hacquemand et al., 2010; Galliot et al., 2012). Using genetic methods, a study reported that mutant mice with dorsal epithelium zone depletion of olfactory neurons exhibited deficits in avoidance behavior when exposed to TMT (Kobayakawa et al., 2007). Another study in mice showed that altered olfactory sensory neuron projections from the dorsal region of the olfactory epithelium to the dorsal olfactory bulb impaired avoidance behavior to TMT (Cho et al., 2011). These results suggest that a circuit in the MOS from the dorsal olfactory epithelium region to the dorsal olfactory bulb play a critical role in processing TMT.

In addition to TMT, the urine of predators is detected by the mouse MOS. One study identified an involvement of TAARs expressed in neurons of the MOE in detecting predator urine odors (Dewan et al., 2013). More specifically, this study found that genetic deletion of the olfactory TAAR gene family abolished the aversion of mice to low concentrations of volatile amines and to the odor of predator urine. Of interest, analysis of predator urine identified the volatile amine β-phenyethylamine (PEA), which is recognized by TAAR4 (Ferrero et al., 2011). Of particular relevance, mice lacking the olfactory TAAR4 exhibit deficits in their aversion to PEA (Ferrero et al., 2011; Dewan et al., 2013). These studies demonstrate predator urine contains a volatile compound, i.e., PEA, detected by a specific receptor in the MOE that induces predator fear-related behavior.

### **THE ACCESSORY OLFACTORY SYSTEM**

The AOS consists of the vomeronasal organ (VNO), a chemoreceptive structure situated at the base of the nasal septum, which houses the microvillar vomeronasal sensory neurons (VSNs). Pheromones are detected by three classes of VSN receptors including vomeronasal receptor type 1 (V1R) and type 2 (V2R) and formyl peptide receptors (Munger et al., 2009; Liberles, 2014). The VNO sends projections to the accessory olfactory bulb (AOB), a forebrain region that serves as the first processing center of vomeronasal information. The AOB then sends information directly or indirectly to a number of brain sites involved in diverse functions.

### **THE RAT AND MOUSE AOS MAY BE NECESSARY BUT NOT SUFFICIENT IN MEDIATING PREDATOR ODOR PROCESSING**

TMT and cat odor are prominently used to investigate the roles of the MOS and AOS in predator odor detection. As previously indicated, rats exposed to TMT, but not cat odor, showed a significant increase in Fos expression in the glomerular cell layer in the MOB. However, in striking contrast, exposure to cat odor, but not TMT, induces a robust increase in Fos expression in the glomerular, mitral, and granule cell layers of the posterior AOB (Staples et al., 2008). These results suggest that TMT is detected primarily by the MOS whereas cat odor is detected by AOS. Cat odor derived from the fur/body appears to act as a kairomone processed by the AOS.

To further assess the potentially independent role of the MOS and AOS in processing predator odor, an interesting study (Masini et al., 2010) exposed rats after destruction occurring in either the olfactory epithelium or VNO to ferret odor collected on a small towel. The study showed that rats rendered anosmic with zinc sulfate or with ablation of the VNO continued to show elevations in stress-induced corticosterone secretion when exposed to ferret odor. Only after joint destruction of the MOB and VNO did significant deficits occur in ferret odor-induced secretion of corticosterone. These results suggest the MOS and AOS have overlapping roles in modulating the ferret odor-induced increase in stress hormone secretion. Due to the reported role of the rat AOB in detecting cat odor as indicated by increased Fos expression (McGregor et al., 2004), removal of the VNO in rats would be useful to determine whether the remaining functional MOS may similarly modulate the processing of fear-related behavior induced by cat fur/body odors.

Several lines of evidence in mice also support an involvement of the VNO chemosensory system in detecting predator odor and activating fear. First, mice exposed to the urine of predators, (e.g., bobcat, fox, rat) exhibit a robust increase in electrical activity in the AOB (Ben-Shaul et al., 2010). Second, removal of the VNO or genetically impairing activation of the VNO increases the time mice spent investigating a collar worn by a cat (Samuelsen and Meredith, 2009) or the bedding of a rat (Isogai et al., 2011). Third, VN1 and VN2 receptors appear to have unique roles in detecting and identifying different types of potential predators, e.g., mammalian, reptiles, or predatory birds (Isogai et al., 2011).

The VNO role in predator odor processing was further investigated in mice lacking the TrpC2, which is the primary signal transduction channel of VNO sensory neurons. This study showed that TrpC2−/<sup>−</sup> mutants exhibited deficits in avoidance and risk assessment behavior when exposed to cat odor (neck swab), snake (shed skin), and rat urine (Papes et al., 2010). Furthermore, consistent with the research implicating the olfactory dorsal epithelium in processing TMT-induced fear in mice (Kobayakawa et al., 2007), TrpC2−/<sup>−</sup> mutants exposed to TMT displayed fear behavior. Thus, in the mouse and to some extent in the rat, the MOS and AOS specifically detect different types of predator odors to facilitate fear.

Of relevance in identifying specific kairomone chemosignals that activate the VNO to induce fear, investigators isolated in cat saliva the protein Feld4, a cat homolog belonging to the mouse and rat major urinary protein (MUP) family (Smith et al., 2004). The MUP family also consists of closely related proteins excreted by rodent exocrine glands (Cavaggioni and Mucignat-Caretta, 2000). Mice exposed to Feld4 exhibit an increase in both c-Fos expression in AOB and defensive behavior (Papes et al., 2010). In addition, exposure to the recombinant Feld4 ligand facilitated fear behavior and stress hormone secretion in wild type but not TrpC2−/<sup>−</sup> mice with deficits in VNO signaling. The MUP ligand, cat Feld4, appears to be a chemosensory signal detected by the VNO that triggers fear in the mouse. The potential presence of Feld4 on cat fur/body during cat grooming may be the basis of the chemosensory signal that activates fear in the mouse. Generalization of the fear-eliciting effects of Feld4 from the mouse to the rat and other prey species requires further investigations.

#### **THE MOUSE GRUENEBERG GANGLION DETECTS A VOLATILE THAT SIGNALS DANGER**

In mice, the GG is a small olfactory structure located on the tip of the nose that detects odors via V2Rs, several TAARs, and ORs (Fleischer and Breer, 2010). GG neurons send axonal projections to dorsal regions of the caudal MOB (Munger et al., 2009). The mouse GG is reported to detect the intraspecific predator odor TMT and conspecific alarm pheromones or odors released when mice are killed, injured, or threatened (Brechbühl et al., 2008, 2013). Threat-induced activation of both conspecific alarm pheromones and TMT via the GG may be linked to the detection of sulfur-containing related volatiles generated by meat digestion of potential predators (Nolte et al., 1994; Brechbühl et al., 2013). These results indicate a dual role of the GG in detecting a volatile produced by both conspecifics and interspecifics that signals danger.

To determine the specific involvement of GG in activating threat, investigators found that sectioning GG axonal projections, which are sent to the MOB (Fuss et al., 2005; Koos and Fraser, 2005), reduced freezing but not risk assessment behavior when mice were exposed to volatile alarm pheromones and TMT (Brechbühl et al., 2013). The researchers suggest that although GG function is impaired, these odorants are still detected, likely by receptors in the MOE (Kobayakawa et al., 2007), which send axonal projections to MOB regions distinct from GG projections (Mamasuew et al., 2011), to maintain activation of some features of fear-related behavior.

#### **THE POTENTIAL ROLE OF THE SEPTAL ORGAN OF MASERA IN MEDIATING PREDATOR ODOR HAS NOT BEEN DETERMINED**

Situated bilaterally in the septal wall between the caudal end of the VNO and the rostrally located MOE lie the septal organ of Masera (SO) (Ma, 2007). The location of the SO in the nasal cavity air path of many rodent species was hypothesized to serve as a general odor detector that alerts the individual (Giannetti et al., 1995) or plays a role in assessing food or social odors (Breer et al., 2006). Of potential interest, in mice, the SO innervates glomeruli in the olfactory bulb from the MOE (Lèvai and Strotmann, 2003), which may have relevance in detecting predator odors. An early study investigated the potential arousing or alerting function of the SO by exposing intact and SO lesioned rats to TMT and food odors as well as non-biological odors such as eucalyptol (Giannetti et al., 1995). Results indicated that both biologically relevant and meaningless odors induced similar awakening and habituation responses in intact and SO lesioned groups, which suggest the SO does not have a unique alerting functional role.

#### **THE TRIGEMINAL SYSTEM**

The trigeminal system is essential in protecting against toxic or irritating odors by triggering reflexes such as apnea and sneezing. In mice and rats, noxious, pungent odors activate the trigeminal system via a large population of chemosensory cells that reach the surface of the nasal epithelium to form synaptic contacts with trigeminal afferent nerve fibers (Munger et al., 2009).

### **THE TRIGEMINAL SYSTEM IS ACTIVATED BY HIGH DOSES OF TMT BUT MAY NOT BE INVOLVED IN MEDIATING UNCONDITIONED FEAR BEHAVIOR**

Some investigators have suggested that TMT should be classified as a general noxious odorant that activates the trigeminal system to trigger avoidance behavior (McGregor et al., 2002; Apfelbach et al., 2005; Fortes-Marco et al., 2013). Evidence suggesting a noxious, irritating role of TMT is based on work showing that exposing rats to low concentrations of TMT (35μmol) is capable of activating the external lateral parabrachial nucleus of the trigeminal sensorial system (Day et al., 2004). In addition, 75μmol or more of TMT stimulates stress hormone ACTH and corticosterone secretion. TMT was also reported to induce nausea in humans (Fendt et al., 2005a). Although many studies were conducted with application of large TMT doses, when low doses are used, investigators observed a pattern of murine behavior akin to unconditioned fear-related responses such as freezing and avoidance (Wallace and Rosen, 2000; Blanchard et al., 2003b; Endres et al., 2005; Hacquemand et al., 2013).

To date, only one study has specifically manipulated the trigeminal system to determine its role in mediating the effects of TMT (Ayers et al., 2013). In this study, rats with trigeminal nerve transection were exposed to TMT (300μmol) or the noxious irritant butyric acid (900μmol). Results indicated that although trigeminal deafferentation effectively impaired freezing to butyric acid, TMT exposed rats were spared and exhibited freezing. On the basis of these observations, TMT odors appear to activate fear via an intact MOS and not through the trigeminal system.

#### **SUMMARY OF THE ROLE OF OLFACTORY SYSTEMS IN MODULATING PREDATOR ODOR FEAR**

In recent years, a number of investigators have begun to unravel the complex olfactory systems that process predator odors. A general conclusion of these studies is that different olfactory subsystems appear to have distinct roles in detecting different predator odors (see summary **Table 1**). For example, receptor systems in the MOS play a key role in mediating the sensory detection of TMT and compounds found in predator urine that activate fear. On the other hand, the AOS appears especially sensitive in processing cat odor or a potential kairomone compound found in cat saliva. However, the majority of studies that yielded insights into the specific role of olfactory subsystems were obtained almost exclusively in the mouse and broad generalization of these novel observations to other small prey species remains to be determined.

Another conclusion is that different olfactory subsystems may have complex overlapping effects in processing predator odors. Studies from the extensive sociosexual literature demonstrate that both the MOS and AOS are capable of detecting and processing both volatile and non-volatile chemosignals (Restrepo et al., 2004; Shepherd, 2006; Spehr et al., 2006; Martinez-Marcos, 2009) but differ in their signaling properties (Xu et al., 2005). Furthermore, although isolated compounds may be specifically detected by one olfactory subsystem, in natural predator odor exposure conditions complex chemosensory compounds may be present and processed by several olfactory subsystems to facilitate unconditioned fear. For example, studies showed the volatile predator urine amine PEA is detected by TAAR4 found in MOE neurons (Ferrero et al., 2011), whereas the nonvolatile kairomone MUP ligand cat Feld4 is detected by VNO neurons (Papes et al., 2010). However, the natural predator scent from the fur/body of cats and ferrets may contain a complex mixture of volatile and non-volatile sensory cues detected by the rodent MOS or AOS (see Masini et al., 2010) to induce unconditioned fear.

Similar overlapping olfactory subsystems may apply to the GG and MOS, where both olfactory subsystems express TARRs, and disruption in GG axonal projections to the MOB do not incur pronounced deficits in predator odor TMT-induced fear. Expression of V2Rs in the GG may play yet another complex role in detecting kairomone-like molecules. Thus, the complex predator body odor fur may be detected by multiple olfactory subsystems to facilitate broad activation of brain circuits that modulate unconditioned fear.

### **PREDATOR ODOR ACTIVATES BRAIN CIRCUITS THAT MODULATE AUTONOMIC, ENDOCRINE, AND FEAR-RELATED RESPONSES**

This section of the review highlights some of the key brain regions linked to olfactory systems, which modulate three major components—autonomic, endocrine, and behavior—of predator odor fear or threat.

### **EXPOSURE TO PREDATOR ODORS ACTIVATE THE AUTONOMIC NERVOUS SYSTEM**

Exposure to threat activates the autonomic nervous system (ANS) to rapidly trigger physiological responses that facilitate immediate survival in a dangerous situation (Ulrich-Lai and Herman, 2009). Secretion of the catecholamine epinephrine and norepinephrine from the sympatho-adrenomedullary system increases heart rate, vasoconstriction, and energy mobilization to support the classic "flight or fight" response. In addition, activation of peripheral β-adrenoceptors on vagal afferents that terminate in the nucleus of the solitary tract located in the brainstem may influence locus coeruleus pathways that secrete NE throughout the brain to further support physiological, behavioral and cognitive functions (Jöels et al., 2006). Here, the few studies that examined the effects of predator odor on ANS activity are discussed (**Table 2**).

One study reported that urethane-anesthetized rats exposed to either TMT or the pungent non-predatory odor butyric acid showed a dramatic increase in adrenal sympathetic nerve activity only to TMT (Horii et al., 2010), which likely increased epinephrine section and elevated blood pressure and heart rate. The authors also showed that live rats exposed to TMT displayed an increase in freezing and a reduction in exploratory behavior. In a subsequent study (Horii et al., 2013), this research group demonstrated that rats exposed to TMT exhibited elevations not only in adrenal sympathetic nerve activity but also in body temperature, an indication of increased metabolism and another measure of elevated autonomic activity (Bouwknecht et al., 2007).

#### **Table 1 | Roles of the olfactory systems in modulating predator odor fear.**


#### **Table 2 | Effects of predator odor on autonomic and endocrine functions.**


Cardiovascular functions were also studied in rats exposed to cat odor emitted from a collar worn by a cat (Dielenberg and McGregor, 2001). In this study, cat odor exposure induced a sustained increase in blood pressure but not heart rate. These autonomic responses were accompanied by a reduction in exploration, avoidance of the cat collar and heightened head out and riskassessment activity.

Information is scarce on the role of specific neural sites underlying predator odor-induced autonomic activity. However, one brain region of interest is the dorsal periaqueductal gray (DPAG), which was implicated a number of years ago to modulate cardiovascular functions (Carobrez et al., 1983; Schenberg et al., 1983; Depaulis et al., 1992). Previous studies reported that rats exposed to cat odor exhibit an increase in Fos-positive cells in the DPAG (Dielenberg and McGregor, 2001; McGregor et al., 2004). To investigate the role of the DPAG in cat odor-induced ANS activation, rats with DPAG lesions were implanted with telemetric probes to measure heart rate and blood pressure (Dielenberg et al., 2004). When exposed to cat odor, DPAG lesioned rats showed a reduction in heart rate and locomotor activity, but no significant decrease in the cat odor-induced rise in blood pressure. The results indicate the DPAG modulates some components of the ANS activated by cat odor.

Although TMT exposure increases sympathetic nerve activity (Horii et al., 2010, 2013), the specific role of the DPAG on ANS functions are not known. Unlike cat odor showing increased Fos expression in the DPAG, exposure to TMT is not accompanied by increases in c-*fos* mRNA (Day et al., 2004), albeit recent functional magnetic resonance imaging revealed a TMT-induced increase in neural activity in the DPAG (Kessler et al., 2012). Perhaps the DPAG modulates ANS functions activated by TMT as reported with cat odor (Dielenberg et al., 2004).

### **EXPOSURE TO PREDATOR ODORS RAPIDLY ACTIVATES THE NEUROENDOCRINE STRESS SYSTEM**

In addition to facilitating of the ANS, a number of studies demonstrated that predator odor activates the hypothalamicpituitary-adrenal system (HPA) in mice and rats (see **Table 2**). For example, exposure to TMT (Morrow et al., 2000; Day et al., 2004; Kobayakawa et al., 2007), cat odor (File et al., 1993; Cohen et al., 2006; Muñoz-Abellán et al., 2011), ferret odor (Masini et al., 2005) or 2-propylthietane, the main constituent of weasel anal gland secretion (Perrot-Sinal et al., 1999) increases adrenocorticotropin (ACTH) or corticosterone secretion.

The hypothalamic paraventricular nucleus (PVN) plays an important integrative role in stress by sending neuronal projections to the median eminence to regulate pituitary-adrenal hormone secretion (Sawchenko et al., 1996; Herman et al., 2002). The PVN also sends projections to brainstem sites including the parabrachial nucleus, the dorsal motor nucleus of the vagus nerve and the nucleus of the solitary tract to regulate autonomic activity (Swanson and Kuypers, 1980).

In relation to predator odor-induced HPA activation, the PVN receives information from the medial amygdala (MeA), a recipient of both direct and indirect MOS and direct AOS projections (Petrovich et al., 2001; Meredith and Westberry, 2004; Pro-Sistiaga et al., 2007). Rats with fiber-sparing lesions of the MeA exhibit significant deficits in ACTH and corticosterone secretion when exposed to ferret odor (Masini et al., 2009). However, the role of the MeA in facilitating HPA hormone secretion is not specific to predator odor. Impairment of the MeA is also reported to attenuate stress hormone secretion induced by restraint stress (Dayas et al., 1999).

Another study found that mice with dorsal epithelium zone depletion of olfactory neurons impaired ACTH secretion when exposed to TMT (Kobayakawa et al., 2007). This impairment in TMT detection and processing may disrupt a MOS-MeA-PVN circuit responsible for facilitating HPA hormone secretion. The study further indicated that dorsal epithelium zone depleted mice showed reductions in TMT-induced Zif268-positive cells in the bed nucleus of the stria terminalis (BST), anterior division, dorsal medial nucleus (BSTdm), which may reflect a disruption in a BST to PVN circuit that facilitates stress hormone secretion (Kobayakawa et al., 2007). The BST also receives direct projections from the AOB and indirect projections from the MeA (Scalia and Winans, 1975; Canteras et al., 1995; Fan and Luo, 2009) and impairments in potential odor processing in the BST may compromise activation of both HPA and ANS functions (Ulrich-Lai and Herman, 2009).

### **THE AMYGDALA, ESPECIALLY THE MEDIAL AMYGDALA, PLAYS AN ESSENTIAL ROLE IN MODULATING PREDATOR ODOR UNCONDITIONED AND CONDITIONED FEAR**

A number of reviews have discussed the neural basis of fear (e.g., LeDoux, 2000; Rosen and Donley, 2006; Pessoa and Adolphs, 2010; Gross and Canteras, 2012). A common theme of these reviews emphasizes the importance of the amygdala in threat detection, the elicitation of fear behavior, and its role in modulating fear learning and memory.

As indicated previously, the MeA receives both direct and indirect projections from olfactory systems and modulates HPA stress hormone secretion induced by predator odor. MeA cells appear sensitive to cat odor as indicated in a study showing impairments in facilitating field excitatory post synaptic potentials in the MeA of rats after exposure to cat odor (Collins, 2011). Behavioral studies demonstrate in rats that fiber-sparing lesions or temporary inactivation the MeA dramatically impair unconditioned freezing when exposed to either cat odor (Li et al., 2004; Blanchard et al., 2005) or TMT (Fendt et al., 2003). Moreover, MeA lesioned rats approached and contacted the cloth that contained cat odor and additional studies indicated MeA lesions did not produce a general increase in locomotor activity or a major deficit in olfactory detection (Li et al., 2004).

The role of the MeA was also studied in relation to predator odor contextual fear consolidation and retrieval (Takahashi et al., 2007). Rats with MeA inactivation immediately after exposure to predator odor exhibited no deficits in the consolidation of contextual fear-related behavior. However, rats with acute inactivation of the MeA immediately prior to retrieval of contextual fear showed increased approach behavior to the apparatus sector that previously contained the cat odor cloth. Thus, the MeA may have a dual role in detecting predator odor, which activates unconditioned fear, and in recalling a previous location associated with predator odor. Of note, a study showed that MeA lesioned rats are capable of freezing to an auditory stimulus paired with footshock (Nader et al., 2001) and suggests that impairment of the MeA does not produce global deficits in learned fear behavior. Rather, the MeA lesion-induced deficit in conditioned fear behavior appears specific to impairments in predator odor-context associations.

The basolateral amygdala (BLA), which is widely implicated in fear conditioning using footshock as the unconditioned stimulus paired with auditory (Fanselow and Poulos, 2005), contextual (Maren et al., 2013), or olfactory cues (Otto et al., 2000; Mouly and Sullivan, 2010), has been the focus of studies that determined whether the BLA is also involve in associating the unconditioned predator odor stimulus with the test context. Using TMT, studies in rats demonstrated that BLA inactivation or fiber-sparing lesions did not induce robust impairments in unconditioned freezing (Wallace and Rosen, 2001; Müller and Fendt, 2006) and produced only mild deficits in conditioned freezing to the context (Wallace and Rosen, 2001). However, other investigators that exposed BLA lesioned rats to cat odor reported deficits in both unconditioned freezing and approach to the cat odor (Vazdarjanova et al., 2001; Takahashi et al., 2007). In addition, temporary inactivation of the BLA immediately after exposure to cat odor impaired contextual avoidance behavior when rats were tested the next day (Takahashi et al., 2007). Thus, the BLA, which is broadly involved in emotional memory consolidation (McGaugh, 2000), is further implicated in the consolidation of cat odor-induced contextual fear.

Unlike the MeA, the BLA in rats does not receive direct projections for either the MOS or AOS (McDonald, 1998; Pitkänen, 2000). Different olfactory detection systems accompanied by downstream indirect projections to the BLA may contribute to the reported differences between TMT and cat odor in unconditioned and conditioned fear. Of possible relevance, cat and ferret fur/body odor, but not TMT, increases c-Fos expression in the BLA (Day et al., 2004; Masini et al., 2005; Staples et al., 2008), which suggests predator odor differences in activating the BLA. Nonetheless, the temporal pattern of c-Fos expression is known to vary (Redburn and Leah, 1997), and a time-course study may reveal temporal increases in c-Fos expression activated by TMT. Furthermore, although the BLA may not modulate an increase in unconditioned fear elicited by TMT, the BLA may play an alternative role in modulating the arousing effects of TMT on dorsolateral striatal-dependent response learning (Leong and Packard, 2014).

The central nucleus of the amygdala (CeA) is another target of interest in predator odor studies due to the broad role of the CeA in modulating autonomic, endocrine, and anxiety and fear behavior (Davis, 2000; LeDoux, 2000). However, studies in rats involving fiber-sparring lesions or temporary inactivation of the CeA demonstrate that exposure to either TMT (Fendt et al., 2003) or cat odor (Li et al., 2004) are ineffective in attenuating unconditioned fear behavior. Thus, not all nuclei in the amygdala are involved in modulating predator odor-induced fear behavior.

#### **THE BED NUCLEUS OF THE STRIA TERMINALS MODULATES PREDATOR ODOR FEAR INDUCED BY OLFACTORY INFORMATION PROCESSED BY THE MAIN OLFACTORY SYSTEM**

Another major direct and indirect projection from olfactory systems is the BST as previously discussed. Studies revealed that inactivation of the BST, especially the ventral BST, reduces freezing when rats are exposed to TMT (Fendt et al., 2003, 2005b), or cat urine (Xu et al., 2012). Notably, the MOS plays a key role in processing both TMT (Kobayakawa et al., 2007) and predator urine (Ferrero et al., 2011; Dewan et al., 2013). Whether kairomone odors derived from predators such as from fur/body odors of the cat or ferret facilitate unconditioned fear behavior via the BST is not known, albeit exposure to both cat odor (Dielenberg and McGregor, 2001) and TMT (Day et al., 2004; Asok et al., 2013) appears to activate the BST as suggested by increased Fos expression.

#### **THE VENTRAL HIPPOCAMPUS MODULATES PREDATOR ODOR UNCONDITIONED AND CONDITIONED FEAR**

The hippocampus plays a prominent role in emotional behavior, especially in processing contextual fear information (Kim and Fanselow, 1992; Phillips and LeDoux, 1992; Zhang et al., 2001). The ventral hippocampus (VHC) has attracted attention in odor studies due to dense reciprocal connections to the MeA and to other amygdalar nuclei such as the cortical nucleus that receives input from the MOS (Scalia and Winans, 1975; McDonald, 1998). The VHC also projects to the AOB and the piriform cortex, a major target of the MOB (Shipley and Adamek, 1984; Illig and Haberly, 2003). In rats, both the olfactory bulb and hippocampal dentate gyrus respond to weasel gland secretion 2-propylthietane and TMT by exhibiting fast wave bursts (Heale et al., 1994). TMT exposure also increases c-*fos* mRNA in the hippocampal dentate gyrus (Day et al., 2004).

Concerning predator odor unconditioned fear behavior, rats with VHC, but not dorsal (DHC), hippocampal lesions exhibited deficits in freezing and crouching when exposed to cat odor (Pentkowski et al., 2006). When tested the next day for contextual fear, VHC lesioned rats continued to show deficits in freezing. Another study in mice exposed to coyote urine showed that VHC lesions impaired avoidance and risk assessment behavior (Wang et al., 2013). In addition, VHC lesioned mice exhibited less freezing than control mice in the contextual fear test. These investigators further reported that exposure to coyote urine activates place cells in the CA1 region of the DHC and modify their firing patterns to stabilize a spatial representation of the fear eliciting encounter (Wang et al., 2012). Together, these studies in the rat and mouse using cat odor and predator urine suggest the VHC plays a role in predator odor processing of unconditioned and contextual fear and the DHC may also contribute to processing contextual fear. However, additional research is required to determine the extent to which the MOS and AOS connected to the hippocampus have overlapping or distinct roles in modulating unconditioned and conditioned predator fear behavior.

### **EXPOSURE TO SPECIFIC PREDATOR ODORS MAY REQUIRE MEDIAL HYPOTHALAMIC NUCLEI TO ACTIVATE FEAR BEHAVIOR**

Another major projection from the MeA, especially from the posteroventral MeA region, and from the BST is to medial hypothalamic nuclei, which broadly regulate reproductive, ingestive, and defensive behavior (Risold et al., 1997). Distinct medial hypothalamic nuclei consisting of the anterior hypothalamic nucleus, dorsomedial part of the ventromedial nucleus, and dorsal premammillary nucleus (PMd) are hypothesized to underlie a medial hypothalamic defensive system (Canteras, 2002).

The PMd stands out in predator studies due to robust increases in Fos expression in rats exposed to cat or cat odor (Canteras et al., 1997; Dielenberg and McGregor, 2001) and dense connections to the PAG (Cezario et al., 2008), which is involved in ANS functions and behavioral expression. Behavioral studies show that PMd lesions significantly impair cat odor-induced fear behavior (Blanchard et al., 2003a; Canteras et al., 2008). Furthermore, βadrenoreceptor blockade in the PMd of rats prior to exposure to cat odor or prior to the context associated with cat odor effectively reduces freezing (Do Monte et al., 2008). These studies demonstrate an important role of the PMd in modulating the occurrence of predator odor unconditioned and conditioned fear.

However, some predator odors do not require the PMd to facilitate fear behavior. A study in rats showed that electrolytic lesions that damage both cells and fibers of passage through the PMd did not impair freezing to TMT (Pagani and Rosen, 2009). Furthermore, fiber-sparring lesions of the anterior and ventromedial hypothalamus failed to disrupt TMT-induced unconditioned freezing. The authors suggest the BST sends projections that pass through the anterior and ventromedial hypothalamic nuclei and circumvent the PMd to terminate in the PAG to modulate freezing expression.

This neural pathway from the BST to the PAG that circumvent connections to nuclei in the medial hypothalamus currently appears specifically activated by TMT. A study comparing the effects of cat odor and TMT on Fos expression found that exposure to cat odor, but not TMT, induced significant Fos expression in both the anterior and dorsal ventromedial hypothalamus as well and the PMd (Staples et al., 2008). This result suggests the medial hypothalamic defensive system is not broadly activated by all predator odors to modulate fear behavior.

#### **THE MEDIAL PREFRONTAL CORTEX MODULATES UNCONDITIONED FEAR ELICITED BY PREDATOR ODORS**

The medial prefrontal cortex (mPFC) is connected to a number of brain structures including the amygdala, hypothalamus, and periaqueductal gray (Gabbott et al., 2005; Price, 2005) that are involved in predator odor fear. In addition, the mPFC is critically involved in fear extinction (Sotres-Bayon et al., 2006; Peters et al., 2009; Marek et al., 2013).

Recent studies implicated the mPFC, consisting of the prelimbic and infralimbic cortex, in predator odor-induced unconditioned fear but the precise role of the mPFC in modulating predator odor fear is not clear. One study reported that temporary inactivation of the prelimbic region increased freezing in rats exposed to TMT (Fitzpatrick et al., 2011). However, another study showed that in 38–42 days old adolescent rats, inactivation of the prelimbic cortex impaired freezing induced by cat odor (Chan et al., 2011). Although both studies reported no significant effects of infralimbic cortex inactivation on freezing, the seemingly opposite effects of prelimbic cortex inactivation on predator odor unconditioned freezing induced by TMT and cat odor is puzzling.

In addition to the predator odor-induced behavioral differences involving the mPFC, studies indicate that exposure to cat odor activates c-Fos expression (Staples et al., 2008; Chan et al., 2011) in the mPFC, whereas no significant increases in mPFC c-Fos were found after exposure to TMT (Day et al., 2004; Staples et al., 2008; Asok et al., 2013). The mPFC of rats also showed elevations in expression of -FosB several days after exposure to cat odor (Mackenzie et al., 2010). In this study, expression of -FosB in the mPFC was associated with long-term effects of predator odor on conditioned fear. Another study measured *egr-1*, a gene transcription factor linked to learning and memory synaptic plasticity (Alberini, 2009). Of interest, rats exposed to TMT showed no significant increase in *egr-1* mRNA in the mPFC (Asok et al., 2013). Research will be required to determine how distinct predator odors such as TMT and cat odor are first processed in olfactory systems that project directly to brain structures for further processing before interacting with the mPFC to modulate unconditioned and conditioned fear behavior.

### **SUMMARY OF BRAIN CIRCUITS THAT MODULATE PREDATOR ODOR FEAR**

Several distinct and overlapping brain regions have been identified that play key roles in predator odor activation of autonomic, endocrine, and fear behavior responses. For example, the MeA, which receive direct projections from MOS and AOS, appears to have a necessary role in general predator odor activation of both HPA hormone secretion and unconditioned fear behavior (see **Tables 2**, **3**). In addition, the MeA and BLA are involved in the retrieval and consolidation, respectively, of predator odor contextual fear. The VHC, another key brain center that interacts with olfactory projection targets such as the MeA and piriform cortex, also appears to have a general role in modulating predator odor unconditioned and contextual fear behavior.

A number of investigators reported that exposure to cat odor induces robust contextual (Blanchard et al., 2001; Dielenberg and McGregor, 2001; Takahashi et al., 2005; Canteras et al., 2008) or auditory (Takahashi et al., 2008) fear conditioning in comparison to the less intense fear conditioning observed with TMT (Blanchard et al., 2003b; Fortes-Marco et al., 2013; Staples et al., 2008; see also ferret odor-induced conditioning in Masini et al., 2006). Notwithstanding the ability of TMT to induced contextual fear conditioning in rats tested under specific environmental conditions such as variations in test cage size (Rosen et al., 2008) or multiple pairings of TMT and the context (Endres and Fendt, 2007), behavioral testing with different predator odors are identifying some neural sites that may account for inconsistencies in displays of predator odor unconditioned and conditioned fear. For example, cat odor, but not TMT, appears to involve the BLA. Although the precise basis underlying this difference is not clear, the possibility exist that vomeronasal signal processing in the BLA may be necessary for predator odor fear conditioning. A study reported that non-volatile pheromones are especially attractive to female mice and the BLA stood out as a distinct region for vomeronal-olfactory associative learning (Moncho-Bogani et al., 2005). Furthermore, another study reported that darcin, an involatile protein sex pheromone in male mouse urine, rapidly induced conditioned preference in female mice that spent time in the site were darcin was previously detected (Roberts et al., 2012). Perhaps exposure to a predator odor kairomone with similar darcin-like properties and/or the MUP ligand, cat Feld4, will trigger rapid and robust fear conditioning in the BLA. Furthermore, predator odor fear conditioning may be enhanced by simultaneous activation of the MeA, VH, medial hypothalamic nuclei, and mPFC. That is, these nuclei are sensitive to the kairomone cat odor (see **Table 3**) and may


require activation in conjunction with the BLA when exposed to predator odor to facilitate unconditioned and conditioned fear.

In stark contrast to the extensively studied olfactory connected neural systems in reproductive behavior or the neural systems in emotional learning and memory, based largely on work involving the application of the unconditioned footshock stimulus, knowledge on biology of predator odor fear is limited. Future predator odor research should address not only the brain circuits that modulate unconditioned and conditioned fear but also the olfactory sensory structures such as the accessory and main olfactory bulbs that are implicated in important developmental, social, and reproductive learning and memory processing (Brennan and Keverne, 1997; Landers and Sullivan, 2012). A recent study in mice reported that fear learning to threat-predictive odors involved changes in synaptic output of olfactory sensory neurons, which suggests that emotional information can be encoded at the level of primary sensory processing (Kass et al., 2013). Thus, characterizing the processing of different predator odors occurring in olfactory structures in conjunction with the complex interconnected neural circuits that mediate both the physiology and behavior of fear will be required to provide a complete picture of how predator chemosignals are uniquely processed to facilitate adaptive fear-related defensive behavior.

### **REFERENCES**


basis of its processing. *Neurosci. Biobehav. Rev.* 29, 1145–1156. doi: 10.1016/j.neubiorev.2005.04.018


by medial, but not central amygdala lesions in rats. *Brain Res.* 1288, 79–87. doi: 10.1016/j.brainres.2009.07.011


agglutinin conjugated to horseradish peroxidase. *Brain Res. Bull.* 12, 669–688. doi: 10.1016/0361-9230(84)90148-5


**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: 29 January 2014; accepted: 20 February 2014; published online: 11 March 2014.*

*Citation: Takahashi LK (2014) Olfactory systems and neural circuits that modulate predator odor fear. Front. Behav. Neurosci. 8:72. doi: 10.3389/fnbeh.2014.00072 This article was submitted to the journal Frontiers in Behavioral Neuroscience.*

*Copyright © 2014 Takahashi. 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.*

## Chronic exposure to a predator or its scent does not inhibit male–male competition in male mice lacking brain serotonin

#### *Ying Huo1,2, Qi Fang1,2, Yao-Long Shi 1,2, Yao-Hua Zhang1 \* and Jian-Xu Zhang1 \**

*<sup>1</sup> State Key Laboratory of Integrated Management of Pest Insects and Rodents in Agriculture, Institute of Zoology, Chinese Academy of Sciences, Beijing, China <sup>2</sup> Department of College of Life Sciences, University of Chinese Academy of Sciences, Beijing, China*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Matthieu Keller, Centre National de la Recherche Scientifique, France*

#### *\*Correspondence:*

*Yao-Hua Zhang and Jian-Xu Zhang, State Key Laboratory of Integrated Management of Pest Insects and Rodents in Agriculture, Institute of Zoology, Chinese Academy of Sciences, 1 Beichen West Road, Chaoyang District, Beijing 100101, China*

*e-mail: zhangyh@ioz.ac.cn; zhangjx@ioz.ac.cn*

Although it is well-known that defective signaling of the 5-HT system in the brain and stressful stimuli can cause psychological disorders, their combined effects on male–male aggression and sexual attractiveness remain unknown. Our research aimed at examining such effects using tryptophan hydroxylase 2 (*Tph2*) knockout male mice vs. a rat- or rat scent-based chronic stress model. *Tph2* <sup>+</sup>/<sup>+</sup> and *Tph2* <sup>−</sup>/<sup>−</sup> male mice were placed individually into the rat home cage (rat), a cage containing soiled rat bedding (rat scent) or a cage containing fresh bedding (control) for 5 h every other day for 56 consecutive days. In *Tph2* <sup>+</sup>/<sup>+</sup> male mice, rat-exposure decreased male–male aggression and sexual attractiveness of urine odor relative to either rat scent-exposure or control; and rat scent-exposure decreased aggression rather than sexual attractiveness of urine odor compared with control. However, such dose-dependent and long-lasting behavioral inhibitory effects vanished in *Tph2* <sup>−</sup>/<sup>−</sup> male mice. RT-PCR assay further revealed that putative regulatory genes, such as AR, ERα and GluR4 in the prefrontal cortex, and TrkB-Tc and 5-HTR1A in the hippocampus, were down-regulated at the mRNA level in either rat- or rat scent-exposed *Tph2* <sup>+</sup>/<sup>+</sup> male mice, but partially in the *Tph2* <sup>−</sup>/<sup>−</sup> ones. Hence, we suggest that the dose-dependent and long-lasting inhibitory effects of chronic predator exposure on male–male aggression, sexual attractiveness of urine odor, and mRNA expression of central regulatory genes might be mediated through the 5-HT system in the brain of male mice.

**Keywords: predator, aggression, sexual attractiveness, central regulatory genes, serotonin, disinhibition**

### **INTRODUCTION**

Rodents rely heavily on chemical senses to detect potential predators and minimize predation risk in natural habitats (Herman and Valone, 2000; Dielenberg and McGregor, 2001; Zhang et al., 2008). The presence of predators or their chemical cues often have non-lethal, but negative, effects on the behavioral, and neurophysiological states of rodents (Dielenberg and McGregor, 2001; Zhang et al., 2003; Adamec et al., 2004; Apfelbach et al., 2005). For example, chronic exposure of rodents to a predator or its chemical cues can enhance anxiety-like behaviors and decrease aggression levels (Francis, 1988; Blanchard et al., 2001; Zhang et al., 2003). Rodents vs. predator-based paradigms have been used extensively to study innate fear- and stress-related psychiatric diseases, such as post-traumatic stress disorder (PTSD) (Hendrie et al., 1996; Adamec et al., 2006a).

Predator and predator scent are evidenced to be graded stress to rodents (Adamec et al., 2008). Unlike the strong stress resulting from close contact with the predator, mild predator scent presumably contains only the pheromones comprised of both volatile and non-volatile molecules from urine and sebaceous gland secretion, and odors emitted from the non-present predator (Kelliher et al., 1999). Previous studies have found that the effect of predator scent-exposure on rodent anxiety and risk assessment fall between controls and those exposed to a predator (Adamec et al., 2004, 2006b). Characterized by mouse-killing behavior (muricide), the rat is often regarded as a mouse predator (Molina et al., 1987; Beekman et al., 2005). Hence, rat- and rat scent-exposure are graded predator stress to mice.

Generally, stress would increase hypothalamic–pituitary– adrenal (HPA) functioning and provoke hyperactivity of the neuropeptide-secreting systems, which eventually lead to the release of stress hormones (Creel, 2001; Beekman et al., 2005). The elevated stress hormones alter metabolic pathways, which exert profound and diffusive effects, such as on reproduction competition ability, including aggressive and defensive levels, pheromone production, sexual attractiveness of urine odor, and related modulations of the neural systems (Creel, 2001; Sands and Creel, 2004; de Kloet et al., 2005; Zhang et al., 2008). To rodents, the released stress hormones induced by predator stimuli would lead to a state that presents reduced aggression levels, as well as less attractiveness to female conspecifics, which might weaken male–male competition, resulting in a reduction of reproductive success (Francis, 1988; Creel, 2001; Zhang et al., 2003).

As one of the most important neurotransmitters, serotonin (5-HT) exerts an extremely wide-ranging influence on rodents, including reproductive activity, aggression, sensory processing, stress adaption, behavioral disinhibition, cognition, memory, and emotion (Canli and Lesch, 2007; Tops et al., 2009; Liu et al., 2011; Kane et al., 2012; Sachs et al., 2013). Tryptophan hydroxylase 2 (Tph2) is required for the synthesis of central 5-HT. *Tph2* knockout mice were recently generated and have been used as a model animal to study 5-HT functions in the brain. *Tph2* <sup>−</sup>/<sup>−</sup> mice are characterized by only minute amounts of brain 5-HT, but normal serotonergic neurons (Liu et al., 2011; Gutknecht et al., 2012). *Tph2* <sup>−</sup>/<sup>−</sup> male mice display social impairments, communication deficits, and behavioral disinhibition indicated by impulsivity (Liu et al., 2011; Angoa-Pérez et al., 2012; Kane et al., 2012). They also show increased aggression in the resident– intruder paradigm and decreased anxiety-like behaviors in the elevated plus-maze (Mosienko et al., 2012). In spite of the 50% decrease in *Tph2* transcriptional activity, *Tph2*+/<sup>−</sup> mice only have a 10% reduction of brain 5-HT, which is insufficient to differentiate them from the *Tph2* <sup>+</sup>/<sup>+</sup> ones on aggression and anxiety-like behaviors (Gutknecht et al., 2012; Mosienko et al., 2012). Currently, little information is available about the modulation effect of 5-HT on mouse urine pheromone; however, our lab has found that *Tph2* knockout would increase the absolute levels of 2-heptanone and *E*-5-hepten-2-one; whereas, it decreases (s)-2-s-butyl-4,5-dihydrothiazole in male urine, indicative of the possible role of central 5-HT in the regulation of pheromone composition (unpublished data).

Previous studies have suggested that the effect of predator threat on rodents may be influenced by the central 5-HT system (Adamec et al., 2006a). In this context, the impulsive *Tph2* <sup>−</sup>/<sup>−</sup> male mice characterized by behavioral disinhibition can be expected to show less stress responses through modulations in the neural circuitry, which may play a causative role in the alterations of aggression levels and urine metabolites (Angoa-Pérez et al., 2012; Mosienko et al., 2012).

Many brain regions are known to play a role in the central processing of stressful stimuli from predators, such as the hippocampus, prefrontal cortex, lateral septum (LAS), and central amygdala (Hayley et al., 2001; Beekman et al., 2005; Joels et al., 2007). Stress hormones feed back to the brain and bind to two types of nuclear receptors serving as transcriptional regulators of brain region-specific susceptible effectors, such as 5-HT receptors, brain-derived neurotrophic factor (BDNF), BDNF receptor TrkB, AMPA (α-amino-3-hydroxy-5-methyl-4 isoxazole propionic acid) glutamate receptor (GluR) subunit, and the receptors of sex steroids that underlie stress-induced responses and behavioral adaptation, all of which function through the 5-HT system in the brain (Nibuya et al., 1999; Shutoh et al., 2000; Nelson and Chiavegatto, 2001; de Kloet et al., 2005; Lund et al., 2006; Nelson and Trainor, 2007). Sex steroid receptors in many brain regions have been suggested to play the modulation role of regulating neuroendocrine stress response and stress-related behaviors such as aggression via effects downstream on sites 5-HTR1A and 5- HTR1B (Simon et al., 1998; Handa and Weiser, 2013). Generally, androgen receptor (AR) and estrogen receptor α (ERα) are positively correlated with aggression levels (Nelson and Chiavegatto, 2001; Li et al., 2004; Scordalakes and Rissman, 2004). 5-HT also can change the expression of AMPA receptor subunits to alter the glutamatergic system activities and, thus, other neuronal functions (Shutoh et al., 2000). For example, GluR4 in the medial prefrontal cortex are in bidirectional control of social dominance hierarchy (Wang et al., 2011). Through 5-HT receptors, particularly 1A and 2A, stress decreases BDNF and increases full-length TrkB (TrkB-FL) expression, respectively, in the hippocampus to modulate learning, adaptive processes, inhibition of aggression, and anxiety- and depression-like behaviors in rodents (Nibuya et al., 1999; Vaidya et al., 2001; Pizarro et al., 2004; Kozlovsky et al., 2007; Martinowich and Lu, 2007; Ito et al., 2011). Unlike TrkB-FL acting as a reverse retrieve for BDNF, TrkB-Tc is speculated to be a negative regulator of TrkB-FL (Eide et al., 1996; Nibuya et al., 1999). Furthermore, 5-HTR1A and 5-HTR1B in the brain, especially in the hippocampus, are closely related to anxiety, aggression, and behavioral disinhibition (de Boer and Koolhaas, 2005; Ambar and Chiavegatto, 2009; Wang et al., 2009).

In rodents, even though there is little known about the central regulation of the urine metabolites, stress would inhibit aggression primarily via inhibitory inputs from the frontal cortex and hippocampus; whereas, the brain areas of medial amygdala (MEA), LAS, bed nucleus of the stria terminalis (BNST), and anterior hypothalamic area (AHA) have been evidenced to prompt the periaquaductal gray (PAG) into promoting aggression (Nelson and Trainor, 2007). In spite of many molecules, such as neurotransmitters, hormones, cytokines, enzymes, growth factors, and signaling molecular affect aggression, 5-HT remains the primary molecular determinant that others may act through the 5-HT signaling system (Nelson and Chiavegatto, 2001).

5-HIAA, the 5-HT metabolite, was found to be elevated only in the hippocampus and prefrontal cortex of both Balb/c and C57BL/6J mice when they were killed 20 min after rat exposure, which might suggest that the hippocampus and prefrontal cortex appear to be regulation sites of the central serotonin system on the effects of predator threat, especially on aggressive behaviors (Hayley et al., 2001). Accordingly, in the current study, prefrontal cortex-specific AR, ERα, and GluR4, and hippocampus-specific AR, ERα, BDNF, TrkB-Tc, 5-HTR1A, and 5-HTR1B that might participate in the stress-induced central responses which function through the 5-HT signaling system in the brain would be selected as the candidate neural substrates.

Previous studies have largely focused on the negative impacts of chronic predator exposure and the modulation roles of the 5- HT system (Apfelbach et al., 2005; Zhou et al., 2008). However, whether the impacts of chronic predator exposure on male–male competition are influenced in the absence of brain 5-HT and possible neural substrates remain unknown. In this study, we used *Tph2* <sup>+</sup>/<sup>+</sup> or *Tph2* <sup>−</sup>/<sup>−</sup> male mice vs. a rat- or rat scent-based paradigms to obtain insight into the differentiation between the two genotypes on the aggressive and defensive levels in staged male–male encounters and sexual attractiveness of urine odor after chronic predator exposure. Then, the mRNA expression of candidate genes in the prefrontal cortex and hippocampus were investigated to explore the possible neural substrates.

#### **MATERIALS AND METHODS EXPERIMENTAL ANIMALS**

Twenty-four male Sprague–Dawley (SD) rats at 8 weeks of age were purchased from the Vital River Laboratories, Beijing, China, and acclimated for 4 weeks prior to use. The rats were housed individually in plastic cages (37 × 26 × 17 cm).

The *Tph2* line was born in our laboratory, while the parental lines were a kind gift from Dr. Yi Rao's laboratory (Peking University, Beijing, China). Mice were weaned at 21 days of age. Then, the male mice were housed in groups with their brothers up to 8 weeks old. After this, male mice were kept singly in plastic cages (27 × 12 × 17 cm). Female mice were always housed in groups of 4 per cage (27 × 12 × 17 cm).

The housing room was under a reversed 14L: 10D light/dark photoperiod (lights on at 7:00 pm), and the temperature was maintained at 23 ± 2◦C. Food (standard mouse chow) and water were provided *ad libitum.* The animal maintenance and handling complied with the Institutional Guidelines for Animal Use and Care at the Institute of Zoology, Chinese Academy of Sciences. Ethics approval was obtained from the Institutional Ethics Committee of the Institute of Zoology, Chinese Academy of Sciences (approval number IOZ12017).

The *Tph2* line was maintained using crossing heterozygotes that the littermates contained wild-type, heterozygotes, and homozygous knockout mice. Genomic DNA was isolated from mouse tails using the phenol/chloroform extracting method at the age of 3 weeks for mouse genotyping. The primers for genotyping were: GCAGCCAGTAGACGTCTCTTAC; GGGCATCTC AGGACGTAGTAG; and GGGCCTGCCGATAGTAACAC. The thermal cycling conditions were as follows: 94◦C for 5 min followed by 35 cycles of 94◦C for 1 min, 62◦C for 30 s, and 72◦C for 1 min 30 s; then 72◦C for 10 min (Liu et al., 2011).

#### **PROCEDURES OF RAT- AND RAT SCENT- EXPOSURE**

Twenty-four *Tph2* <sup>+</sup>/<sup>+</sup> or *Tph2* <sup>−</sup>/<sup>−</sup> male mice aged 12 weeks were randomly assigned into 3 groups of 8 mice each: rat group (each mouse was put into the rat home cage (37 × 26 × 17 cm) where the rat was individually housed and the bedding (250 g) was changed every week, and no barriers existed between the mouse and rat during rat exposure); rat scent group (each mouse was put into a cage (37 × 26 × 17 cm) containing 250 g soiled rat bedding that was used for 1 week by one rat); control group (each mouse was put into a cage (37 × 26 × 17 cm) that contained 250 g fresh bedding). Each *Tph2* <sup>+</sup>/<sup>+</sup> or *Tph2* <sup>−</sup>/<sup>−</sup> male mouse was regularly put into the corresponding standard plastic cage from 9:00 am to 2:00 pm, every other day for 56 consecutive days in a fixed donoracceptor manner. Such treatments were alternately imposed on *Tph2* <sup>+</sup>/<sup>+</sup> or *Tph2* <sup>−</sup>/<sup>−</sup> male mice. In these rat-exposure pairings, the responses of rats to the mice ranged from watching, sniffing, and chasing with attacks. One assaultive rat exhibiting muricide was excluded.

Additionally, another 8 raw *Tph2* <sup>+</sup>/<sup>+</sup> male mice and 6 raw *Tph2* <sup>−</sup>/<sup>−</sup> male mice without any treatment (including exposure procedures and the following urine collection and behavioral tests presented below) were maintained in parallel for the comparison of gene expression between the two genotypes in the raw state.

#### **URINE COLLECTION**

On days 57–59, the next 3 days after completing the 56 consecutive treatment days, the urine of all of the male mice subjected to the exposure procedures were collected in the dark phase, as we previously described (Zhang et al., 2007, 2008). In brief, each donor mouse was placed in a clean mouse cage (27 × 12 × 17 cm) with a wire grid 1 cm above the bottom. Once the animal urinated, the urine was immediately absorbed and transferred to an eppendorf tube in ice, using a disposable glass capillary (i.d. 1.8 mm, 15 cm long). Urine that was deposited with or next to feces was not collected. Urine was individually sealed and kept at −20◦C until use.

#### **ENCOUNTERING TEST**

On days 60–62, the next 3 days after urine collection, male–male encountering tests were performed between male mice subjected to different exposure procedures within the same genotype, as previously described (Clancy et al., 1984; Zhang et al., 2008). Weight-matched male mice from different groups within the same genotype were paired and simultaneously placed into a clean mouse cage (27 × 12 × 17 cm) for 10 min continuous recording after initial aggressive behavior or defensive behavior. Such records were conducted by hand on a data sheet with a precalibrated time scale in units of 10 s using a stopwatch. A behavior pattern that lasted 10 s or less was treated as one unit. Each mouse was used only once a day. Aggressive behaviors included tail rattles, sideway postures, pushing, chasing, and biting; defensive behaviors included fleeing and upright postures.

#### **BINARY TEST OF URINARY ATTRACTIVENESS**

Binary choice tests via two capillaries were conducted to explore urinary attractiveness, as we previously described (Zhang et al., 2007, 2008). In brief, 14–15 adult female *Tph2* <sup>+</sup>/<sup>+</sup> mice in estrus (6–9 months of age) were used as test recipient subjects, each of which was used only once per day. The estrous status of female mice was determined by microscopic examination of vaginal epithelium. *Tph2* <sup>+</sup>/<sup>+</sup> female mice were given 1 day of rest between each test. Urine samples from different groups were randomly paired and presented to female mice by two identical disposable glass capillaries (internal diameter = 1.8 mm, length = 15 cm), which contained 2μL urine about 1 cm from the samplecontaining end, and the other end sealed by odorless gum. The sample-containing end was presented to the test mice in their home cages simultaneously and kept approximately 2 cm apart from each other. We recorded investigating behavior (sniff within 1 cm of the tips or lick the end of the capillaries) for 3 min after the initial sniffing response. The durations that the test mouse spent investigating each odor were recorded using two hand-held stopwatches.

#### **TISSUE SAMPLING**

On day 70, after 1 week of rest after the encountering tests, all of the male mice, including the ones subjected to the exposure procedure, urine collection and behavioral tests, and the raw ones without any treatment were directly and quickly (within 3 s) decapitated using a pair of sterile operating scissors. The prefrontal cortex and hippocampus were dissected in a mouse brain matrix on ice for the following real-time PCR analysis. The whole process was conducted in a separate dim room in the dark phase. The tissues were frozen in liquid nitrogen immediately and stored at −80◦C until use.

#### **REAL-TIME PCR**

Total RNA of the prefrontal cortex and unilateral hippocampus were isolated using Trizol (Invitrogen), and cDNA was reversetranscribed from total RNA (2μg) using PrimeScript® RT reagent Kit With gDNA Eraser (Perfect Real Time) (Takara), following the manufacturer's instructions. Real-time PCR was performed using the RealMasterMix (SYBR Green) (Tiangen). Specific primers were designed for genes, while the housekeeping gene β-actin was chosen as a control for normalizing the relative mRNA level (the primer sequences are available upon request). Twenty μL reaction agents were carried out in the Mx3005P quantitative PCR system (Stratagene, La Jolla, CA, USA) comprised of 9μL of 2.5 × RealMasterMix/20 × SYBR solution, 1μL of template cDNA, 0.5μM of each primer, and 9μL sterile water. Negative controls containing no template were also performed for each primer pair. The thermal cycling conditions were as follows: 95◦C for 2 min followed by 40 cycles of 95◦C for 20 s, 60◦C for 20 s, and 68◦C for 40 s. The melting curve analysis was performed to eliminate the presence of unspecific products by a high-resolution data collection during an incremental temperature change from 55 to 95◦C with a ramp rate of 0.2◦C/s. The data derived from the Mx3005P quantitative software were calculated using the 2−-*CT* formula (Livak and Schmittgen, 2001; Wang et al., 2006).

#### **STATISTICAL ANALYSIS**

Behavioral data were analyzed by the Wilcoxon matched-pairs signed-rank test because of the non-normal distribution characteristics (the normality was checked using the Kolmogorov– Smirnov test). The comparisons of gene expression among the three groups of *Tph2* <sup>+</sup>/<sup>+</sup> male mice or the *Tph2* <sup>−</sup>/<sup>−</sup> ones that were subjected to exposure procedures, and the following urine collection and behavioral tests were analyzed by One-Way analysis of variance (ANOVA) with least significant difference (LSD) tests. The comparisons of gene expression between the raw *Tph2* <sup>+</sup>/<sup>+</sup> male mice and the raw *Tph2* <sup>−</sup>/<sup>−</sup> ones were analyzed by independent-samples *t*-test. All statistical analyses were conducted using SPSS 16.0 software (SPSS Inc., Chicago, IL, USA) with the critical value of α = 0.05.

#### **RESULTS**

#### **RAT- AND RAT SCENT-EXPOSURE INHIBITED MALE–MALE AGGRESSION IN** *TPH2* **<sup>+</sup>***/***<sup>+</sup> MALE MICE, BUT NOT IN THE** *TPH2* **<sup>−</sup>***/***<sup>−</sup> ONES**

In staged dyadic encounters between *Tph2* <sup>+</sup>/<sup>+</sup> male mice, both the control [**Figure 1A**, Wilcoxon rank sum test, median 13.00 (interquartile range 6.000–18.00) vs. median 2.000 (interquartile range 0.000–12.00), *Z* = −2.201, *N* = 7, *P* = 0.028] and rat scent-exposed ones [**Figure 1C**, Wilcoxon rank sum test, median 16.00 (interquartile range 4.000–25.00) vs. median 0.000 (interquartile range 0.000–12.00), *Z* = −2.028, *N* = 7, *P* = 0.043] showed higher levels of aggression than the rat-exposed ones; and the control ones showed higher levels of aggression than the rat scent-exposed ones [**Figure 1B**, Wilcoxon rank sum test, median 12.50 (interquartile range 6.000–22.00) vs. median 6.500 (interquartile range 0.000–9.000), *Z* = −2.106, *N* = 8, *P* = 0.035].

Correspondingly, rat-exposed *Tph2* <sup>+</sup>/<sup>+</sup> males [**Figure 1A**, Wilcoxon rank sum test, median 13.00 (interquartile range 0.000–17.00) vs. median 0.000 (interquartile range 0.000–6.000), *Z* = −1.997, *N* = 7, *P* = 0.046] and the rat scent-exposed ones [**Figure 1B**, Wilcoxon rank sum test, median 4.500 (interquartile range 0.000–21.00) vs. median 0.000 (interquartile range 0.000– 1.000), *Z* = −2.023, *N* = 8, *P* = 0.043] showed more defensive behaviors than the control ones; and the rat-exposed ones showed more defensive behaviors than the rat scent-exposed ones [**Figure 1C**, Wilcoxon rank sum test, median 12.00 (interquartile range 0.000–19.00) vs. median 0.000 (interquartile range 0.000–3.000), *Z* = −2.120, *N* = 7, *P* = 0.034].

However, such treatments did not differentiate aggressive and defensive behaviors in dyadic interactions between *Tph2* <sup>−</sup>/<sup>−</sup> male mice (**Figures 1D–F**).

#### **RAT-EXPOSURE DECREASED SEXUAL ATTRACTIVENESS OF URINE ODOR IN** *TPH2* **<sup>+</sup>***/***<sup>+</sup> MALE MICE, BUT NOT IN THE** *TPH2* **<sup>−</sup>***/***<sup>−</sup> ONES**

Two-choice tests unraveled that the urine odor from either rat scent-exposed *Tph2* <sup>+</sup>/<sup>+</sup> males [**Figure 2A**, Wilcoxon rank sum test, median 3.390 (interquartile range 0.000–15.76) vs. median 1.280 (interquartile range 0.000–12.25), *Z* = −2.073, *N* = 15, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>.038] or control *Tph2* <sup>+</sup>/<sup>+</sup> males [**Figure 2A**, Wilcoxon rank sum test, median 4.780 (interquartile range 0.400–44.87) vs. median 3.495 (interquartile range 0.000–30.79), *Z* = −2.103, *<sup>N</sup>* <sup>=</sup> 14, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>.035] induced higher attraction of *Tph2* <sup>+</sup>/<sup>+</sup> females than that from rat-exposed *Tph2* <sup>+</sup>/<sup>+</sup> males. However, *Tph2* <sup>+</sup>/<sup>+</sup> females showed no olfactory preferences between rat scent-exposed and control *Tph2* <sup>+</sup>/<sup>+</sup> males [**Figure 2A**, Wilcoxon rank sum test, median 3.140 (interquartile range 0.290–24.94) vs. median 3.585 (interquartile range 0.000–52.18), *Z* = −0.722, *N* = 14, *P* = 0.470].

Even so, two-choice tests revealed that the urine odor between any two groups of *Tph2* <sup>−</sup>/<sup>−</sup> males induced no apparent differences in attraction of *Tph2* <sup>+</sup>/<sup>+</sup> females (**Figure 2B**).

#### **BOTH RAT- AND RAT SCENT-EXPOSURE DOWN-REGULATED THE GENE EXPRESSION IN THE PREFRONTAL CORTEX OF** *TPH2* **<sup>+</sup>***/***<sup>+</sup> MALE MICE, BUT NOT IN THE** *TPH2* **<sup>−</sup>***/***<sup>−</sup> ONES**

In the prefrontal cortex of *Tph2* <sup>+</sup>/<sup>+</sup> male mice, the expression of AR [**Figure 3A**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 6.913, *P* = 0.015], ERα [**Figure 3A**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 55.31, *P* = 0.000] and GluR4 [**Figure 3A**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 4.594, *P* = 0.042] apparently differed among the 3 groups. Compared with the control ones, *Tph2* <sup>+</sup>/<sup>+</sup> AR (**Figure 3A**, LSD *post-hoc t*-tests, rat-exposure: *P* = 0.007; rat scent-exposure: *P* = 0.016), ERα (**Figure 3A**, LSD *post-hoc t*-tests, rat-exposure: *P* = 0.000; rat scent-exposure: *P* = 0.000) and GluR4 (**Figure 3A**, LSD *post-hoc t*-tests, rat-exposure: *P* = 0.016; rat scentexposure: *P* = 0.073, marginal significance) expression were down-regulated.

However, unlike *Tph2* <sup>+</sup>/<sup>+</sup> male mice, the *Tph2* <sup>−</sup>/<sup>−</sup> ones upregulated AR expression [**Figure 3B,** One-Way ANOVA, *F*(2, <sup>20</sup>) = 42.51, *P* = 0.000] in the rat- (**Figure 3B**, LSD *post-hoc t*-tests, *P* = 0.000) and rat scent-exposed male mice (**Figure 3B**, LSD *post-hoc t*-tests, *P* = 0.000) as compared with the control ones. In addition, AR was up-regulated in the rat-exposed *Tph2* <sup>−</sup>/<sup>−</sup> male mice more than the rat scent-exposed ones (**Figure 3B**, LSD *posthoc t*-tests, *P* = 0.027). The expression of other genes, including

ERα and GluR4 did not differ among the 3 groups of *Tph2* <sup>−</sup>/<sup>−</sup> male mice (**Figure 3B**).

Additionally, the expression of AR, ERα, and GluR4 in the prefrontal cortex did not differ between the raw *Tph2* <sup>+</sup>/<sup>+</sup> male mice and the raw *Tph2* <sup>−</sup>/<sup>−</sup> ones (**Figure 4**).

#### **BOTH RAT- AND RAT SCENT-EXPOSURE DOWN-REGULATED HIPPOCAMPAL GENE EXPRESSION MORE IN** *TPH2* **<sup>+</sup>***/***<sup>+</sup> MALE MICE THAN IN THE** *TPH2* **<sup>−</sup>***/***<sup>−</sup> ONES**

In the hippocampus of *Tph2* <sup>+</sup>/<sup>+</sup> male mice, the AR [**Figure 5A**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 12.62, *P* = 0.002], ERα [**Figure 5A**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 46.55, *P* = 0.000], TrkB-Tc [**Figure 5B**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 75.98, *P* = 0.000] and 5-HTR1A [**Figure 5C**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 5.875, *P* = 0.023] expression were significantly different among the 3 groups. As compared with the control ones, *Tph2* <sup>+</sup>/<sup>+</sup> hippocampal AR (**Figure 5A**, LSD *post-hoc t*-tests, rat group: *P* = 0.003; rat-scent group: *P* = 0.001), ERα (**Figure 5A**, LSD *post-hoc t*-tests, rat group: *P* = 0.000; rat-scent group: *P* = 0.000), TrkB-Tc (**Figure 5B**, LSD *post-hoc t*-tests, rat group: *P* = 0.000; rat-scent group: *P* = 0.000), and 5-HTR1A (**Figure 5C**, LSD *posthoc t*-tests, rat group: *P* = 0.033; rat-scent group: *P* = 0.010) expression were down-regulated.

In the hippocampus of *Tph2* <sup>−</sup>/<sup>−</sup> male mice, the AR [**Figure 5D**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 14.43, *P* = 0.002], ERα [**Figure 5D**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 147.0, *P* = 0.000] and BDNF [**Figure 5E**, One-Way ANOVA, *F*(2, <sup>9</sup>) = 4.142, *P* = 0.053, marginal significance] expression also presented significant

differentiation among the 3 groups. As compared with the control ones, *Tph2* <sup>−</sup>/<sup>−</sup> hippocampal ERα was both down-regulated in the mice exposed to either rat (**Figure 5D**, LSD *post-hoc t*-tests, *P* = 0.000) or rat scent (**Figure 5D**, LSD *post-hoc t*-tests, *<sup>P</sup>* <sup>=</sup> <sup>0</sup>.000). However, *Tph2* <sup>−</sup>/<sup>−</sup> hippocampal AR of rat scentexposed mice (**Figure 5D**, LSD *post-hoc t*-tests, *P* = 0.001) and BDNF of rat-exposed mice (**Figure 5E**, LSD *post-hoc t*-tests, *P* = 0.019) were only down-regulated relative to the control ones. Additionally, the strong rat-exposure stress increased *Tph2* <sup>−</sup>/<sup>−</sup> hippocampal AR expression than the milder rat scent-exposure stress (**Figure 5D**, LSD *post-hoc t*-tests, *P* = 0.004).

*Tph2* <sup>+</sup>/<sup>+</sup> BDNF, *Tph2* <sup>+</sup>/<sup>+</sup> 5-HTR1B, *Tph2* <sup>−</sup>/<sup>−</sup> TrkB-Tc, *Tph2* <sup>−</sup>/<sup>−</sup> 5-HTR1A, and *Tph2* <sup>−</sup>/<sup>−</sup> 5-HTR1B expression in the hippocampus were not changed by rat- or rat scent- exposure (**Figures 5B,C,E,F**).

However, compared with the raw *Tph2* <sup>+</sup>/<sup>+</sup> males, hippocampal ERα (**Figure 6A**, independent-samples *t*-test, *t* = 4.464, *df* = 12, *P* = 0.001) and 5-HTR1A (**Figure 6C**, independentsamples *t*-test, *t* = 2.888, *df* = 12, *P* = 0.014) expression of the raw *Tph2* <sup>−</sup>/<sup>−</sup> ones decreased; whereas, the BDNF (**Figure 6B**, independent-samples *t*-test, *t* = −3.571, *df* = 12, *P* = 0.004) expression increased. The expression of other genes, including AR, TrkB-Tc, and 5-HTR1B in the hippocampus did not differ between the two genotypes in the raw state (**Figures 6A–C**).

#### **DISCUSSION**

Our data showed that chronic exposure to rat or rat scent both could produce long-lasting inhibitory effects on *Tph2* <sup>+</sup>/<sup>+</sup> male mice, reflected in the down-regulated male–male aggression, sexual attractiveness of urine odor, and mRNA expression of central regulatory genes. However, such inhibitory effects were reduced or abolished in *Tph2* <sup>−</sup>/<sup>−</sup> male mice.

One of the primary responses to stress is an increase in the concentration of circulating adrenal glucocorticooids (GC). Instead of GC elevations lasting for hours, chronic exposure to a predator or its scent causes GC levels to remain high for more than a few days, causing a broad and long-lasting negative effects (Creel, 2001). Such impacts have been widely proven in rats and mice, focusing on anxiogenic effects as measured in the EPM, light/dark box, and acoustic startle tests (Dielenberg and McGregor, 2001; Adamec et al., 2006a). For example, the anxiety state of rats exposed to a cat lasted 21 days or more, similar to long-lasting effects on acoustic startle responses of CD-1 mice exposed to rat odor (Adamec and Shallow, 1993; Hebb et al., 2003a).

It is a general belief that psychological stressors, such as the threat of a predator, would negatively affect social and sexual behaviors, such as aggression levels and sexual attractiveness (Francis, 1988; Blanchard et al., 2001; Zhang et al., 2003; Apfelbach et al., 2005). Receptive females expend much more

**FIGURE 3 | Comparison of the relative expression of genes in the prefrontal cortex of different groups. (A)** Expression patterns of AR, ERα and GluR4 in the prefrontal cortex of the rat-exposed, rat scent-exposed and control *Tph2*+/<sup>+</sup> male mice (mean <sup>±</sup> SE, *<sup>n</sup>*Rat <sup>=</sup> 4, *n*Rat scent = 4, *n*Control = 4). In order to obtain a large amount for each *Tph2*+/<sup>+</sup> biological replicate, the prefrontal cortex from two male mice in the same group were ground together for RNA extraction. **(B)** Expression patterns of AR, ERα and GluR4 in the prefrontal cortex of the rat-exposed, rat scent-exposed and control *Tph2*−/<sup>−</sup> male mice (mean ± SE, *n*Rat = 7, *n*Rat scent = 8, *n*Control = 8). The data were analyzed by One-Way ANOVA, followed by LSD *post-hoc t*-tests ( ∗*P* < 0.05; ∗∗*p* < 0.01).

rat-exposed, rat scent-exposed and control *Tph2*−/<sup>−</sup> male mice. Mean <sup>±</sup> SE, *n*Rat = 4, *n*Rat scent = 4, *n*Control = 4. In order to obtain a large amount for each biological replicate, the hippocampi from two male mice in the same group were ground together for RNA extraction. The data were analyzed by One-Way ANOVA, followed by LSD *post-hoc t*-tests ( ∗*P* < 0.05; ∗∗*p* < 0.01).

effort during reproduction than do males; hence, they prefer to select high-quality males, in order to increase the probability of reproductive success and offspring survival (Huck and Banks, 1982). Both aggression levels and urinary attractiveness to female conspecifics are representatively manifested in the ability of male mice in male–male competition. In our current study, the descending degree in aggression levels of *Tph2* <sup>+</sup>/<sup>+</sup> males, dependent upon stimulation intensity, strongly supported the dose-dependent negative influences and predicted competitive ability. Furthermore, such inhibition lasted for a minimum of 6 days, which is consistent with the proven long-lasting effects on anxiogenic states, as stated above.

However, rat scent-exposure did not reduce the urinary attractiveness of *Tph2* <sup>+</sup>/<sup>+</sup> male mice. Such an anti-intuitive result might be ascribed to the insufficient intensity of the rat scent used, since the intensity of predator stress has been suggested as an important factor in the severity of long-lasting effects (Hebb et al., 2003b). For instance, we previously reported that male mice exposed to a low dose of cat urine showed higher sexual attractiveness to female conspecifics than the control ones (Zhang et al., 2008). Even so, the results that urine odor from both rat scent-exposed and control *Tph2* <sup>+</sup>/<sup>+</sup> males were more attractive to females than the rat-exposed ones still demonstrated the dose-dependent inhibitory effects of chronic predator exposure on male–male competition.

Accordingly, male–male aggression and sexual attractiveness of urine odor of *Tph2* <sup>+</sup>/<sup>+</sup> male mice were long-lastingly inhibited by graded stress from rat and rat scent, depending on the stimulation intensity in the current study, which is consistent with and expands on previously published reports.

The pharmacological challenge of neuronal 5-HT, such as treatment with 5-HTR1A and 5-HTR1B agonists, depletion using *p*-Chlorophenylalanine (PCPA; irreversible inhibitor of the 5-HT synthesizing enzyme tryptophan hydroxylase), or intracerebral injection of the 5-HT neurotoxic agent 5,7-dihydroxytryptamine (5,7-DHT), also induces multiple alterations, such as aggression levels and hormone release (Willoughby et al., 1982; Chiavegatto et al., 2001; de Boer and Koolhaas, 2005). Even so, these

manipulations have some limitations to consider, including time (stages of pre- or post-natal development), target tissue (brain or periphery), route, effective time, and complex mechanisms regulating 5-HT neuron firing, presynaptic release, and post-synaptic receptor expression (Miyazaki et al., 2011; van Kleef et al., 2012). However, genetic disruption in the *Tph2* affects the central 5-HT system throughout the lifespan, leading to possible alterations in the formation of the neural circuitry mediating emotion and stress adaption during a critical period of brain development (Adamec et al., 2006a; Liu et al., 2011).

Accordingly, in the current study, the predator or its scent exposure did not alter aggression and urinary attractiveness of *Tph2* <sup>−</sup>/<sup>−</sup> male mice, suggesting that brain 5HT deficiency might prevent the inhibitory effects by affecting sensory processing and, thus, induce behavioral disinhibition (Raleigh et al., 1991; Nelson and Chiavegatto, 2001; Freichel et al., 2004; Canli and Lesch, 2007; Tops et al., 2009; Liu et al., 2011; Kane et al., 2012). Moreover, the reduced expression of ERα and 5-HTR1A accompanied by increased BDNF in the hippocampus of the raw *Tph2* <sup>−</sup>/<sup>−</sup> male mice, which may suggest that deficiency of brain 5-HT synthesis likely extends beyond the serotonin system itself (Murphy et al., 2003; Adamec et al., 2006a).

The expression of the prefrontal cortex- and hippocampusspecific genes, which were selectively used in the current study, can reasonably reflect stress-induced central responses and elucidate the mechanisms underlying behavioral alterations. In the current study, the down-regulation of AR, ERα and GluR4 in the prefrontal cortex, and TrkB-Tc and 5-HTR1A in the hippocampus might result from stress responses and, in turn, provide the neural substrates for the alterations of male–male aggression and sexual attractiveness in the *Tph2* <sup>+</sup>/<sup>+</sup> male mice exposed to a rat or rat scent. The decreased expression of GluR4 in the prefrontal cortex and 5-HTR1A in the hippocampus of stressed *Tph2* <sup>+</sup>/<sup>+</sup> male mice was consistent with previous literature (Wang et al., 2009, 2011). Despite being in line with the expected low expression in the hippocampus of stressed *Tph2* <sup>+</sup>/<sup>+</sup> TrkB-Tc, the results of the indifferent *Tph2* <sup>+</sup>/<sup>+</sup> BDNF expression were surprising (Nibuya et al., 1999; Vaidya et al., 2001; Pizarro et al., 2004; Kozlovsky et al., 2007). Such bifurcations originated mostly from differences in genetic background, as previous studies on the impacts of predator stress on hippocampal BDNF expression focusing on rats other than the C57BL/6J × 129S5/S mice used in our study. Furthermore, the decreased hippocampal BDNF expression of C57BL/6 mice followed acute social stress other than chronic predator stress in our study. In this respect, it is necessary to note, instead of decreased hippocampal BDNF mRNA expression in rat vs. predator-based models, the reduced hippocampal TrkB-Tc mRNA expression might play an important role in modulation of stress response in mice vs. predator-based paradigms, according to our results. AR and ERα have been suggested to be positively correlated with aggression level, as stated above (Li et al., 2004; Scordalakes and Rissman, 2004). In this research, AR and ERα in the prefrontal cortex, rather than in the hippocampus, might contribute to central regulation and related behavioral alterations after predator stress, thus confirming and extending the aforementioned work.

In addition, although rat- and rat-scent exposure could have different effects on male–male competition of *Tph2* <sup>+</sup>/<sup>+</sup> male mice, their differences in threat intensity were not sufficient to differentiate the mRNA expression of the examined genes.

*Tph2* <sup>−</sup>/<sup>−</sup> male mice showed different mRNA expression patterns from the *Tph2* <sup>+</sup>/<sup>+</sup> ones. Specifically, rat- and/or rat scent- exposure up-regulated prefrontal AR expression, but down-regulated hippocampal AR, ERα, and BDNF expression, suggesting that 5-HT might mediate central regulation of responses to predation risk. However, such alterations of gene expression appeared to exert no impacts on male–male aggression and sexual attractiveness in *Tph2* <sup>−</sup>/<sup>−</sup> male mice, likely reflecting that these genes were not able to work through the defective 5-HT system (Nelson and Chiavegatto, 2001).

In conclusion, the current results are consistent with previous work that exposure to a predator or predator scent causes dose-dependent and long-lasting negative impacts on rodents. However, we particulalrly relate predator-based chronic stress to brian 5-HT-deficent male mice and male–male competition ability for potential female mates, as indicated by male–male aggression and urinary attractiveness to female conspecifics. For the first time, inhibitory effects on these behavioral measures are found to be abolished in *Tph2* <sup>−</sup>/<sup>−</sup> male mice. We also demonstrate that AR, ERα and GluR4 expression in the prefrontal cortex, and TrkB-Tc and 5-HTR1A expression in the hippocampus are changed in response to predation risk. In turn, they may regulate the alterations of male–male aggression and urinary attractiveness to female conspecifics through the 5-HT system in the brain of male mice.

#### **ACKNOWLEDGMENTS**

We thank Yi Rao's laboratory (Peking University, Beijing) and Yan Liu (Peking University, Beijing) for the kind gift of the *Tph2* line mice. We also thank Jin-Hua Zhang for animal breeding and behavioral tests, and Jin-Long Han for his assistance in early PCR assay. This work was supported primarily by grants from the National Basic Research Program of China [973 Program, No. 2010CB833900], the China National Science Foundation [31272322], the Chinese Academy of Sciences [KSCX2-EW-N-5], and the State Key Laboratory of Integrated Management of Pest Insects and Rodents in Agriculture (Chinese IPM1208).

#### **REFERENCES**


of field and laboratory studies. *Neurosci. Biobehav. Rev.* 29, 1123–1144. doi: 10.1016/j.neubiorev.2005.05.005


disinhibition and anxiety-like behaviour following mild early life stress. *Int. J. Neuropsychopharmacol.* 16, 2081–2094. doi: 10.1017/S1461145713000321


**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 November 2013; accepted: 19 March 2014; published online: 08 April 2014.*

*Citation: Huo Y, Fang Q, Shi Y-L, Zhang Y-H and Zhang J-X (2014) Chronic exposure to a predator or its scent does not inhibit male–male competition in male mice lacking brain serotonin. Front. Behav. Neurosci. 8:116. doi: 10.3389/fnbeh.2014.00116 This article was submitted to the journal Frontiers in Behavioral Neuroscience.*

*Copyright © 2014 Huo, Fang, Shi, Zhang and Zhang. 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.*

## Mouse Grueneberg ganglion neurons share molecular and functional features with *C. elegans* amphid neurons

### *Julien Brechbühl †, Fabian Moine† and Marie-Christine Broillet\**

*Department of Pharmacology and Toxicology, Faculty of Biology and Medicine, University of Lausanne, Lausanne, Switzerland*

#### *Edited by:*

*Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Joerg Fleischer, Universität of Hohenheim, Germany*

#### *\*Correspondence:*

*Marie-Christine Broillet, Department of Pharmacology and Toxicology, Faculty of Biology and Medicine, University of Lausanne, Bugnon 27, CH-1005 Lausanne, Switzerland e-mail: mbroille@unil.ch*

*†These authors have contributed equally to this work.*

The mouse Grueneberg ganglion (GG) is an olfactory subsystem located at the tip of the nose close to the entry of the naris. It comprises neurons that are both sensitive to cold temperature and play an important role in the detection of alarm pheromones (APs). This chemical modality may be essential for species survival. Interestingly, GG neurons display an atypical mammalian olfactory morphology with neurons bearing deeply invaginated cilia mostly covered by ensheathing glial cells. We had previously noticed their morphological resemblance with the chemosensory amphid neurons found in the anterior region of the head of *Caenorhabditis elegans* (*C. elegans*). We demonstrate here further molecular and functional similarities. Thus, we found an orthologous expression of molecular signaling elements that was furthermore restricted to similar specific subcellular localizations. Calcium imaging also revealed a ligand selectivity for the methylated thiazole odorants that amphid neurons are known to detect. Cellular responses from GG neurons evoked by chemical or temperature stimuli were also partially cGMP-dependent. In addition, we found that, although behaviors depending on temperature sensing in the mouse, such as huddling and thermotaxis did not implicate the GG, the thermosensitivity modulated the chemosensitivity at the level of single GG neurons. Thus, the striking similarities with the chemosensory amphid neurons of *C. elegans* conferred to the mouse GG neurons unique multimodal sensory properties.

**Keywords: olfactory, amphid neurons, Grueneberg ganglion, behavior, calcium imaging, temperature sensing, alarm pheromone**

### **INTRODUCTION**

Organisms have evolved specialized populations of sensory receptor neurons to detect the chemical information that is present in their environment (Ache and Young, 2005). In the model organism *Caenorhabditis elegans* (*C. elegans*), chemo- and thermosensing are performed by 12 pairs of ciliated sensory neurons, the amphid neurons, that are found in the anterior region of the animal head (Mori and Ohshima, 1995; Bargmann, 2006). Amphid neurons include three principal olfactory classes named amphid wing neurons of type A (AWA), B (AWB), and C (AWC) (**Figure 1A**). They play a fundamental role in mate identification, in food finding and in noxious conditions avoidance by odortaxis (Bargmann, 2006). They also participate in the development of the organism by navigating through spatial thermal gradients by thermotaxis (Mori and Ohshima, 1995). In the mouse, the olfactory sensory neurons are dispatched in four distinct subsystems: the main olfactory epithelium (MOE), the septal organ of Masera (SO), the vomeronasal organ (VNO) and the most recently described Grueneberg ganglion (GG) (Gruneberg, 1973) (**Figure 1B**). These subsystems are implicated in the detection of molecules carrying chemical messages, such as odorants and pheromones that play a role in behaviors ranging from food finding to social communication (Munger et al., 2009). We have identified one of these subsystems, the GG, as a chemodetector of alarm pheromones (APs) (Brechbühl et al., 2008). APs signal injury, distress or the presence of predators (Kiyokawa et al., 2004). A wide variety of changes in behaviors can be observed in the presence of APs such as a decreased exploratory activity or increase in vigilance and freezing behaviors (Zalaquett and Thiessen, 1991; Kiyokawa et al., 2006). We have recently isolated and identified the chemical structure of one mouse AP, 2-*sec*butyl-4,5-dihydrothiazole (SBT) (Brechbühl et al., 2013). This volatile APs is produced by both male and female mice under different alarm conditions and resembles the sulfur-containing volatiles present in predator scents. In addition to its reported chemosensory modality (Brechbühl et al., 2008, 2013; Mamasuew et al., 2011a; Hanke et al., 2013), the mouse GG has also been implicated in sensing cold temperatures (Mamasuew et al., 2008; Schmid et al., 2010).

We noticed the morphological resemblance between the mouse GG neurons and the olfactory AWA, AWB, and AWC amphid neurons from the nematode *C. elegans* (Brechbühl et al., 2008). Indeed, mouse GG neurons display an atypical olfactory morphology with neurons that bear deeply invaginated cilia and that are mostly covered by glial cells (Gruneberg, 1973; Tachibana et al., 1990; Brechbühl et al., 2008; Liu et al., 2009). They have a rostral localization in a water permeant epithelium and they lack direct contact with the nasal cavity (Gruneberg, 1973; Fuss et al., 2005; Koos and Fraser, 2005; Fleischer et al., 2006a; Roppolo et al., 2006; Storan and Key, 2006; Brechbühl et al., 2008; Liu et al., 2009). In *C. elegans*, pairs of either AWA, AWB, or AWC neurons detect volatile odorants. They are found under a water permeant

septal organ; MOE, main olfactory epithelium. The signaling elements are indicated as follows: cGMP, cyclic guanosine monophosphate; cGKII, cGMP-dependent protein kinase of type 2; CNGA3, cyclic nucleotide-gated channels 3; DAF-11/ODR-1, potential receptor-like transmembranous guanylyl cyclase; egl-4, cGMP-dependent protein kinase; GC-G, particulate guanylyl

cuticle and are wrapped by a single ensheathing glial cell that also surrounds the modified cilia (Bargmann et al., 1993; Bargmann, 2006; Inglis et al., 2007; Bacaj et al., 2008).

These morphological similarities might also indicate molecular similarities between mouse GG neurons and *C. elegans* amphid neurons. Indeed, recent studies have revealed that canonical and non-canonical signaling elements are expressed in GG neurons (Fleischer et al., 2006b, 2007, 2009; Pyrski et al., 2007; Brechbühl et al., 2008; Liu et al., 2009, 2012). They could, as in *C. elegans*, be potentially implicated in parallel and/or convergent G proteincoupled receptors (GPCRs)- and cyclic guanosine monophosphate (cGMP)-dependent signaling pathways for thermo- and chemodetection (Mamasuew et al., 2010, 2011a,b; Schmid et al., 2010; Hanke et al., 2013).

Here, we investigated the conserved multisensory modalities of mouse GG neurons. We found striking similarities between mouse GG neurons and nematode amphid neurons, especially

cyclase G; GPCRs, G protein coupled receptors; PDE, phosphodiesterase; PDE2A, phosphodiesterase 2A; TAX-2/4, cyclic nucleotide-gated (CNG)-like channels. **(C)** Immunohistochemistry experiments on coronal slices of the GG of OMP-GFP mice for homologous signaling proteins found in *C. elegans* amphid neurons. GC-G was found to be expressed in the cilia. CNGA3 was found to be expressed principally in cilia and axons; some somatic expression could also be observed. PDE2A was found in soma and in axons. cGKII was found in soma. The specific subcellular localizations are shown in high power views (white dashed rectangles). White arrowheads indicate ciliary processes, black arrowheads indicate soma and white arrows indicate axons. A minimum of 2 animals (from P0–P29) and 6 slices were used for each antibody staining tested. Nuclei are shown in blue (DAPI counterstain). Scale bars, 20µm.

the AWC class. We found that 2,4,5-trimethylthiazole, a known AWC ligand, was able to initiate neuronal responses in mouse GG neurons in a cGMP-dependent manner. AWC neurons are also known to act as thermosensors and we found here that the temperature can modulate the chemosensitivity of GG neurons. Thus, GG neurons, through their position at the tip of the nose, are able to integrate multiple sensory inputs thereby allowing an animal to assess additional aspects of its olfactory environment.

### **EXPERIMENTAL PROCEDURES**

#### **ANIMALS AND TISSUE PREPARATION**

OMP-GFP mice (Potter et al., 2001) from pups to adult stages were used for all experimental investigations. This particular gene-targeted mouse strain expresses the green fluorescent protein (GFP) as a histological reporter under the control of the olfactory marker protein (OMP) promoter (Mombaerts et al., 1996; Potter et al., 2001). OMP is a marker specific for mature olfactory sensory neurons (Margolis, 1972). Animal care was in accordance with the Swiss legislation and the veterinary authority. Mice were killed by CO2 or cervical dislocation. Nasal cavities were prepared in ice-cold artificial cerebrospinal fluid (ACSF), containing 118 mM NaCl, 25 mM NaHCO3, 10 mM D-glucose, 2 mM KCl, 2 mM MgCl2, 1.2 mM NaH2PO4, and 2 mM CaCl2 (pH 7.4) saturated with oxycarbon gas [95% O2: 5% CO2; (vol/vol)] under a fluorescence-equipped dissecting microscope (M165 FC; Leica). For specific experiments, ACSF calcium free solution was also used and it was composed of NaCl 118 mM, NaHCO3 25 mM, D-Glucose 10 mM, KCl 2 mM, MgCl2 2 mM, NaH2PO4 1.2 mM, EDTA 10 mM and EGTA 10 mM saturated with oxycarbon gas.

#### **IMMUNOHISTOCHEMISTRY**

Protocol for floating immunohistochemistry was adapted from (Brechbühl et al., 2011, 2013). Briefly, the tip of the nose was carefully dissected in PBS (138 mM NaCl, 2.7 mM KCl, 0.9 mM KH2PO4, and 0.8 mM Na2HPO4, pH 7.6) before being fixed in 4% paraformaldehyde (PAF 4%, in PBS pH 7.4; 158127, Sigma) at 4◦C for 3 h. Fixed tissue preparations were embedded in 5% agar (A7002, Sigma) prepared in PBS. Agar blocks were transferred on ice for 30 s for solidification. The agar blocks were fixed vertically with cyanacrylat glue (Roti coll 1, Carl Roth) onto the holder of the vibroslicer (VT1200S, Leica). 80µm coronal sections were cut in PBS and were selected with a fluorescence-equipped dissecting microscope (M165 FC; Leica). Slices were blocked overnight at 4◦C in a PBS solution containing 10% NGS (normal goat serum, Jackson ImmunoResearch) and Triton X-100 0.5%. Primary antibodies were applied to the slices for 16 h at RT in a PBS solution containing NGS 5% and Triton X-100 0.25%. Slices were washed in NGS 2% and were incubated in the dark with the secondary antibody in a PBS solution containing NGS 2% for 1 h at RT. Slices were finally washed and mounted in Vectashield (H-1200, Vector Labs) with DAPI mounting medium. The primary antibodies used for the detection of signaling proteins were the particulate guanylyl cyclase G [GC-G (PGCG-701AP); 1:300, Rabbit, FabGennix], the cyclic nucleotide-gated channel type 3 [CNGA3 (LS-C14509); 1:300, Rabbit, Lifespan Bioscience], the phosphodiesterase 2A [PDE2A (PD2A-101AP); 1:500, Rabbit, FabGennix] and the cGMP-dependent protein kinase type II [cGKII (H-120, sc-25430); 1:50, Rabbit, Santa Cruz biotechnology]. The secondary antibody used was coupled to Cy3 [Cy3 conjugated AffiniPure anti-Rabbit (111-165-144); 1:200, Goat, Jackson ImmunoResearch]. Control experiments were performed by omitting primary antibodies. Observations and acquisitions were made by confocal microscopy (SP5, Leica) under objectives of 40–100×. Post-analysis and reconstructions were made with Imaris (Bitplane IMARIS 6.3).

#### **PROTEIN SEQUENCE ANALYSES**

The protein sequences were obtained from the National Centre for Biotechnology Information (NCBI) database. We chose the following *C. elegans* sequences DAF-11 (gi: 198447220); TAX-4 (gi: 13548488); pde-2 (gi: 71989276); egl-4 (gi: 71989393) and obtained, respectively, the following mouse homologous sequences after BLASTP algorithm: GC-G (gi: 124487301); CNGA3 (gi: 530537234); PDE2A (gi: 344217717); cGKII (gi: 188219585). Scores of identity and similarity were used to evaluate the sequence homologies.

#### **CALCIUM IMAGING**

Calcium imaging experiments were performed on acute tissue slices of mouse GG (Brechbühl et al., 2008, 2013). Briefly, pups and adult mice were killed and dissected in fresh ACSF solution at 4◦C. The tip of the mouse nose was included in a block of low melting 5% agar at 41◦C and directly placed on ice for solidification. Blocks were fixed vertically with cyanacrylat glue (Roti coll 1, Carl Roth) on the object holder and coronal slices from 60–80µm were cut at 4◦C with a vibroslicer (VT1200S, Leica). Multiple slices could be obtained from one mouse GG. Slices were selected for their GFP expression with a fluorescent stereomicroscope (M165 FC, Leica). Selected slices were loaded with Fura-2 acetoxymethyl ester (AM) (5µM; TEFLabs) and pluronic acid (0.1%; Pluronic F-127, Invitrogen) in an incubator for 45 min (37◦C, 5% CO2). Slices were placed in the bath chamber (RC-26, Warner Instruments) and immobilized with a slice anchor. Observations were made under an inverted fluorescence microscope (Axiovert 135, Zeiss) with a 25 or 63x objective and Cool SNA-HQ camera. A bipolar temperature controller (SC-20/CL-100, Warner instruments) was used to control the bath temperature. RT (room temperature) corresponded to 23–25◦C. The software MetaFluor (MetaFluor, Visitron Systems) was used to monitor intracellular calcium and to acquire images (Brechbühl et al., 2011).

#### **CHEMOSTIMULATION**

Odorant and drugs were obtained from Sigma-Aldrich and were prepared freshly before each experiment by direct dilution in ACSF. For pyrazine, a stock solution was prepared in alcohol (w:v; 1:2) before final dilution (Bargmann et al., 1993; Tonkin et al., 2002). The osmolarity of solutions was between 285 and 300 Osm/L. These chemical cues are known to reversibly mediate intracellular calcium increases in *C. elegans* when used at 1:1,000–1:1,000,000 dilutions (Lans et al., 2004; Chalasani et al., 2007). For this reason, for each tested cue, a range of concentrations from 100–1µM was used in our experiments. A short exposure to an extracellular potassium concentration of 20 mM was used as a viability test and standard reference. The percentages of responses were standardized by comparing the calcium increases observed with KCl vs. the ones observed with the tested cue. Fura-2AM ratio (F340/380 nm) observed during the perfusion of ACSF was considered as baseline activity; it corresponded to ∼5% of a KCl response. Calcium increases twice larger than this baseline activity (10% of the KCl response) were considered as responses (Brechbühl et al., 2013). 8-bromoguanosine 3 ,5 -cyclic monophosphate (8-Br; 500µM) was used to mimic the intracellular source of cGMP in GG (Schmid et al., 2010). To inhibit CNGA3 channels, the CNG inhibitor L-*cis* diltiazem hydrochloride (Dilt) was used (100–500µM) and continuously perfused on tissue slices during a minimum of 5 min before perfusing any tested cues for inhibition tests (Frings et al., 1992). In calcium free experiments, the ACSF calcium free solution was perfused during a minimum 5 min before any cues were perfused.

#### **THERMOTAXIS AND HUDDLING**

A total of 31 pups from OMP-GFP mice from 5 different litters were used to study thermotaxis and huddling behaviors. As previously described (Roppolo et al., 2006; Brechbühl et al., 2008), surgical ablation of the GG was performed in 25 P0 mice by cutting the GG axon bundles. Untreated mice did not undergo the surgical procedure (*n* = 6). In order to assess the effective surgical procedure, mice were phenotyped at P15 (after the behavioral sessions), using a stereomicroscope (MZ16FA, Leica). GFP fluorescence of the GG and, as control, the MOE was observed. Mice were considered as axotomized mice (Axo; *n* = 18*/*25) in case of total absence of GFP fluorescence at the normal localization of the GG, the other mice were considered as sham control mice (Ctrl; *n* = 7*/*25). Between behavioral tests, pups were returned to their mothers and littermates. By precaution, adult male mice were kept separate from the litters. Because no differences were measured between untreated and Ctrl mice, only results from Ctrl and Axo mice are presented for clarity purposes. The thermotaxis protocol was adapted from (Pacheco-Cobos et al., 2003; Serra and Nowak, 2008). Untreated, Ctrl and Axo mice were placed at P5, P9, and P12 in a Plexiglas arena (13 × 13 cm) in which a thermal gradient (37–0◦C) was generated by controlling the temperatures of the walls. To prevent place preference, a four-session test design was performed for each pup, where the temperatures of the walls changed clockwise. Between each session, the Plexiglas arena was cleaned with alcohol and water. Sessions of 3 min were recorded with a standard HD camera and/or a thermal camera (ThermaCAM™ E45, FLIR Systems). Post-analysis was performed with computer assistance, the position of the body center (green dots) as well as the tip of the nose (red dots, corresponding to the GG region) were reported as a function of time (*t* = 0s, *t* = 30s, *t* = 60s, *t* = 120s) and the first contact of the nose with the hottest wall was measured. To determine the huddling behavior, six pups (at P5, P9, and P12) were placed in the center of a Plexiglas arena (13 × 13 cm) at RT (23–25◦C). Sessions of 3 min were recorded with a standard HD camera and/or a thermal camera (ThermaCAM™ E45, FLIR Systems). Active behaviors (climbing, contact maintenance) as well as body temperature were measured.

#### **STATISTICS**

For statistical comparisons, open source statistical package R version 3.0.2 was used. Normality and homogeneity were evaluated by the Shapiro test. Monofactorial comparisons were done with student's *t*-tests or Wilcoxon *w*-tests. Multifactorial comparisons were done by ANOVA. Values are expressed as mean ± s.e.m. Significance levels are indicated as follows: ∗*p <* 0*.*05; ∗∗*p <* 0*.*01; ∗∗∗*p <* 0*.*001; ns for non-significant.

#### **RESULTS**

#### **CONSERVED MOLECULAR SIGNALING IN MOUSE GG NEURONS**

The amphid AWA, AWB, and AWC neurons of *C. elegans* express multiple molecular signaling proteins that are directly related to their functional roles (Bargmann, 2006). For example, in a single AWC neuron, one can find, in specific subcellular localizations, both canonical and non-canonical GPCRs (Sengupta et al., 1996; Battu et al., 2003; Alcedo and Kenyon, 2004), potential receptorlike transmembranous guanylyl cyclase DAF-11/ODR-1 (Vowels and Thomas, 1994; Birnby et al., 2000), downstream elements G*i*-like proteins (Roayaie et al., 1998; Jansen et al., 1999), phosphodiesterases (PDE) (Bargmann, 2006; O'halloran et al., 2012), cGMP-dependent protein kinase (egl-4) (L'etoile et al., 2002; O'halloran et al., 2009; Lee et al., 2010) and cyclic nucleotidegated (CNG)-like channels TAX-2/4 (L'etoile and Bargmann, 2000; Kaupp and Seifert, 2002; Bargmann, 2006). Interestingly, similar signaling proteins are also present in mouse GG neurons (Fleischer et al., 2006b, 2007, 2009; Brechbühl et al., 2008; Liu et al., 2009) (**Figures 1A,B**). In a first approach, we focused on the cGMP-dependent proteins and verified by immunohistochemistry their neuronal expression and localization in mouse GG neurons (**Figure 1C**). GG tissue slices were obtained from OMP-GFP transgenic mice (Mombaerts et al., 1996; Potter et al., 2001). We looked for the presence of the particulate guanylyl cyclase G (GC-G), a potential transmembrane receptor (Fleischer et al., 2009; Liu et al., 2009), the cyclic nucleotide-gated channels 3 (CNGA3) (Liu et al., 2009) as well as downstream regulatory elements such as the phosphodiesterase 2A (PDE2A) (Fleischer et al., 2009; Liu et al., 2009; Matsuo et al., 2012) and the cGMPdependent protein kinase of type 2 (cGKII) (Liu et al., 2009). We found GC-G exclusively in ciliary structures but CNGA3 in cilia, cell bodies and axons. The downstream element PDE2A was expressed in cell bodies and axons but cGKII was located exclusively in the cell bodies. These specific subcellular localizations were conserved among GG neurons and mice (*>*90 neurons were checked in each GG slice corresponding to *>*1000 neurons observed for the expression of each investigated proteins). Moreover, these expression patterns were identical to those found in amphid neurons (Coburn and Bargmann, 1996; Mccleskey, 1997; Coburn et al., 1998; Dwyer et al., 1998; Bargmann, 2006). We next evaluated, by BLAST algorithm, the identity and similarity of these GG-expressed proteins with the ones of amphid neurons. We found that the mouse GC-G shares 29% identity and 65% similarity with the *C. elegans* DAF-11, CNGA3 and TAX-4 share 44% identity and 91% similarity, PDE2A and pde-2 share 38% identity and 58% similarity and cGKII and egl-4 share 47% identity and 96% similarity. The similar expression patterns as well as their high similarity scores suggest that these molecular signaling elements might be an orthologous set of proteins (Altenhoff and Dessimoz, 2009).

#### **CONSERVED CHEMOSENSITIVITY OF MOUSE GG NEURONS**

In amphid neurons, these cGMP-related proteins participate in chemosensing (Bargmann, 2006). Indeed, volatile water soluble cues such as pyrazine for AWA neurons, 2-nonanone for AWB neurons, thiazole and 2,4,5-trimethylthiazole for AWC neurons (Bargmann, 2006) are known to reversibly induce intracellular calcium increases. This neuronal stimulation is partially mediated by the second messenger cGMP (Coburn et al., 1998; Bargmann, 2006). To verify if these known ligands of amphid neurons could also activate GG neurons, we performed calcium imaging experiments on GG coronal slices from OMP-GFP mice (Brechbühl et al., 2008). Tissue slices were incubated in Fura-2AM, a ratiometric calcium-sensitive dye. GG cells were identified by the intrinsic green fluorescence of GFP in their cell bodies and by their specific morphology (**Figure 2A**). The uptake of the dye was confirmed by fluorescence observations (**Figure 2B**). Chemical stimuli were delivered at room temperature in oxycarbonated ACSF continuously perfused on the tissue slices in the imaging chamber and the neuronal viability was evaluated by a brief stimulation of KCl (**Figures 2B,C**). We found that stimulation with these ligands of amphid neurons evoked calcium transients of different amplitudes in most mouse GG neurons. Interestingly, the AWA and AWC ligand 2,4,5-trimethylthiazole induced the largest responses (mT; 66.2 ± 3.6%; *n* = 66 responding neurons/66 tested neurons) (**Figure 2D**). Smaller responses were observed with thiazole (Th; 35.2 ± 3.9%; *n* = 14*/*16) and pyrazine (Py; 26.7 ± 2.0%; *n* = 32*/*34). Single GG neurons could be stimulated by all tested AWA and AWC ligands (**Figure 2C**; *n* = 14*/*14). On the other hand, in responding neurons no activation was observed with the AWB ligand 2-nonanone (No; 4.2 ± 1.3%; *n* = 0*/*18), thus conferring to GG neurons a selectivity for relatedchemical structures (Brechbühl et al., 2013). Calcium increases due to mT were rapidly reversible and reproducible (**Figure 2E**). No adaptation was observed in the presence of the stimulus for a period of 10 min (**Figure 2E**). Responses occurred over a broad range of concentrations (tested from 100–1µM) (**Figure 2F**). Recently, dependence of GG chemosensitivity on cGMP signaling has been shown using animals lacking cGMP-associated signaling proteins (Mamasuew et al., 2011b; Hanke et al., 2013). The calcium transients we observed could be mimicked by perfusion of the cGMP membrane-permeable analog 8-bromoguanosine 3 , 5 -cyclic monophosphate (8-Br; 500µM; *n* = 88*/*95) (Zufall and Munger, 2010) (**Figure 2G**). Thus, in addition to previous reports, we showed that the cyclic nucleotide-gated channel blocker L-*cis* Diltiazem (Dilt; 500µM) was able to inhibit completely the calcium transients generated by 8-Br (**Figure 2G**; *n* = 11*/*11) but only partially those generated by mT (58%) (**Figure 2H**; *n* = 6*/*6). The presence of extracellular calcium was necessary to observe an 8-Br or mT response (**Figures 2G,H**). It therefore appears that mouse GG neurons also partially display a cGMP-dependent chemosensitivity resembling the one of AWC amphid neurons (Coburn et al., 1998; Fujiwara et al., 2002; Bargmann, 2006; Tsunozaki et al., 2008).

#### **TEMPERATURE-DEPENDENT CHEMOSENSITIVITY OF GG NEURONS**

Interestingly, the chemosensitive amphid neurons, especially the AWC neurons, are also sensitive to temperature variations, which influence animal behavior (Biron et al., 2008; Kuhara et al., 2008). Indeed, genetic or laser deletion of AWC neurons, or members of their signaling pathway, demonstrated their contribution to the animal thermotactic behavior (Biron et al., 2008). Mouse GG neurons are also known to be sensitive to temperature changes (Mamasuew et al., 2008; Schmid et al., 2010) and in mice, two essential behaviors depend on temperature sensing, thermotaxis and huddling (Pacheco-Cobos et al., 2003; Alberts, 2007). We therefore tested the potential role of the GG in these two behaviors, comparing sham control mice (Ctrl) with GG axotomized mice (Axo) (Roppolo et al., 2006; Brechbühl et al., 2008, 2013). We focused on mouse pups, as the maintenance of body temperature within narrow limits is one of their most basic homeostatic needs (Pacheco-Cobos et al., 2003). Contrary to their homoiothermic mother, mouse pups are poikilothermic, which means that their body temperature varies with the temperature of their surroundings (Alberts, 2007). When mouse pups are separated from their mother, their internal temperature drops. Thus, when pups are placed in a thermal gradient, they naturally search by active movements the warmest region (Pacheco-Cobos et al., 2003). To evaluate the potential implication of the GG in mouse thermotaxis, we used a behavioral arena where a thermal gradient was generated and we filmed the pups performance with a combination of a normal and a thermal camera (**Figure 3A**). Performance was assessed in Ctrl (*n* = 7) and Axo (*n* = 18) animals of different ages by placing the pups in the middle of the arena and subsequently measuring the time of the first contact with the 37◦C wall at different ages (**Figure 3B**). As expected, age was a critical factor for the mice performance (ANOVA: ∗∗∗), but phenotype was not relevant (ANOVA: ns). Indeed, no significant differences were observed between Ctrl and Axo pups at P5 (Ctrl: 36.7 ± 5.5 s; Axo: 43.7 ± 3.9 s; *w*-test: ns), P9 (Ctrl: 18.2 ± 3.7 s; Axo: 11.1 ± 1.2 s; *w*-test: ns) nor at P12 (Ctrl: 11.5 ± 2.3 s; Axo: 15.7 ± 1.7 s; *w*-test: ns). Temporal localizations of the tip of the nose (corresponding to the GG location) as well as the body center were plotted (**Figure 3C** and supplementary **Movies 1**, **2**), which demonstrated that both phenotypes were equally efficient in thermotaxis. Indeed, after 30 s the majority of the pups had found their final position. Furthermore, for both Ctrl and Axo mice, the tip of the nose was the body part that was found physically closest to the 37◦C wall.

In addition to thermotaxis, mouse pups have to develop efficient huddling behaviors (Alberts, 2007). Huddles of pups are aggregations established by the tendency of pups to approach one another and then actively maintain contact. By huddling, mouse pups are able to preserve their body heat and thus reduce their metabolic expenditure in relation to the ambient temperature, saving their energy for growth (Alberts, 2007). Pups within a huddle are observed to root and burrow between other bodies, climb on one another, crawl around the periphery and then re-enter the huddle (Schank and Alberts, 1997). A multitude of sensory cues, including tactile, thermal and olfactory stimuli, govern the behavior of pups (Alberts, 1978). We evaluated the huddling behaviors of Ctrl (*n* = 6) and Axo (*n* = 6) mice pups (at P5 and P9) placed in the middle of an arena at room temperature and recorded for 3 min (**Figures 3D,E**). Both phenotypes were active and typical features of huddling as "climbing" and "contact maintenance" were observed. In addition, the body temperatures as well as the size of the "mice aggregates" were similar between both groups (supplementary **Movies 3**, **4**). Thus, the absence of a functional GG did not seem to affect the natural huddling behavior of mice pups. The evaluation of the phenotype was done by stereomicroscopy observations after the thermotaxis and huddling behavioral sessions at P15 (**Figure 3F**). Based on these results, these two behaviors are not governed by thermal sensors expressed in GG neurons.

At the single amphid neuron level, experiments have shown the modulation of neuronal responses by temperature (Biron et al., 2008; Kuhara et al., 2008). Moreover, it has been reported that cool temperature enhances the number

of odorant-responsive GG neurons (Mamasuew et al., 2011a). We therefore tested the presence of a temperature-dependent chemosensitivity of GG neurons by calcium imaging on GG slices from adult and young OMP-GFP mice (**Figures 4A,B**). We first exposed GG neurons to continuous temperature variations of the bath perfusion (7–39◦C) and observed a fine adjustment of the calcium level in the majority of GG neurons (**Figure 4C**; 36/48 cells). The recovery of the initial intracellular calcium level was only observed with a return to the basal temperature

(in black), AWB (in red) or AWC (in green) ligand's relationship. 2–9 mice (P1–P26) were used for each tested chemical. Values are expressed as mean

> (**Figure 4C**). In an additional set of experiment, we observed that, as in amphid neurons, the coolness-evoked GG responses were also partially dependent on cGMP signaling. Indeed, L*cis* Diltiazem (Dilt; 500µM) was able to inhibit partially (69%) the calcium transients generated by a decrease in temperature from 25–7◦C in responding neurons (*n* = 13*/*16) (**Figure 4D**; *n* = 13*/*13). The presence of extracellular calcium was necessary to observe the coolness-evoked response (**Figure 4D**; *n* = 13*/*13). We next evaluated the modulation of chemosensitivity

representative responses observed in single GG neurons for each panel.

Scale bars, 20µm in **(A)** and **(B)**.

**FIGURE 3 | A functional GG is not necessary for thermotaxis and huddling behaviors. (A–C)** Thermotaxis analysis. **(A)** Schematic representation of the arena visualized with a thermal camera. The pup is placed in the middle of the arena and, during a session of 3 min, localization is recorded. Red dot represents the GG localization and, the green dot, the body center. **(B)** The time necessary for the tip of the nose (red dot) to be in contact with the 37◦C wall is measured, for P5, P9, and P12. Experiments were done with sham control (Ctrl; *n* = 7) pups and GG axotomized (Axo; *n* = 18) pups; each pup has been used in 4 sessions. **(C)** Merge view of the

localization of the tip of the nose (red dots) and body center (green dots) after 0, 30, 60, and 120 s for P5, P9, and P12. For clarity, only the 4 sessions of 6 different pups (at the three tested ages) per phenotype were plotted **(C)**. **(D–F)** Huddling analysis. **(D)** Thermal view of 6 pups per phenotype at P5, after 40, 80, and 120 s. **(E)** Normal view at P9 after 40 s. **(F)** Phenotyping was done at the end of the thermotaxis and huddling behavioral sessions, at P15. The Ctrl (presence of GG) and Axo (no GG) mice were grouped for post-analysis. Scale bars, 500µm in **(F)**. Values are expressed as mean ± s.e.m.; *w*-test, ∗*P <* 0*.*05; ∗∗*P <* 0*.*01; ∗∗∗*P <* 0*.*001; ns, not significant.

**FIGURE 4 | The chemosensitivity of GG neurons is modulated by temperature variations. (A)** GG coronal slice from an OMP-GFP mouse, GG neurons can be observed with their intrinsic GFP fluorescence and Hoffman modulation view (Hv). **(B)** Representative Fura-2AM loaded slice observed at 380 nm in color encoded map for unbound Fura during different combination of temperature and mT perfusion. **(C–D)** Fine adjustment of the calcium level in the majority of GG neurons by temperature variations (*n* number of responding neurons/number of total neurons = 36/48; 7 mice from P1–P19). Gray lines indicate bath temperature variations. **(C)** In perfusions of 10 min, a linear intracellular adaptation of the calcium level in a single GG neuron is observed during variations in bath temperature (*n* number of tested neurons = 23). **(D)** cGMP dependence of coolness-evoked GG responses. Calcium transients could be partially inhibited by L-*cis* diltiazem [Dilt 100µM (32%, *n* = 5); 500µM (69%, *n* = 16)] and totally in calcium-free medium (*n* = 16). **(E)** Representative calcium transients induced in GG neurons by perfusion of

of mouse GG neurons by temperature. We performed calcium imaging experiments and perfused mT at different temperatures (from 25–13◦C). Interestingly, we observed in single neurons, increased mT responses with a decrease of the bath temperature (**Figure 4E**). This temperature modulation of the chemical response was significant when a 10◦C difference was applied mT (100µM) at different temperatures. Perfusions of KCl (20 mM) were used as viability control. **(F)** Normalized increased response to mT in function of the temperature (*n* = 25; 3 mice from P4–P23). **(G,H)** The apparent temperature of the tip of the mouse nose (GG region; white arrowheads) is dependent on the ambient temperature. Thermal view of a female mouse head (P21). **(G)** The GG region is close to the ambient temperature (25 ± 3◦C). **(H)** The same mouse is observed while sniffing a cold (upper panel, white square, 4◦C) or a hot (lower panel, black square, 37◦C) tube. The observed temperatures of the tip of the nose are 18 ± 2◦C and 29 ± 3◦C, respectively. Thermal gradient scales are indicated under the mouse head pictures. Fluorescence intensity Fura-2 ratio = F340/F380 is indicated by arbitrary units (a.u.). Perfusion times are indicated by horizontal bars. Traces illustrated in **(C)**, **(D),** and **(E)** are representative responses observed in single GG neurons for each panel. Scale bars, 20µm in **(A)** and **(B)**. Values are expressed as mean ± s.e.m.; *t*-test, ∗*P <* 0*.*05; ∗∗*P <* 0*.*01; ∗∗∗*P <* 0*.*001; ns, not significant.

(**Figure 4F**). Thus, we here demonstrate that, not only the number of odorant-responsive GG neurons is increased by cool temperatures (Mamasuew et al., 2011a) but also the neuronal signal intensity.

Interestingly, thermal observations of the adult mouse head (**Figures 4G,H**) allowed us to estimate the temperature of the tip of the nose (corresponding to the GG localization). It appeared to vary with the environmental temperature, confirming the fact that GG neurons may be influenced by ambient temperature. Taken together, these results strongly support the notion of conserved multisensory modalities present in the mouse GG.

### **DISCUSSION**

In *C. elegans*, the AWC class of amphid neurons are known to express different sets of proteins in parallel and convergent signaling pathways (Bargmann, 2006). These neurons possess canonical transduction cascade elements such as multi GPCRs odorant-like receptors, G*i*-like proteins as well as non-canonical cGMP effectors such as membrane potential receptor guanylyl cyclases or downstream proteins like CNG channels and regulatory enzymes (Bargmann, 2006). In mouse GG neurons, we report here the expression of a set of homologous proteins that probably retained similar functions. GG neurons have indeed preserved their cGMP-dependent activities (Mamasuew et al., 2010, 2011b; Hanke et al., 2013) and are implicated both in thermo- and chemo- sensing suggesting an orthologous status for the involved signaling proteins (Altenhoff and Dessimoz, 2009). The relevance of cGMP involvement for both chemo and temperature sensing in GG neurons (Mamasuew et al., 2011b; Hanke et al., 2013) was verified in our study. We also found that extracellular calcium was necessary to record both types of cellular responses but, since L-*cis* Diltiazem is not only a selective inhibitor of cyclic-nucleotide gated channels (Kraus et al., 1998; Gomora and Enyeart, 1999; Takahira et al., 2005) we cannot rule out that ion channels other than CNGA3 are implicated in the observed neuronal activations as suggested previously (Schmid et al., 2010).

Multimodalities are known to be important for the animal to sense its olfactory environment (Ma, 2010). In addition to chemical stimulation, physical stimuli can elicit responses in the different mouse olfactory subsystems. In the MOE as well as in the SO, neurons respond to both odorants and pressure using the same signaling pathway (Grosmaitre et al., 2007). Moreover, TRPM5 positive neurons found in mouse nasal cavities respond to a large variety of irritant odorants and may also be able to detect differences in temperature (Talavera et al., 2005; Lin et al., 2008a,b; Tizzano et al., 2010). Similarly to these examples it appears that GG neurons are both chemo- and thermosensitive (Brechbühl et al., 2008, 2013; Mamasuew et al., 2008, 2011a,b; Schmid et al., 2010; Hanke et al., 2013). We found that mouse GG neurons were activated by ligands of AWC amphid neurons in particular by the 2,4,5-trimethylthiazole. Moreover, GG neurons, like AWC neurons, display a chemosensitivity, that also responds to temperature variations. Nevertheless, fundamental differences exist between the behavior of amphid and GG neurons that limit comparisons. Thus, the neuronal responses in GG neurons occur during exposure to chemical cues, but they were observed afterwards in absence of chemical cues for amphid neurons (Chalasani et al., 2007). In addition, methylated thiazole structures seem to be rather more attractive for the nematode (Bargmann, 2006), while they are repulsive for the mouse (Apfelbach et al., 2005). These differences might be explained by evolutionary adaptation such as the loss or gain of unknown cellular or molecular switches (Tsunozaki et al., 2008).

The localization of the GG, close to the entry of the naris, is influenced by changes in the environmental temperature. Nevertheless, we showed here that the absence of a functional GG does not interfere with two specific pup behaviors namely thermotaxis and huddling. However, at the single neuronal level, temperature can modulate the chemosensitivity of GG neurons; a mechanism that could be implicated in the fine adjustments necessary to modulate the sensitivity of the neurons exposed to the outside environment. Interestingly, this thermotuning of olfactory sensing is present throughout the animal kingdom such as in nematodes (Adachi et al., 2008) or in insects (Zeiner and Tichy, 2000; Riveron et al., 2009), indicating an inherited and conserved adaptation to the environmental pressure.

Danger cues such as APs and predator scents share a structural similarity that is detected by GG neurons (Brechbühl et al., 2013). The identified mouse APs, the 2-*sec*-butyl-4,5 dihydrothiazole as well as predator scents such as the fox 2,4,5-trimethylthiazoline and the bobcat 2,6-dimethylpyrazine or other pyrazine-related cues (Mamasuew et al., 2011a) are closely related to the methyl thiazole structure that activates *C. elegans* amphid neurons. Interestingly, most of these cues are natural products of bacterial metabolism (Brown, 1979; Schellinck and Brown, 2000; Apfelbach et al., 2005; Bargmann, 2006; Zhang, 2008). The detection of these products is known to be important for odortaxis in *C. elegans* (Bargmann, 2006). In rodents, these products are, for example, generated in the guts of predators and induce upon sensing immediate attention as well as avoidance and survival strategies (Apfelbach et al., 2005). Thus, we may speculate, that the ability of an organism to detect cues from similar origin (bacterial degradation) occurs in a cluster of specialized olfactory neurons (Enjin and Suh, 2013) that has been conserved throughout evolution.

#### **AUTHOR CONTRIBUTIONS**

Author contributions: Julien Brechbühl, Fabian Moine and Marie-Christine Broillet designed research; Julien Brechbühl, Fabian Moine, and Marie-Christine Broillet performed research; Julien Brechbühl, Fabian Moine and Marie-Christine Broillet analyzed data; and Julien Brechbühl and Marie-Christine Broillet wrote the paper.

#### **ACKNOWLEDGMENTS**

We thank I. Rodriguez for the OMP-GFP mice; the Cellular Imaging Facility of the University of Lausanne and its coordinator J.-Y. Chatton; V. Ackermann and S. Citherlet from the LESBAT of the EIVD for the thermal camera and practical advices; G. Luyet, N. Hurni and M. Nenniger Tosato for their excellent technical support; R. Stoop and I. Rodriguez for fruitful discussions on the manuscript. This work is supported by the Department of Pharmacology and Toxicology, University of Lausanne, and by Swiss National Foundation Grant FNS 3100A0-125192 (to Marie-Christine Broillet).

### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at: http://www.frontiersin.org/journal/10.3389/fnbeh. 2013.00193/abstract

**Movie 1 | Thermotaxis in a sham Ctrl mouse.** A thermal gradient is generated in the behavioral arena. Here the Ctrl mouse pup (P5) is placed in the middle of the arena and the thermotaxis behavior is recorded with a thermal camera. Temperature is indicated by the pseudocolorized scale.

**Movie 2 | Thermotaxis in an Axo mouse.** A thermal gradient is generated in the behavioral arena. Here the Axo mouse pup (P5) is placed in the middle of the arena and the thermotaxis behavior is recorded with a thermal camera. Temperature is indicated by the pseudocolorized scale.

**Movie 3 | Huddling in sham Ctrl mice.** Here 6 Ctrl mouse pups (P5) are placed in the middle of an arena at RT. Huddling behaviors are recorded with a thermal camera. Temperature is indicated by the pseudocolorized scale.

**Movie 4 | Huddling in Axo mice.** Here 6 Axo mouse pups (P5) are placed in the middle of an arena at RT. Huddling behaviors are recorded with a thermal camera. Temperature is indicated by the pseudocolorized scale.

### **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: 10 October 2013; accepted: 20 November 2013; published online: 09 December 2013.*

*Citation: Brechbühl J, Moine F and Broillet M-C (2013) Mouse Grueneberg ganglion neurons share molecular and functional features with C. elegans amphid neurons. Front. Behav. Neurosci. 7:193. doi: 10.3389/fnbeh.2013.00193*

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

*Copyright © 2013 Brechbühl, Moine and Broillet. 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.*

## Sniff adjustment in an odor discrimination task in the rat: analytical or synthetic strategy?

### *Emmanuelle Courtiol †, Laura Lefèvre , Samuel Garcia , Marc Thévenet , Belkacem Messaoudi and Nathalie Buonviso\**

*Centre de Recherche en Neurosciences de Lyon, Equipe Olfaction: du codage à la mémoire, CNRS UMR 5292—INSERM U1028—Université Lyon1, Lyon, France*

#### *Edited by:*

*Regina M. Sullivan, Nathan Kline Institute and NYU School of Medicine, USA*

#### *Reviewed by:*

*Christiane Linster, Cornell University, USA Daniel W. Wesson, Case Western Reserve University, USA*

#### *\*Correspondence:*

*Nathalie Buonviso, Centre de Recherche en Neurosciences de Lyon, Equipe Olfaction: du codage à la mémoire, CNRS UMR 5292—INSERM U1028—Université Lyon1, 50 Avenue Tony Garnier, 69366 Lyon, Cedex 07, France e-mail: buonviso@olfac.univ-lyon1.fr*

#### *†Present address:*

*Emmanuelle Courtiol, Emotional Brain Institute, New York University Langone Medical Center, Department of Child and Adolescent Psychiatry and the Nathan Kline Institute for Psychiatric Research, NY, USA*

A growing body of evidence suggests that sniffing is not only the mode of delivery for odorant molecules but also contributes to olfactory perception. However, the precise role of sniffing variations remains unknown. The zonation hypothesis suggests that animals use sniffing variations to optimize the deposition of odorant molecules on the most receptive areas of the olfactory epithelium (OE). Sniffing would thus depend on the physicochemical properties of odorants, particularly their sorption. Rojas-Líbano and Kay (2012) tested this hypothesis and showed that rats used different sniff strategies when they had to target a high-sorption (HS) molecule or a low-sorption (LS) molecule in a binary mixture. Which sniffing strategy is used by rats when they are confronted to discrimination between two similarly sorbent odorants remains unanswered. Particularly, is sniffing adjusted independently for each odorant according to its sorption properties (analytical processing), or is sniffing adjusted based on the pairing context (synthetic processing)? We tested these hypotheses on rats performing a two-alternative choice discrimination of odorants with similar sorption properties. We recorded sniffing in a non-invasive manner using whole-body plethysmography during the behavioral task. We found that sniffing variations were not only a matter of odorant sorption properties and that the same odorant was sniffed differently depending on the odor pair in which it was presented. These results suggest that rather than being adjusted analytically, sniffing is instead adjusted synthetically and depends on the pair of odorants presented during the discrimination task. Our results show that sniffing is a specific sensorimotor act that depends on complex synthetic processes.

#### **Keywords: sniffing, olfaction, rat, sorption properties, discrimination, olfactomotor act**

### **INTRODUCTION**

Sampling of sensory information is achieved through dedicated motor systems. In olfaction, sniffing allows a rhythmic sampling of the environment and constrains both the timing and the intensity of the input to the olfactory structures. A remarkable feature of sniffing is its highly dynamic nature; sniffing in rats varies both in frequency and flow rate (Youngentob et al., 1987). This characteristic raises the important question of what the implications of these variations are on olfactory processing. The zonation hypothesis, proposed by Schoenfeld and Cleland (2005, 2006), is based on the observation that the olfactory epithelium (OE) is activated by both imposed and inherent patterns (for review, see Scott, 2006). Imposed patterns are determined by a complex interplay between the sorption properties of odorant molecules and the rate of nasal airflow, which affects the deposition of molecules across the OE (Mozell, 1964a,b; Mozell and Jagodowicz, 1973). In contrast, inherent patterns are determined by populations of olfactory neurons with different receptive properties distributed in distinct regions of the OE (Moulton, 1976; Kent and Mozell, 1992; Vassar et al., 1993; Freitag et al., 1995; Yoshihara and Mori, 1997). Hence, modifications of sniffing parameters affect the deposition of odorant molecules and the activation of the OE (Ezeh et al., 1995; Scott-Johnson et al., 2000; Scott et al., 2006; Yang et al., 2007; Scott et al., 2014). This body of evidence led Schoenfeld and Cleland (2005, 2006) to propose that during odor discrimination, an animal may adapt its sniffing parameters to optimize the deposition of odorant molecules on the most receptive OE areas. Sniffing would thus depend on the physicochemical properties of the odorant molecules, particularly their sorption (which depends notably on their water solubility and volatility). Rojas-Líbano and Kay (2012) tested this hypothesis and showed that rats used different sniff strategies when targeting a high-sorption (HS) molecule or a low-sorption (LS) molecule in a binary mixture. However, the strategy used for sniff adjustment when animal is confronted to a choice discrimination between two similarly sorbent odorants is still unknown. Indeed, at least two alternative strategies might exist for performing a two-alternative choice odor discrimination: (1) sniffing is adjusted independently for each odorant according to its sorption properties (analytical processing), or (2) sniffing is not adjusted independently for each odorant but is instead adjusted based on the pairing context (synthetic processing). The goal of this report was to test these possible sniffing strategies in the rat. We developed a method to non-invasively record sniffing to maintain physiological sniffing dynamics (Teichner, 1966). We used whole-body plethysmography with a two-alternative choice odor discrimination (Uchida and Mainen, 2003). We showed that similar sorption properties did not inevitably endow molecules with the property to be similarly sniffed. Moreover, we showed that the same odorant was sniffed differently depending on the odor pair in which it was presented. These results suggest that sniffing is adjusted in a synthetic manner that is dependent on the context in which the odorant is presented.

### **MATERIALS AND METHODS**

#### **ANIMALS**

Data were obtained from five male Long–Evans rats (Charles River, l- Arbresle, France) that weighed 250–300 g at the start of experimentation. Animals were housed individually at 23◦C and were maintained under a 12 h light–dark cycle (lights on from 6:00 a.m. to 6:00 p.m.). Food was available *ad libitum* during the experiment. Rats were placed under water restriction, with access to water provided during the behavioral session and for 1 h after each session. Experiments were performed in strict accordance with the European Community Council directive of November 24, 1986 (86/609/EEC), and the guidelines of the French Ethical Committee and French Legislation.

#### **SNIFF RECORDING**

The recording apparatus consisted of a whole-body customized plethysmograph (diameter: 20 cm, height: 30 cm, EMKA Technologies, France; **Figure 1A**) placed in a homemade soundattenuating cage (length: 60 cm, width: 60 cm, height: 70 cm). The apparatus was composed of two independent airtight chambers: the animal chamber and the reference chamber. The pressure changes that resulted from animal respiration were measured by a differential pressure transducer (Model dpt, EMKA Technologies) with one sensor in the animal chamber and another in the reference chamber. The measured signal was amplified, digitally sampled at 1 kHz and acquired with a PC using an acquisition card (MC-1608FS, Measurement Computing, USA).

The whole-body plethysmograph was equipped with three ports (inner diameter: 2 cm, depth: 2.5 cm; **Figure 1A**). The odor port was elevated 8 cm from the floor and was between two reward ports that were located 6 cm to the left and right of the odor port. This central odor port was connected to a homemade olfactometer. Odorants were delivered at a constant flow of 400 mL/min. Constant deodorized air also flowed through the top of the plethysmograph at a rate of 1100 mL/min.

To maintain a constant flow through the plethysmograph and to leave the sniffing signal undisturbed, a ventilation pump was connected to the whole body plethysmograph to vacuum out the equivalent of the air pushed into the chamber at 1500 mL/min (1100 + 400 mL/min).

The reward ports contained a pipette connected to a water pump. Each port was equipped with a capacitive sensor that allowed nose poke detection in the odor port and lick detection in each of the reward ports.

**FIGURE 1 | Non-invasive sniff recording. (A)** Schematic representation of the whole-body plethysmography. This setup allows us to access the respiratory signal of the animal in a non-invasive manner. The plethysmograph is equipped with three ports: an odor port (pink, central) connected to the olfactometer and two reward ports (black and white circles surrounded with blue lines) equidistant from the central odor port. **(B)** Example of one trial. The rat starts the trial by poking its nose into the central port; this motion triggers the delivery of an odorant for 3 s (green). Each odorant was associated with a reward port: odorant A/left port and odorant B/right port. If the rat makes the correct choice, water is available at the reward port for 6 s from the beginning of the trial. The first lick (light green) triggers the delivery of 60µL of water (blue).

### **BEHAVIORAL TRAINING** *Task*

We adapted a two-alternative choice odor discrimination task developed by Uchida and Mainen (2003). Rats started a trial by poking their nose into the central port. The nose poke triggered the delivery of an odorant for 3 s. Each odorant of an odor pair was associated with a reward port: odorant A/left port and odorant B/right port (**Figure 1B**). The rat had 6 s to reach a reward port and lick for water. If the rat made the correct choice, the first lick triggered the delivery of 60µL of water for 2 s. The inter-trial interval was at least 7 s. Each day, the rats performed a session with an odor pair. Each session was composed of 60–100 trials.

Animals were considered successful at discriminating an odor pair when they achieved the training criterion of 70% correct trials for each reward port on two consecutive days.

#### *Odorants*

We used the following odorants at saturated vapor pressure (Sigma-Aldrich, Fluka): methyl benzoate (mbz), ethyl benzoate (etbz), both enantiomers of carvone (L-car/D-car), both enantiomers of limonene (L-lim/D-lim), isoamyl acetate (iso), heptanol (hept), cumene (cum), and cyclooctane (cyc). Odorants were classified as LS or HS, as in Rojas-Líbano and Kay (2012). The sorption coefficients for all molecules are listed in **Table 1**. The odor pairs used were (**Table 1**):


In the discrimination task, seven pairs of odorants were used. All rats first learned the rule of the task with the odor pair mbz/etbz. Data acquired with this pair were not included in the analysis. The other odor pairs were randomly presented once the rats reached the criterion performance.

#### **DATA PROCESSING**

The sampling duration (Sd), time of odorant delivery, number of licks on each reward port, time of water delivery, and respiratory signal were recorded and stored in an SQL database using OpenElectrophy (Garcia and Fourcaud-Trocmé, 2009).

#### *Respiratory signal*

Using the whole-body plethysmography setup, the natural respiratory signal was a periodic function showing alternating negative (inspiration) and positive (expiration) deflections (**Figure 2A**). A key aspect of respiratory signal analysis was the detection of these deflections to measure respiratory cycles, which was achieved using an algorithm described in Roux et al. (2006). The **Table 1 | Physicochemical properties of the odorant molecules: vapor pressure (VP, in mm Hg at 25◦C), Henry's law constant (Kaw1 from the group method and Kaw2 from the bond method; in atm-m3/mole) and S (sorptiveness).**


*All parameters were obtained from ChemSpider. Data were generated from the ACD/Labs PhysChem Suite and US Environmental Protection Agency's EPISuitetm. The odorants used are L-carvone (L-car), D-carvone (D-car), L-limonene (L-lim), D-limonene (D-lim), heptanol (hept), methyl benzoate (mbz), isoamyl acetate (iso), cumene (cum), and cyclooctane (cyc). Odorants were classified either as low sorption (LS) or high sorption (HS).*

algorithm performed signal smoothing for noise reduction and detection of zero-crossing points to accurately define the inspiration and expiration phases. The inspiration phase started at the zero-crossing point of the falling phase and ended at the zerocrossing point of the rising phase. The expiration phase started at the zero-crossing point of the rising phase and ended at the zero-crossing point of the falling phase (**Figure 2A**). In addition, to eliminate detection artifacts, we determined a cut-off value for signal duration (rejection if value *<* median/4) and for signal amplitude (rejection if value *<* median/6).

#### *Sniffing parameters*

Before each session, the plethysmograph was calibrated by pushing 1 mL of air into the rat chamber. The resulting pressure variation in the experimental box was recorded, allowing measurement of the respiratory volume. Respiratory cycles were measured during the sampling period (Sd, **Figure 2B1**), and the number of sniffs (Ns, **Figure 2B1**) during this period was collected. Sniffs that were considered as occurring during the sampling period were cycles in which the inspiration was included in the sampling period (**Figure 2B1**). For each respiratory cycle, inspiration peak flow rate (IPF), inspiration duration (ID), and expiration duration (ED; **Figure 2B2**) were measured. To compare the values

**FIGURE 2 | Sniffing signal processing. (A)** Top: Raw sniffing signal recorded by the plethysmograph. An algorithm was applied to detect the zero-crossing points. Bottom: The blue squares represent the detection of the beginning of the inspiratory phase, and the violet squares represent the beginning of the expiratory phase. **(B1)** Sampling duration and number of sniffs. Sampling duration (Sd) is defined as the time spent in the odor port. The number of sniffs (Ns) is defined as the number of sniffs occurring during the sampling period (pink square). **(B2)** A representative sniff cycle is shown to illustrate the parameters measured: inspiration duration (ID), expiration duration (ED), and inspiration peak flow rate (IPF).

between rats and across time, we performed data normalization for each cycle by dividing the value of each parameter for each sniff by the mean value of each parameter over the whole session.

#### **STATISTICS**

Python scripts (scipy.stats) and Statview were used for all statistical analyses. We first measured the average duration of the sampling period for all the experimental sessions. We removed trials whose duration was greater than the mean sampling time plus 2 standard deviations.

The analysis focused on the respiratory signals of sessions in which the animals achieved ≥70% accuracy. Only periods resulting in a correct behavioral response were considered. To compare sniffing or sampling parameters between two odorants, we used a paired *t*-test, with the level of significance set at 0.05 (∗*p <* 0*.*05, ∗∗*p <* 0*.*01, and ∗∗∗*p <* 0*.*001).

#### **VALIDATION**

To validate our setup, we presented D-lim at two different concentrations. As demonstrated previously (Youngentob et al., 1987), we found that a decrease in odorant concentration led to an increase in IPF (data not shown). This control shows that our experimental setup allowed us to accurately measure sniffing and reproduce data obtained by others.

### **RESULTS**

We measured sniffing in unrestrained animals performing a twoalternative choice odor discrimination task (**Figure 1**). Global sniffing strategies were measured according to the Sd (time spent in the odor port) and the Ns (number of sniffs occurring during the Sd, **Figure 2B1**). We also individually analyzed each sniff cycle during the Sd. For all pairs combined, animals sampled odorants with 3*.*3 ± 0*.*026 sniffs (mean ± s.e.m.). We therefore focused our analysis on the first three sniffs following the odor onset (first, second, and third sniffs). For these three cycles, we analyzed the normalized ID, the normalized IPF, and the normalized ED (**Figure 2B2**).

Sniffing parameters were compared between odorants presented in a pair of odorants with similar sorption properties. Pairs were either HS enantiomeric (L-car/D-car), LS enantiomeric (L-lim/D-lim), HS non-enantiomeric (hept/mbz and hept/iso, with comparable and different vapor pressures in the odor pair, respectively), or LS non-enantiomeric (cum/cyc) molecules. On the whole, five pairs of odorants with similar sorption properties were tested. To test the effect of the context, we additionally used the odor pair D-car/D-lim.

### **ODORANT SORPTION AND SNIFF ADJUSTMENT: AN ANALYTICAL STRATEGY IS NOT SUPPORTED**

#### *Pairs of enantiomeric odorants*

For the two pairs of enantiomers we tested (LS/LS and HS/HS pairs), the global sampling parameters were similar; animals sniffed these enantiomers with a similar Sd and a similar Ns [**Figure 3A**, Sd: L-car/D-car pair *t*(194) = −0*.*323, *p* = 0*.*74; L-lim/D-lim pair *t*(210) = −0*.*085, *p* = 0*.*93, Ns: L-car/D-car pair *t*(194) *<* 0*.*001, *p >* 0*.*05; L-lim/D-lim pair *t*(210) = −0*.*592, *p* = 0*.*56]. Similarly, an analysis of the fine sniffing parameters revealed few or no significant differences between enantiomeric odorants, regardless of the sorption properties, as shown in **Figure 3B**. For the L-car/D-car pair (**Figure 3B**, left), a significant difference appeared only in the ID during the second cycle [*t*(187) = 2*.*994, *p <* 0*.*01] and in the IPF during the first cycle [*t*(194) = −2*.*044, *p <* 0*.*05]. We also observed few differences between L-lim and D-lim (**Figure 3B**, right) with a significant difference only in the ID during the second cycle [*t*(200) = 2*.*207, *p <* 0*.*05]. Thus, very similar molecules, such as the two pairs of enantiomers tested, induce similar sniffing strategies.

#### *Pairs of non-enantiomeric odorants*

In a second step, we analyzed sniffing strategies when the animals had to discriminate between odorants that had similar sorption properties but were non-enantiomers (**Figure 4**). Here, we surprisingly observed more heterogeneous results; the odorants of the LS/LS pair (cum/cyc) were sampled similarly, but the odorants of the HS/HS pair (hept/mbz) were sampled differently. As shown in **Figure 4**, both global sampling parameters [**Figure 4A**, Sd: *t*(165) = −0*.*513, *p* = 0*.*6084; Ns: *t*(165) = −0*.*985, *p* = 0*.*3261] and fine parameters (**Figure 4B**, right) of individual sniffs were similar for cum and cyc. In contrast, both the Sd and Ns were significantly different between HS odorants hept and mbz [**Figure 4A**, Sd: *t*(167) = −4*.*295, *p <* 0*.*001; Ns: *t*(167) = −3*.*553, *p <* 0*.*001]. When we examined each cycle individually

L-car/D-car pair is: *n*cycle 1 = 195, *n*cycle 2 = 188, and *n*cycle 3 = 175 and in L-lim/D-lim pair is: *n*cycle 1 = 211, *n*cycle 2 = 201, and *n*cycle 3 = 162. Data were analyzed using a paired *t*-test; ∗*p <* 0*.*05; and ∗∗*p <* 0*.*01.

(**Figure 4B**, left), we observed only a few differences between hept and mbz with only one significant difference in the ID during the third cycle [*t*(143) = −2*.*691, *p <* 0*.*01]. Thus, for the two different pairs of odorant molecules used that were nonenantiomers, we did not observe a sorption-based rule; molecules could be sniffed either similarly or differently even if they were endowed with similar sorption properties. This finding suggests that sorption is not the only parameter involved in sniffing adjustment and is confirmed by results from another pair of odorants, hept/iso, which have similar sorption properties (HS/HS, see **Table 1**) but differ in their vapor pressure. The results presented in **Figure 5** reveal that both global sampling parameters [**Figure 5A**, Sd: *t*(149) = 7*.*603, *p <* 0*.*001; Ns: *t*(149) = 10*.*049, *p <* 0*.*001] and fine parameters of individual sniffs [**Figure 5B**, IPF second cycle: *t*(136) = 3*.*551, *p <* 0*.*001; third cycle: *t*(64) = 6*.*962, *p <* 0*.*001; ED first cycle: *t*(149) = − 2*.*553, *p <* 0*.*05; second cycle: *t*(136) = − 7*.*941, *p <* 0*.*001; third cycle: *t*(64) = −7*.*952, *p <* 0*.*001] were significantly different between hept and iso. Thus, vapor pressure seems to enhance differences between sniffing strategies.

Taken together, these results show that odorants with similar sorption properties can be sniffed similarly (three pairs out of five) or differently (two pairs out of five). This finding suggests that sorption is not the only parameter involved in sniffing adjustment. Sniffing may not be restricted to analytical processing but instead may be adjusted synthetically, taking into account the pair in which the odorant is presented. We tested this possibility by analyzing sniffing strategies when the same molecule was presented in two different odor pairs.

#### **THE SAME ODORANT INDUCES DIFFERENT SNIFFING STRATEGIES WHEN PRESENTED IN DIFFERENT ODOR PAIRS: A SYNTHETIC STRATEGY IS LIKELY**

Three odorants, D-car, hept and D-lim, were presented in two different pairs (**Figure 6**), which allowed us to compare the sniffing pattern for the same odorant when it was presented in two different pairing contexts (see Materials and Methods). As shown in **Figure 6A**, except for D-car [Sd: *t*(205) = −1*.*424, *p* = 0*.*156; Ns: *t*(205) = −1*.*618, *p* = 0*.*107], the global sampling parameters were significantly affected by the pair in which the odorant was presented. Animals took more sniffs and remained longer in the odor port when hept was presented in the hept/iso pair than in the hept/mbz pair [**Figure 6A**, Sd: *t*(154) = 4*.*335, *p <* 0*.*001; Ns: *t*(154) = 3*.*104, *p <* 0*.*01]. Similarly, animals took more sniffs and remained longer in the odor port when D-lim was presented in the D-car/D-lim pair than in the L-lim/D-lim pair [Sd: *t*(205) = − 2*.*191, *p <* 0*.*05; Ns: *t*(205) = −1*.*976, *p <* 0*.*05].

We further measured the respiratory cycle parameters. Here too, we observed significant variations in sniffing cycle parameters depending on the pair in which the odorant was presented. As shown in **Figure 6B**, D-car induced significant differences in the ID, IPF, and ED parameters when presented in different pairs. Animals sniffed longer (higher ID and ED) and with a lower IPF when D-car was presented in the L-car/D-car pair than in the D-car/D-lim pair [**Figure 6B**, left; ID: first cycle *t*(205) = 3*.*734, *p <* 0*.*001; second cycle *t*(196) = 3*.*164, *p <* 0*.*01; third cycle *t*(180) = 8*.*022, *p <* 0*.*001; IPF: first cycle *t*(205) = −3*.*440, *p <* 0*.*001; second cycle *t*(196) = −3*.*165, *p <* 0*.*01; ED:

#### **FIGURE 5 | Continued**

*n* = 150. **(B)** Modulation of respiratory parameters in the first, second, and third cycles for hept/iso. From top to bottom: mean (± s.e.m.) normalized ID, IPF, and ED. Same colors as in **(A)**. The number of trials for each odorant and cycle in hept/iso pair is: *n*cycle 1 = 150, *n*cycle 2 = 137, *n*cycle 3 = 65. Data were analyzed using a paired *t*-test; ∗*p <* 0*.*05; and ∗∗∗*p <* 0*.*001.

first cycle *t*(205) = 4*.*702, *p <* 0*.*001; third cycle *t*(180) = 5*.*322, *p <* 0*.*001]. Similarly, hept induced a significant difference in the ID and IPF of the three cycles and in the ED of the first cycle when presented in two different pairs [**Figure 6B**, middle; ID: first cycle *t*(154) = 3*.*132, *p <* 0*.*01; second cycle *t*(147) = 3*.*775, *p <* 0*.*001; third cycle *t*(120) = 3*.*641, *p <* 0*.*001; IPF: first cycle *t*(154) = 2*.*77, *p <* 0*.*01; second cycle *t*(147) = 2*.*687, *p <* 0*.*01; third cycle *t*(120) = 2*.*27, *p <* 0*.*05; ED: first cycle *t*(154) = 2*.*041, *p <* 0*.*05]. In this case, animals sniffed hept with a higher ID, ED, and IPF in the hept/iso pair than in the hept/mbz pair. For D-lim, the respiratory cycle variations were more modest, with significant differences only in the IPF of the third sniff [**Figure 6B**, right; *t*(156) = 3*.*836, *p <* 0*.*001] and in the ED of the first cycle [*t*(205) = 3*.*891, *p <* 0*.*001].

As a whole, these results show that the same odorant presented in two different odor pairs was sniffed differently by changing the global sampling and/or the individual sniff cycle features. This finding suggests that there is no absolute sniff pattern corresponding to one odorant or a category of odorants but rather a relative pattern based on the pairing context of the odorants.

### **DISCUSSION**

By finely measuring the sniff parameters of rats performing a two-alternative odor choice discrimination task, we showed that molecules with similar sorption properties can be sniffed similarly (three out of five odor pairs tested) or differently (two out of five odor pairs tested) indicating that sniffing variations are not only governed by odorant sorption. Further, we provided new evidence that sniffing adjustment is a synthetic process dependent on the pair of odorants presented in the discrimination task.

#### **RELATION BETWEEN SNIFFING PATTERN AND SORPTION PROPERTIES**

Based on pioneering studies of OE function, the zonation hypothesis was proposed to explain the role of sniffing in olfaction (Schoenfeld and Cleland, 2005, 2006). This hypothesis proposes that sniffing optimizes the deposition of odorant molecules through the OE and is dependent on the sorption properties of the odorant molecules. Rojas-Líbano and Kay (2012) showed that odorants with different sorption properties induce different sniffing strategies. However, they did not show that odorants with similar sorption properties induce similar sniffing patterns. For the first time, we showed that enantiomers were sniffed similarly, at least true for the two odor pairs used (**Figure 3**), confirming the relationship between molecular properties and sniff parameters. However, we also demonstrated that such a strong relationship does not exist for non-enantiomeric odorants with similar sorption properties; some non-enantiomers were sniffed similarly, whereas others were sniffed differently (**Figure 4**). This observation may explain why significant sniffing variations between odorants differing in their sorption properties were not observed by Cenier et al. (2013). Several factors can account for the lack of a strict relationship between sorption properties and sniffing strategy in non-enantiomeric molecules. First, as shown recently by Scott et al. (2014), the electroolfactogram responses of medial and lateral recesses of the OE—which are anatomically optimized for odorants with different solubility—are differently affected by high nasal flow rates in active sniffing. Thus, using a larger set of odorants and/or odorants specifically activating the central zone may have led to different results. Second, physicochemical properties of the odorant other than sorption are likely components of sniff adjustment. For example, we showed that a difference in vapor pressure seems to enhance the difference of sniffing strategy between two odorants with similar sorption properties (**Figure 5**). Third, other factors related to the interaction between odorant molecules and their receptors may be involved, such as molecule/receptor affinity, molecule/odorant binding protein interactions (Pelosi, 2001), the number of receptors accessible to the molecule or enzymatic degradation of the odorant molecule (Thiebaud et al., 2013). Fourth, it is highly likely that sniffing is not adjusted in an absolute manner that is dependent on the properties of an individual odorant but instead takes into account the context (here, the pair of odorants) in which an odorant is presented.

#### **SNIFFING AS A SYNTHETIC STRATEGY**

We observed that a same odorant could induce different sniffing pattern depending on the odor pair in which it is presented. It seems that, in our experimental conditions, sniff modulation is a synthetic process taking into account not only the individual odorant properties but also the context in which an odorant is presented. This result was unexpected, and the physiological role of such a strategy is therefore of interest. We propose that a synthetic strategy may optimize sniffing to achieve the maximal decorrelation of OE activation between odorants. Indeed, in our task, the goal of the animal was to act rapidly and successfully to gain a reward. In such conditions, the aim is not to clearly identify the odorant but rather to find the most reliable clue to make the correct decision. Thus, the animal may adjust its sniffing to achieve the greatest difference between the two OE activation patterns and could thus adopt a type of intermediate sniffing pattern. This hypothesis could explain why the same odorant was sniffed differently when presented in two different pairs. We do not claim that the olfactory system does not use the sorption properties of the odorants. Rather, the system likely uses these properties in combination with other contextual information to quickly perform correct odorant discrimination.

Our observation that sniffing is synthetically modulated fits well with the concept that sniffing is modulated by higher functions such as emotion (Hegoburu et al., 2011), context (Wesson et al., 2008a), social behavior (Wesson, 2013), or attentional demand (Plailly et al., 2008). This concept implies that we could observe different results using other experimental conditions such as a different behavioral paradigm. Interestingly, the activity of the first olfactory brain relay, the olfactory bulb, is modulated by context and learning (Kay and Laurent, 1999;

Doucette et al., 2011). The dependence of sniff modulation and olfactory bulb activity on contextual clues likely originates in the complex relationship among perception, motor, motivation and respiratory pathways (Clarke and Trowill, 1971; Ikemoto and Panksepp, 1994; Kepecs et al., 2007). For example, inputs from these centers may modify olfactory bulb activity as a function of attention, motivation and learning (Gray and Skinner, 1988). These centers may also act on the respiratory system and be part of the network involved in controlling olfactomotor action.

#### **SNIFFING ADJUSTMENT: A RAPID AND FINE OLFACTOMOTOR ACT**

The olfactomotor act could be defined as a modulation of sniffing by the olfactory system. Different authors have shown that odorant presentation can modify or trigger sniffing (Welker, 1964; Alberts and May, 1980) as well as induces a concomitant modification of the firing pattern of respiratory center neurons (du Pont, 1987). The effect on the sniff is extremely fast and appears in the 50 ms following the olfactory bulb activation by an odorant (Wesson et al., 2008a,b) or even earlier if the olfactory bulb is electrically stimulated (45 ms, Monod et al., 1989). In humans, Johnson et al. (2003) also showed that sniffing could be quickly adapted, i.e., within 160 ms, depending on odorant concentration. In accordance with these observations, we showed that sniffing can vary during the first cycle following odor onset (**Figures 3**–**5**). The animal thus has the possibility to adjust his sniffing very quickly. Moreover, when we analyzed the global sniffing parameters (Sd and Ns) and the fine parameters of the first, second, and third cycles (ID, IPF, and ED), we observed that all these parameters can be modulated individually or concomitantly. As a comparison, sampling frequency and flow rate can act either independently or synergistically on bulbar output to shape the neuronal message (Courtiol et al., 2011; Esclassan et al., 2012). The animal could use each parameter independently or combine some of them to achieve specific functions. Those various modulations reveal a high flexibility in sniff adjustment and the olfactory system may need to use all the possible adjustments to improve odor representation in the olfactory center and therefore odor discrimination.

#### **CONCLUSIONS**

In conclusion, we showed that rats use a synthetic sniffing strategy that considers the pair of odorants to be discriminated. The system likely uses the odorant properties in combination with other contextual information to quickly perform correct odorant discrimination. Our results provide an additional argument demonstrating that sniffing is a specific, quickly and finely adapted sensorimotor act. Future studies will investigate the extent to which sniffing variations help or are mandatory to perform correct olfactory discrimination.

#### **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: 04 February 2014; accepted: 10 April 2014; published online: 05 May 2014. Citation: Courtiol E, Lefèvre L, Garcia S, Thévenet M, Messaoudi B and Buonviso N (2014) Sniff adjustment in an odor discrimination task in the rat: analytical or synthetic strategy? Front. Behav. Neurosci. 8:145. doi: 10.3389/fnbeh.2014.00145 This article was submitted to the journal Frontiers in Behavioral Neuroscience.*

*Copyright © 2014 Courtiol, Lefèvre, Garcia, Thévenet, Messaoudi and Buonviso. 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 olfactory bulb theta rhythm follows all frequencies of diaphragmatic respiration in the freely behaving rat

### *Daniel Rojas-Líbano1,2†, Donald E. Frederick2,3, José I. Egaña4 and Leslie M. Kay1,2,3\**

*<sup>1</sup> Committee on Neurobiology, The University of Chicago, Chicago, IL, USA*

*<sup>2</sup> Institute for Mind and Biology, The University of Chicago, Chicago, IL, USA*

*<sup>3</sup> Department of Psychology, The University of Chicago, Chicago, IL, USA*

*<sup>4</sup> Departamento de Anestesiología y Reanimación, Facultad de Medicina, Universidad de Chile, Santiago, Chile*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Thomas A. Cleland, Cornell University, USA Emmanuelle Courtiol, New York University Langone Medical Center, USA*

#### *\*Correspondence:*

*Leslie M. Kay, Institute for Mind and Biology, The University of Chicago, 940 E 57th St., Chicago, IL 60637, USA e-mail: lkay@uchicago.edu*

#### *†Present address:*

*Daniel Rojas-Líbano, Laboratorio de Neurosistemas, Facultad de Medicina, Universidad de Chile, Santiago, Chile; Facultad de Educación, Universidad Alberto Hurtado, Santiago, Chile*

### **INTRODUCTION**

Whenever the performance of a sensory system becomes crucial for an animal, it is always in the context of interaction with its immediate environment, such as when searching for food, hunting, escaping, mating, or simply exploring its surroundings. Animals execute these tasks by actively manipulating sensory stimuli through motor activities that sample physical features in time and space. Well known examples are eye movements and finger movements in primates, whisking in rodents, and sniffing in mammals. Active stimulus manipulation facilitates feature selection from complex stimuli during learning processes, ongoing adaptation to the task at hand and optimization of stimulus discrimination (Mitchinson et al., 2007; Rojas-Líbano and Kay, 2012). Neural circuits implicated in the different motor sampling modalities comprise almost all brain systems, display profuse interconnection with sensory areas and underlie higher-level cognitive functions (Liversedge and Findlay, 2000; Diamond et al., 2008; Wachowiak, 2011).

In olfaction, stimuli must be carried from external air to the sensory epithelium inside the nose. Mammals accomplish this through sniffing, a voluntary modification of regular breathing. Sniffing motor patterns have effects on the temporal structure of sensory input (Carey et al., 2009), subsequent brain processing (Carey and Wachowiak, 2011) and gain modulation of incoming sensory signals (Verhagen et al., 2007; Courtiol et al., 2011a;

Sensory-motor relationships are part of the normal operation of sensory systems. Sensing occurs in the context of active sensor movement, which in turn influences sensory processing. We address such a process in the rat olfactory system. Through recordings of the diaphragm electromyogram (EMG), we monitored the motor output of the respiratory circuit involved in sniffing behavior, simultaneously with the local field potential (LFP) of the olfactory bulb (OB) in rats moving freely in a familiar environment, where they display a wide range of respiratory frequencies. We show that the OB LFP represents the sniff cycle with high reliability at every sniff frequency and can therefore be used to study the neural representation of motor drive in a sensory cortex.

**Keywords: local field potential, olfactory bulb, respiration, rat, theta rhythm**

Rosero and Aylwin, 2011; Esclassan et al., 2012). However, there are no systematic studies of the basal conditions of the system and its internal neural representation, close to its ecological mode of operation. For example, rodents navigate through an environment constantly sensing and learning, mainly through whisking and sniffing. This allows them to gauge object distances, track entities of interest, and adapt their displacement accordingly and generally to learn about their immediate environments.

In mammals, it is hypothesized that air entry on each respiration cycle causes either a mechanical or odor signal, or both, detected by olfactory sensory neurons in the nose (Ueki and Domino, 1961; Onoda and Mori, 1980; Grosmaitre et al., 2007). This, in turn, produces a strong local field potential (LFP) oscillation in the olfactory bulb (OB) that represents the sniff cycle in voltage change over time. Early efforts by Walter Freeman combining modeling and recording hypothesized that the transient coordinated volley of impulses arriving to the OB with each inspiration would modify gain parameters of the OB neuronal network. This in turn would cause a transient increase in OB activity, reflected in LFP oscillations at the respiratory frequency, a phase-locked burst of gamma activity, and an increase in single neuron activities (Freeman, 1975). Empirical evidence for this relationship comes from anesthetized preparations (Adrian, 1950; Fontanini et al., 2003) or from short-time example traces in awake animals (Freeman, 1976; Courtiol et al., 2011b), and most of the time in the context of odorant stimulation. The normal activity range within which this respiratory-neural correlation is valid is therefore unknown, as is whether it holds in conditions of regular exploratory behavior, with non-specific odorant stimulation.

Given this lack of systematic evidence, we wanted to address at least two experimental questions. Does the strong coupling between respiration and the OB LFP hold for the continuallyadapting sniffing activity displayed in free exploratory behavior? Does this relationship change with the animal's behavior?

We addressed these questions in freely behaving rats using a task-free paradigm to encourage the rats to use a wide range of exploratory respiratory behaviors. On the motor side, we monitored the diaphragm, the main muscle target of the respiration network. Simultaneously we monitored the sensory side by recording LFPs from the olfactory bulb. We found that the relationship between motor sampling behavior and sensory activity is consistent across all respiratory frequencies without specific odorant stimulation other than the normal olfactory environment. We also found that several features of respiration are present in and therefore can be reliably extracted from OB oscillations, and that different behavioral states modulate the relationship. Our data show an ongoing sensory-motor coupling across a range of normal behaviors. In a situation like this, when there are no clearcut, experimenter-controlled stimulus delivery times, and the animal is moving freely, our results also illustrate the usefulness of investigating sensory responses using the temporal frame of the relevant motor sampling activity.

### **MATERIALS AND METHODS SUBJECTS**

Five adult male Sprague Dawley rats were used in the experiments (Harlan). Of these, two rats also participated in an olfactory discrimination protocol published elsewhere (Rojas-Líbano and Kay, 2012). All rats were housed individually in standard clear polycarbonate home cages with filter tops and maintained on a 14/10 h light/dark cycle [lights on at 8:00 AM central standard time (CST)]. All recording sessions were performed during the light phase, between 9:00 AM and 5:00 PM CST. All experimental procedures were done with approval and oversight by the University of Chicago Institutional Animal Care and Use Committee, according to Association for Assessment and Accreditation of Laboratory Animal Care guidelines.

#### **ELECTRODE FABRICATION**

LFP electrodes were made using 100μm diameter, formvarcoated stainless steel wire (304 HFV, California Fine Wire, Grover Beach, CA). On each recording site were implanted two electrodes previously glued together longitudinally, with an inter-tip separation of ∼0.5 mm. Impedance of LFP electrodes was in the range of 50–100 k-.

Ground and reference electrodes were made using 250μm diameter, polyimide-coated stainless steel wire (304 H-ML, California Fine Wire, Grover Beach, CA). A ∼10 cm long piece was cut and ∼2–3 cm of insulation removed at one end, which was wrapped around a small stainless steel screw (MPX-080-2P, Size:#0−80 × 1/8", Pan head style, Small Parts Inc., Miami Lakes, FL). At the other end, ∼0.5 cm of insulation was removed and the exposed wire was crimped to a gold connector.

EMG electrodes were made following the design of Shafford et al., designed originally for recording of the diaphragm EMG in awake rabbits (Shafford et al., 2006). A full description of the fabrication can be found elsewhere (Rojas-Líbano and Kay, 2012). Briefly, the EMG recording electrode was made using a piece of 250μm diameter stainless steel wire forming a coil of length 1–2 mm. This piece was attached to a flexible, multistranded stainless steel wire (AS 155-30, Cooner Wire, Chatsworth, CA).

Prior to surgery, EMG electrodes and their wires were disinfected in a solution of benzalkonium chloride (diluted 1:750).

#### **SURGERY**

Rats were anesthetized initially using a Ketamine-Xylazine-Acepromazine cocktail injected subcutaneously. Anesthesia during surgery was maintained using Pentobarbital sodium solution administered intraperitoneally. Rodent aseptic surgery guidelines were followed for all surgical procedures (Cunliffe-Beamer, 1993). A midline incision was done to gain access to the abdominal cavity. EMG electrodes were held by the heat shrink tubing segment using a hemostat and inserted into the diaphragm, perpendicular to the diaphragm surface. Three EMG leads were implanted in each rat, at the right costal diaphragm. The flexible wire was passed through the abdominal muscles, 2–3 cm lateral to the midline incision, and knots were tied with the same wire cable on both sides of the abdominal muscle to prevent excess wire movement. The abdominal muscle incision was sutured using absorbable polyglycolic acid suture (Surgical Specialties Co, Reading, PA). Electrode wires were tunneled under the skin and exteriorized at the back of the head. The abdominal skin incision was sutured with non-absorbable polyamide suture (Henry Schein, Melville, NY). Placement of diaphragm electrodes took ∼2.5 h from incision to suturing.

After placement of EMG electrodes, rats were placed in a stereotaxic frame. Electrodes were positioned using stereotaxic coordinates with respect to Bregma, in the OB (8.5 mm anterior, 1.5 mm lateral) and the PC (0.5 mm anterior, 3 mm lateral). The reference and ground electrodes were screwed into the skull. All gold connectors, from EMG, LFP and ground/reference electrodes, were inserted into a threaded round nine-pin socket (GS09SKT-220; Ginder Scientific) and fixed onto the rat's head with dental acrylic.

Postsurgical analgesia was provided via subcutaneous injection of buprenorphine (Buprenex®). Recording sessions were performed 15 days after surgery at the earliest.

#### **EXPERIMENTAL PROTOCOLS AND DATA ACQUISITION**

For the recording sessions, rats were connected to the recording system and then placed back in their home cages, using a modified cage top to pass the cable through it. The arrangement allowed the rat to move freely in the cage. Recordings lasted 10–25 min.

Electrophysiological data were acquired using a Cheetah-32 (Neuralynx, Bozeman, MT) amplifier system of 2.5 M input impedance, and a unity gain headstage preamplifier (HS-18- CNR, Neuralynx). During acquisition, signals were amplified (2000×) and filtered online with an analog Butterworth filter (Rolloff 12 DB per Octave). LFP channels were filtered between 1 and 200 Hz or between 1 and 475 Hz. EMG raw channels were filtered between 1 and 3000 Hz and EMG bipolar channels were filtered between 300 and 3000 Hz. In some sessions, all channels were sampled at 1892 Hz. In other sessions, LFP channels were sampled at 2000 Hz and EMG channels at 8000 Hz.

#### **DATA ANALYSIS**

All data analysis was done offline using MATLAB®. Code was written to store and process the data. Neuralynx MATLAB function Nlx2MatCSC was used to import the data into MATLAB.

#### *EMG data processing*

In order to extract the respiratory parameters (i.e., inhalation duration, tidal volume, cycle duration, inhalation strength), the EMG signal was processed. Full details of this processing are described elsewhere (Rojas-Líbano and Kay, 2012). First, the signal was rectified by taking its absolute value. Second, a moving mean filter of the signal was calculated using a window of 40 ms, with a step size of 1 sample, to produce the rectified and smoothed EMG (rsEMG). Each point of the rsEMG was calculated according to the formula:

$$\nu(t) = \frac{1}{2w+1} \sum\_{t=w}^{t+w} \varkappa(t)$$

where *x* (*t*) is the original rectified EMG, 2*w* + 1 is the dimension of the window and *y* (*t*) is rsEMG, with *t* representing the index of the sample at any given time.

#### *Inhalation-detection algorithm*

An algorithm was written in MATLAB to automate the detection of inhalation episodes. The timing of inhalation (start and end points) is the only parameter required to calculate all the respiratory timing variables. The algorithm operated on the rsEMG signal and compared it with a moving threshold signal (movingmean filter applied to rsEMG with window of 80 ms). In a first pass, any points on the rsEMG above the threshold were kept and the rest were made equal to zero. Then a collection of indices was obtained by searching for all the transitions from zero to positive (inhalation start), and from positive to zero (inhalation end). In a second pass, a minimum inhalation duration (30 ms) value was defined and all the inhalation events shorter than the minimum were removed. In a third and final pass, a minimum exhalation duration (10 ms) value was defined and all shorter events were removed. The collection of indices was used then in the original rsEMG signal to determine all respiratory variables. All inhalation episodes detected by the algorithm were visually inspected.

#### *Spectral analysis*

Spectral analysis was carried out with open-source Chronux algorithms (http://chronux.org). The multitaper method was used to estimate frequency spectra. Theory for multitaper estimates can be found in several articles and books (Thomson, 1982, 2001; Percival and Walden, 1993). In the multitaper procedure, a time-bandwidth *NW* product is defined, where *N* is the number of samples of the electrophysiological time series and *W* is the bandwidth of interest. Then multiple data tapers are selected and computed. The tapers correspond to a set of functions called discrete prolate spheroidal, or Slepian, sequences (Slepian, 1978), and usually the number of tapers *K* is selected to be not larger than 2*NW* − 1, with *NW* ≈ 4–6 a typical choice. Spectrum estimates are computed by taking the discrete Fourier transform of the time series multiplied by the selected tapers and averaging over tapers (Thomson, 1982).

Spectra estimates *S*ˆ were calculated as the average over *K* tapers as

$$\hat{S}\left(f\right) = \frac{1}{N} \frac{1}{K} \sum\_{k=1}^{K} \widetilde{\mathbf{V}}^{(n,k)} \cdot \widetilde{\mathbf{V}}^{\*(n,k)}$$

where <sup>∗</sup> means complex conjugate and **V**(*n*,*k*) is the discrete-time Fourier transform of the product between the time series *x* (*t*) and the *k*-th taper *w*(*k*) (*t*):

$$\widetilde{\mathbf{V}}^{(n,k)} = \sum\_{t=0}^{T} e^{i2\pi ft} \boldsymbol{\nu}^{(k)}\left(t\right) \boldsymbol{V}^{(n)}(t) \cdot t\_{\mathbb{S}}\left(\frac{t}{\cdot}\right)$$

where *f* represents frequency in Hz, *t* represents time and *T* is the duration of the time series in seconds (i.e., *T* = *N* · *tS*, with *tS* being the sampling interval). Spectrograms were constructed by plotting spectral power during a series of overlapping constantwidth time windows (overlap: 0.1 s; width: 0.4 s). Spectral coherence *C* between two signals (e.g., between rsEMG and LFP) was computed as

$$C\_1(f) = \frac{\hat{\mathbb{S}}\_{12}(f)}{\sqrt{\hat{\mathbb{S}}\_1(f)\,\hat{\mathbb{S}}\_2(f)}}$$

where *<sup>S</sup>*ˆ<sup>12</sup> *f* corresponds to the cross-spectrum calculated from the fast-Fourier transforms (FFTs) of the time-tapered vector signals, and *S*ˆ<sup>1</sup> *f* , *S*ˆ2(*f*) correspond to the individual power spectra of the two signals. Confidence limits (95%) were estimated for coherence magnitude by a jackknife procedure. Coherencegrams were constructed by plotting spectral coherence during a series of overlapping constant-width time windows (overlap: 2 s; width: 0.3 s).

#### *LFP polarity and LFP Phase calculations*

Our LFP recordings from the OB were made with electrodes penetrating perpendicular to the dorsal surface of the bulb. In these conditions, electrode tips ended either dorsal (D) or ventral (V) to the mitral cell layer. Recordings in D and V positions have almost identical shape but inverted polarity, since the layered architecture of the bulb forms an electrical dipole with zero isopotential surface near the mitral cell layer (Freeman, 1972). For purposes of the analysis, all LFP recordings were transformed off-line to the same polarity, by multiplying the time series by −1 where appropriate. The polarity chosen was the polarity of D recordings, i.e., where the characteristic gamma bursts occur in the falling phase of the Theta-LFP, as shown in **Figure 1A**. In this orientation, the end or peak of inhalation is coincident with the theta positive peak, and

exhalation occurs on the downward trajectory, ending with the theta trough.

After polarity correction, we first low-pass filtered the raw LFP at 15 Hz to calculate theta-LFP phase. Then we calculated the instantaneous phase of the filtered signal using the discrete Hilbert transform, from the MATLAB Signal Processing Toolbox. We considered as the start of the theta cycle the point of phase = −π of the instantaneous phase time series, i.e., the local minima of the theta-LFP, and the point of phase = π the end of the theta cycle (i.e., the next local minimum of the theta-LFP after cycle start).

#### *LFP Phase-Amplitude coupling measure*

We calculated a measure of phase-amplitude coupling to study the relationship between the phase of slow (1–15 Hz) oscillations and the amplitude of fast (40–100 Hz) oscillations in the OB LFP. Our measure was based on the methods described by Tort and colleagues in several articles (Tort et al., 2009, 2010). Briefly, we first filtered the raw LFP signal to obtain two time series containing separately the slow and fast oscillations. Then we calculated the instantaneous phase for the slow oscillation and the amplitude envelope for the fast oscillation by means of the Hilbert transform, obtaining two new time series containing the phase and amplitude measures. Out of these we calculated two estimates of phase-amplitude coupling. By binning the phase, we produced histograms of amplitude distribution over the phase. And by measuring the difference between this distribution from a uniform distribution, we also calculated a Modulation Index (MI) for the phase-amplitude coupling (Tort et al., 2010).

### **RESULTS**

Relationships between respiratory behavior and what is termed the respiratory rhythm in the OB LFP have been studied only within a narrow range of frequencies. In order to study this relationship over a wide range of respiratory frequencies we used simple free behavior in the rats' home cage. Our previous research showed a tightly defined set of respiratory behaviors during odor discrimination (Rojas-Líbano and Kay, 2012), and we have observed that OB LFP properties are highly variable dependent on behavior (Kay, 2003). We recorded from chronically implanted rats during free behavior in their home cages, using a modified cage top to allow for recording. We simultaneously acquired the olfactory bulb LFP and the diaphragm EMG signals to observe the sensory and motor activities of the animals while they explored (**Figure 1**). Cages did not contain specific odorous stimuli, but only objects usually present: feces, bedding, and food pieces. In our experience, this odorous environment produces a full range of sniffing behaviors.

#### **GENERAL DESCRIPTION OF RESPIRATORY PATTERNS**

We used the processed diaphragm signal, i.e., the rsEMG, to extract the relevant respiratory variables (see Materials and Methods and **Figure 1**). **Table 1** shows central tendency and variability measures for all rats and variables. We show a total of five rats with diaphragm (respiratory) data, one of which (rat 5) did not have LFP data and therefore is not shown in figures with LFP results (**Figures 2**–**7**). Rats displayed a continuously varying respiratory frequency, consistent with free behavior, ranging approximately from 2 to 12 Hz, and could switch between


frequencies rapidly, in 1–2 respiratory cycles (see **Figure 2**). While the entire waking respiratory range is represented in the data, all rats spent the vast majority of their time during the recorded epochs engaging in sniffing behavior (5–10 Hz), which was the object of this study. Only one rat (rat 3) produced low respiratory frequencies for a large amount of the time, and this rat had a lower mean frequency and a much larger variance in frequency than the other four rats (see **Table 1**). As we have previously shown for odorant sampling periods (Rojas-Líbano and Kay, 2012), the diaphragm signal allows us to monitor sniffing activity, and we extend it now to the spontaneously exhibited respiratory range of freely behaving rats, without interfering with airflow at the nasal cavity.

#### **RELATIONSHIP BETWEEN DIAPHRAGMATIC RESPIRATION AND THE OLFACTORY BULB LFP**

Central representation of the respiratory event in cortical signals allows us to infer how much information about the motor drive is present in the sensory signal. In this case the motor activity (inhalation, exhalation) produces changes in the external immediate environment (airflow and pressure changes in nasal cavity), which result in activation of the sensory neurons and the corresponding brain structure. We therefore assess the strength of sensorimotor coupling between the OB, a cortical area that receives the sensory signal from the olfactory nerve, and the diaphragm, the muscle driven by the output of the neuronal respiratory center (Feldman et al., 2013). We first consider whether there is a general association at all respiratory frequencies associated with many types of behaviors between diaphragmatic EMG activity and the LFP of neuronal populations in the OB. In the frequency domain, we observe a strong relationship between the signals, with high spectral coherence precisely within the range of respiration exhibited by the rats (see **Figure 3**). For each rat we calculated the spectral frequency of maximum coherence (fmaxC) and the median respiratory frequency (medianRF), and the values obtained where (in Hz): Rat 1, fmaxC = 5.67, medianRF = 5.85; Rat2, fmaxC = 8.46, medianRF = 7.84; Rat 3, fmaxC = 8.26, medianRF = 7.8; Rat 4, fmaxC = 5.82, medianRF = 5.88. This coupling is sustained over time, as shown by spectral coherence in a time-frequency chart (**Figure 3**).

We also analyzed LFP epochs surrounding the diaphragmatic respiration cycle using the motor period as a temporal frame to study the sensory signal. **Figure 4C** shows these data in raster plots of LFP segments, sorted by respiratory cycle duration. This sorting, based exclusively on EMG data, also produced a sorting of the LFP epochs, reflecting that the temporal structure of respiratory cycles was conserved in the LFP across all cycle durations. Inspection of these rasters suggests a correlation between the respiratory cycle duration (vertical black curves in rasters) and LFP cycles starting at approximately 80 ms after inhalation onset or a constant delay between the start of respiration in the

diaphragm and arrival of signal at the olfactory bulb (**Figure 4B**). We estimated this delay in all the rats by extracting the timepoints of the relevant LFP cycles. We used two approaches to extract cycles from the LFP time series. First, we calculated the maxima and minima from the low pass-filtered LFP (1–15 Hz) and defined one cycle to be the LFP epoch going from one minimum to the next. We also calculated the instantaneous phase of the LFP using the Hilbert transform, and defined one cycle to be the LFP epoch going between points of instantaneous phase = −π to phase = π. We obtained identical results with both techniques.

All the rats displayed a relatively constant delay between inhalation onset at the diaphragm and the start of a new LFP theta cycle in the OB (**Figure 4D**, blue dots). Mean and standard deviations for the delay were: Rat 1, 78.5 ± 34.1 ms; Rat 2, 79.2 ± 15.3 ms; Rat 3, 82.6 ± 12.1 ms; Rat 4, 78.8 ± 15.2 ms; Sample mean, 79.8 ± 1.9 ms. Consistent with a delay independent of respiration duration, in the linear least-squares model (LFP cycle duration) = β0+ β1.(EMG cycle duration) we found low Pearson's product-moment correlation coefficients and regression coefficients when we tried to correlate the EMG cycle duration with the simultaneously measured (i.e., exactly corresponding segment in time) LFP cycle: Rat 1, β<sup>1</sup> = 0.077, β<sup>0</sup> = 0.066, *r* = 0.0679; Rat 2, β<sup>1</sup> = 0.069, β<sup>0</sup> = 0.07, *r* = 0.095; Rat 3, β<sup>1</sup> = 0.12, β<sup>0</sup> = 0.064, *r* = 0.29; Rat 4, β<sup>1</sup> = 0.094, β<sup>0</sup> = 0.065, *r* = 0.21. (see **Figure 4D**, blue-dots scatter plots)

Conversely, when we assessed the correlation between respiration duration of individual cycles in the diaphragm and the duration of the LFP cycle shifted in time (i.e., measured from the start of a new LFP cycle and not exactly simultaneous with EMG cycle), we found that data in all rats were consistent with a cycle-by-cycle representation of diaphragm respiration in the theta LFP cycles of the rats' OB. In this case the coefficients were: Rat 1, β<sup>1</sup> = 0.78, β<sup>0</sup> = 0.029, *r* = 0.64; Rat 2, β<sup>1</sup> = 0.76, β<sup>0</sup> = 0.032, *r* = 0.62; Rat 3, β<sup>1</sup> = 0.91, β<sup>0</sup> = 0.016, *r* = 0.86; Rat 4, β<sup>1</sup> = 0.93, β<sup>0</sup> = 0.011, *r* = 0.84. (see **Figure 4D**, red-dots scatter plots).

We also analyzed the low frequency (1–2 Hz) respiratory cycles from Rat 3. These slower cycles showed an EMG-LFP delay of about 200 ms, longer than the rest of the rats and himself at higher (4–12 Hz) respiratory frequencies (**Figure 5**). Regardless of this difference, it was clear from the data that the LFP oscillations were following the respiratory cycles at these low frequencies (**Figure 5A**) as well as they did at higher respiratory rates. The relationship between respiration and the LFP is also clear in inhalation-triggered LFP means, both in low (**Figure 5B**) and high (**Figure 5C**) respiratory frequencies.

We analyzed a previously described relationship between low and high frequencies within the OB LFP. Gamma activity bursts (40–100 Hz) in the OB LFP are reported to occur phase-locked to theta oscillations (Freeman, 1976; Rojas-Líbano and Kay, 2008). This can be quantified using phase-amplitude coupling (PAC) measures such as the modulation index (Tort et al., 2010) between the user-defined low and high frequencies. We computed these coupling measures for the rats' LFP and found a consistent PAC

for all the rats with LFP and EMG data (**Figure 6**). Gamma activity, which in our recordings was 80–100 Hz for the most part, was consistently modulated by theta phase, increasing in amplitude in the second half (from phase = 0 to phase = π) of each theta cycle.

epoch. Note that the end of the EMG cycle precedes the end of the LFP

During the recording sessions, all rats alternated between two main behavioral modes: at first, they actively explored the cage, covering most of its area. Occasionally they stopped and sniffed with the tip of the nose close to the bedding, the walls or the air space above, and then resumed body movement. In the second behavioral mode, rats nose poked at a hole in one of cage walls and stayed still with the nose protruding a little to the outside of the cage, sniffing. The wall hole is the place where the rat normally finds the drinking tube of a water bottle, which was removed for the recording session. In general, rats spent 30% of the session in the nose-poke mode and the remaining 70% in the exploratory mode (see **Figure 7A**).

Respiratory behavior was considerably modified as a function of the behavioral modes. This is illustrated in **Figure 7**. In the exploratory mode, respiratory frequency varied between 4 and 12 Hz. In contrast, during nose-poke behavior it remained between 5.5 and 6.5 Hz, with very little variation throughout the period (**Figure 7B**). LFP power decreased drastically during nose-poke, both in theta and gamma ranges (**Figures 7C,D**). Coupling between LFP and EMG, however, remained high in both behavioral modes (**Figure 7E**). It is interesting to note that the reduction in LFP power seen during nosepoke periods was not the result of reduced inhalation strength, as rsEMG amplitude did not decrease compared to exploration periods (**Figure 7F**).

delay from **(C)** is also present in the non-zero slope of the line fit.

### **DISCUSSION**

The diaphragm, one of the main muscle targets of the neuronal respiratory network, and the olfactory bulb, the first brain structure to receive input from olfactory sensory neurons, were recorded during free behavior of rats exploring their home cages. We found that rats spontaneously displayed a full range of respiratory frequencies from 1 to 12 Hz and could switch rapidly between frequencies, in 1–2 respiratory cycles (**Figures 1**–**2**). The higher end of this frequency band is also associated with sniffing behavior seen in learning and odor discrimination contexts (Rojas-Líbano and Kay, 2012). We found a close relationship between motor respiratory activity and sensory LFPs in the OB, expressed as spectral coherence between the signals (**Figure 3**) and as a correlation of the cycle durations measured from both

**FIGURE 6 | Phase-Amplitude Coupling (PAC) measures for the OB LFP. Top:** example trace showing raw LFP and its Gamma (40–100 Hz) and Theta (1–15 Hz) filtered versions, from Rat 1. **Middle row:** mean LFP Gamma amplitude distribution over LFP Theta phase bins. For clarity, two cycles are shown (1 cycle = 360◦). Black horizontal line shows the expected amplitude values for a uniform distribution (i.e., no modulation). **Bottom row:** contour plots of Modulation Index (MI) for all pairs of Gamma and Theta LFP frequencies. The data are dominated in the gamma band by high frequency activity, so little coupling is shown for lower gamma frequencies. Respiration frequency distributions from EMG data are shown as the 25th and 75th quartiles (white vertical lines) and the median (white circle).

signals (**Figure 4**). Our results indicate a delay of about 80 ms from inhalation onset at the diaphragm and signal arrival in the OB (**Figure 4**), for sniffing (4–12 Hz) respiratory frequencies, and a delay of 200 ms for lower (<1.5 Hz) respiratory frequencies. We found as well a consistent LFP phase-amplitude coupling in the OB between respiratory-related frequencies (1–15 Hz) and gamma frequencies (40–100 Hz), as shown in **Figure 6**. In addition, we found a regular association between the behavioral modes exhibited in our setup and specific motor and sensory patterns of the recorded structures.

A sequence of several events is known to translate the diaphragm motor output to the respiratory rhythm in the OB. As the diaphragm contracts, air is drawn into the nose and generates a pressure wave that mechanically stimulates the OSNs and brings odorant molecules into contact with the mucosa. The OSNs appear to respond to both odorant binding at the receptors and mechanical stimulation (Grosmaitre et al., 2007), which in turn sends a volley of activity through the olfactory nerve that reaches the OB. This volley is translated by intraglomerular circuits and read by electrodes as a respiratory-linked oscillation of the LFP. If sensory drive by the olfactory nerve is interrupted by tracheotomy, naris closure or nerve transection, OB theta oscillations are almost completely abolished (Domino and Ueki, 1960; Gault and Leaton, 1963; Gray and Skinner, 1988; Courtiol et al., 2011a). Changes in respiratory cycles through the manipulation of pressure changes in the nasopharynx, change the corresponding theta frequency in the OB (Fontanini et al., 2003; Courtiol et al., 2011b; Esclassan et al., 2012).

Respiratory modulation of OB neural activity also occurs at the single neuron level (Walsh, 1956; Macrides and Chorover, 1972; Onoda and Mori, 1980; Ravel et al., 1987), with most reports recording from mitral/tufted cells. In this context, modulation means that the spiking probability of a fraction of the recorded cells is a function of the respiratory phase. Whereas some studies report that a little modulation survives after bypassing nasal breathing (Ravel et al., 1987), some others report complete abolition of modulation in conditions of no nasal breathing (Phillips et al., 2012). At any rate, it seems that the majority of the modulation comes from nasal neural input and a relatively small fraction from centrifugal inputs to the OB.

Despite reference to the respiratory rhythm in the literature over many decades, it was not known if the respiratory-related theta oscillation in the OB would follow the changing respiratory rhythms in the normal range of frequencies displayed spontaneously by behaving rats. Previous studies that had addressed this question were done in anesthetized animals that breathe at very low rates. If the rhythms are tightly linked at all frequencies, then this indicates that the OB circuit has access to and incorporates the motor program at all times. In all the rats we recorded we were able to show not only a spectral coherence between the OB LFP and the diaphragm EMG, but also a cycle-by-cycle correlation between the duration of respiratory and theta LFP cycles, with the LFP cycle lagging respiration, for all the respiratory frequencies exhibited by the rats, which in our case had the range 1–15 Hz. We found a consistent delay estimate of approximately 80 ms from the diaphragm (motor output) signal to the OB theta band LFP (cortical) signal at sniffing frequencies. It takes on the order of 20–40 ms for the diaphragm contraction to begin changing the airflow in the nose (estimated from Katz et al., 1962); we therefore conclude that the remainder of the 80 ms delay between the diaphragm signal and the OB theta oscillation is due to sorption and diffusion of odorants, transduction of odorant binding and mechanical stimulation by ORNs, transduction delays of the sensory signal and the time it takes for the target neural circuits in the OB to respond to the sensory volley.

Another prominent feature of the OB theta oscillation previously described is the phase-amplitude coupling (PAC) between the theta (1–15 Hz in the rat) and the faster gamma (40–100 Hz in the rat) oscillations. The PAC is characterized by bursts of gamma that occur phase-locked to the second half (from phase = 0 to phase = π) of the theta cycle. Extensive work done in anesthetized rats has confirmed this PAC, at least under urethane anesthesia (Buonviso et al., 2003; Cenier et al., 2009). In the awake rat, it is customary to assume that the PAC is always present, although only example traces are available in the literature (Freeman, 1976; Kay et al., 1996; Martin et al., 2004; Courtiol et al., 2011b). Again, in the entire range of frequencies exhibited by our rats we were able to detect this PAC, quantified by the Modulation Index (MI) created by Tort et al. (2010). The MI analysis of the OB LFPs showed that the bursts of gamma activity (40–100 Hz in our rats) occurred during the second half of the theta cycle (**Figure 6**) irrespective of the theta frequency. A recent report shows that this PAC is abolished during sleep (Manabe and Mori, 2013), where the "nested" gamma oscillations, as the authors refer to, could not be detected either during short-wave sleep or during rapid-eye movement sleep. These authors also report high and low frequencies for gamma oscillations, peaking around 80 and 60 Hz respectively and appearing sequentially and link these to a recent report about firing probabilities of tufted vs. mitral cells at the inhalation to exhalation transition and during late exhalation, respectively (Fukunaga et al., 2012). In our sessions, we did not see much low-frequency gamma, with most activity located between 80 and 100 Hz. However, we have reported previously the occurrence of these two types of gamma (Kay, 2003; Rojas-Líbano and Kay, 2008), especially during grooming, waiting or immobility behaviors. The fact that we did not report them here is mainly related to the types of spontaneous behaviors displayed by our rats within the span of the 10–15 min of our recordings, which was for the most part active exploration. In addition, in our rats we saw little or no beta LFP activity (20–35 Hz), which has been reported to occur mainly in response to high volatility chemicals delivered close to the animal's nostrils or as a result of operant associative olfactory learning (Ravel et al., 2003; Lowry and Kay, 2007; Martin et al., 2007; Aylwin et al., 2009).

We also report a spontaneous switch between two behavioral modes, which we found in all recorded rats. They either explored the cage, moving around it actively sniffing, or they nose-poked at a hole in the cage and stayed there, locking their respiration frequency at about 6 Hz. The switch between these respiration/behavioral modes was reflected clearly in the OB LFP, with marked changes in power at all frequencies. In turn, spectral coherence between respiration and LFP was not affected by these modes, reflecting again the capabilities of the system to maintain coupling at different frequencies (**Figure 7**).

Sniffing strategies show many connections to behaviors associated with learning and remembering odors. The sniffing rhythm itself, although it overlaps in frequency with the hippocampal theta rhythm implicated in learning and memory processes, is not normally coherent with the hippocampal rhythm. This means that there is no consistent phase relationship between the signals. However, in some learning conditions the rhythms do become coherent. One study showed that when rats learned a reward contingency reversal, sniffing showed coherence with the hippocampal theta rhythm which decayed with learning (Macrides et al., 1982). The opposite effect was shown during learning and performance of a difficult go/no-go task in which rats had to keep track of timing (Kay, 2005). In this condition OB and hippocampal theta rhythm coherence was positively correlated with performance.

Rats also adjust their sniffs as they learn; recent work from our laboratory has shown that rats use variable sniff flow strategies to detect odors dependent on their sorptive properties (Rojas-Líbano and Kay, 2012). As rats learn to discriminate odors, they increase sampling duration and use longer inhalations for low than high sorption odorants, but upon reaching criterion performance they produce lower flow sniffs for low sorption than for high sorption odorants. The current study now shows that the theta rhythm in the OB can be used to study *respiratory frequency* on a cycle-by-cycle basis and to track the sensorimotor representation within the neural circuit.

#### **ACKNOWLEDGMENTS**

Funding was provided by the NIDCD, R01-DC007995 (Leslie M. Kay). We thank James Schadt for advice on EMG electrode fabrication, and Meagen Scott for help in animal care, behavior and surgery. We also thank Cristóbal Córdova for help with **Figures 1**, **7**.

#### **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: 25 April 2014; accepted: 27 May 2014; published online: 11 June 2014. Citation: Rojas-Líbano D, Frederick DE, Egaña JI and Kay LM (2014) The olfactory bulb theta rhythm follows all frequencies of diaphragmatic respiration in the freely behaving rat. Front. Behav. Neurosci. 8:214. doi: 10.3389/fnbeh.2014.00214 This article was submitted to the journal Frontiers in Behavioral Neuroscience.*

*Copyright © 2014 Rojas-Líbano, Frederick, Egaña and Kay. 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.*

## Infant rats can learn time intervals before the maturation of the striatum: evidence from odor fear conditioning

*Julie Boulanger Bertolus <sup>1</sup> \*, Chloe Hegoburu1, Jessica L. Ahers 2, Elizabeth Londen2, Juliette Rousselot 1, Karina Szyba2, Marc Thévenet 1, Tristan A. Sullivan-Wilson2, Valérie Doyère3, Regina M. Sullivan2 and Anne-Marie Mouly1 \**

*<sup>1</sup> Lyon Neuroscience Research Center, INSERM U1028, CNRS UMR5292, University Lyon1, Lyon, France*

*<sup>2</sup> Child and Adolescent Psychiatry, Emotional Brain Institute, Nathan Kline Institute, New York University School of Medicine, New York, NY, USA*

*<sup>3</sup> Centre de Neurosciences Paris-Sud, CNRS UMR 8195, University Paris-Sud, Orsay, France*

#### *Edited by:*

*Carmen Sandi, Ecole Polytechnique Federale De Lausanne, Switzerland*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Antonios Stamatakis, National and Kapodistrian University of Athens, Greece*

#### *\*Correspondence:*

*Julie Boulanger Bertolus and Anne-Marie Mouly, CRNL, UMR 5292, INSERM 1028, Université Lyon1, 50 Avenue Tony Garnier, 69633 Lyon Cedex 7, France e-mail: julie.boulanger-bertolus@ olfac.univ-lyon1.fr; ammouly@olfac.univ-lyon1.fr*

Interval timing refers to the ability to perceive, estimate and discriminate durations in the range of seconds to minutes. Very little is currently known about the ontogeny of interval timing throughout development. On the other hand, even though the neural circuit sustaining interval timing is a matter of debate, the striatum has been suggested to be an important component of the system and its maturation occurs around the third post-natal (PN) week in rats. The global aim of the present study was to investigate interval timing abilities at an age for which striatum is not yet mature. We used odor fear conditioning, as it can be applied to very young animals. In odor fear conditioning, an odor is presented to the animal and a mild footshock is delivered after a fixed interval. Adult rats have been shown to learn the temporal relationships between the odor and the shock after a few associations. The first aim of the present study was to assess the activity of the striatum during odor fear conditioning using 2-Deoxyglucose autoradiography during development in rats. The data showed that although fear learning was displayed at all tested ages, activation of the striatum was observed in adults but not in juvenile animals. Next, we assessed the presence of evidence of interval timing in ages before and after the inclusion of the striatum into the fear conditioning circuit. We used an experimental setup allowing the simultaneous recording of freezing and respiration that have been demonstrated to be sensitive to interval timing in adult rats. This enabled the detection of duration-related temporal patterns for freezing and/or respiration curves in infants as young as 12 days PN during odor fear conditioning. This suggests that infants are able to encode time durations as well as and as quickly as adults while their striatum is not yet functional. Alternative networks possibly sustaining interval timing in infant rats are discussed.

**Keywords: olfactory fear conditioning, ontogeny, memory, interval timing, striatum, respiration, freezing, infant rats**

#### **INTRODUCTION**

The ability to time events is continuously used in humans and other animals. It leads to the prediction of events, to the production of appropriate responses, and to the detection of errors in usual temporal patterns. It allows us, for example, to estimate if we have time to cross the street or if we have to stop when the traffic light turns yellow by estimating when it will turn red. Pavlov (1927) was the first to describe the encoding of the temporal relationships between events. Pavlovian conditioning is now defined as the pairing of an initially neutral stimulus (the conditioned stimulus, CS) with an unconditioned one (US). After repeated presentations of this association, the CS elicits conditioned responses which appear to be anticipatory to the arrival of the US (Pavlov, 1927). This suggests an encoding of the temporal relationships between the CS and the US.

The ability to perceive, estimate and discriminate durations in the range of seconds to minutes is referred to as interval timing. Very little is currently known about the ontogeny of interval timing throughout development. Fitzgerald et al. (1967) demonstrated that human infants as young as 1-month-old can be conditioned to temporal regularities. Indeed, when exposed to repeated light/dark switches, they exhibit pupillary dilatation, which becomes regular regardless if the light stimulus is presented or not. Since then, several studies have compared time judgments during development and showed that time estimation improves throughout childhood (see a meta-analysis by Block et al., 1999; a review by Droit-Volet, 2011). In animals, most of the studies carried out to investigate the ontogeny of time encoding in rats have used eyeblink conditioning, which involves interval durations in the milliseconds range. This procedure usually consists of pairing an auditory stimulus with an eyeblink-eliciting US (e.g., a mild air-puff to the eye or a mild shock to the eyelid). After many CS-US pairings (usually 100–300), a conditioned eyeblink response occurs such that the peak amplitude of the conditioned response occurs at or just before the onset time of the US (Smith, 1968; Smith et al., 1969). Eyeblink conditioning has been shown to emerge gradually between post-natal (PN) day 17 and PN24 (Stanton et al., 1992; Freeman et al., 1993). However, while interstimulus interval discrimination of the conditioned eyeblink response occurs in juvenile rats, performance shows protracted development through adulthood (Freeman et al., 2003; Brown et al., 2006). Concerning the timing of longer duration intervals (seconds to minutes range), only two studies (Lejeune et al., 1986; Lejeune, 1992) investigated interval timing in juvenile (20–30 days of age) rats using a Fixed Interval procedure, where a food reinforcer was delivered in response to lever presses on a fixed 60 s schedule. This procedure is used to investigate the ability of animals to adjust to the temporal regularities of their environment. The studies by Lejeune et al. (1986) and Lejeune (1992) reported excellent temporal regulation of behavior in juvenile rats. However, since Fixed Interval procedures require multiple training days, juvenile rats reached the age of 30 days at the end of training. To our knowledge, no experiment has been conducted to test interval timing at younger ages in rats.

The neural circuits sustaining the encoding and memorization of temporal information are still a matter of debate. Numerous structures have been suggested to be involved, with considerable differences depending on the paradigms, and no structure has been exclusively related to time encoding. Yet, there is a consensus in the literature that the areas critically involved in timing depend on the timescale considered. Indeed, the cerebellum might be crucial for sub-second durations, while the striatum and prefrontal cortex might be more involved in processing seconds to minutes interval durations (for reviews, see Buhusi and Meck, 2005; Meck et al., 2008). More specifically, a number of recent studies have demonstrated that the striatum is an important component of the interval timing system both in humans and animals (reviewed in Buhusi and Meck, 2005; Coull et al., 2011). However, anatomical studies showed that the striatum morphological maturation occurs around PN15 (Chronister et al., 1976; Tepper and Trent, 1993). Moreover, Tepper et al. (1998) reported that the electrophysiological characteristics of the striatal neurons continue to mature throughout the preweaning period. This delayed ontogenesis of the striatum raises the question of whether rats are capable of timing intervals prior to striatum development. The global aim of the present study was to investigate interval timing abilities in rats at an age for which the striatum was not yet mature.

We addressed this question using odor fear conditioning. Indeed, in comparison with the Fixed Interval procedure, fear conditioning makes very little demand on the rat motoric capabilities. Furthermore, odor fear conditioning can be applied to very young animals (Sullivan et al., 2000) as their sense of smell is fully functional at birth, contrary to vision and audition.

We first investigated whether the striatum, known for its role in time processing, was activated by odor fear conditioning in developing rats using 2-DG autoradiography. Although fear learning was shown at all tested ages, activation of the striatum was observed in adults but not in juvenile animals. Next, we assessed the presence of evidence of interval timing in ages before and after the inclusion of the striatum into the fear conditioning circuit. For this, we used an experimental setup allowing the simultaneous recording of freezing and respiration which have been demonstrated to be sensitive to interval timing in adult rats (Hegoburu et al., 2011; Shionoya et al., 2013). This enabled us to assess the emergence of temporal patterns during the acquisition of the odor-shock association in both adult and infant rats. Such analysis revealed evidence of interval timing in ages prior to and after the functional maturation of the striatum.

### **METHODS**

#### **ANIMALS**

The subjects were male and female Long Evans rats born and bred either in the Nathan Kline Institute colony (originally from Harlan, USA) or in the Lyon Neuroscience Research Center (originally from Janvier, France). Only one female and one male pup per litter per treatment/test condition were used for all experiments and animals from the same litters were used in the different treatment/test conditions and ages. A total of 20 litters were used. Three groups of ages were used: PN day 12–15 (PN12– 15, infants), PN22–24 (juveniles) and older than PN75 (adults). Day of birth was considered PN0. Pups were maintained with their litters up to the end of the experiments, including juvenile pups. Adults were housed by pairs at 23◦C and maintained under a 12 h light-dark cycle (lights on from 6:00 am to 6:00 pm). Food and water were available *ad libitum* and abundance of wood shavings was supplied for nest building. All experiments were conducted in strict accordance with the Institutional Animal Care and Use Committee of the Nathan Kline Institute, which follows the guidelines from the American National Institutes of Health, and with the European Community Council Directive of November 24, 1984 (84/609/EEC) and the French National Committee (87/848) for care and use of laboratory animals. Care was taken at all stages to minimize stress and discomfort to the animals. An overlap in personnel conditioning/testing both infant and adult rats in France and the USA ensured consistency of conditioning and testing of infant and adult animals between labs.

#### **TRAINING APPARATUS**

Equivalent conditioning apparatus and procedures were used between labs as previously described (USA: Coulbourn equipment, described in Sevelinges et al., 2008; France: Emka and Coulbourn equipment, described in Hegoburu et al., 2011). In the USA, a standard Coulbourn shock chamber was used. In France, a custom built Plexiglas conditioning chamber for fear conditioning was used and equipped with features allowing fine-grain freezing analysis (4 camera views, B/W CMOS PINHOLE camera, Velleman, Belgium, and an homemade acquisition software using Matrox Imaging Library and acquisition card, Matrox video, UK) and a plethysmograph (diameter 30 cm, Emka Technologies, France) for measuring respiratory frequency (see Hegoburu et al., 2011 for further description of the plethysmograph). The height of the plethysmograph was adapted to the age of the animal in order to optimize the signal-noise ratio, leading to a height of 30 cm for the adults and 16.5 cm for the infants.

#### **ODOR FEAR CONDITIONING PROCEDURE**

CS-US parameters were standardized between labs. The American lab used the Coulbourn FreezeFrame software for stimuli delivery control and video recording. The French lab used a custom build program for stimuli delivery and recording. Conditioning took place in a sound attenuation chamber with deodorized air constantly flowing through the cage (2 L/min). The odor CS was a 30 s peppermint odor (McCormick Pure Peppermint; 2 L/min; 1:10 peppermint vapor to air) and was controlled with a solenoid valve that diverted the airflow to the peppermint air stream, thus minimizing pressure change. The 1-s mild electric shock was delivered through a grid floor. In the US, for the youngest pups, the shock was delivered through an electrode to one hind limb. At both sites, adult rats were handled for about 4 days and placed into the conditioning chamber for context habituation. Juveniles received only 1 day of handling and habituation while infants, for which conditioning to context is not yet developed (Raineki et al., 2010), were not handled to minimize distress from separation from the mother.

Three training conditions were used throughout the experiments: Odor-shock pairings (Paired condition), Odor-shock unpaired presentation (Unpaired condition) and Odor-alone presentation (Odor group). In the paired groups, during the first 10 min of the conditioning session, the animals were allowed an adaptation period of free exploration. Then the CS odor was introduced into the cage for 30 s, the last second of which overlapped with the shock. The animals received ten odor-shock trials, with an inter-trial interval of 4 min. In the Unpaired groups, the same procedure was carried out except that the shock and the odor were explicitly unpaired using a fixed long duration (180 s) between the odor onset and the shock arrival. In the Odor groups, the odor was presented alone for 30 s.

#### **RETENTION TEST**

A subset of animals at the three ages (Infants: *n* = 17; Juveniles: *n* = 22; Adults: *n* = 26) were tested the day after conditioning for learned fear responses. Testing was done by an experimenter blind to the training conditions.

Infants were tested in a y-maze (start box: 8*.*5 × 10 × 8 cm; choice arms: 8*.*5 × 24 × 8 cm), one arm containing the peppermint odor CS (20μL peppermint on Kim Wipe), and the other containing familiar pine shavings (20 mL clean bedding). Pups were given 5 trials. For each trial, the pup was placed in the start box (5 s), the door to each alley opened and the animals were given 60 s to choose an arm. The number of choices of the arm containing the CS odor was compared between groups using a One-Way ANOVA with the Group as an independent factor.

For adults and juveniles, the rat was placed in a novel experimental cage with new contextual (tactile and visual) cues and allowed a 10-min odor-free period. The CS odor was then presented five times (30 s with a 4-min inter-trial interval) and the animal's freezing response was hand scored by an experimenter blind to the training conditions. The average freezing behavior of a 25 s period preceding odor delivery (Pre-CS period) was compared with the full odor presentation period (CS period) using a Two-Way ANOVA with Group (Paired, Unpaired and Odor) as an independent factor and Period (Pre-CS vs. CS) as a repeated measures factor. Significant ANOVAs were followed by *Post-hoc* Fisher comparisons.

#### **STRIATUM 2-DG AUTORADIOGRAPHY**

In the 14C 2-DG autoradiography, the animals (adults: *<sup>n</sup>* <sup>=</sup> 17, juveniles: *n* = 18) were injected with 2-DG (20μCi/100 g) 5 min prior to conditioning (Paired, Unpaired and Odor). Brains were removed immediately after the 45 min of conditioning, frozen in 2-methylbutane (−45◦C) and stored at −70◦C. Brains were then cut into 20μm coronal sections (following equilibration to −20◦C) and placed in exposure cassettes for 4 days with standards (14C methylmethacrylate standard 10 <sup>×</sup> <sup>0</sup>*.*02 mCi; American Radiolabeled Chemicals, Inc.), scanned (Epson) and analyzed using NIH Image J Software for quantitative optical densitometry. The dorsal striatum was divided into anterior (Bregma +1.9 mm, Paxinos and Watson, 1986) and posterior dorsal striatum (Bregma +0.1). Within each part, the lateral and medial dorsal striatum were differentiated, as they receive inputs from different parts of the brain (McGeorge and Faull, 1989). Coordinates were adjusted in pups to obtain similar measure locations in both adults and pups. To control for potential differences in section thickness and autoradiograph exposure, 2-DG uptake was measured relative to 2-DG uptake in the corpus callosum that did not vary between conditioning groups (Sullivan and Leon, 1986; Sullivan et al., 2000). For each developmental age, the data from the anterior and posterior parts of the dorsal striatum were analyzed using a Three-Way ANOVA with Group as an independent factor and Anteriority (anterior vs. posterior part) and Laterality (lateral vs. medial part) as repeated measures factors. A One-Way ANOVA was then carried out on the data of each of the four subparts, followed by *post-hoc* Fisher comparisons allowed by the ANOVA results.

#### **ASSESSMENT OF TEMPORAL PATTERNS OF FREEZING AND RESPIRATION DURING ACQUISITION**

To assess interval timing during development we used olfactory fear conditioning in PN12–15 infant (*n* = 30) and adult rats (*n* = 25), ages representative of before and after functional striatal development. Three experimental groups of animals per age were trained using either 30 s (Paired 30 s group), or 20 s (Paired 20 s group) Odor-Shock interval duration, or Odor alone presentation (Odor group).

In each experimental group, the time course of respiration and behavior was monitored throughout the acquisition session. Offline, the respiratory signal was analyzed and momentary respiratory frequency was determined. The animal's freezing behavior was automatically detected using a LabView homemade software that had been validated by comparison to hand scoring by an experimenter blind to the rats' condition. Definition of the freezing at the different ages followed the methods defined by Takahashi (1992) and takes into account the immaturity of infants' musculoskeletal system. Instant respiratory frequency and freezing were averaged on a second by second basis, leading to 1-s time bins curves. The resulting individual curves were then averaged among animals of the same experimental group.

The temporal dynamics of the recorded parameters in presence of the CS odor was compared using a Two-Way ANOVA with Group as an independent factor and Time as a repeated measures factor. *Post-hoc* pairwise comparisons were then carried out when allowed by the ANOVA results. In addition in each group, a One-Way ANOVA for repeated measures was carried out to compare each CS-bin (i.e., seconds 1–19 in the Paired 20 s group and seconds 1–29 in the Paired 30 s group) to the averaged parameter value during 25 s prior the odor arrival as a baseline. *Post-hoc* within group comparisons were then carried out when allowed by the ANOVA results. For all the statistical comparisons performed, the significance level was set at 0.05.

In order to assess whether the temporal patterns of freezing and respiratory rate observed for the Paired 20 s and Paired 30 s groups are related to the CS-US interval, we assessed whether they respected the scalar property. Indeed, a remarkable property of interval timing is that the error of time estimation varies linearly with the estimated interval. This is known as the scalar property of time estimation (Gibbon, 1977). As a consequence, when the behavioral temporal response functions are normalized by the criterion time (multiplicative rescaling), they superimpose (reviewed in Matell and Meck, 2000). To test if this property was respected for the freezing and respiratory rate temporal responses, the time axis for the individual curves of the Paired 30 s rats was multiplicatively rescaled so that the 19 bins of time of CS for both groups represented the same proportions of elapsed time from CS onset to shock presentation. The scalar timing rule predicts superior superposition of the functions in relative time (multiplicative transform, comparison of the Paired 20 s average curve to the rescaled Paired 30 s average curve), compared to no rescaling (comparison of the Paired 20 s average curve to bins 1–19 of the Paired 30 s average curve) or rescaling by an additive shift of x-axes (additive transform, comparison of the Paired 20 s average curve to bins 11 to 29 of the Paired 30 s average curve). Superposition was indexed by eta-square (η2), a measure of the proportion of variance accounted for by the mean of the two functions (Brown et al., 1992). When superposition is perfect, η<sup>2</sup> is at its maximum value of 1. Superposition was assessed on curves obtained with absolute freezing and respiratory rate values as well as with data normalized in relative response rate (maximum minus minimum values) on the Y axis.

### **RESULTS**

#### **FEAR LEARNING: ODOR-SHOCK CONDITIONING PRODUCED LEARNING AT ALL AGES**

To assess learning from odor-shock conditioning at all ages, some animals were conditioned and tested for retention approximately 24 h later. As illustrated on **Figure 1**, all paired odorshock conditioning animals showed learning at testing. Infants tested in a Y-Maze (**Figure 1A**) showed a significant Group effect [*F*(2*,* 14) = 13*.*404, *p* = 0*.*001] and *post-hoc* Fisher analysis indicated that the paired groups were significantly different from each of the controls (Paired vs. Unpaired *p* = 0*.*002; Paired vs. Odor *p <* 0*.*001). Juvenile and adult rats received a cue test (CS presentations) in a novel context with learning assessed through freezing rate and analyzed with Two-Way ANOVA with repeated measures (Pre-CS vs. CS). Juvenile animals (**Figure 1B**) showed a significant Group × Period interaction [*F*(2*,* 20) = 4*.*234, *p* = 0*.*029]. *Post-hoc* analysis evidenced a significant freezing increase during the CS odor in the Paired group (*p* = 0*.*02) while this increase was not present in the Unpaired and Odor groups (*p >* 0*.*05). Adult rats (**Figure 1C**) showed a significant Group × Period interaction [*F*(2*,* 23) = 6*.*462, *p* = 0*.*006]. In the Paired group, the level of freezing was higher during the CS presentation than during the Pre-CS period (*p <* 0*.*05), while it remained unchanged in the Unpaired and Odor groups (*p >* 0*.*05).

Therefore, the training paradigm used in the present study resulted in good memory at the three developmental ages considered.

Unpaired, *n* = 9; Odor, *n* = 6) and **(C)** adults (Paired, *n* = 8; Unpaired, *n* = 8; Odor, *n* = 10) received a cue test in a novel environment and the freezing rate (mean <sup>±</sup> s.e.m.) was compared between pre-CS and CS Odor. #Significant intragroup difference (*p <* 0*.*05); ∗significant intergroup difference (*p <* 0*.*05).

#### **STRIATAL14C 2-DG AUTORADIOGRAPHY: THE POSTERIOR DORSAL STRIATUM IS ACTIVATED DURING ODOR-SHOCK CONDITIONING IN ADULTS BUT NOT IN JUVENILES**

The level of activation of the dorsal striatum (**Figure 2A**) during odor fear acquisition was assessed in adults and juveniles using 14C 2-DG autoradiography.

In adults (**Figure 2B**), the Three-Way ANOVA revealed a significant effect of Anteriority [*F*(1*,* 14) = 21*.*65, *p <* 0*.*001], Laterality [*F*(1*,* 14) = 19*.*67, *p* = 0*.*001] and Laterality × Group interaction [*F*(1*,* 14) = 4*.*51, *p* = 0*.*03]. Further One-Way ANOVA revealed that in the anterior part of the dorsal striatum (upper panel) the level of activation did not vary between groups for both the lateral [*F*(2*,*14) *<* 1] and medial [*F*(2*,* 14) = 1*.*24, *p* = 0*.*32] parts. In the posterior part of the striatum (lower panel), the One-Way ANOVA revealed a main effect of the Group in the medial part [*F*(2*,* 14) = 3*.*62, *p* = 0*.*05] but no significant effect in the lateral part [*F*(2*,* 14) = 2*.*50, *p* = 0*.*12] although the same tendency was observed. *Post-hoc* Fisher tests revealed that in the medial part, the Paired group exhibited a higher level of 2-DG uptake compared to the Odor group (*p* = 0*.*021) and the unpaired group (tendency, *p* = 0*.*066).

In juvenile rats (**Figure 2C**), the activation of the striatum was at a similar level between all groups regardless if the measure was taken on the anterior or the posterior dorsal striatum, medial or lateral parts (for all comparisons: *F <* 1).

In summary, in adult rats, odor fear conditioning was associated with an increased activation in the posterior part of the dorsal striatum, while in juveniles no increase was observed.

#### **TEMPORAL PATTERNS FOR FREEZING AND RESPIRATION DURING ACQUISITION: INFANTS SHOW A TEMPORAL PATTERN OF US EXPECTANCY**

In order to assess the ability of animals of the different ages to encode interval duration in our odor fear conditioning paradigm, the temporal patterns of responses of animals conditioned to a 30 s and to a 20 s CS-US interval were compared in adult and infant rats. For both ages, we examined the temporal patterns of freezing and respiration during the CS presentation in 1-s time bins. The average curves of the last four odor-shock pairings were pooled together for each parameter (i.e., freezing and respiratory rate). Both parameters were compared between and within groups using a Two-Way ANOVA with factor Group (Paired 20 s, Paired 30 s, and Odor) and repeated factor Time. The comparison was made on the common CS duration between the Paired 20 s and Paired 30 s groups (1–19 s).

**rats.** The temporal pattern is represented with a 1-s bin precision, from the odor onset (green vertical line on each graph) to shock arrival (red and blue

#### *Adults*

For freezing (**Figure 3**, left panel), the ANOVA revealed a significant effect of Group [*F*(2*,* 22) = 51*.*709, *p <* 0*.*001], Time [*F*(18*,* 396) = 1*.*906, *p* = 0*.*014] and Group × Time interaction [*F*(36*,* 396) = 1*.*831, *p* = 0*.*003]. Further analyses revealed that both Paired groups showed higher levels of freezing than the Odor group [Paired 20 s vs. Odor, Group: *F*(1*,* 15) = 52*.*737, *p <* 0*.*001; Paired 30 s vs. Odor, Group: *F*(1*,* 15) = 125*.*292, *p <* 0*.*001]. In addition, there was a significant difference between Paired 20 s and Paired 30 s groups temporal patterns during CS presentation [Paired 20 s vs. Paired 30 s, Group × Time: *F*(18*,* 252) = 2*.*456, *p* = 0*.*001]. *Post-hoc* within group comparisons showed that, in Paired 30 s animals, introduction of the CS induced a significant decrease in freezing (at seconds 10, 14 to 19, 21, 22, and 28 of the CS, *p <* 0*.*05) compared to baseline. In the Paired 20 s or in the Odor group no change in freezing rate was observed in presence of the CS odor.

Concerning respiration (**Figure 3**, right panel), the ANOVA revealed a significant effect of Group [*F*(2*,* 22) = 14*.*498, *p <* 0*.*001], Time [*F*(18*,* 396) = 13*.*033, *p <* 0*.*001] and Group × Time interaction [*F*(36*,* 396) = 3*.*341, *p <* 0*.*001]. Further analyses revealed a significantly higher respiratory rate in the Paired groups compared to the Odor group [Paired 20 s vs. Odor, Group: *F*(1*,* 15) = 20*.*85, *p <* 0*.*001; Paired 30 s vs. Odor, Group: *F*(1*,* 15) = 22*.*927, *p <* 0*.*001], as well as a significant difference between Paired 20 s and Paired 30 s groups temporal patterns [Paired 20 s vs. Paired 30 s, Group × Time: *F*(18*,* 252) = 2*.*551, *p* = 0*.*001]. In addition, within-group comparisons showed that while Odor animals showed almost no reaction upon odor arrival compared to baseline due to habituation, Paired animals' respiration rate increased significantly following odor onset. In particular, in Paired 20 s animals the increase was significant from seconds 5 to 14 of the CS (*p <* 0*.*05), after which the rate returned to baseline value (*p >* 0*.*05). In Paired 30 s animals, the respiratory rate increased significantly from second 4 of the CS until the shock arrival (*p <* 0*.*05).

In order to assess whether the temporal patterns of freezing and respiratory rate described above for the Paired groups are related to the CS-US interval, we tested whether they respected the scalar property as explained in the Methods.

When considering the freezing curves of Paired 20 s and Paired 30 s groups (**Figure 4**, upper part), superposition was better under the additive transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*49, right side) than under the multiplicative transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*26, middle) or no transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*24, left side). In contrast, for respiration (**Figure 4**, lower part), superposition was better under the multiplicative transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*85, middle) than under the additive transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*58, right side) or no transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*82, left side). The foregoing analyses were based on absolute response values for each dependent measure. When data were re-plotted in relative response rate on the Y axis, the results were similar, showing superior superposition with additive transform for freezing (additive: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*67, multiplicative: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*36, no transform: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*20), and with multiplicative transform for respiration (multiplicative: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*82, additive: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*53, no transform: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*80). Thus, the scalar property was best respected for respiration while it was not observed for freezing.

In summary, adult rats exhibited different temporal patterns of freezing and respiration rates for the two CS-US intervals. Scalar rules were respected for the respiratory rate but not for freezing, which confirms our previous data (Shionoya et al., 2013).

#### *Infants*

The temporal patterns of freezing and respiration rate were assessed in infants using the same analysis parameters as for adults.

Concerning freezing (**Figure 5**, left side), the ANOVA evidenced a tendency for a Group effect [*F*(2*,* 27) = 3*.*01, *p* = 0*.*066],

**respiration (lower part) temporal patterns in adult rats.** The scalar timing rule predicts better superposition of the curves in relative time (multiplicative transform, **middle panel**), compared to no rescaling (**left** Superposition between the two curves was indexed by eta-square (η2) indicated in the bottom right of each graph, the highest values being highlighted in bold characters.

and a significant effect of Time [*F*(18*,* 486) = 12*.*988, *p <* 0*.*001] and Group × Time interaction [*F*(36*,* 486) = 6*.*656, *p <* 0*.*001]. Further analysis revealed that both Paired groups showed significantly different levels of freezing than Odor group [Paired 20 s vs. Odor, Group: *F*(1*,* 18) = 4*.*256; *p* = 0*.*054; Paired 30 s vs. Odor, Group: *F*(1*,* 18) = 4*.*489, *p* = 0*.*048]. In addition, the temporal patterns of Paired 20 s and Paired 30 s animals were significantly different [Paired 20 s vs. Paired 30 s, Group × Time: *F*(18*,* 324) = 4*.*811, *p <* 0*.*001], although similar levels of baseline and preshock (3 last seconds before shock) freezing were observed in both groups (*p >* 0*.*05). In infants, introduction of the CS odor induced a strong decrease in freezing in the Paired groups. Indeed, within group comparisons showed that, in Paired 20 s animals, this decrease was significantly different from baseline from second 6 until shock arrival (*p <* 0*.*05), whereas in the Paired 30 s group significance was reached from second 17 until shock arrival (*p <* 0*.*05).

Concerning the respiration (**Figure 5**, right panel), the ANOVA revealed an effect of Group [*F*(2*,* 27) = 17*.*588, *p <* 0*.*001], Time [*F*(18*,* 486) = 22*.*209, *p <* 0*.*001] and of the Group × Time interaction [*F*(36*,* 486) = 5*.*123, *p <* 0*.*001]. Further analysis evidenced that in both Paired groups the respiratory rate was significantly higher than in the Odor group [Paired 20 s vs. Odor, Group: *F*(1*,* 18) = 98*.*076, *p <* 0*.*001; Paired 30 s vs. Odor, Group: *F*(1*,* 18) = 10*.*581, *p* = 0*.*004]. In addition, the temporal patterns of Paired 20 s and Paired 30 s animals were significantly different [Paired 20 s vs. Paired 30 s, Group × Time: *F*(18*,* 324) = 3*.*753, *p <* 0*.*001], although similar respiratory rates were observed at baseline and prior the shock arrival in the two groups (*p >* 0*.*05). Within group comparisons showed that, in the Paired 20 s group, introduction of the CS Odor induced a significant increase in respiratory rates compared to baseline from second 3 of the CS until shock arrival (*p <* 0*.*05), whereas in Paired 30 s animals, the increase reached significance from second 6 until the end of the CS (*p <* 0*.*05).

When the scalar property was tested on the freezing curves (**Figure 6**, upper part), the highest superposition index was obtained for the multiplicative transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*82, middle)

represents the averaged baseline. Paired 20 s group: red (*n* = 10), Paired 30 s group: blue (*n* = 10) and Odor group: green (*n* = 10). Filled circles on each curve indicate the points that are significantly different from the baseline (*p <* 0*.*05).

compared to additive transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*71, right side) or no transform (η<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*70, left side). Similar results were observed for the respiratory rate curves (**Figure 6**, lower panel; multiplicative: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*61, additive: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*53, no transform: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*48). When response rates were re-plotted in relative values on the Y axis, the results were similar, showing superior superposition with multiplicative transform for freezing (multiplicative: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*85, additive: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*78, no transform: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*67), and for respiration (multiplicative: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*91, additive: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*63, no transform: <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>*.*82). Thus, in infant rats, the scalar property was respected for both respiration and freezing.

In summary, infant rats showed different temporal patterns of freezing and respiration rates for the two duration intervals. Both freezing and respiration curves followed the scalar rules, thus supporting the hypothesis that infant pups expressed temporal learning.

#### **DISCUSSION**

The present study investigated the ontogeny of time durations encoding in odor fear conditioning. We first assessed whether the striatum, which is a brain area implicated in timing, was differentially activated in odor fear conditioning throughout development. This is the first study to assess striatum activity across development within this paradigm. The 2-DG metabolic mapping study revealed that, while dorsal striatum was activated during odor fear acquisition in adults, its activity remained unchanged in juveniles. These data are in agreement with data on striatum development, which indicates the juvenile striatum is still maturing (Chronister et al., 1976; Tepper and Trent, 1993; Tepper et al., 1998). To investigate interval timing abilities before and after the striatum functional maturation, we performed finegrain analysis of behavioral and physiological responses shown to be valuable for assessment of timing in adults during the early stages of odor fear learning. Specifically, we monitored the time course of the animal's respiratory rate and freezing behavior during conditioning at two developmental ages: one with striatum activity correlated with learning (adult) and the other for which the striatum has been shown to be immature in the literature (infant). The data showed that for both adult and infant rats, duration-related temporal patterns can be detected for freezing and/or respiration curves suggesting that infants are able to encode time durations as well and as quickly as adults while their striatum is not yet functional.

#### **IN ADULT ANIMALS RESPIRATION RATE IS A RELIABLE INDEX OF TIME ENCODING**

The present data show that in adult animals, the temporal patterns of respiration rate in the Paired 20 and 30 s groups were significantly different and followed the scalar property. In contrast, freezing temporal patterns, although different, did not respect scalar rules. These data confirm the results we obtained in a previous study using a slightly different paradigm (Shionoya et al., 2013). Indeed in that study, the same animals were first conditioned to a 20 s odor-shock interval and then shifted to a 30 s interval. This resulted in a shift of the temporal pattern of respiration toward the new duration. However, while for analytical reasons studies of timing usually manipulate the interval durations within rather than between subjects (Brown et al., 1992), this procedure may lead to biases in the temporal pattern of conditioned responses due to the previously learned duration. Here we show that differential patterns are observed when different groups of animals are used for the 20 and 30 s CS-US intervals, thus when the two groups underwent an equivalent amount of training.

The present study also confirms that, in adults, respiration is a more sensitive index than freezing to investigate the emergence of duration-related temporal patterns within a few trials (Shionoya et al., 2013). Indeed, the respiratory rate pattern respected the

scalar property (Gibbon, 1977). This property refers to the observation that, in interval timing, the variability in the temporal behavior of an animal grows proportionally with the duration of the timed stimulus. This was not observed for the freezing temporal patterns suggesting that respiration is a more reliable index than freezing to assess interval timing in odor fear conditioning. This property might be due to the fact that respiratory rate is a highly fluctuant signal which can be modulated both by the sampling of odorants (Macrides et al., 1982; Youngentob et al., 1987; Kepecs et al., 2007; Wesson et al., 2008) and by the acquired emotional valence of the stimulus (Freeman et al., 1983; Monod et al., 1989; Nsegbe et al., 1997) thus allowing the observation of subtle transient variations in the animal's fear levels.

#### **INFANT ANIMALS SHOW INTERVAL TIMING ABILITIES**

In infants, both the respiration and the freezing rates showed clear temporal patterns respecting the scalar property. To our knowledge, our study is the first to show interval timing in 12–15 days old infant rats. Indeed previous studies devoted to investigate the ontogenesis of temporal learning of seconds-to-minutes intervals in rats were performed on animals aged of 21 days at the beginning of training (Lejeune et al., 1986; Lejeune, 1992). Only slightly younger ages had been investigated on the sub-second range (17 day-old: Stanton et al., 1992). The lack of data in the literature concerning younger ages can be explained by the fact that classical paradigms used to investigate interval timing in animals use peak interval procedures (Catania, 1971) or temporal discrimination tasks (Stubbs, 1968), both of which require numerous conditioning sessions and behavioral responses beyond the infants' motor abilities. In the present study, the use of fine-grain analysis of the temporal patterns of both respiration and freezing permitted us to highlight interval timing abilities occurring after only a few odor-shock pairings. This observation in rats is in line with data collected in human babies showing that they can be conditioned to temporal regularities within a few reinforced trials as early as at 1 month of age (Fitzgerald et al., 1967). In this study, Fitzgerald and colleagues used light/dark regularly spaced switches to evidence that, after about 10 switches, human infants show regular pupillary constriction or dilatation regardless if the switch occurs or not. A recent study using operant discriminative conditioning in 4-months-old babies confirmed a relatively high sensitivity to time at early ages (Provasi et al., 2011).

Interestingly in our study the use of complementary indices such as freezing and respiration permitted us to highlight parameters that revealed timing indices changes throughout development. Indeed, in infants, contrary to adults, freezing appears to be a good index of interval timing. Infant rats respond to the CS odor by a strong decrease in freezing. While it is unclear why freezing was a better measure of timing in pups compared to adults, it may be related to pups' immature freezing response (Hunt and Campbell, 1997).

#### **INVOLVEMENT OF THE DORSAL STRIATUM IN ODOR FEAR CONDITIONING**

The 2-DG autoradiograph revealed that, in adult Paired rats, the medial part of the posterior striatum showed an increased activity compared to Unpaired and Odor animals. A growing literature suggests the involvement of the striatum in interval timing (Allen et al., 1972; Hikosaka et al., 1989; Matell et al., 2003; Höhn et al., 2011). According to McGeorge and Faull (1989), the lateral dorsal striatum receives projections from the sensori-motor cortex, while the medial posterior dorsal striatum receives inputs from the piriform cortex (McGeorge and Faull, 1989) and the amygdala (McDonald, 1991). Therefore, the difference we observe between localizations of the striatum activation might be due to the paradigm and modalities used.

The dorsal striatum has also been suggested to be involved in aversive learning in general such as in auditory fear conditioning (Ferreira et al., 2003; Kishioka et al., 2009; Wendler et al., 2014), or in a two-way active avoidance task (Darvas et al., 2011; Wendler et al., 2014) but not in contextual fear conditioning (Ferreira et al., 2003). Interestingly, both auditory fear conditioning and two-way avoidance present some temporal regularities for the animal to learn, while contextual fear conditioning usually does not. This suggests that the striatum might be preferentially involved in time learning rather than in the learning of the aversion. This is also supported by the results in Unpaired animals of the present study, which showed a similar level of activation in the striatum as the control animals.

In juveniles, our 2-DG metabolic mapping study showed similar levels of activation in the striatum of Paired animals compared to Unpaired and Odor animals. In infants, both morphological (Chronister et al., 1976; Tepper and Trent, 1993) and electrophysiological studies (Tepper et al., 1998) suggest that the striatum is immature. Therefore, at these two developmental ages, the striatum does not seem to be involved in the learning while infant animals in our study and juvenile animals in the literature (Lejeune et al., 1986; Lejeune, 1992; Stanton et al., 1992; Freeman et al., 1993) show clear evidence of learning of interval duration. This raises the question of the neuronal network supporting timing in young animals. It has been demonstrated that neural circuits underlying learning can evolve throughout the development of the organism. The neural substrate of odor-shock associative learning, for example, changes dramatically around PN10. Prior to this age, the association is supported by the olfactory bulb and the anterior piriform cortex (Moriceau et al., 2006). After PN10, the olfactory bulb disengages (Rangel and Leon, 1995), the piriform activation switches from anterior to posterior (Roth and Sullivan, 2005) and the amygdala gets involved in the encoding of the association (Moriceau et al., 2006). Another example of modification of the neural circuit underlying a cognitive process can be found for extinction of conditioned fear. Indeed, while PN24 rats present an adult-like extinction that requires the ventromedial prefrontal cortex (vmPFC), extinction in PN17 rats does not involve the vmPFC (for a review, see Kim and Richardson, 2010). A similar switch throughout the brain maturation could be suggested to underlie time encoding in the brain. While the current model of interval timing encoding proposed in the literature requires a complex communication between the cortex and the striatum (Matell and Meck, 2004), an alternative pathway could be involved in the encoding of interval durations in pups. The olfactory cortex in particular could play a role as it is functional at birth and is known to be involved in odor fear conditioning (Sevelinges et al., 2004, 2008; Jones et al., 2007). Interestingly in a previous study we showed that glutamate and GABA release in the olfactory cortex during odor fear conditioning was correlated to the time of arrival of the CS-US trial suggesting a role for this structure in time encoding (Hegoburu et al., 2009). This is in line with data from the literature showing that sensory cortices are implicated in the processing of temporal information (Quirk et al., 1997; Shuler and Bear, 2006; Bueti et al., 2008). The amygdala could also be involved in interval timing in infants. Indeed a recent study carried out in adult rats reported that changing the CS–US interval during auditory fear memory reactivation induced a selective increase in Zif-268 activity in the lateral nucleus of the amygdala (Díaz-Mataix et al., 2013), and a growing literature suggest that the amygdala may play a role in timing the CS-US interval (Díaz-Mataix et al., 2014). As mentioned above the amygdala is involved in odor fear conditioning from the age of 10 days PN and could thus take part in time processing in infant animals. Finally, alternatively to the striatum, the olfactory tubercle could be considered. Indeed, based on its embryological, anatomical and neuro-chemical properties, this structure is considered as part of the striatum, and has been shown to be functional at early developmental ages (Alheid and Heimer, 1988; Voorn et al., 2004). A recent review by Wesson and Wilson (2011) highlighted the involvement of this structure in both basic and complex olfactory functions and its potentially critical role in interfacing sensory processing and behavioral response. Indeed the olfactory tubercle receives direct olfactory sensory input from the olfactory bulb and piriform cortex (Luskin and Price, 1983) and additional input from the olfactory amygdala (Krettek and Price, 1978; Ubeda-Bañon et al., 2007), thus enabling the association of a given olfactory stimulus with its learned emotional valence. Therefore, although the involvement of the olfactory tubercle in timing has not been investigated yet, this structure could act as a short pathway to link perception with the production of temporally structured actions, specifically at early developmental ages.

In conclusion, the present study shows that in odor fear conditioning, interval durations are learned after only few trials from a very early age. Although the underlying neural network remains to be elucidated, and may evolve with ontogeny, these findings support the hypothesis of a simultaneous encoding of the associative link between two events together with their temporal relationships (Balsam and Gallistel, 2009; Balsam et al., 2010). While further experiments are needed to assess temporal learning at earlier ages than those used in the present study, our data suggest that associative and temporal learning might be two sides of the same coin.

#### **ACKNOWLEDGMENTS**

This work was supported by CNRS-PICS program, Partner University Fund Emotion & Time, LIA CNRS-NYU LearnEmoTime, ANR-Memotime, ANR-TDE, COST-Action TIMELY, NIH-DC009910, NIH-MH091451. This work was performed within the framework of the LABEX CORTEX (ANR-11-LABX-0042) of Université de Lyon, within the program "Investissements d'Avenir" (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR). The authors gratefully acknowledge Ounsa Ben-Hellal for taking care of the animals.

#### **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: 28 March 2014; paper pending published: 07 April 2014; accepted: 25 April 2014; published online: 15 May 2014.*

*Citation: Boulanger Bertolus J, Hegoburu C, Ahers JL, Londen E, Rousselot J, Szyba K, Thévenet M, Sullivan-Wilson TA, Doyère V, Sullivan RM and Mouly A-M (2014) Infant rats can learn time intervals before the maturation of the striatum: evidence from odor fear conditioning. Front. Behav. Neurosci. 8:176. doi: 10.3389/fnbeh. 2014.00176*

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

*Copyright © 2014 Boulanger Bertolus, Hegoburu, Ahers, Londen, Rousselot, Szyba, Thévenet, Sullivan-Wilson, Doyère, Sullivan and Mouly. 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.*

## Rodent ultrasonic vocalizations are bound to active sniffing behavior

### *Yevgeniy B. Sirotin1†, Martín Elias Costa2 and Diego A. Laplagne1,3\*†*

*<sup>1</sup> Shelby White and Leon Levy Center for Brain, Mind and Behavior, The Rockefeller University, New York, NY, USA*

*<sup>2</sup> Integrative Neuroscience Lab, Department of Physics, University of Buenos Aires, Buenos Aires, Argentina*

*<sup>3</sup> Brain Institute, Federal University of Rio Grande do Norte, Natal, Brazil*

#### *Edited by:*

*Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France*

#### *Reviewed by:*

*Markus Wöhr, Philipps-University of Marburg, Germany Stefan Brudzynski, Brock University, Canada*

#### *\*Correspondence:*

*Diego A. Laplagne, Brain Institute, Federal University of Rio Grande do Norte, Av. Nascimento de Castro, 2155, Natal, RN 59056-450, Brazil e-mail: dlaplagne@neuro.ufrn.edu*

*†These authors have contributed equally to this work.*

During rodent active behavior, multiple orofacial sensorimotor behaviors, including sniffing and whisking, display rhythmicity in the theta range (∼5–10 Hz). During specific behaviors, these rhythmic patterns interlock, such that execution of individual motor programs becomes dependent on the state of the others. Here we performed simultaneous recordings of the respiratory cycle and ultrasonic vocalization emission by adult rats and mice in social settings. We used automated analysis to examine the relationship between breathing patterns and vocalization over long time periods. Rat ultrasonic vocalizations (USVs, "50 kHz") were emitted within stretches of active sniffing (5–10 Hz) and were largely absent during periods of passive breathing (1–4 Hz). Because ultrasound was tightly linked to the exhalation phase, the sniffing cycle segmented vocal production into discrete calls and imposed its theta rhythmicity on their timing. In turn, calls briefly prolonged exhalations, causing an immediate drop in sniffing rate. Similar results were obtained in mice. Our results show that ultrasonic vocalizations are an integral part of the rhythmic orofacial behavioral ensemble. This complex behavioral program is thus involved not only in active sensing but also in the temporal structuring of social communication signals. Many other social signals of mammals, including monkey calls and human speech, show structure in the theta range. Our work points to a mechanism for such structuring in rodent ultrasonic vocalizations.

**Keywords: ultrasonic vocalizations, rat, mouse, respiration, speech breathing, theta, rhythm, orofacial**

### **INTRODUCTION**

Many behaviors are organized into repetitive cycles. In active rodents, orofacial sensorimotor behaviors like sniffing, whisking, and head movements are organized into cycles with a characteristic frequency in the theta range ∼5–10 Hz (Welker, 1964; Macrides, 1975; Deschênes et al., 2012). The cyclical nature of these behaviors serves to structure both sensory input and motor output (Ganguly and Kleinfeld, 2004; Kepecs et al., 2006). However, while each behavior can independently display characteristic patterns, they often phase lock to each other (Welker, 1964; Moore et al., 2013; Ranade et al., 2013). This not only yields coordinated patterns of behavior, but also coordinated activity in associated neural circuits (Kay, 2005; Grosmaitre et al., 2007; Cury and Uchida, 2010; Shusterman et al., 2011; Deschênes et al., 2012; Miura et al., 2012; Moore et al., 2013). Indeed, both hippocampal and cortical theta rhythms can transiently phase lock to motor theta rhythms during specific behaviors (Komisaruk, 1970; Macrides et al., 1982; Ganguly and Kleinfeld, 2004; Kay, 2005; Shusterman et al., 2011). Such structuring suggests that our understanding of each individual behavior can benefit from consideration of the broader behavioral context.

The vocal behavior of rats and mice is proposed to feature two mechanisms of sound production. Audible vocal output of fundamental frequency below 20 kHz is produced, as in human speech, when air flowing out through tensed vocal folds causes them to vibrate resulting in sound pressure waves of rich harmonic content (Roberts, 1975a). Vocalization of fundamental frequency in the ultrasonic range (*>*20 kHz) is believed to be produced when air flowing through a small orifice formed by tight vocal folds produces ultrasound of nearly pure single frequencies via an aerodynamic whistle mechanism (Roberts, 1975b; Riede, 2011). Rat ultrasonic vocalization falls in two families with distinct ethological and neurophysiological parallels (Brudzynski, 2009). Aversive settings such as the anticipation of pain or danger can result in prolonged emission of ultrasound in the 20–25 kHz range with little or no frequency modulation, named "22 kHz" ultrasonic vocalizations (USVs). Ultrasound in the ∼30–90 kHz range ("50 kHz USV") is generally emitted by males and females in mating and other social interactions. Emission of 50 kHz USVs has been further linked to expectation of reward and activation of mesolimbic dopaminergic pathways (reviewed in Brudzynski, 2013). In turn, listening to 50 kHz USVs effectively induces approach behavior in both male and female rats, suggesting they may promote social contact (Wöhr and Schwarting, 2007; Seffer et al., 2014; Willadsen et al., 2014). Mice lack a 22 kHzlike alarm vocalization, and emit brief USVs in the ∼50–100 kHz range, mostly studied in the context of mating (Holy and Guo, 2005). Vocalizations are usually segmented by experimenters into individual packets ("calls" or "syllables") based on silences and/or spectral discontinuities (Liu et al., 2003; Wright et al., 2010). Interestingly, when segmenting by silences of 40 ms and over, adult rat and mouse calls are found to come in bouts with instantaneous rates in the theta range (Liu et al., 2003; Kim and Bao, 2009).

Vocal output depends critically on air flowing through the larynx, which is temporally structured by the breathing cycle (Roberts, 1975a). As in birds and humans, ultrasonic vocalizations in rats have been shown to be associated with increased subglottal pressure, indicating a phasic relationship with the breathing cycle (Roberts, 1972; Hegoburu et al., 2011; Riede, 2011, 2013). Highly vocal animals like humans and birds developed exquisite control mechanisms that coordinate breathing with activity in muscles used for vocalization in order to produce complex vocal output (MacLarnon and Hewitt, 1999; Andalman et al., 2011). As previously shown by us and others, rats show this control to some degree as they are able to maintain exhalations of over 2 s during the emission of prolonged 22 kHz alarm calls (Hegoburu et al., 2011; Assini et al., 2013). Rat breathing patterns are additionally constrained by sniffing, which is an active breathing behavior used to sample the olfactory environment (Welker, 1964; Wachowiak, 2011). Breathing patterns associated with normal respiration can be distinguished from active sniffing based on their frequency. Normal respiration in adult rats is typically below 3 Hz whereas active sniffing is typically in the theta range (Welker, 1964; Hegoburu et al., 2011; Wachowiak, 2011). However, despite clear dependence of vocalizations on breathing, the interplay between 50 kHz USVs and respiratory dynamics has not been previously investigated.

Here we examined, in detail, the relationship between respiration and ultrasonic vocal output of rats in a social environment. We find that ultrasonic vocalization of the 50 kHz family is largely restricted to periods of active sniffing (5–10 Hz). Within each sniff, both the initiation and cessation of vocal output was precisely linked to specific phases of the sniff, initiating just after the end of the inhalation and finishing just prior to the peak of the exhalation. As a result, the sniff cycle segments ultrasound production into individual calls, which inherit its theta rhythmicity. In turn, vocal output deforms ongoing sniff rhythms, briefly stretching the exhalation period as necessary to accommodate the full duration of the produced vocalization.

Our results show that orofacial behaviors with theta rhythmicity are not only involved in active sampling but also temporally structure outgoing communication signals at this rate. Moreover, we show that the sniffing and ultrasound production systems in rodents are linked on a millisecond scale, suggesting a tight coupling between the neural centers controlling sniffing and vocalizations.

#### **MATERIALS AND METHODS**

#### **ANIMAL SUBJECTS**

All procedures were approved by The Rockefeller University Institutional Animal Care and Use Committee. Simultaneous recording of ultrasonic vocalizations and intranasal pressure were carried out on 5 Long Evans adult male rats (Charles River, ages 3–8 months, single housed from 2 months of age), and 2 CBA/CaJ adult male mice (Jackson Labs, ages 10–11 weeks, pair housed). Male mice were recorded in the presence of an adult female C57 mouse. Rats were held on an inverted light cycle and all recordings were carried out during the dark phase under infrared illumination.

#### **RECORDING SESSIONS**

Rats were placed in a custom built social arena in a single-walled soundproof room. The purpose of this setup was to promote vocal production from social interaction while still being able to unequivocally assign each call to the rat it originated from. The arena (see **Figure 1A**) was split in two halves, 46 × 33 × 74 cm (W × L × H) each, 25 cm apart on the wide side. Walls were made of thin vertical bars and surrounded by 5 cm thick wedged foam to minimize echoes. The separation between halves was packed with foam from 20 cm above the floor to the top to minimize cross-talk between microphones (see below). The acrylic floor was covered with Aspen Chips bedding (NEPCO, Warrensburg, NY, USA), chosen to minimize locomotion related noise (the same bedding was used in the home cages). One rat was placed on each side of the arena where they could hear and smell each other for sessions lasting up to 2 h. Male-female mice pairs were recorded together in a 20 × 40 × 30 cm (W × L × H) acrylic box with Aspen Chips bedding. The respiration of the female mouse was not monitored. Intranasal pressure and ultrasound signals were simultaneously digitized by a data acquisition board at 250 kHz sampling frequency (PCIe-6259 DAQ with BNC-2110 connector, National Instruments). Animals were monitored from outside the room through video under infrared illumination.

## **ULTRASONIC VOCALIZATIONS**

#### *Recording and detection*

One condenser microphone with nearly flat (±5 dB) response from 10 to 150 kHz (CM16/CMPA-5V, Avisoft Bioacustics) was positioned above each rat at a height of 72 cm to selectively pick up calls from the rat beneath (**Figure 1A**). All USV analysis was performed on the raw sound recordings with custom built MATLAB routines (The Mathworks). To efficiently handle the large recorded datasets, we developed automated techniques for detecting ultrasound emissions and assigning them to the rat of origin (Figure S1). The performance of our detection and assignment methodology was assessed in an independent set of recordings (see below). We first obtain the sonogram for each microphone (Figure S1A, 2 ms time window, 0.25 ms time step, 1 kHz bandwidth, 3 tapers; http://chronux*.*org/; Mitra and Bokil, 2007). Each time step of the spectrogram constitutes a vector *P* where each point is the power at a given frequency (18–100 kHz). We next normalize this vector by its sum (to ensure all values span between 0 and 1) and calculate the entropy of this normalized vector *Pn* as *H* = −*Pn* · *log*2*Pn*. For rodent vocalizations, sound power is concentrated at a single frequency, reducing the entropy, while unwanted noise is typically broadband and thus of high entropy (Figure S1A). Segments lasting at least 3 ms with entropy below a fixed threshold of 6.5 bits and bounded by silences of *>*20 ms are selected as putative USVs. These are then curated by automatically discarding as noise those with high power in the sonic range (5–18 kHz) and visually inspecting those

**FIGURE 1 | Simultaneous recording of respiration and ultrasonic vocalization. (A)** Left: Schematic of the recording arena as viewed from the top (top) and side (bottom). The position of the ultrasonic microphones (red) and video cameras (green) is shown. Tubing (gray) connects the nasal cannulae with pressure sensors. Right: Snapshot of rats simultaneously behaving in the arena. **(B)** Segment of intranasal pressure (black) recorded from a rat in a social setting. Red bars: periods of ultrasonic vocal output detected for this rat. Scale bar: 2 s.

From here on, inhalations are plotted as positive deflections of the pressure trace. **(C)** Detailed view of respiration (bottom) and ultrasonic vocalizations (top; sonogram). From here on, black arrowheads denote zero relative intranasal pressure. Scale bar: 250 ms. **(D)** Autocorrelations of respiration (black) and ultrasonic vocalizations (red) from a 10 min recording segment. Note signals show similar periodicity, with first peaks at 130 and 150 ms respectively (eq. 7.7 and 6.7 Hz). **(A–C)** same data set.

with intermediate levels of ultrasonic entropy and sonic power. In a dataset of 31 recording sessions we estimated 94% of emitted USVs (47866 of an estimated total of 51095) were effectively detected in this way (Figure S1B).

Detected USVs are assigned to the emitting rat by comparing the signals from both microphones. When ultrasound is detected (crosses the entropy threshold) at only one microphone, the USV is assigned to the rat on the same side of the arena. If the same USV is detected at both microphones, it is assigned to the rat under the microphone with lowest entropy (examples in Figure S1A). To assess the accuracy of the USV assignment we analyzed 11 recording sessions with just one rat in the arena. 77% of calls (20653 of 26815) were detected only by the microphone on the rat's side (Figure S1C). Of those detected in both, the entropy difference was large enough to unambiguously assign them to the correct side of the arena (Figure S1D). Overall, 99.8 ± 0.1% of USVs were properly assigned at each session. In the special case of two rats vocalizing at the same time, they will typically produce USVs with different fundamental frequency profiles at each microphone. When these profiles are found to differ by *>*1 kHz during *>*3 ms we deduce both rats vocalized simultaneously and assign to each the USV detected by the microphone on its side (Figure S1E).

Mice USVs were recorded from a single condenser microphone positioned 30 cm above the floor and detected in a similar fashion. As justified in section Structuring of Mouse Ultrasonic Vocalizations by Sniffing, all calls were assigned to the male mouse.

#### *Analysis*

"Vocal ratio" was defined as the fraction of time (0–1) spent producing ultrasound in a window of 3 s. This measurement is independent of any segmentation of vocal production. A "call" was defined as the ultrasound emitted within an individual sniff. "Call rate" as the number of detected calls per second in a 3 s window. "Instant call rate" was calculated for calls occurring on consecutive sniffs as the reciprocal of the time between the onsets of the two calls (**Figure 6D**).

### **SNIFFING**

#### *Cannula implantation*

To monitor respiration, the end of a thin 2-cm-long stainless cannula (gage 22) was implanted in the nasal cavity. The cannula was bent to an S-shape so as to end above the temporal bone. Animals were anesthetized using isoflurane gas anesthesia. A skin incision was made exposing the frontal bone and most of the nasal bone. A small hole was drilled in either the left or the right nasal bone, into which the tip of the cannula was inserted from above so as to protrude into the nasal cavity. The cannula was affixed to the hole with a small drop of cyanoacrylate glue (All-purpose Krazy Glue), and stabilized on the skull with methyl methacrylate dental cement around skull screws. Animals were given at least 2 days after a surgery for recovery.

#### *Data acquisition and pre-processing*

During experiments, the cannula was connected to a pressure sensor located above the arena (24PCAFA6G, Honeywell; modified to reduce internal air volume) with ∼100 cm of Teflon tubing (AWG# 22 STD, Pennsylvania Fluorocarbon) via a plastic fluid swivel (375/22PS, Instech). The output of the pressure sensor bridge was coupled to an instrumentation amplifier (AD620, Analog Devices) for recording. For analysis, signals were downsampled to 1 kHz Inhalations caused an inward flow of air through the nose that resulted in a decrease in measured pressure whereas exhalations caused an outward flow of air through the nose resulting in an increase in the measured pressure signal. Throughout the figures, inhalations are shown as upward deflections and zero denotes atmospheric pressure.

The tubing connecting the cannula to the pressure sensor filters down fast fluctuations and imposes a time delay to the pressure signal. To measure this distortion we generated broadband pressure signals with an electrodynamic transducer (ET-132-203; Labworks Inc.) driven by a linear power amplifier (PA119; Labworks Inc.). We then recorded the same signal with our pressure sensor directly at the output of the transducer and after distortion by the tubing (Figure S2A). We used these two signals to calculate the transfer function of the tubing through Fourier deconvolution (http:// terpconnect.umd.edu/∼toh/spectrum/Deconvolution.html) and used this transfer function to reconstruct the undistorted intranasal pressure signal in all recordings (see Figure S2 for validation).

#### *Analysis*

To identify individual respiratory cycles ("sniffs"), we developed MATLAB routines to segment the recorded pressure traces as follows. Slow drifts in sensor output were removed (400 Hz low pass Butterworth filter). Signals were then mean subtracted and divided by their standard deviation. Sniff cycles were defined to start at the inhalation onset and end at the exhalation offset (onset of the next inhalation). Inhalation onsets were detected as positive slope crossings of a fixed threshold. The end of each inhalation was defined as the negative slope crossing of the same threshold. Sniffs with aberrant inhalation durations (*<*20 ms) were rejected from subsequent analyses.

The phase within the sniffing cycle was computed using a previously described algorithm (Shusterman et al., 2011). Briefly, we determined three points in time for each cycle: inhalation onset, inhalation offset (exhalation onset), and exhalation offset, as described above. We then morphed each sniff cycle so that the duration of its inhalation and exhalation matched the average durations across all recorded sniffs. Phase within the sniff was then defined as the normalized time (0–1) within the morphed sniff (see Figures 1A,B in Shusterman et al., 2011).

The instant rate of a sniff cycle was defined as the reciprocal of the time between the start of its inhalation and that of the next cycle. "Ongoing sniff rate" is calculated as the mean instant rate in 3 s windows. Only silent sniffs were included to specifically quantify the respiratory rhythm without direct effects from USVs (see **Figures 6A,D**).

#### **BOUT ANALYSIS**

For the analysis of call bouts, a binary vector was constructed for each recording session. Each vector element corresponded to a single sniff and was assigned 1 if the sniff was vocal and 0 if the sniff was silent. A call bout was defined as a stretch of calls occurring over consecutive sniff cycles (a stretch of ones in the vector). Distributions of bout lengths were obtained by pooling across sessions for each rat. Two random models were used to generate surrogate binary vectors. First, we constructed a constant probability model, where a single call probability was used for each vector element (i.e., sniff). Each sniff was randomly assigned a call with a fixed probability obtained by dividing the total number of calls over the total number of sniffs. For the variable probability model, we simulated the effect of a varying call production rate within a session. The probability of assigning a call to each surrogate element was obtained from the measured data as follows. We convolved the observed binary vector with a Gaussian kernel to estimate an underlying local call production probability. In this analysis, "rate estimation window" corresponds to the full width at half maximum of this kernel (measured in number of sniffs). To capture potential call probability fluctuations at different time scales, we generated surrogate datasets with models of different rate estimation window from 4 to 256 sniffs. For each session and model, we generated 1000 pseudorandom surrogate vectors, calculating the distribution of bout lengths for each. For each session, we calculated the log likelihood of observing a given bout length in the real vs. surrogate data as log10 of the ratio between the probability of observing a bout of a given length in the real data and that of the surrogates. For example, a value of 1 is obtained if a given bout length is 10 times more likely in the real data.

### **STATISTICAL ANALYSIS**

Relationships showing apparent linearity were analyzed with linear regression (**Figures 3B**, **6E,F**, **7B**). Others with repeated measures ANOVA (**Figures 2C**, **6B,C**).

### **RESULTS**

To examine the relationship between respiration dynamics and ultrasonic vocal output of rats, we developed a split social arena. In the arena, adult male rats separated by a wire divider could hear and smell each other in the dark (**Figure 1A**). Analysis of audio from a pair of overhead microphones allowed us to unequivocally assign vocalizations to each rat. To monitor respiration, we implanted the rats with intranasal cannulae coupled to pressure sensors (see Materials and Methods). We recorded respiration and vocalizations for extended periods of time (30–120 min) at high sampling frequency (250 kHz), which allowed us to examine

**FIGURE 2 | Ultrasonic vocalization occurs during periods of fast sniffing. (A)** Top: Spectrogram of a section of the recorded respiration. Warmer tones denote higher power (AU). Note the alternation between periods of fast (∼7 Hz) and slow (∼2 Hz) respiration. Bottom: simultaneous vocal production from this rat quantified as fraction of time spent vocalizing within a 3 s sliding window (vocal ratio). Blue shading: periods of silence (vocal ratio = 0). Red shading: high vocal production (vocal ratio *>* 0.025). Top right: mean frequency spectrum of respiration for periods of high vocal production (red; peak = 6.8 Hz) and silence (blue; peak = 2.2 Hz) in the example. **(B)** Distribution of sniff rates during periods of high vocal production (red) or silence (blue). Mean ± s.e.m., *N* = 5 rats. **(C)** Vocal ratio as a function of sniff rate. To account for varying average vocal output of individual rats, curves were normalized by their maximum prior to averaging. Effect of sniff rate on mean vocal ratio: *p <* 0*.*0001 (ANOVA, *N* = 5 rats). **(D)** Autocorrelations of vocal ratio (red) and sniff rate (black), averaged in 3 s intervals.

relationships between these behaviors across multiple timescales (**Figure 1**). Rats showed large variations in the rate of respiration and ultrasonic vocalization (**Figure 1B**). Under these conditions, all vocal output was restricted to USVs of the 50 kHz family (**Figure 1C**). As expected, intranasal pressure traces showed strong periodicity in the theta range imposed by the inhalationexhalation cycle. Interestingly, vocal output was also periodic at theta (**Figure 1D**).

### **RATS PRODUCE ULTRASOUND DURING FAST SNIFFING**

Respiration rate in awake rats varies with behavioral state over a wide range (1–10 Hz) (Wachowiak, 2011). In our recordings, rats also alternated between periods of silence and high vocal production (**Figure 2A**). Visual inspection of respiration and vocalization records suggested that rats vocalized mostly during periods of active sniffing (e.g., **Figure 1B**). To quantify this relationship, we computed "vocal ratio" as the fraction of time spent producing ultrasound in a sliding window of 3 s (**Figure 2A** bottom; Methods). We calculated average ongoing sniff rate in this same window by segmenting the continuous intranasal pressure traces into individual sniff cycles (sniffs) and computing their average instantaneous rate (Methods). To avoid possible interactions between ultrasound production and sniffing, we excluded sniff cycles associated with vocal production from the calculation of sniff rate. During silent periods (vocal ratio = 0), rats were either breathing passively (rate *<* 4 Hz) or actively sniffing (rate *>* 5 Hz), spending similar periods of time in each mode. In contrast, periods of high vocal output (vocal ratio *>* 0.025) were exclusively associated with active sniffing (**Figure 2B**). Overall, this results in a strong positive correlation between vocal production and ongoing sniff rate with maximal vocal output during periods of 8 Hz sniffing (**Figure 2C**). Changes in vocal ratio were, however, faster than those of respiratory rate (**Figure 2D**), reflecting that brief periods of high vocal production occurred within longer periods of fast sniffing (e.g., **Figure 2A**).

### **ULTRASOUND PRODUCTION PROLONGS THE SNIFF CYCLE**

Mammalian vocalization usually prolongs the respiratory cycle (Smotherman et al., 2010). We analyzed whether this is also the case for the brief rat vocalizations of the 50 kHz family. During silent respiration, recorded intranasal pressure typically followed a sinusoidal pattern, indicating roughly equal time spent inhaling and exhaling (e.g., **Figure 1**, blue trace in **Figure 3A**). Of our full population of recorded sniffs (*N* = 256991 sniffs in 5 rats), vocal sniffs accounted for 15 percent (*N* = 37593). Despite our observation that ultrasound is produced during periods of high ongoing sniff rate, vocal sniffs were on average longer than silent sniffs (163 ± 64 vs. 131 ± 55 ms; median ± inter-quartile-range; *p* ∼= 0, Wilcoxon rank sum test for equal medians). Within each vocal sniff, we quantified the total duration of ultrasound production as the difference between the first and last time-point having ultrasound. We found that overall sniff length increased with ultrasound duration (**Figure 3A**). Specifically, it was exhalation durations that increased, while inhalations remained largely unaltered (**Figure 3B**). Exhalations grew with ultrasound duration with a mean linear slope of 0.85 (**Figure 3C**). As a consequence, the emission of ultrasound during a given sniff cycle

was accompanied by an instantaneous drop in the sniffing rate (**Figure 3D**).

#### **ULTRASONIC VOCALIZATION OCCURS AT SPECIFIC PHASES OF THE SNIFF CYCLE**

We next examined the detailed temporal alignment between ultrasound production and the inhalation-exhalation cycle. Prior work established that ultrasound is produced during exhalations, corresponding to periods of high subglottal pressure (Riede, 2011). Interestingly, during production of ultrasound, relative intranasal pressure remained close to zero, indicating reduced airflow through the nose (**Figure 4A**). This relationship held up to the millisecond timescale as brief drops in the power of the emitted ultrasound co-occurred with sharp peaks in nasal flow (Figure S3). We examined the coupling of ultrasound production to inhalations and exhalations by warping each sniff to a common phase axis aligning inhalation onsets, inhalation-exhalation transitions, and exhalation offsets (Methods). The average vocal sniff had a distinctly different shape than a silent sniff, with a pronounced deviation from a sinusoid after inhalation corresponding to the period of low airflow through the nose (**Figure 4B**, top). Indeed these shape differences were so pronounced that sniff shape alone was often an excellent predictor of the presence

mark exhalation onset. Note ultrasound is produced during low-pressure region following exhalation onset. **(B)** Top: mean sniff waveforms from silent (blue) or vocal (red) sniffs for one example rat. All waveforms were warped to align at three points: onsets of inhalation and exhalation and the end of exhalation. Bottom: distribution of ultrasound onset (black) and offset (gray) phases in the vocal sniffs. Inhalation onset: phase = 0, exhalation onset: phase = 0.33, end of exhalation: phase = 1. Gray line: exhalation onset. Time between most frequent vocalization onset and offset marked in pink. **(C)** Distribution of ultrasound onset (black) and offset (gray) phases for all rats. Boxes: median and 25–75th percentiles. Whiskers: 10–90th percentiles.

of vocalization (Figure S4). For all vocal sniffs, ultrasound production onsets and offsets were tightly coupled to sniff phase. Ultrasound production began shortly after the end of inhalation and ended prior to the peak of exhalation (**Figure 4B**, bottom). This tight coupling was observed in each of our tested animals (**Figure 4C**).

#### **THE SNIFF CYCLE NATURALLY SEGMENTS EMITTED ULTRASOUND INTO CALLS**

Ultrasound appears to be emitted in brief units separated by silences, usually named "calls" or "syllables." A clear rationale for this segmentation is, however, missing. It is clear from our data that rats are silent during inhalations. To understand how this structures the emission of ultrasound in time, we quantified the distribution of silence durations and its relation to the sniff cycle. We defined silences as intervals longer than 2 ms with no detectable vocal output. The analysis revealed identical multimodal distributions for all rats (**Figure 5A**). Silences were either shorter than 20 ms (58 ± 3%) or longer than 60 ms (41 ± 3%). Short silences occurred between ultrasound emissions within a single sniff cycle whereas long silences included at least one inhalation and thus separated emissions across sniffs (**Figure 5B**). In consequence, segmenting calls by a minimum silence of 20–60 ms is equivalent to segmenting by sniff cycle as all calls are moored to a single sniff and each sniff harbors at most one call (**Figure 5C**). The sniff cycle thus provides a natural segmentation of ultrasound production into individual calls.

#### **ONGOING SNIFF RATE MODULATES CALL DYNAMICS**

Studies on USVs typically correlate measurements like call rate and duration with experimental conditions. Having now defined a "call," we analyzed to what extent their properties depend on the ongoing respiratory rate, assessed in neighboring silent sniffs (**Figure 6A**). As expected from our previous results, ongoing sniff rate strongly influenced measured call rates, which were maximal when sniffing at theta frequency (**Figure 6B**). The probability of emitting a call on each sniff also peaked during theta sniffing

**FIGURE 5 | Sniff cycles segment ultrasonic vocalization into calls. (A)** Distribution of silence durations (Mean ± s.e.m.; *N* = 5 rats). Inset: detail of short silences. **(B)** Example of ultrasonic vocalizations (top) and simultaneous sniffing (bottom). Gray and black bars mark the occurrence of long (*>*40 ms) and short (*<*40 ms) silences. Note short silences are contained within single exhalations while long ones span more than one sniff cycle. Scale bar: 100 ms. **(C)** Segmentation of calls as a function of silence duration threshold. Orange: Percentage of segmented calls that do not share a sniff cycle with other calls. Green: Percentage of calls that do not span more than one sniff cycle. The gray area shows the range of silence duration thresholds that effectively segment over 95% of calls by sniff cycles (20–80 ms).

demonstrating that increased call rates were not trivially due to having more sniffs per unit time (**Figure 6C**).

So far we showed that sniff frequency strongly alters the quantity of calls produced. Does sniffing also alter the detailed dynamics of call production (**Figure 6D**)? We found that calls had a characteristic duration that was largely independent of sniff rate up to 8 Hz sniffing. However, for faster rates mean duration dropped by 25%, highlighting an interaction between the ongoing sniffing behavior and the vocal motor plan (**Figure 6E**).

We studied call rates in finer temporal detail by measuring the instant rate between calls occurring in consecutive sniffs (**Figure 6D**). As previously observed (Kim and Bao, 2009), rat calls have a characteristic instant rate of ∼6 Hz (**Figure 6F**, inset). If this was a fixed property of USV emission mechanisms, instant call rate should be largely independent of ongoing respiratory rates. On the contrary, it was positively correlated to the rate of the immediately preceding silent sniff (**Figure 6F**). Thus, instant call rates carry information about ongoing sniffing frequency. This interaction is bidirectional, as calling immediately affects respiratory rate, bringing it to a narrower range centered at 6 Hz (**Figure 6F**).

#### **STRUCTURING OF MOUSE ULTRASONIC VOCALIZATIONS BY SNIFFING**

We next extended our analysis to the ultrasonic vocalizations of the laboratory mouse (*Mus musculus*). We simultaneously recorded vocal output with intranasal pressure in male CBA/CaJ adults (*N* = 2) during encounters with a female. Previous studies have concluded females rarely, if ever, emit USVs during mating so detected ultrasonic calls can be assigned to the male partner (White et al., 1998). Indeed, all calls detected from our recordings matched the breathing pattern of the male (**Figure 7A**). The sniff cycles of mice differed from that of rats in that even for silent sniffs, inhalations were followed by a brief period of constant low relative intranasal pressure before going into full exhalation (**Figure 7A**), whereas in the rat this pattern was strongly indicative of USVs (see **Figures 1C**, **3A**, **4A,B** and Figure S2). As in the rats, the emission of USVs significantly prolonged the sniff cycle, with a positive correlation between exhalation duration and the duration of USV (**Figure 7B**). The slope of this relationship was less pronounced (compare **Figures 7B**, **3A,B**). Nonetheless, the locking of the ultrasound production to the phase of the sniff cycle was comparable to that found for rats, with USVs starting after the end of the inhalation and ending prior to the peak of the exhalation (**Figure 7C**).

The temporal properties of ultrasonic calls in the mouse were qualitatively similar to the rat. Silence durations of at least 40–60 ms segmented ultrasonic output into calls (mean duration = 46 ms) occurring within a single sniff (**Figure 7D**). The distribution of instantaneous rates of calls produced on consecutive sniffs peaked at 6.5 Hz whereas instantaneous rates of silent sniffs peaked at 8 Hz (**Figure 7E**). This shift is a direct result of prolongation of exhalations by calls, as also observed for the rats.

#### **CALL BOUTS ARE DIFFERENT IN RATS AND MICE**

While rodent USVs appear to cluster in time (Nyby and Whitney, 1978; Brudzynski and Pniak, 2002), it is not clear whether the call "bout" is a fundamental unit of their vocal production.

defined as the mean instant rate (1/sniff duration) of all sniffs with no USV (black sniffs) in the same window. **(B)** Call rate vs. ongoing sniff rate. Effect of sniff rate *p <* 0*.*0001, repeated measures ANOVA, *N* = 5 rats. **(C)** Percentage of sniffs accompanied by calls vs. ongoing sniff rate. Effect of sniff rate

sniffs ("t" in figure). Instant sniff rate is that of the immediately preceding silent sniff. **(E)** Call duration vs. instant sniff rate. Red: linear regression; *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*12, *<sup>p</sup> <sup>&</sup>lt;* <sup>0</sup>*.*001. **(F)** Instant call rate vs. instant sniff rate. *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*67, *p <* 0*.*0001. Inset: distribution of instant call rates.

Alternatively, calls could appear to be grouped in time simply because of continuous fluctuations in call rate (Nawrot, 2010). We took advantage of the natural segmentation provided by the sniff cycle to explore this in rats and mice. We defined a bout as a series of calls emitted on consecutive sniffs and asked whether their occurrence was a statistically significant event. At first glance, no strong tendency for emitting bouts was observed for rats, as the distribution of bout lengths decayed monotonically with 72 ± 4% (*N* = 5 rats) composed of a single call and only 2.5 ± 0.7% containing 5 or more calls (**Figure 8A**). To test for structure in the vocal production we compared this distribution with a random model where rats have a constant probability of emitting a call on each sniff given by their mean call rate (see Materials and Methods). Bouts of 3 or more calls occurred more frequently than chance, while isolated calls were in fact less probable (**Figure 8A**). However, when comparing with a family of random models that account for call rate variations, the grouping of calls into bouts matched models where calls are randomly emitted with a probability fluctuating with a temporal resolution of 1–2 s (**Figure 8A**, inset). This analysis suggests that call bouts defined in this way are not a fundamental feature of rat vocal production but rather reflect fast modulations in their behavioral state. Mouse calls were emitted in strikingly longer bouts than for those of rats, with only ∼45% of them composed of a single call and ∼20% containing 5 calls or more (**Figure 8B**). This high structuring could not be accounted for by random models with slow call rate fluctuations (**Figure 8B**, inset), suggesting mice USVs are indeed preferentially grouped into bouts.

### **DISCUSSION**

By examining long periods of simultaneously recorded respiration and ultrasonic vocalization patterns we found a profound relationship between these two behaviors across timescales. Overall, vocal production is largely restricted to periods of active sniffing. During these periods, both sniffs and calls are periodic at theta frequencies (6–8 Hz). USVs are not, however, a byproduct of olfactory behavior as rats can sniff fast without vocalizing. Calls are produced exclusively during exhalations and prolong sniffs causing an instantaneous reduction in sniff rate. Most calls are, however, brief, producing only a modest drop in sniff rate from 8 to 6 Hz. In this way, the rate of ongoing sniffing effectively imparts its theta rhythmicity onto calls.

Though it is commonplace in the field to talk about rodent "calls," a proper delineation of the term is missing. Segmenting a stream of vocal output into meaningful units is an important first step in any semantic or syntactic study. The working hypothesis behind defining animal "calls" is that there are a finite number of distinct motor plans for the production of vocalizations which could differentially correlate with the emitter's physiological or behavioral state and the receiver's responses. Segmentation of the produced sound by this underlying structure results in a more compact description of the vocal repertoire and aids in the

**FIGURE 7 | Structuring of mouse ultrasonic vocalizations by sniffing. (A)** Detailed view of ultrasonic vocalizations (top; sonogram) and respiration (bottom) for a mouse. Scale bar: 250 ms. Compare to **Figure 1C**. **(B)** Top: average waveforms for silent sniff cycles (blue) or cycles simultaneous to the emission of ultrasonic vocalizations of increasing duration (reds; vocal sniffs) for a mouse. Data was binned by ultrasound duration (mean durations: 14, 42, and 81 ms). Traces are aligned to inhalation onset (dotted line). Compare with **Figure 3A**. Bottom: Inhalation (gray) and exhalation (black) durations for individual vocal sniff cycles vs. vocalization duration across mice. Lines: linear regressions; Slope = 0.32 (exh) and 0.02 (inh), *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*08 (exh) and 0.01 (inh). Compare with **Figure 3B**. **(C)** Top: mean sniff waveforms from silent (blue) or vocal (red) sniffs across mice. All waveforms were warped to align at three

points: onsets of inhalation and exhalation and the end of exhalation. Bottom: Distribution of ultrasound onset (black) and offset (gray) phases in the vocal sniffs. Inhalation onset: phase = 0, exhalation onset: phase = 0.2, end of exhalation: phase = 1. Gray line: exhalation onset. Time between most frequent vocalization onset and offset marked in pink. Compare with **Figure 4B**. **(D)** Segmentation of calls as a function of silence duration threshold in mice. Orange: percentage of segmented calls that do not share a sniff cycle with other calls. Green: percentage of calls that do not span more than one sniff cycle. The gray area shows the range of silence duration thresholds that effectively segment over 95% of calls by sniff cycles (40–60 ms). Compare with **Figure 5C**. **(E)** Blue: distribution of silent sniff rates. Red: distribution of instant call rates for calls made on consecutive sniffs.

understanding of vocal communication systems. Animal vocalization is usually broken up in calls at spectrotemporal discontinuities, but the choice of parameters is not trivial. We propose a physiologically grounded segmentation strategy such that a call is defined as the ultrasound emitted during a single exhalation. We further show this rule can be accurately implemented without recording respiration by choosing a silence duration threshold between 20 and 60 ms for rats and 40–60 ms for mice. Of those studies where the segmentation method is reported, some used silence durations within or close to these ranges (Liu et al., 2003; Holy and Guo, 2005; Wright et al., 2010) while others used thresholds too short to match the sniffing structure (Sewell, 1970; Takahashi et al., 2010).

Welker's detailed examination of rat behavior demonstrated the phasic relationship between sniffing, whisking, and head movement. During active periods, these behaviors are produced in cycles coherent at theta frequency (Welker, 1964; Deschênes et al., 2012; Moore et al., 2013; Ranade et al., 2013). This shared oscillatory patterning has been proposed to be relevant for information exchange between brain areas (Kay, 2005; Kepecs et al., 2006). Our results add the emission of ultrasonic vocalizations to the family of orofacial behaviors with theta rhythmicity observed in rodents (**Figure 9**). As such, the detailed properties of USVs are not independent but bounded by this rhythmic frame. Any research into the neural or broader behavioral correlates of any such motor behaviors would thus benefit from considering the broad context of the others to identify any individual contributions and interactions (Assini et al., 2013; Moore et al., 2014). During ultrasound production, motoneurons in the nucleus ambiguus control larynx geometry via activation of specific muscles (Yajima and Hayashi, 1983; Riede, 2011). The observed phase locking of vocalizations with the sniff cycle suggests a precise coordination between activity in this motoneuron pool and the brainstem nuclei responsible for orchestrating the respiratory rhythm (Moore et al., 2014). The mechanistic links posited by our observations should be confirmed by experimental manipulation of activity in these nuclei, as is being done for dissecting the interactions between the sniffing and whisking rhythms (Moore et al., 2013). Our results show that constriction of the larynx associated with ultrasound production is associated with a delay in the onset of the following respiratory cycle, similar to that observed for swallowing (McFarland and Lund, 1993). USVs are natural and frequent perturbations of the sniffing cycle. Understanding how they affect (and are affected by) the instantaneous phase of other orofacial rhythms like whisking and head movements could aid in understanding the hierarchical organization of their associated motor nuclei. Of particular interest is the coordination of ultrasonic vocalization with active whisking, as it is likely that both are simultaneously acting as rhythmic communication signals during close distance social interactions (Wolfe et al., 2011).

The rate of respiration is strongly correlated with the behavioral state of the animal (Welker, 1964; Hegoburu et al., 2011). We show that calls carry detailed information about sniff dynamics at both slow and fast timescales. At slow scales, the co-occurrence of high rates of 50 kHz USVs and fast sniffing could reflect their common drive by the ascending dopaminergic system (Costall and Naylor, 1975; Brudzynski, 2007). Given this link, 50 kHz

**FIGURE 8 | Call bouts are different in rats and mice. (A)** Probability of observing rat call bouts of a given length (i.e., the number of consecutive sniffs with calls). Blue: real measured data. Red: surrogate data constructed assuming constant vocalization rate (see Materials and Methods). Inset: Comparison of measured bout length probabilities to a family of surrogate models with varying rate estimation windows (4–256 sniffs; x-axis; see Materials and Methods). Y-axis: log likelihood ratio between measured and surrogate bout length probabilities (for bout length 1–5). Positive values indicate that bouts of a given length are more likely in real vs. surrogate data. Red arrowhead: surrogate model with a rate estimation window of width 12 sniffs matches real data for all bout lengths (log-likelihood ∼= 0). Panels show mean ± s.e.m.; *N* = 5 rats. **(B)** Same analysis as in A for mice. Note lower probability of bout length = 1 for mice (46%) than for rats (72%). Surrogate data with a 4-sniff rate estimation window approximates observed bout distribution in mice, compared with 12-sniff window for rats.

USVs could preferentially promote social contact in individuals in positively aroused, exploratory states. At faster time scales, calls group together in time resulting in bouts where calls are emitted in consecutive sniffs. We found that the statistics of rat call bouts do not support their status as a fundamental unit of vocal production, but rather appear secondary to changes in the drive to produce calls on the timescale of 1–2 s. In contrast, mouse calls are organized into longer bouts that cannot be accounted for by slow rate fluctuations, in agreement with a proposed song-like production (Holy and Guo, 2005). Call instant rates within bouts are centered on theta, with their precise value closely reflecting the underlying sniffing rate. Thus, the instantaneous call rate could transmit detailed information about the ongoing sniffing rate of the emitter, which is intimately linked with behavioral state. Interestingly, sounds presented at these rates are privileged in their processing by the auditory system of rats. During development, the auditory cortex selectively enhances the representation of sounds presented within theta band ∼7 Hz (Kim and Bao, 2009), suggesting that theta patterning is important for the learning of species specific vocalizations. In adults, auditory responses to sounds are heavily attenuated at presentation rates above 10 Hz (Kilgard and Merzenich, 1998), which corresponds to the upper limit of our observed distribution of instantaneous call rates. Thus, the auditory system of rodents may be preferentially tuned to the sniff-driven dynamics of conspecific vocalizations.

Other mammalian orofacial communication signals are temporally structured at theta frequencies, such as marmoset twitter calls (Wang et al., 1995), macaque lip-smacking (Ghazanfar et al., 2010) and human speech (Chandrasekaran et al., 2009). Specific disruption of this rhythmicity results in impaired intelligibility (Saberi and Perrott, 1999; Ghitza and Greenberg, 2009; Ghazanfar et al., 2013) and cortical oscillations at matching frequencies are proposed to play a role in their selective perception (Giraud and Poeppel, 2012). Whether theta rhythms in primate and rodent social signals are evolutionarily linked and whether emission and perception of all of them are linked to cortical theta oscillations remains unknown.

#### **ACKNOWLEDGMENTS**

Experimental work was conducted at the Levy Center for Mind, Brain and Behavior of The Rockefeller University, New York, NY, USA. Final analysis and writing was carried out at the Brain Institute, Federal University of Rio Grande do Norte, Natal, RN, Brazil. Funding was provided by the Leon Levy Foundation. The authors would like to thank Andrew Widmer and Robert Assini for assistance with the recordings and Pawel Wojcik for contribution in the respiration sensor design.

#### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at: http://www.frontiersin.org/journal/10.3389/fnbeh. 2014.00399/abstract

#### **REFERENCES**

Andalman, A. S., Foerster, J. N., and Fee, M. S. (2011). Control of vocal and respiratory patterns in birdsong: dissection of forebrain and brainstem mechanisms using temperature. *PLoS ONE* 6:e25461. doi: 10.1371/journal.pone.0025461


Sewell, G. D. (1970). Ultrasonic signals from rodents. *Ultrasonics* 8, 26–30.

Shusterman, R., Smear, M. C., Koulakov, A. A., and Rinberg, D. (2011). Precise olfactory responses tile the sniff cycle. *Nat. Neurosci.* 14, 1039–1044. doi: 10.1038/nn.2877


**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 July 2014; accepted: 30 October 2014; published online: 18 November 2014.*

*Citation: Sirotin YB, Elias Costa M and Laplagne DA (2014) Rodent ultrasonic vocalizations are bound to active sniffing behavior. Front. Behav. Neurosci. 8:399. doi: 10.3389/fnbeh.2014.00399*

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

*Copyright © 2014 Sirotin, Elias Costa and Laplagne. 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.*

## Repeated exposure to odors induces affective habituation of perception and sniffing

### *Camille Ferdenzi\*, Johan Poncelet , Catherine Rouby and Moustafa Bensafi*

*Centre National de la Recherche Scientifique UMR5292, INSERM U1028, Centre de Recherche en Neurosciences de Lyon, Université Claude Bernard Lyon 1, Lyon, France*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Ilona Croy, University of Gothenburg, Sweden Monique A. Smeets, Utrecht University, Netherlands*

#### *\*Correspondence:*

*Camille Ferdenzi, Centre National de la Recherche Scientifique UMR5292, INSERM U1028, Centre de Recherche en Neurosciences de Lyon, Université Claude Bernard Lyon 1, 50 Avenue Tony Garnier, F-69366 Lyon Cedex 07, France e-mail: cferdenzi@crnl.cnrs.fr*

Olfactory perception, and especially hedonic evaluation of odors, is highly flexible, but some mechanisms involved in this flexibility remain to be elucidated. In the present study we aimed at better understanding how repeated exposure to odors can affect their pleasantness. We tested the hypothesis of an affective habituation to the stimuli, namely a decrease of emotional intensity over repetitions. More specifically, we tested whether this effect is subject to inter-individual variability and whether it can also be observed at the olfactomotor level. Twenty-six participants took part in the experiment during which they had to smell two odorants, anise and chocolate, presented 20 times each. On each trial, sniff duration and volume were recorded and paired with ratings of odor pleasantness and intensity. For each smell, we distinguished between "likers" and "dislikers," namely individuals giving positive and negative initial hedonic evaluations. Results showed a significant decrease in pleasantness with time when the odor was initially pleasant ("likers"), while unpleasantness remained stable or slightly decreased when the odor was initially unpleasant ("dislikers"). This deviation toward neutrality was interpreted as affective habituation. This effect was all the more robust as it was observed for both odors and corroborated by sniffing, an objective measurement of odor pleasantness. Affective habituation to odors can be interpreted as an adaptive response to stimuli that prove over time to be devoid of positive or negative outcome on the organism. This study contributes to a better understanding of how olfactory preferences are shaped through exposure, depending on the individual's own initial perception of the odor.

#### **Keywords: pleasantness, smell, repeated exposure, sniff, habituation**

### **INTRODUCTION**

Olfactory perception is known to be highly flexible as a function of perceiver's age, sex, or motivation state, of the context where the odor is perceived or of the characteristic of the odorant itself like its structure or its concentration. Another prominent factor of variations in odor perception is repeated exposure, which is able to improve olfactory detection thresholds (Stevens and O'Connell, 1995; Dalton et al., 2002) and can even boost olfactory sensitivity in seemingly anosmic participants (Wysocki et al., 1989; Mainland et al., 2002). Some studies also investigated the effect of exposure on discrimination abilities. There is now clear evidence that unreinforced exposure to odors can improve discrimination between odorants in humans (Rabin, 1988; Jehl et al., 1995). In line with this, it has been shown that exposure to odor mixtures can alter the perceived quality of the individual components (Stevenson, 2001). For instance, exposure to wine or beer through personal experience or through controlled training improves the ability to discriminate between different wines or beers (Owen and Machamer, 1979; Peron and Allen, 1988; Melcher and Schooler, 1996).

Although these studies show that exposure improves odor perception through differentiation of stimulus features, dimensions, or categories, how repeated exposure to odors affects one of the most prominent dimension of olfaction, namely pleasantness, remains understudied. What happens when we are exposed to the same odorant repeatedly? Do we like it more, or on the contrary do we like it less, or does liking remain stable overtime? A pioneer work in the field conducted by Cain and Johnson (1978) showed that mere presentation of a given odor significantly changed its hedonic value. More specifically, repeated presentation of a pleasant odor (citral) led to a decreased pleasantness whereas repeated presentation of an unpleasant odor (isobutyric acid) led to a decreased unpleasantness. In others words, repeated exposure shifted odor pleasantness ratings toward neutrality, a phenomenon called by Cain and Johnson "affective habituation." However, to the best of our knowledge, non-verbal correlates of self-reported decrease in pleasantness (for pleasant odors) and unpleasantness (for unpleasant ones), such as psychophysiological responses, remain very scarcely investigated in the olfactory domain (but see evoked potentials for the unpleasant pole in Croy et al., 2013). Such a physiological indicator would be of particular interest, because it would strengthen the notion that affective habituation phenomenon is not due to experimental demand or even to a change in the use of the subjective scale over time.

Non-verbal measures of odor hedonics include autonomous (Alaoui-Ismaili et al., 1997; Bensafi et al., 2002a), or motor responses as reflected by reaction time studies (Bensafi et al., 2002b; Jacob and Wang, 2006; Boesveldt et al., 2010) and by sniffing responses to odors (Bensafi et al., 2003, 2007). Indeed, research in animals and humans has shown that sniffing behavior, i.e., the motor component of olfaction, is of considerable importance in odor perception. Sniffing is driven by stimulus attributes such as odor concentration (Laing, 1983; Frank et al., 2003; Johnson et al., 2003), and induces by itself activation in human primary olfactory cortex (Sobel et al., 1998). Furthermore, there is psychophysiological evidence that sniffing is modulated by subjective pleasantness of an odor: sniff duration and sniff volume increase when pleasant odors are sampled compared to unpleasant ones (Frank et al., 2003; Mainland and Sobel, 2006; Bensafi et al., 2007, 2003). Moreover, even when participants are asked to maintain their sniff for a specific duration irrespective of odor content, they sniff pleasant odors stronger and for a longer time (Bensafi et al., 2007). Thus, measuring sniffing patterns has two main advantages in studies on odor hedonics. First, it allows testing whether modulations in pleasantness are consistent with modulations in physiological/motor response. Second, this measure appears less vulnerable than verbal ratings to modulation by explicit or voluntary strategies, which makes it a more objective measure of hedonic responses. The main aim of the present study was therefore to examine whether affective habituation is not only observed at the self-reported level but also reflected at the psychophysiological level, by modulating sniffing responses to pleasant and unpleasant odors.

One striking particularity of odor hedonic responses is their variation between individuals: whereas affective evaluation of a given odor is positive for some individuals, the same smell may be considered unpleasant by others. For example, Doty (1975) emphasized the "large differences between observers in regard to the assessment of odorant hedonicity" (p. 495) based on 10 odorants, and noted for example that benzaldehyde had a bimodal distribution with half of the participants describing it as unpleasant and the other half as pleasant. In the same line, Bensafi et al. (2012) showed that two CO2-odor mixtures received varied hedonic ratings from one participant to another, and revealed differential activations in the brain according to whether the stimulation was perceived as pleasant or unpleasant. Moreover, Lundström et al. (2006) evidenced variability between individuals as regards pleasantness of the smell of androstenone, going from unpleasant to neutral. These hedonic rating differences were accompanied by distinctive verbal descriptions and neural responses in olfactory evoked potentials: Individuals who gave the lowest pleasantness ratings described the smell as "sweaty" and "urinous" and showed larger P3 amplitudes than individuals who gave higher pleasantness ratings and who described the smell with non-body descriptors ("smoky," "fresh," "sweet," and "chemical"). In accordance with this finding, Keller et al. (2007) showed that variation of olfactory receptor expression accounted for a significant part of olfactory perceptual differences, especially between likers and dislikers of androstenone. Neuroimaging studies also shed light on these inter-individual differences in hedonic ratings of smells. In an fMRI study, Rolls and McCabe (2007) showed that chocolate cravers rated this flavor as more pleasant than non-cravers, and that an increasing level of pleasantness was associated with an enhanced activity in the pregenual cingulate cortex, the medial orbitofrontal cortex, and the dorsolateral prefrontal cortex. Altogether, these findings suggest that in olfaction studies, and especially in these dealing with pleasantness, it is of the utmost importance to take into account inter-individual differences because they have significant implications at the peripheral and central levels of olfactory processing.

Therefore, the secondary aim of the present study was to investigate inter-individual variability of the effect of repeated exposure on perceptual ratings and sniffing activity. To this end, participants were exposed to odors for which a previous study revealed large hedonic variability between raters (anise and chocolate; Barkat et al., 2008). In our study, participants were classified as "likers" or "dislikers" for each particular smell based on their initial hedonic ratings. They were then exposed 20 times to each odorant while hedonic ratings and sniffing behavior were recorded. We hypothesized that: (1) olfactory repeated exposure should decrease odor pleasantness in "likers" and odor unpleasantness in "dislikers," (2) such affective habituation should be accompanied by changes in sniff parameters, namely decreased sniff volume and duration in "likers" and increased sniff volume and duration in "dislikers."

## **MATERIALS AND METHODS**

#### **PARTICIPANTS**

Twenty-six young adults (mean age ± s.e.m: 21*.*5 ± 0*.*46, range 19–29; 18 women and 8 men) attending the Claude Bernard University of Lyon (France) participated in the experiment. The experimental procedure was explained in great detail to the participants, who provided written consent prior to participation. The study was conducted according to the Declaration of Helsinki and was approved by the local ethical committee. Based on participant's reports, exclusion criteria were: abnormal olfaction, history of neurological disease or injury, or history of nasal insult (broken nose or surgery).

### **ODORANTS AND OLFACTOMETRY**

Based on the results of a hedonic ranking task involving 8 odorants in a previous study (Barkat et al., 2008), two odorants, anise and chocolate (Euracli, France), were chosen because (1) they received a medium mean rank, and (2) they exhibited a large inter-individual variability. Odorants were diluted in mineral oil (10%) and presented to both nostrils via a nasal mask (**Figure 1A**). They were presented 20 times each in a random order, with an inter-stimulus interval of 30 s and duration of 3 s. Stimulations were delivered via a computer-controlled air-dilution olfactometer whereby odorants were diffused synchronously with the beginning of participant's inspiration (respiration was recorded continuously during the study).

The general principle of the olfactometer is to mix two airflows (odorized and pure air) to deliver a constant odorized or non-odorized airflow to the participant's nose. Pure air is sent by a compressor and cleaned by an activated carbon filter before being carried to the olfactometer input line (6 mm diameter, 5 m length tube). A manometer allows selecting the air input pressure. Then, air enters two channels: (1) a channel where it works as air carrier, and (2) an "odor" channel (one per odorant). For each odorant, a glass tube is set with polypropylene marbles where

**FIGURE 1 | (A)** Experimental device ensuring odor delivery and sniff recording. Nasal respiration was monitored with a flow sensor connected both to the subject's nose via a nasal mask and to the olfactometer. The nasal inspiration, detected by the flow sensor, triggers the sending of the odorant by the olfactometer during the requested duration to the subject's nose. To clean the mask chamber between stimulations, and to decrease

the risk of odor contamination, the mask was connected to a Ruben valve (Intersurgical®, 7562700, 22F-22M/15F, UK) so that the odorized air contained in the chamber was sent out on each expiration. **(B)** Grand average of the sniffs for the odors of anise (AN) and chocolate (CH) across all trials and all participants, showing the maximum flow rate, duration, and volume.

the odor is adsorbed. At the exit of each channel, an electric valve is programmed to be closed or open so that the odorant is pushed into the airflow for a given duration and pressure. The output odorous air is led by a 4 mm tube (20 cm length) into the nasal mask.

The experimental room was well-ventilated and included two areas, one for the experimenter and one for the participant. The experimenter area contained the computer controlling the olfactometer and two control screens showing the processing of the olfactometer and the answers the participant was giving on his/her own screen. The participant's area included the olfactometer output, as well as a screen and a mouse allowing them to read the instructions and give their ratings after each olfactory stimulation.

#### **PROCEDURE**

After providing written informed consent and reading instructions, participants were taken into the testing room. At this point, the experimenter fitted the sniffing equipment to the participants. Sniffing was recorded using an airflow sensor (TSL®, 4000 series, Model 40211, USA) connected to the nasal mask delivering odors to both nostrils. Sniffing signal was amplified and digitally recorded at 100 Hz using Python software®.

Upon installation of the nasal mask, the experiment started. Each trial was timed, and cued by the computer-generated visual instructions "please prepare to smell," displayed for 3 s and announcing odor delivery. Once the instruction disappeared, participants were to sniff, which enabled the airflow sensor to detect the beginning of subject's inspiration and trigger odor delivery via the olfactometer. Following each odor presentation, participants rated stimulus pleasantness and intensity on an on-screen visual analog scale: the left end of the scale was labeled "extremely unpleasant" or "no stimulus perceived" (0), and the right end "extremely pleasant" or "extremely strong" (100). Instructions, odor presentation and sniffing recordings were all time-locked through one central computer.

#### **DATA ANALYSIS**

For each participant, we recorded intensity and pleasantness ratings (0–100) and sniff parameters on 20 occasions per odor (T1 to T20, for anise and chocolate). Sniffs (see **Figure 1B**) were preprocessed by removing baseline offsets and aligned in time by setting the point where the sniff entered the inspiratory phase as time zero. Sniff maximum flow rate, duration, and volume (see **Figure 1B**) were calculated for the first sniff of every trial, for every participant. Before analyzing how the ratings and sniff parameters changed with repeated exposure, outliers defined as values exceeding three standard deviations from the participant's mean were removed (0.65% of the trials). Then, analyses of the time-related changes in ratings and sniff parameters were performed (1) at the group level, by comparing time-related changes of "likers" (participants giving the highest pleasantness scores) and "dislikers" (participants giving the lowest pleasantness scores), and (2) at the individual level, by correlating each participant's initial pleasantness rating at T1 with the time-related changes across the trials T1 to T20. Time-related changes in hedonic and intensity ratings, sniff maximum flow rate, duration, and volume were represented by the slope of each variable as a function of trial number (1 to 20). A positive slope and a negative slope, respectively correspond to an increase and a decrease of the measured variable over time. For the group analysis, slopes were computed on average scores for "likers" and "dislikers" on each trial T1 to T20. The significance of the increases/decreases was assessed by using linear regressions with trial number as individual pleasantness rating at T1 and the slopes of pleasantness, intensity and sniff parameters were investigated using Spearman rank correlation coefficient. Here, we expect "likers" to exhibit negative slopes and "dislikers" to display positive slopes in both perceptual and sniffing variables. This should be confirmed at the individual level by negative correlations between individual hedonic scores at T1 and individual slopes.

### **RESULTS**

#### **INTER-RATER VARIABILITY IN ODOR PLEASANTNESS**

Anise and chocolate were selected for their average neutral valence and the variability of pleasantness ratings they receive in the population. To verify that this was true in our sample, we examined both the average and individual ratings on the very first trial of each odor (i.e., at T1). As expected, anise and chocolate had moderate average pleasantness on the 0–100 hedonic scale, with large inter-individual variations (anise: *M* ± *SD* = 40*.*5 ± 25*.*2, range 1–100; chocolate *M* ± *SD* = 40*.*5 ± 25*.*2, range 1–85). The large variations in pleasantness ratings across participants allowed categorizing them as either "dislikers" or "likers" for each odorant. There were 14 "dislikers" (pleasantness ratings between 1 and 31 at T1) and 12 "likers" (ratings 47–100) for anise, and 13 "dislikers" (ratings 1–47) and 13 "likers" (ratings 50–85) for chocolate.

#### **GROUP ANALYSES: "LIKERS" AND "DISLIKERS"**

Average pleasantness, intensity, sniff maximum flow rate, sniff duration, and sniff volume of "likers" and "dislikers" across the 20 trials are shown in **Figure 2**. Results of the linear regressions between the five variables and time (**Table 1**) suggest that repeated exposure induced a significant decrease in pleasantness and intensity ratings, sniff duration, and sniff volume in "likers," while these variables increased without reaching statistical significance in "dislikers." In both groups, repeated-exposure resulted in a convergence of pleasantness ratings toward neutrality. Indeed, while hedonic ratings of "likers" and "dislikers" significantly differed at T1 (*t*-tests for independent samples, **Table 1**), they did not differ any more at T20. "Likers" and "dislikers" did not significantly differ on the other variables at T1 or T20, except for sniff maximum flow rate, higher in "dislikers" at T20 for chocolate. Finally, pleasantness ratings did not significantly correlate with intensity nor with sniffing parameters at T1 (Spearman rank correlations).

#### **CORRELATION BETWEEN INDIVIDUAL INITIAL PLEASANTNESS AND TIME-RELATED PERCEPTUAL CHANGES**

To go further, we then focused on each participant's pleasantness ratings at T1 and we correlated it with the time-related changes in pleasantness, intensity, and sniff parameters represented by the slopes of these variables as a function of trial number. The slopes were positive or negative depending on the participants (e.g., pleasantness ratings: range = −2*.*21 to +3.57, mean = 0.00 for anise, and range = −3*.*35 to +2.06, mean = −0*.*50 for chocolate). As expected, Spearman coefficients showed significant negative correlations between initial pleasantness and the


**Table 1 | (A) Linear regressions between pleasantness, intensity, sniff maximum flow rate, sniff duration, sniff volume, and trial number (1 to 20) for "likers" and "dislikers" separately, and for the odors of anise and chocolate separately and together. (B)** *t***-tests for independent samples between "likers" (L) and "dislikers" (D) at trial 1 and trial 20.**

*\*\*\*p < 0.001; \*\*p < 0.01; \*p < 0.05.*

slopes of the variables—except sniff maximum flow rate—for one or both odors. These results, illustrated in **Figures 3**, **4**, mean that: (i) higher initial odor pleasantness ratings were associated with larger decreases of pleasantness, intensity, sniff volume and duration during repeated exposure (more negative slopes), and (ii) lower initial odor pleasantness ratings were associated with smaller decreases (slopes closer to zero) and even to increases of these variables (positive slopes), especially for the pleasantness ratings (**Figures 3A,B**) and the sniff volume (**Figures 4E,F**).

### **DISCUSSION**

In the present study, we aimed at testing how hedonic perception of odors varies with repeated exposure, and whether inter-individual differences in hedonic perception of a given odor can modulate this variation. Namely, we used two odors people did not agree to find pleasant or unpleasant and presented them twenty times each (T1 to T20) in a random sequence. We explored time-related perceptual and motor (sniffing) changes for each odor, according to the participant's initial hedonic judgment. First, when considering the groups of "likers" (who rated the odor as pleasant at T1) and of "dislikers" (who rated the odor as unpleasant at T1), we found that pleasantness significantly decreased with time in "likers." In "dislikers," unpleasantness tended to decrease with time but the effect did not reach significance. These effects were paralleled by similar changes in intensity ratings, sniff duration, and sniff volume. We noticed that these effects led to a decrease in affective responsiveness since pleasantness ratings of both groups did not differ any more after 20 odor presentations. Second, when investigating more precisely the level of initial pleasantness rating at T1, we found negative correlations with the slopes (or time-related changes) of pleasantness, intensity, and sniff volume and duration for at least one odor. Correlation graphs (**Figures 3**, **4**) show that higher initial pleasantness was mostly associated with more negative slopes (decrease in ratings and sniffing) and lower initial pleasantness was mostly associated with more positive slopes (increase in ratings and sniffing). In sum, we showed that affective habituation occurs with repeated exposure, which can be observed both at the self-reported level and at the olfactomotor level. We also provided evidence that repeated exposure influences individuals differently according to whether they initially liked or disliked the odor, affective habituation being more significant for odor "likers."

One can wonder whether peripheral mechanisms such as olfactory adaptation may explain the present findings. Peripheral olfactory *adaptation* (or olfactory fatigue) is a phenomenon characterized by a decrease of the olfactory receptors' sensitivity due to prolonged or repeated exposure. Our experimental procedure was designed to limit such phenomenon by using appropriate inter-stimulus intervals (minimum 30 s) and by presenting two different odors randomly. Moreover, olfactory adaptation is characterized by a decrease in perceived intensity (Cain, 1969). Thus, if adaptation had occurred in our study, all participants should have displayed a decrease in perceived intensity, paralleled with an increase in sniff magnitude (see Laing, 1983; Frank et al., 2003, for the link between odor intensity and sniff volume/duration). However, this was not the case since a substantial number of participants displayed positive slopes over time for intensity, sniff volume and sniff duration (see **Figures 3C,D**, **4C–F**). Rather, the

time-related variation of pleasantness toward neutrality observed in our study is likely due to more central processes, and may therefore be preferentially qualified of *affective habituation*. It must be kept in mind that both processes are not independent (central processing can reflect changes in peripheral response) and the origin of response reduction due to repeated exposure remains unclear (Dalton, 2000).

Affective habituation is a form of learning that has been observed in previous studies, through decreasing strength of responses to repeated emotional stimuli of various nature, at the psychophysiological level (reduction of the electrodermal and electromyographic response: Bradley et al., 1993) and at the neurophysiological level (decrement in amydgala activation: Wright et al., 2001; Mutschler et al., 2010). At the behavioral level, few studies have described affective habituation using odors with contrasted pleasantness. Cain and Johnson (1978), who measured pleasantness of odors before and after repeated exposure, found a shift in the direction of hedonic neutrality: the positively valenced odor of citral became less pleasant and the negatively valenced odor of isobutyric acid became less unpleasant after exposure. Similarly, Prescott et al. (2008)showed an increase of pleasantness of two (neutral and unpleasant) odors after an exposure phase, as did Croy et al. (2013) after three presentations of the unpleasant odor of H2S. The latter result was corroborated by a reduced neuronal activation at the cerebral level and was interpreted as a decrease in emotional salience. With a more time-related approach, our study provided further evidence that this effect exists and is gradual: using a linear model of the pleasantness change across 20 odor presentations, we showed that pleasantness follows different trajectories, depending on the initial hedonic rating of the participants.

In this study, sniffing behaviors followed the same pattern as pleasantness ratings. This result reinforces the hypothesis that affective habituation occurs when an odor is repeated in a short period of time. Odor pleasantness is known to co-vary with sniffing behavior parameters, whether the odor is really smelled or whether it is imagined: compared to unpleasant odors (like rotten egg or fish), pleasant ones (like strawberry or rose) have been repeatedly found to be associated with larger and longer sniffs (Bensafi et al., 2003, 2007; Joussain et al., 2013). This motor

correlate of odor pleasantness seems to be a robust mechanism since it is observed even when participants are asked to maintain constant sniffs across conditions (Bensafi et al., 2007). In line with this, we found that, as for pleasantness, sniff volume and duration mostly decreased over time in "likers" and tended to increase or stagnated in "dislikers." In sum, not only did repeated exposure cause pleasantness to become more neutral, it also caused more involuntary parameters of olfactory perception (sniff duration and volume) to reflect this tendency toward neutrality. One may be surprised by the fact that "likers" and "dislikers" did not differ in their sniffing patterns for any of the two odors, and that pleasantness ratings did not correlate with sniffing volume or duration at T1. Relationship between sniff and pleasantness reported in the literature was usually found in response to odors with different qualities and more importantly, with highly contrasted valence (e.g., rose vs. rotten egg in Bensafi et al., 2003). In our study we compared individual responses to the same odor: not only are differences thus likely to be less marked but also interindividual variability may have prevented the difference between "likers" and "dislikers" to reach statistical significance. Sniffing may nonetheless be considered as a reliable measure because, for a given individual, fine time-related changes paralleling changes in pleasantness were found in our study.

Why would affective habituation occur when odors are presented repeatedly? And how can this be interpreted in relation to another apparently contradictory theory, the mere exposure effect, according to which exposure leads to familiarization and higher liking (Zajonc, 1968)? In the conditions of our experimental design, namely 20 repeated presentations of two odors within about an hour, responsiveness to the repeated stimuli decreased. As nicely explained by Dijksterhuis and Smith (2002), habituation is a very useful mechanism that prevents us to be overwhelmed by the numerous stimulations of our environment. When encountering an emotional stimulus, such as an appetitive or a repulsive odor, we may first react intensely, but if subsequent repeated or prolonged exposure proves not to have any positive or negative consequence on the organism, such an intense response becomes unnecessary. On the course of time, the stimulus becomes less relevant, leading to reduced responsiveness. The effect of repeated exposure can be more pronounced or even reversed if the stimulus has effective or supposed consequences on the organism. For example, repeated chocolate ingestion, which has physiological outcomes, leads an initially very positive stimulus (chocolate) to lose its pleasantness (like in our study) and even to become aversive, and activates accordingly two different cerebral substrates related to reward and punishment, respectively (Small et al., 2001). Another example refers to unpleasant odors. If the odor were associated with the belief that it is harmful, by itself or via its source, responsiveness to the odor would then be more likely to increase rather than to decrease or remain stable like in our study. Indeed, in a study by Dalton (1996), perceived intensity of an odor increased over time for an odorous substance presented as being hazardous (sensitization), whereas it decreased in participants who believed that this substance was healthy. If pleasantness of the repeated harmful substance were measured, it probably would decrease over time (instead of increasing or remaining stable like in our study). These results highlight the importance of cognitive influences on odor perception, both at a given time (Herz and von Clef, 2001; De Araujo et al., 2005) and over time. The mere exposure effect, where novel (never encountered) stimuli that become more familiar with exposure also become more appreciated (Zajonc, 1968), may have the same origins as the habituation pattern of initially negative stimuli found in Cain and Johnson (1978) and more moderately in "dislikers" in our study. It is also the phenomenon that might occur in the case of cultural influences on odor perception: learning to associate initially negative smells with positive consequences (taste enjoyment of smelly cheese in France or of the foul-smelling durian fruit in Asia; Ayabe-Kanamura et al., 1998; Ferdenzi et al., 2013) may decrease its unpleasantness possibly to the point where it even reaches the positive side of the pleasantness scale.

In sum, repeated presentation of emotional stimuli such as odors may produce gradual decrease in responsiveness (tendency to neutral hedonic valence), but cognitive influences related to the consequences on the organism can modulate this pattern, by increasing responsiveness to repeated stimuli that have harmful or beneficial outcomes. In future studies, the asymmetry between affective habituation to pleasant and unpleasant odors (or of "likers" and "dislikers") should be investigated further. Indeed, our study suggested that habituation was much less pronounced in "dislikers" than in "likers." In "dislikers," pleasantness and sniff magnitude seemed to have a tendency to increase with repeated exposure but the effect did not reach significance (**Table 1**), while reverse time-related changes were highly significant in "likers." It might be that unpleasant odors are more resistant to the effect of familiarization, because maintaining an aversion for potentially harmful stimuli is an adaptive behavior (Delplanque et al., 2008; Ferdenzi et al., 2013). Affective habituation to unpleasant odors may thus be more limited in amplitude and/or might require longer exposure to reach the same magnitude as with pleasant stimuli, but this remains to be tested.

Finally, our study shows that it is highly relevant for olfaction studies to take into account inter-individual differences in hedonic perception. Agreement between raters and between cultures seems to be lower for neutral and pleasant odors than for unpleasant ones (Schaal et al., 1998). Hedonically neutral odors, in particular, may not be truly "neutral" and may rather receive highly contrasted odor ratings with some participants finding them pleasant and others finding them unpleasant (as in Doty, 1975), which leads to a moderate average score. Our study highlights significant differences from one person to another in the changes of perception and sniffing over time, for the same odor. When investigating odor hedonics, it is hazardous to consider the object *per se* independently of the perceiver (Robin et al., 1999; Rouby and Bensafi, 2002; Forestell and Mennella, 2005) because pleasantness is subjective and depends on personal past experience, current needs and goals.

#### **AUTHOR CONTRIBUTIONS**

Moustafa Bensafi and Catherine Rouby designed the research, Camille Ferdenzi and Moustafa Bensafi analyzed the data and wrote the paper, Johan Poncelet conducted the data collection.

### **ACKNOWLEDGMENTS**

This study was supported by a grant from the ANR to Camille Ferdenzi (PDOC Program, ATTRASENS Project).

### **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: 11 February 2014; accepted: 21 March 2014; published online: 10 April 2014. Citation: Ferdenzi C, Poncelet J, Rouby C and Bensafi M (2014) Repeated exposure to odors induces affective habituation of perception and sniffing. Front. Behav. Neurosci. 8:119. doi: 10.3389/fnbeh.2014.00119*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Ferdenzi, Poncelet, Rouby and Bensafi. 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.*

## Properties and mechanisms of olfactory learning and memory

#### *Michelle T. Tong1 \*, Shane T. Peace2 and Thomas A. Cleland1*

*<sup>1</sup> Computational Physiology Lab, Department of Psychology, Cornell University, Ithaca, NY, USA*

*<sup>2</sup> Computational Physiology Lab, Department of Neurobiology and Behavior, Cornell University, Ithaca, NY, USA*

#### *Edited by:*

*Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France*

#### *Reviewed by:*

*Justus V. Verhagen, The John B. Pierce Laboratory, USA Carolyn Harley, Memorial University of Newfoundland, Canada*

#### *\*Correspondence:*

*Michelle T. Tong, Department of Psychology, Uris Hall, Cornell University, Ithaca, NY 14853, USA e-mail: tt389@cornell.edu*

Memories are dynamic physical phenomena with psychometric forms as well as characteristic timescales. Most of our understanding of the cellular mechanisms underlying the neurophysiology of memory, however, derives from one-trial learning paradigms that, while powerful, do not fully embody the gradual, representational, and statistical aspects of cumulative learning. The early olfactory system—particularly olfactory bulb—comprises a reasonably well-understood and experimentally accessible neuronal network with intrinsic plasticity that underlies both one-trial (adult aversive, neonatal) and cumulative (adult appetitive) odor learning. These olfactory circuits employ many of the same molecular and structural mechanisms of memory as, for example, hippocampal circuits following inhibitory avoidance conditioning, but the temporal sequences of post-conditioning molecular events are likely to differ owing to the need to incorporate new information from ongoing learning events into the evolving memory trace. Moreover, the shapes of acquired odor representations, and their gradual transformation over the course of cumulative learning, also can be directly measured, adding an additional representational dimension to the traditional metrics of memory strength and persistence. In this review, we describe some established molecular and structural mechanisms of memory with a focus on the timecourses of post-conditioning molecular processes. We describe the properties of odor learning intrinsic to the olfactory bulb and review the utility of the olfactory system of adult rodents as a memory system in which to study the cellular mechanisms of cumulative learning.

**Keywords: olfactory bulb, synaptic plasticity, generalization, representational learning, behavior**

#### **1. INTRODUCTION**

Odor learning, like all learning, is distributed across multiple regions of the brain. Studies of learning within associative brain regions such as the hippocampus and prefrontal cortex particularly in rodents—can utilize and manipulate olfactory stimuli just as they do other forms of sensory input (Eichenbaum et al., 1996; Eichenbaum, 1998; Law and Smith, 2012; Peters et al., 2013). Importantly, however, a substantial component of odor learning is intrinsic to the olfactory bulb (OB), and to its interactions with the piriform (olfactory) cortex to which OB mitral cells (second-order sensory neurons of the OB) project (**Figure 1**). Within OB proper, several lines of evidence, including N-methyl-D-aspartate (NMDA)-based synaptic plasticity (Wilson, 1995; McNamara et al., 2008), the long-term potentiation of ascending piriform pyramidal projections onto OB granule cells (Gao and Strowbridge, 2009) and odor memory persistence linked to the selective retention of adult-born interneurons in the OB (Moreno et al., 2009; Kermen et al., 2010; Sultan et al., 2010) indicate that the OB itself supports sophisticated intrinsic plasticity mechanisms that regulate the transformation of olfactory signals across the first principal sensory synapse.

The elucidation of these intrinsic learning mechanisms within OB presents both theoretical and practical opportunities. While the OB is highly interconnected with multiple cortical and subcortical regions, it is morphologically isolated. This facilitates, for example, the specific delivery of neurochemicals or virallypackaged transgenes to the OB via cannulation. The neural circuitry of the OB and the physiology of its diverse neurons are reasonably well-described (**Figure 1**), enabling the development of biophysically realistic models of OB function that can associate specific cellular properties and mechanisms with systems-level function and performance (Migliore and Shepherd, 2002; Li and Cleland, 2013). Specific odor-dependent behavioral paradigms have been developed that are strongly sensitive to OB manipulations and are likely to depend on OB intrinsic learning, enabling some segregation of OB-specific learning from odor learning dependent on other brain regions (Wilson and Linster, 2008). As the direct target of primary sensory neurons, the OB responds to, and differentiates among, the physical stimulus representations of odorants, but also closely apposes these bottom-up inputs with powerful top-down state-dependent and neuromodulatory influences. The olfactory system thus provides a powerful model to study *representational learning* (Bieszczad and Weinberger,

**FIGURE 1 | Circuit diagram of the mammalian olfactory bulb.** The axons of olfactory sensory neurons (OSNs) expressing the same odorant receptor type converge and arborize together to form *glomeruli* (shaded ovals; two depicted) on the surface layer of the olfactory bulb. Intrinsic OB interneurons innervate each glomerulus, including olfactory nerve driven periglomerular cells (PGo), external tufted cell-driven periglomerular cells (PGe), and external tufted cells (ET). Superficial short-axon cells (sSA), closely related to PG cells and possibly part of the same heterogeneous population, are not associated with specific glomeruli but project broadly and laterally within the deep glomerular layer. Principal neurons include mitral cells and tufted cells (collectively depicted as M/T), which interact via reciprocal connections in the external plexiform layer (EPL) with the dendrites of inhibitory granule cells (Gr), thereby receiving

recurrent and lateral inhibition, and project out of the OB to several regions of the brain. The heterogeneous deep short-axon cell population (dSAC) includes cells that deliver GABAergic inhibition onto granule cells and one another, and, along with granule cells, receive centrifugal cortical input from piriform pyramidal cells. OE, olfactory epithelium (in the nasal cavity); GL, glomerular layer; EPL, external plexiform layer; MCL, mitral cell layer; IPL, internal plexiform layer; GCL, granule cell layer. Filled triangles denote excitatory synapses; open circles denote inhibitory synapses. Speckles surrounding OSN terminals denote volume-released GABA and dopamine approaching presynaptic GABAB and dopamine D2 receptors. Note that sSA-PG and sSA-sSA synapses are depicted as excitatory despite being GABAergic (discussed in Cleland, 2014). Figure adapted from Cleland et al. (2012).

2010)—that is, the integrated effects of learning on the *forms* (or shapes) of neural representations as well as their persistence (see section 2.1). However, the OB remains underdeveloped as a model system for the neuroscientific study of learning and memory circuits. Critical elements such as the factors influencing memory persistence, the mechanistic differences between associative and nonassociative conditioning, and the signature molecular mechanisms of cellular and synaptic learning have been observed in OB but require further exploration. We here outline the features of OB-dependent intrinsic learning and review work on the structural and molecular mechanisms of memory formation, with a focus on the timecourses of learning-initiated signaling cascades and the roles of extracellular signals such as classical neuromodulators and the peptide brain-derived neurotrophic factor (BDNF). In particular, we discuss how research into representational, appetitive and cumulative learning mediated by plasticity in the olfactory system can most productively contribute to a broad understanding of general learning and memory mechanisms.

#### **2. LEARNING IN THE OLFACTORY BULB**

#### **2.1. ODOR LEARNING IS REPRESENTATIONAL**

*Learning* alters the transformation of information by a neural circuit, and *memory* refers to the persistence of that altered transformation function over time. In olfactory representational learning, the *forms* of odor representations are sensitive to learning and can be measured using behavioral generalization gradients (Cleland et al., 2002, 2009; Fernandez et al., 2009). Olfactory generalization gradients define the range of variance in odor quality that an animal will respond to as representative of a given odor, and reflect the statistical reliability of odor features (Wright and Smith, 2004). The area under the gradient, or *consequential region* (Shepard, 1987), describes the degree of certainty expressed by the animal that a stimulus of a given quality is likely to represent that learned odor or its implications. Increased pairings of odor with reward progressively sharpen the generalization gradient (**Figure 2A**), and manipulations of other training parameters indicate that factors that increase classical learning also increase the rate of sharpening of olfactory generalization gradients (Cleland et al., 2009). If the odor being paired with reward is itself variable in quality, however, it becomes clear that the generalization gradient does not sharpen *per se*, but progressively conforms to the actual environmental distribution of reward-predicting odor qualities as experienced by the animal (Cleland et al., 2012, **Figure 2B**). That is, the learning-dependent regulation of generalization gradients describes a statistical learning process by which an animal's internal odor representations become gradually and probabilistically categorical (Tenenbaum and Griffiths, 2001), evolving to correctly reflect the meaningful categories of the external olfactory environment (Cleland et al., 2012). This aspect of odor learning has been hypothesized to rely on OB circuitry both for theoretical reasons and based on results from the experimental manipulation of OB circuit function (Mandairon et al., 2006; Guérin et al., 2008; McNamara et al., 2008; Linster and Cleland, 2010; Devore and Linster, 2012; de Almeida et al., 2013; Dillon et al., 2013). Hence, in contrast to *odorants*—which are chemical stimuli, whether simple or complex—*odors* here are psychometrically defined as probability density functions of odorant quality that the animal has learned imply the same consequences, embedded within a high-dimensional similarity space that is best defined by odorant receptor activation levels (Cleland, 2008, 2014). Behaviorally-measured generalization gradients constitute one-dimensional trajectories within this high-dimensional space, essentially estimating the changing form of the odor representation via sampling. The key point is that memory content is not a constant, but changes with learning and over time just as memory strength and persistence do. By providing quantifiable, interpretable measures of memory content as it evolves, odor generalization gradients illustrate the advantages of representational learning systems for the study of learning and memory mechanisms.

#### **2.2. APPETITIVE ODOR LEARNING IS CUMULATIVE AND INCREMENTAL**

The cellular and synaptic neurophysiology of mammalian learning and memory is substantially based on fear conditioning. The advantage of the conditioned fear model is that strong, discrete, and easily measurable memories can be generated by single learning trials, avoiding the complexity and additional questions imposed by the need to integrate the cumulative effects of multiple learning events. The persistence of these memories is a function of the unconditioned stimulus amplitude—e.g., footshock current—but commonly extends to several days (Bekinschtein et al., 2007), enabling study of the sequential transitions in their structural, biochemical, and molecular substrates that occur over time. The clearest example of these gradually transforming dependencies is the protein synthesis requirement for long-term (many hours to days or more) but not short-term (up to a few hours) memory (Davis and Squire, 1984; DeZazzo and Tully, 1995), though additional phases of memory have been defined in some systems. Moreover, many specific cellular signaling cascades, induced by fear conditioning events and underlying the relevant learning, have been described; whereas most of these processes are initiated immediately after the causal event, several have been described that are initiated minutes or even hours later (**Figure 3**).

**FIGURE 2 | Olfactory generalization gradients in mice. (A)** Mice received either 3, 6, or 12 pairings of an odor CS with buried reward, after which their perseverance in digging following randomized presentation of the odor CS, a similar odorant (S1) and a dissimilar odorant (D) was measured. Increasing the number of training trials prior to testing progressively increased perseverance and sharpened associative generalization gradients. 3x: three training trials; 6x: six training trials; 12x: 12 training trials. Asterisks denote significant pairwise differences. Figure adapted from Cleland et al. (2009). **(B)** Generalization

gradients adapt to the variance of the conditioning odor. Mice received 12 pairings of an odor CS with buried reward; this CS was either a 50:50 mixture of odorants C4 and C5 (*Trained on 50:50 mix* group) or was a variable mixture of the same two odorants, varying from 95:5 to 5:95 on different trials (*Trained on high variance* group). The high-variance training group generalized significantly more broadly than the low-variance training group. Abscissa comprises a sequentially-similar homologous series of different odorants (C3, C4, C5, C6) and a dissimilar odorant (D). Figure adapted from Cleland et al. (2012).

Olfactory appetitive learning, in contrast, is gradual, cumulative, and statistical (Cleland et al., 2009, 2012, **Figure 2**). The richness of the OB learning model that is gained by its statistical and representational character also imposes a cost in terms of unavoidable complexity. For example, the distribution of repeated training events in time will always be a factor; *massed* versus *spaced* learning schedules are well known to affect memory persistence (Tsao, 1948; Menzel et al., 2001; Kermen et al., 2010), and intertrial interval timing can even determine which areas of the brain are most immediately responsible for nonassociative odor learning (McNamara et al., 2008). This complexity, however, is manageable, and is substantially mitigated by the theoretical tractability and experimental accessibility of the OB as well as the elucidation of plasticity-related molecular cascades in other cortical memory systems. Studies of representational plasticity and memory processes in primary and secondary sensory cortices offer a singular opportunity to understand a new dimension of memory—plasticity in form as well as persistence—in preparations within which its form can be measured physiologically as well as psychophysically (Bieszczad and Weinberger, 2010). This is critical for a mechanistic understanding of statistical learning, in which repeated stimulus experiences are not identical and part of the challenge of learning is to estimate the intrinsic variability of meaningful stimuli and the relationship between stimulus quality and outcome.

It is worth noting that odor learning in OB is not always gradual and cumulative. Fear conditioning to odorants induces strong one-trial learning, which appears to involve plasticityrelated molecular changes in the OB (Jones et al., 2007). However, this aversive learning appears to substantially *broaden* olfactory generalization, rather than sharpen it as is observed in appetitive odor learning (Chen et al., 2011). This can be interpreted as adaptive, in that broad generalization is the safe, conservative response to a dangerous odor of uncertain variance, but it raises important mechanistic questions: what would be the cumulative effect of multiple aversive conditioning trials, for example, or how would this broad generalization gradient interact with strong, preexisting, non-aversive odor representations? It also suggests that the mechanisms underlying the effects of aversive conditioning in OB may differ qualitatively from those underlying appetitive conditioning at some level of organization.

Odor learning in neonates also is qualitatively different from that described in adults. First, neonatal odor learning is more heavily OB-dependent than it is in adults, in part because downstream learning areas such as the amygdala are not yet functional (Berdel et al., 1997; Sullivan, 2003). Second, neonatal odor learning is substantially stronger and less conditional than it is in adults, exhibiting a nearly all-or-none quality that contributes to the rapid learning of maternal and nest odors (Sullivan et al., 1991, 2000). Accordingly, studies of the molecular cascades and mechanisms underlying intrinsic OB learning are most advanced in neonates (McLean and Harley, 2004; McLean et al., 2005; Grimes et al., 2012; Lethbridge et al., 2012). Many of these mechanisms, however, are likely to be conserved in some form to underlie the incremental appetitive learning exhibited by adults. For example, norepinephrine (NE) and the intracellular cascades that it initiates play a central role in neonatal olfactory learning (Sullivan et al., 1989; Yuan et al., 2003). Indeed, the innate hyperresponsivity of the neonatal locus coeruleus (LC) is the underlying mechanism of one-trial odor learning in neonates (Sullivan, 2001, 2003; Moriceau and Sullivan, 2005), and NE delivery to the OB is sufficient to induce odor preference learning in neonates (see section 4.1). In adults, in contrast, LC responsivity is much more measured and conditional. Nevertheless, NE in the adult OB is essential for even the basic nonassociative learning processes underlying odor habituation (Guérin et al., 2008; Shea et al., 2008; Moreno et al., 2012), and selective NE receptor antagonists infused into OB impair conditioned odor preference learning, recognition memory, and near-threshold odor identification (Guérin et al., 2008; Escanilla et al., 2010, 2012; Linster et al., 2011; Manella et al., 2013). In this way, the cellular mechanisms of memory elaborated by one-trial learning paradigms can serve as the basis for study of the more complex problem of ongoing appetitive learning.

#### **2.3. ADVANTAGES OF OLFACTORY BULB LEARNING MODELS**

The OB provides both practical and theoretical advantages for study of the molecular and structural mechanisms involved in memory. Practically, pharmacological agents can be infused selectively and locally into the OB. Intrinsic OB circuits display functional plasticity similar to other regions of the brain, including long-term synaptic potentiation (Gao and Strowbridge, 2009) and adult neurogenesis (Lledo et al., 2006), and are reconfigured substantially by neuromodulatory inputs (Devore and Linster, 2012). Established behavioral paradigms enable insight into the changing form as well as the persistence of odor representations over time, and physiological studies enable measurements of direct correspondence between environmental changes, behavioral performance, and the synaptic and molecular changes that occur in neural circuitry (Abraham et al., 2012, 2014; Qiu et al., 2014). In particular, odor learning exhibits varying memory durations that are related to behavioral task parameters and depend on evolving physiological substrates for short-term memory (**Figure 4**; McNamara et al., 2008), intermediate-term memory (Grimes et al., 2011), and long-term memory (**Figure 5**; Lazarini and Lledo, 2011).

#### *2.3.1. Habituation and cross-habituation*

In this non-associative olfactory learning paradigm, animals first are habituated to an odorant, responding to repeated presentations with progressively lower investigation times. Some time after habituation, they are presented again with that odorant, or with a series of structurally and perceptually similar odorants (*cross-habituation*, also referred to as spontaneous discrimination). Perceptually distinct odors elicit normal, non-habituated investigation times, but odorants similar to the habituated odorant elicit reduced, partially-habituated responses depending on the degree of similarity between the habituated and test odorants. A generalization gradient therefore can be constructed by presenting a battery of similar odorants to habituated animals and measuring the pattern of cross-habituation among odorants (Cleland et al., 2002). Interestingly, memory for odorant habituation acquired on short timescales (tens of seconds) is predominantly mediated within piriform cortex (Wilson, 2009), whereas habituation on the minutes timescale is localized within OB (McNamara et al., 2008; Chaudhury et al., 2010). Habituation and cross-habituation memory persistence is sensitive to the degree of habituation, declining over a 10–20 min period in standard protocols (Freedman et al., 2013). Both the extent and persistence of cross-habituation memory are regulated by neuromodulatory and hormonal effects in the OB as well as task parameters and

**FIGURE 4 | Olfactory generalization gradients measured at various latencies after conditioning. (A)** Progressive decay of a newly learned olfactory generalization gradient over an STM timescale. Mice received 12 massed training trials in which they dug in a dish of sand scented with a 1.0 Pa conditioned odor to retrieve a food reward. Separate cohorts of conditioned mice then were tested at different latencies for their perseverative responses (digging times) to the odor CS, a highly similar odorant S1, a moderately similar odorant S2, and a structurally and perceptually different odorant D. Responses declined and generalization gradients flattened with greater training-testing latencies. Methodology follows that of Cleland et al. (2009). Four of six latencies tested are depicted for clarity. **(B)** Lin-log plot of digging time in the CS during testing at all six latencies tested (2, 10, 30, 60, 180, 1440 min). Data are fit with the regression line *<sup>y</sup>* = −0.687ln(x) <sup>+</sup> 9.278, *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>.683.

state variables (Mandairon et al., 2006, 2008; Dillon et al., 2013; Manella et al., 2013).

#### *2.3.2. Associative generalization*

Generalization gradients also can be measured in response to odorants that are conditioned via associative pairing with reward (Cleland et al., 2002, 2009). After conditioning, animals are tested with batteries of structurally and perceptually similar odorants, often in a digging task where the odorant cue signals a buried reward. The animals' perseverance, measured as time spent digging, in pursuit of an expected reward (that is not present in test trials) declines with increasing perceptual dissimilarity between the conditioned and test odorants. The breadths and forms of these gradients are sensitive to determinants of learning and to the statistical variance in odorant CS quality across conditioning trials (Cleland et al., 2009, 2012) and also are sensitive to the pharmacological and neuromodulatory manipulation of OB circuitry (Zimering and Cleland, 2011). Associative odor learning based on a standard short-term conditioning paradigm (a single series of up to twelve massed conditioning trials) progressively decays over a timescale of several hours (**Figure 4**), though this timescale is likely to be sensitive to training parameters.

#### *2.3.3. Odor discrimination*

Odor discrimination is the most commonly used olfactory learning model, and subsumes many radically different conditioning paradigms and performance metrics. The distinguishing feature of this task is that animals are motivated to distinguish between two or more odors with different learned contingencies (e.g., one is rewarded and the other not), such that it tends to measure an animal's capacity to learn a given discrimination rather than to measure an odor representation *per se*. Automated tasks with relatively nonintuitive metrics (e.g., odor-specified left-right selection or go/no-go tasks) may utilize hundreds of training trials, whereas tasks with more intuitive (to the animal) metrics such as odor-cued digging often require less than 20 trials to reach criterion. The dependence of odor discrimination performance on OB circuitry corresponds closely with the difficulty of the discrimination (Rinberg et al., 2006), which corroborates theoretical proposals that OB circuitry serves in large part to identify which statistical differences among inputs correspond to meaningfully different odorants, and which are simply variations of a single odor that should be generalized (Cleland et al., 2012).

### *2.3.4. Olfactory performance depends on memory*

In olfaction, memory does not serve only to remember odors past, but is also a critical factor in realtime perceptual processing, even within OB and piriform cortex (Wilson and Stevenson, 2003a,b; Zucco et al., 2014). Hence, short-term and long-term memory processes are likely to be highly interactive and conditional; e.g., the form of a long-term memory should acquire the evolving characteristics of accumulating short-term memory processes during multitrial odor learning tasks or natural learning scenarios. That is, though it is established in general that STM and LTM processes are initiated separately—i.e., LTM is not simply a continuation of STM (**Figure 6**; Izquierdo et al., 1999)—it also is true that LTM must be able to be repeatedly updated based on new information even before it is first behaviorally expressed. One likely scenario is that short-term learning and memory processes contribute to this updating—a hypothesis that the olfactory appetitive learning and memory model is well-poised to test.

### **3. MOLECULAR AND STRUCTURAL MECHANISMS OF LEARNING AND MEMORY**

Memory mechanisms are heterogeneous in form, structure, and timecourse, yet exhibit many commonalities across regions of the brain. We here separate these mechanisms into two broad categories: *molecular*, which includes intracellular cascades, molecular signaling, neuromodulatory influences, activity-dependent protein synthesis, and epigenetic modifications, and *structural*, which includes physiological changes such as long-term potentiation or other synaptic weight modifications, alterations to neuronal morphology such as dendritic branching, changes to terminal shapes or numbers, and ancillary modifications such as effects on glia or cell adhesion to the extracellular matrix, as well as changes to neuron number via adult neurogenesis or selective apoptosis. Mechanisms from these categories often are interdependent, and exhibit characteristic response timecourses that underlie memory-related changes. In this section, we review selected learning models and mechanisms drawn primarily from the hippocampal literature, focusing on models with well-developed response timecourses and signaling mechanisms for which there is evidence of relevance to OB learning as well.

#### **3.1. MOLECULAR MECHANISMS**

Inhibitory avoidance (IA) is a well-established behavioral paradigm for one-trial fear conditioning that offers a simple

analog measure of memory "strength." IA memories can persist strongly for days, enabling study of both short-term and longterm memory mechanisms. If entering a darkened chamber or stepping down from a platform results in footshock on the conditioning trial, a normal animal will hesitate, in subsequent test trials, before again entering that chamber or stepping down. The delay in seconds before again entering the chamber or stepping down is a robust measure of the strength of the actionconsequence association. Much of what is known about the molecular mechanisms of memory and their timecourses in mammalian systems has been developed using this task.

#### *3.1.1. Long-term memory*

IA conditioning leads to a rapid elevation in calcium/calmodulindependent protein kinase II (CaMKII) levels in the hippocampus. This in turn enhances the phosphorylation of cyclic AMP (cAMP) response element binding protein (CREB) (Miyamoto, 2006) and promotes the formation of complexes with ionotropic glutamate NMDA receptors (Sanhueza and Lisman, 2013), which have been shown to play a functional role in learning and memory (reviewed in Danysz et al., 1995). Blocking CaMKII activity immediately after IA training substantially reduced animals' fear responses when measured 24 h later (i.e., LTM). However, blocking CaMKII activity 30 min after IA resulted in a weaker LTM deficit, and blockade 2–4 h after IA had no effect on LTM at all (**Figure 3**; Wolfman et al., 1994). These findings indicate that CaMKII plays a crucial role early in the memory induction process, and that its functional role in LTM formation is confined to a specific period following learning. The neurotrophin BDNF also plays a critical signaling role in LTM induction (**Figure 3**). For example, blockade of BDNF signaling through its TrkB receptor, or through function-blocking anti-BDNF, disrupted LTM but not STM for a conditioned IA event, whereas infusion of recombinant BDNF into hippocampus rescued IA memory from amnesia induced by glucocorticoid receptor blockade (Chen et al., 2012).

Other studies have demonstrated the early involvement of the cAMP–protein kinase A (PKA)–CREB pathway in LTM formation. Cyclic AMP levels in the hippocampus begin to rise about 30 min following IA training, peak at 3 h after training, and decrease to baseline levels circa 6 h after training (Bernabeu et al., 1996, 1997a). PKA activity and CREB phosphorylation (pCREB levels), in contrast, both exhibit two distinct peaks: one immediately following training and another beginning roughly 3 h thereafter and persisting until 6 h, but not 9 h, post-conditioning. The second of these peaks coincides with peak hippocampal cAMP levels (Bernabeu et al., 1997a). Injection of the PKA inhibitor KT5720 into the hippocampus 0–6 h after conditioning impairs IA memory when tested 24 h after training (Bernabeu et al., 1997a). Similarly, injections of CREB antisense oligonucleotides into the amygdala impaired LTM in the IA task (Canal et al., 2008), and infusions of antisense CREB oligonucleotides into the hippocampus prior to water maze training blocked 48 h LTM while sparing 4-h STM (Guzowski and McGaugh, 1997). Mutant mice that lack the α and β isoforms of CREB also exhibit impaired LTM consolidation, but normal STM, on a contextual fear conditioning task (Bourtchuladze et al., 1994).

In a multi-trial, appetitive learning paradigm based on the radial arm maze task, increased PKA activity and CREB phosphorylation levels were observed in the hippocampus after the fourth consecutive day of training, but not after the first day, in contrast to the immediate same-day effects observed in IA studies (Mizuno et al., 2002). A similar contrast between IA conditioning and appetitive learning effects has been described with learning-associated BDNF activation. Infusions of functionblocking anti-BDNF antibody into the hippocampus prior to, but not 6 h after, IA training block LTM, indicating that BDNF activity around the time of learning sets the stage for eventual LTM consolidation (Alonso et al., 2002). In contrast, on an appetitive radial arm maze task, BDNF mRNA levels in hippocampus increased only after 8, but not 4, consecutive days of conditioning, and even then mRNA levels were significantly elevated only after 15 min, but not immediately, following training (Mizuno et al., 2002). A similar increase in BDNF mRNA levels, with a comparable 15-min delay, also was observed after 28 days of training with this task (Mizuno et al., 2000). (It remains unclear whether the levels observed at 8 and 28 days represent a continuous elevation in task-induced BDNF mRNA transcription across those days or reflect multiple peaks in BDNF mRNA activity). These findings suggest that similar molecular mechanisms can mediate multi-trial appetitive learning as underlie fear-based single-trial learning, but that the timecourses can differ. It is in these latter contingencies that the richness of appetitive learning studies is likely to contribute most significantly to general studies of learning and memory. Should LTM be modeled as a statistical evidence accumulation system, in which LTM is formed only after enough evidence has accumulated that the cue-reward association is reliable and likely to remain true over time? How is this compatible with the evidence that, in IA training, LTM induction is initiated immediately after learning (and is not dependent upon intact STM), even though it cannot govern behavioral responses until hours later? Once LTM is induced, how is the persistence of that memory governed? How does existing LTM contribute to STM formation over multi-day training sequences? What factors contribute to the timescales, selectivity, and stringency of new memory formation?

#### *3.1.2. Short-term memory*

The formation and maintenance of short-term memory (STM), which are independent of protein synthesis, rely on different molecular mechanisms than those underlying LTM (Izquierdo et al., 1999). To elucidate these different mechanisms, animals were conditioned using the IA protocol, immediately infused with one of a battery of antagonists into the hippocampus, and behaviorally tested for memory retention at 1.5 h (STM) and 24 h (LTM) after training (Izquierdo, 2000). The study showed that STM formation required cyclic GMP (cGMP), mitogenactivated protein kinase kinase (MAPKK), and PKA, but did not depend on protein kinase G (PKG), protein kinase C (PKC), or CaMKII. (Note that this dependence of STM on PKA differs from the PKA-independent STM of neonatal OB as described above). Additionally, infusions of either an AMPA receptor antagonist or a γ -aminobutyric acid (GABA) subtype A receptor agonist into the entorhinal cortex prior to training impaired STM when tested at 1.5 h following training, but did not impair LTM when tested 24 h after conditioning (Izquierdo et al., 1998). Critically, these results demonstrated that LTM does not derive from STM representations, in that LTM formation does not depend on intact STM. Instead, at least two distinct cascades of events are set in motion after training (**Figure 6**), one of which enables rapid behavioral adjustment but decays in several hours (STM), and the other of which is longer-lasting but cannot be behaviorally expressed for the first few hours after conditioning (LTM). Interestingly, the distinct STM and LTM pathways - and many of their mechanistic elements - are common across widely divergent clades, including mollusks and insects as well as vertebrates (Davis and Squire, 1984; DeZazzo and Tully, 1995; Blum et al., 2009), suggesting that these properties have been strongly conserved.

A distinct, intermediate phase of memory, termed intermediate-term memory (ITM), also has been defined, originally in *Aplysia californica*. It is characterized primarily by its dependence on protein translation but not on transcription (Sutton et al., 2001, 2002), although a separate, mechanistically distinct form of ITM in *Aplysia* also has been described (Sutton et al., 2004). Though the timescales of these memory phases in *Aplysia* differ from their mammalian analogs, a translationdependent, transcription-independent ITM for conditioned odor preference also has been identified in neonatal rat OB (Grimes et al., 2011). Infusion of anisomycin, a translation blocker, immediately after conditioning had no effect on odor memory when the rat pups were tested one or 3 h later, but eliminated the memory when tested 5 or 20 h after conditioning. When actinomycin, a transcription blocker, was similarly infused, memory at 1, 3, and 5 h was comparable to control animals, but an impairment of odor memory was observed at 24 h. It is likely that the mechanism underlying memory during this intermediate period (∼5 h after conditioning) is simply the LTM mechanism; that is, it is in this time window that new proteins begin to be required for memory maintenance, but the translation of existing mRNA transcripts provides a sufficient supply for a limited time. This interpretation is supported by subsequent results demonstrating that PKA blockade blocks 5-h ITM and 24-h LTM, but not 3-h STM (Grimes et al., 2012).

#### **3.2. STRUCTURAL MECHANISMS**

Long-term memory has long been associated with persistent structural changes in specific brain regions involved in the formation of the memory. Many of the molecular mechanisms characterized in LTM induction and maintenance also have been shown to influence these structural changes, which may in some cases be the primary effectors of the memory. We briefly review some of these structural mechanisms here.

#### *3.2.1. Long-term potentiation*

Over decades of research, considerable debate has arisen about whether, and to what extent, long-term synaptic potentiation (LTP, Bliss and Lømo, 1973) underlies or otherwise corresponds to behaviorally-measured LTM (Izquierdo, 1993). The arguments in favor of their relationship were strengthened by the elucidation of two distinct forms of hippocampal long-term plasticity (LTP), a short-duration early form (E-LTP) and a longer-lasting late form (L-LTP) distinguished primarily by the latter's dependence on protein synthesis. Specifically, the persistence of LTP in the CA1 region beyond roughly 4 h depends on mRNA and protein synthesis (Frey et al., 1988; Bliss and Collingridge, 1993); translation blockers injected into rat dentate gyrus during *in vivo* LTP induction caused synaptic potentiation to decay within 3–4 h (Krug et al., 1984). This timescale closely resembles the proteinsynthesis dependency of LTM observed in behavioral studies. Similarly, after LTP induction by a tetanic stimulation of afferent fibers in hippocampal slices, any further tetanus to the afferent within 3 h generates only short-term plasticity across the synapse, whereas after 4 h the same tetanus could generate a longer-lasting potentiation over and above the initially induced LTP level (Frey et al., 1995). The timescale of this effect also corresponds with the STM/LTM distinctions described above, and additionally suggests that LTM expression may free up resources needed for further learning. Finally, several molecular mechanisms associated with memory induction and persistence also regulate LTP. For example, CaMKII activity is necessary for LTP induction (Malinow et al., 1989), PKC inhibition immediately following induction leads to early decay of potentiation (Wang and Feng, 1992), and PKA inhibition prior to LTP induction limits the persistence of LTP to roughly 3 h (Frey et al., 1993). BDNF also facilitates the induction of LTP in hippocampal slices (Korte et al., 1995), and the application of BDNF in the presence of protein synthesis inhibitors is sufficient to transform a short-lasting LTP to a longer-lasting form (Lu et al., 2008), suggesting a role for BDNF in the determination of long-term functional plasticity that is comparable to its necessary and sufficient role in determining LTM persistence (Bekinschtein et al., 2008). Moreover, blockade of BDNF signaling immediately following LTP induction reduced LTP persistence. Specifically, LTP induction in slices generated a transient peak in the phosphorylated form of the TrkB receptor for BDNF; pTrkB levels rose 15 min following induction, peaked at 30 min, and slowly declined to baseline over 2 h (Lu et al., 2011). Preventing TrkB activation with TrkB-IgG at the 30-min peak, but not at 60 min post-induction, inhibited persistent LTP (Kang et al., 1997). The timecourses of these interactions also correspond to those of the early biochemical cascades involved in LTM formation as discussed above.

#### *3.2.2. Neuronal and synaptic morphology*

Changes in neuronal morphology, such as the growth of new dendritic spines, have been shown to accompany novel experiences (Leggio et al., 2005; Jung and Herms, 2014). Importantly, the stabilization of new dendritic spines underlies at least some LTMs (Yang et al., 2009), indicating that durable modifications of the synaptic weights within neuronal networks mediated by physical spines is a structural mechanism underlying memory persistence (reviewed in Ramiro-Cortés et al., 2014; Sotelo and Dusart, 2014). The specific roles of these morphological elements are further emphasized by the dependence of LTM on intact cytoskeletal dynamics (Lamprecht, 2014). Notably, BDNF and other neurotrophins associated with memory regulation have been strongly implicated in the modification and maintenance of both synaptic efficacy and dendritic morphology (reviewed in McAllister et al., 1995; Castello et al., 2014; Zagrebelsky and Korte, 2014).

#### *3.2.3. Adult neurogenesis*

Learning and memory in the hippocampus and olfactory bulb also are associated with the incorporation of new adult-born neurons. The proliferation of new neurons ceases prior to adulthood in most brain regions, with the exception of the hippocampus and OB, and possibly the hypothalamus (Cheng, 2013). Hippocampal progenitor cells are produced in the subgranular zone (SGZ) of the hippocampus and migrate a short distance to the granule cell layer of the dentate gyrus (DG); in contrast, OB progenitor

cells are produced in the subventricular zone (SVZ) and migrate to the OB along the rostral migratory stream for 10–14 days before arriving in the OB and differentiating within the granule cell and glomerular layers (Petreanu and Alvarez-Buylla, 2002). The observation that olfactory learning increases the odorspecific survival of adult-born neurons in OB (Alonso et al., 2006; Kermen et al., 2010; Sultan et al., 2010) and, conversely, that the selective activation of these adult-born neurons facilitates olfactory performance and memory (Alonso et al., 2012), has led to a broad and well-supported hypothesis that adult neurogenesis underlies LTM in OB as it does in the hippocampus (reviewed in Sahay et al., 2011; Gheusi et al., 2013; Lepousez et al., 2013). However, the observation that this constant integration of new neurons does not result in a progressively increasing total neuron number in the OB (Mouret et al., 2008) suggests that these new neurons may be relatively short-lived, or may replace older neurons, or both, rendering unclear some essential aspects of the role of adult neurogenesis in long-term odor memory within OB.

In the hippocampus, environmental enrichment and experience increase the survival rates of adult-generated neurons within the dentate gyrus (Kee et al., 2007; Tashiro et al., 2007). Moreover, critically, the selective destruction of adult-born neurons that recently had been incorporated into the hippocampal network impaired spatial memory in the Morris water maze task when animals were tested seven days after training (Arruda-Carvalho et al., 2011). This latter result indicates that these newly-incorporated neurons were substantially mediating the new spatial memory; indeed, it has been suggested that adultborn neurons in HPC are employed specifically for new learning (i.e., initial acquisition), as opposed to the expression or reacquisition of memory (Anderson et al., 2011). A similar principle is emerging in the OB, within which the selective ablation of newlyincorporated adult-born neurons following appetitive odor conditioning eliminated animals' memory for that odor (Akers et al., 2011).

Interestingly, some of the signaling mechanisms most strongly associated with LTM formation also appear to be involved in the learning-dependent survival of adult-born neurons. Besides a basic activity-dependence arising from glutamate and GABA receptor activation (Platel et al., 2010; Platel and Bordey, 2011), the survival of adult-born neurons is also enhanced by stimulation with NE (Veyrac et al., 2009; Moreno et al., 2012) or BDNF (Scharfman et al., 2005). For example, infusions of BDNF into the hippocampus, when delivered to adult rats over 2 weeks, increased the number of adult-born granule cells when compared against control animals infused with saline vehicle or bovine serum albumin (Scharfman et al., 2005). In heterozygous BDNF knockout mice, the number of surviving new neurons in the hippocampus did not change (despite increased proliferation in the SGZ); however, adult-born neurons continued to express markers of immature neurons as well as reduced dendritic growth, suggesting that reduced BDNF levels impaired their processes of maturation and differentiation. Other studies have emphasized a role for BDNF in the survival, rather than the proliferation or differentiation, of adult-born neurons (e.g., Sairanen et al., 2005).

### **4. MECHANISMS OF ODOR LEARNING IN THE OB**

Odor learning in the OB offers rare opportunities to study the molecular and structural mechanisms of learning and memory in concert with well-controlled perceptual and behavioral tasks. During appetitive learning, OB circuitry integrates information about the statistical properties of the conditioned stimulus, perhaps also incorporating other features of the odor environment, and supports persistent representations of this learning. Insofar as has been studied, the molecular and structural determinants of OB memory appear similar to those described for hippocampal fear conditioning and other memory systems. The particular value of OB-dependent behavioral learning paradigms is that they enable study of these molecular and structural mechanisms in the more complex milieu of cumulative, multi-trial, representational learning, in which the instantiation of LTM is delayed and conditional in nature, and based on information acquired over time. The representational aspect of OB learning further enables study of how learning alters the form, as well as the strength and persistence, of acquired memories.

#### **4.1. MOLECULAR MECHANISMS IN THE OB**

Intrinsic memory mechanisms within the OB appear to share common pathways and adhere to similar pharmacologicallyelaborated phases as have been elucidated in IA-based neural plasticity and memory studies. For example, PKA activity in the neonatal rat OB increases 10 min after one-trial olfactory appetitive conditioning, and blocking PKA activation in the OB with the competitive inhibitor Rp-cAMPS disrupted odor preference memory when tested 5 or 24 h, but not 3 h, after training. Moreover, exogenous administration of the PKA activator Sp-cAMPs into the OB prior to odor exposure sufficed to induce intermediate (5 h) and long-term (24 h) odor preference memory. Higher dosages of Sp-cAMPs into the OB further extended the persistence of this odor preference memory up to 72 h (Grimes et al., 2012). Odor-reward conditioning, but not odor or reward alone, also induced increased CREB phosphorylation in neonatal OB mitral cell nuclei 10 min after training, suggesting that pCREB-related plasticity in mitral cells may be important for the formation of odor LTM (McLean et al., 1999). The MAPK/extracellular signal-related kinase (ERK) pathway also is activated by odor learning in neonates; odor stimulation induced ERK phosphorylation in selective populations of OB neurons related to the identity of the learned odor (Mirich et al., 2004). Neonatal odor learning, like hippocampal LTM, appears to rely on NMDA receptor activation (Lethbridge et al., 2012) and the increased expression of synaptic AMPA receptors (Cui et al., 2011); notably, the PKA-dependent phosphorylation of AMPA receptor subunit GluA1 rises with a similar timecourse as does the level of CREB phosphorylation in mitral cells, peaking at about 10 min post-conditioning (Cui et al., 2011). BDNF mRNA levels increase in the OB and piriform cortex within 2 h of olfactory fear learning (Jones et al., 2007). To the extent that a substantially common set of essential molecular mechanisms is employed, the important distinctions between one-trial learning and appetitive statistical learning become within which neurons, under what conditions, and to what extent these mechanisms are invoked.

Most studies of olfactory learning and memory that measure the form of the odor memory (typically via generalization gradients) have been performed in adult animals and at STM timescales. There is little research to date on the molecular mechanisms underlying bulbar STM, though there is a substantial literature on the effects of neuromodulators, hormones (Dillon et al., 2013), and other extracellular signaling molecules. Noradrenergic effects within OB, in particular, have been studied in both nonassociative and associative olfactory representational learning studies (reviewed in Linster et al., 2011), which suggest that NE in the OB may be necessary for even the simplest forms of odor learning. Notably, a nonspecific infusion of NE into OB suffices to restore the nonassociative learning deficits arising from depletion of cortically-projecting NE fibers (Guérin et al., 2008), though dosage is critical, and bulbar NE levels induced by moderate stress can suppress OB-dependent STM in some contexts (Manella et al., 2013). In neonatal rats, as noted in section 2.2, bulbar NE is heavily released into OB during odor learning, and exogenous application of NE into the OB suffices to induce odor preference learning when paired with an odorant, essentially serving as an unconditioned stimulus as no external source of reward is required (Sullivan et al., 1989; Harley, 2004; Grimes et al., 2012). Activation of the PKA pathway with Sp-cAMPs also acts as an unconditioned stimulus in neonatal OB in this context (Grimes et al., 2012). There is no evidence, however, that bulbar NE can serve as an unconditioned stimulus for adult odor learning, and even in neonates this property may be epiphenomenal. If NE serves to gate activity-dependent plasticity in OB circuits, then known properties of neonatal physiology ensure that in neonates this learning will always be strong, always depend on odor-induced activation of OB circuits, and always be appetitive (neonates respond appetitively and can be positively conditioned to even normally-aversive unconditioned stimuli such as electric shocks; Sullivan, 2001). Consequently, simply gating circuit plasticity in the OB could be expected to directly generate a positive association in neonates. In any event, analogous pairings of odor presentation with bulbar NE infusions in adult mice demonstrate that NE facilitates habituation to presented odors, but does not innately generate odor preferences as it does in neonatal animals (Shea et al., 2008). Of course, other classical neuromodulators, notably acetylcholine acting at muscarinic receptors within OB, also exert effects within OB circuitry on odor learning and STM maintenance (Devore and Linster, 2012; Devore et al., 2012).

BDNF also is clearly implicated in LTM formation for IA learning. While it has been much less thoroughly studied in the olfactory system, BDNF transcription is activated in OB and piriform cortex after odor conditioning (Jones et al., 2007), and olfactory sensory deprivation reduces BDNF expression in neonatal OB (McLean et al., 2001). BDNF and its precursor proBDNF exert distinct physiological effects on OB neuronal excitability and plasticity (Mast and Fadool, 2012). BDNF heterozygous knockout mice and BDNF(Val66Met) point mutants exhibit reduced activity-dependent secretion of BDNF and behavioral deficits in an OB-dependent task. Both mutants habituate normally to odors but exhibit greatly reduced spontaneous discrimination in the cross-habituation task (Bath et al., 2008), suggesting an impairment in their ability to form specific odor representations. The clearest effects of BDNF on the OB, however, are structural in nature, substantially affecting dendritic arborization and adult neurogenesis.

#### **4.2. STRUCTURAL MECHANISMS IN THE OB**

#### *4.2.1. Long-term potentiation*

Long-term potentiation has been clearly if sparsely observed in the early olfactory system, notably within piriform cortex and its ascending synapses into OB. NMDA receptor-dependent LTP has been demonstrated at afferent and associative fiber synapses within piriform cortical slices, and coactivation of the two can facilitate a form of associative LTP if local inhibition is suppressed (Kanter and Haberly, 1990, 1993). Piriform pyramidal neuron feedback projections onto OB granule cells also exhibit spike timing-dependent LTP (Gao and Strowbridge, 2009), which may be a particularly powerful computational element given the importance of dynamical, timing-dependent interactions within OB circuitry. Contemporary models of OBpiriform computations have regarded these circuits as a pattern separation/completion network not unlike the dentate gyrus/CA1 circuit of hippocampus, in which piriform association fibers underlie pattern completion (Hasselmo et al., 1992; Barnes et al., 2008) and their feedback projections onto inhibitory granule cells within OB underlie pattern separation (Strowbridge, 2009), within a common recurrent circuit. This rich and structured plasticity requires further experimental and theoretical development, but exemplifies the capacities of the olfactory system as a model for understanding complex memory systems.

#### *4.2.2. Neuronal and synaptic morphology*

Spine densities in OB and piriform cortex are affected by odor learning and by learning-associated trophic factors, notably BDNF. In piriform cortex, spine density on pyramidal neurons increased in odor-conditioned rats compared with pseudoconditioned or naïve controls, an effect potentially corresponding to increased synaptic weights in the association fiber network (Knafo et al., 2001). In the neonatal OB, dendritic branching and spine morphology is substantially regulated by BDNF signaling mediated by the TrkB receptor (Matsutani and Yamamoto, 2004; Imamura and Greer, 2009). In adults, BDNF expression in the OB persists (Malkovska et al., 2006), and continues to regulate OB dendritogenesis, at least among parvalbumin-expressing neurons of the external plexiform layer (Berghuis et al., 2006). In combination with the integration of adult-born neurons into the OB network (below), it is clear that the regulation of dendritic connectivity among OB neurons is a significant determinant of OB functional plasticity, and that BDNF is a crucial regulator of the underlying mechanisms.

#### *4.2.3. Adult neurogenesis*

Adult neurogenesis in the OB has been studied extensively with regard to its effects on, and mediation of, odor learning. The differentiation of adult-born neurons within OB and its relevance for olfactory perception and odor learning have been extensively studied and reviewed elsewhere (Lazarini and Lledo, 2011; Lepousez et al., 2013; Gheusi et al., 2013). Of particular interest for present purposes, though, is the regulation of these neuronal differentiation processes by signaling molecules and other established mediators of olfactory learning, as well as timing and task dependencies that may suggest points of particular mechanistic importance.

The incorporation of new neurons is most widely associated with olfactory LTM; as described above, the selective ablation of newly differentiated OB neurons specifically disrupted a longterm odor memory (Akers et al., 2011). However, there also are indications that adult-born neurons may participate in STM processes. Infusions of the antimitotic drug AraC into the lateral ventricle of rats abolished the arrival of new neurons into the OB, while largely sparing hippocampal neurogenesis, and impaired short-term nonassociative memory for odors learned thereafter (Breton-Provencher et al., 2009). Specifically, the absence of new neurons in OB did not affect memory for a habituated odor after 30 min, but 60-, 90-, and 120-min odor memories were disrupted compared with control animals. In contrast, AraC treatment did not affect 24-h or 7-day preference memory for an odorant paired with reward over 4 days. It remains unclear whether this difference depends more on the multi-day spacing of the trials or on the associative nature of the task.

Interestingly, it has been proposed that nonassociative and associative odor learning preferentially activate neurons of different ages within OB (Belnoue et al., 2011). Specifically, nonassociative perceptual learning preferentially activated newly-arrived neurons (∼2 weeks old), as measured by c-Fos immunoreactivity, whereas water-rewarded odor discrimination training in a go/no-go task preferentially activated more mature, though still recently generated, interneurons (5–9 weeks of age). This result is consistent with the results described above, in that the OBs of AraC-infused mice in that study were devoid of neurons younger than 3–4 weeks, as required for nonassociative odor learning, but possessed a full complement of neurons in the 5–9 week age range, as were most heavily utilized in the rewarded task. (Also of potential interest is that activation does not necessarily correspond to increased survival; olfactory go/no-go training has been associated with enhancing the survival of 2–4 week old neurons in OB, while increasing apoptosis in 5-week old neurons, and not affecting fully mature interneurons 9 weeks of age or older; Mouret et al., 2008). These results still beg the question, of course, of what factors in these different training paradigms underlie the selective recruitment of different cohorts of new neurons. These results illustrate another advantage of the olfactory system for studies of complex and naturalistic learning, in which task parameters may determine the differential utilization of OB (and non-OB; Luu et al., 2012) circuit elements for odor-dependent learning.

BDNF signaling is a significant contributor to the survival of new neurons in the OB. BDNF levels are similar in both the site of neurogenesis in the SVZ and in the OB, the target of migration, and regulate both neuronal migration and differentiation (the latter via the MAPK pathway) (Petridis and El Maarouf, 2011). Infusions of BDNF into the lateral ventricle of adult rats significantly increased the generation and/or survival of adult-born neurons in the OB (Zigova et al., 1998; Benraiss et al., 2001); in an analogous *in vitro* study, BDNF administered to neurons arising from the subependymal zone of rats promoted their survival (Kirschenbaum and Goldman, 1995). Mice heterozygous for either the BDNF gene or its TrkB receptor exhibit reduced neuron survival in the OB, as do mice with the Val66Met point mutation in the BDNF gene, which impairs activity-dependent BDNF secretion (Bath et al., 2008); these mutants also exhibited impaired nonassociative odor learning as described above. Neuronal proliferation was not affected by these mutations, suggesting that the effects of BDNF primarily relate to survival and differentiation. The powerful effects of this neurotrophin on olfactory learning and neuronal differentiation, and its association with established learning-associated molecular cascades, render it a strong candidate for study in order to elucidate the complex relationships underlying these representational, statistical learning processes in naturalistic contexts.

### **5. IN SUMMARY**

Understanding the neurophysiological basis of natural learning and memory is one of the great challenges of neuroscience. Much of what is known about the cellular mechanisms underlying learning derives from one-trial learning paradigms of inhibitory avoidance (fear conditioning), though research in other plastic neural systems has indicated that they share many, though not all, of the same underlying molecular and structural mechanisms of plasticity. One-trial odor learning studies, which induce plasticity in olfactory bulb, suggest that these cortical circuits also rely on these common mechanisms for plasticity—although bulbar memory also depends on adult neurogenesis, a structural mechanism which it shares only with the hippocampus.

Most natural learning, however, is less categorical than these one-trial paradigms, requiring multiple encounters in order to elucidate relevant stimuli and learn appropriate associations. Appetitive learning in adults, for example, tends to be gradual, conditional, and statistical in nature. This raises new mechanistic questions: how does learning accumulate over multiple trials? How do STM and LTM mechanisms interact over the extended timescales of natural experience? How are the relevant features of the sensory scene identified, selected, and represented? How does learning change the form, or quality, of a sensory representation in response to accumulating information? Developing the olfactory system as a neurophysiological learning and memory model enables engagement with these rich questions.

### **AUTHOR CONTRIBUTIONS**

Michelle T. Tong and Thomas A. Cleland conceived of and wrote the paper, and designed the figures. Shane T. Peace designed and performed the research featured in **Figure 4**. Michelle T. Tong designed and performed the research featured in **Figure 5**.

### **FUNDING**

This work was supported by NIH/NIDCD grant DC012249 to Thomas A. Cleland.

#### **ACKNOWLEDGMENTS**

The authors would like to thank undergraduate research assistants Alan Leung, Chin Ho Fung, Rahul Krishnan, Dmitriy Migdalovich, Simon Wong, and Zhongming Chen for help with the research conducted in **Figures 4**, **5**. We would also like to thank SiWei Luo (Department of Psychology, Cornell University) for providing valuable discussion and references.

### **REFERENCES**


cells, but not guidance to the olfactory bulb. *J. Clin. Neurosci.* 18, 265–270. doi: 10.1016/j.jocn.2010.06.021


and long-term memory for sensitization in *Aplysia*. *Learn. Mem.* 9, 29–40. doi: 10.1101/lm.44802


**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 2014; accepted: 16 June 2014; published online: 07 July 2014. Citation: Tong MT, Peace ST and Cleland TA (2014) Properties and mechanisms of olfactory learning and memory. Front. Behav. Neurosci. 8:238. doi: 10.3389/fnbeh. 2014.00238*

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

*Copyright © 2014 Tong, Peace and Cleland. 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.*

## Sleep and olfactory cortical plasticity

### **Dylan C. Barnes 1,2,3 and Donald A. Wilson1,2,3,4\***

<sup>1</sup> Emotional Brain Institute, Nathan Kline Institute for Psychiatric Research, Orangeburg, NY, USA

<sup>2</sup> Behavioral and Cognitive Neuroscience Program, City University of New York, New York, NY, USA

<sup>3</sup> Department of Biology, University of Oklahoma, Norman, OK, USA

<sup>4</sup> Department of Child and Adolescent Psychiatry, New York University Langone School of Medicine, New York, NY, USA

#### **Edited by:**

Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France

#### **Reviewed by:**

Kensaku Mori, University of Tokyo, Japan Nadine Ravel, Center for Research in Neuroscience of LYON (CRNL), France

#### **\*Correspondence:**

Donald A. Wilson, Department of Child and Adolescent Psychiatry, New York University Langone School of Medicine, 1 Park Avenue, 7th floor, New York, NY 10016, USA e-mail: donald.wilson@nyumc.org

In many systems, sleep plays a vital role in memory consolidation and synaptic homeostasis. These processes together help store information of biological significance and reset synaptic circuits to facilitate acquisition of information in the future. In this review, we describe recent evidence of sleep-dependent changes in olfactory system structure and function which contribute to odor memory and perception. During slow-wave sleep, the piriform cortex becomes hypo-responsive to odor stimulation and instead displays sharp-wave activity similar to that observed within the hippocampal formation. Furthermore, the functional connectivity between the piriform cortex and other cortical and limbic regions is enhanced during slow-wave sleep compared to waking. This combination of conditions may allow odor memory consolidation to occur during a state of reduced external interference and facilitate association of odor memories with stored hedonic and contextual cues. Evidence consistent with sleep-dependent odor replay within olfactory cortical circuits is presented. These data suggest that both the strength and precision of odor memories is sleep-dependent. The work further emphasizes the critical role of synaptic plasticity and memory in not only odor memory but also basic odor perception. The work also suggests a possible link between sleep disturbances that are frequently co-morbid with a wide range of pathologies including Alzheimer's disease, schizophrenia and depression and the known olfactory impairments associated with those disorders.

**Keywords: olfaction, piriform cortex, slow-wave sleep, odor memory, odor perception, memory consolidation**

#### **INTRODUCTION**

The olfactory system (**Figure 1**) is remarkably plastic. This experience-dependent plasticity is not reserved for higher-order olfactory cortical areas or zones of multisensory integration, but rather is expressed throughout the olfactory pathway from the nose to cortex. For example, in the olfactory epithelium, odor experience and associative training can modify olfactory sensory neuron receptor gene expression, survival and axonal targeting (Tyler et al., 2007; Jones et al., 2008; Tian and Ma, 2008; Kass et al., 2013). In the olfactory bulb (OB), such experience can modify OB glomerular size, number of juxtaglomerular neurons and glomerular responses (Fletcher, 2012; Miura et al., 2012). In addition, second-order neuron (mitral/tufted cell) structure (Gheusi et al., 2000; Davison and Ehlers, 2011), sensory physiology and excitability (Moreno et al., 2009; Boland and Alloy, 2013; Tononi and Cirelli, 2014) have also been shown to be sensitive to odor experience. Not only are OB principal neurons changed, but also the large population of OB inhibitory interneurons, granule cells undergo extreme odor experiencedependent changes in survival and physiology (Killgore and McBride, 2006; Moreno et al., 2009; Yokoyama et al., 2011). These learned changes in OB structure lead to changes in OB function, often measured as changes in local circuit oscillations (Grajski and Freeman, 1989; Martin et al., 2006; Chapuis et al., 2009; Kay and Beshel, 2010). Beyond the OB, odor experience has been shown to modify piriform cortical pyramidal cell synaptic and sensory physiology (Mouly et al., 2001; Wilson, 2003, 2010; Roesch et al., 2007; Cohen et al., 2008; Chapuis and Wilson, 2011; Saar et al., 2012), membrane excitability (Saar and Barkai, 2003) and dendritic structure (Knafo et al., 2004), as well as piriform cortical local circuit oscillations (Martin et al., 2006; Kay and Beshel, 2010; Chapuis and Wilson, 2011). Finally, functional connectivity between the OB and olfactory cortex (Martin et al., 2006), and between these primary olfactory structures and higher-order cortical areas such as the hippocampus (Martin et al., 2007) and orbitofrontal cortex (Cohen et al., 2008) are also highly experience-dependent. This list is not meant to be exhaustive of all possible identified changes, but rather meant to exemplify the extent and diversity of changes induced by either passive odor exposure (or lack thereof) or associative conditioning.

Together, these experience-induced changes can modify olfactory system sensitivity and acuity, as well as link odor quality to associative meaning or hedonic valence. The changes can last as short as a few seconds, in the case of short-term odor adaptation (Best and Wilson, 2004) to as long as days, weeks or years in the case of associative memory (Fillion and Blass, 1986) or following long-term odor-exposure (Dalton and Wysocki, 1996).

Apparently the most stable feature of the olfactory system is its constant change. Thus, as described elsewhere (Wilson and Stevenson, 2003; Wilson and Sullivan, 2011), olfactory perception lies at the interface between memory and sensation. That is, experience-dependent plasticity is required not only to allow the associative memory that strawberry aroma signals a pleasant, energy dense fruit, but also to allow the implicit memory underlying perception of the complex mixture of odorants emanating from the fruit as a unique odor object which English-speaking humans label "strawberry".

Most forms of long-term memory require a process of consolidation, wherein temporary traces of encoded information become more permanently stored through post-experience modulation (McGaugh, 2013). Post-experience consolidation is in essence a "save now" process. Not all experiences need be permanently stored. Those events or stimuli that signaled something biologically significant (e.g., a source of food, safety or danger, or mating opportunity) may be more likely to be consolidated than other events. Thus for example, events that elevate peripheral epinephrine or central norepinephrine are more strongly consolidated than those that do not (McGaugh, 2013).

However, it is increasingly apparent that memory consolidation can also involve a sleep-dependent stage. Sleeping within a few hours of learning a new skill or new information enhances memory for that skill or information, compared to staying awake after the initial learning. This consolidation enhancement (enhanced long-term memory) can occur for motor skill memories, declarative/episodic memories, emotional memories and sensory memories/perceptual learning (Stickgold, 2005; Diekelmann and Born, 2010; Rasch and Born, 2013; Tononi and Cirelli, 2014).

Recent work has begun exploring the role of sleep in olfactory memory. Olfaction is interesting in this regard given the unusual anatomy of the olfactory pathway compared to all other sensory systems, most notably the relatively limited involvement of a thalamic nucleus prior to the primary sensory cortex. Two major approaches have been used in this research. First, odors have been used as contextual cues while some other task was ongoing, such as declarative or emotional memory task (Rasch et al., 2007; Eschenko et al., 2008; Hauner et al., 2013). This research has primarily utilized the odor context to manipulate sleep-dependent memory consolidation of these other forms of information (Rasch et al., 2007; Diekelmann et al., 2012; Hauner et al., 2013). Several recent reviews that include this work have been published (Rasch and Born, 2013; Shanahan and Gottfried, 2014) and this work will not be the focus of this paper. The second approach has been to focus on odor memory and the olfactory system itself. This work has examined state-dependent changes in olfactory system activity and connectivity, and explored how sleep directly modifies the olfactory system after odor experience. Based on this work, we argue here that olfactory system structure and function, and thus olfactory perception itself, is shaped in an odor-specific manner during post-odor experience sleep. These modifications are critical to both the associative meaning of learned odors, as well as the acuity of odor perception and memory. The work suggests that odor perception not only depends on the odors we smell, but also on the odors of which we dream.

### **SLEEP**

Sleep or sleep-like states appear to occur in all animals, including invertebrates and vertebrates. Sleep is generally defined as a period of behavioral dormancy, though is not necessarily associated with a quiescent nervous system. In fact, the mammalian brain is highly active during sleep and undergoes shifts between several different sleep-related physiological states that together are referred to as sleep structure. Although sleep-related activity can occur unilaterally in marine mammals and some birds, generally the entire brain enters specific stages of sleep nearly simultaneously.

#### **SLEEP STRUCTURE**

In non-human mammals, there are two widely accepted behavioral phases of sleep: rapid eye movement (REM) or paradoxical sleep and slow wave sleep (SWS), also known as non-paradoxical or NREM sleep (**Figure 2**). In the circadian rhythm of most mammals, early nocturnal and diurnal sleep contains mostly SWS and late sleep is largely devoted to REM sleep (Tobler, 1995). Both of these sleep stages are characterized by specific field potential oscillations of brain activity. REM sleep is characterized by fast, low-amplitude oscillatory activity in the θ-band (4–8 Hz) and higher frequency bands characteristic of waking (van der Helm et al., 2011). REM sleep also includes ponto-geniculo-occipital (PGO) waves that are intense bursts of synchronized activity propagating from the pontine region of the brain stem to the lateral geniculate nucleus and visual cortex (Datta and O'Malley, 2013). Behavioral hallmarks of REM sleep are phasic bouts of REM (hence the phase's name) and muscle atonia (Jones, 1979).

The predominant oscillatory activity in SWS is in the δ-band (0.5–4.0 Hz), which includes the less than 1 Hz slow oscillation (Buzsáki, 1986; Mölle et al., 2002; Montgomery et al., 2008). These oscillations are comprised of alternating periods of membrane depolarization, the "up" states, and hyperpolarization, "down" states. Most cortical neurons engage in slow oscillations and the firing patterns produce high synchrony across cellular populations in different brain regions (Amzica and Steriade, 1998; Steriade et al., 2001; Volgushev et al., 2006). The widespread synchrony of neuronal activity during SWS is the backbone of the slow oscillation and allows for a relatively global time scale in which activity is limited to the depolarized up states and terminated with the hyperpolarized down states (Luczak et al., 2007; Mölle and Born, 2011). Early sleep SWS is classically thought of as important for processing declarative and hippocampal based memories (Wagner and Born, 2008), although recent studies may point to complementary, rather than separate processes for memory consolidation (see below) (Rasch and Born, 2013).

A special type of oscillation seen in the hippocampus (Eschenko et al., 2008) and the piriform cortex (Wilson, 2010; Manabe et al., 2011; Narikiyo et al., 2014) during periods of quiet wakefulness and, more predominantly during SWS, is the sharp-wave ripple (SPW-R). These are fast depolarizing waves that were first seen in the CA3 region of the hippocampus and are superimposed by high frequency (100–300 Hz) ripple oscillations (Diekelmann and Born, 2010). During sharp-wave ripple events, small subpopulations of pyramidal cells that were active during prior wakefulness spontaneously fire in the same pattern as during learning (Pavlides and Winson, 1989) in a much shorter amount of time as compared to their original timing during encoding. The number of sharp-wave ripple events during post-learning sleep is also significantly correlated with the formation and strength of memories (Axmacher et al., 2008; Ramadan et al., 2009). The olfactory system is special in relation to other sensory systems as sharp-waves have also been described within the main olfactory cortex, the piriform cortex (Wilson and Yan, 2010; Manabe et al., 2011; Narikiyo et al., 2014) that are similar to those described in the hippocampus (Buzsáki, 1986), and that these sharp-waves appear to be generated independent of hippocampal origination, perhaps originating in the endopiriform nucleus (Behan and Haberly, 1999).

#### **SLEEP IS IMPLICATED IN SYNAPTIC HOMEOSTASIS**

SWS is traditionally thought of as the marker for homeostatically regulated sleep pressure. After prolonged periods of wakefulness, sleep pressure is greatest as evidenced by a shorter latency to enter SWS following sleep onset and an increased power of slow wave activity (SWA), which gradually decreases from early to late sleep (Borbély, 1982; Borbély and Achermann, 1999; Riedner et al., 2007). However, in addition to sleep homeostasis, SWS may also play an important role in the homeostasis of synaptic weight. That is, sleep may be a time when the brain can re-adjust synaptic weights according to their recent history of use, returning them to a state where they can be maximally efficient at encoding new information—avoiding saturation and incapable of further strengthening (Tononi and Cirelli, 2003). This hypothesis of synaptic homeostasis allows for a widespread decrease in the strength of synaptic connections that take place during SWS. Plasticity-related gene expression follows a circadian rhythm and sleep deprivation causes a decrease in the expression of these genes (Guzman-Marin et al., 2006). During wakefulness, and especially during learning, there is a buildup of molecular correlates associated with long-term potentiation (LTP) such as BDNF and cAMP response element-binding protein (Silva, 2003). During SWS, however, the expression of these genes is significantly reduced or downscaled (Tononi and Cirelli, 2003). LTP also involves an increase in dendritic spine growth. Sleep, however, seems to act as a pruning mechanism during which time there is a decrease in the number of spines (Bushey et al., 2011). Synaptic downscaling, therefore, prevents saturation of the synaptic weight and reduces place and energy demands of a neural network, thereby allowing the system to reset to prepare for the encoding of new information during succeeding wakefulness (Horn et al., 1998a,b). The hypothesis promotes the consolidation of stronger memories, however, that were encoded during the pre-sleep period to remain and replay spontaneously while more trivial synapses can be reset. This active system consolidation implicates a certain selectivity of which memories to consolidate. It is likely that the widespread consolidation of everything recently learned would produce a system overflow. In fact, sleep does not benefit all memories, although the mechanisms that determine which memory will be tagged for consolidation and which will be forgotten during sleep is still unclear.

### **SLEEP AND PERCEPTUAL LEARNING IN THALAMOCORTICAL SYSTEMS**

As noted above, sleep-dependent consolidation has been demonstrated for a variety of forms of information including declarative/episodic, motor, emotional and sensory memory. The prevailing view is that during sleep, recently acquired information is replayed, both within local neural networks (e.g., hippocampus or sensory cortex) and across networks (e.g., linking hippocampal and neocortical networks). This replay allows recently strengthened synapses to solidify those changes either biochemically or structurally, while other less critical synapses are reset to basal strengths, ready for plasticity another day (Stickgold, 2005; Cirelli and Tononi, 2008; Rasch and Born, 2013; Tononi and Cirelli, 2014). Performing replay during sleep, especially SWS, may be ideal as it is a period of hypo-responsiveness to external sensory inputs, and thus replay may be less prone to external interference.

Although sleep-dependent consolidation influences most forms of memory, here we focus on the perceptual learning to highlight evidence of sleep-dependent consolidation within thalamocortical sensory systems. Perceptual learning is an enhancement in perceptual acuity or sensitivity induced by training. Perceptual learning is associated with sensory system plasticity and is remarkably specific. For example, learning to make very precise judgments of the alignment of two vertical lines (e.g., a vernier scale) does not transfer to judgments of diagonal or horizontal lines. Similar selective improvements can be made in the other thalamocortical senses with appropriate training. The learned enhancement in sensory acuity is associated with narrowed receptive fields in sensory neocortex and/or enlargement of the sensory cortical region devoted to that segment of the sensory world (e.g., Recanzone et al., 1992; Godde et al., 1996; Kilgard and Merzenich, 1998; Xerri et al., 1999). The improvement in sensory acuity following perceptual learning is facilitated by sleep (Karni et al., 1994; Allen, 2003; Fenn et al., 2003; Atienza et al., 2004; Gottselig et al., 2004; Censor et al., 2006; Yotsumoto et al., 2009), although see Hussain et al. (2008) and Aberg et al. (2009) for counter–examples. Both REM (Karni et al., 1994) and SWS (Aeschbach et al., 2008) have been implicated in consolidation of perceptual learning. As noted above, SWS may be a very effective period for sleep-dependent replay of learned stimuli given that the sensory evoked activity in the thalamus becomes highly variable and reduced during SWS (McCormick and Bal, 1994; Steriade et al., 2001). This shift away from thalamic monitoring of external events may reduce interference during memory replay.

In a variety of paradigms, sleep structure during the posttraining period is modified, with prolonged bouts of either REM or SWS, depending on the specific task (Tononi and Cirelli, 2014). This prolongation of sleep bouts is hypothesized to reflect the additional pressure required for memory storage. In addition, perceptual learning can influence sleep related activity in neocortical sensory systems in other ways. First, SWS related activity (EEG oscillation power or fMRI activation) is enhanced in sensory cortex during sleep after perceptual learning (Cantero et al., 2002; Yotsumoto et al., 2009; Bang et al., 2013), and specifically in those regions of sensory neocortex encoding the learned stimulus (Yotsumoto et al., 2009). Second, while slow waves are cortical-wide events generally driven by thalamocortical oscillations, recent evidence suggests that following visual perceptual learning, slow-waves may be preferentially initiated in primary visual cortex (Mascetti et al., 2013).

Together, these changes in sleep structure, and sleep-related oscillatory activity within the primary sensory system may promote consolidation and refinement of newly learned sensory representations, allowing enhanced perceptual acuity after sleep compared to a similar period of waking (Karni et al., 1994; Mednick et al., 2002; Allen, 2003; Fenn et al., 2003; Aeschbach et al., 2008). However, neocortical sensory systems evolved well after the primary olfactory system. Can the mammalian olfactory system, with a very different structure and relationship to thalamus, also support sleep-dependent memory consolidation?

### **SLEEP AND OLFACTORY SYSTEM PHYSIOLOGY**

Sleep-dependent changes in neocortical function are largely shaped by changes in thalamic activity (Steriade et al., 2001; Buzsaki, 2006). For example, slow-waves derive from a synchronization of slow, δ-frequency oscillations generated in thalamic and neocortical networks. During this thalamocortical slow-wave activity, sensory evoked activity in the thalamus is reduced and more variable, resulting in reduced sensory cortical activation. In contrast, although there is an olfactory thalamic nucleus that contributes to odor processing and memory (Lu and Slotnick, 1990; Courtiol and Wilson, 2014), it is downstream of the primary olfactory cortex, not between the cortex and the periphery. Thus, the state-dependent gating performed by the thalamus in thalamocortical systems is missing in olfaction. Furthermore, rather than displaying slow-wave activity during SWS as in thalamocortical systems, the olfactory cortex (piriform cortex) displays sharpwave ripples, similar to that observed in the hippocampal formation. Nonetheless, important parallels exist between the olfactory cortex and thalamocortical systems in sleep-wake state dependent physiology.

Although there is no thalamic state-dependent gate between the nose and the piriform cortex, the piriform cortex shows robust state-dependent fluctuations in odor responsiveness. During SWS, both piriform cortical single-units (Murakami et al., 2005; Wilson, 2010) and local field potentials (Barnes et al., 2011) show greatly attenuated odor responses. This SWS-dependent modulation is expressed both in unanesthetized animals spontaneously cycling between waking and SWS (Barnes et al., 2011; Manabe et al., 2011), and in urethane-anesthetized animals (Murakami et al., 2005; Wilson, 2010) that also show spontaneous fast-wave/slow-wave cycling. It should be emphasized that cortical odor-evoked activity is not completely eliminated during SWS, but is greatly reduced. In unanesthetized animals, odor responses appear unaffected by REM sleep states (Barnes et al., 2011). The human olfactory system appears similarly depressed during SWS (Carskadon and Herz, 2004).

During SWS, while the piriform cortex responsivity to the outside world is reduced, its activity shifts to sharp-wave/ripple like activity, reminiscent of that observed in the hippocampal formation. However, the piriform cortical sharp-wave activity is relatively independent of that observed simultaneously in the hippocampus (Manabe et al., 2011). Although the generator of these large cortical sharpwaves in unknown, one hypothesis is that they are driven by the highly auto-excitatory endopiriform nucleus, which has broad excitatory connections throughout piriform cortex (Behan and Haberly, 1999). In fact, the current sink for piriform cortical sharp-waves is located in layers II/III which is consistent with a intracortical association fiber/endopiriform driven potential (Wilson, 2010; Manabe et al., 2011). Piriform cortical layer II/III single-unit activity during these sharpwaves coincides with the deep recorded negative peak (Wilson, 2010; Manabe et al., 2011; Narikiyo et al., 2014). That is, units shift from primarily firing in phase with the respiratory cycle during waking or fast-wave states to firing en masse in phase with the sharpwaves during SWS.

This coherent firing of large piriform cortical pyramidal cell ensembles in phase with sharpwave evokes strong responses in monosynaptic targets of the piriform, including feedback to the OB (Manabe et al., 2011; Narikiyo et al., 2014). Functional connectivity/coherence of piriform cortex with limbic structures such as the basolateral amygdala and dorsal hippocampus, as well as neocortical areas is significantly enhanced during SWS compared to fast-wave states (Wilson and Yan, 2010; Wilson et al., 2011). Combined with the reduced response to sensory afferents, these changes in functional connectivity suggest a turning inward, perhaps consistent with the needs of replay and strengthening odor associations with meaning and hedonics.

One contributor to the state-dependent shift in piriform cortical activity is a change in neuromodulatory tone over the sleep-wake cycle, particularly acetylcholine (ACh). ACh plays a major role in odor processing and plasticity throughout the olfactory pathway from the OB (Ravel et al., 1994; Tsuno et al., 2008; Chaudhury et al., 2009), to olfactory cortex (Barkai et al., 1994; Wilson, 2001; Chapuis and Wilson, 2013). In the piriform cortex, ACh muscarinic receptor (AChmR) activation suppresses association fiber synaptic efficacy through a reduction in presynaptic glutamate release, with minimal effects on afferent fiber synapses from mitral cells (Hasselmo and Bower, 1992). This means that during waking or vigilance, when ACh levels are high, the piriform cortex would be strongly driven by afferent input from the bulb, while intracortical association fibers would be suppressed. During SWS however, when ACh levels drop, the intracortical association fiber system would be released from cholinergic suppression and could more effectively influence cortical activity. Such intracortical excitation between recently activated piriform cortical pyramidal cells during SWS may be important in replay of recently experienced odors.

### **SLEEP AND ODOR MEMORY**

As noted above, a number of groups have used odors and odor contexts as cues in declarative, procedural or emotional memory tasks (Rasch et al., 2007; Eschenko et al., 2008; Diekelmann et al., 2011; Arzi et al., 2012; Hauner et al., 2013). This work has emphasized how odors delivered during sleep can modulate consolidation of various forms of memory in limbic or other brain regions. Here, we focus specifically on work examining how sleep related activity shapes the olfactory system itself, in turn shaping odor coding, perception and memory.

#### **ODOR EXPERIENCE AFFECTS SLEEP STRUCTURE**

Laboratory housed rats and mice spend approximately 40% of a 24 h day in SWS (Barnes et al., 2011). Left alone for a 4 h period during the light cycle, rodents enter SWS within 5–15 min and then cycle between SWS, REM sleep and waking, with REM sleep bouts emerging after the initial SWS bouts. As is true in a variety of conditioning tasks, during a 4 h period following odorfear conditioning rats spend significantly more time in SWS than control rats, as recorded within the piriform cortex (Barnes et al., 2011; Barnes and Wilson, 2014). In our paradigm, there was a slight, non-significant decrease in REM sleep, thus the total duration of sleep did not change (Barnes et al., 2011). Importantly, the percentage increase in post-training SWS duration strongly correlated with the strength of fear memory 24 h later (Barnes et al., 2011).

Thus, during SWS, the primary olfactory cortex is hyporesponsive to environmental stimulation, instead displaying sharp-wave activity, and following fear conditioning, animals spend more time in this state. It is hypothesized that the piriform cortical sharpwave activity during SWS may contribute to olfactory system reorganization and odor memory consolidation during a time of reduced interference from new odor stimulation.

#### **ODOR REPLAY MODULATES MEMORY STRENGTH AND PRECISION**

As an initial examination of whether odors are replayed in the piriform cortex during sharp-wave activity, we asked whether the temporal structure of single-unit activity during sharp-waves is affected by recent (15–30 min) odor experience that occurred during fast-wave state (Wilson, 2010). Piriform cortical singleunits were recorded in anesthetized rats, which, though surgically anesthetized, naturally cycle between fast-wave and slow-wave states (Fontanini and Bower, 2005; Murakami et al., 2005; Wilson, 2010). Single-unit activity during piriform cortical sharpwaves was characterized and then the animal was allowed to spontaneously shift to fastwave state. During fastwave state, the animal was repeatedly exposed to an odor to which the recorded unit responded, or in control animals to no odor or to an odor to which the cell did not respond. After the stimulation, the animal was allowed to spontaneously return to slow-wave state, and again unit activity was characterized relative to sharpwave events. In cells that had been stimulated with odor during the fast-wave state, the temporal structure of sharp-wave related unit activity shifted and became more variable, compared to either control group (Wilson, 2010). Thus, the temporal structure of piriform cortical single-unit activity during slow-wave state was shaped by past odor experience. These data do not confirm that the cells were replaying the odor during a sharpwave, but are consistent with that interpretation. Unfortunately spontaneous single-unit activity in the piriform cortex is very slow (Roesch et al., 2007; Chen et al., 2011), thus some of the analytical techniques to discern actual "replay" that are used in other brain regions are currently less effective here. To address this problem, we decided to impose replay directly following odor fear conditioning.

Odor fear conditioning can induce either highly odor-specific fear responses (e.g., freezing) or more generalized odor-evoked freezing, depending on the nature of the training protocol (Chen et al., 2011). Differential conditioning, involving both a CS+ (predicts footshock) and a CS− (does not predict footshock) induces fear responses selectively to the CS+. Conditioning with a CS+ but no CS−, in contrast, induces generalized fear responses to a wide range of odors. These different behavioral outcomes are associated with distinct changes in piriform cortical single-unit odor coding (Chen et al., 2011). Differential conditioning, which induces odor specific fear, results in a narrowing of piriform cortical single-unit odor receptive fields compared to pseudoconditioned controls, i.e., an enhancement in odor acuity. In contrast, conditioning without a CS−, which results in generalized odor fear, is associated with a broadening of piriform cortical single-unit odor receptive fields, i.e., a reduction in acuity. This suggests that the precision of the odor memory is at least in part due to changes in piriform cortical odor coding. Similar results can be observed after appetitive conditioning (Chapuis and Wilson, 2011). Does cortical activity during post-training sleep contribute to the strength and/or precision of this odor memory?

To test this, we utilized olfactomimetic electrical stimulation of the OB (Mouly et al., 2001) as the CS+ and CS− in a differential fear conditioning paradigm, with different spatial patterns of electrical stimulation serving as the different stimuli. We chose this type of stimulus because it allowed us to deliver identical stimuli, with precise temporal control during training and different posttraining states, regardless of the animal's posture or respiration in the different states. We further hypothesized that the strength of the direct OB stimulation would allow activation of the piriform cortex regardless of behavioral state.

Rats were trained using these olfactomimetic stimuli in a differential fear conditioning task, and then placed in a quiet chamber for the 4 h post-training. Piriform cortical local field potentials and neck muscle EMG were recorded to identify sleep/wake states. Animals that were trained and then left alone during the 4 h post-training period displayed CS+ specific freezing the following day. However, the strength and accuracy of this memory could be modulated by post-training CS+ imposed replay (Barnes and Wilson, 2014). For example, animals that received imposed post-training replay (i.e., OB stimulation identical to the CS+ delivered during training) while they were awake showed reduced CS+ evoked freezing the following day, consistent with extinction. In contrast, animals that received the identical imposed replay selectively during slow-wave sleep showed a significant enhancement in the strength of the CS+ evoked response. This enhanced memory was selective for the CS+. Importantly, imposing replay during slow-wave sleep of a novel olfactomimetic stimulus not previously encountered, while the animal was hypothetically spontaneously replaying the learned stimulus resulted in normal memory strength for the CS+, but now the memory was generalized to all olfactomimetic stimuli tested (Barnes and Wilson, 2014).

We hypothesize that during post-training slow-wave sleep, ensembles of piriform cortical neurons that had been recently co-activated during training, preferentially fire together during piriform cortical sharp-waves in a form of spontaneous replay. This co-activation is facilitated by a release of intracortical association fiber synapse suppression due to low levels of cholinergic activity during slow-wave sleep (Hasselmo and Bower, 1992; Hasselmo and McGaughy, 2004). The co-activation during sharp-waves should strengthen excitatory synapses between neurons within the ensemble, enhancing representation of the learned odor (Linster et al., 2003, 2009). Adding noise to the replay by imposing the novel stimulation during slow-wave sleep should expand the membership of the ensemble to include irrelevant neurons, thus degrading the precision of the odor representation and inducing generalized fear responses.

If this model is correct, then disrupting intracortical association fiber synapses during the post-training period should impair odor memory strength and/or memory precision. To test this, we infused the GABA<sup>B</sup> receptor agonist baclofen bilaterally into the piriform cortex during the posttraining period. Baclofen selectively suppresses association fiber synapses in the piriform cortex (Tang and Hasselmo, 1994; Poo and Isaacson, 2011; Barnes and Wilson, 2014), and suppresses sharp-wave amplitude (Barnes and Wilson, 2014). Animals with suppressed association fiber synapses during the 4 h post-training period showed normal freezing to the CS+, however the freezing strongly generalized across odors. Thus, the post-training association fiber activity, presumably during spontaneous sleep-dependent replay, is necessary for odor memory precision the following day (Barnes and Wilson, 2014).

Together, this work suggests that piriform cortical sharpwave activity during post-training slow-wave sleep allows for a strengthening of ensemble representation of the learned odor. This post-training replay can modify both the strength and the precision of the odor memory, at least in part via plasticity within the piriform cortex itself. However, it must be noted that co-activation of large piriform cortical ensembles during sharp-waves also result in propagation of spike trains out of the cortex to its monosynaptic neighbors. Thus, this provides an opportunity for piriform cortical manipulation not only of intra-cortical synapses, but also of its efferent targets during slow-wave sleep. These "downstream" effects have been the focus of recent work in Kensaku Mori's group (Manabe et al., 2011; Yokoyama et al., 2011; Yamaguchi et al., 2013; Narikiyo et al., 2014).

### **ODOR REPLAY MODULATES ADULT-BORN OLFACTORY BULB NEURON SURVIVAL**

One important role for odor learning is for memory of novel flavors. The majority of the perception of flavor is derived from volatile food odorants delivered to the olfactory epithelium via retronasal smell. While humans are believed to have the most advanced retronasal olfactory abilities (Shepherd, 2012), rodents also experience retronasal smell (Chapuis et al., 2009). In omnivore's such as mus musculus and rattus norvegicus, learning about food odors, and their nutritional and/or illness producing characteristics is critical for survival. Interestingly, a common behavioral response in many mammals following a satiating meal is drowsiness and sleep; referred to a post-prandial sleep. For example, 50% or more of food deprived mice begin resting or fall asleep within 1 h of being given access to food (Yokoyama et al., 2011). Could post-prandial sleep contribute to memory for the odors and flavors of the consumed food?

As noted above, odor memory is associated with changes throughout the olfactory pathway. Perhaps the most extreme neural correlate of odor memory is differential survival of OB granule cells. OB granule cells are inhibitory interneurons which modulate the excitability of OB mitral and tufted cells and are the primary target of descending inputs from olfactory cortex. Importantly, granule cells display adult neurogenesis (Bayer, 1983), with survival of adult born neurons dependent on olfactory stimulation and activity (Killgore and McBride, 2006; Moreno et al., 2009). Differential experience-dependent granule cell survival may contribute to olfactory acuity and information storage (Gheusi et al., 2013). Yokoyama et al. (2011) have demonstrated that there is enhanced granule cell death in the few hours postfeeding. The extent of cell death is correlated with the amount of time spent in slow-wave sleep (though not REM sleep) (Yokoyama et al., 2011). Granule cell death is enhanced even more if the bulb is odor deprived during the food exposure. This suggests that those granule cells not activated by the food odors are selectively targeted for apoptosis during post-prandial sleep.

Mori's group suggests that the strong descending sharpwaveassociated pyramidal cell spiking from the olfactory cortex to the OB during slow-wave sleep (Manabe et al., 2011) may be the critical signal differentiating granule cell death and survival. Thus, it is hypothesized that during waking odor exposure, for example during the meal, granule cells activated by the odors are tagged by descending piriform cortical axonal input. Those cells not activated by the odor do not receive the same tagging. During subsequent post-prandial slow-wave sleep, cortical sharpwave evoked descending activity, perhaps in concert with sleepassociated neuromodulatory inputs initiate cascades leading to non-tagged granule cell apoptosis (Yokoyama et al., 2011). Similar events may occur following other odor learning experiences, for example the fear conditioning protocols described above.

### **SUMMARY OF MAJOR EFFECTS OF SLEEP ON OLFACTORY CORTEX**

Combining the findings from these two paradigms leads to the suggestion that post-odor exposure slow-wave sleep contributes to changes across the olfactory system that contribute to both the strength and precision of odor memory (**Figure 3**). Slow-wave sleep associated piriform cortical sharp-waves

odors during waking. These strong, synchronous sharp-wave events help strengthen synaptic connections within odor-coding ensembles, as well as help shape OB granule cell survival in an odor-specific manner. Abbreviations: GL = glomerular layer, M = mitral cell layer, G = granule cell layer.

allow strengthening of associations within cortical ensembles encoding specific odors in a replay-like manner. Reducing this association or imposing noise during replay impairs the precision of the odor memory. Sharp-wave evoked piriform cortical activity also induces strong activation of cortical efferent targets, such as the OB and perhaps other monosynaptic targets (Courtiol and Wilson, 2014; Narikiyo et al., 2014), contributing to memory-associated changes in those structures.

#### **SLEEP, PATHOLOGY AND ODOR PERCEPTION**

Olfactory deficits are associated with a variety of disorders including, but not limited to Alzheimer's Disease (Murphy, 1999), Parkinson's disease (Doty, 2012), schizophrenia (Malaspina et al., 2012) and major depression (van Mill et al., 2010). All of these disorders are also associated with sleep disturbances such as insomnia and sleep fragmentation (Spiegelhalder et al., 2013). While originally thought of as side effects of the primary disorder, sleep disturbance is increasingly seen as integral component of many disorders. For example, specifically treating sleep disorders in individuals with major depression helps alleviate depressive symptoms (Sánchez-Ortuño and Edinger, 2012; Spiegelhalder et al., 2013). Furthermore, given the importance of sleep related memory consolidation and synaptic homeostasis as described here, it is easy to see how sleep disturbance could contribute to cognitive deficits.

Thus, we speculate that a contributing factor to the widespread occurrence of olfactory disorders, particularly odor identification, across diverse pathologies may be related to underlying sleep disorders. Memory and neural plasticity are integral not only to odor memory, but also to basic odor perception (Wilson and Stevenson, 2003; Wilson and Sullivan, 2011). As described here and elsewhere, sleep, especially slow-wave sleep, is now known to play an important role in that plasticity, including modulation of synaptic connectivity (Tsuno et al., 2008) and survival (Yokoyama et al., 2011) of OB neurons known to be critical for precise odor discrimination (Gheusi et al., 2000; Moreno et al., 2009). Furthermore, disruption of normal sleep-related activity within the olfactory cortex can impair the strength and accuracy of odor memory, leading to impaired odor-guided behavior (Barnes and Wilson, 2014). In addition, 24 h of sleep deprivation has been shown to impair odor identification in humans (Killgore and McBride, 2006). Thus, olfactory perception and memory may benefit from a good night's sleep.

### **ACKNOWLEDGMENTS**

Dylan C. Barnes was supported by a pre-doctoral fellowship from NIDCD (F31-DC012284). Donald A. Wilson was funded by grants from NIDCD (R01- DC03906) and NIA (R01-AG037693).

### **REFERENCES**


Silva, A. J. (2003). Molecular and cellular cognitive studies of the role of synaptic plasticity in memory. *J. Neurobiol.* 54, 224–237. doi: 10.1002/neu.10169


**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 March 2014; accepted: 03 April 2014; published online: 22 April 2014. Citation: Barnes DC and Wilson DA (2014) Sleep and olfactory cortical plasticity. Front. Behav. Neurosci. 8:134. doi: 10.3389/fnbeh.2014.00134*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Barnes and Wilson. 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*.

## Beta and gamma oscillatory activities associated with olfactory memory tasks: different rhythms for different functional networks?

#### *Claire Martin1 \* and Nadine Ravel <sup>2</sup> \**

*<sup>1</sup> Laboratory Imagerie et Modélisation en Neurobiologie et Cancérologie, CNRS UMR 8165, Université Paris Sud, Université Paris Diderot, Orsay, France <sup>2</sup> Team "Olfaction: Du codage à la mémoire," Centre de Recherche en Neurosciences de Lyon CNRS UMR 5292, INSERM U1028, Université Lyon 1, Lyon, France*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Daniel W. Wesson, Case Western Reserve University, USA Maria Luz Aylwin, University of Chile, Chile*

#### *\*Correspondence:*

*Claire Martin, Laboratory Imagerie et Modélisation en Neurobiologie et Cancérologie, CNRS UMR 8165, Université Paris Sud, Université Paris Diderot, Batiment 440, 91405 Orsay, France e-mail: martin@imnc.in2p3.fr; Nadine Ravel, Equipe "Olfaction: Du codage à la mémoire," Centre de Recherche en Neurosciences de Lyon, 50 Avenue Tony Garnier, 69366 Lyon Cedex 07, France e-mail: nravel@olfac.univ-lyon1.fr*

Olfactory processing in behaving animals, even at early stages, is inextricable from top down influences associated with odor perception. The anatomy of the olfactory network (olfactory bulb, piriform, and entorhinal cortices) and its unique direct access to the limbic system makes it particularly attractive to study how sensory processing could be modulated by learning and memory. Moreover, olfactory structures have been early reported to exhibit oscillatory population activities easy to capture through local field potential recordings. An attractive hypothesis is that neuronal oscillations would serve to "bind" distant structures to reach a unified and coherent perception. In relation to this hypothesis, we will assess the functional relevance of different types of oscillatory activity observed in the olfactory system of behaving animals. This review will focus primarily on two types of oscillatory activities: beta (15–40 Hz) and gamma (60–100 Hz). While gamma oscillations are dominant in the olfactory system in the absence of odorant, both beta and gamma rhythms have been reported to be modulated depending on the nature of the olfactory task. Studies from the authors of the present review and other groups brought evidence for a link between these oscillations and behavioral changes induced by olfactory learning. However, differences in studies led to divergent interpretations concerning the respective role of these oscillations in olfactory processing. Based on a critical reexamination of those data, we propose hypotheses on the functional involvement of beta and gamma oscillations for odor perception and memory.

**Keywords: beta and gamma oscillations, odor learning, behavior, olfactory bulb, piriform cortex**

#### **INTRODUCTION**

Among the functional particularities of the olfactory system, we wish to stress its privileged access to limbic structures and its predisposition to rhythmicity. In the absence of thalamic relay, the olfactory receptors are only two and three synapses distant from the cortical amygdala and the hippocampus respectively. This singularity could partly explain why olfactory experience has been reported to be so efficient to shape odor representations (Wilson and Stevenson, 2003; Davis, 2004). In adults, anatomic and functional plasticity related to odor learning occur at every step of the olfactory system. As early as in the olfactory mucosa, olfactory learning increases the number of sensory neurons specific to a trained odorant (Jones et al., 2008; Dias and Ressler, 2014). Studies carried out in the main olfactory bulb (MOB) and the piriform cortex (PCx) reported long-lasting modifications of neuronal activity and synaptic efficiency in various learning contexts (Barkai and Saar, 2001; Mouly et al., 2001; Mouly and Gervais, 2002; Martin et al., 2004b; Sevelinges et al., 2004; Mandairon and Linster, 2009; Restrepo et al., 2009; Wilson and Sullivan, 2011; Royet et al., 2013).

The olfactory system is also highly dynamic. On the one hand, odorant detection and coding are constrained by respiratory modulation through breathing. The sniff cycle controls the firing pattern of olfactory neurons in time and is suggested to be the functional time unit for odor processing (Buonviso et al., 2006; Kepecs et al., 2006; Wachowiak, 2011). On the other hand, odor processing has been associated with oscillations of the local field potential (LFP) both in insects (Perez-Orive et al., 2002) and mammals (Kay et al., 2009). Those signals reflect a weighted average of synchronized dendro-somatic components of neuronal processing within a neural population (Buzsáki et al., 2012). Because they underlie coincident activity, oscillations would favor temporal coordination of sensory information within brain areas and facilitation of its transfer across regions (Varela et al., 2001; Siegel et al., 2012). Accordingly, they are ideally suited to subserve memory processes such as encoding, consolidation and retrieval (Engel et al., 2001; Varela et al., 2001; Tallon-Baudry et al., 2004; Fell and Axmacher, 2011).

The present review will leave apart the respiratory modulation which has already been the object of several recent reviews (i.e., Buonviso et al., 2006; Kepecs et al., 2006; Scott, 2006; Wachowiak, 2011). The aim here is to synthesize our current knowledge about the conditions in which the other two main oscillatory rhythms linked to odor processing, namely beta (15–40 Hz) and gamma (60–90 Hz), are observed at the first stages of olfactory processing, the MOB and the PCx. A majority of studies designed to decipher odor coding have been performed in anesthetized animals. These studies have been essential for understanding the activity of single neurons in response to odorants both in the MOB and the PCx (Buonviso et al., 2003; Fletcher and Wilson, 2003; Litaudon et al., 2003; Chapuis and Wilson, 2012; Fukunaga et al., 2012). In the present review, we will focus exclusively on LFP recordings performed in awake, behaving animals. One benefit of chronic LFP recordings as compared to single unit recordings is the ability to follow the evolution of rhythmic activities across cerebral areas throughout training as a means of tracking learning-related changes. Because oscillatory activities are transient, their detection and precise description requires operant devices, in which the timing of odorant onset and offset can be precisely controlled. Comparing the relation between behavior and LFP oscillations in various conditions, we came to propose hypotheses on the functional involvement of beta and gamma oscillations in the context of odor processing. Far to be exhaustive, the scope of this review is to consider the respective putative role of these two oscillatory rhythms in odor coding and memory.

### **THE OLFACTORY SYSTEM AND ITS OSCILLATIONS**

More than any other sensory system the olfactory system has early been reported to be oscillatory (Adrian, 1942; Freeman and Schneider, 1982; Gray, 1994). This specificity is most probably due to two parameters, the nature of the stimulus and the organization of olfactory areas. First, odorant molecules are slow to reach the detector, compared to sound or light. They travel with nasal airflow, and do not reach simultaneously the different parts of the nasal cavity. Because odorant onset cannot be sharp, it most often fails to elicit evoked potential. Second, as it will be described below, the central olfactory relays (MOB and PCx) are tightly interconnected and host specific features; the mitralgranule dendrodendritic reciprocal synapses in the MOB, and a dense network of associational fibers in the PCx.

The olfactory sensory neurons present in the nasal cavity are the point where the odorant chemical information is transduced and transmitted to the brain (Zufall and Munger, 2001). All the olfactory sensory neurons that express the same molecular receptor converge onto a few glomeruli, well identified microdomains containing the first synapse of the olfactory information path (Zou et al., 2009). In the absence of thalamic relay, the MOB has been considered already as an associative structure where inhibition plays a major role. Olfactory signal that travels in the principal excitatory neurons, mitral and tufted (MT) cells, is gated at two levels within the structure. Surrounding glomeruli, juxtaglomerular cells include astrocytes and various types of neurons: excitatory external tufted cells, periglomerular cells and short axon cells. While periglomerular are GABAergic cells, short axon cells, which processes extend across several glomeruli, have two opposite actions by releasing both GABA and Dopamine (Liu et al., 2013). Deeper in the structure, the specific interaction between granules and MT cells via dendrodendritic reciprocal synapses is a key element for the large oscillatory activity displayed in the olfactory system. MT cells axons coalesce into the lateral olfactory tract and project to numerous areas termed as the olfactory cortex. Privileged targets of the MOB are the anterior olfactory nucleus and the anterior PCx (Haberly, 2001; Cleland and Linster, 2003; Hintiryan et al., 2012). MT cells also contact in a lesser extent, the posterior PCx, the lateral entorhinal cortex, the olfactory tubercle, the ventral tenia tecta and the anterior cortical complex of the amygdala.

The PCx is anatomically and functionally divided into two parts: a rostral region (anterior PCx) mostly connected to the other olfactory areas and a caudal region (posterior PCx) in connection with higher cognitive regions and characterized by dense associational connectivity (Haberly, 2001; Litaudon et al., 2003; Bekkers and Suzuki, 2013). Indeed, the anterior part of the PCx has strong bidirectional projections to the posterior part of the structure and to the MOB and the anterior olfactory nucleus. On the contrary, the posterior PCx has dense feedforward projections to numerous cortical and subcortical regions including high-order association areas, but lacks functional projections to the anterior PCx. In addition, feedback projections from the posterior PCx to the MOB are sparse. A noticeable point in the functional anatomy of the PCx is the presence of abundant associational connections, sparser in the anterior than in the posterior PCx (Hagiwara et al., 2012). Anatomo-functional connectivity of the PCx already suggests a key role of this structure in the elaboration of complex mechanisms of olfactory perception and memory (Gottfried, 2010; Wilson and Sullivan, 2011). Differences between anterior and posterior PCx would sustain complementary memory processes: as suggested in the literature, the anterior PCx would mediate odors matching such as generalization, discrimination or pattern completion (Wilson and Stevenson, 2003; Chapuis and Wilson, 2012) whereas the posterior PCx would rather link odor to previously learned non-olfactory information (Haberly, 2001).

The MOB and the PCx are densely interconnected. The lateral olfactory tract carries odor information from the MT cells to pyramidal cells. In turn, pyramidal cells send axon collaterals to the MOB. These glutamatergic fibers synapse almost exclusively on the different type of inhibitory interneurons contained in the MOB. They have a major inhibitory effect on the structure at two levels: the glomerulus via periglomerular cells, and the mitral cell via granule cells. Interestingly, the strongest drive is to deep and superficial short axon cells, the main source of inhibition onto granule and periglomerular cells (Boyd et al., 2012). Centrifugal projections to the MOB do not only originate from the PCx (Matsutani, 2010). Moreover, acetylcholine, norepinephrine, and serotonin projections modulate the activity of the MOB and the PCx (Linster and Hasselmo, 2001; Ennis et al., 2007; Rothermel et al., 2014). They can inhibit as well as disinhibit glomerular activity and MT cells.

Dense interconnections between and within olfactory structures are conducive to the emergence of oscillations. Presumably for this reason, the MOB and the PCx have been very early seen as good models to study rhythmic activities in the brain, and have been the target of pioneering electrical recording of brain oscillations (Adrian, 1942, 1950). Few years later, Lavin et al. (1959) performed the first recording of the electrical activity of the MOB in awake, unrestrained cats chronically implanted with electrodes. They reported bursts of activity related to the arousal of the animal. Since that time, numerous studies have recorded intracerebral LFPs in the MOB and the PCx in behaving animals. In these brain areas, even raw signals overtly display different types of oscillations that can be easily defined in sub-classes according to their frequency range and to the moment they occur in relation with external events.

Oscillatory activities in the olfactory system covers a broad frequency band comprised between 1 and 150 Hz. In the MOB and the PCx, three rhythms dominate. The larger and most obvious is linked to the respiration and occurs in a frequency range overlapping with the hippocampal theta rhythm (∼1–10 Hz) (Kay et al., 2009). In awake and motivated animals, regular bursts of fast oscillations, i.e., the gamma rhythm (∼60–90 Hz) are nested onto the respiratory modulation, occurring at the transition between inspiration and expiration (Buonviso et al., 2006; Manabe and Mori, 2013) (for an example see **Figures 1A**, **3A**). Odorant presentation most often elicits beta oscillations (∼15–40 Hz) of variable amplitude, but has also been associated with gamma increase restrained to the MOB (Beshel et al., 2007). Finally, in the MOB, some sporadic long lasting bursts of low frequency gamma (∼35–65 Hz) can occur during exploration (Kay, 2003). The boundaries of these rhythms are sometimes variable in frequency, depending on the animal species; as a consequence we will consider their functional condition of emergence rather than their absolute frequency.

#### **ODOR-EVOKED MODULATION OF LFPs, INFLUENCES ON BETA AND GAMMA BAND OSCILLATIONS**

The rhythm classically studied in the olfactory system has been the gamma band (∼60–90 Hz). In the absence of imposed odorant stimulation, in particular when animals freely explore their environment, the presence of gamma bursts, regularly nested at

respiratory modulation. **(B)** Raw LFP signal (0.1–300 Hz) and corresponding

the transition between inspiration and expiration characterizes the LFP in the MOB (**Figure 1A**) and at least in the anterior PCx. Beside the ubiquitous nature of gamma bursts in the olfactory system, the fact that gamma frequency has been recognized as the gold standard for sensory coding following the work of Wolf Singer in the visual system (Singer, 1993) probably drew the attention of the community on this frequency range. A detailed historical review about gamma oscillations in the olfactory system can be found in Rojas-Líbano and Kay (2008).

Gamma oscillations have been extensively analyzed in several regions of the olfactory system by Walter Freeman (Freeman, 1960; Freeman and Schneider, 1982; Eeckman and Freeman, 1990), who focused analyses on a rather large frequency range (20–90 Hz). As it has been reported in numerous studies, gamma bursts amplitude increases mostly when the animal is in an attentive state (Bressler, 1984; Eeckman and Freeman, 1990). This relationship to attention and motivation is easy to observe in the initial phase of any training. Indeed, our data suggest that when a rat is placed in a novel environment, the amplitude of spontaneous activity related gamma bursts increases very rapidly as the animal becomes more familiar with the arena and aware of what is going to happen (Martin et al., unpublished data). By recording EEG using a 64-electrodes grid at the surface of the MOB of small mammals, Freeman analyzed the spatial distribution of odor-induced gamma bursts amplitude, considering the brain as a chaotic system. In two of the most famous papers (Freeman and Schneider, 1982; Di Prisco and Freeman, 1985), authors examined this pattern during and after either an aversive or appetitive odor conditioning. They reported that spatio-temporal motifs emerging from gamma bursts analysis were relatively independent of odor presentation and more related to the significance of the odor. They proposed that gamma oscillatory activity modulations were

gamma band and an increase in the beta band power.

mostly related to the context and expectations of animals. These studies were pioneering in considering the MOB as a central element of a broader network underlying odor representations, rather than a passive odor relay. They were also the first evidence that odor processing at this early sensory stage already takes into account the context and the experience of the animal. Interestingly, as reported in several articles from the same group (Di Prisco and Freeman, 1985), the fact that spatial distribution of iso-amplitude gamma bursts is indeed modified by experience does not mean that odorant presentation increases gamma oscillations amplitude. On the contrary, the mean gamma power over the MOB remained stable or decreased by 15–35% during odorant sampling. In a series of experiments we performed at the beginning of the 2000's we observed that odorant sampling in the context of a Go/No-Go task was first associated with a strong and transient decrease of gamma oscillatory activities at the onset of odor. This power decrease was observed in naïve animals and was amplified as rats became experts for the odorants used in the task. Gamma depression was transient and most often followed by a light rebound effect before a return to baseline activity with calibrated regular gamma bursts nested on a slower respiration-related activity (Ravel et al., 2003; Martin et al., 2004b, 2006).

The decrease in gamma activity is most often replaced by the emergence of an activity in the beta band (15–40 Hz, centered around 25 Hz) that is never observed in the absence of odorant in normal condition (**Figure 1B**). This shift in the oscillatory dynamics between gamma and beta frequencies is characteristic of odorant sampling in awake animals and have been reported in numerous studies in the MOB (Gray and Skinner, 1988b; Martin et al., 2004b; Lowry and Kay, 2007; Lepousez and Lledo, 2013; Chery et al., 2014) and the PCx (Martin et al., 2006). Interestingly, beta oscillations elicited by odorant stimulations have been characterized for the first time in the dentate gyrus of the hippocampus where inhalation of toluene by the rat produced fast-wave bursts (Vanderwolf, 1992). This group conducted many studies in awake rats submitted to passive presentations of odorants that were supposed to be innately relevant or naturally aversive to the animals (urine, feces, toluene, predator odors. . . ) (Vanderwolf, 1992; Heale and Vanderwolf, 1994; Zibrowski and Vanderwolf, 1997; Chapman et al., 1998; Zibrowski et al., 1998; Vanderwolf and Zibrowski, 2001; Vanderwolf et al., 2002). They showed that a low frequency wave (around 20 Hz) was elicited by these odorants in a large network covering the MOB, the PCx, and limbic structures (entorhinal cortex, dentate gyrus). They also observed in the PCx that the repeated presentation of odorants (10–15 trials) leads to a gradual enhancement of beta wave amplitude that persists for several days (Vanderwolf and Zibrowski, 2001). In a subsequent study, Lowry and Kay (2007) have also found large beta activity during passive presentation of some specific odorants. However, they reported that all the odorants that showed significantly higher beta power were in a certain range of vapor pressure, between 1 and 120 mmHg. Interestingly, this range includes TMT, a component of fox feces, and toluene. In urethane anesthetized rats, similar observations were made that the molecular feature of odorants influenced the probability of emergence of beta oscillations (Cenier et al., 2008). Consequently, the reason why some odorants elicited higher beta power could be due to their volatility rather than their innate value. However these studies have been conducted using cotton swab presentations, a condition in which the odorant concentration and duration is more difficult to calibrate. Elevated odorant concentration may by itself be fearful and/or aversive for macrosmatic animals such as rodents. Indeed, in both sets of data, beta power enhancement induced by repeated exposure to the same series of odorants suggests that other processes than pure olfactory detection occur like some odor recognition and classification. Oscillatory activities related to naturally meaningful odorant molecules have also been found in the accessory olfactory bulb, which receives its sensory input from the vomeronasal organ. In awake female mice, male urine exposure significantly increases LFP power in frequencies overlapping with beta rhythm (ranging from 8 to 24 Hz) at the vicinity of the MT cells layer. Interestingly, following mating, the power of the LFP oscillations recorded under baseline conditions is dramatically increased across all frequency bands, suggesting that some form of synaptic plasticity has occurred (Binns and Brennan, 2005).

In conclusion, as presented in **Table 1**, most of the studies using odorant presentation have shown that they elicits a shift in frequency for the oscillatory activities recorded in the MOB and the PCx. Respiration locked gamma band activity (60–90 Hz) decreases and a slower beta oscillation (15–40 Hz) emerges. The reason why the pioneering studies led by Walter Freeman did not described such a systematic shift could be explained by the fact that recording were not performed in deep layers but at the surface of the cortex (Buffalo et al., 2011).

### **NETWORK SUSTAINING BETA AND GAMMA RHYTHMS IN THE OLFACTORY BULB**

Is the same network involved in the generation of beta and gamma oscillations? As presented above, odorant presentation often leads to a gamma decrease coupled to a beta increase suggesting that the two rhythms share a common cellular substrate. If this is easily noticeable during odor-reward learning tasks, some studies involving passive and non-reinforced odor presentations find both gamma and beta enhancement during odor sampling (Lowry and Kay, 2007; Carlson et al., 2014). Stimulus delivery, not constraint by a nose poke may not be continuous, which could explain this discrepancy. Indeed, Lowry and Kay (2007) mention that within single investigation period, bursts at each frequency actually alternate, as it is reported in urethane anesthetized rats. In our hands, passive odor presentations induced the same shift from gamma to beta rhythm (Chabaud et al., 2000). The conditions of generation of gamma oscillations in the MOB have been extensively studied by computational modeling and electrophysiology *in vivo* or *in vitro*. Much less data have been collected concerning beta oscillations.

Gamma bursts present during spontaneous activity are generated in the MOB under the influence of spontaneous input from the neuroreceptors located in the nasal cavity (Hernandez-Peon et al., 1960; Gray and Skinner, 1988a), and are then transmitted to the PCx (Bressler, 1984; Mori et al., 2013). Indeed, blocking descending centrifugal influences by cooling or local



infusion of anesthetic leads to an increase and not a decrease of MOB gamma bursts amplitude (Gray and Skinner, 1988a; Martin et al., 2006). In addition, the section of the lateral olfactory tract, which interrupts the transmission of the olfactory signal from the MOB to the PCx, selectively abolishes gamma bursts in the PCx (Neville and Haberly, 2003). On the contrary, pharmacological removal of centrifugal influences to the MOB abolishes beta oscillations in both the MOB and the PCx (Martin et al., 2006). Therefore, the major difference between the two rhythms is that gamma oscillations are generated locally within the MOB, whereas beta oscillations require intact bidirectional connectivity at least between the MOB and the PCx.

Within the MOB, gamma oscillations have been shown to be supported by the reciprocal synapse between mitral and granule cells (Nusser et al., 2001; Bathellier et al., 2006; Schoppa, 2006; Lagier et al., 2007; David et al., 2009). Computational models agree with the fact that these oscillations require an appropriate balance between excitation and inhibition, consistent with the mechanism proposed for gamma-band generation in other cortical areas (Cannon et al., 2014). Recent data have confirmed these mechanisms in awake mouse. They show that increasing the excitation/inhibition balance of MT cells via a decrease of GABAa receptors inhibition or local injection of glutamatergic agonists boosts gamma oscillatory power (Lepousez and Lledo, 2013). Consistently, selective MT cells drive using optogenetic technique causes a 5–10 fold increase of gamma oscillations without affecting other frequency bands. By scanning different frequency for light pulses (between 25 and 90 Hz), the authors show that the maximal response of the LFP occurs around 66 Hz, which corresponds to the dominant frequency of spontaneous gamma oscillations. Interestingly, the same GABAa receptors antagonist picrotoxin, which enhances gamma oscillations, leads to a reduction of beta oscillations power by more than 65%. On the contrary injection of MK801, an NMDA receptor antagonist, reduce gamma oscillations power without affecting beta oscillations (Lepousez and Lledo, 2013). Finally, modifications of beta and gamma oscillations observed in the presence of glutamate reuptake blockers argue for a role of glutamate spillover in constraining synaptic time constants. They suggest that MT cells glutamate release may locally change NMDA and AMPA mediated excitation (Martin et al., 2012; Lepousez and Lledo, 2013).

Taken together, these data imply that the two rhythms require the MOB network to exist. Even if they are both constrained by inhibition onto MT cells, this inhibition is likely to occur under different forms: either locally within the MOB network (granule and periglomerular cells) or remotely through centrifugal feedback. Because beta oscillations require intact connections between the MOB and at least the PCx, they are likely to emerge when the cerebral network engaged is broader. In the following part, we will examine which behavioral conditions are associated with either of the two oscillations, and how the distinction can have a functional readout in the context of learning.

### **ODOR LEARNING INDUCED MODIFICATIONS: DIFFERENT RHYTHMS FOR DIFFERENT LEARNING TASKS?**

As we discussed earlier, beta oscillatory activity has been observed in the olfactory system in naïve animals in response to toxic or aversive odorants (Zibrowski et al., 1998; Vanderwolf et al., 2002). When exposed to a neutral unfamiliar odorant, only weak beta oscillations are observed but their amplitude increases through training as soon as this odor starts to acquire a behavioral meaning for the animal (Ravel et al., 2003; Martin et al., 2004b). Such a learning-induced increase in beta power has been observed in several structures associated with odor processing (MOB, PCx, entorhinal cortex, and hippocampus) and for a variety of behavioral paradigms (see **Table 1**): Go/No-Go task (Ravel et al., 2003; Martin et al., 2004b, 2007; Gourévitch et al., 2010; Lepousez and Lledo, 2013), two-alternative choice task (Fuentes et al., 2008) and aversive learning (Chapuis et al., 2009). However, a few studies, with similar operant conditioning, report an odor evoked gamma increase instead of a change in beta activity (Beshel et al., 2007; Rosero and Aylwin, 2011).

It is easily arguable, when comparing the studies where LFPs have been recorded in olfactory structures in different operant tasks that the presence or not of substantial beta oscillations seems to be strongly dependent on the behavioral context of the task, and the cognitive strategy required to solve it (Kay et al., 2006). Go/No-Go and two-alternative choice have been the two main tools used to assess odor discrimination and learning in rodents. In the case of the Go/No-Go task, two odorants are delivered in a random order, one is positively rewarded (CS+; sucrose) and another is not rewarded or associated with a negative reinforcement (CS-; quinine). Initially, both odorants are neutral to the animals, and do not elicit any particular behavior. Over the course of training, animals learn to associate each odorant with the corresponding reward, and exhibit a differential behavior in response to the two odorant stimuli. Reaching the behavioral criterion for good performances takes several sessions, a duration that can vary with the difficulty of the task, which depends itself on the qualitative proximity of the odorants used.

In the Go/No-Go task, we have constantly found beta power increase during learning for both the CS+ and CS− (Martin et al., 2004a,b, 2006), raising the question of the link between this activity and the chemical feature of the odorant in one hand, or the odor meaning on the other hand. Oscillations in the olfactory system are triggered by odorant sampling and are likely to carry some aspects of odorants, as it has been demonstrated in anesthetized animals (Cenier et al., 2008). Indeed, by recording LFP signals from four different locations within the MOB, we showed that the main characteristics of beta oscillations (frequency and amplitude) are not homogeneous across the MOB, contrarily to gamma bursts recorded during spontaneous activity. Moreover, during learning, a stronger beta power is found in the posterior part of the structure (Martin et al., 2004b). Distinct odors evoke different amplitude levels of beta oscillations, irrespectively of the reward they are associated to. For a given animal, two different CS+ odorants can evoke distinct beta amplitude (Martin et al., 2004a), and a reversal procedure for an odorant pair (inverse learning contingencies) does not lead to the mirror image of the beta activity for each odor (Martin et al., 2007). Taken together, these data show that specificity of beta oscillations after learning would convey some feature of the odorant. However, it is likely that beta rhythm also reflects the odor signification acquired through learning.

Contrarily to the Go/No-Go, where only one odor is reinforced, the two-alternative choice task is symmetrically rewarded and a pellet is delivered for each correct response. This task, that seems more demanding for a rodent, is indeed often acquired slowly by the animals, and with a lower final performance (Friedrich, 2006; Slotnick, 2007). However, a recent study has shown that adjustment of parameters could allow to attain the same level of accuracy than for the Go/No-Go task in the same laps of time (Frederick et al., 2011). Besides the difficulty of the task, we can make the hypothesis that these two tasks involve different strategies and thus activate different brain circuits. In the two-alternative choice paradigm, odors can elicit high amplitude beta oscillations (10–30 Hz) and a significant decrease in the gamma band (70–100 Hz) (Fuentes et al., 2008). However, Beshel et al. (2007) using this task to compare successive odor pairs discriminations obtained different results. In this study, as expected, the animals are faster to reach the criterion for molecularly dissimilar odorants than for similar ones. Moreover, once animals are at the criterion for the discrimination, odor evoked gamma (60–85 Hz) power is very high for fine discriminations and almost absent for coarse ones. Within a given session, gamma power increases almost linearly across trial block but resets at the beginning of each session even if performances are improved. Interestingly, gamma increase is restricted to the MOB and does not propagate to the PCx. However, besides this gamma response, beta oscillations are also observed in three interconnected olfactory areas (MOB and anterior and posterior PCx) and only the beta band exhibits consistently elevated coherence levels between these three areas during odor sniffing across all odor pairs, classes (alcohols and ketones), and discrimination types (fine and coarse) (Kay and Beshel, 2010).

As mentioned earlier, the respiratory modulation influences odor processing. In anesthetized-tracheotomized animals, an airflow change is sufficient to change the relative power of beta and gamma frequency bursts (Courtiol et al., 2011). The direct relation between sniffing properties and oscillatory patterns during olfactory conditioning is still an open question in awake animals, who can actively tune their respiratory modulation. Still, we cannot exclude that sniffing properties affect oscillatory activities during odor sampling in the context of learning. A very recent paper showed that sniffing properties can be modulated by the context of the discrimination, i.e., which odor pair is presented during the test (Courtiol et al., 2014). However, the adjustment of sniffing parameters during odorant mixtures discrimination seems to rely largely on differences in sorption quality of the elements (Rojas-Líbano and Kay, 2012). Evolution of sniffing frequency and/or duration during the acquisition of the task is more likely to affect the intensity or the length of oscillations rather than its frequency.

We can rule out the hypothesis that only the concentration of odorant would turn beta into gamma in some conditions as it has been reported in anesthetized preparations (Neville and Haberly, 2003) and suggested in other studies (Rosero and Aylwin, 2011). Indeed, Go/No-Go and two-alternative choice tests were in this case performed in the same laboratory, using the same apparatus and the same odorant concentrations (Beshel et al., 2007; Martin et al., 2007). Beshel et al. (2007) show that the relatedness of the two odors involved in the discrimination increase gamma power in the MOB. However the relationship between the elevation of gamma power and the chemical proximity of the odorants, directly linked with the difficulty of the task, seems to be task specific. Indeed, as illustrated in **Figure 2**, the use of two chemically related odors heptanol and hexanol leads to different results in the two paradigms: beta band (15–40 Hz) power increase in the Go/No-Go paradigm (Martin et al., 2007) and enhanced gamma power (65–85 Hz) in the two-alternative choice task (Beshel et al., 2007).

How can we explain these discrepancies? One possibility could be that beta and gamma oscillations do correspond to distinct odor-related cognitive processing occurring at different stages of the training. Beta rhythm would be necessary, during the acquisition of the discrimination, to set up a broad network of distant brain structures required for specific rules and odor encoding. Indeed, one consistency across studies is that beta connects different brain areas (olfactory areas and beyond). Gamma could be required only in a subsequent stage of training, when odor discrimination has been learned by the rat, but its resolution is more challenging to reach the criterion. At this stage, the olfactory network has already been modified by learning and odor processing required for a fine sensory discrimination is rather supported by a local network and sustained by gamma band oscillations. This shift could allow a faster and more efficient treatment of odor, requiring less energy expenditure.

In the context of two-alternative choice task, beta rhythm is indeed present during the first learning sessions and disappears in subsequent sessions (Beshel et al., 2007; Kay and Beshel, 2010). In this condition, the number of trials required to reach the criterion for a given odor pair is enhanced compared to Go/No-Go for instance for the pair hexanol/heptanol, 350 trials (Beshel et al., 2007) vs. 72 trials (Martin et al., 2007) respectively (in both case, after rule transfer from previous odor pair discrimination). It is likely that the rat learns the discrimination between the odors during the first session, but that further sessions are necessary for the acquisition of the sensory-motor association. In the case where gamma oscillations are recorded in the context of a Go/No-Go task, the number of trials to criterion is also elevated (close to 1000) (Rosero and Aylwin, 2011).

The hypothesis that beta and gamma are two distinct mechanisms occurring at different time scale of the learning process is consistent with the idea proposed by Engel and Fries (2010) that beta band activity would dominate when top down input are the majority, whereas gamma band would rather reflects bottom-up local processing of sensory input. We will argue this hypothesis in the following part.

### **GAMMA AND BETA OSCILLATIONS, LOCAL vs. DISTAL NETWORKS?**

In agreement with the notion that in brain circuits beta rhythms coordinate long-range communication whereas faster gamma rhythms are more related to local intra structure processing (Kopell et al., 2000; Siegel et al., 2012; Cannon et al., 2014), the emergence in the MOB and PCx of these two rhythms in odordriven behavioral tasks is thus likely to sustain different network properties and processing.

In line with their implication in memory processes, beta oscillations have been found to sustain long range interactions. They have been recorded in many distant brain structures related to olfactory-driven behavior. Beyond the MOB and the PCx, they have been found in the lateral entorhinal cortex (Martin et al., 2004a; Igarashi et al., 2014), in the tubercle (Carlson et al., 2014), the hippocampus (Martin et al., 2007; Igarashi et al., 2014), in motor cortex M1 (Hermer-Vazquez et al., 2007) in

different parts of the prefrontal cortex (infralimbic and orbitofrontal cortex), the basolateral amygdala, and the insular cortex (Chapuis et al., 2009). Interestingly, using an olfactory discrimination Go/No-Go task, van Wingerden et al. (2010) reported an increase in gamma oscillations in the orbitofrontal cortex, where power was correlated with rat training and performance, as shown by Beshel et al. (2007) in the MOB. However, in the same study, the authors also observe some late beta oscillatory activity more associated with odorant sampling and very similar to what was reported in the MOB or the hippocampus (Martin et al., 2004b, 2007). This shift from gamma to beta rhythm observed in several areas associated with odorant sampling just before the animal makes a decision is in agreement with the hypothesis of a general beta synchronization across odor-processing areas that could be the signature of a functional network set up through learning.

Whereas gamma oscillations recorded in these structures are likely to reflect local processing and thus to have a distinct origin from that recorded in primary olfactory structures, we propose that beyond the MOB and PCx, beta oscillations would tag brain structures involved in the behavioral task that the animal is performing and form a unique representation of the odor in this task. Studies where multielectrode recordings have been performed have shown that beta increase occurred specifically in brain regions involved in the task performed by the animal. Indeed, in Martin et al. (2007), beta does not increase in the hippocampus for the first odor discrimination but for the transfer that is more likely to involve the structure. In the same way, after odor aversive conditioning, beta oscillations increase in insular and infralimbic cortices when the odor is ingested but not when it is delivered by airflow (Chapuis et al., 2009). By extension, we postulate that other brain areas, not yet studied, are capable of joining beta oscillatory network if involved in a given olfactory task.

One striking and stable characteristic of the emergence of beta rhythm in olfactory structures is that it is narrowly linked to behavioral output. Interestingly, beta power modulations seem to follow some aspects of the learning curve dynamics at least in the MOB and PCx. Indeed, beta gradually increases across training sessions (Martin et al., 2004b) and a strong beta oscillatory activity is observed just one or two sessions before the learning criterion was reached regardless of the time needed by the rat to acquire the discrimination (Ravel et al., 2003; Martin et al., 2004b, 2007). This is also true when odor learning is achieved in 1 day in the context of aversive odor learning (Chapuis et al., 2009) or following rule transfer to a new odor pair discrimination, which is done within one learning session (Martin et al., 2007). It is important to underline that beta power increases specifically for the learnt odor pair and falls down at the beginning of each new odor pair presentation. If training is continued post-criterion, beta power decreases as a function of overtraining (Martin et al., 2007). On the contrary, we observe that once the discrimination is achieved, if animals are put at rest and not tested for a long period (from a week to a month), beta oscillation emerges again stronger than ever. This latter effect is observed both after appetitive and aversive conditioning (Martin et al., 2004b; Chapuis et al., 2009). Finally, the emergence of beta oscillatory activity in a network seems to be highly specific of the conditioning procedure. Taking advantage of two different experimental situations suitable to induce a conditioned olfactory aversion we were able to demonstrate two different odor-evoked beta networks according to how the odor has been previously experienced by the animal (Chapuis et al., 2009).

All together these observations have therefore spawned the idea that beta rhythm might be necessary to bind together elements of a broad network and contribute to the build-up of memory. Indeed, we make the hypothesis that such a coordinated oscillatory activity could be used to tag preferentially inter or intra area connections that need to be reinforced to be efficiently and rapidly reactivated when the odor is further encountered. This idea has been strongly reinforced by a recent article that identified beta oscillations coupling between the entorhinal cortex and the hippocampus as a mechanism for the emergence of a functional circuit during encoding of odor associative memory (Igarashi et al., 2014).

The shift between gamma and beta when an odor is processed is likely due to a change in bidirectional connections between the MOB and other cerebral structures (**Figure 3**).

It has been reported for a long time that centrifugal influences are gating synaptic plasticity processes in both MOB and PCx (for a review see Mandairon and Linster, 2009). As expected, manipulations of centrifugal projections alter behavior. The lesion of efferent inputs to the MOB prevents the formation of odorreward associations, but has no effect on the resolution of spontaneous habituation experiment (Kiselycznyk et al., 2006). Besides, specific manipulation of noradrenergic action to the MOB impairs mice in discrimination learning in a Go/No-Go paradigm (Doucette et al., 2007). In this same task, MT cells undergo a profound change in odor responsiveness throughout a learning session (Doucette and Restrepo, 2008), that is dependent on centrifugal feedback (Restrepo et al., 2009). Both cortical feedback and neuromodulatory influences play a determinant role in the shift between gamma and beta rhythms as indicated by the impact of their local blockade on the odor-evoked activity in both the MOB and the PCx in a Go/No-Go task (Martin et al., 2006), reinforcing their link with expression of plasticity and memory processes. Interestingly, directed coherence analyses have shown that during odorant sampling, the MOB would drive odor related beta activity to the PCx (Boeijinga and Lopes da Silva, 1989; Kay and Beshel, 2010) and to the hippocampus (Chapman et al., 1998; Gourévitch et al., 2010), carrying relevant information in the

bottom up direction instead of the reverse. In contrast, it could be the opposite during memory consolidation since in a slowwave sleep-like state induced under anesthesia, the functional link based on slow waves LFP recordings (*<*15 Hz) is in the direction of the hippocampus to the PCx (Wilson and Yan, 2010). That beta rhythm could play a role in memory consolidation during sleep remains an open question.

The ultimate argument for a causal link between emergence of oscillatory activities and improvement of behavioral performance would be to degrade beta/gamma oscillatory dynamic in the network and observe behavioral impairment. Few studies have been conducted in MOB mammals that mainly addressed gamma activity. Local injection of low doses of picrotoxin, a GABAa receptor antagonist were reported to enhance gamma oscillations and also led to behavioral modifications: mice displayed an increased odorant sampling time, and their performances were selectively altered in the case of a fine odor discrimination in a Go/No-Go paradigm (Lepousez and Lledo, 2013). Nusser et al. (2001) used a transgenic mouse model, in which GABAa receptors were disrupted specifically on granule cells, i.e., those cells where centrifugal feedback targets and reported gamma oscillations power was enhanced in the MOB. Behavioral testing concluded that whereas mice seemed to perform better on a simple odor identification task, they were impaired on a mixture discrimination test. Interestingly, these two studies show that network modifications that lead to gamma band increase also result in behavioral impairment. Using pharmacological blockade, we have reported that inactivation of feedback projections abolish beta oscillations and conversely increases gamma power (Martin et al., 2006). Consequently, we postulate that those modifications that increase gamma oscillations and also lead to beta weakening impair behavior.

### **CONCLUSION**

In this review, we focused on two different oscillatory rhythms beta (15–40 Hz) and gamma (60–100 Hz) that have been associated with olfactory stimulus processing. We propose that gamma activity is associated with the resting state of a network limited to the first two steps of the olfactory system (MOB and anterior part of the PCx). As reported, this basic activity could be modulated in power during learning according to some experimental conditions but rather reflects the involvement of a limited local network under the control of higher cortical feedback and neuromodulators. In behaving animals, as soon as an odor is processed, this local coordination is disrupted and replaced by a lower frequency oscillation in the beta range (15–40 Hz). Most of the data reported in this review lead to the hypothesis that beta activity is the signature of a larger network including not only olfactory sensory areas but also each structure involved in the processing of the odorant stimulus, which could differ according to the behavioral situation. As stated in the present article, a decent amount of data is in favor of a strong correlation between beta oscillation modulation in power and learning-induced changes, in both rats and mice. Beta rhythm frequency is well suited for long range interactions (Kopell et al., 2000; Von Stein and Sarnthein, 2000) and thus for sustaining memory processes. The presence of

**FIGURE 3 | Schematic illustration of hypotheses for the generation of beta (15–35 Hz) and gamma (60–90 Hz) rhythms in the olfactory bulb and the piriform cortex in awake behaving animals.** Example of raw LFP traces recorded **(A)** in the MOB in the absence of olfactory stimulation and **(B)** in MOB and two different regions of the PCx before learning and **(C)** after learning during a conditioned discrimination paradigm (Go/No-Go). **a–c**: Corresponding schematic representation of MOB and PCx interconnected networks and centrifugal modulation. The level of neuromodulation is represented by the red arrow on the left, the level of cortical feedback by green arrow on the right. **(A)** Spontaneous activity: in the absence of olfactory stimulation. Observe the regular theta respiratory modulation (around 2 Hz) and the associated bursts of gamma activity (60–90 Hz). See also in **(B)** and **(C)** the portion just preceding the odorant sampling (green square area) how the gamma bursts decrease in the posterior part of the PCx compare to the MOB and anterior part of PCx. **a**: In the absence of olfactory stimulation, the level of activation in both networks is weak and variable and both structures are dominated by theta and gamma activity. Gamma activity is transmitted from the MOB to the PCx. **(B)** Before learning: during odorant

sampling, occurrence of gamma bursts is reduced but recovered after the animal has left the odor port. **b**: During this phase, a population of mitral cells of the MOB becomes active, this input is transmitted to a corresponding population of pyramidal cells. Both neuromodulatory and cortical feedback are exerted on the networks. However, no real coordination is set up in the network. **(C)** During training: In addition to a strong and sustained decrease in gamma activity, a clear beta oscillation is observed in the MOB and two regions of PCx associated to odorant sampling. **c**: During this phase, we propose that both assemblies of active mitral cells and pyramidal cells reinforce their connections. The result could be a more efficient and rapid transfer of olfactory information. This coordination is under the influence of both cortical feedback and neuromodulatory fibers as suggested by the results we observed with lidocaine inactivation of the peduncle (Martin et al., 2006). Once synaptic contacts are established, if the training is maintained to get over training, the amplitude of beta oscillatory activity decreases. On the contrary, if the animal is left in his home cage for a long interval without training and tested again, both structures exhibit a very strong beta oscillatory activity.

beta oscillatory rhythm within and between neuronal networks would optimize information processing, representing a framework for neuronal synchronization. By this mean odor coding would be more efficient and temporal simultaneity would favor hebbian mechanism of plasticity (Cassenaer and Laurent, 2007). However, we still lack evidence to disambiguate whether beta oscillations are instrumental for processes like spike timing plasticity in the network or if on the contrary they are just reflecting these changes. Nevertheless, we propose the idea that mapping such oscillatory activities in a neural network could be a good way to assess learning-induced brain plasticity at least in the context of odor-guided tasks. Recently, beta oscillations have also been used as a tool to reveal impaired network activity preceding behavioral dysfunctions (Wesson et al., 2011) and evaluate the impact of a treatment to enhance clearance of beta-amyloid protein in a mouse model of Alzheimer disease (Cramer et al., 2012). Providing experimental evidence to support a causal link between oscillatory binding and inter area synchronization will be one of the main goal for the future.

#### **ACKNOWLEDGMENT**

This review has received the financial support of the excellence network LABEX Cortex.

#### **REFERENCES**


of a respiratory cycle. *Eur. J. Neurosci.* 17, 1811–1819. doi: 10.1046/j.1460- 9568.2003.02619.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: 14 April 2014; accepted: 28 May 2014; published online: 23 June 2014. Citation: Martin C and Ravel N (2014) Beta and gamma oscillatory activities associated with olfactory memory tasks: different rhythms for different functional networks? Front. Behav. Neurosci. 8:218. doi: 10.3389/fnbeh.2014.00218*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Martin and Ravel. 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.*

## Context-driven activation of odor representations in the absence of olfactory stimuli in the olfactory bulb and piriform cortex

#### *Nathalie Mandairon1 \*, Florence Kermen1†, Caroline Charpentier 1†, Joelle Sacquet 1, Christiane Linster 2† and Anne Didier 1†*

*<sup>1</sup> Centre de Recherche en Neurosciences de Lyon, UMR CNRS 5292 INSERM 1028, Université Lyon1, Lyon, France <sup>2</sup> Computational Physiology Lab, Neurobiology and Behavior, Cornell University, Ithaca, NY, USA*

#### *Edited by:*

*Donald A. Wilson, New York University School of Medicine, USA*

#### *Reviewed by:*

*Frederic Levy, Institut National de la Recherche Agronomique, France Leslie M. Kay, The University of Chicago, USA*

#### *\*Correspondence:*

*Nathalie Mandairon, Centre de Recherche en Neurosciences de Lyon, UMR CNRS 5292 INSERM 1028, Université Lyon1, 50 Avenue Tony Garnier, F-69007 Lyon, France e-mail: nathalie.mandairon@ olfac.univ-lyon1.fr*

*†These authors have contributed equally to this work.*

Sensory neural activity is highly context dependent and shaped by experience and expectation. In the olfactory bulb (OB), the first cerebral relay of olfactory processing, responses to odorants are shaped by previous experiences including contextual information thanks to strong feedback connections. In the present experiment, mice were conditioned to associate an odorant with a visual context and were then exposed to the visual context alone. We found that the visual context alone elicited exploration of the odor port similar to that elicited by the stimulus when it was initially presented. In the OB, the visual context alone elicited a neural activation pattern, assessed by mapping the expression of the immediate early gene zif268 (egr-1) that was highly similar to that evoked by the conditioned odorant, but not other odorants. This OB activation was processed by olfactory network as it was transmitted to the piriform cortex. Interestingly, a novel context abolished neural and behavioral responses. In addition, the neural representation in response to the context was dependent on top-down inputs, suggesting that context-dependent representation is initiated in cortex. Modeling of the experimental data suggests that odor representations are stored in cortical networks, reactivated by the context and activate bulbar representations. Activation of the OB and the associated behavioral response in the absence of physical stimulus showed that mice are capable of internal representations of sensory stimuli. The similarity of activation patterns induced by imaged and the corresponding physical stimulus, triggered only by the relevant context provides evidence for an odor-specific internal representation.

**Keywords: olfactory bulb, Zif268, cell mapping, conditioning, visual context, modeling**

### **INTRODUCTION**

Sensory neural activity is highly context dependent and shaped by experience and expectation. For example, throughout the olfactory system, neural responses to odors are shaped by behavioral relevance of the odor (Kay and Laurent, 1999; Martin et al., 2004; Doucette and Restrepo, 2008; Doucette et al., 2011; Wilson and Sullivan, 2011), by previous experience (Buonviso and Chaput, 2000; Moreno et al., 2009; Wilson, 2009; Chaudhury et al., 2010) and task difficulty (Mandairon et al., 2006; Li et al., 2008). Changes in neural responses to odors can be seen as early as in the olfactory bulb (OB), the target of sensory neurons (Freeman and Schneider, 1982; Mandairon and Linster, 2009). Contextual information is presumably shaped by previous experience and expectation and mediated to first order sensory structures by feedback projections from higher brain areas. The OB is an ideal target structure for the integration of sensory and contextual information because it receives direct inputs from sensory neurons, without thalamic detour, as well as a massive inputs from higher order brain areas such as noradrenergic and cholinergic nuclei, amygdala, piriform cortex, enthorinal cortex (Shipley and Ennis, 1996). Inspired by paradigms used in human imagery experiments which showed that a visual stimulus previously associated with an odorant is able to activate primary olfactory cortical regions (Gottfried et al., 2002, 2004) in the absence of olfactory stimulation, we here tested if association of an odorant with a visual context in mice would allow the visual context alone to elicit a behavioral responses usually associated with an olfactory stimulus as well as neural activation in olfactory pathways. In the OB, odor quality is represented by distributed patterns of activity in both the glomerular and granule cell layers. Each odor is represented by a unique pattern of relative activity across the OB, as visualized by 2DG (Johnson and Leon, 2007) or immediate early gene (IEG) mapping (Inaki et al., 2002) as well as partial visualization using optical methods. In addition to these spatial activity patterns, odors also evoke unique temporal patterns across the OB, as well as stimulus induced oscillations and synchronization (Kay et al., 2009). Because activity in the granule cell layer is susceptible to activity dependent plasticity, is highly odor and experience specific, we here chose relative activation patterns of granule cells as a measure for odor representations in the OB. We find that in mice which had been presented with an odorant repeatedly in the same visual context, the visual context alone elicited a behavioral response similar to that elicited by the stimulus when it was initially presented. In the OB, the visual context alone elicited a neural activation pattern, assessed by mapping the expression of the immediate early gene zif268 (egr-1) that was highly correlated with that elicited by the associated odorant, but not other odorants. Both behavioral and neural activation was not elicited by a novel context and both dependent on intact feedback to the OB from higher brain areas. A computational model of the OB and cortex which incorporated known features of the interactions between these two areas showed that experimentally described plasticity in projections from cortex to bulb, paired with "context neurons" previously used in models of hippocampal processing (Hasselmo and Wyble, 1997) sufficed to reproduce the observed experimental results. We conclude that in rodents, neural representation of an odorant in primary sensory areas can be elicited in its absence by exposure to the context to which the odorant was previously associated. This further suggests that rodents can build internal representation of the olfactory stimulus.

### **METHODS**

#### **SUBJECTS**

Sixty adult male C57BL/6J mice (8 weeks old, Charles River, L'Arbresles, France) were used in accordance with the European Community Council Directive of November 24, 1986 (86/609/EEC). Mice were kept in standard mouse cages with full access to food and water. Experimental group contained 5 to 13 animals.

#### **EXPERIMENTAL SET UP**

All behavioral experiments were conducted in individual training cages (20 × 27 cm) with visual cues differing in shape, color and pattern added to the outside walls of the cage. The cage lids were pierced in their center and a 3-cm diameter tube was pushed inside the cage through this hole. The tubes were transparent and had 3-mm holes in their bottom through which odor diffused. This system allowed the introduction of the odor without opening the cages during the experiment (**Figure 1A**).

#### **BEHAVIORAL TRAINING**

Thirty minutes per day during 10 days, mice were placed in the training cage (**Figure 1A**). Five minutes after their introduction, a non-odorized swab or an odorized swab was introduced into the tube.

the granule cell layer of the OB.

For odorant presentation, the swab was impregnated with 60μL of pure +Limonene (Sigma-Aldrich, purity 97%) or pure Decanal (Fluka, purity ≥95%).

#### **BEHAVIORAL TEST**

On day 11, mice were placed again into the training cage or novel context cage. After 5 min, the non-odorized or odorized swab introduced, depending on the group. The amount of time that the mice investigated the tube was manually recorded during 10 min after the introduction of the stimulus (**Figure 1A**). Investigation was defined as active exploration within 1 cm around the odor port.

#### **DATA ANALYSIS**

All mice were included in the analyses. Results are expressed as mean ± s.e.m. For behavioral data, Kruskall–Wallis (multiple comparisons) Mann–Whitney (two-group comparisons) tests were applied. For Zif268 expression maps, data showed normal distribution and between-group differences were assessed using ANOVA followed by Fisher *post-hoc* tests for pair comparisons or unilateral *t*-tests when appropriate (Systat software). The level of significance was set to 0.05.

#### **Zif268 EXPRESSION MAPPING**

One hour after the test session, animals were killed by intracardiac perfusion (under deep anesthesia, Pentobarbital 3.64 mg/kg) and brains were sectioned using a cryostat. Zif268 immunochemistry (Mandairon et al., 2008) was performed for each animal on serial coronal sections (sampling interval = 70μm). Zif268-positive cells were counted automatically in the granule cell layer of the OB using mapping software (Mercator, Explora Nova, La Rochelle, France; **Figure 1B**) coupled to a Zeiss microscope. The cell counts were conducted by experimenters who were blind to the experimental condition of the mice. The number of labeled profiles was divided by the surface of the region of interest to yield the total densities of labeled cells. Maps of Zif268-positive cells were constructed as previously described (Mandairon et al., 2006). Briefly, the granule cell layer was divided into 36 sectors of 10◦. The number of labeled cells/μm<sup>2</sup> was calculated for each sector and measurements were then merged into arrays of 10◦ × 70-μm bins yielding a 2-D map of the granule cell layer. Arrays were averaged within each group, and a colored image plot of the data was constructed. To compare odor and context evoked activation maps, we calculated the pairwise overlap between maps (Python scripts associated to the Scipy library). Following standard procedures, maps were first threshold to keep the 30% highest values (percentile). This threshold was set as best fitted to the clusters delineated by visual inspection of the Zif268 expression maps. We analyzed the similarities between maps by counting the number of overlapping pixels and calculating a percentage of overlap. Between groups comparisons were done using *t*-tests for comparison of proportions (Mandairon et al., 2006; Sultan et al., 2011).

Zif268 labeling was analyzed in the anterior piriform cortex (layer II) in about eight sections per mice (distributed between 1.18 and 2.46 mm anterior to Bregma, Paxinos Atlas). The boundary between the anterior and posterior cortex was located at the level of the anterior commissure.

#### **CANNULATION**

Mice were anesthetized (100 mg/kg ketamine and 6 mg/kg xylazine, i.p.) and implanted as described earlier (Kermen et al., 2011) into both olfactory peduncles at the following coordinates with respect to bregma: AP = +2.4 mm; ML = ± 0.75 mm; DV = −3 mm. Following surgery, mice were allowed to recover for 10 days before beginning the training. Lidocaine (Sigma) (2%, 1μl/side) was freshly prepared and infused into the each medial peduncle. Behavioral testing began 10 min after lidocaine administration was completed.

### **COMPUTATIONAL MODELING**

We used a computational model of olfactory sensory neurons, OB and piriform cortex (**Figure 2A**). The individual elements of this model have been described in detail before (Linster and Cleland, 2001, 2002, 2004; Linster et al., 2003) and have been adapted. Synaptic plasticity between pyramidal cells and granule cells, as described experimentally (Gao and Strowbridge, 2009) is new to this model as is the introduction of "context" neurons (Hasselmo and Wyble, 1997). Context neurons here represent the context of the behavioral experiments, or a combination of features of the cage in which odor exposure happened. These context neurons, after training, can drive activity in olfactory cortex, creating context dependent responses as described experimentally (Calu et al., 2007). To enable context learning, synaptic plasticity was also introduced between context neurons and pyramidal cells.

In a model simulating 100 OSNs, 100 mitral (Mi), granule (Gr), periglomerular (PG), 100 pyramidal (Pyr) cells and 10 context neurons, synapses between mitral and pyramidal cells were created randomly with each mitral cell projecting to any pyramidal cell with an equal probability of PMit-Pyr = 0*.*1. Intra-cortical connections and interneurons were omitted from this model. Pyramidal cells projected back to randomly chosen granule cells with an overall connection probability of 0.4 and initially weak synapses. 10 context neurons activated by behavioral context connected to pyramidal cells in an all to all fashion with initially weak synapses. Information flow in the model is both feedforward (OSNs activate mitral cells, mitral cells activate granule cells

synapses exhibiting synaptic plasticity to pyramidal cells. Pyramidal cell (pyr) outputs project back to OB granule cells with initially weak synapses exhibiting plasticity (lot, lateral olfactory tract). **(B)** Neural activity patterns in response to stimulation with a randomly chosen odorant before training. Activity is color-coded from low (blue) to high (red). **(C)** Membrane potential and action potentials of 10 neurons in response to a 200 ms stimulation.

and cortical pyramidal cells) and feedback (context cells activate pyramidal cells and pyramidal cells activate granule cells). Representative spatial activation patterns and neural firing patterns are depicted in **Figures 2B,C**.

In the simulations presented here (**Figure 2B**), simulated exposure to an odorant in a specific context induced activity dependent plasticity of synapses from pyramidal to granule and from context to pyramidal cells. Synaptic strengths were first calculated from the parameters given in **Table 1**, and responses to simulated odorants were obtained. To simulate perceptual learning in response to repeated exposure to an odorant, synapses between pyramidal and granule and between context and pyramidal cells underwent synaptic potentiation:

$$\boldsymbol{\nu}\_{ij}^{tmined} = \boldsymbol{\nu}\_{ij}^{\text{naive}} + \eta \ast \sum\_{i,j=0}^{N} \mathbf{x}\_i \mathbf{x}\_j$$

where *wij* is the synaptic strength between the presynaptic pyramidal cell (context cell) *j* and the postsynaptic granule cell (pyramidal cell) *i*, η is the rate of potentiation and *xj* and *xi* are the total numbers of spikes emitted by the pre and postsynaptic cells during the period of odor stimulation.

#### **RESULTS**

#### **CONTEXTUAL PRIMING PRODUCES SIMILAR PATTERNS OF RESPONSIVENESS IN THE OB AS ODOR STIMULATION**

To test for contextual activation of OB neurons, mice were trained to associate an odorant (+limonene) with a visual context (visual cues added to a transparent cage, context A) by being introduced into the cage for 30 min per day during 10 consecutive days (**Figure 3A**). On day 11 (test), one group of mice was placed in the same visual context with the same odorant (Lim-Lim, ctxA), a second group of mice was placed in the same visual context with no odorant (Lim-NO, ctxA) and a third group was placed

#### **Table 1 | Model parameters.**


in novel visual context (context B) with no odorant (Lim-NOctxB) (**Figure 3A**). Upon testing, the group exposed to the visual context only (Lim-NO) investigated the odor delivery device significantly more than both other groups (Lim-Lim and Lim-NOctxB) (group effect *p* = 0*.*008; Lim-Lim vs. Lim-NO *p* = 0*.*011, Lim-NO vs. Lim-NO-ctxB *p* = 0*.*004) (**Figure 3B**), suggesting an expectation of the stimulus. This expectation was specific to the context previously associated with the odor stimulus, because the increased sniffing did not occur in the novel context (Lim-No ctx B) (**Figure 3B**). At the neural level, the overall Zif268-positive cell density in the granule cell layer of the OB did not vary among the three groups [*F*(2*,* 12) = 1*.*16, *p* = 0*.*34] (**Figure 3C**); however, the similarity between patterns evoked by the training context only (Lim-NO) and odor stimulus (Lim-Lim) was high (72% overlap), whereas overlap between patterns evoked by the training context (Lim-NO) and a new context (Lim-NO-ctxB) was significantly less (47% overlap, *p <* 0*.*0001) (**Figures 3D,E**). These findings show that the training context previously associated with the odorant induced a pattern of activity mimicking odor specific activity as well as a significant behavioral response.

#### **CONTEXT-EVOKED NEURAL ACTIVATION PATTERNS ARE SPECIFIC TO THE ASSOCIATED ODOR**

We confirmed this result with an additional set of mice in which context A was associated with a different odorant (decanal, **Figure 4A**). Results were similar to those found with the previous odorant (**Figure 3**). Indeed, presentation of the context alone (Dec-NO) induced a significantly increased investigation of the odor delivery device (*p* = 0*.*01) (**Figure 4B**), suggesting an expectation of the odorant. As with limonene, overall levels of Zif268 expression were not significantly different between the two groups (**Figure 4C**). The patterns of neural activation in the OB evoked by the context alone (Dec-NO) were very similar to those evoked by the odor in the same context (Dec-Dec) (62% overlap) (**Figures 4D,E**). In contrast, the patterns evoked in mice who associated decanal with context A were significantly different from those evoked by context A in mice who associated context A with limonene [44% overlap between Dec-NO and Lim-NO and 43% between Dec-Dec and Lim-Lim compared to the 62% overlap between Dec-Dec and Dec-NO (*p <* 0*.*0004) or compared to the 72% overlap between Lim-NO and Lim-Lim (*p <* 0*.*0001)], showing that the activity pattern evoked by context A alone was specific to the odor associated with that context by the mice.

#### **CONTEXT-EVOKED NEURAL ACTIVATION IN ANTERIOR PIRIFORM CORTEX IS SPECIFIC TO THE ASSOCIATED CONTEXT**

In anterior piriform cortex, to which the OB projects (Shipley and Adamek, 1984; Haberly, 1998), we found that Zif268-positive cell density did not differ between groups smelling the odor and groups exposed to the associated context alone. Because no specific spatial activity pattern is associated with an odorant in cortex (Isaacson, 2010) we did not analyze the overlap between maps but rather compared to a naive control group not exposed to an odorant during training and testing (NO-NO) and found more activation in the limonene-context associated groups compared to naive [*F*(2*,* 6) = 13*.*70, *p* = 0*.*006, NO-NO vs. Lim-Lim *p* = 0*.*002; NO-NO vs. Lim-NO *p* = 0*.*019] (**Figure 5A**). The

**FIGURE 3 | Context-evoked behavioral and OB neural responses in the absence of olfactory stimulus. (A)** Mice were trained to associate a visual context to an odorant (+limonene: Lim) 30 min per day during 10 days and tested on day 11. The day of the test, the same odorant as during training ("Lim-Lim" group), or an empty swab ("Lim-NO" group, NO: no odor) was introduced in the cage. In a third experimental group, mice were trained with +limonene, but tested in a different context without the odorant ("Lim-NO-ctxB" group). Mice were sacrificed (S) 1 h after the test. **(B)** Investigation time of the odor delivery device. Mice trained during 10 days with Lim and tested with no odor (Lim-NO) showed an investigation time of the odor port superior to the mice trained and tested with the same odorant

same results were obtained in the group of mice presented with decanal during context association [*F*(2*,* 5) = 18*.*07, *p* = 0*.*005, NO-NO vs. Dec-Dec *p* = 0*.*004; NO-NO vs. Dec-NO *p* = 0*.*004] (**Figure 5B**).

#### **CONTEXT-ACTIVATED BEHAVIORAL AND NEURAL RESPONSES DEPEND ON CENTRAL INPUTS TO THE OLFACTORY BULB**

Information about a learned context is likely to be transmitted as top-down information to the OB (Gilbert and Sigman, (Lim-Lim). Moreover, when the visual context changed the day of the test, the increase of investigation time was no longer observed. **(C)** The density of Zif268-positive cells was similar between all groups. **(D)** The overlap between maps of Lim-Lim and Lim-NO was high (72%) indicating the similarity between those two maps. Overlap significantly decreased when Lim-NO was compared to Lim-NO-ctxB. (∗*p <* 0*.*05; ∗∗*p <* 0*.*005). **(E)** Normalized 2-D maps of the density of Zif268-positive cells in the granule cell layer of Lim-Lim and Lim-NO groups showed a similar pattern of Zif268 expression. Cell density in this figure and in following figures is color-coded from low (blue) to high (red). When the context was changed the day of the test, this pattern was altered (Lim-NO-ctxB).

2007). Mice with surgically implanted cannula in the medial olfactory peduncle were exposed to the visual context and odorant during 1 h daily for 10 days. On day 11, mice were exposed to the context alone. In this experimental group, lidocaine (or saline) was infused into the olfactory peduncle in such a manner as to decrease central inputs to the OB without affecting olfactory input (Martin et al., 2006) (**Figure 6A**). In response to the context, lidocaine-induced decrease of peduncle activity (Lim-NO-Lido) led to a significant decrease in context-induced

investigation time compared to saline-infused animals (Lim-NO-Sal) (*p <* 0*.*0005) (**Figure 6B**). The overall level of Zif268 expression in the OB was not significantly different between animals infused with lidocaine or saline (*t*-test, *p* = 0*.*11) (**Figure 6C**), suggesting no direct effect of lidocaine in the OB through diffusion from the infusion site. However, the distribution of Zif268 positive cells was deeply altered in Lim-NO-Lido compared to Lim-NO-Sal (47% overlap between Lim-NO-Lido and Lim-NO-Sal vs. 72% overlap between Lim-NO and Lim-Lim, *p <* 0*.*0001) (**Figures 6D,E**). This result shows that context evoked activity in the OB depends on functional feedback inputs from other brain centers.

(Dec-NO) showed an investigation time of the odor port superior to the mice

#### **COMPUTATIONAL MODELING OF THE NEURAL CIRCUITS UNDERLYING CONTEXT-DRIVEN OB ACTIVATION**

Dec-NO groups showed similar patterns of Zif268 expression. (∗*p <* 0*.*05).

We then used a well-described computational model of the olfactory system (Linster et al., 2007; Linster and Cleland, 2009), to which we added an abstract set of context neurons encoding the visual features of the context (**Figures 2A–C**) (Gao and Strowbridge, 2009) and a connection between context neurons and cortical pyramidal cells. The behavioral association between the context and odor was simulated by stimulating olfactory sensory neurons with an "odor" while simultaneously stimulating the context neurons with a "context;" during the formation of this association the activity-dependent learning rule is turned

on (**Figures 2B,C**). After training, stimulation with odorant and the context drove granule cell activation (**Figure 7A**, Odor+ctxA, corresponding to Lim-Lim in experimental data). Presence of the trained context only (**Figure 7A**; No-odor+ctxA, corresponding to Lim-NO in experimental data) stimulated a very similar pattern of activity (*r*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*96) (**Figure 7A**). When a novel context was presented to the network in the absence of odorant, pyramidal cells were not activated and did not shape granule cell activation: the activation pattern of granule cells was nonspecific and the overlap with the representation triggered by the trained context was low (*r*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*14, **Figure 7A**, No-odor-ctxb corresponding to NO-Lim-ctx B in experimental data). A set of ten simulations with novel networks and randomly chosen odorants confirmed these results to be independent of the choice of odorant. There was a statistically significant difference between the two sets of overlap [*F*(1*,* 17) = 816*.*16; *p <* 0*.*001], showing that while the trained context evokes granule cell activity resembling that in response to the odorant, the novel context does not (**Figure 7B**). We observed a significant effect of group on the discharge rates of granule cells (no-odor naive, odor+ctxA, no-odor+ctxA; [*F*(2*,* 27) = 21*.*285, *p <* 0*.*001] (**Figure 7C**); with individual significant differences between the naive network and both trained networks (*p <* 0*.*001) but not between the trained networks (*p >* 0*.*2) (**Figure 7C**), as observed experimentally (**Figure 3**). Cortical pyramidal cells exhibited significantly higher spike rates in response to the trained odor (**Figure 7D**, Odor\_ctxA) or the context alone (**Figure 7D**, N-Odor\_ctx) than in response to no odor in an untrained network (**Figure 7D**, No-Odor\_naive), as shown experimentally (**Figure 5**). When cortical feedback inactivation were simulated by decreasing the synaptic weights from pyramidal neurons to granule cells, the overlap between granule cell activation patterns in response to context alone with intact feedback (No−odor+ctxA) and context alone with "lesioned" feedback (No-odor+ctxA-no feedback) was very low (*r*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*24 in the example in **Figures 7E,F**).

#### **DISCUSSION**

In this study, in order to trigger an internal odor representation, we developed a behavioral paradigm which allowed inducing odor expectations by exposing animals to a context previously associated to an odorant. The increase in investigation time directed to the empty odor source the day of testing strongly supports the view that animals actually expected, and even searched for the odorant.

Using this original paradigm, we demonstrated a patterned, odor and context specific activation of olfactory cortices by contextual information in the absence of a physical stimulus. This result is in accordance with a previous fMRI study in humans, in which the same brain regions were activated by imagining visual (Halpern and Zatorre, 1999), auditory (Kosslyn et al., 2001), or olfactory stimuli (Bensafi et al., 2007) and actually viewing, hearing or smelling them. In humans, the previous studies showed that imagining odors activated olfactory structures as the piriform cortex, left insula and amygdala. Here, we showed that not only the piriform cortex but also the OB were activated during context evoked odor expectation. This finding is reminiscent of the "search image" revealed by EEG recordings in rabbits (Freeman, 1983). The "search image" was defined by Freeman as a large-scale pattern of strengthened connections (synaptic template) that could serve to represent an expected stimulus even if it is not present. We mapped Zif268 expression in granule cells because bulbar patterned expressions are odor-specific and replicable across individuals in these cells (Inaki et al., 2002; Mandairon et al., 2006; Busto et al., 2009) and hence allowed us to compare odor and context evoked activation patterns. We found that the context induced activation pattern in the OB was highly similar to the one observed after odor stimulation and was specific to each odorant tested. Taken together, these findings strongly suggested that odor expectation induced an internal neural representation of the odorant in the OB which was context and odor specific and resulted from the specific odor-context association the animal was exposed to.

Both behavioral and OB neural context-driven responses depend on intact centrifugal projections to the OB. When we blocked top-down fibers by infusing mice with an anesthetic (Lidocaine) in the medial part of the olfactory peduncle 10 min before testing we eliminated the context driven behavioral and neural responses. The olfactory peduncle contains fibers projecting from the rest of the brain to the OB including the anterior olfactory nucleus, glutamatergic fibers from the piriform and entorhinal cortices and cortical amygdaloid nuclei (Haberly and Price, 1977, 1978), cholinergic neurons from basal forebrain

or noradrenergic (McLean and Shipley, 1991), serotoninergic (McLean and Shipley, 1987) neurons as well as sparse projections from the hypothalamus. We assume that a high percentage of these incoming fibers are blocked by our lidocaine injection; we can therefore not speculate on which centrifugal fibers contribute to our observations. From a computational point of view, any secondary or tertiary olfactory structures receiving odor and context information and projecting back the OB could perform this function. Overall, we observed that top-down inactivation resulted in a decreased investigation time compared to non-injected mice, as if they were not expecting the odor. This result is consistent with data showing that blockade of central inputs to the OB using an infusion of Lidocaine in the olfactory peduncle reduced the amplitude of odor-induced oscillatory beta responses (Martin et al., 2006) which are involved in odor associative learning and in anticipation of odor stimuli (Kay et al., 2009). The simulations presented showed that a combined model of OB, olfactory cortex and context neurons, including experimentally described synaptic plasticity between pyramidal and granule cells (Gao and Strowbridge, 2009) can reproduce the described priming of

granule cell activation by visual context. The model suggests that during the learning of the odor-context association, information flows from bulb to cortex and from cortex to bulb, and that activity dependent plasticity in both pathways can suffice to support context-driven bulbar activity in the absence of olfactory stimulation. The association with context needs to be performed outside the OB to prevent changes in odor processing in the absence of the learned context. Piriform cortex is only one of many candidate structures to perform this function; it was chosen for these simulations because (a) cortical pyramidal cells project onto bulbar granule cells with synapses that have been shown to undergo activity dependent plasticity (Gao and Strowbridge, 2009) and (b) because neural activity in piriform cortex has been shown to be modulated by behavioral context (Calu et al., 2007). In theory, any brain area receiving odor inputs from the OB and projecting back to OB granule cells and capable of forming an association with context information would yield equivalent results. Our simulations results are not dependent in a specific structure being implemented.

The data presented, together with the computational results suggest that both odor and contextual information shape rather than create OB neural responses. Odor quality, or the expectation thereof, would therefore be encoded in the spatiotemporal patterns of bulbar activity (Kay et al., 2009; Mori and Sakano, 2011) and can be elicited by sensory or central inputs.

### **ACKNOWLEDGMENTS**

This work was supported by the CNRS and Lyon1 University.

### **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: 31 January 2014; accepted: 04 April 2014; published online: 29 April 2014. Citation: Mandairon N, Kermen F, Charpentier C, Sacquet J, Linster C and Didier A (2014) Context-driven activation of odor representations in the absence of olfactory stimuli in the olfactory bulb and piriform cortex. Front. Behav. Neurosci. 8:138. doi: 10.3389/fnbeh.2014.00138*

*This article was submitted to the journal Frontiers in Behavioral Neuroscience. Copyright © 2014 Mandairon, Kermen, Charpentier, Sacquet, Linster and Didier. 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.*

## Comparative chemosensory cognition

### *Alan Gelperin\**

*Department of Molecular Biology, Princeton Neuroscience Institute, Princeton University, Princeton, NJ, USA \*Correspondence: gelperin@princeton.edu*

#### *Edited by:*

*Anne-Marie Mouly, Centre de Recherche en Neurosciences de Lyon, France*

*Reviewed by:*

*Frederic Levy, Institut National de la Recherche Agronomique, France*

**Keywords: comparative cognition, olfactory learning, invertebrate consciousness, higher order conditioning, genetic model systems**

Don Griffin, one of the foremost students of comparative cognition, offered the following definition of cognition: "The term cognition is ordinarily taken to mean information processing in human and nonhuman central nervous systems that often leads to choices and decisions" (Griffin and Speck, 2004). Clearly we can study cognition by analyzing animal information processing systems and the manner in which behavioral choices and decision making are altered by experience. The use of olfactory information for guiding a wide range of basic biological decisions is ubiquitous in animals, including humans (Gelperin, 2010). Chemosensory processing, and particularly olfactory information processing, is a particularly attractive modality within which to seek comparative insights into cognitive processes underlying learning and memory.

The advent of modern molecular and genetic tools for selectively modifying and perturbing functionally identified groups of neurons, particularly optogenetic methods, has led to a focus on chemosensory processing in a limited number of species for which genetic tools are well developed, including *C. elegans* (Glater et al., 2014) and *D. melanogaster* (Wilson, 2013), among others. Looking more broadly for instances of chemosensory learning reveals a remarkable diversity of cognitive plasticity from ants to zebrafish. The renewed interest in comparative cognition (Shettleworth, 2012) plus glimmerings of recognition that a voluminous and highly folded cerebral cortex is not required for implementation of higher order logic operations (Watanabe et al., 2008) is producing renewed interest in exploration of chemosensory processing and learning in an increasingly diverse set of species.

Another impetus for comparative studies is the quest to identify the minimum essential neural circuit that can implement the synaptic operations required for higher-order learning and decision making involved in choosing among alternative neural outputs based on previously learned weighting factors, as in reinforcement learning (Wilson et al., 2014). Modeling studies tightly constrained by demonstrated or plausible neuronal and synaptic properties are essential to this enterprise, having been used to show for example that five neurons with suitable dynamics and plausible synaptic plasticity functions can demonstrate the Kamin blocking effect, previously shown behaviorally in *Limax* (Goel and Gelperin, 2006). In the face of overwhelming evidence that neuronal properties and synaptic communication are essentially similar from ants to zebrafish (Llinas, 2008), the hope is that gaining new insights into the minimum essential circuitry for higher order information processing functions may be more straightforward in compact nervous systems comprised of fewer computing elements compared to the mammalian central nervous system. The use of transgenic *Drosophila* mutants as models of human disease (Chen and Crowther, 2012; van Alphen and van Swinderen, 2013) is also fueling this enterprise.

A remarkable example of olfactory learning in a compact brain is the demonstration of food imprinting and prenatal chemosensory conditioning in the predatory mite *Neoseiulus* (Peralta Quesada and Schausberger, 2012), whose entire brain of 10,325 cells occupies a single synganglion, containing a prominent olfactory lobe, measuring only 100 by 65 microns (van Wijk et al., 2006). A more familiar example of a compact brain with excellent olfactory abilities is that of the no-see-um, e.g., *Culicoides sonorensis*, best described as a flying nose with gonads and biting mouthparts measuring only a few hundred microns in overall body size. The success of these tiny cognitive engines is shown by the fact that the family containing the nosee-ums, the Ceratopogonidae, contains over 4000 species worldwide. No-see-um olfactory learning is likely as members of a species of larger size, also with obligate blood-sucking by females, are know to modulate their olfactory responses due to associative learning (Sanford et al., 2013).

Consider the noble nematode, *C. elegans*, possessed of 302 neurons, only a subset of which is needed to implement associative chemosensory conditioning modulated by dopamine (Mersha et al., 2013). Dopamine is a familiar neuromodulator in the mammalian CNS implicated in the processing of signals and synapses mediating reward and expectation (Kobayashi and Schultz, 2014). *C. elegans* possesses multiple learning types modifying its information processing in non-chemosensory modalities (Ardiel and Rankin, 2010). This suggests that a search for higher order conditioning such as second order conditioning and blocking using chemosensory stimuli might be fruitful.

A detailed prescription for comparing cognitive abilities across wide phyletic boundaries was provided by Bullock (1993), who emphasized the need for quantitative measurement of higher order cognitive operations by invertebrate brains. He also highlighted the need to frame experimental questions assessing complex information processing with special regard to the neuroethological context within which experimental questions are most effectively and insightfully asked of the animal subject. The use of olfactory stimuli to ask questions about cognitive aspects of higher order learning has been particularly fruitful for terrestrial gastropods in general (Gelperin, 2013) and *Limax maximus* in particular (Watanabe et al., 2008). *Limax* has been shown to be capable of a variety of higher order learned logic operations on olfactory stimuli, most recently involving the ability to learn the association between olfactory stimuli and the reward of access to water after the animal was subjected to rapid and severe dehydration, a normal stressor for terrestrial slugs. Olfaction is the dominant sensory modality for distance perception in terrestrial mollusks and a brain region with unique neuronal architectures and dynamics, the procerebral lobe, is devoted to learning about odors (Matsuo et al., 2011). Like the mushroom bodies of insects and the vertical lobes of octopi, identification of the sensory inputs to these distinctive central information processing centers in invertebrates can help guide the choice of sensory modalities within which to look for complex cognitive logic operations.

The cognitive abilities of honeybees in the chemosensory domain include not only higher order chemosensory learning, but also the construction and use of cognitive maps incorporating multiple domains of sensory input during their construction, particularly visual and chemosensory inputs (Gould, 1990). In 1973 Karl von Frisch shared the Nobel Prize for his pioneering work on the multimodal communication methods used by bees in communicating and receiving information on potential food sources in the environment surrounding the hive. Several generations of von Frisch's scientific descendants have continued his neuroethological tradition in selecting behavioral questions arising from the use of natural odor, taste, color, and shape information contained in the floral fountains exploited by bees. Processing of these proximate cues is interlaced with the processing of visual and magnetic information for navigation from hive to food source. Thus, chemosensory cues are only one aspect of a multimodal processing system that is capable of concept learning, a form of higher order learning that relies on the relationships between objects (e.g., same/different, left/right, above/below) rather than the specific properties of individual stimuli. These studies used the training protocols of delayed matching to sample and delayed non-matching to sample (Reinhard et al., 2004), protocols that are widely used for probing cognitive aspects of stimulus representation in primates. A recent summary of the cognitive architecture of the honeybee brain, which contains 950,000 neurons packaged in a volume of 1 mm3, identifies more than 17 discrete categories of computations as demonstrated abilities of the honeybee brain, including a requirement for neuronal circuitry tasked with assigning value to stimulus configurations, a value assignment that changes with experience (Menzel, 2001). Another mammalian cognitive parallel is the demonstration that honeybees consolidate a novel navigation memory during sleep (Menzel and Giurfa, 2001).

Chemosensory cognition in *Drosophila* has only recently come under experimental examination, although the study of olfactory learning is well developed (Young et al., 2011; Beyaert et al., 2012), augmented by experimentally useful modeling work (Wessnitzer et al., 2012). The seminal initial work on *Drosophila* learning was done by molecular biologists (Quinn et al., 1974) so work focused initially on development of high throughput screens for memory mutants rather than identification of natural stimulus configurations and contingencies that would provide natural experimental approaches to asking cognitive questions (Tomchik and Davis, 2013). Nonetheless in more recent work behavioral and neurobiological aspects of sleep, dopaminergic arousal, aggression, selective attention and courtship in *Drosophila* have been identified. Both aggression and courtship in *Drosophila* have critical chemosensory components involving both olfactory and gustatory receptors that allow male flies to distinguish between potential mates and competing conspecific males (Hollis and Kawecki, 2014). The maintenance of a range of cognitive tasks in male *Drosophila*, including but not limited to olfactory learning ability, was significantly reduced after 100 generations of enforced monogamy (Anderson and Adolphs, 2014).

Another example of a genetic model system that has engendered work on olfactory cognition is the zebrafish, *Danio* (Friedrich, 2013). Odors are known to be critical cues for guiding a number of behaviors in fish, including homing, reproduction, ingestion and social and avoidance behaviors. Aqueous odors can also participate in eliciting food-aversion conditioning, accompanied by induction of the immediate early response gene Egr-1 in gustatory areas of the zebrafish brain (Boyer et al., 2013). High throughput methods have been developed for assessment of the effects of genetic and pharmacological manipulations on visual responses of larval zebrafish behavior. Extension of these methods to olfactory conditioning will allow assessment of olfactory learning in a quantitative and unbiased fashion and facilitate the search for higher order learning about odors.

The power of genetic tools has promoted a focus on a limited number of animal species, known as genetic model systems, that enable use of the socalled genetic toolbox. This trend is augmented by multiple demonstrations that human disease genes and their downstream effects can be usefully studied in some of these genetic model systems, particularly *Drosophila*. Work on comparative cognition among vertebrate species provides more and more examples of cognitive skills among non-hominids, particularly but not exclusively elephants, birds and cetaceans. An interesting generalization from this effort is to look for further examples of higher order learning and other cognitive skills among invertebrates. For example, the debate on whether *Octopus* is conscious has already begun (Mather, 2008), fueled by descriptions of unique personality types among captive specimens (Mather and Kuba, 2013).

The unique relationship between neural circuits processing olfactory memory and circuits controlling emotions may provide yet another unique vantage point for a comparative approach to chemosensory cognition (Anderson and Adolphs, 2014). The proposal stresses the common features and evolutionary advantages of modes of behavior that are commonly identified as outward manifestations of emotions. An example is provided by negative or positive odor conditioning in male *Drosophila*, where flies show a conditioned positive place preference to an odor previously paired with mating with a virgin female (Shohat-Ophir et al., 2012). Mating responses to virgin females require neuropeptide Y, also involved in reward learning to ethanol. The concatenation of these findings suggests that these learned responses involve a rewarding internal state, a potential substrate for positive emotions. The idea that certain brain states can be rewarding, as indexed by their ability to promote the increased probability of discrete behaviors with which they are associated, is also supported by the finding that direct electrical stimulation of discrete brain areas in the cerebral ganglion of the terrestrial snail *Helix* can lead to significant increases in the probability of occurrence of the behaviors yoked to the application of brain stimulation (Balaban and Chase, 1989; Balaban and Maksimova, 1993). The use of implanted electrodes in minimally restrained animals (Cooke and Gelperin, 2001) increases the range of possible analyses of the rewarding effects of direct brain stimulation.

If the concept of consciousness loses its uniquely mammalian brand, the study of olfactory information processing may be the most general and fruitful approach to the study of comparative cognition, including consciousness, in the 96% of animal species in the Invertebrata. Some invertebrates could have a brain state representing a precursor of consciousness, as recently suggested for *Drosophila* (van Swinderen, 2005). Thus, understanding chemosensory cognition could help unravel some of the mechanisms underlying an evolutionary precursor to hominid consciousness.

This effort was presaged by Vince Dethier in his paper on Microscopic Brains, which ends with the following: "Perhaps these insects are little machines in a deep sleep, but looking at their rigidly armored bodies, their staring eyes, and their mute performances, one cannot help at times wondering if there is anyone inside" (Dethier, 1964).

#### **REFERENCES**


*elegans*. *Learn. Mem.* 17, 191–201. doi: 10.1101/lm.960510


*Rev. Neurosci.* 36, 217–241. doi: 10.1146/annurevneuro-062111-150533

Young, J. M., Wessnitzer, J., Armstrong, J. D., and Webb, B. (2011). Elemental and non-elemental olfactory learning in *Drosophila*. *Neurobiol. Learn. Mem.* 96, 339–352. doi: 10.1016/j.nlm.2011.06.009

**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: 13 April 2014; accepted: 07 May 2014; published online: 23 May 2014.*

*Citation: Gelperin A (2014) Comparative chemosensory cognition. Front. Behav. Neurosci. 8:190. doi: 10.3389/ fnbeh.2014.00190*

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

*Copyright © 2014 Gelperin. 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.*