NEUROVISION: NEURAL BASES OF BINOCULAR VISION AND COORDINATION AND THEIR IMPLICATIONS IN VISUAL TRAINING PROGRAMS

EDITED BY: Olivier A. Coubard PUBLISHED IN: Frontiers in Integrative 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.*

ISSN 1664-8714 ISBN 978-2-88919-655-5 DOI 10.3389/978-2-88919-655-5

## 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

## **NEUROVISION: NEURAL BASES OF BINOCULAR VISION AND COORDINATION AND THEIR IMPLICATIONS IN VISUAL TRAINING PROGRAMS**

Topic Editor:

**Olivier A. Coubard,** The Neuropsychological Laboratory, CNS-Fed, France

Binocular vision is achieved by five neurovisual systems originating in the retina but varying in their destination within the brain. Two systems have been widely studied: the retino-tectal or retino-collicular route, which subserves an expedient and raw estimate of the visual scene through the magnocellular pathway, and the retino-occipital or retino-cortical route, which allows slower but refined analysis of the visual scene through the parvocellular pathway. But there also exist further neurovisual systems: the retino-hypothalamic, retino-pretectal, and accessory optic systems, which play a crucial role in vision though they are less understood. The retino-pretectal pathway projecting onto the pretectum is critical for the pupillary or photomotor reflex. The retino-hypothalamic pathway projecting onto the suprachiasmatic nucleus regulates numerous behavioral and biological functions as well as circadian rhythms. The accessory optic system targeting terminal lateral, medial and dorsal nuclei through the paraoptic fasciculus plays a role in head and gaze orientation as well as slow movements. Taken together, these neurovisual systems involve 60% of brain activity, thus highlighting the importance of vision in the functioning and regulation of the central nervous system.

But vision is first and foremost action, which makes perception impossible without movement. Binocular coordination is a prerequisite for binocular fusion of the object of interest on the two foveas, thus ensuring visual perception. The retino-collicular pathway is sufficient to elicit reflexive eye movements with short latencies. Thanks to its motor neurons, the superior colliculus activates premotor neurons, which themselves activate motor neurons of the oculomotor, trochlear and abducens nuclei. At a higher level, a cascade of neural mechanisms participates in the control of decisional eye movements. The superior colliculus is controlled by the substancia nigra pars reticulata, which is itself gated by subcortical structures such as the dorsal striatum. The superior colliculus is also inhibited by the dorsolateral prefrontal cortex through a direct prefrontotectal tract. Cortical areas are crucial for the triggering of eye movements: the frontal eye field, supplementary eye field, and parietal eye field. Finally the cerebellum maintains accuracy.

The focus of the present research topic, entitled Neural bases of binocular vision and coordination and their implications in visual training programs, is to review the most recent findings in brain imaging and neurophysiology of binocular vision and coordination in humans and animals with frontally-placed eyes. The emphasis is put on studies that enable transfer of knowledge toward visual training programs targeting visual field defects (e.g., hemianopia) and binocular functional disorders (e.g., amblyopia).

**Citation:** Coubard, O. A., ed. (2015). Neurovision: Neural Bases of Binocular Vision and Coordination and their Implications in Visual Training Programs. Lausanne: Frontiers Media. doi: 10.3389/978-2-88919-655-5

Image by darrenwhi / bigstockphoto.com http://www.bigstockphoto.com/image-349230/stock-photo-retro-eyes

# Table of Contents

*06 Editorial: Neural bases of binocular vision and coordination and their implications in visual training programs* Olivier A. Coubard

## **PANORAMA**

*10 Educating the blind brain: a panorama of neural bases of vision and of training programs in organic neurovisual deficits*

Olivier A. Coubard, Marika Urbanski, Clémence Bourlon and Marie Gaumet

## **PART I – NEURAL BASES OF EYE MOVEMENTS AND BINOCULAR COORDINATION**


Magali Seassau, Christophe Loic Gérard, Emmanuel Bui-Quoc and Maria Pia Bucci

*80 LATER models of neural decision behavior in choice tasks* Imran Noorani

## **PART II – NEURAL BASES OF VISUAL PERCEPTION AND BINOCULAR VISION**

*90 Cortical and white matter mapping in the visual system-more than meets the eye: on the importance of functional imaging to understand visual system pathologies*

Noa Raz and Netta Levin


## **PART III – VISUAL TRAINING PROGRAMS IN ORGANIC DEFICITS AND THEIR NEURAL BASES**


Tara L. Alvarez, Raj Jaswal, Suril Gohel and Bharat B. Biswal

*232 Frontal eye field, where art thou? Anatomy, function, and non-invasive manipulation of frontal regions involved in eye movements and associated cognitive operations*

Marine Vernet, Romain Quentin, Lorena Chanes, Andres Mitsumasu and Antoni Valero-Cabré

*258 What saccadic eye movements tell us about TMS-induced neuromodulation of the DLPFC and mood changes: a pilot study in bipolar disorders*

Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel, David Szekely, Thierry Bougerol and Christian Marendaz

## Editorial: Neural bases of binocular vision and coordination and their implications in visual training programs

Olivier A. Coubard\*

*The Neuropsychological Laboratory, CNS-Fed, Paris, France*

Keywords: binocular vision, eye movements, visual pathways, neurovisual disorders, visual rehabilitation

## Opening

To see or not to see? That is the question of this research topic. How do human beings see not with their eyes but with their brain, which lies in a moving body, itself evolving in a continuously changing environment? What and how do humans see in the context of a particular task at a given moment? How do humans cease to see after some damage in the brain or neurofunctional disorder? And how may the basic science of eye movements and vision help to develop efficient visual training programs?

The present research topic, entitled Neural bases of binocular vision and coordination and their implications in visual training programs, aims at putting forward our knowledge of the neural underpinnings of vision in its motor, sensory, cognitive, emotional, and vegetative expressions. It does not target an exhaustive collection of what we know in the field of visual neurosciences. For that purpose, the reader may refer to the volume sets by Chalupa and Werner (2003). Rather, this research topic focuses on the latest findings on the neural aspects of eye movements and visual perception that directly help to understand and improve visual training programs in pathological conditions. Such disorders follow damages of the cerebral visual pathways (e.g., hemianopia) or refer to syndromes hitherto believed to be peripheral but in which neurophysiology and brain imaging are uncovering neural correlates or causes (e.g., amblyopia).

Edited and reviewed by: *Sidney A. Simon, Duke University, USA*

\*Correspondence: *Olivier A. Coubard, olivier.coubard@cns-fed.com*

Received: *15 July 2015* Accepted: *23 July 2015* Published: *13 August 2015*

#### Citation:

*Coubard OA (2015) Editorial: Neural bases of binocular vision and coordination and their implications in visual training programs. Front. Integr. Neurosci. 9:47. doi: 10.3389/fnint.2015.00047*

The research topic is divided into three parts respectively dedicated to eye movements, visual perception, and visual training programs, each having six chapters, and starts with an overview. In the introductory chapter, Coubard, Urbanski, Bourlon and Gaumet (Coubard et al., 2014) remind the reader of the importance of action in visual processing before describing the cascade of physiological mechanisms underlying eye movements, followed by a description of the five main neurovisual systems. After an overview of pathological conditions causing not eye but brain blindness—also called neurovisual disorders—the authors end by describing the disciplines of visual rehabilitation.

## Part I—Neural Bases of Eye Movements and Binocular Coordination

The important role of fixational eye movements in binocular vision is developed by Otero-Millan, Macknik and Martinez-Conde (Otero-Millan et al., 2014). The authors review the past 50 years of research on binocular coordination of each fixation eye movement type and how pathologies of binocular vision impact such coordination. Percheron, François and Pouget (Percheron et al., 2015) challenge histological, anatomical, and functional definitions of the frontal eye field (FEF) in primate species—a critical area in eye movement control subjected to passionate debates. This review will certainly help redefining new methodologies and thinking paradigms for future research. The role of prefrontal neurons, particularly of the dorsolateral prefrontal cortex (DLPFC), is reviewed by Funahashi (2014). The author describes prefrontal neuron pre-saccadic and post-saccadic memory-related activities, and offers insightful perspectives about attentional control processes such as working memory updating and the control of performance. The impact of visual pathology like amblyopia on saccadic behavior is addressed in an original study by Perdziak, Witkowska, Gryncewicz, Przekoracka-Krawczyk and Ober (Perdziak et al., 2014). Using a delayed saccade task, the authors show that the amblyopic eye is slower to respond than the non-amblyopic one. In dyslexia, eye movement behavior and visual search ability are explored by Seassau, Gérard, Bui Quoc and Bucci (Seassau et al., 2014). The authors show that oculomotor control in reading and binocular coordination of saccades in visual search are impaired in dyslexic children as compared to typical readers. Finally, Noorani (2014) reviews 34 years of research on the LATER (linear approach to threshold with ergodic rate) physiological model. The author points out that saccadic behavior, even in advanced decision tasks such as antisaccade and sequential decisions, can be accurately predicted by the model.

## Part II—Neural Bases of Visual Perception and Binocular Vision

To examine visual pathologies, Raz and Levin (2014) recommend using functional magnetic resonance imaging (fMRI) and Diffusion TensorImaging (DTI). Illustrating damages at different levels of the visual pathways, the authors show the advantages of combining the topological and hodological approaches to understand visual phenomena. Strabismus has until recently been considered only as a peripheral visual pathology. Bui Quoc and Milleret (2014) show how vision science and clinical ophthalmology are now advanced enough to comprehend possible cerebral origins of strabismus, whose complexity mirrors that of its polymorphism and various expressions. Amblyopia, deriving from abnormal visual experience (strabismus, anisometropia), has also been preferentially examined through the ophthalmological lens. Joly and Frankó (2014) review neuroscientific discoveries in amblyopia and the way functional brain imaging helps to map deficits in 3D vision networks within occipital-parietal and occipital-temporal pathways. Visual scene perception has been subjected to major advances for the last three decades. Kauffmann, Ramanoël and Peyrin (Kauffmann et al., 2014) describe that scenes are processed in terms of low/high spatial frequencies activating occipital areas in relation to peripheral/foveal representations and specific areas within the occipital-temporal cortex. In an original study using a saccade choice task, Boucart, Calais, Lenoble, Moroni and Pasquier (Boucart et al., 2014) show that participants suffering from posterior cortical atrophy compared to participants with Alzheimer's disease are impaired in detecting targets in a scene and do not benefit from contextual information. Vision is also a major issue in psychiatry. Notredame, Pins, Deneve and Jardri (Notredame et al., 2014) review psychophysical and neurophysiological findings about visual perception in schizophrenia. Through illusory paradigms and Bayesian inference framework, they explain perceptual changes in schizophrenia providing insightful directions to study its neural mechanisms.

## Part III—Visual Training Programs in Organic Deficits and Their Neural Bases

Taub, Mark and Uswatte (Taub et al., 2014) examine the possibility that techniques of Constraint Induced (CI) therapy previously employed in the rehabilitation of movement may prove useful for the treatment of visual deficits. The strengthening of diminished neural connections is developed to account for CI-induced neurovisual improvements. Homonymous hemianopia (HH) is a symmetric loss of vision in both eyes following unilateral retrochiasmatic lesion. Perez and Chokron (2014) review the different visual rehabilitation techniques in HH and the potentials of blindsight, the implicit visual function enabling hemianopic patients of performance escaping their consciousness. Describing the neural mechanisms of plasticity and reorganization after brain damage and those induced by visual training programs is a challenging issue, addressed by Urbanski, Coubard and Bourlon (Urbanski et al., 2014), suggesting that future research cannot dispense with studying the cerebral changes in relationship to behavioral improvements. In an original study, Alvarez, Jaswal, Gohel and Biswal (Alvarez et al., 2014) measured brain changes induced by vergence training in participants with convergence insufficiency. They demonstrate that the vergence peak velocity correlates with blood-oxygen-level dependent (BOLD) signal changes within the FEF, posterior parietal cortex and cerebellar vermis. Non-invasive manipulation techniques have become interesting tools to modulate vision in humans. Vernet, Quentin, Chanes, Mitsumasu and Valero-Cabré (Vernet et al., 2014) review the malleability of regions involved in eye movements and visual perception, particularly the FEF, and extend to the therapeutic interest of these modulatory techniques. In psychiatry, a study by Beynel, Chauvin, Guyader, Harquel, Szekely, Bougerol and Marendaz (Beynel et al., 2014) explores the effects of DLPFC intermittent theta burst stimulation in participants suffering from bipolar disorder. They show that neuromodulation improves performance in the antisaccade task, which also correlates with mood changes.

## Closure

In a world where the human factor introduces unlimited degrees of complexity, science has an ability to focus the attention of scientists and clinicians on restricted objects and questions. It does it over time, worldwide, disregarding daily life contingencies, keeping ideas strong and emotions hot. To see or not to see was the question of the research topic (see Section Opening). To address it, 59 authors from six countries— France, Israel, Japan, Poland, UK and USA—have generously accepted to contribute so as to put forth our knowledge of the neural underpinnings of eye movements and visual perception. Thirty-six other scientists from sixteen countries— Brazil, Canada, China, Finland, France, Germany, Hungary, Israel, Italy, Netherlands, Poland, Portugal, Spain, Switzerland, UK, and USA—have also offered their expertise to challenge authors within the limits that scientific exercise imposes. The outcome of this synergy is the object the reader now holds before his/her eyes. There is no doubt that this research topic will inspire both scientists and clinicians investigating and helping patients in the field of neurovision for the next 10 years.

## References


## Acknowledgments

The author thanks his editor in chief, Sidney A. Simon (Duke University, Durham, USA) for his support during this two-year two-month two-day editing journey. The author is grateful to Emiliano Macaluso (Fondazione Santa Lucia, Roma, Italy) and John J. Foxe (Albert Einstein College of Medicine, New York, USA) for editing two articles of the research topic. The author gives thanks to people of the Frontiers Neuroscience Editorial Office, particularly Lucia Brandi, Carina Paraíso, Sara Fahmy and Camilla Drury. The authors also thanks the Frontiers Science Production Office.


**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.

Copyright © 2015 Coubard. 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.

# PANORAMA

## Educating the blind brain: a panorama of neural bases of vision and of training programs in organic neurovisual deficits

#### **Olivier A. Coubard1,2\*, Marika Urbanski 3,4 , Clémence Bourlon<sup>5</sup> and Marie Gaumet <sup>1</sup>**

<sup>1</sup> The Neuropsychological Laboratory, CNS-Fed, Paris, France

<sup>2</sup> Laboratoire Psychologie de la Perception, UMR 8242 CNRS-Université Paris Descartes, Paris, France

<sup>3</sup> Service de Médecine et de Réadaptation Gériatrique et Neurologique, Hôpitaux de Saint-Maurice, Saint-Maurice, France

4 Institut du Cerveau et de la Moelle Epinière (ICM), Sorbonne Universités, Université Pierre et Marie Curie UM 75, Inserm U 1127, CNRS UMR 7225, Paris, France <sup>5</sup> Service de Médecine et de Réadaptation, Clinique Les Trois Soleils, Boissise-le-Roi, France

#### **Edited by:**

Emiliano Macaluso, Fondazione Santa Lucia, Italy

#### **Reviewed by:**

Marcelo Fernandes Costa, Universidade de São Paulo, Brazil Mark Richard Harwood, City College of New York, USA

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

Olivier A. Coubard, The Neuropsychological Laboratory, CNS-Fed, 14 rue du Regard, 75006 Paris, France e-mail: olivier.coubard@cns-fed.com Vision is a complex function, which is achieved by movements of the eyes to properly foveate targets at any location in 3D space and to continuously refresh neural information in the different visual pathways. The visual system involves five main routes originating in the retinas but varying in their destination within the brain: the occipital cortex, but also the superior colliculus (SC), the pretectum, the supra-chiasmatic nucleus, the nucleus of the optic tract and terminal dorsal, medial and lateral nuclei. Visual pathway architecture obeys systematization in sagittal and transversal planes so that visual information from left/right and upper/lower hemi-retinas, corresponding respectively to right/left and lower/upper visual fields, is processed ipsilaterally and ipsialtitudinally to hemi-retinas in left/right hemispheres and upper/lower fibers. Organic neurovisual deficits may occur at any level of this circuitry from the optic nerve to subcortical and cortical destinations, resulting in low or high-level visual deficits. In this didactic review article, we provide a panorama of the neural bases of eye movements and visual systems, and of related neurovisual deficits. Additionally, we briefly review the different schools of rehabilitation of organic neurovisual deficits, and show that whatever the emphasis is put on action or perception, benefits may be observed at both motor and perceptual levels. Given the extent of its neural bases in the brain, vision in its motor and perceptual aspects is also a useful tool to assess and modulate central nervous system (CNS) in general.

**Keywords: binocular vision, eye movements, visual pathways, neurovisual disorders, visual rehabilitation**

### **INTRODUCTION**

Born in Canada and USA at the beginning of the 20th century and in France in the 1950s under the impulsion of Henri Hécaen, neuropsychology examines the relationship between cognitive activity (attention, perception, gesture, memory, language, etc.) and corresponding cerebral condition (the different areas of the central nervous system—CNS—from spinal cord to cortex). At the frontiers of neurology, psychology and psychiatry, neuropsychology explores how brain and function may be correlated in brain-damaged patients or by using functional brain imaging (Diffusion Tensor Imaging, DTI; functional Magnetic Resonance Imaging, fMRI; Positron Emission Tomography, PET; Single Photon Emission Computed Tomography, SPECT). In close collaboration with neurologists and psychiatrists, neuropsychologists assess and rehabilitate brain-damaged patients by acting onto sensory, motor, cognitive and emotional spheres. As a branch of neuropsychology, visual neuropsychology specifically studies vision in its sensory, motor, cognitive and emotional dimensions. As such, visual neuropsychology focuses on the nervous part of visual function, from retina to the multiple areas of the brain it involves. Given that 60% of the brain participates in vision (Orban et al., 2004; Orban, 2007), the extent of this study field is wide, from the most elementary visual functions (visual acuity, contrast sensitivity, visual field, color, depth, movement or visuo-spatial perception) to the most complex ones (object and face identification, perception of scenes and of emotions, written language processing, action-perception interaction, etc.). Similarly, neurovisual pathologies extend from low-level (partial or complete loss of visual field, achromatopsia, astereopsia, akinetopsia, etc.) to high-level disorders (visual agnosias, prosopagnosia, visual alexia, Balint syndrome, etc.). In this didactic review for both experts and novices, we provide a panorama of the neural bases of vision in its motor and perceptual aspects: eye movements and visual systems, respectively. Based on this knowledge, we briefly review the different damages that can occur in the visual systems, before overviewing the different rehabilitation schools of visual neuropsychology, which were developed in Europe and USA since the 1970s. The present review belongs to a Frontiers in Integrative Neuroscience e-book containing eighteen other contributions, and as such will offer reference throughout

the text to those articles related to either Eye movements or Visual Perception or Visual training programs (Coubard, in press).

#### **FROM ACTION TO PERCEPTION**

"In the beginning was the act" (Von Goethe, 1808–1832/2014). In line with von Goethe, we point out in this review that vision is first and foremost action. The reason why the eyes move is twofold. First they move as direct consequence of retina morphophysiology (see **Figure 1A**). Only the fovea containing a high density of cones allows humans to perceive visual stimuli with high acuity, while the rest of the retina containing less cones but high density of rods perceives blur. For that reason, the eyes have to move to foveate visual stimuli in eccentricity or in depth. Second the eyes move as visual perception is impossible as soon as movement is absent, which has been demonstrated different ways since the seminal work by Yarbus (1967). Indeed when fixational eye movements are suppressed and the visual stimulus stabilized on the retina, perception just vanishes in a few seconds. This is due to the fact that one function of fixational eye movements, among other functions, is to continuously refresh neural activity in visual pathways (for a review see Martinez-Conde et al., 2013).

To visually explore the world, humans make a variety of eye movements. Saccades are step movements in direction to foveate targets at different locations, while smooth pursuit aims at following a moving target in direction (see **Figure 1B**). Vergence is a movement of the eyes in depth, which can be step as saccades or smooth as pursuit (see **Figure 1C**). For an original research on functional brain imaging of vergence eye movements, see in the present e-book the article by Alvarez et al. (2014). Eye movements can be not only volitional but also automatic in response to sudden stimuli in any modality (reflexive eye movements), to stabilize images on the retina during head movements (vestibulo-ocular reflex), or to gaze moving visual patterns (optokinetic reflex). Even when fixating a stationary point, the eyes are never at rest but move through micromovements, tremor, drifts, microsaccades, which are critical for vision as mentioned above (for reviews see Carpenter, 1988; Leigh and Zee, 1999; Coubard, 2011; Martinez-Conde et al., 2013). For a review on fixational eye movements and binocular vision, see in the present e-book the article by Otero-Millan et al. (2014).

For didactic purpose, **Figure 2** illustrates the brain and its subcortical and cortical architecture. Eye movements are performed thanks to extraocular muscles, which are controlled by a cascade of physiological mechanisms (Hikosaka and Isoda, 2010; see **Figures 2**, **3**). At the lowest level, extraocular muscles are directed by motoneurons: the lateral rectus is innervated by abducens nerve (VI), the superior oblique by trochlear nerve (IV), and other muscles (medial rectus, superior and inferior recti, inferior oblique, as well as intrinsic muscles) by the oculomotor nerve (III). Motoneurons are themselves directed by premotor or burst neurons, which generate three patterns of innervation: the pulse is the velocity to rotate the eye; the step is the position to maintain the eye in its new position; the slide, between the pulse and the step, counteracts viscoelastic forces of the oculomotor muscles and globe in the orbit (for reviews see Scudder et al., 2002; Coubard, 2013; see **Figure 3**). Premotor neurons for horizontal saccades are located in the paramedian pontine reticular formation (PPRF), the medullary reticular formation (medRF), the nucleus prepositus hypoglossi (NPH), and the medial vestibular nucleus (MVN). PPRF and medRF premotor neurons provide the pulse force, whereas the step force is achieved by bilateral NPH and adjacent MVN. PPRF premotor neurons are excitatory and contact ipsilateral motoneurons. Premotor neurons of medRF are inhibitory contacting contralateral motoneurons (Fuchs et al., 1985; Langer et al., 1986; Moschovakis et al., 1996; Scudder et al., 2002). Premotor neurons for vertical saccades are located in the rostral interstitial nucleus of the medial longitudinal fasciculus (riMLF; Büttner-Ennever and Büttner, 1978; King and Fuchs, 1979), while those for vergence

have been found in the mesencephalic reticular formation (MRF; Mays, 1984; Judge and Cumming, 1986; Mays et al., 1986). Premotor neurons of any type of eye movements in any direction are under common inhibitory control of so-called omnipause neurons (OPN) confined in the nucleus raphe interpositus in the brainstem (Büttner-Ennever et al., 1988; see **Figures 2**, **3**).

(CN: caudate nucleus; LN: lentiform nucleus) is in close vicinity of the

At a higher level, the premotor circuitry is controlled by the superior colliculus (SC), which can elicit on its own reflexive eye movements (Schiller et al., 1987; Chaturvedi and van Gisbergen, 1999), i.e., eye movements with reaction time from 60 ms in monkeys (Fischer and Boch, 1983) and 80 ms in humans (Fischer and Ramsperger, 1984). This is possible thanks to the direct retinotectal pathway to the superficial, visual layers of the SC and then direct activation of SC motor neurons by SC visual neurons, probably through the interlaminar connection between its superficial and intermediate layers (Mooney et al., 1988; Isa and Kobayashi, 2004). SC movement-related neurons located in its caudal pole activate premotor neurons (Moschovakis et al., 1996; Chaturvedi and van Gisbergen, 1999), while SC fixation neurons in the rostral pole prevent premotor neurons through

Coubard, with permission).

OPNs (Munoz and Wurtz, 1993a,b; Chaturvedi and Van Gisbergen, 2000; see **Figures 2**, **3**). Whether the SC neurons in the rostral pole play the same role but simply for small movement amplitudes (Hafed and Krauzlis, 2008), or are a separate fixation population through their connections to the OPNs (Munoz and Wurtz, 1993a,b) remains an open question.

At the cortical level, the SC is controlled by the posterior parietal cortex (PPC) and more specifically parietal eye field (PEF) for triggering reflexive movements (Paré and Wurtz, 2001) and the frontal eye field (FEF) for intentional movements (Schall et al., 2011), while the supplementary eye field (SEF) plays a role in movement preparing (Nachev et al., 2008). PEF is also involved in visual attention and in spatial updating of visual information (Pierrot-Deseilligny et al., 2004). For a review on the role of FEF in eye movements, see in the present e-book the article by Percheron, François and Pouget (What makes a frontal eye field area the frontal eye field area?, under review). The inhibitory control of SC is achieved by the substantia nigra pars reticulata (SNpr), which is itself gated by subcortical structures such as the dorsal striatum (Hikosaka et al., 2000). The SC also receives inhibition from the dorsolateral prefrontal cortex (DLPFC) through a direct prefrontotectal tract (Goldman and Nauta, 1976; Leichnetz et al., 1981; Gaymard et al., 2003). For a review on the role of DLPFC in eye movements, see in the present e-book the article by Funahashi (2014). Finally, cerebellum maintains accuracy (Prsa and Thier, 2011; see **Figures 2**, **3**).

In summary, the subtle cascade of physiological excitatory and inhibitory mechanisms, allowing eye movements to be performed with appropriate timing and precision, reveals the complexity of oculomotor control. For a review on decisional aspects of eye movements, see in the present e-book the article by Noorani (2014). The quality of eye movements and of binocular coordination is a prerequisite for fusion of the object of interest on the two foveas thus ensuring binocular visual perception.

#### **THE SEEING BRAIN FROM EYE TO CORTEX**

Binocular vision is achieved by five main neurovisual systems originating in the retina but varying in their destination within the brain (see **Figure 4**). Two systems have been widely studied: the retino-occipital or retino-cortical visual pathway

(see **Figure 4A**) and the retino-collicular or retino-tectal visual pathway (see **Figure 4B**). But there also exist three other systems: the retino-pretectal (see **Figure 4C**) and the retino-hypothalamic (see **Figure 4D**) visual pathways, as well as the accessory optic system (AOS) (not illustrated), which play a crucial role in vision though they are less known. The first neuron that is given the information from sensory cells—cones and rods—is the bipolar neuron. Interestingly, bipolar neurons transfer information to the second neuron, the ganglion neuron or retinal ganglion cell (RGC), within the retina. This means that CNS is already present in the peripheral organ for vision, reminding us that the eye is ontogenetically a differentiation of the diencephalon (see **Figures 2A,B**). This is how authors and artists see in the eye a door directly open to the mind (e.g., Marendaz et al., 2007). On the scientific viewpoint, it will be our rationale for using eye movements and vision as useful indicators of CNS (dys)functioning.

In mammals, RGCs include at least 20 different subtypes and directly target more than 24 brain areas. Each RGC subtype responds to a specific feature in the visual scene, though the function of many of them remains to be elucidated (for a recent review see Dhande and Huberman, 2014).

The retino-occipital visual system is the most evolved in phylogenesis. In this system (see **Figure 4A**), RGCs are retino-thalamic as they exit the eye through the optic nerve, go through the optic tract to reach the lateral geniculate nucleus (LGN), where they contact the third neuron, the thalamo-cortical neuron. Importantly between the optic nerve and tract, ganglion fibers originating in nasal hemi-retinas cross the optic chiasma whereas those originating in temporal hemi-retinas do not (see **Figure 4**). Such lateral systematization has consequences that will be described below (see Section The Blind Brain from Optic Neuritis to Neglect). Thalamo-cortical neurons exit LGN, go through the optic radiation or geniculo-calcarine tract to reach the primary visual or occipital striate cortex or V1 or Brodmann's area 17, where preliminary processing of visual information is achieved (Merigan and Maunsell, 1993; Bullier, 2001; Kaplan, 2003; see **Figure 4**). The retino-occipital visual system is mainly made of parvocellular neurons (P/X cells in animals), which convey high spatial frequencies allowing fine analysis of the visual scene (Merigan and Maunsell, 1993; Espinosa and Stryker, 2012). The corollary of such sophistication is its slowness as compared to other visual pathways in which visual information is processed faster and earlier but with lower resolution (Bullier, 2001; Kaplan, 2003). Some magnocellular neurons (M/Y cells in animals) are also present but in lower proportion. Recent studies using transgenic labeling of specific RGCs in mouse have demonstrated that LGN contains at least two functional categories of cells: direction selective RGCs (DSGCs) and non-DSGCs in respectively the shell and the core of LGN (Dhande and Huberman, 2014). W-like cells (small diameter) reside in LGN shell where DSGCs terminate, whereas Y-like cells (large diameter) reside in LGN core where alpha RGCs and non-DSGCs terminate, while X cells (medium diameter) are found both in LGN shell an core (Dhande and Huberman, 2014). From V1, the information is sent to extrastriate (i.e., other than V1) cortex: that from parvocellular neurons project mostly to V2, V4, and the occipital-temporal (ventral) stream involved in object identification; that from magnocellular neurons projet mostly to V2, V3, MT, and the occipital-parietal (dorsal) stream involved in object localization, action and action-perception integration (Ungerleider and Mishkin, 1982; Merigan and Maunsell, 1993; Kravitz et al., 2013; see **Figure 4**). For a review on neural bases of spatial frequency processing in scene perception, see in the present e-book the article by Kauffmann et al. (2014).

Phylogenetically less evolved than the retino-occipital route, other visual systems play nevertheless an important role in vision. Some of ganglion neurons do not reach LGN but fork earlier in direction of visual neurons located in superficial layers of the caudal pole of the SC (Dhande and Huberman, 2014; see **Figure 4B**). As evoked in Section From Action to Perception, the SC is responsible for the triggering of reflexive eye movements thanks to motor neurons of its intermediate layers (Schiller et al., 1987; Moschovakis et al., 1996). This contingent of ganglion neurons is made up with magnocellular cells and projects also to the pulvinar, to dorsal stream (Berson, 1988), as well as temporal (Sugase et al., 1999) and frontal (Bullier, 2001) areas of the brain. Such magnocellular pathway subserves an expedient but raw estimate of the visual scene by conveying low spatial frequencies (Bullier, 2001; Bar, 2003; Isa and Kobayashi, 2004). Importantly, the retino-collicular visual system is also linked to limbic structures such as the amygdala and the orbitofrontal cortex, which it is able to rapidly activate in response to fearful stimuli (Krolak-Salmon et al., 2004). This particular feature explains how visual stimuli may have emotional effects. Recent advances using genetic marking of RGCs have identified at least four parallel retinotopically complete maps in mouse SC, but it remains unknown whether these different maps are separate or combined within the network of collicular neurons (Dhande and Huberman, 2014).

Intrinsically photosensitive RGCs (ipRGCs) reach neither LGN nor SC but the pretectum, located between the mesencephalon and diencephalon (Dhande and Huberman, 2014; see **Figure 4C**). Pretectum contains several nuclei: the pretectal nucleus and tegmental nuclei such as interstitial nucleus, preinterstitial nucleus and epithalamic nucleus. The pretectal nucleus receives direct afferences from the retina and indirect ones from LGN and projects to the autonomic nervous system. Fibers of tegmental nuclei of the reticular formation belong to medial longitudinal fasciculus (MLF; Sprague, 1972; Büttner-Ennever and Horn, 1997). Functionally, the pretectum is critical for the pupillary light or photomotor reflex (Clarke et al., 2003).

A contingent of ipRGCs is directed to hypothalamus, specifically to the supra-chiasmatic nucleus, dorsally to the optic chiasma (Dhande and Huberman, 2014; see **Figure 4D**). Through melanopsin, this visual system projects to the pineal gland, which itself produces melatonin. As such, the retino-hypothalamic pathway regulates numerous behavioral and biological functions as well as circadian rhythms: temperature, wake/sleep, cortisol, reproduction, autonomic and hormonal functions, etc. (Trachtman, 2010). This explains how visual stimuli may have influence on various biological rhythms and functions.

Finally, the AOS (not illustrated) plays a role in head and gaze orientation as well as slow movements, and generate reflexive eye movements that compensate for retinal slip (Brodsky, 2012). AOS consists of On-DSGCs projecting to two brainstem targets: the nucleus of the optic tract and the dorsal terminal nucleus on the one hand, the medial and lateral terminal nuclei on the other, for controlling respectively horizontal and vertical slip compensation (Dhande and Huberman, 2014).

In recent years, brain functional imaging in humans has put forth our understanding of the different visual systems and has allowed researchers to explore new subsystems. This is how specialized visual areas have been discovered, such as the Fusiform Face Area (FFA) for faces (Sergent et al., 1992), the Parahippocampal Place Area (PPA) for navigation (Epstein et al., 1999), the Lateral Occipital Cortex (LOC) for objects and tools (Grill-Spector et al., 2001), the Extrastriate Body Area (EBA) for human body (Downing et al., 2001), or the Visual Word Form Area (VWFA) within the fusiform gyrus for reading (McCandliss et al., 2003; Dehaene and Cohen, 2011). For recent findings on functional brain imaging of visual pathways, see in the present e-book the article by Raz and Levin (2014).

To summarize, five visual systems participate in vision, two of which are widely studied (the retino-occipital and the retinocollicular systems) and three of which are less known (the retinopretectal, the retino-hypothalamic and the AOS). Taken together, this neurovisual circuitery represents 60% of brain activity in humans (Orban et al., 2004; Orban, 2007). The ubiquity of vision in the CNS explains on the one hand potential impact on vision that may have any brain injury, and reveals on the other hand the importance of vision in the functioning of the CNS in general and how useful vision may be in neuropsychological assessment and rehabilitation.

#### **THE BLIND BRAIN FROM OPTIC NEURITIS TO NEGLECT**

We now move to the damages that can occur in the neural visual pathways and rendering blind not the eye but the brain. Before entering the core of visual disorders, we remind two physiological properties of the visual system.

A first striking feature of vision is retinal inversion (see **Figure 5A**). After light information has crossed the cornea, the anterior chamber, the pupil, the lens, the posterior chamber, it has to cross all layers of the retina, as it is inversed, to reach sensory cells. Indeed, cones and rods are opposite to the light for a reason that is hitherto unknown, except that their metabolic and photopigment regeneration requirements need ready access to the choroidal blood supply in the deepness of the retina. Once sensory cells have transformed light into neural information, the latter is transferred to bipolar neuron then to ganglion ones as described above (see Section The Seeing Brain from Eye to Cortex). Due to retinal inversion, ganglion neuron fibers exit the eye making a hole in the retina, the blind spot, to merge into the optic nerve (see **Figure 5A**). Retinotopy, that is the way

information is spatially organized on the retina, is preserved throughout visual pathways and is retrieved particularly in SC and primary visual cortex (Dowling, 1970; Tamraz et al., 1999; Chalupa and Werner, 2003; Podoleanu, 2012; see **Figure 2C**).

A second striking feature of vision is visual pathways' systematization in sagittal (left vs. right) and transversal (up vs. down) planes (see **Figures 2C**, **5B**). In the sagittal plane, physiological architecture is such that visual information from left hemi-retinas is processed by the left hemisphere, whereas that from right hemiretinas remains in the right hemisphere. To achieve this feat, fibers from left (temporal) hemi-retina of left eye remain in the left hemisphere between the optic nerve and tract, while fibers from left (nasal) hemi-retina of the right eye cross the median plane in the chiasma optic to be processed by the left hemisphere. In mirror, fibers from right (temporal) hemi-retina of right eye remain in the right hemisphere, whilst fibers from right (nasal) hemi-retina of left eye cross the median plane in the chiasma optic to be processed by the right hemisphere. Importantly, the fact that information from left/right visual field is processed by right/left hemispheres is only due to the fact that the eyes are globular and, as a direct consequence, visual information is reversed in both sagittal and transversal planes between the physical world and the retinas (Tamraz et al., 1999; Chalupa and Werner, 2003; see **Figures 2C**, **5B**).

In transversal plane, fibers from lower hemi-retinas of the eyes is processed by lower fibers of the optic tracts and radiations, whereas fibers from upper hemi-retinas of the eyes is processed by upper fibers of the optic tracts and radiations. Such transversal distinction has some consequences in terms of damage (see below). As example, optic radiations between LGN and V1 project in such a 3D wave that upper fibers are parietal while lower fibers are temporal. Once again, only because of eyes' globularity, information from upper visual field is processed in fine in the part of the primary visual cortex that is below the calcarine sulcus, while information from lower visual field is processed in the upper part of V1 above the calcarine fissure (Tamraz et al., 1999; Chalupa and Werner, 2003; see **Figures 2C**, **5**).

These physiological constraints in mind, visual consequences may be more easily inferred from organic damages that can occur at any level of the visual pathways, as exemplified here for the retino-occipital route (see **Figure 5B**). At the lowest level of visual processing, unilateral lesion of the optic nerve is responsible for monocular blindness. Optic neuritis as observed in multiple sclerosis can result in monocular blindness (Viret et al., 2013; see **Figure 5B1**). Lesion of optic chiasma, which can be seen in hypophyseal adenoma, results in bitemporal hemianopia: only fibers from nasal retinas decussating in the optic chiasma are injured, resulting in temporal loss of vision bilaterally (Schneider, 1979; Foroozan, 2003; see **Figure 5B2**). If unilateral lesion concerns only direct fibers of optic chiasma, it results in monocular nasal blindness (see **Figure 5B3**). Homonymous hemianopia (HH) is inherent in unilateral lesion of either optic tract, or LGN, or optic radiations, or occipital primary visual cortex. HH causes loss of vision in ipsilateral hemi-retinas, thus in contralateral visual field (Grunda et al., 2013; de Haan et al., 2014; see **Figures 5B4–7/8**). Most common aetiologies for HH are stroke of the posterior cerebral artery (70%), traumatic brain injury (11–14%) and tumors (11%) (Zhang et al., 2006). Visual field defects can be more restricted than HH. Indeed unilateral retrochiasmatric lesion can be restricted to lower (temporal) or upper (parietal) optic radiations, resulting respectively in homonymous superior quadrantanopia (see **Figure 5B5**) or homonymous inferior quadrantanopia (see **Figure 5B6**). According to Zhang et al. (2006), 37% and 62% of HH would be complete and incomplete, respectively, with 29% for quadrantanopia. To end, it is worth noting that these low-level visual field defects (HH, quadrantanopia, etc.) have also deleterious impact in high-level functions such as reading or scene perception. As such, hemianopic dyslexia refers to reading deficit inherent in HH without lesion of high-level area related to reading such as VWFA (Schuett et al., 2008). For an original research on eye movements in dyslexia, see in the present e-book the article by Seassau et al. (2014). For scene perception, Perez et al. (2009, 2013) showed that left/right HH differently affect scene detection and categorization.

From V1, damages of the visual pathways yield visual impairments of higher level and complexity. Lesion may be restricted to any part of extrastriate cortex such as V2, V3, V4 or V5/MT. As examples, a specific lesion of V4 results in achromatopsia, the inability to see colors in visual stimuli (Duvelleroy-Hommet et al., 1997; Heywood and Kentridge, 2003), while a specific lesion of V5/MT results in akinetopsia, the inability to see motion in visual stimuli (Zeki, 1991; Nawrot, 2003). At a higher level, organic lesion of the brain may concern either the inferior temporal cortex (the ventral stream or "what" pathway) or PPC (the dorsal stream or "where" or "how" pathway). Ventral stream lesions lead to the constellation of visual agnosias gathering defects of identification, which may specifically concern objects (Konen et al., 2011), faces (Gainotti, 2014), reading (Cohen et al., 2004; Epelbaum et al., 2008), etc. Dorsal stream lesions result in defects in action and visuo-spatial attention such as Balint syndrome (Biotti et al., 2012), or spatial neglect (Corbetta and Shulman, 2011; Harvey and Rossit, 2012).

To end, other neurovisual systems can also be impaired, resulting in specific impairment of the functions related to the retino-collicular (Gaymard et al., 2003), retino-pretectal (Papageorgiou et al., 2009), retino-hypothalamic (Fraser et al., 2012) visual pathways, as well as to AOS (Brodsky, 2012). However these studies are scarce given the low frequency of such focal lesions in humans.

In summary, the different neurovisual systems (retinooccipital, retino-collicular, retino-pretectal, retino-hypothalamic, optic accessory) can be specifically impaired following organic lesion of the brain. Depending on the level at which the lesion occurs in the visual pathway from the optic nerve to their destination within the brain, visual defects concern different levels of visual processing, but low-level deficits may also impact high-level functions.

## **EDUCATING THE BLIND BRAIN: THE DIFFERENT SCHOOLS OF REHABILITATION**

Given our knowledge of eye movement and visual perception psychophysics and physiology and of the different ways visual systems may be injured, cognitive neuropsychology has enabled researchers to basically examine separately cortical vs. subcortical pathways of vision, and to clinically develop tools to rehabilitate impaired visual functions and/or boost spared ones. For a review on principles underlying visual training programs, see in the present e-book the article by Taub et al. (2014). Since the late 1970s and early 1980s, different schools have emerged in Europe and USA, which we now briefly review.

Recovery of some residual capabilities in patients suffering from organic lesion of one or several visual pathways (so-called neurovisual disorders) has been well documented. Pöppel et al. (1973) were the first to report that a patient suffering from a right HH following a left retrochiasmatic lesion was nevertheless able to accurately saccade to a flash presented in his blind visual field, though he had no conscious experience of his performance. Weiskrantz et al. (1974) called "blindsight" the ability of HH patients to perform well in visual tasks though they have no conscious experience of such performance. Blindsight includes eye movement abilities but also pointing, reaching and prehension tasks, identification of one of two objects or discrimination between shape and color attributes in forced-choice tasks (Weiskrantz et al., 1974).Whether blindsight is due to either spared cortical areas or functional subcortical routes is still under debate. For a review on blindsight, see in the present e-book the article by Perez and Chokron (2014). Consistent with the subcortical hypothesis, particularly the retino-collicular route linked to limbic system, HH patients can discriminate between fearful vs. neutral faces presented in their blind visual field, again without any consciousness of their performance, indicating that blindsight extends to affective performance (de Gelder et al., 1999). The first evidence of the involvement of the subcortical visual pathway in HH was brought by Sahraie et al. (1997). In this study, fMRI evidenced the involvement of the SC in HH patient GY performing a task well in which he had to decide the direction of a stimulus presented in his blind visual field. Additionally, Morris et al. (2001)invited patient GY to be presented with fearful vs. happy faces in his blind visual field. Using PET, the authors evidenced that fearful faces activated the neural network involving the retino-collicular visual pathway and the amygdala, providing further direct evidence that the colliculo-thalamo-amygdala complex is associated with unconscious visual perception of fearful face stimuli.

Rehabilitative techniques and methods in visual neuropsychology have been first initiated in Germany, USA, and France. In Germany, Zihl has developed visual training based on reading, pointing and eye movements (Zihl, 1980, 1995a,b). In this training, HH patients are invited to read a text moving on a screen, to point towards visual targets, and to make auditory saccades at different locations in space. Such training has been evidenced to enlarge the visual field as assessed by perimetry and to improve saccadic behavior. Based on this research, Kasten and Sabel introduced a standardized visual training so-called "Visual Restoration Therapy" (VRT, NovaVision© ) (Kasten and Sabel, 1995). In VRT, home-training of the blind visual field using computer-controlled stimuli allows patients suffering from HH or more restricted visual field defects to significantly enlarge their visual field size (Sabel et al., 2004; Jobke et al., 2009). Such improvement in vision seems to be independent of eye movements (Kasten et al., 2006) though it remains under debate (Horton, 2005). In line with these studies, Schuett (2009) has demonstrated the benefits of such rehabilitation in hemianopic dyslexia.

In USA, based on the observation that right traumatic brain injury patients exhibited deficits in visual exploration and in attentional span size, Ben-Yishay and Diller (1981, 1993) have developped visual training focused on attention. Specifically, the authors have used reinforcement of selective attention to improve visual exploration on the one hand, and of working memory, particulary central executive (the attentional control part of working memory), to boost planning, self-efficacy and consciousness on the other. This pioneer research has influenced visual entrainments focusing on executive functioning to rehabilitate perceptual deficits (e.g., Blázquez-Alisente et al., 2004).

In UK, since the seminal work by Weiskrantz et al. (1974) mentioned above, new tools have been discovered to assess the blind visual field of neurovisual patients. The latest one is the blindfield pupil response, which is attenuated in amplitude as compared to that of the sighted field and may be a predictor of intact psychophysical capacity in cortical visual field defects (Sahraie et al., 2013). On the other hand, Kennard and Pampakian have studied in HH the impact of saccadic eye movements in visual trainings. They have showed that saccades are improved with visual training, but eye movements may be more the effect than the cause of visual improvement (Pambakian and Kennard, 1997; Pambakian et al., 2004; Mannan et al., 2010). Additionally and consistent with Sahraie's pioneer research (Sahraie et al., 1997), the English school has been putting emphasis on the retinocollicular route to account for blindsight capabilities, particularly through its links with limbic system (Tamietto et al., 2009, 2010, 2012). On a pragmatic note, Leff et al. (Koiava et al., 2012; Ong et al., 2012) have proposed Read-Right©, a web application for diagnosing HH and rehabilitating hemianopic alexia.

In France, Ducarne, Bergego, and Barbeau were the first to develop visual neuropsychology in brain-damaged children and adults (Ducarne and Barbeau, 1981; Ducarne et al., 1981, 1983). Based on their research, they formalized assessment and rehabilitation protocols of visual field defects and neglect (Barbeau, 1992; Ducarne De Ribaucourt and Barbeau, 1993; Vital-Durand and Barbeau, 1995). In this visual training program, rehabilitation is organized around somatognosic (body, posture) and proprioceptive processes, reinforcement of visual afferences through acoustic and tactile afferences, residual visual capabilities using light, movement and colors, cognitive processes involving extrabody space representations, and verbal instructions. Special interest is also given to eye movements through the training of visual alert and smooth pursuit. Technically, reinforcement involves one-target or two or more targets' stimuli in tactile, proprioceptive, acoustic, manual and lower limb (leg, foot) modalities in experimental, pragmatic, and visuo-constructional situations (Ducarne De Ribaucourt and Barbeau, 1993). In line with this work, Chokron et al. showed in HH enlargement of contralesional visual field between pre- and post-training periods as assessed by automated perimetry (Chokron et al., 2008). Using fMRI and putting emphasis on retino-occipital route, these authors have also reported differential reorganization of visual cortical areas depending on lesion hemispheric side (Perez et al., 2009, 2013). For a review on functional brain imaging of visual training programs, see in the present e-book the article by Urbanski et al. (2014).

To end, it is worth noting that new technologies have also revolutionized visual training programs. As examples, repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS) have emerged as useful techniques to boost cortical areas in the neighborhood of lesioned occipital cortex and indirectly neuromodulate visual capabilities (Valero-Cabré et al., 2011; Brunoni et al., 2012; Afifi et al., 2013). For a review on non-invasive manipulation of frontal regions and eye movements, see in the present e-book the article by Vernet et al. (2014). Furthermore, Amedi et al. have shown that so-called visual areas (e.g., EBA) may not be visual but supramodal as evidenced by functional brain imaging of congenitally blind patients (Striem-Amit and Amedi, 2014). Indeed these areas can support sensory substitution and allow blind patients to see through nonvisual information like sounds or music (Maidenbaum et al., 2014; Proulx et al., 2014).

To summarize, various visual training programs have been developped since the 1970s in USA and Europe based on psychophysical, functional brain imaging, neuromodulation and substitution discoveries. Depending on theoretical frameworks, trainings have focused on action (the French school), on perception (the German and English schools), or on attention (the American school), resulting in each case in improvement of both eye movement behavior and visual perception.

## **CONCLUSION**

Making ours von Goethe's percept (Von Goethe, 1808– 1832/2014), we have emphasized in this review that vision is first and foremost action. The variety and subtlety of eye movements (saccades, pursuit, vergences, etc.) require the perfect orchestration of a cascade of physiological mechanisms from motoneurons to cortical areas. This movement machinery not only ensures binocular coordination to foveate targets at any location in 3D space, but is also a prerequisite for visual perception since action precedes perception. We are aware that many actions may be based on the preceding perceptual information, or in a stabilization context after removing movement there is still perception before the visual input fades. But because movement is essentially unavoidable (e.g., oculomotor tremor, drifts, head/body movements) and desirable for exploration, movement has always existed in animals and the evolution of perception has always dealt with this reality. In that sense movement has always preceded the evolution of our perception. In this context, we suggest that "visual action" is a more physiologically plausible expression than "active vision", which appears to be in most cases a pleonasm. Visual perception is achieved through five main systems and, importantly, any visual stimulus activates these different pathways. The retino-occipital and retino-collicular routes are widely studied, whereas the retino-pretectal, the retinohypothalamic, and the AOS are less explored. Therefore, our knowledge of complete or partial visual field defects inherent in antero- or retro-chiasmatic lesions, and that of high-level visual deficits (agnosias and visual action disorders) following a lesion of ventral or dorsal streams is well advanced. But studies exploring the impact of visual disorders on pupillary response, rhythms, biological functions (e.g., sleep or hormonal disorders), or on slow movements are lacking. Consistent with the neural bases of eye movements and visual pathways, action-perception integrated and multimodal interventions seem to provide the best results in visual rehabilitation. This suggests that any rehabilitative training in neuropsychology should first take into account the cognitive and cerebral constraints. In other words, any training should fit the physiology in its resources, plasticity, and limitations. Because vision recruits 60% of the brain (Orban et al., 2004; Orban, 2007), eye movements and visual perception are useful tools to assess and rehabilitate the CNS in general. In other words, vision in its motor and perceptual aspects may be useful biomarker and neuromodulator of CNS functioning: education in children and in normal aging, rehabilitation in CNS functional disorders, in neurological and psychiatric diseases, and in pathological aging. For an original research on eye movements and visual perception in Alzheimer's disease, see in the present e-book the article by Boucart et al. (2014). For a review on visual perception in schizophrenia, see in the present e-book the article by Notredame et al. (2014). For an original research on eye movements in bipolar disorder, see in the present e-book the article by Beynel et al. (2014). With respect to functional disorders, recent findings have suggested that visual disorders hitherto supposed to be peripheral may have cerebral causes and/or effects. This might be the case of anisometropia, amblyopia, strabismus, which involve CNS dysfunctioning as revealed by neurophysiological studies in animals and functional brain imaging studies in humans. For a review on functional brain imaging of amblyopia, see in the present e-book the article by Joly and Frankó (2014). For an original research on eye movements in amblyopia, see in the present e-book the article by Perdziak et al. (2014). For a review on neurophysiology of amblyopia, see in the present e-book the article by Bui Quoc and Milleret (2014). Thus vision in its complexity and richness offers exciting directions for future basic and clinical research.

## **ACKNOWLEDGMENTS**

Marika Urbanski was supported by the French Agence Nationale de la Recherche (project CAFO-RPFC, No: ANR-09-RPDOC-004-01).

## **REFERENCES**


Yarbus, A. L. (1967). *Eye Movements and Vision.* New York: Plenum.


**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 July 2014; accepted: 31 October 2014; published online: 05 December 2014*.

*Citation: Coubard OA, Urbanski M, Bourlon C and Gaumet M (2014) Educating the blind brain: a panorama of neural bases of vision and of training programs in organic neurovisual deficits. Front. Integr. Neurosci. 8:89. doi: 10.3389/fnint.2014.00089 This article was submitted to the journal Frontiers in Integrative Neuroscience*.

*Copyright © 2014 Coubard, Urbanski, Bourlon and Gaumet. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and 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*.

# PART I – NEURAL BASES OF EYE MOVEMENTS AND BINOCULAR COORDINATION

## Fixational eye movements and binocular vision

#### **Jorge Otero-Millan1,2 , Stephen L. Macknik1,3 and Susana Martinez-Conde<sup>1</sup>\***

<sup>1</sup> Department of Neurobiology, Barrow Neurological Institute, Phoenix, AZ, USA

<sup>2</sup> Department of Neurology, Johns Hopkins University, Baltimore, MD, USA

<sup>3</sup> Department of Neurosurgery, Barrow Neurological Institute, Phoenix, AZ, USA

#### **Edited by:**

Olivier A. Coubard, CNS-fed, France

#### **Reviewed by:**

Alessandra Rufa, Università di Siena, Italy Matteo Valsecchi, Justus-Liebig Universität Giessen, Germany

**\*Correspondence:**

#### Susana Martinez-Conde,

Department of Neurobiology, Barrow Neurological Institute, 350 W Thomas Rd., Phoenix, AZ 85013, USA

e-mail: smart@neuralcorrelate.com

During attempted visual fixation, small involuntary eye movements–called fixational eye movements–continuously change of our gaze's position. Disagreement between the left and right eye positions during such motions can produce diplopia (double vision). Thus, the ability to properly coordinate the two eyes during gaze fixation is critical for stable perception. For the last 50 years, researchers have studied the binocular characteristics of fixational eye movements. Here we review classical and recent studies on the binocular coordination (i.e., degree of conjugacy) of each fixational eye movement type: microsaccades, drift and tremor, and its perceptual contribution to increasing or reducing binocular disparity. We also discuss how amblyopia and other visual pathologies affect the binocular coordination of fixational eye movements.

**Keywords: microsaccades, drift, disparity, fixation, ocular, amblyopia**

#### **INTRODUCTION**

Binocular vision is a sensorimotor process: eye movements work to keep the lines of sight of left and right eye pointing to the same target, and the visual system combines the resultant, slightly different retinal images, to form a single percept (i.e., binocular fusion) and create a sensation of depth (i.e., stereopsis). Correspondence between the left and right retinal images is complicated by the fact that our eyes are never perfectly still, even when we attempt to maintain our gaze on an object of interest. Small fixational eye movements change the degree of alignment between the two eyes and continuously move the retinal images (**Figure 1**). In spite of this constant motion, we rarely suffer from diplopia (double vision), indicating that the motor system and the visual system are finely tuned to each other. Thus, normal fixational eye movements do not preclude binocular fusion; in other words, fixation disparity (disagreement between the alignment of the left and right eye) stays below a certain threshold that would preclude fusion from taking place. In the presence of pathologies that interfere with proper functioning of the visual or motor mechanisms, such us amblyopia or strabismus, subjects may suffer from diplopia and lack stereoscopic vision. Here we review the oculomotor characteristics of binocular fixation, the perceptual consequences of fixational eye movements on binocular vision, and the clinical aspects of pathological instability during binocular fixation.

The small eye movements that occur during attempted visual fixation consist of an alternation of quick motions called microsaccades (which occur once or twice per second) and periods of relative stability where the eye drifts slowly (**Figure 2**). A third type of fixational eye movement, beyond the measuring ability of most eye tracking systems, is called tremor, and is characterized by a very small quick oscillation that occurs simultaneously with drifts. Numerous studies have addressed the binocular properties of each kind of fixational eye movement. Most reports agree that microsaccades are generally conjugate, that is, that during microsaccades the two eyes move towards the same direction and by a similar amount, but there is less consensus about drifts.

From a perceptual standpoint, microsaccades have been shown to counteract visual fading and filling-in (Martinez-Conde et al., 2006; Troncoso et al., 2008a; McCamy et al., 2012; Costela et al., 2013), scan small and informative visual regions (Otero-Millan et al., 2008, 2013; McCamy et al., 2014b), improve visual acuity by precisely relocating the fovea (Ko et al., 2010; Poletti et al., 2013), and trigger perceptual transitions in a number of bistable illusions, including binocular rivalry (van Dam and van Ee, 2005; Troncoso et al., 2008b; Otero-Millan et al., 2012). Drifts and tremors are thought to enhance the processing of high spatial frequencies (Kuang et al., 2012).

#### **BINOCULAR CONTROL OF FIXATION EYE MOVEMENTS**

In this section we ask two main questions: First, are fixational eye movements conjugate? That is, do they have similar magnitudes and directions in both eyes? And second, does the difference in fixational eye movement directions and magnitudes between the two eyes serve to reduce or to increase fixation disparity? These are two related, but independent, questions. Disconjugate eye movements will reduce or increase disparity as a function of the vector difference between the movements in the two eyes, rather than of each absolute vector.

#### **MICROSACCADES**

Microsaccades are small saccades that occur 1–3 times per second during attempted fixation. They tend to be less than 0.5◦ in

**FIGURE 1 | Demonstration of fixation disparity. (A)** Example of nonius lines stimulus used to measure fixation disparity subjectively (Adapted from Jaschinski et al., 2005). Fixate on one of the central circles and diverge your gaze to achieve double vision, then try to match each circle with each circle and each X with each X. When you attain fusion, pay attention to the vertical lines. Misalignment between the top and the bottom line will be due to fixation disparity caused by fixational eye movements. **(B)** Schematic of the perception after fusion. Each of the central vertical lines is seen by one eye only and therefore do not fuse, whereas the central circles and Xs are seen binocularly. Fusion of the central circle indicates approximate alignment between the two eyes. Any simultaneous misalignment or motion of the vertical lines relative to each other will denote the corresponding fixation disparity.

amplitude, but can go up to 1◦ or more (Rolfs, 2009; Martinez-Conde et al., 2013; Otero-Millan et al., 2013).

Multiple studies, using different recording systems, have concluded that microsaccades are mostly conjugate eye movements (Krauskopf et al., 1960; Yarbus, 1967; St.Cyr and Fender, 1969; Schulz, 1984; Møller et al., 2002; Engbert and Kliegl, 2004). Indeed, most contemporary microsaccade studies use a binocular criterion (i.e., they only analyze microsaccades detected in both eyes) to reduce the amount of false positives resulting from the use of automatic microsaccade detection algorithms (Laubrock et al., 2005; Engbert and Mergenthaler, 2006; Engbert, 2006; Rolfs et al., 2006).

The first binocular recordings of microsaccades, performed in the early 1950s (Lord, 1951; Riggs and Ratliff, 1951; Ditchburn and Ginsborg, 1953), showed that a microsaccade in one eye was almost always accompanied by a microsaccade in the other eye, and that there was an overall correspondence between their respective magnitudes and directions.

Krauskopf et al. (1960) performed the first comprehensive and quantitative study of the binocular properties of microsaccades. They found that more than 95% of microsaccades had the same direction in both eyes and that the microsaccadic magnitudes in the two eyes were highly correlated. They showed that when the amplitude of the movement differed in the two eyes, the resulting difference tended to correct for errors in vergence. These results were later confirmed by St.Cyr and Fender (1969).

High-speed and high-resolution noninvasive video-trackers brought about a renewed interest in the binocular characteristics of fixational eye movements in the last decade. Møller et al. (2002, 2006) showed that microsaccades are generally conjugate. Engbert and Kliegl (2004) found that microsaccades tend to correct binocular disparity: on average, they reduced disparity by about 2 min of arc, with an approximate standard deviation of 6 min of arc. Around 35% of the microsaccades were errorproducing, however. van Horn and Cullen (2012) recently showed that only 7–8% of monkey microsaccades have complete opposite directions.

Microsaccades and saccades are often immediately followed by a fast smaller movement in the opposite direction, called a dynamic overshoot. Dynamic overshoots are also saccadic in nature, i.e., they follow the same main peak velocity/magnitude relationship as saccades, and therefore differ from the glissades or vergence eye movements that can also occur after saccades (Kapoula et al., 1986). Dynamic overshoots can be monocular and tend to be more common in the abducting eye (Abadi et al., 2000). It remains unclear why overshoots are more common in the abducting eye, but it could be related to the fact that saccades are generally asymmetric, being slightly faster and shorter in the abducting eye (Collewijn et al., 1988). Due to an oscillation of the lens in the eye, dynamic overshoots may appear larger in recordings performed with videooculography or Dual Purkinje eye tracking systems than in recordings obtained with scleral search coils (Kimmel et al., 2012; Nyström et al., 2013).

#### **DRIFT**

Drift refers to the slow eye movements that occur in between microsaccades during attempted fixation. Drifts are typically smaller and slower than microsaccades (typically less than 0.13 degrees in size, less than 0.5◦ /per second in speed (Rolfs, 2009)).

Eye drifts during fixation may not be a specific kind of eye movement, but result from the combined action of the gaze holding and retinal stabilization systems: The eye tends to drift slowly towards a "central position", especially in the darkness and when fixating eccentric targets (Leigh and Zee, 2006). In the presence of a visual stimulus, the pursuit and optokinetic systems compensate for any retinal slip, and the vergence system compensates for binocular disparities. If the head is not completely fixed, the vestibulo-ocular reflex will moreover compensate for head movements. In addition, vergence eye movements or glissades can follow saccades (Kapoula et al., 1986). All these systems are subject to neural and sensory noise and thus may produce additional undesired drift.

Different studies have obtained discrepant results regarding the binocular coordination of drifts and its role in correcting fixation disparity: Ditchburn and Ginsborg (1953), reported that drifts are mainly conjugate in the vertical component, with the two eyes moving up or down simultaneously. The horizontal component presented lesser conjugacy, and sometimes had a "wave-like" appearance with alternating periods of convergence and divergence. Simon et al. found that drift appeared to occur synchronously between the two eyes, although sometimes diverging and sometimes converging (Simon et al., 1984). Multiple studies using different and complex analyses (Spauschus et al.,

1999; Thiel et al., 2006, 2008) have found some level of synchronization between the two eyes during drifts. However, it is important to note that the fact that eye movements are synchronized according to these measurements is independent of the movements being conjugate and their effect on increasing or reducing disparity (Rolfs, 2009).

Research on the respective roles of drift and microsaccades on correcting fixation disparity has followed a similar path as studies on their role on correcting overall fixation position. Early on, Cornsweet et al. found that microsaccades, but not drift, had a corrective role in both overall fixation position and binocular disparity (Cornsweet, 1956; Krauskopf et al., 1960). Later studies found drifts to correct both fixation position (Steinman et al., 1967) and fixation disparity (St.Cyr and Fender, 1969), however. Specifically, St.Cyr and Fender (1969) found that drifts corrected errors in binocular disparity only in the horizontal direction. More recently, Engbert and Kliegl (2004) studied separately the contribution of microsaccades and drift to the correction of monocular fixation error and to the correction of binocular disparity, using random walk modeling and measuring the temporal correlations of eye positions for different timescales. They found that both microsaccades and drifts corrected fixation position on a long timescale (more than 100 ms), but only microsaccades corrected fixation disparity on a long timescale. Both microsaccades and drift produced random changes in disparity on short timescales (>20 ms).

#### **TREMOR**

Ocular microtremor is a small wave-like movement of just a few seconds of arc in amplitude and a frequency around 90 Hz (Martinez-Conde et al., 2004; Rolfs, 2009). Given tremor's small amplitude and fast frequency, only the most accurate eye tracking systems are able to measure it (in most standard systems it falls within the noise level). Some specific devices have been developed to measure tremor (Bengi and Thomas, 1968; Bolger et al., 1992, 1999; McCamy et al., 2013a, 2014a). Early studies found that tremor was independent in the two eyes (Riggs and Ratliff, 1951), but more recent research has found a peak of energy in the spectral coherence of tremor in the two eyes, indicating some level of synchronization that could be due to motor neuron activity (Spauschus et al., 1999).

#### **TORSION**

Human eyes have 3◦ of freedom: they can move not only horizontally and vertically, but also in the torsional plane. Torsional eye movements are rotations of the eye around the line of sight so the direction of gaze does not change. Torsional eye movements can induce disparities between the two eyes, especially in the periphery, and affect the 3D perception of slant (Enright, 1990). Van Rijn et al. measured spontaneous torsional eye movements during fixation and found that they were largely conjugate (Van Rijn et al., 1994). Cyclovergence, the difference between the torsional positions of the two eyes, was more stable than cycloversion, the average torsional position of the two eyes (0.07 vs. 0.2◦ ). They also found that the presence of a background improved cyclovergence stability. Zhang and Li (2012) observed small torsional movements associated with microsaccades.

#### **FIXATIONAL EYE MOVEMENTS IN BINOCULAR VS. MONOCULAR VIEWING**

Binocular performance can be superior to monocular performance of the same visual task, a phenomenon related to the brain's ability to combine effectively the information from the two eyes, known as binocular summation. Binocular summation predicts improved fixation stability under binocular viewing as compared to monocular viewing. Accordingly, González et al. (2012) found increased fixation instability during monocular viewing, especially for the occluded eye. They also showed that microsaccade rate is lower during binocular viewing, in agreement with Krauskopf et al.'s previous finding that microsaccades are larger and less frequent during monocular viewing (Krauskopf et al., 1960). González et al.'s results are also consistent with the observation that subjects make larger and less frequent microsaccades when they fixate larger and less precise targets (Steinman, 1965; McCamy et al., 2013b). Motter and Poggio (1984) moreover found that binocular viewing of a fixation target in monkeys produced a small but consistent reduction in the variability of eye positioning, when compared to either eye alone. Other studies found that microsaccade properties did not differ for monocular and binocular viewing, however (Schulz, 1984; Nallour Raveendran, 2013).

#### **FIXATIONAL EYE MOVEMENTS IN NEAR VS. FAR VIEWING**

Few studies have performed direct measurements of the parameters of fixational eye movements at different viewing distances. One might expect such parameters to change with the vergence effort demanded at each distance. Krauskopf et al. (1960) found no differences in fixational eye movement characteristics between far and near viewing, however.

#### **IS DISPARITY A STIMULUS FOR FIXATIONAL EYE MOVEMENTS?**

The fact that microsaccades and drifts correct disparity on average does not necessarily mean that disparity information is used in microsaccade or drift generation. Disparity correction by fixational eye movements could be accomplished in two different ways. First, each eye could act independently to reduce its own fixation position error. Second, visual system's estimation of disparity estimated could be used to produce a binocular eye movement that reduces such disparity.

Krauskopf et al. (1960) first set out to address this issue and found that both microsaccade magnitude and the probability of a microsaccade being triggered depended on gaze position error, but not on disparity error. Because Krauskopf did not find drifts to be corrective, he did not conduct similar analyses for drifts.

Later, St.Cyr and Fender (1969) confirmed Krauskopf's microsaccade findings. They found that the microsaccadic correction of fixation position error did not improve when also considering disparity. However, they did find that disparity information contributed to drift control.

#### **NEURAL CONTROL OF BINOCULAR FIXATIONAL EYE MOVEMENTS**

Because microsaccades are very brief, they must be controlled without visual feedback (i.e., the time lag of visual feedback is longer than the duration of a regular microsaccade, i.e., <30 ms (Otero-Millan et al., 2008). Slow eye movements such as drift can be continuously controlled by visual feedback, however. These two types of control systems are commonly referred to as open-loop and closed-loop. Correspondingly, two different vergence systems, fast and slow, are said to control the binocular coordination of eye movements (Cullen and Van Horn, 2011). During fixation, these two systems might control the respective conjugacies of microsaccades and drifts.

Recent neurophysiological evidence indicates that microsaccades are generated by the same circuit as saccades (Hafed et al., 2009; Guerrasio et al., 2010; Hafed and Krauzlis, 2012; van Horn and Cullen, 2012). Neural control of saccade-vergence interactions has been controversial since the times of Hering and Helmholtz. Hering believed that both eyes were controlled by a combination of binocular commands of the same amplitude for each eye (Hering's law of equal innervation (Hering, 1977)). In Hering's framework, apparently monocular or disconjugate eye movements were explained by the mathematical combination of version and vergence movements. Helmholtz believed instead that the two eyes were controlled independently, and that binocular coordination was a learned behavior (Coubard, 2013).

Horizontal eye movements are controlled by motor neurons in the abducens nucleus (innervating the lateral rectus) and in the oculomotor nucleus (innervating the medial rectus). Neurons in the abducens nucleus project through the medial longitudinal fasciculus (MLF) onto the contralateral oculomotor nucleus. Thus, during a saccade towards the right, neurons in the right abducens and left oculomotor nucleus will show very similar discharge patterns driving the movements of the right and left eyes respectively. In this circuit, the controversy regarding Hering's and Hemholhtz ideas translates into two possible implementations of disconjugate saccades. Following from Hering's law, a third group of neurons should modulate the discharge of neurons in the oculomotor nucleus, whereas Helmholtz's proposal requires two populations of neurons in the abducens nucleus, each corresponding to one eye (**Figure 3**). Zee et al. have proposed several models implementing both possibilities (Zee et al., 1992).

Studies have provided evidence in support of both Hering and Helmholtz's proposals. Neurons in the mesencefalic reticular formation (MRF) encode vergence commands and project to the oculomotor nucleus (Mays, 1984). Other research has shown neurons in the abducens nucleus encoding the monocular saccadic command (Cullen and Van Horn, 2011), a finding that also applies to microsaccades (van Horn and Cullen, 2012). One possible explanation for these apparently contradictory results is that, whereas the slow vergence is controlled by the vergence neurons in the MRF, saccades are encoded monocularly in the abducens (Cullen and Van Horn, 2011; Coubard, 2013). A recent study has shown that neurons in the rostral superior colliculus, typically associated with conjugate eye movements only, also encode changes in vergence angle (Van Horn et al., 2013).

The generation mechanisms of tremor are unknown, but some studies have proposed that it originates in the ocular motor neurons (OMN; Spauschus et al., 1999). If so, the synchrony between the left and right eye tremor reported by some studies, may result from the synchrony among the motorneurons that drive each eye.

## **FIXATIONAL EYE MOVEMENTS AND BINOCULAR PERCEPTION**

#### **STEREOPSIS AND FIXATION INSTABILITY**

Our visual system creates the perception of depth based on the small horizontal differences between the images projected onto each eye. This phenomenon is called stereopsis and it requires that both eyes are directed to the same target, so that the two retinal images can be fused into a single percept. If the two eyes are not properly aligned, double vision (diplopia) occurs.

The fact that our eyes move continuously during fixation and we rarely suffer from diplopia limits the possible mechanisms responsible for stereoscopic vision. Thus, to study binocular fusion, stereopsis and diplopia, one must know how much the

misalignment between the two eyes varies during fixation, and how much misalignment will result in diplopia.

The maximum amount of disparity or misalignment between the two eyes that the visual system can fuse into a single percept is called Panum's area, which is classically considered to range between 2 and 20 min of arc (Fender and Julesz, 1967; Duwaer and Brink, 1981a; Schor and Tyler, 1981). Panum's area varies for different stimuli, and differs in the horizontal and vertical axis (Fender and Julesz, 1967; Qin et al., 2006). Fender and Julesz (1967), using retinal stabilization, found that the perception of stereopsis presents the properties of hysteresis (**Figure 4**). That is, once the visual system achieves fusion, the perception of stereopsis continues even if disparity increases, a phenomenon that is particularly noticeable with random dot stereograms. In such cases, the maximum disparity permitting fusion is only around 7 min of arch, but once fusion is achieved, the perception of stereopsis continues while separating slowly the two stimuli up to 2◦ . In normal viewing conditions (i.e., without retinal stabilization) the two values are much closer to each other, because the visual system will use vergence movements to correct the disparity. With respect to fixational eye movements, Fender and Julesz (1967) concluded that hysteresis could compensate for the disparity introduced by slow drift (but not for large disparities caused by fast microsaccades, which could only be corrected with vergence movements).

The standard deviation of disparity during human fixation is between 1 and 7 min of arc, depending on the study (Duwaer and van den Brink, 1981b; Steinman et al., 1982), which would mean that the sensory system is capable to achieve fusion with disparities between 3 and 21 min of arc (3 standard deviations) to avoid diplopia during normal vision. This values are comparable to Panum's area's measurements (2–21 min of arc) and consistent with variability across human studies (Duwaer and Brink, 1981a). In cases when eye movements introduce larger disparities, the hysteresis properties described above could help to maintain fusion for disparities of up to 2◦ . Motter and Poggio (1984) obtained larger values for the standard deviation of disparity during fixation in the primate (i.e., about 10 min of arc).

The continuous motion of the eyes and the related changes in disparity make it unlikely that the visual system relies solely on retinal correspondence between the left and the right eye to achieve stereopsis. At least two mechanisms have been proposed: One possibility, put forward by Anderson and Van Essen (1987) and supported by neuronal data from Motter and Poggio (1990), involves visual receptive field shifting based on a signal carrying eye velocity information (either from corollary discharge or from

global motion estimation). Another mechanism, proposed by Howard and Rogers (1996), considers that the stereoscopic system only relies on first or higher order spatial derivatives of disparity. Small eye movements produce homogenous changes in disparity across the visual field, leaving the spatial derivatives unchanged. Because such a mechanism would be insensitive to small eye movements, it would not require additional signals to account for them and maintain fusion.

#### **BINOCULAR RIVALRY**

Binocular rivalry refers to the perceptual phenomenon that occurs when two very different visual stimuli are presented to each eye at corresponding retinal locations. In such cases, fusion does not take place, but the observer perceives an alternation of the two stimuli (rather than a mixture of both). Multiple studies have studied the potential relationship between eye movement production and the timing of the perceptual transitions in binocular rivalry (Sabrin and Kertesz, 1980; van Dam and van Ee, 2005, 2006a).

Binocular rivalry is present in stabilized images, arguing against a causal role of eye movements in driving the perceptual transitions (Blake et al., 1971; Wade, 1973) although the distribution of the durations of the intervals is different from the distribution during non-stabilized vision. Thus, it is unlikely that eye movements are the sole source of the transitions, but they may play a modulatory role. It is also possible that the transitions themselves affect the eye movements or that a third process of voluntary or involuntary control drives both the transitions and the eye movements.

Sabrin and Kertesz (1980) fount that microsaccade rates increased by 50% during binocular rivalry conditions vs. nonrivarly conditions, and that the increase happened mainly at the beginning of the periods of right eye dominance. Later, Sabrin and Kertesz (1983) found that simulated microsaccades with parameters matching real microsaccades while viewing stabilized rival stimuli best replicated the transitions occurring during nonstabilized viewing. This suggested that the oculomotor system and the rivalry system are tuned to each other. van Dam and van Ee (2005, 2006b) used orthogonal gratings as binocularly rivalrous stimuli, so that eye movements might produce retinal changes or not depending on their size relative to the grating frequency. They found that only microsaccades that led to retinal shifts were correlated with perceptual transitions during binocular rivalry.

## **FIXATIONAL EYE MOVEMENTS IN AMBLYOPIA AND STRABISMUS**

Ciuffreda et al. studied the fixational eye movements of subjects affected with amblyopia and strabismus. Their main finding was increased drift in the amblyopic eye during monocular viewing. If the amblyopia was due to strabismus, or in cases of alternating strabismus, the size and frequency of saccadic intrusions also increased (Ciuffreda et al., 1979a,b, 1980). More recently, Shi et al. (2012) found larger and less frequent microsaccades during monocular viewing with the amblyopic eye than during viewing with the fellow eye. In the case of viewing with the fellow eye, microsaccade parameters were comparable to those in subjects with normal vision. It is interesting to note that the characteristics of microsaccades during the fixation of large targets (i.e., in normal vision) (Steinman, 1965; McCamy et al., 2013b) resemble the microsaccadic parameters observed by Shi et al. in the amblyopic eye, suggesting that decreased fixation precision could be a common underlying mechanism.

Fixation stability is another related eye movement metric affected by amblyopia. Fixation stability is typically measured as the dispersion of the eye position during attempted fixation, for example BCEA (bivariate contour elliptical area). This parameter combines the influences of microsaccades and drifts, however; thus it cannot differentiate between the effects of either eye movement. González et al. (2012) found decreased fixation stability in the amblyopic eye, when compared to the eyes of healthy observers. Fixation stability in the fellow eye (non-amblyopic eye) was comparable to that in healthy observers, under both binocular and monocular viewing. Monocular viewing with the amblyopic eye decreased fixation stability of the fellow eye as compared to the eyes of control subjects (**Figure 5**).

Amblyopia is typically accompanied by poor visual acuity. Subramanian et al. (2013) studied fixation instability in amblyopic eyes of children with strabismus and/or anisometropia (when the two eyes have unequal refractive power). They found that the BCEA was larger in the amblyopic eye than in the fellow eye, especially along the horizontal axis. Fixation instability was correlated with visual acuity, that is, patients with larger BCEA had lower acuity.

eye, and the right eye is the fellow eye (modified from González et al.,

In an attempt to improve fixation stability for the amblyopic eye and achieve bifoveal fixation (Raveendran et al., 2014), reduced the contrast of the image shown to the fellow eye so that it was comparable to that perceived via the amblyopic eye. They found that, despite improvement in fixation stability in the amblyopic eye, bifoveal fixation is transient, with the strabismic eye drifting away from foveal alignment.

Another abnormal pattern of binocular fixational eye movements in amblyopic patients is fusion maldevelopment nystagmus syndrome (FMNS; Birch, 2013). FNMS is characterized by a horizontal conjugate nasalward slow-phase and a corrective temporalward quick-phase.

Recent research has shown that fixation instability can be used to detect amblyopia in early ages (when it can go undetected up to a third of the times (Loudon et al., 2011). This study used a binocularity score that measured how well subjects could fixate with both eyes on a target.

## **GENERAL DISCUSSION**

amblyopic eye results in increased instability in both eyes.

We have reviewed current knowledge about fixational eye movements in relation to binocular vision. An important recurrent theme is the two-way interaction between sensory systems and motor systems. Both sensory and motor aspects must be taken into account when studying visual perception and eye movement control: eye movements affect the sensory input, and the sensory input affects the eye movements in turn. The mutual tuning of fixation instability and the fusional system is a prime example of this interface. The relationship between fixation instability and decreased visual acuity in amblyopia is also indicative of the tight bond between the motor and the sensory facets of fixational eye movements.

Contemporary videooculography techniques are non-invasive and easy to operate, but do not have the same level of precision and accuracy as the scleral-search coil technique (McCamy et al., 2014a). To calculate the vergence position of the eyes one must determine the difference between the positions of the two eyes. Thus, vergence uncertainty will always double the uncertainty of the monocular eye position. This poses a challenge to the accurate measurement of disparity, and new studies conducted with videooculography techniques tend to report larger disparity values than those found in earlier research. Special care should be taken in calibrating the eye tracking set up in order to study vergence eye movements (De Luca et al., 2009).

Whereas recent studies have shed light on the generation and roles of microsaccades (see Martinez-Conde et al., 2013, for a review) much less is known about drift. Future studies should clarify the current discrepancies in results and determine how each of the gaze holding systems (vestibular, optokinetic, vergence or common integrator) contributes to drift. Can drift be generated purposely (i.e., to maintain a certain degree of disparity)? Is drift mainly a miss-calibration of either system (i.e., such as a lower gain in the integrator or an inappropriate gain of the optokinetic or vestibular systems)? Or is drift a mere manifestation of the noise level of one or all of those systems?

## **REFERENCES**


Wade, N. J. (1973). Binocular rivalry and binocular fusion of after-images. *Vision Res.* 13, 999–1000. doi: 10.1016/0042-6989(73)90080-1

Yarbus, A. L. (1967). "Eye movements during fixation on stationary objects," in *Eye Movements and Vision* (L. A. Riggs, Trans.) (NY: Plenum Press), 24–33.

Zee, D. S., Fitzgibbon, E. J., and Optican, L. M. (1992). Saccade-vergence interactions in humans. *J. Neurophysiol.* 68, 1624–1641.

Zhang, X., and Li, J. (2012). A novel methodology for high accuracy fixational eye movements detection. *Bioinformatics Biomed. Technology* 29, 133–140.

**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 March 2014; accepted: 03 June 2014; published online: 07 July 2014*.

*Citation: Otero-Millan J, Macknik SL and Martinez-Conde S (2014) Fixational eye movements and binocular vision. Front. Integr. Neurosci. 8:52. doi: 10.3389/ fnint.2014.00052*

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

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

## What makes a frontal area of primate brain the frontal eye field?

#### Gérard Percheron† , Chantal François and Pierre Pouget\*

Sorbonne Universités, UPMC Univ Paris 06, INSERM, CNRS, UM75, U1127, UMR 7225, ICM, Paris, France

The frontal eye field region (FEF) of the oculomotor pathways has been intensely studied. The primary goal of this review is to illustrate the phylogenetic displacement of the FEF locus in primate species. The locus is arrayed along the arcuate sulcus in monkeys and abuts into the primary motor strip region in humans. The strengths and limitations of the various functional, anatomical and histological methodologies used to identify such regions are also discussed.

Keywords: FEF, saccades, attention, anatomy, comparative, cortex

## Introduction

#### Edited by:

Olivier A. Coubard, CNS-Fed, France

#### Reviewed by:

Suliann Ben Hamed, Centre de Neuroscience Cognitive, France Vincent Prevosto, Duke University, USA

#### \*Correspondence:

Pierre Pouget, Sorbonne Universités, UPMC Univ Paris 06, INSERM, CNRS, UM75, U1127, UMR 7225, ICM, F-75013 Paris, France pierre.pouget@upmc.fr

†Deceased

Received: 29 March 2014 Accepted: 12 April 2015 Published: 18 May 2015

#### Citation:

Percheron G, François C and Pouget P (2015) What makes a frontal area of primate brain the frontal eye field? Front. Integr. Neurosci. 9:33. doi: 10.3389/fnint.2015.00033 Since the earliest studies of the brain as an organ, many criteria have been used to define subtypes and hierarchical organization within and between distinct brain regions. Historically, this has meant describing configurations of multiple brain areas believed to be associated with distinct functions and/or sub-functions according to anatomical landmarks. However, functional classifications have also been widely developed to allow existing definitions of anatomical brain regions to be questioned and reexamined (Brodmann, 1909; von Economo and Koskinas, 1925; Kononova, 1949, 1955; Bailey and von Bonin, 1951; Sarkissov et al., 1955; Sanides, 1964).

Classical definitions of brain region are based on many cytoarchitectonic and comparative anatomical studies. One important criterion of classification is the delineation of areas whose neurons have been retrogradely labeled after injecting another distinct region in the brain (Kitai and Bishop, 1981; Mesulam, 1982). This represents a connective criterion. This connectivity criterion has been used, for example, to define the axons projecting from the motor cortex to downstream motor centers. Another long-established tool that remains valuable is the delineation of zones where low-threshold currents produce functional modulation, for example evoked movements or perceptions. These represent functionally evoked criteria. In primates, such criteria have been used, for example, to define the region of the FEF by evoking eye movement while stimulating regions rostral to the arcuate sulci (Ferrier, 1874; Foerster, 1931). More recently, and particularly in humans, novel functional imaging methods relying on different criteria, such as changes in blood flow, diffusion of water molecules, or other concepts have been used (Price, 2012). The primary goal of this review is to illustrate the phylogenetic displacement of the FEF locus in primate species. We will illustrate our reasoning with a discussion of the location of the FEF in different primate brains according to various anatomical and functional criteria collected in recent decades in different primate brains (see for example Huerta et al., 1987; Tehovnik et al., 2000; Amiez and Petrides, 2009).

Some of this discussion will be based on the hitherto unpublished work of Dr. Gérard Percheron, who died in January 2011. His notes have been revisited and translated as part of the preparation of this review. A neurologist by training, and a former intern in Paris hospitals, Percheron had an early passion for basal ganglia morphology. He was one of the founders of International Basal Ganglia society, first established at a 1983 meeting in Lorne, in the state of Victoria, Australia. His enthusiasm for the study of the thalamus endured, and he described a functionally oriented partitioning of the thalamus in primates. Upon retirement, he continued to write on topics related to the anatomical organization of the cerebral cortex, with a particular interest in phylogenetic development across the primate order.

## The Architectonic Criteria and Position in Relation to Sulci

At the beginning of the 20th century, cortical architectonic studies began to attempt to characterize histologic entities according to their function. Various cytoarchitectonic (Brodmann, 1909; von Economo and Koskinas, 1925; Kononova, 1949, 1955; Bailey and von Bonin, 1951; Sarkissov et al., 1955; Sanides, 1964) and myeloarchitectonic (Campbell, 1905; Vogt, 1927; Strasburger, 1937; Filimonoff, 1949; Sanides, 1964) criteria enabled the division of the cerebral cortex into ''areas''. The cortex is classically characterized as a stratified or laminated neuronal assembly, whose parallel, superimposed layers allow organization perpendicular to the superficial laminae, and make columnar differentiation possible (**Figure 1**). Most regions of the cerebral cortex consist of six laminae (isocortex), with some variations and exceptions. This six-layer cerebral cortex is referred to as homotypical or eulaminate—the socalled association cortex (Brodmann, 1909; von Economo and Koskinas, 1925; von Economo, 1927; von Bonin and Bailey, 1947; Bailey and von Bonin, 1951). Heterotypic cortex refers to areas possessing fewer than six layers: the so-called granular and agranular types. The granular type (or koniocortex) is found in sensory cortices and is characterized by closely packed non-pyramidal cells in layers II–IV, which make it difficult to distinguish the layers in these areas. Afferent cortical fibers synapse in layers II and IV, with layer IV being the privileged stage of reception of specific thalamic axonal terminations. The agranular type is found in various regions, such as in the anterior part of the cingulate cortex, which lacks a granular layer IV, and the motor cortex, though a distinct interneuron layer IV has been recently and very elegantly described by Garcia-Cabezas in the motor cortices (Garcia-Cabeza and Barbas, 2014). Efferent cortical fibers exit from layers III and V (White, 1989; Snell, 1997). The dysgranular cortex is the cortical region that is transitional between the agranular and the granular cortex (von Bonin and Bailey, 1947). After von Economo (1927), we consider gyral parts to be superficial in gyri, sulcal parts to be located on the banks of the sulci, and fundic parts to be located in the depths of the sulcus.

## The Histotype and Anatomy of the Frontal Eye Field Region

#### Cercopithecidae

Old World monkeys (Cercopithecidae) are a group of simians native to Old World regions including Africa, India and Southeast Asia. Old World monkeys are medium to large in size. Some species are arboreal while others are terrestrial. Cercopithecids are almost exclusively diurnal. Most of the

six horizontal layers perpendicular to the superficial laminae. (A) Laminar organization of the primary motor cortex of macaque. (B) Laminar organization of the frontal eye field (FEF) of macaque.

anatomical studies on area 8 have been carried out on Cercopithecidae, which includes the diverse genera of the macaques. It is now more than a century since the oculomotor cortical areas in Cercopithecidae were first discovered and described. The cerebral sulcal pattern is very stable across the whole family (Falk, 1978), with an obvious and deep arcuate sulcus and a straight sulcus principalis or rectus. The oculomotor cortex is mainly gyral but also partially sulcal (Brodmann, 1905). Most investigators, including Brodmann, have recognized at least two cytoarchitectonic areas on this gyrus. Walker (1940) distinguished area 8a close to the superior branch of the arcuate sulcus and area 45 (along the inferior part), which is a part of area 8 (Brodmann, 1905; Vogt and Vogt, 1919). Area 8 is characterized by a thin but evident granular layer (Mauss, 1908) with large pyramidal cells in layers III and V (Brodmann, 1905; von Bonin and Bailey, 1947). Briefly, the oculomotor cortex of Cercopithecidae is mainly gyral, partially sulcal, and granular.

The oculomotor effect of cortical stimulation described by Ferrier (1874, 1875), Beevor and Horsley (1888), Mott and Schaefer (1890), and Vogt and Vogt (1907) was found to be the most effective just anterior to the arcuate sulcus (**Figure 2**). All stimulation studies since agree that the frontal oculomotor cortex has its core in the inferior arcuate sulcus but some (probably depending on stimulation parameters) delineated a wider area. Smith (1936, 1949) described an oculomotor area extending dorsally (to upper 6 and 9). This was reduced by Crosby (1953) and Brucher (1955, 1964), extended by Robinson and Fuchs, even more restricted by Bruce et al. (1985) (to the posterior portion of the arcuate sulcus, and mainly its anterior bank), before being re-extended again by Moschovakis et al. (2004). What is most commonly named today in macaques as the FEF (Huerta et al., 1986) mainly corresponds to cytoarchitectonic

area 8 or FDΓ. FEF or area 8 (if they are the same) has been functionally subdivided into two parts (though this distinction was not retained by Huerta et al., 1986): one dealing with pursuit and the other with saccades (Bruce et al., 1985). The saccadic region is located in a restricted area along the anterior wall of the arcuate sulcus, whereas the pursuit part is located deeper in the sulcus close to the fundus (Fukushima et al., 1999, 2008).

#### Cebidae

The Cebidae family of monkeys are mostly diurnal, but one genus, the Aotus, is primarily active at night. The oculomotor cortex of Cebidae has not been studied as widely or over such a long period of time as that of other monkeys or apes. It will be presented here in more detail. Compared to the gyral pattern of Cercopithecidae, the gyral pattern of Cebidae is variable. Owl monkeys (Aotus trivarigatus) are lissencephalic anterior to the central sulcus (**Figure 3**), as they only seldom have an inferior arcuate dimple (Huerta et al., 1986, 1987), and the oculomotor cortex is rather frontal. In squirrel monkeys (Saimiri sciureus) gyral variations range from no sulcus at all (**Figure 3**), to a simple dimple (Emmers and Akert, 1963; Huerta et al., 1987), to a small arcuate sulcus (Akert, 1964). The FEF is close to the dimple with considerable variation between individuals (Huerta et al., 1987). A small dysgranular area close to a particularly large dimple has been reported (Akert, 1964). The gyral pattern (**Figure 3**) of Cebus (apella or unspecified) is close to that of Cercopithecidae with a more accentuated sulcus arcuatus only (Sanides, 1970) or with a sulcus principalis (Tian and Lynch, 1997). As in the macaque, FEF in Cebus monkeys has been subdivided into two parts, one for smooth movement (FEMsem) and the other for saccadic movements (FEFsacc) (Tian and Lynch, 1996, 1997). In the Cebus, the FEFsacc is located at the apex of the arcuate sulcus on its anterior wall, a position close to that of Old World monkeys (**Figure 3**).

#### Apes

Hominoidea are a branch of Old World tailless anthropoid primates native to Africa and Southeast Asia. In comparison to the Cercopithecidae, major changes have occurred in the cortex of apes, especially in the areas anterior to the central sulcus. There is no sulcus resembling the arcuate sulcus of Cercipithecidae in apes. Differential studies carried out on a large number of chimpanzee brains showed considerable individual variation, particularly in the inferior precentral region (Sherwood et al., 2003). These authors insisted on the variability of the Broca's area homologue in great apes in area 44 and had some difficulty in accurately identifying the inferior part of the precentral sulcus. They concluded that the inferior part of the precentral sulcus is not a reliable criterion for delimiting area 44. The location of the oculomotor area in chimpanzees has mainly been mapped using electrical stimulation (Grünbaum and Sherrington, 1903; Hines, 1940; Dusser de Barenne et al., 1941; Bailey et al., 1950). Despite these differences in cerebral sulcal patterns in chimpanzees, the position of the oculomotor area in the gorilla (Sherwood et al., 2003, 2004) and the orangutan (Beevor and Horsley, 1890) is about the same as that of the chimpanzee (**Figure 7**). Sherwood et al. (2004) suggested that this uniformity might

monkey (Saimiri sciureus), Cebus monkey (apella or unspecified) and Macaque monkey brains. Note that the brain of Owl monkey is mostly lissencephalic anterior to the central sulcus. The gyral variations of squirrel monkey ranges from no sulcus at all to a clear gyral pattern. Note that the gyral pattern is close to that of Cercopithecidae with an accentuated sulcus arcuatus only, or with a sulcus principalis. The gyral pattern of macaque monkey is close to that of Cebus monkeys and other Cercopithecidae with an accentuated sulcus arcuatus only, or with a sulcus principalis.

reflect a common Bauplan<sup>1</sup> to great ape brain macrostructural organization.

The individual variability of cerebral sulcal patterns makes it difficult to examine this idea closely. For, chimpanzees, gorillas and orangutans, the sulcus containing the oculomotor FD is always distant from the motor cortex, is agranular (von Bonin and Bailey, 1947), and has been functionally defined using microstimulation as homologous to the FD of macaques.

#### Humans

There are major differences between apes and humans. However, even today, the location of the FEF in humans still raises intriguing problems. The human cerebral sulcal pattern is discernibly different from that of apes. The significant individual gyral variation, sometimes even between one hemisphere and the other, may partly explain the evident discrepancies in historical interpretations, which is particularly noticeable when comparing maps. Histologic studies have shown that the human FEF is not linked to any major sulcus (Pandya and Yeterian, 1996; Amiez et al., 2006) and that FEF is not located in Brodmann area 8 (Brodmann, 1909) (part of the granular cortex) but within the agranular cortex (**Figure 4**). Both PET and fMRI studies suggest that the activity of Brodmann area 8 is more associated with working memory, handling uncertainty, and analyzing coherent movements in the visual field, than eye movement per se (Cheng et al., 1995; Hyder et al., 1997; du Boisgueheneuc et al., 2006; Janata, 2009). Finally, some imaging studies have localized the human FEF in the precentral sulcus, abutting or within the primary motor strip.

These imaging studies have tended to characterize the human FEF as mainly precentral, premotor, and agranular (**Figure 5**). According to these properties, the human FEF might have been associated with the premotor cortex in Cercopithecidae (area 6 of Brodmann, 1909), area FB of von Economo and Koskinas (1925), area 4 s of von Bonin (1949), between FA and FB for Bailey and von Bonin (1951), and in area 6 for Sarkissov et al. (1955). However, it is important to question the assumption that the measured neuronal activity is only related to moving the eyes (Kawashima et al., 1996). Images derived from control scans performed while subjects fixate are compared with images gathered during saccadic test scans. However, as Kawashima et al. (1996) mention, unless subjects are specifically instructed to inhibit blinking, it is common for blinking to occur when saccades are made. In contrast, subjects blink less frequently during steady fixation or at rest. It is therefore possible that imaging studies may have located the FEF too far caudally, toward a motor strip containing a region that mediates blinking responses. An imaging study designed to examine

FIGURE 4 | Brodmann human brain areas defined and numbered based on the cytoarchitectural organization of neurons observed in the cerebral cortex using the Nissl stain. Note that functional FEF does not denote regions that can be distinguished by the morphology of the cells contained within it.

<sup>1</sup>Bauplan is a German word meaning an architectural plan. For FEF and primates, that meant a basic body plan for FEF structures and functions might have been modified by differences in sizes, proportions and by fusions and losses of other brain areas.

saccade generation failed to find the FEF near the expected precentral sulcus location. Instead insignificant precentral sulcus activation, with marked activation in the middle frontal gyrus was observed (Kawashima et al., 1996; Sugiura et al., 2004). An active region located in the middle frontal gyrus suggests anatomy homologous to monkey FEF. However, when Guipponi and colleagues tackled this question again recently, they showed that the identification of the neural correlates of spontaneous blinks in macaque monkeys does not map to the anterior bank of the arcuate sulcus (Guipponi et al., 2014). The measured fMRI activation has been identified as belonging to area 3b and not to motor primary cortex or to premotor area 6, calling into question the possible confounding factors revealed by previous studies. Evidence gathered using transcranial magnetic stimulation combined with structural MRI (Müri et al., 1991; Wessel and Kömpf, 1991) or imaging methods (Paus et al., 1996; Luna et al., 1998; Tehovnik et al., 2000) lends support to the location of the human FEF in the middle frontal gyrus.

#### Important Discrepancies

In animal models, there is a major discrepancy between two major classical anatomical works. For Bailey and von Bonin (1951), macaque FDΓ is almost entirely buried in the inferior frontal sulcus, and is thus much smaller than von Economo's area (**Figure 6**). None of the three areas—Brodmann's area 8, von Economo's area FdΓ, nor Bailey and von Bonin's FdΓ—has a topographical relation with human FEF. Bailey and von Bonin (1951) noted that Brodmann's work on the human brain was not extensive, and followed lengthy study of lemurs and monkeys. He published only a few figures of the human structures. Today, it is agreed that Brodmann's human area 8 is not functionally homologous to simian area 8. Bailey et al. (1950) had already expressed doubt as to whether this area (in macaques) was homologous to FDΓ in the human brain. Foerster published on two occasions Foerster (1931, 1936) two maps drawn after direct stimulations in man.

In his 1931 work, reproduced by Tehovnik et al. (2000) and redrawn by Brucher (1964), the FEF area is shown just in front of the precentral sulcus, close to the inferior frontal sulcus. This is approximately the position of area D that Dejerine (Dejerine and Roussy, 1906; Dejerine, 1914), claimed dealt with conjugate deviation of the head and eyes. The 2000 redrawing of Foerster's (1936) map by Blanke and colleagues, places the eye field more medially, even crossing the superior frontal sulcus. Penfield and Rasmussen (1957) later showed that the sites that effectively stimulate eyelid movements or eye rotations were located more posteriorly just in front of the central sulcus or, more precisely, just in front of the motor cortex where movement of the arms, face, and mouth could be elicited (**Figure 5**). The only area of controversy was the evoked head movement that was sometimes located more anteriorly, close to, and around, the precentral sulcus. Penfield and Rasmussen's (1957) localizations appear more posteriorly in comparison to recent maps (Chica et al., 2014). Blanke et al. (2000) also applied direct electrical stimulation to localize the FEF. They placed it in front of the precentral sulcus on both sides of the superior frontal sulcus, extending to the middle frontal gyrus. More recently, imaging methods have placed the human FEF in the precentral sulcus (Paus et al., 1996; Luna et al., 1998; Tehovnik et al., 2000).

Returning to histoanatomical data, the human FEF, as located by the recent histologic studies, is not located in Brodmann area 8 nor in von Economo FDΓ (Brodmann, 1909; von Economo, 1929; Huerta et al., 1987; Tehovnik et al., 2000; Amiez and Petrides, 2009). Surprisingly, the FEF of humans, in contrast to that of monkeys and apes, is no longer in the granular part of the cortex. Reviewing the data published on classical maps, FEF appears to be located within the agranular cortex—not in the giganto-pyramidal part of the motor cortex, but in the isocortex agranularis simplex of Bailey and von Bonin (1951). This region may be associated with the premotor cortex in Cercopithecidae: Brodmann's (1909) area 6, von Economo and Koskinas's (1925) area FB, von Bonin's (1949) area 4 s, between FA and FB for Bailey and von Bonin (1951), and in area 6 for Sarkissov et al. (1955). This observation has also been made by Tehovnik et al. (2000). In all these species there are connections with the frontal cortex just anterior to, and within, the premotor cortex, just caudal to the FEF, though the connections to the supplementary motor area are controversial in macaques (Stanton et al., 1995) and differences in cortico-cortical connections have been reported between Cebidae and macaques (Huerta et al., 1987).

In addition, within the striatum, it is also known that in macaques the FEF sends axons to the dorsal part of the caudate nucleus (Künzle and Akert, 1977; Huerta et al., 1986; Stanton et al., 1988; Pouget et al., 2009) interspaced with the frontal granular islands, which are in the associative part of the striatum not the sensorimotor part. There is no indication that this would be the case in humans. In fact the thalamic territories where the largest differences between macaques and humans were observed, were those involved in oculomotor function. A recent review confirmed these observations by comparing diffusion tractography imaging of FEF-striatal motor pathways in humans and macaques (Neggers et al., 2015). The authors confirmed that in macaques FEF is connected with the head of the caudate and anterior putamen, and M1 is connected with more posterior sections of caudate and putamen, corroborating neuroanatomical tract tracing findings. In humans FEF and M1 are connected to largely overlapping portions of posterior putamen and only a small portion of the caudate. In that respect, some of the hypothetical differences between the cortico-subcortical connections with the FEF might explain some of the functional differences between the FEF in humans and macaques. This different position and histology of FEF in humans and macaques should discourage making of assumptions about its connections to other cortical and subcortical regions. Most of these connections in apes are unknown.

## Discussion

One point is constant the FEF is always sulcal: to be more precise, it is located on one wall of the cerebral sulcus (**Figure 7**). This is obvious in Cercopithecidae that have a well delineated arcuate sulcus. In humans, it is located on the anterior bank of the superior precentral sulcus, close to the intersection between the precentral sulcus (vertical) and the superior frontal sulcus. This almost angular position is not sufficient to suggest that the superior precentral sulcus or the angle between this sulcus and the superior frontal sulcus is a remnant of the arcuate nucleus in Cercopithecidae (Blanke et al., 2000; Petrides and Pandya, 2002). Though the FEF of chimpanzees or gorillas is also sulcal, it has a different location; the sulcus does not evoke a transitory position and is does not lend itself to such an interpretation. Its sulcal position is likely due to the fact that its axons arrive early during the development and anchor in a still expanding cortex (e.g., Abeles, 1991).

The second point is the major change that has occurred between apes and humans. The FEF in apes is composed of a moderately thick and granular layer IV both rostrally and

caudally. This granular layer IV becomes almost invisible at the fundus and into the posterior bank, whereas in humans, FEF is pre-central, pre-motor, and agranular. This makes it difficult to conclude that there is topological equivalence between the simian and human FEF position in the agranular—dysgranular—eugranular sequence. Leaving aside the location of the FEF in human, apes and macaques, these discrepancies between cytoarchitectonic classifications, together with the functional delimitations established with the use of fMRI, are raising fundamental questions about the importance of the topological position of cortical areas within the Sanides cytoarchitectonic gradients, but also on the core measures extracted from the fMRI technique by itself.

In 2000 Tehovnik and colleagues concluded that ''the anatomy of FEFs is an enigma.'' In 2015, we find ourselves sharing this view. Major studies in human and animals still need to be performed to reach any other position. Four questions still need to be answered (Tinbergen, 1951): firstly, how did the FEF evolve (phylogeny)? How does the FEF promote fitness (selection)? How does the FEF develop (ontogeny)? And finally, how does the FEF system work (mechanism)? Some elements of the first two questions have been addressed in comparative neuro-anatomy, where it has been found that the organization of the basal ganglia among birds, mammals and other vertebrates is similar. In contrast, the organization of the pallial domains of these groups is more varied (e.g., Jarvis et al., 2005). Some common pathways are also preserved in the sensorimotor domain. All vertebrates have a circuit dedicated to the processing of spatial sensory information and orienting responses, which is commonly centered on the optic tectum. Reptiles and birds have evolved a highly laminated optic tectum that is much more developed than the top-down control of the optic tectum in mammals. In particular the acropallial gaze fields are a major center in the bird gaze control circuitry, exerting top-down gain control of the brain stem spatial map via a parallel projection to the deep layers of the optic tectum and to the saccadegenerating premotor neurons in the brain stem. In primates the projection to the superior colliculus of the FEF also exerts an indirect control on brain stem activity. Inter- and intra- species variability within these top-down oculomotor pathways have also been reported (Huerta et al., 1986; Tehovnik et al., 2000; Amiez et al., 2006). This review has mainly focussed on the displacement of the locus of FEF in primates, which lies on the arcuate sulcus in monkeys, but abuts the primary motor strip region in humans.

How does this displacement of FEF promote selection among primates? For a given species, how does the FEF develop ontogenetically? We believe that a synthesis of the responses to these questions holds the key to understanding the function of the FEF.

## References


Dejerine, J. (1914). Sémiologie des Affections du Système Nerveux. Masson: Paris.


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

Copyright © 2015 Percheron, François and Pouget. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and 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.

## Saccade-related activity in the prefrontal cortex: its role in eye movement control and cognitive functions

#### **Shintaro Funahashi \***

Kokoro Research Center, Kyoto University, Kyoto, Japan

#### **Edited by:**

Olivier A. Coubard, CNS-Fed, France

#### **Reviewed by:**

Nandakumar Narayanan, University of Iowa Carver College of Medicine, USA Vincent Prevosto, Duke University, USA

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

Shintaro Funahashi, Kokoro Research Center, Kyoto University, 46 Yoshida-Shimoadachi, Sakyo-ku, Kyoto 606-8501, Japan e-mail: funahashi.shintaro.2z@ kyoto-u.ac.jp

Prefrontal neurons exhibit saccade-related activity and pre-saccadic memory-related activity often encodes the directions of forthcoming eye movements, in line with demonstrated prefrontal contribution to flexible control of voluntary eye movements. However, many prefrontal neurons exhibit post-saccadic activity that is initiated well after the initiation of eye movement. Although post-saccadic activity has been observed in the frontal eye field, this activity is thought to be a corollary discharge from oculomotor centers, because this activity shows no directional tuning and is observed whenever the monkeys perform eye movements regardless of goal-directed or not. However, prefrontal post-saccadic activities exhibit directional tunings similar as pre-saccadic activities and show context dependency, such that post-saccadic activity is observed only when monkeys perform goal-directed saccades. Context-dependency of prefrontal post-saccadic activity suggests that this activity is not a result of corollary signals from oculomotor centers, but contributes to other functions of the prefrontal cortex. One function might be the termination of memory-related activity after a behavioral response is done. This is supported by the observation that the termination of memory-related activity coincides with the initiation of post-saccadic activity in population analyses of prefrontal activities. The termination of memory-related activity at the end of the trial ensures that the subjects can prepare to receive new and updated information. Another function might be the monitoring of behavioral performance, since the termination of memory-related activity by post-saccadic activity could be associated with informing the correctness of the response and the termination of the trial. However, further studies are needed to examine the characteristics of saccade-related activities in the prefrontal cortex and their functions in eye movement control and a variety of cognitive functions.

**Keywords: prefrontal cortex, saccadic eye movement, post-saccadic activity, context dependency, directional selectivity, frontal eye field**

#### **INTRODUCTION**

It is well known that both the frontal and supplementary eye fields participate in eye movement control, since (1) electrical stimulation of these areas evokes saccadic eye movements (e.g., Bruce et al., 1985); (2) lesion of these areas causes deficits in voluntary eye movements (e.g., Latto and Cowey, 1971a,b); and (3) eye movement-related activity is observed in many neurons in these areas (e.g., Bizzi, 1967, 1968; Bizzi and Schiller, 1970; Bruce and Goldberg, 1985). However, evoked eye movements have been observed by electrical stimulation not just within these eye fields but also in a rather wide area of the lateral prefrontal cortex. In recent neurophysiological studies using non-human primates, both pre- and post-saccadic activities were observed in the lateral prefrontal cortex (Funahashi et al., 1991; Takeda and Funahashi, 2002), which is located more anterior than the frontal eye field, although a great majority of saccade-related activities were post-saccadic; i.e., saccade-related activity was initiated after the initiation of saccadic eye movement or after the termination of eye movement. Most of these post-saccadic activities exhibited directional selectivity, in that this activity was observed only when the monkey performed saccades toward a certain direction. In addition, these activities exhibited context dependency, in that this activity was observed only when the saccade was goal directed, and not during spontaneous saccades during the inter-trial interval. Even though post-saccadic activity apparently has no functional significance in eye movement control, a large number of prefrontal neurons exhibit directional and contextdependent post-saccadic activities. This suggests that these postsaccadic activities must have some functional significance while the subject performs cognitive behaviors in which the prefrontal cortex participates (e.g., working memory, attention, decisionmaking). This article describes the characteristics of post-saccadic activity observed in the prefrontal cortex and considers possible functions of this activity in relation to the functions of the prefrontal cortex. The classification of pre- and post-saccadic activity and detailed analysis of saccade-related activities were performed in primate neurophysiological studies. Therefore, I focused on these primate physiological studies in this article.

## **HISTORICAL PERSPECTIVE ON THE PREFRONTAL CONTRIBUTION TO EYE MOVEMENT CONTROL**

#### **EFFECTS OF ELECTRICAL STIMULATION IN THE FRONTAL LOBE**

Ferrier (1875) used monkeys and reported that the frontal lobe participates in eye movement control. He applied electrical stimulation to part of the monkey's frontal lobe and observed eye movements that were directed toward the side opposite the stimulated hemisphere. He reported his observations in Philosophical Transactions (Ferrier, 1875), as follows:

"It has already been stated that the antero-frontal regions of the hemispheres give no response to electrical stimulation. Only one exception to this statement is to be made, viz. that in one case irritation of these regions caused the eyes to be tuned to one or other side, according as the electrodes were placed on the opposite hemisphere (p. 433)."

Subsequently, evidence that electrical stimulation of the frontal lobe, especially of the frontal eye fields, produced eye movements has been observed in other species such as cats, dogs, and apes (chimpanzee and orang-utan) (see Crosby et al., 1952 for details). In humans, Holmes (1938) reported that a patient with damage to the frontal lobe was unable to move his eyes in response to a command or to look at an object in any direction. He concluded "the frontal centers make possible the tuning of gaze in any desired direction and the exploration of space, but they also keep under control, or inhibit, reflexes that are not appropriate to our conduct or our reactions to the world around us". It had been known that the cortical area within which electrical stimulation produced hand or arm movements was in the pre-central sulcus. Therefore, the frontal lobe had been known to include the cortical area that supported eye movement control separately from the area that was responsible for skeletomotor control. This notion was supported by both animal and human studies. However, the size of the cortical area within which electrical stimulation could effectively evoke eye movements and the types of evoked eye movements were different among species and among investigators (Crosby et al., 1952). These differences could be due to several experimental factors. For example, since animals were anesthetized in these experiments, the animal's state of alertness or degree of anesthesia could affect eye movements evoked by electrical stimulation. In addition, the stimulus intensity used in these experiments was not always reported. Therefore, the stimulus intensity may have been different between these studies. Further, eye movements were not recorded and analyzed in detail. These differences in experimental conditions may have resulted in the inconsistent results regarding the characteristics of eye movements evoked by electrical stimulation of the frontal lobe.

Krieger et al. (1958) examined the effects of anesthesia on electrically evoked eye movements and found that the state of consciousness was important for determining the patterns of electrically evoked eye movements, as were the location stimulated and the parameters of the stimulation. Therefore, to exclude the effects of anesthesia, Wagman et al. (1961) used rhesus monkeys with high cervical transection and examined eye movements evoked by electrical stimulation of the frontal lobe. As shown in **Figure 1A**, they found that electrical stimulation had effects over a rather wide area of the frontal lobe. Eye movements were evoked by electrical stimulation of the cortex near the arcuate sulcus as well as the principal sulcus. The most commonly observed eye movements they found were horizontal conjugate, contralateral saccades. They also frequently observed oblique, either upward or downward, contralateral saccades. They examined the characteristics of evoked eye movements by stimulating not only the surface of the frontal lobe but also within the cortex. They also observed contralateral conjugate saccades by stimulation of the subsurface cortex. Based on these observations, they concluded that the frontal eye field extended anteriorly to the rostral limit of the principal sulcus, medially to the medial surface, and laterally to the lower limit of the arcuate sulcus.

While Wagman et al. (1961) examined electrically evoked eye movements, they neither stated parameters of electrical stimulation, nor recorded evoked eye movements. Robinson and Fuchs (1969) developed a method for accurately recording eye movements of awake monkeys and found that brief electrical stimulation (30-ms pulse train, 1-ms pulse at 200 Hz, 0.1–0.5 mA) of the frontal eye field (Brodmann's area 8) produced a single contralateral saccadic eye movement with a typical latency of 25 ms after stimulation onset. Most of the evoked saccades were single contralateral saccades. They found that electrical stimulation of the frontal eye field produced only saccadic eye movements, and did not produce smooth pursuit, vergence, or centering eye movements or nystagmus, although these eye movements were often observed in anesthetized animals and were later found in electrophysiological studies.

Electrical stimulation of the frontal eye field consistently produced saccadic eye movements, indicating that the frontal eye field acts as a premotor region for saccadic eye movements. However, the effects of frontal eye field lesion in monkeys did not directly support this notion. For example, Latto and Cowey (1971a,b) showed that, although lesion of the frontal eye field in monkeys produced temporary visual neglect in the contralateral visual field, recovery from neglect was rapid and there was little effect on saccadic eye movements. Schiller et al. (1980) also showed that both unilateral and bilateral lesions of the frontal eye field produced only temporary deficits in eye movements. Monkeys with these lesions exhibited neglect of the contralateral visual field and made fewer saccades toward targets in that field. These results suggested that the frontal eye field participated in the initiation of visually guided saccadic eye movements. However, they showed that paired lesions of both the frontal eye field and the superior colliculus produced severe deficits in eye movements. These monkeys could no longer perform saccadic eye movements toward visual targets and the deficits showed little recovery with time. In summary, although both electrical stimulation studies and lesion studies suggested that the frontal eye field apparently participated in the initiation of visually guided saccadic eye movements, further neurophysiological studies were needed to show the timing of the discharge of neurons in the frontal eye field during saccadic eye movements.

**FIGURE 1 | Prefrontal cortical area where electrical stimulations evoked saccadic eye movements**. **(A)** Results from Wagman et al. (1961). The locations of the stimulation are shown as black dots and the directions of the arrows.

#### **SACCADE-RELATED SINGLE-NEURON ACTIVITY IN THE FRONTAL EYE FIELD**

Evarts (1966) developed a method for recording single-neuron activity from the motor cortex of awake behaving monkeys. Using this method, Bizzi (1967) recorded single-neuron activities in the frontal eye field (area 8) of awake monkeys and found that the majority of frontal eye field neurons discharged in relation to eye position and that the neurons in this area discharged only after the initiation of eye movements. Subsequently, Bizzi (1968) found two types of neural activity (Type I and Type II) in the frontal eye field of monkeys. Type I neurons exhibited burst activity only after the initiation of saccadic eye movement, and this activity lasted for a very short time at the beginning of fixation. These neurons also exhibited burst activity in the fast phase of nystagmus. In addition, this burst activity was observed in complete darkness. On the other hand, Type II neurons exhibited a steady discharge when the eyes were oriented in a specific direction and were silent during saccadic eye movements. These neurons were also active during smooth pursuit eye movement and during the slow phase of nystagmus. All of these experiments were performed with the monkey's head immobilized. Bizzi and Schiller (1970) found that the characteristic discharge patterns of Type I and Type II neurons in the frontal eye field neurons were maintained regardless of head movements.

Neurophysiological studies using awake behaving monkeys revealed that neurons in the frontal eye field (area 8) exhibit activation in relation to eye movements. Distinct groups of frontal eye field neurons participate in either saccadic eye movements or smooth pursuit eye movements. Eye movement-related activity was observed in complete darkness. These results supported the

notion that the frontal eye field participates in eye movement control. However, these neurophysiological studies also showed that none of the neurons exhibited activation before the initiation of eye movement. Therefore, it was suggested that these neurons participated in some complex coordinate system used for orienting, since lesions of the prefrontal cortex had been shown to exhibit characteristic difficulties of visuomotor performances, such as difficulty in orienting vertical direction under tilted body conditions (Teuber, 1964).

Although none of the neurons in the frontal eye field exhibited responses that were exclusively related to saccadic eye movements, visually responsive neurons had been observed in the frontal eye field (Mohler et al., 1973) and were shown to exhibit enhanced visual responses when the visual stimulus was presented as the target for the saccadic eye movement (Wurtz and Mohler, 1976). Up to that point, this enhancement of visual responses was the only identified pre-saccadic activity observed in the frontal eye field in relation to eye movement. Since enhancement of the visual response was observed only when the visual stimulus was presented as the target of saccadic eye movement, this enhancement was considered to be a pre-eye movement signal and was exclusively related to eye movement. Goldberg and Bushnell (1981) further examined pre-saccadic enhancement of the visual response under a variety of task conditions and found that enhancement of the visual response was specifically observed only when the presentation of the visual stimulus preceded eye movement. Enhancement of the visual response was not observed when the visual stimulus presented within the visual receptive field was not the target of eye movement. In addition, enhancement of the visual response was not observed when the visual stimulus was the target of reaching behavior without movement of the eyes. Based on these observations, they concluded that pre-saccadic enhancement of the visual responses could be a cortical component of the neural events that preceded purposeful visually-guided saccades. Although pre-saccadic activity was found in the frontal eye field, this activity was still related to presentation of the visual stimulus and was not directly related to the control of eye movement.

Bruce and Goldberg (1985) first reported pre-saccadic activity that was directly related to saccadic eye movement in the frontal eye field. They strictly defined the location of the frontal eye field by intra-cortical micro-stimulation (**Figure 1B**; Bruce et al., 1985) and found that over half of the sampled neurons recorded from this area exhibited pre-saccadic activity (Bruce and Goldberg, 1985). These pre-saccadic neurons were classified into three groups (*visual*, *movement*, and *visuomovement* neurons). Among these, movement and visuomovement neurons exhibited saccaderelated activity without visual stimuli. Therefore, these neurons must be directly related to eye movement control. Interestingly, movement neurons exhibited significant pre-saccadic activity only during performance of the learned saccade tasks. These neurons exhibited either no activity or significantly weaker and less consistent activity before spontaneous saccades in the dark. In addition, each neuron that showed pre-saccadic activity exhibited a broad tuning with respect to saccade direction and had an optimal direction in which the magnitude of pre-saccadic activity was maximal. The neurons that exhibited pre-saccadic activity were mostly observed within the cortical area where a low (<50 micro A) threshold electrical current evoked saccadic eye movements (Bruce et al., 1985). The directions of saccades evoked by intra-cortical micro-stimulation were usually the same as the optimal direction of pre-saccadic activity at a given recording site.

They also observed post-saccadic neurons in the frontal eye field. These neurons exhibited only post-saccadic activity following saccades made in conjunction with the tasks as well as spontaneous saccades made outside the tasks (Bruce and Goldberg, 1985). In addition, neurons that exhibited only post-saccadic activity were observed in the cortical area where higher (>100 micro A) threshold current was required to evoke saccadic eye movements. The directions of evoked saccades were different than the optimal direction of post-saccadic activity (Bruce et al., 1985).

It has been known that some frontal eye field neurons exhibited both pre- and post-saccadic activities (Bruce and Goldberg, 1985). In these neurons, both pre- and post-saccadic activity exhibited directional tuning and the optimal direction of presaccadic activity was usually opposite of the optimal direction of post-saccadic activity. In visually guided or memory-guided saccade tasks, monkeys usually made a saccade toward the central fixation target just after making a saccade toward the peripheral target to get a reward. Therefore, post-saccadic activity observed in these neurons was not pure post-saccadic activity but could be pre-saccadic activity related to the eye movement toward the central fixation target. This suggests that neurons exhibiting both pre- and post-saccadic activities are distinct from neurons that exhibit only post-saccadic activity.

To summarize, Bizzi (1967, 1968) initially found no neurons that exhibited pre-saccadic activity in the frontal eye field. However, an evidence that a large number of neurons with pre-saccadic activity can be found in the frontal eye field as defined by lowthreshold intra-cortical micro-stimulation proved that the frontal eye field plays an important role in generating voluntary and purposive saccadic eye movements. This notion is supported by the findings that pre-saccadic activity is directionally tuned and context-dependent, such that this activity can be observed only when the subject performs a purposive eye movement and is not observed during spontaneous eye movements. On the other hand, many neurons that exhibit only post-saccadic activity are also present in the frontal eye field. This post-saccadic activity has been shown to be directionally un-tuned and not context-dependent. Therefore, neurons that exhibit pre-saccadic activity are distinct from neurons that exhibit only post-saccadic activity.

## **SACCADE-RELATED ACTIVITY IN THE DORSOLATERAL PREFRONTAL CORTEX**

The cortical area that is limited to eye movement control was defined within the classical "frontal eye field" (Brodmann's area 8; **Figure 1B**), which is a rather small area located at the caudal end of the principal sulcus and can be defined by whether or not saccadic eye movement is evoked by low-threshold intracortical micro-stimulation (<50 micro A) (Bruce et al., 1985). However, as seen in **Figure 1A**, electrically evoked eye movements had been observed over wide area of the lateral prefrontal cortex. Prefrontal participation in the control of motor behavior was known, since movement-related activity had been observed in the dorsolateral prefrontal cortex while monkeys performed a variety of motor tasks using the hand or arm. For example, Kubota and Niki (1971) first reported the transient excitation of prefrontal neurons when monkeys pressed a lever with their hand to get a reward in the delayed alternation task. Subsequently, Fuster (1973) observed similar movement-related activity while monkeys performed a delayed-response task with hand. Since then, activity related to behavioral responses has been widely observed when monkeys performed cognitive tasks as well as when monkeys performed simple motor tasks (see Funahashi and Takeda, 2002). Most movement-related activity exhibited selectivity, in that movement-related activity was observed only when monkeys performed a movement toward a particular direction (Kubota and Funahashi, 1982) or a certain type of behavior (e.g., go response) (Watanabe, 1986). It had been observed that movement-related activity started well before the initiation of the behavioral response (Kubota and Niki, 1971). Since the response characteristics of movement-related activity observed in the prefrontal cortex were similar to those of movement-related activities observed in the primary motor cortex (Kubota and Funahashi, 1982), this activity has been considered to participate in the initiation and execution of behavioral responses appropriate for the task.

Eye movement-related activity was also observed in the dorsolateral prefrontal cortex. Kojima (1980) used a delayed saccade task and reported saccade-related activity in the monkey prefrontal cortex. He observed a higher discharge rate during task-related saccades than during spontaneous saccades in the inter-trial interval. Joseph and Barone (1987) used a delayed oculomotor task, in which monkeys were required to perform a saccadic eye movement and an arm movement separately following visual and auditory stimuli, and examined single-neuron activity in the dorsolateral prefrontal cortex. Although they observed several types of task-related activities (e.g., signal-related pre-saccadic tonic cells, post-saccadic tonic cells, and signalrelated phasic cells), they observed few saccade-related neurons, most of which exhibited post-saccadic activation. Subsequently, Barone and Joseph (1989) used a spatial sequencing task, in which monkeys were required to press three targets in the order of their illumination under either a visually guided condition or a memory-guided condition, and examined prefrontal neural activities during the monkey's performance of this task. Although they analyzed visual tonic cells and context cells in detail, they also found that saccade-related neurons comprised 11% of their total sample (*n* = 302). They reported that, among these neurons, 85% exhibited post-saccadic activation and the remaining neurons exhibited some evidence of pre-saccadic activation. Although Joseph and Barone (1987) and Barone and Joseph (1989) used oculomotor tasks for their experiment, they mainly focused on neural activities related to cognitive functions. They did not focus on the mechanism of eye movement control by the prefrontal cortex. However, these studies showed that the dorsolateral prefrontal cortex also contained saccaderelated neurons, most of which exhibited post-saccadic activation, although the proportion of saccade-related neurons was small.

Boch and Goldberg (1989) directly examined prefrontal neural activity by using oculomotor tasks to examine the characteristics of saccade-related activity (e.g., visually guided saccade task, delayed saccade task, and memory-guided saccade task). They collected single-neuron activities from the posterior third of the principal sulcal area. As with frontal eye field neurons, visual response was enhanced when the visual stimulus presented within the visual receptive field became a target of eye movement. No evoked eye movement was observed with electrical stimulation using a current as high as 150 micro A in the recording area. They observed post-saccadic activation in 11% of neurons. Postsaccadic activation had no selectivity with respect to saccade direction or amplitude and was observed during purposive saccades as well as spontaneous saccades in total darkness.

Funahashi et al. (1991) examined saccade-related activity of dorsolateral prefrontal neurons using a memory-guided saccade task (oculomotor delayed-response task, ODR task) and a visually guided saccade task, and found that a large number of neurons exhibited saccade-related activation in dark. Among 434 neurons recorded from the principal sulcal area, one third exhibited saccade-related responses in the ODR task. Among the neurons with saccade-related activity, a great majority (78%) exhibited post-saccadic activity and the remaining 22% exhibited pre-saccadic activity. **Figure 2** shows an example of postsaccadic activity. Nearly all of the neurons that exhibited either pre-saccadic or post-saccadic activity showed directional selectivity (**Figure 3**). For the majority (62%) of neurons with presaccadic activity, the best direction was toward the visual field contralateral to the recording hemisphere, and the best direction for the remaining neurons was toward the ipsilateral field (23%) or along the vertical meridian (15%). On the other hand, 48% of neurons with post-saccadic activity had a best direction toward the contralateral visual field, and the best direction for the remaining neurons was toward the ipsilateral field (36%) or along the vertical meridian (16%). The comparison of saccaderelated activity between the ODR task and visually guided saccade task revealed that most of the neurons exhibited highly similar profiles of directional selectivity and response magnitude in the two tasks. In addition, both pre- and post-saccadic activities were observed only in conjunction with task-related saccades. As shown in **Figure 2**, prefrontal neurons exhibited post-saccadic activity only when the monkey performed purposive saccades leading to a reward. Comparable activity was not observed in association with saccades performed during the inter-trial interval in dark. This context dependency of postsaccadic activity is one of the important features of prefrontal neurons.

The original observation by Funahashi et al. (1991) was confirmed by Takeda and Funahashi (2002), who examined information represented in pre- and post-saccadic activities using two oculomotor tasks; the ODR task and the rotatory ODR (R-ODR) task. The direction of the saccade was the same as the direction of the visual cue in the ODR task but 90◦ clockwise from the direction of the visual cue in the R-ODR task. By comparing the directional tuning of saccade-related activity in the ODR and R-ODR conditions, we could determine whether each saccade-related activity encoded the direction of the visual cue or the direction of the saccade. For example, if the preferred direction of saccade-related activity was the same between these two tasks, we could conclude that this activity encoded the direction of the visual cue. On the other hand, if the preferred direction of saccade-related activity in the R-ODR task was 90◦ clockwise from the preferred direction in the ODR task, we could conclude that this activity encoded the direction of the saccade. Takeda and Funahashi (2002) showed that, among 57 neurons that exhibited directional activity during the response period, 58% encoded the direction of the saccade and 35% encoded the direction of the visual cue (**Figure 4**). Both preand post-saccadic neurons were included in the latter group. Interestingly, most of the neurons with saccade-related activity that encoded the direction of the visual cue also exhibited directional delay-period activity which encoded the direction of the visual cue, and the preferred directions of both activities were almost identical. These results indicate that not all saccade-related activities are related to the execution or control of saccadic eye movement. Although the activity is generated in relation to eye movement, some saccade-related activities of prefrotal neurons apparently participate in functions other than eye movement control.

In summary, it has been shown that the dorsolateral prefrontal cortex includes a large number of saccade-related neurons. Although most saccade-related activity was post-saccadic, a significant proportion of saccade-related neurons exhibit presaccadic activity. The characteristics of directional selectivity and context dependency are similar between pre- and post-saccadic activities in the prefrontal cortex. These results indicate that the prefrontal cortex participates in the control of purposive saccadic eye movements, similar to the frontal eye field. However, since most saccade-related activity in the prefrontal cortex was post-saccadic and since some saccade-related activities of

prefrontal neurons apparently participate in functions other than eye movement control, the prefrontal cortex and the frontal eye field may make different contributions to the control of saccadic eye movements.

### **COMPARISON OF SACCADE-RELATED ACTIVITY BETWEEN THE PREFRONTAL CORTEX AND THE FRONTAL EYE FIELD**

An important feature of the saccade-related activity of prefrontal neurons is that both pre- and post-saccadic activities are contextdependent, i.e., prefrontal neurons exhibited saccade-related activity only when the monkey performed purposive saccades (Funahashi et al., 1991). Both pre- and post-saccadic activities were either greatly reduced or absent during spontaneous saccades. However, neurons in the frontal eye field exhibit context dependency only for pre-saccadic activities, and not for post-saccadic activity (Bruce and Goldberg, 1985). Contextdependency only for pre-saccadic activity had also been shown in other brain areas including the posterior parietal cortex (area 7a) (Lynch et al., 1977), the caudate nucleus (Hikosaka et al.,

1989), and the substantia nigra pars reticulata (Hikosaka and Wurtz, 1983). In these brain areas, most neurons exhibited presaccadic activity during goal-directed saccades, but only weak or no activity during spontaneous saccades with the same direction and magnitude in the dark.

movements but the cumulative curves of neuron activities. Adapted

from Funahashi et al. (1991).

Another important feature of saccade-related activity in the prefrontal cortex is that most neurons exhibited only postsaccadic activity (Funahashi et al., 1991). Post-saccadic activity began at the same time of the initiation or well after the termination of saccadic eye movements. The median latency of post-saccadic activity was 122 ms after the initiation of the eye movement. The mean duration of saccadic eye movements was 45 ms (Funahashi et al., 1989). Therefore, most post-saccadic activity in the prefrontal cortex initiated more than 50 ms after the completion of saccadic eye movements in the dark.

Although post-saccadic activity is neglected in many reports because this activity does not have any clear role in the initiation and control of saccadic eye movements, post-saccadic activity has usually been considered to be an evidence of a corollary discharge

from structures that are directly responsible for eye movement. If post-saccadic activity is a corollary discharge from the oculomotor centers, then it should be observed with any saccade regardless of whether it is purposive or spontaneous. In fact, neurons in the oculomotor centers, such as the brainstem premotor nuclei (Hepp et al., 1989) and the superior colliculus (Sparks and Mays, 1980), were active during every saccade regardless of whether it was rewarded or spontaneous. In fact, post-saccadic activity in the inferior and lateral pulvinar (Robinson et al., 1986) and the frontal eye field (Bizzi, 1968; Bizzi and Schiller, 1970; Bruce and Goldberg, 1985; Bruce et al., 1985) was observed during saccades regardless of whether it was rewarded or spontaneous. Therefore, post-saccadic activity in these structures is likely to be a corollary discharge. On the other hand, post-saccadic activity observed in the prefrontal cortex was context-dependent, in that postsaccadic activity was observed only during saccades in relation to the reward and became significantly weak or was not observed during spontaneous saccades. Therefore, post-saccadic activity observed in the prefrontal cortex is not a corollary discharge from oculomotor structures in the brainstem.

In addition, in the frontal eye field, directional preference and contralateral bias were observed for pre-saccadic activity, but not for post-saccadic activity (Bruce and Goldberg, 1985). However, in the prefrontal cortex, most pre- and post-saccadic activities exhibited directional preference and contralateral bias, although the contralateral bias for pre-saccadic activity was stronger than that for post-saccadic activity (**Figure 3**; Funahashi et al., 1991).

In summary, the prefrontal cortex includes a large number of saccade-related neurons, most of which exhibit post-saccadic activity. Although post-saccadic activity is initiated well after the initiation of saccadic eye movement, this activity is clearly related to saccadic eye movement, since it exhibits selectivity with

respect to the saccade direction and directional tuning. However, in contrast to the traditional idea that post-saccadic activity is a corollary discharge from oculomotor centers, post-saccadic activity observed in the prefrontal cortex is not a corollary discharge because of its context-dependent nature. These results indicate that post-saccadic activity plays some specific roles in cognitive functions in which the prefrontal cortex participates.

## **CONTRIBUTION OF POST-SACCADIC ACTIVITY TO COGNITIVE FUNCTIONS**

#### **POST-TRIAL ACTIVITIES OBSERVED IN THE PREFRONTAL CORTEX**

Post-saccadic activity is initiated after the initiation of saccadic eye movement. Some neurons with post-saccadic activity start firing well after the termination of saccadic eye movement. Several post-trial activities have been found in the prefrontal cortex, and post-saccadic activity could be classified as one of these. For example, Rosenkilde et al. (1981) reported three types of posttrial activities while monkeys performed delayed-response and delayed matching-to-sample tasks. The post-trial activities that they observed depended on the presence or absence of the reward. One type of post-trial activity (Type I cells) was related to reward delivery. This activity was observed after correct, reinforced responses regardless of the task conditions. This activity was also observed after the delivery of free reward without any response. No activity was observed after unreinforced responses regardless of whether the response was correct or not. Therefore, this activity is considered to be directly related to reward delivery. Another type of post-trial activity (Type II cells) was not observed after correct, reinforced responses, but was observed in extiction trials with correct but unreinforced responses. They stated that this type of activity encoded deviations from expectancy of the reward. The third type of post-trial activity (Type III cells) was observed after all responses regardless of the presence or absence of the reward. They concluded that this activity encoded termination of a trial.

Watanabe (1989) further characterized post-trial activities observed in the prefrontal cortex. In addition to neurons (*juicerelated units* and *end-of-trial-related units*) with responses similar to those of the Type I and III cells described by Rosenkilde et al. (1981), several other kinds of neurons that exhibit posttrial activities have been reported. For example, *reinforcementrelated units* responded only when the reward was given for the correct response and did not respond to the free reward. *Reinforcement-error-related units*responded after both correct and error responses, but the responses in each case were different. He found some reinforcement- and reinforcement-error-related units that encoded the correctness of the response, since the responses of these units depended only on the correctness of the response and were independent of the presence and absence of the reward. In addition, he found two kinds of *error-related units* that responded only when the monkey made an error: one was related to error recognition and the other was related to encoding the absence of an expected reward, which is similar to the Type II cells described by Rosenkilde et al. (1981).

Neurons in the prefrontal cortex exhibit several kinds of post-trial activities. Since reward delivery is always provided at the end of a correct trial, reward-related activity (e.g., Type I cells, Type II cells, juice-related units, reinforcement-related units, and error-related units) could be observed in any type of experiment performed in the prefrontal cortex. Actually, reward-related activity was observed when the ODR task was used to examine prefrontal activities (Ichihara-Takeda and Funahashi, 2006). In the ODR task, the monkey was required to perform memoryguided saccades toward one of eight directions. The same amount of reward was delivered in every correct trial immediately after the end point of the saccade entered within the target zone. Therefore, no directional tuning should be observed in reward-related activity. As was explained before, a great majority of post-saccadic activity exhibited directional selectivity in the prefrontal cortex (Funahashi et al., 1991; Takeda and Funahashi, 2002). Therefore, post-saccadic activity is not a kind of rewardrelated activity. On the other hand, activity that encoded the end of the trial (Type III activity and end-of-trial-related activity) and activity that encoded the correctness of the response were observed in the prefrontal cortex when the monkeys performed cognitive tasks. Activity that encoded the end of the trial would be important for neural systems related to cognitive functions, since these systems need to prepare to receive information that is necessary for the next new trial immediately after the preceding trial is complete. Similarly, activity that encodes the correctness of the response and error-related activity that detects an error should also be important for monitoring the responses and for making flexible changes in behavior. Post-saccadic activity observed in the prefrontal cortex seems to have features similar to the activity that encodes the end of the trial and that encodes correctness, due to the context-dependent nature of post-saccadic activity.

### **POST-SACCADIC ACTIVITY COULD TERMINATE DELAY-PERIOD ACTIVITY**

Funahashi et al. (1991) observed a large number of prefrontal neurons that exhibited post-saccadic activity while monkeys performed an oculomotor version of the delayed-response task (ODR task). This post-saccadic activity could be important for the prefrontal cortex to prepare to receive updated information for a new trial and to monitor responses and make flexible changes in behavior.

Using the ODR task, Funahashi et al. (1989) found that a large number of prefrontal neurons exhibited tonic, sustained activation during the delay period (delay-period activity). This delayperiod activity was directionally tuned, in that it was observed only when visual cues were presented at a particular area in the visual field, usually the visual field contralateral to the recording hemisphere. Delay-period activity has several important features (Funahashi et al., 1989). First, the duration of delay-period activity is either prolonged or shortened depending on the length of the delay period. Second, this activity is observed only when monkeys perform correct behavioral responses. Third, a great majority of delay-period activity exhibits a directional preference, in that it is observed only when a visual cue is presented at a particular area in the visual field. The preferred direction of this activity differs from neuron to neuron. Therefore, it has been proposed that neurons that exhibit directional delay-period activity have mnemonic receptive fields (memory fields) in the visual field, analogous to visual receptive fields. Fourth, it has been shown that the great majority of delay-period activity represents information regarding the direction of the visual cue (retrospective information), whereas the minority represents information regarding the direction of the saccade (prospective information) (Funahashi et al., 1993; Takeda and Funahashi, 2002). Based on these observations, delay-period activity has been considered to be a neural correlate of temporary information-storage processes in working memory (Goldman-Rakic, 1987; Funahashi, 2001; Fuster, 2008).

While a monkey performs the delayed-response task, the ability to retain spatial information regarding the visual cue "on line" as directional delay-period activity is essential for the monkey to correctly perform eye movements in the response period. However, directional delay-period activity is no longer necessary once a saccadic eye movement has been made. The presence of delay-period activity may even be counterproductive if it persists into the next trial, since the subject must refresh his memory in each new trial. Funahashi et al. (1989) showed that delay-period activity was actively terminated immediately after the subject made saccadic eye movements. A great majority of post-saccadic activity exhibited directional selectivity. The distributions of the preferred directions and tuning widths were similar between postand pre-saccadic activities in the prefrontal cortex. These results suggest that post-saccadic activity is a feedback signal from the brain areas that are responsible for the initiation of purposive saccades. Therefore, post-saccadic activation observed in the prefrontal cortex may be a neural signal from these brain areas and may play a critical role in terminating delay-period activity after the execution of eye movement.

In fact, as shown in **Figure 5**, delay-period activity terminated rapidly when saccadic eye movement was initiated. At the same time, post-saccadic activity started increasing. Thus, the termination of excitatory delay-period activity coincided with the initiation of post-saccadic activity based on population analyses of prefrontal activities (Goldman-Rakic et al., 1990; Funahashi and Takeda, 2002). This result supports the idea that post-saccadic activity could play a critical role in terminating delay-period activity after the execution of eye movement. The termination of delay-period activity at the end of the trial ensures that the subject is prepared to receive updated information. This should contribute to the flexible modulation of behavior.

#### **POST-SACCADIC ACTIVITY MAY PARTICIPATE IN PERFORMANCE MONITORING**

Activity that encodes the correctness of the response and activity that detects errors have been observed in the prefrontal cortex in monkeys as well as humans (see reviews by Ridderinkhof et al., 2004; Mansouri et al., 2009). These activities should be important for monitoring responses and enabling the flexible modulation of behavior. It has been proposed that the anterior cingulate cortex contributes to monitoring performance by detecting errors. For example, Carter et al. (1998) observed activation of the anterior cingulate area during erroneous responses in the continuous performance task by human fMRI studies. However, the same area was also activated during correct responses under conditions of increased response competition, in which errors were likely to occur. Similar results have also been reported by Botvinick et al. (1999) using a version of the flanker task. They observed anterior cingulate activation even in correct trials and this activation was greater in high conflict and incongruent trials than in low conflct and congruent trials. The anterior cingulate cortex does not just detect errors, it also participates in the online monitoring of performance.

Post-saccadic activity related to the detection of errors and the monitoring of conflict among processes has been observed in the supplementary eye field. Stuphorn et al. (2000) analyzed the activities of supplementary eye field neurons while monkeys performed a countermanding task and found several types of neurons that exhibited post-saccadic activities. One type of neuron was strongly activated after the initiation of the planned eye movement specifically when monkeys failed to cancel it. Although this activity can be classified as post-saccadic activity, they stated that this activity is distinct from post-saccadic activity observed in the frontal eye field because this activity was not observed in trials without the stop signal, in which monkeys made visually-guided saccades toward the target, and because this activation was observed in both ipsiversive and contraversive saccades. Based on these observations, they concluded that this activity is a signal of the occurrence of the error. A second type of neurons was also activated only when the initiation of the planned eye movement was cancelled. However, this activation was not likely to be related to the cancelling of eye movement because the activity occurred after the stop-signal reaction time, which is the timing after the initiation of eye movement in noncancelled stop-signal trials. When eye movements are cancelled in the countermanding task, gaze-shifting neurons (saccade-related neurons) and gaze-holding neurons need to be co-activated, but this co-activation produces a conflict in processing and the balance of activation between these two groups of neurons is critical for determining whether the eye movement will be cancelled or not. Therefore, they suggested that the second type of activity signals a conflict in processing. Although post-saccadelike activity was observed in the supplementary eye field while monkeys performed a countermanding task, Stuphorn et al. (2000) concluded that these activities represented performance monitoring, such as signals that detect errors and a conflict in processing.

Prefrontal participation in flexible behavioral control and behavioral adjustment to dissolve the conflict has been examined using the anti-saccade task and categorization tasks (Munoz and Everling, 2004; Mansouri et al., 2009). Recently, Teichert et al. (2014) focused on post-saccadic activities in the frontal eye field and tried to understand these activities in more cognitive terms, such as reward expectation, decision uncertainty, or response conflict. If post-saccadic activity in the frontal eye field is related to functions such as reward expectation, decision uncertainty, or response conflict, this activity should differ between correct and error trials. To determine whether the activity in the frontal eye field reflects choice errors, they examined error-related activity while monkeys performed a reward-biased speed-categorization task with delayed feedback. Since monkeys often made impulsive errors in this task, they could analyze a sufficient amount of internally generated error-related activities in the frontal eye field. As a result, they found that error-related activity was observed in a large fraction of frontal eye field neurons and that the magnitude of error-related activity was strongly correlated with the difficulty of the task and not with eye movements. Therefore, post-saccadic activity observed in the frontal eye field might be related to the evaluation of post-decision outcome. Since the evaluation of the preceding choice could play a role in optimizing the subsequent behavior, post-saccadic activity might play a role in monitoring performance.

As stated before, activity that encodes the end of the trial and activity that encodes the correctness of the response have been observed in the prefrontal cortex while monkeys performed cognitive tasks using hand or arm movements. These activities clearly exhibit context-dependent features. In addition, error-related activity that is modulated by the difficulty of the trial also exhibits context-dependent features, as does the post-saccadic activity observed in the prefrontal cortex. Since post-saccadic activity observed in the prefrontal cortex exhibited directional selectivity and was observed only during the response period of the task, this activity might have a specific role, such as terminating delay-period activity. This function could also be considered a type of performance monitoring, since the termination of delay-period activity by post-saccadic activity would be associated with informing the correctness of the response and the termination of the trial. The termination of memory-related activity at the end of the trial ensures that the subject is prepared to receive new and updated information. Therefore, post-saccadic activity observed in the prefrontal cortex may also participate in performance monitoring.

## **CONCLUSIONS**

Neurophysiological studies have revealed that neurons in the lateral prefrontal cortex exhibit saccade-related activity. Although a significant proportion of saccade-related neurons exhibited presaccadic activity, most neurons exhibited post-saccadic activity. Saccade-related activity in the prefrontal cortex showed directional selectivity and context dependency, such that the activity was observed only when the monkey performed goal-directed saccades. The characteristics of directional selectivity and context dependency were similar between pre- and post-saccadic activities. These results indicate that the prefrontal cortex participates in the control of purposive saccadic eye movements, similar to the frontal eye field. Based on the traditional idea that post-saccadic activity is a corollary discharge from oculomotor centers, post-saccadic activity observed in the frontal eye field is a corollary discharge because it has no directional tuning and no context dependency. However, post-saccadic activity observed in the prefrontal cortex is not a corollary discharge because of its context-dependent nature. Therefore, the prefrontal cortex and the frontal eye field might contribute differently to the control of saccadic eye movements. Post-saccadic activity in the prefrontal cortex could play some specific roles in cognitive functions in which the prefrontal cortex participates. For example, the activity that encoded the end of the trial, the activity that encoded the correctness of the response, and error-related activity were observed in the prefrontal cortex when the monkeys performed cognitive tasks. These activities could be important for monitoring the responses and for making flexible changes in behavior. Post-saccadic activity observed in the prefrontal cortex seems to have features similar to these activities, due to the contextdependent nature of post-saccadic activity. In addition, postsaccadic activity might act to terminate memory-related activity after a behavioral response is done. The termination of memoryrelated activity at the end of the response behavior ensures that the subject can prepare to receive updated information. This idea was supported by the observation that the termination of memory-related activity coincided with the initiation of post-saccadic activity. In addition, a recent study shows that post-saccadic activity observed in the frontal eye field as well as the supplementary eye field might be related to the evaluation of post-decision outcome. The evaluation of the preceding choice could play a role in optimizing the subsequent behavior. Therefore, post-saccadic activity might play a role in monitoring performance and making flexible changes in behavior. However, further investigations are needed to examine the characteristics of saccade-related activities observed in the prefrontal cortex and their possible contributions to eye movement control and prefrontal functions.

#### **ACKNOWLEDGMENTS**

This work was supported by Grants-in-Aid for Scientific Research from the Japanese Ministry of Education, Culture, Sports, Science, and Technology (MEXT) (21240024 and 25240021) to Shintaro Funahashi.

#### **REFERENCES**


**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: 10 April 2014; accepted: 10 June 2014; published online: 30 June 2014*. *Citation: Funahashi S (2014) Saccade-related activity in the prefrontal cortex: its role in eye movement control and cognitive functions. Front. Integr. Neurosci. 8:54. doi: 10.3389/fnint.2014.00054*

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

*Copyright © 2014 Funahashi. 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 amblyopic eye in subjects with anisometropia show increased saccadic latency in the delayed saccade task

#### *Maciej Perdziak1 \*, Dagmara Witkowska1, Wojciech Gryncewicz 1, Anna Przekoracka-Krawczyk2 and Jan Ober <sup>1</sup>*

*<sup>1</sup> Laboratory for Oculomotor Research, Department for Biophysical Measurements and Imaging, Nał ˛ecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland*

*<sup>2</sup> Laboratory of Vision Science and Optometry, Faculty of Physics, Adam Mickiewicz University of Poznan, Poznan, Poland*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

*Reviewed by:*

*Xuefeng Shi, Tianjin Medical University, China Simon Clavagnier, McGill University, Canada Marianne Emma Florence Piano, Glasgow Caledonian University, Scotland*

*\*Correspondence:*

*Maciej Perdziak, Laboratory for Oculomotor Research, Polish Academy of Sciences, 60-195 Poznan, Szeherezady 132, Poland e-mail: mperdziak@ibib.waw.pl*

The term *amblyopia* is used to describe reduced visual function in one eye (or both eyes, though not so often) which cannot be fully improved by refractive correction and explained by the organic cause observed during regular eye examination. Amblyopia is associated with abnormal visual experience (e.g., anisometropia) during infancy or early childhood. Several studies have shown prolongation of saccadic latency time in amblyopic eye. In our opinion, study of saccadic latency in the context of central vision deficits assessment, should be based on central retina stimulation. For this reason, we proposed saccade delayed task. It requires inhibitory processing for maintaining fixation on the central target until it disappears—what constitutes the GO signal for saccade. The experiment consisted of 100 trials for each eye and was performed under two viewing conditions: monocular amblyopic/non-dominant eye and monocular dominant eye. We examined saccadic latency in 16 subjects (mean age 30 ± 11 years) with anisometropic amblyopia (two subjects had also microtropia) and in 17 control subjects (mean age 28 ± 8 years). Participants were instructed to look at central (fixation) target and when it disappears, to make the saccade toward the periphery (10◦) as fast as possible, either left or the right target. The study results have proved the significant difference in saccadic latency between the amblyopic (mean 262 ± 48 ms) and dominant (mean 237 ± 45 ms) eye, in anisometropic group. In the control group, the saccadic latency for dominant (mean 226 ± 32 ms) and non-dominant (mean 230 ± 29 ms) eye was not significantly different. By the use of LATER (Linear Approach to the Threshold with Ergodic Rate) decision model we interpret our findings as a decrease in accumulation of visual information acquired by means of central retina in subjects with anisometropic amblyopia.

**Keywords: amblyopia, anisometropia, saccadic latency, anisometropic amblyopia, delayed saccade, saccadometry**

## **INTRODUCTION**

The term *amblyopia* is used to describe reduced visual function in one eye (or both eyes, though not so often) which cannot be fully improved by refractive correction and explained by the organic cause observed during regular eye examination (Barrett et al., 2004). However, amblyopia is reversible (especially when detected in early childhood) in most cases by proper therapy. This is usually a unilateral developmental disorder of spatial vision which affects about 2–5% of the population and is associated with abnormal visual experience during infancy or early childhood (Roper-Hall, 2007). The most common risk factors for developing amblyopia are anisometropia (unequal refractive error in the two eyes), strabismus (misalignment of visual axes) and another form of deprivation (e.g., congenital cataract) early in life (Kiorpes and McKee, 1999; Roper-Hall, 2007). In the case of anisometropic amblyopia, peripheral retina is generally normal (Yu et al., 1998; Pardhan and Whitaker, 2000), and spatial vision deficits are the consequence of chronic blur in the area of central retina during sensitive period for development of visual acuity (see Daw, 1998). Anisometropia becomes clinically significant when its magnitude reaches approximately 1 D in either one or both meridians (Benjamin, 2006). It is generally accepted that strabismic and anisometropic amblyopia are associated with an active inhibition of visual input originating in the fovea of the deviating and more ametropic eye respectively (Von Noorden and Campos, 2002). In practice, amblyopia is recognized on the basis of reduced visual acuity on Snellen chart, despite the optimal refractive correction and a period of refractive adaptation (Stewart et al., 2004). Apart from evident deficits in optotype acuity, subjects with amblyopia also manifest deficits in other visual functions including reduced contrast sensitivity (Bradley and Freemen, 1981; McKee et al., 2003), spatial and temporal crowding (Bonneh et al., 2007), reduced grating and Vernier acuity (McKee et al., 2003), spatial uncertainty (Demanins and Hess, 1996), temporal instability and spatial distortions (Sireteanu et al., 2008) or prolongation of visual reaction time for amblyopic eye (Hamasaki and Flynn, 1981; Nuzzi et al., 2007). Increased latency of the evoked potentials was also reported in amblyopia (Sokol, 1983; Parisi et al., 2010) but features such as critical flicker frequency and color vision remains generally normal (Roper-Hall, 2007). Moreover, subjects with amblyopia often suffer lack of (or reduced) binocular function, especially stereopsis (Roper-Hall, 2007).

Quite a high number of neuroimaging (see Anderson and Swettenham, 2006) and psychophysical (see Von Noorden and Campos, 2002) studies were conducted in amblyopic subjects but surprisingly only few of them were focused on oculomotor behavior (especially saccades). Saccades are fast (up to ∼500◦/s) and brief (typically ∼30–100 ms) eye movements which redirect the fovea between successive points of visual scene (Leigh and Zee, 2006; Munoz et al., 2007). The central retina (more precisely *fovea centralis* with the central pit called *foveola* ∼0.35 mm across and ∼1.2◦ of the visual field) is physiologically predestined for acquisition of the visual image with high resolution and greatest visual acuity (Moses and Hart, 1987; Munoz et al., 2007). Apart from obvious gaze shifting function, human oculomotor system also ensures the stable retinal image and prevents visual fading. Clear and stable retinal image is needed for proper visual development and perception. Saccades are generated in the brainstem and triggered at the level of cerebral hemispheres (Pierrot-Deseilligny et al., 2002). There is a specific three–level hierarchy in oculomotor system, that can be summarized as *what* (recognition), *where* (localization) and *how* (pattern of neuronal excitation required for execution) (Carpenter, 2004a) (**Figure 1**). At the lowest level of this hierarchy are horizontal and vertical gaze centers, respectively the burst neurons in the paramedian pontine reticular formation (PPRF) and the burst neurons in the midbrain rostral interstitial nucleus of the medial longitudinal fasciculus (riMLF) (Leigh and Zee, 2006). They send saccadic command to the motoneurons innervating horizontally and vertically acting extraocular muscles (Leigh and Zee, 2006). Above the PPRF and riMLF level are the superior colliculus (SC)—area important for target selection and initiation of eye movement (this task is

eye field (FEF), the parietal eye field (PEF) and the supplementary eye field (SEF) involved in recognition, inhibition, and decision about the movement. supplemented also by the cerebellum and the cortex) (Carpenter, 2004a; Leigh and Zee, 2006). At the top of this hierarchy are the cortical areas involved in saccade control by inhibition of unnecessary collicular mechanisms. They can be thought of as preventing the collicular route from operating, so the saccadic reaction times are longer than they might otherwise be (Carpenter, 2004a). Experimental studies (Schall and Hanes, 1993; Gaymard et al., 1998; Johnston and Everling, 2008) have shown that three cerebral regions are involved in triggering saccades: the frontal eye field (FEF), the parietal eye field (PEF) and the supplementary eye field (SEF). The FEF is mainly involved in the control of voluntary saccades (Vernet et al., 2014) (e.g., intentional—visually guided saccades, antisaccades, delayed saccades, memory guided saccades). The SEF appears crucial in learning phase (presentation of the visual target) (Müri et al., 1995) and just after the GO signal for saccade (Müri et al., 1994). The PEF is believed to be involved in the triggering of reflexive, visually guided saccades (Hanes and Schall, 1996; Pierrot-Deseilligny et al., 2002). One of the most interesting parameters of saccadic eye movement is the saccadic reaction time (latency). The interval between stimuli presentation and the beginning of eye movement is typically about 200–250 ms and is surprisingly variable (Hanes and Schall, 1996; Carpenter, 1999, 2004b; Liversedge et al., 2011) Saccadic latency is also much longer than it would be expected from synaptic delays and nerve conduction (the retino–collicular route takes about 60 ms) (Carpenter, 1999) because of higher, cortical levels involved in decisions about whether to respond to a stimulus or not (Carpenter, 1988, 1999; Reddi and Carpenter, 2000).

Generation of saccades is closely connected with decision making processes (reflected by the saccadic latency). Carpenter's model LATER (Linear Approach to Threshold with Ergodic Rate) suggests that at the moment of target presentation, the decision signal starts from initial level S0 and rises linearly with a rate (r) until it reaches the decision threshold ST which causes the generation of saccade (**Figure 2**). The rate of rise (r) varies randomly from trial to trial (about a mean μ and with variance σ2) and its variability exhibits characteristics of a normal distribution. According to the model LATER the level S0 represents the logarithm of the prior probability whilst the threshold ST reflects the urgency of reaction. The μ (mean of the rate of rise of the decision signal as well as the mean reciprocal latency) (Noorani, 2014) can be treated as the supply of information (Carpenter, 1981; Reddi and Carpenter, 2000) whereas the σ reflects the variability in the latency distributions (Carpenter and Williams, 1995; Noorani, 2014). Saccadic latency, is characterized by a skewness of distribution toward the greater values (Reddi and Carpenter, 2000). Although the variability of saccadic reaction time (latency) is large, in general, under high-contrast conditions, it follows a simple rule: the reciprocal of latency obeys a Gaussian (*recinormal*) distribution (Carpenter, 1988). Since latency and rate (r) are reciprocally related, Carpenter (1981) proposed to analyze not the distribution of latency (T) but the distribution of its reciprocal 1/T (*promptness*). When we plot the latency distribution as a function of its promptness, we obtain a curve that is very close to normal distribution (Carpenter, 1988). Using a probit scale, we can transform the cumulative Gaussian distribution to a straight line (a *reciprobit plot*, where the *x* axis has a reciprocal

scale and the *y* axis has a probit scale) whose intercept with the 50% line represents the median and whose slope is directly related to standard deviation (SD) (Reddi et al., 2003; Coubard, 2012). The intercept represents the probability of not making a saccade at all (Carpenter, 1981; Reddi and Carpenter, 2000). Changes in the parameters of the LATER model have distinct effects on the reciprobit plot (**Figure 2**). Alternations in the distance between ST and SO causes swiveling of the distribution line about the point (I) where it intersects the vertical infinite-time axis. Parameter ST can be modified due to a different instruction given to the subjects ("react as fast as possible" or "react as accurate as possible"). Variation in S0 can be caused by a different probability of the target dislocations (Carpenter, 1981, 1999; Reddi and Carpenter,

by So change, but ST may also by modified) causes swiveling of the

2000; Reddi et al., 2003). Changes in average rate of rise (μ) shifts the plot parallel along the time axis (leftward-reduction of latency or rightward-increase of latency) without a change in the slope (Carpenter, 1999; Reddi and Carpenter, 2000; Reddi et al., 2003; Coubard, 2012). The parallel shift of reciprobit plot reflects changes in the rate of information supply to the visual system (Reddi and Carpenter, 2000; Coubard, 2012).

Only few studies have dealt with saccadic latency in amblyopia and most of them have shown the increased latency during viewing with the amblyopic eye (see Mackensen, 1958; Ciuffreda et al., 1978; Niechwiej-Szwedo et al., 2010). However, previous studies of saccadic latency in amblyopia were focused on reflexive saccades, which are initiated in response to novel exogenous stimuli

change in the slope.

(Johnson et al., 2012), and neither of them have examined more complex saccadic responses (e.g., delayed saccades), which additionally involve the frontal cortex (FEF) and enable the central (visually deprived) retina to be engaged in a higher degree in the programming and saccade execution process. Reflexive saccades to the location of novel target depend primarily upon the direct projections from the occipital (visual) and parietal cortices to the superior colliculus (LeVasseur et al., 2001; McDowell et al., 2008). More complex, volitional saccades (e.g., delayed saccades) require additional neural regions to support the higher level processes (e.g., inhibition) and depend more upon the frontal cortex and its direct or indirect (via basal ganglia) projections to the superior colliculus (LeVasseur et al., 2001; McDowell et al., 2008).

In our opinion, the study of saccadic latency in the context of central vision deficits assessment, should be based on central retina stimulation. For this reason, we decided to study more complex saccadic responses during saccade delayed task. Proposed saccadic paradigm, requires inhibitory processing for maintaining fixation on the central target until it disappears, what constitutes the GO signal for saccade (Pierrot-Deseilligny et al., 2002; Munoz et al., 2007). In order to execute this kind of saccade, several processes are believed to occur: computation of the parameters of the movement, inhibition of the already prepared saccade, disengagement of visual attention from fixation position and finally decision to move (Coubard et al., 2003). In proposed saccadic paradigm, we may distinguish two phases:


Taking all into consideration, the central retina has a dominant effect on observed visual deficits in subjects with anisometropic amblyopia (Yu et al., 1998; Pardhan and Whitaker, 2000), thus we expect that the loss of physiological function of foveal vision in these individuals may affect specifically saccades in terms of their spatial and temporal properties during saccade delayed task.

Our hypothesis is, that subjects with anisometropic amblyopia will reveal differences in the saccadic latency distribution (increased latency in amblyopic eye in comparison to dominant eye), predominantly by parallel (rightward) shifting of the reciprobit plot. According to the LATER model this should be interpreted as a difference in supply of information (decrease in accumulation of visual information what delays the saccade execution decision) between the amblyopic and dominant eye.

Under physiological conditions, saccadic latencies in both eyes are generally equal (Carpenter, 1988), so it is reasonable (especially in cases of unilateral amblyopia) to compare saccadic latency distribution between the eyes. Thus, the aim of the current study is to investigate saccadic latency (by the use of the delayed task) in subjects with anisometropic amblyopia.

## **METHODS**

### **PARTICIPANTS**

Sixteen participants with anisometropic amblyopia (mean age 30 ± 11 years) and seventeen control subjects (mean age 28 ± 8 years) took part in the study. All subjects underwent standard optometric examination, including the measurement of visual acuity (Snellen chart – decimal notation), refractive error examination (static retinoscopy and subjective refraction), the binocular vision examination (cover test at distance, Worth 4 dot test at far and near, stereopsis-stereo Fly test, phoria measurement at distance – Maddox) and slit lamp (anterior segment) examination. Since anisometropia may be associated with microtropia (Hardman Lea et al., 1991), we performed also several additional tests (Bagolini test, 4 base-out prism test and fixation pattern – visuscopy) in order to diagnose the existence of microtropia, defined as small angle heterotropia of less than 5◦ associated with harmonious anomalous retinal correspondence (ARC) (Lang, 1974). In addition, in subjects with microtropia amblyopia, normal or near normal peripheral fusion, reduced or absent stereoacuity, eccentric fixation, foveal (central) suppression scotoma are often present (Houston et al., 1998; Von Noorden and Campos, 2002).

In our experimental group fourteen subjects were orthotropic and two (No. 7 and No. 8) had microtropia "with identity" (no manifest movement on cover test, the eccentric fixation point coinciding with the angle of ARC) (Houston et al., 1998) (see **Table 1** for clinical characteristic of individuals with amblyopia). Anisometropic (experimental) group (AG) consisted of subjects with anisometropic amblyopia according to the following criteria:


Seventeen subjects in the control group (CG) had normal or corrected to normal visual acuity (20/20 or better) in each eye and normal binocular vision with good stereopsis (at least 40 arcseconds).

Taking into consideration that two lines of Snellen acuity (BCVA) difference is considered as amblyopia, we classified amblyopia as severe (BCVA worse than 0.2), moderate (BCVA: 0.2–0.5) and mild (BCVA: 0.6–0.9) (Wright et al., 2006; Sapkota et al., 2013).

The experimental protocols were approved by the local ethic committee at Poznan University of Medical Sciences and all


*eye; LE, left eye;* # *represent subject with no differences between the dominant and non-dominant eye in saccadic latency.* experiments were conducted in accordance with the ethical standards laid down in the 1964 Declaration of Helsinki.

#### **EXPERIMENT PROCEDURE**

The subjects were seated in front of the white, uniform wall at the distance of 3 m. The overall room illumination was medium. The visual stimuli for horizontal saccade subtended ∼4.5 arcminute. Before the experiment, participants performed the practice trials, consisted with 10 saccades either to the left or right. The experiment consisted of 100 trials for each eye. The experiment was performed under two viewing conditions: monocular amblyopic/non-dominant eye and monocular dominant eye. There was one minute break between trials. The order of viewing conditions was randomized across the subjects, starting either with the dominant or non-dominant eye. The experiment was performed in silence and all subjects have used their optimal refractive correction. In the course of the experiment, the central fixation stimuli (red laser spot) were generated at the beginning, and after 200 ms the peripheral (10◦) either left or right stimuli for saccadic refixation response is displayed. The signal for saccadic refixation is given by the disappearance (randomized in time) of central stimuli (GO signal for saccade). Participants were instructed to look at central (fixation) target and when it disappears to make as quickly as possible the saccade toward the peripheral target. The graphical representation of experimental procedure is presented in **Figure 3**.

#### **APPARATUS**

The stimulus, for subject's visual system, is displayed using the miniature laser projectors mounted on the sensor forehead plate. Eye movements (saccadic latencies) were recorded using the Saccadometer Ober Consulting (Ober et al., 2009). The system measures the eye movements using direct infra-red oculography along the horizontal axis (±35◦) with high temporal (1 kHz) and spatial (0.1◦) resolution. The onset of saccadic response was detected on-line by the measuring system, using 5◦/s velocity threshold. The saccadic latencies were measured as the time between the onset of the GO signal (disappearance of the central fixation stimuli) and the onset of saccadic response and stored in Saccadometer memory. They were later transferred to the PC using optical-USB connection, and analyzed there by means of the software provided by the Saccadometer manufacturer.

#### **DATA ANALYSIS AND STATISTICS**

Blinks, error responses (saccades characterized by incorrect direction or amplitude) as well as trials without any responses were rejected from the analysis. Moreover we excluded saccades characterized by latencies smaller than 80 ms and greater than 700 ms. Additionally, the saccades with latencies above and under 2.5 SD were rejected (Van der Stigchel et al., 2010).

For every subject mean, and coefficient of variation (CV) of saccadic latency for dominant eye (DE) and amblyopic eye (AE)/non-dominant eye (NDE) was calculated. Also, we calculated also best fit LATER parameters (σ and μ of the rate of rise) by minimization of the Kolmogorov-Smirnov one-sample test (using SPIC software; Carpenter, 1994).

Data was analyzed using IBM SPSS Statistics. The normality of the gathered data was evaluated using Kolmogorov-Smirnov

**Table 1 | Clinical** 

**characteristic**

 **of individuals**

 **with** 

**anisometropic**

 **amblyopia.**

*RE, right* 

test. To analyze average values of saccadic latency, its CV and LATER parameters (σ and μ of the rate of rise) we used repeated-measures ANOVA, with one between-subjects factor (group [control subjects and anisometropic subject]) and one within-subjects factors (viewing condition [DE and AE / NDE]). Significant effects of interaction were analyzed further using *posthoc* Bonferroni-corrected Fisher's LSD test. In the anisometropic group, using Spearman's Rho we calculated correlation between the degree of amblyopia, the mean and variability of saccadic reaction time, as well as σ and μ of the rate of rise for amblyopic eye viewing. For each subject, we also tested an individual differences for DE and AE / NDE (for every subject separately) using SPIC software. Two-sample Kolmogorov-Smirnov was used to compare the observed distributions, while using Student's *t*-test we analyzed individual differences between means of distributions.

#### **RESULTS**

#### **MEAN SACCADIC LATENCY**

The mean saccadic reaction times for particular groups in the condition of DE and AE/NDE viewing are presented in **Figure 4**. The differences in latencies between both eyes were observed. Latency for AE/NDE was longer than for DE [AE/NDE = 246 ± 42 ms vs. DE <sup>=</sup> <sup>231</sup> <sup>±</sup> 38 ms; *<sup>F</sup>*(1, 31) <sup>=</sup> <sup>19</sup>.586, *<sup>p</sup>* <sup>&</sup>lt; <sup>0</sup>.001, <sup>η</sup><sup>2</sup> <sup>=</sup> 0.387], which was proved by significant main effect of *visual condition*.

Results from the statistical tests have shown that in the anisometropic group (AG), the difference in saccade latency between two eyes was increased (262 ± 48 ms vs. 237 ± 45 ms for AE and DE, respectively) compared to the control group (CG) (230 ± 29 ms vs. 226 ± 32 ms for NDE and DE, respectively),

which was confirmed by the significant *group* x *viewing condition* interaction [*F*(1, 31) <sup>=</sup> <sup>10</sup>.572, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.003, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.254]. *Post-hoc* Bonferroni test gave a significant result only for the AG (mean difference = 25.000, *p* < 0.001), but not for the CG (mean difference = 3.824, *p* = 0.406). Saccadic reaction times for AE in the anisometropic group was increased as compared to latencies in the NDE in control group, which was revealed by the *posthoc* test (mean difference = 32,669, *p* = 0.024). Latencies for DE did not differ between the groups (mean difference = 11.493, *p* = 0.397).

#### **VARIABILITY OF SACCADIC LATENCY**

The CV of reaction time for particular groups in the condition of DE and AE/NDE viewing is presented in **Figure 5**. In the anisometropic group CV was comparable between DE and AE viewing conditions (AE: 22 ± 8%; DE: 20 ± 7%). The same was in the case of control group (CV for NDE viewing 20 ± 6%; for DE viewing 19 ± 6%), which was also observed in the analysis that has shown insignificant main effect of *visual condition* [*F*(1, 31) = <sup>1</sup>, 903, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.178, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.058] and insignificant *group* <sup>×</sup> *viewing condition* interaction [*F*(1, 31)= 0.256, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.616, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.008]. The CV was comparable between the groups, which was proved by the no effect of the group [*F*(1, 31) = 0.601, *p* = 0.444, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.019].

#### **INDIVIDUAL ANALYSIS**

In the experimental group, significant differences were observed in eight subjects in the latency distribution, which was manifested by the significant increase in the mean latency for the AE. Additionally, the increase in the mean latency for AE (without differences in the distributions) was found in three subjects. Only five out of sixteen subjects did not demonstrate any differences between AE and DE. In the control group, significant differences in the latency distribution were observed only in one subject, manifested by the significant increase in the mean latency for NDE. Additionally, the increase in the mean latency for NDE

without differences in the distributions were found in four subjects. Twelve out of seventeen controls did not demonstrate any differences between DE and NDE.

#### **CORRELATIONS WITH THE DEGREE OF AMBLYOPIA**

The analysis (using Spearman's Rho) did not reveal any correlations between the degree of amblyopia and the results of mean value and variability in reaction time, as well as σ and μ of the rate of rise for amblyopic eye viewing.

#### **LATER PARAMETERS**

#### *Mean μ of the rate of rise*

The mean (μ) of the rate of rise for particular groups in the condition of DE and NDE/AE viewing is presented in **Figure 6**. The differences in μ of the rate of rise between both eyes were observed. The μ decreased for the NDE / AE, which was confirmed by the main effect of viewing condition [AE/NDE = 4.46 ± 0.67 vs. DE = 4.71 ± 0.67; *F*(1, 31) = 21.720, *p* < 0.001, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.412]. As can be seen in the anisometropic group, the difference in the μ between both *visual condition* was higher (AE: 4.20 ± 0.67 Hz; DE: 4.60 ± 0.72 Hz) than in the control group (NDE: 4.70 ± 0.61 Hz; *DE*: 4.81 ± 0.64 Hz), which was proved by the significant *group* × *viewing condition* interaction [*F*(1, 31) = <sup>6</sup>.677, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.015, <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.117]. *Post-hoc* Bonferroni test revealed significant effect only in the anisometropic group (mean difference = 0.389, *p* < 0.001) but not in the control group (mean difference = 0.114, *p* = 0.146). The μ of the rate of rise for AE in the experimental group decreased when compared to μ for the NDE in the control group which was confirmed by the *post-hoc* analysis (significant only for the AE/NDE condition, mean difference = −0.496, *p* = 0.032). The μ of the rate of rise for DE did not differ between the groups (mean difference = −0.212, *p* = 0.376).

#### *Variability (σ) of the rate of rise*

The σ of the rate of rise for particular groups in the condition of DE and NDE viewing is presented in **Figure 7**. The σ of the rate

**subjects for particular viewing condition.** Anisometropic subjects performed decreased μ during viewing with the amblyopic eye as compared to viewing with the dominant eye. Error bars, ±1 s.e.m.

**FIGURE 7 | Variability (***σ***) of the rate of rise for controls and anisometropic subjects in particular viewing condition.** There were no significant differences between viewing conditions neither in amblyopes nor in controls. Error bars, ±1 s.e.m.

of rise was comparable between two viewing conditions both in the anisometropic group (for AE: 0.85 ± 0.21 Hz; for DE: 0.83 ± 0.19) and in the control group (for NDE: 0.89 ± 0.27 Hz; for DE: 0.84 ± 0.26 Hz). It was confirmed by the statistical analysis, showing insignificant main effect of viewing condition [*F*(1, 31) = <sup>1</sup>.381, error *df* <sup>=</sup> 31, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.<sup>249</sup> <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.043] and insignificant *group* × *viewing condition* interaction [*F*(1, 31) = 0.369, error *df* <sup>=</sup> 31, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.<sup>548</sup> <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.012). The CV did not differ between the groups, which was proved by the insignificant effect in the group [*F*(1, 31)= 0.075, error *df* <sup>=</sup> 31, *<sup>p</sup>* <sup>=</sup> <sup>0</sup>.<sup>786</sup> <sup>η</sup><sup>2</sup> <sup>=</sup> <sup>0</sup>.002].

#### **DISCUSSION**

The main finding of this study is the significant difference of saccadic reaction time between the amblyopic and dominant eye in subjects with anisometropic amblyopia. In the control group, saccadic latency for dominant and non-dominant eye was not significantly different. Other authors also reported prolongation of saccadic latency in subjects with amblyopia (see Ciuffreda et al., 1978; Niechwiej-Szwedo et al., 2010). To the best of our knowledge, this is the first study of saccadic latency, carried out using the delayed task in individuals with anisometropic amblyopia. Although delayed saccades are usually applicable when investigating the ability to suppress "automatic" visually triggered saccades (e.g., in Parkinson's disease, Tourette Syndrome or ADHD) (see Munoz et al., 2007), we implement this paradigm in order to engage the central retina in the programming and saccade execution process. However, during preliminary phase of this study, we observed that simultaneous presentation of the central and peripheral target, causes difficulties in the selection of the central point/target during amblyopic eye viewing. For this reason, we decided to use a 200 ms interval between the appearance of the central (as first) and peripheral target in order to indicate which of them is the central. Such a short time interval should not affect the preparatory phase and from the practical point of view appeared very helpful.

Ciuffreda et al. (1978) studied saccadic latencies in 13 subjects (six with constant strabismic amblyopia, three with amblyopia without strabismus and four with intermittent strabismus with or without amblyopia). In the group of constant strabismic amblyopia, the researchers reported an increased latency in three out of six subjects. In the group of amblyopia without strabismus (anisometropic type) they observed increased saccadic latency in two out of three subjects. In the group of intermittent strabismus two out of four subjects had amblyopia and also had longer latency. Unfortunately in Ciuffreda's study (1978) there was no control group. Moreover, there were only three subjects with anisometropic amblyopia and such a small group is not statistically significant. They reported an overall increase of saccadic latency in seven out of eleven amblyopic patients, but they have used different saccadic paradigm, intended for reflexive saccades (fixation target disappears and at the same time appear the peripheral target appears). It is worth to mention, that both in our and Ciuffreda et al. (1978) study not every amblyopic eye showed increased saccadic latency. We cannot exclude the situation that speediness of saccadic decision making is modulated by the compensatory mechanisms (positive and negative outcome of cortical adaptation), what may explain the lack of increase in saccadic latency in some of the amblyopes. Ciuffreda et al. (1978) reported also that after successful amblyopia treatment saccadic latency remained abnormally high. However, they examined only one subject after therapy and future studies should focus on exploring the influence of amblyopia treatment on saccadic latency (we are going to study this topic in the near future).

The study by Niechwiej-Szwedo et al. (2010) included 13 subjects with anisometropic amblyopia and, in contrast to the previous studies, they examined saccadic latency and additionally the saccadic amplitudes and peak velocities during visuomotor task including looking at and pointing with right index finger to a visual target. The saccadic paradigm used by Niechwiej-Szwedo et al. (2010) was also intended for reflexive saccades. Despite of increase saccadic latencies for amblyopic eye (AE: 236 ± 72 ms, DE: 191 ± 38 ms), they reported also an increase in the variability of saccadic latency and no binocular advantage, usually manifested as the reduction of saccadic latency under binocular viewing conditions. Reaching accuracy and reaction time (353 ± 66 ms for patients and 334 ± 86 ms for controls subjects) was comparable between patients and control group but anisometropic amblyopes had significantly longer total mean movement time (Niechwiej-Szwedo et al., 2011). The mean amplitudes and peak velocities were comparable between patients and controls, however they reported greater variability of amplitudes in amblyopic subjects. In the control group, mean saccadic latency was comparable for both right (195 ± 31 ms) and left (192 ± 29 ms) eye (which is in good agreement with our observation) and was significantly shorter for binocular viewing condition (175 ± 32 ms). It is worth to mention that pointing and reaching task requires additional involvement of the dorsal visual stream that has been primarily associated with visually guided reaching and grasping (Hebart and Hesselmann, 2012). To the needs of this study we have assumed that, the releasing of saccadic response is the process taking place exclusively within visual-oculomotor system, and for this reason we decided to compare the findings of saccadic latency measured during visuomotor task with our results. Niechwiej-Szwedo et al. (2010) recorded eye movements binocularly with sampling rate 200 Hz, which results in 5 ms temporal resolution. In our opinion, when we want to analyze differences in saccadic latency in scale of several milliseconds, it is reasonable to use higher sampling rate (at least 300 Hz acc. Juhola et al., 1985; 1000 Hz acc. Ober et al., 2009). Moreover, in the Niechwiej-Szwedo et al. (2010) study there were only 20 saccades for each direction and amplitude, which provided overall 80 saccades per eye. Saccadic latency time constitutes rather a capricious object for the investigator and in order to achieve confidence intervals at the level of ±3 ms we need to acquire at least 100 trials (Ober et al., 2009). Niechwiej-Szwedo et al. (2010) concluded that longer saccadic latency for amblyopic eyes reflects rather the slower afferent (sensory) visual processing than a deficit in the efferent (motor) pathway of the saccadic system. The mean value of saccadic latency reported by Niechwiej-Szwedo et al. (2010) is reduced in comparison to our finding. This is not surprising because it is well known that for the young healthy subject, reflexive saccade latency is significantly faster (typically below 200 ms) than the latency of all forms of voluntary saccade (typically above 200 ms) (Walker et al., 2000; Van Stockum et al., 2011). Both the study by Ciuffreda et al. (1978) and Niechwiej-Szwedo et al. (2010) reported an increase in the variability of saccadic latency during amblyopic eye viewing. This finding was not confirmed by our study. Authors of mentioned studies evaluated variability of latency using its standard deviation (SD). We decided to estimate the coefficient of variation (CV) which, as opposed to SD (that determines absolute differences in characteristics) is a relative measure of the features variability (CV = SD/Mean × 100%). Therefore, the increase of SD may not entail the increase of CV when is accompanied with the adequate growth of mean value (both mentioned studies reported about the simultaneous increase of SD and mean value in the amblyopic viewing condition).

The function of visual pathway from photoreceptors to the visual cortex can be evaluated by means of visual evoked potentials (VEP) recordings. The waveform of VEP contains several characteristic peaks, and in the context of our findings, the most interesting are the peaks C1 (typically 55–70 ms post stimuli onset, generated in primary visual cortex) (Jaskowski, 2009; ´ Kolb et al., 2013) and P100 (typically 95–110 ms post stimuli onset, generated in dorsal extrastriate cortex of the middle occipital gyrus) (Jaskowski, 2009; Kolb et al., 2013 ´ ). Results of VEP responses in amblyopia are rather conflicting and most of the existing studies were focused on the amplitude and latency of the P100. Some authors reported abnormal VEP responses (reduced amplitude and increased P100 latency) in amblyopic subjects (see Sokol, 1983; Parisi et al., 2010) and some did not find any differences between the amblyopic and the sound eye (especially for anisometropic amblyopia) (see Chung et al., 2008; Halfeld Furtado de Mendonca et al., 2013). Parisi et al. (2010) suggested that a delay in postretinal conduction in amblyopia may be responsible for abnormal cortical VEP responses. They reported mean increase ∼10 ms (for 25 subjects with anisometropic amblyopia) of the P100 latency. Despite the lack of coherent VEP results identifying the simple neural transportation delay, we cannot exclude the potential influence of such delay on our results.

It is well known, that amblyopia is a cortical deficit, and it is widely accepted view that the primary site of neural deficit in amblyopia is the primary visual cortex (V1), where the information from two eyes is first combined (Hendrickson et al., 1987; Barrett et al., 2004; Roper-Hall, 2007). However, physiological deficits in area V1 are not sufficient to fully explain the whole perceptual deficits observed in amblyopic subjects (Anderson and Swettenham, 2006). Several recent neuroimaging studies indicates that cortical deficits associated with amblyopia are localized within and beyond area V1 (see Anderson and Swettenham, 2006). Hence, it seems to be reasonable to study not only the latency of reflexive saccades but also the latency of voluntary saccades (e.g., delayed saccades) that involves additionally higher cortical processing in the greater extend than the simple saccadic refixation.

The supply of information to the visual system can be considered in two dimensions: the speediness and equally important correctness/completeness and both are required for effective saccades programming. For this reason we interpret our finding on the basis of reduced rate of rise of neural activation in the cells involved in saccade initiation for amblyopic eye viewing. As we will discuss in more detail later, the rate of rise of neural activation represents the rate of rise of the decision signal, which (according to the Carpenter's LATER decision model) depends on the rate of information supply to the visual system. We have applied the distributional analysis carried out by means of reciprobit plot, and combined it with the LATER model (Carpenter, 1988; Wardak et al., 2012), in order to plot, compare and interpret the changes in the saccadic reaction time for subjects with anisometropic amblyopia. To the best of our knowledge this the first study that makes such an effort. Neurophysiological evidences supporting the Carpenter's model comes from the studies on rhesus monkeys. Hanes and Schall (1996) studied neural activity of single cells in the FEF, an area that plays a central role in production of voluntary saccades. They reported that neural activity of the cell began to increase ∼100 ms before the saccade initiation and peaked shortly after saccade initiation. Saccadic eye movements were initiated only if the neural activity reached a specific and constant threshold level (Hanes and Schall, 1996). The study by Hanes and Schall (1996) has revealed the population of saccadic movement-related neurons, whose activity corresponds closely with the rise-to-threshold of LATER's decision signal (Hanes and Schall, 1996; Reddi et al., 2003). According to the LATER model, the neural signal rises linearly in each trial from an initial level SO to a threshold level ST, which initiates the saccade. The rate of rise varies randomly between trials in a Gaussian fashion with mean μ (Reddi et al., 2003). Based on the 100 saccadic responses for each eye, we found significant decrease in mean (μ) rate of rise in anisometropic group (**Figure 6**) during amblyopic eye viewing, evidenced by the rightward parallel shift of corresponding reciprobit plots (**Figure 8**). This may be interpreted as a difference in supply of information (decrease in accumulation of visual information) acquired by means of the central retina, that delays the saccade execution decision. In the control group, the mean rate of rise was not significantly different between the eyes (**Figure 6**), evidenced by the superposition of the corresponding reciprobit plots (**Figure 9**). **Figure 10** presents the subject without any detectable visual problems, however we reported surprisingly significant difference in saccadic latency between the eyes (manifested by the significant increase of mean latency for non-dominant eye). We cannot exclude the situation that this subject was not properly motivated during experimental procedure because when we re-examined this subject several weeks later there were no differences in saccadic latency between the eyes.

We would like also to briefly describe the history of subject No. 15. In this case, the visual acuities of RE and LE were 0.7 and 1.2, respectively, during the subject's first visit, and then a proper refractive correction (including full astigmatism correction of the RE) was prescribed in the secondary school. After several months of regular wearing the new glasses, the visual acuity of the RE improved but still was slightly reduced. Although the experimental test for this subject was performed after treatment, we still decided to join this subject to the experimental group since there are still two lines difference in the Snellen visual acuity between RE and LE. Abnormal VEP response was also recently reported in this subject. In **Figure 11**, the subject with mild amblyopia (No. 3) in the right eye was presented. In this case, saccadic latency was paradoxically symmetrical in the right and left eye. The individual differences observed in some subjects are quite likely to rise from the compensatory mechanisms at the cortical level (e.g., subject No. 3), or psychophysical higher level factors (e.g., control subject in **Figure 10**). Furthermore, we should always remember that saccadic latency depends on many factors, including the lowest-level factors such as luminance and contrast of the stimulus and the higher-level factors such as urgency or prior probability. Hence, in our experiment the room illumination and stimulus properties was kept uniform in order to provide repeatable testing conditions.

The increase of saccadic decision-making time (latency) in amblyopia may result from a less efficient (slower) processing of visual information and/or from the poorer quality of information acquired by means of central retina. However, on the basis of our

**FIGURE 10 | Example of swiveling and parallel shifting of reciprobit plot for control subject.** In this case we reported significant differences in the latency distribution, manifested by the significant increase of mean latency for NDE.

findings we cannot conclusively state which of those two potential mechanisms have a dominant effect on the increase in saccadic latency in our amblyopes. It might be interesting for future studies to explore this topic.

The nature of spatial vision degradation in amblyopia is yet not fully understood. Spatial undersampling (i.e., reduced number of neurons and alternation in the spacing of retinal and cortical receptors) (Levi and Klein, 1996) and uncalibrated neural disarray (i.e., normal number of neurons but lack of calibration in the spatial array of cells covering the visual field) (Field and Hess, 1996) are the two major hypothesis for the losses of spatial vision in amblyopia (Field and Hess, 1996; Levi and Klein, 1996; Wang et al., 1998). However, we cannot decide which of those two proposed explanation have a dominant effect on our result. Both neural undersampling because of a decrease in neuronal spatial sampling density (Wang et al., 1998) and neural disarray (irregular sampling) because of disordered spatial arrangements of cells (Demanins et al., 1999) may reduce the speediness of saccadic decision making time.

Under physiological condition, the attempted steady fixation on a stationary visual target, does not cause the eye to remain motionless. High resolution oculomotor recordings with search coil, allowed to distinguish three types of fixational (involuntary) eye movements: tremor (very fast ∼90 Hz oscilations superimposed on drifts), drifts (slow meandering movement that occur between microsaccades) and microsaccades (small, typically less than 0.5◦ in amplitude involuntary saccades that occur during attempted fixation) (Martinez-Conde and Macknik, 2009; Otero-Millan et al., 2014). See Tables 1–3 from Martinez-Conde et al. (2004) for the detailed characteristics of fixational eye movements. Ciuffreda et al. (1979) studied fixational eye movements in amblyopic subjects. They reported increased drift during monocular viewing with the amblyopic eye. Several authors reported also fixation instability (measured as the dispersion of the eye position during attempted fixation) during amblyopic eye viewing (see Gonzalez et al., 2012; Subramanian et al., 2013). The increase of fixation instability during amblyopic eye viewing may result from increased tolerance of target eccentricity, allowing the observed target to depart further away from the fovea, before calling for the corrective, re-centering saccade. More recently, Shi et al. (2012) found increased amplitude and reduced frequency of microsaccades during monocular viewing with the amblyopic eye in subjects with anisometropic amblyopia. Microsaccade parameters for viewing with the fellow eye were comparable to those in subjects with normal vision. Rolfs et al. (2006) by the use of delayed saccade task have shown that saccadic latency is increased for saccades that occur shortly after microsaccades and is decreased when the microsaccades occurred up to 50ms before the GO signal (target disappearance) for saccades (Rolfs et al., 2006). For that reason the post saccadic refractive periods, which are generated intrinsically by the eye fixation function, can interfere with the externally applied saccadic stimuli being non-time coherent with the intrinsic refractive periods. Assuming that this kind of overlap may appear purely randomly, it should not contribute to the absolute values of the mean latency time but may increase its variability. However, the potential impact of abnormal fixational eye movements in amblyopia on saccadic latency was not confirmed by our results—we did not report any differences in saccadic latency variability between the amblyopic and control group. To the best of our knowledge there is a lack of studies exploring the influence of abnormal fixational eye movements (e.g., increased fixation instability) often observed in amblyopia on the latency of the subsequent saccade. It seems to be interesting for future studies to explore this topic.

Amblyopia is often also associated with temporal instability and spatial distortion (Sireteanu et al., 2008). Although Sireteanu et al. (2008) observed these deficits mainly in strabismic and deep amblyopia, we cannot exclude their effect on our results. This deficits may affect the perception of the disappearance of the central target what may in result delay the saccade initiation in our experiment.

## **CONCLUSIONS**

We found that there is no difference in saccade latency between the right and left eye in the control (without detectable visual deficits) group and significant difference in the anisometropic group. The comparison of saccadic latency between the right and left eye seems to be a useful extension of standard optometric/orthoptic or ophthalmologic examination, especially in subjects with amblyopia. Still, the potential use of distributional analysis of saccadic latencies as a diagnostic tool in amblyopia, should take into account large physiological variability of saccadic latency time and requires the acquisition of at least 100 saccadic responses for each eye. Additionally, it is important to remember that saccadic latency constitutes rather a capricious object for the investigator and depends upon many factors, including the nature of the stimulus, amount of available information, urgency, or prior probability. We hope, that this study contribute to a better understanding of the visual deficits and neural mechanisms underlying them in amblyopia.

#### **ACKNOWLEDGMENTS**

We would like to thanks all colleagues from our oculomotor laboratory in Poznan, especially P. Czarnecki, J. Lopatka, and J. Dylak for their technical and information technology support. We would like also to thanks Agata Gryc for correcting the revised version of the manuscript and Joanna Perdziak for graphical support.

#### **REFERENCES**


Carpenter, R. H. S. (1988). *Movements of the Eyes.* 2nd Edn. London: Pion.


**Conflict of Interest Statement:** Witkowska D., Gryncewicz W., Gryncewicz W., and Ober J. declare work for the Ober Consulting company. The coauthors have declared that no competing interests exist. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest

*Received: 30 April 2014; accepted: 16 September 2014; published online: 14 October 2014.*

*Citation: Perdziak M, Witkowska D, Gryncewicz W, Przekoracka-Krawczyk A and Ober J (2014) The amblyopic eye in subjects with anisometropia show increased saccadic latency in the delayed saccade task. Front. Integr. Neurosci. 8:77. doi: 10.3389/ fnint.2014.00077*

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

*Copyright © 2014 Perdziak, Witkowska, Gryncewicz, Przekoracka-Krawczyk and Ober. 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.*

## Binocular saccade coordination in reading and visual search: a developmental study in typical reader and dyslexic children

#### **Magali Seassau<sup>1</sup>\*, Christophe Loic Gérard<sup>2</sup> , Emmanuel Bui-Quoc <sup>3</sup> and Maria Pia Bucci <sup>4</sup>**

<sup>1</sup> e(ye)BRAIN, Ivry-sur-Seine, France

<sup>2</sup> Service de Psychopathologie de l'Enfant et de l'Adolescent, Hôpital Robert Debré, Paris, France

<sup>3</sup> Service d'Ophtalmologie, Hôpital Robert Debré, Paris, France

<sup>4</sup> UMR 1141 Inserm-Paris 7-Hôpital Robert Debré, Paris, France

#### **Edited by:**

Olivier A. Coubard, CNS-Fed, France

#### **Reviewed by:**

Marcelo Fernandes Costa, Universidade de São Paulo, Brazil Catalina Palomo-Alvarez, Complutense University, Spain

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

Magali Seassau, e(ye)BRAIN, 1 bis rue Jean Le Galleu, 94200 Ivry-sur-Seine, France e-mail: magali.seassau@ eye-brain.com

Studies dealing with developmental aspects of binocular eye movement behavior during reading are scarce. In this study we have explored binocular strategies during reading and visual search tasks in a large population of dyslexic and typical readers. Binocular eye movements were recorded using a video-oculography system in 43 dyslexic children (aged 8–13) and in a group of 42 age-matched typical readers. The main findings are: (i) ocular motor characteristics of dyslexic children are impaired in comparison to those reported in typical children in reading task; (ii) a developmental effect exists in reading in control children, in dyslexic children the effect of development was observed only on fixation durations; and (iii) ocular motor behavior in the visual search tasks is similar for dyslexic children and for typical readers, except for the disconjugacy during and after the saccade: dyslexic children are impaired in comparison to typical children. Data reported here confirms and expands previous studies on children's reading. Both reading skills and binocular saccades coordination improve with age in typical readers. The atypical eye movement's patterns observed in dyslexic children suggest a deficiency in the visual attentional processing as well as an impairment of the ocular motor saccade and vergence systems interaction.

**Keywords: binocular coordination, dyslexia, reading, visual search, development**

#### **INTRODUCTION**

Reading is a higher cognitive process depending on multiple processes: sensory perception, eye movements, linguistic and semantic capacities (Rayner et al., 2011). Furthermore, it is well known that a good control of the ocular motor system, in particular saccades, convergence and fixations, is essential for reading (Levy-Schoen and O'Regan, 1979; Seassau and Bucci, 2013). Indeed, deficits in one or more of these mechanisms could be at the origin of dyslexia. Despite intensive research on eye movements in dyslexic subjects, the origin of dyslexia is still debated, and other theories have been proposed which do not agree with an ocular motor impairment in dyslexic population (Lyon et al., 2003).

Abnormal eye movement performance observed in dyslexic children could be due to poor strategy of visual information processing, as it is not dependant on the language. A high number of regressive saccades and unstable fixation were observed by Pavlidis (1981) in Greek dyslexic children; in English dyslexic children, Rayner (1985) reported frequent saccades of smaller amplitude and longer duration fixations; in Italian dyslexic children, De Luca et al. (1999) observed frequent fixations with longer durations. More recently, slower reading speed and high number of saccades and regressions was reported in German dyslexic children (Trauzettel-Klosinski et al., 2010). Abnormal eye movements in picture searching was also reported in Chinese dyslexic children by Li et al. (2009), showing more fixations and frequent saccades of small amplitude.

Galaburda et al. (1985) was the first to show dysfunction at the level of the magnocellular system in dyslexics. Following this study, many researchers confirmed this hypothesis, showing in dyslexic population: poor binocular coordination during prolonged fixations (Stein and Fowler, 1993); visual confusion during reading (Stein and Walsh, 1997); and poor eye alignment during fixation after the saccade (Eden et al., 1994).

Confirming and extending the magnocellular hypothesis of dyslexia, impairment in visual search performance was also reported in dyslexic adults (Iles et al., 2000) with a motion coherence deficit. Even with these results, the existence of a deficiency in the magnocellular system in dyslexia is still under debate and some research does not share the hypothesis of poor visual system (Dhar et al., 2010; Skottun, 2010).

Jointly with such findings, a reduced visual attentional window size hypothesis was proposed. Bosse et al. (2007) reported that some dyslexic children have a limitation in the number of letters which can be processed in parallel. Consequently, dyslexics will make shorter saccades and frequent fixations in comparison to non dyslexic children. An fMRI study of this group (Peyrin et al., 2010) provided evidence on the role of parietal regions, particularly the left superior parietal area, in the visual attentional span and its deficiency in dyslexics. A recent study from Schneps et al. (2013) reported that young dyslexic students with reduced visual attentional span showed poor reading capabilities, suggesting a link between visual attentional spatial capabilities and dyslexia.

It is necessary to recall that the majority of research dealing with dyslexics with eye movements in reading was limited to measure movements from only one eye. Reading is, however, an activity requiring saccades and convergent eye movements. Horizontal saccades bring the eyes to successive words. For appropriate fusion of the two retinal images, the convergence angle between the two eyes needs to be well adjusted to the distance of the word. Only two studies explored binocular performance during reading in dyslexic children. Kirkby et al. (2011) reported poor binocular saccade coordination and poor fixation in dyslexic children and they suggested that reading itself could be responsible for such impairment. In contrast, a study from our group (Bucci et al., 2012) explored the quality of binocular coordination during reading and during visual search in groups of dyslexic and non dyslexic children. Disconjugacy measured during and after the saccade was significantly smaller in 10–12 year-olds than in 8–9 year-old non-dyslexic children. Furthermore, young children's saccades were smaller in amplitude; young children fixate more often and for longer than older children. Such ocular motor behavior has been observed both while reading and searching, suggesting an immaturity of the ocular motor saccade and interaction of vergence systems.

In the present study, we have attempted to assess ocular motor performance in reading a text and visual search task in a large population of dyslexic children and compared these findings to those from typical reader children. The novelty of the present study is that we analyzed the developmental aspects of binocular eye movement behavior during reading and visual search tasks.

## **MATERIALS AND METHODS**

#### **SUBJECTS**

Forty three dyslexic children participated in the study. Dyslexic children were recruited from the pediatric hospital where they were referred for a complete evaluation of their dyslexic state with an extensive examination including neurological/psychological and phonological capabilities. For each child the time of reading a text, its comprehension, and the capacity of reading word/pseudowords were evaluated by using the L2MA battery (Chevrie-Muller et al., 1997). This is the standard test developed by the Centre de Psychologie appliquée de Paris, often used in France and already employed in our previous studies for selecting dyslexic population (Bucci et al., 2008, 2009). Inclusion criteria were: scores of this test beyond 2 standard deviations; a normal mean intelligence quotient (IQ, evaluated with WISC-IV; between 80 and 115). Ages of dyslexic children



Mean values of: chronological and reading age, IQ, binocular vision (stereoacuity test, TNO measured in seconds of arc; near point of convergence, NPC measured in cm; heterophoria at near distance measured in prism diopters; vergence fusional amplitudes (divergence and convergence) at near distance measured in prism diopters. Asterisks indicate that value is significantly different between the two groups of children (p ≤ 0.002).

were comprised between 7 and 13 years (see clinical scores in **Table 1**). A carefully selected chronological age-matched control group (ages comprised between 7 and 13 years) of 42 typical reader children was selected. The control children had to satisfy the following criteria: no known neurological or psychiatric abnormalities, no history of reading difficulty, no visual impairment or difficulty with near vision. Also, reading capabilities within the normal range. Both the similitude test of the WISC IV assessing the verbal capability, and the matrix test of the WISC IV assessing the logic capability were performed. Normal range for both tests is 10 ± 3 (Wechsler intelligence scale for children—fourth edition, 2004). The control group was normal for verbal and for logic capabilities (see clinical scores in **Table 1**).

Both typical reader and dyslexic children underwent an ophthalmologic examination of their visual sensorial and motor function (mean values showed in **Table 1**). The stereoacuity threshold based on disparity detection was tested with the TNO random dot test for stereoscopic depth discrimination (Netherlands Organization, Richmond Products, Boca Raton, FL, USA). To avoid the lengthy time taken by that visual test, we limited on measure convergence and divergence fusional amplitude at near distance. All children had normal binocular vision (mean value of 55 s of arc or better). Visual acuity was normal (≥20/20) for all children, dyslexic as well as typical reader. The near point of convergence was normal for both groups of children tested (mean value of 2 cm). Heterophoria at near distance (i.e., latent deviation of one eye when the other eye is covered, using the cover-uncover test) was normal for both groups of children tested (≤ exophoria of 3.5 prism D). Moreover, an evaluation of vergence fusion capability using a prisms bar was done at near distance. The divergence and convergence amplitudes were significantly different in the dyslexic group in comparison to the non dyslexic children. ANOVA showed a significant group effect for the divergence and convergence amplitudes (respectively, *F*(1,83) = 13.56, *p* < 0.0005 and *F*(1,83) = 11.44, *p* < 0.002). The dyslexic group had significantly


**FIGURE 1 | Reading (A,B) and visual search (C,D) task respectively used for children with reading age of 8–9 and 10–13 years, respectively**.

smaller values of divergence and convergence compared to the typical group.

In summary, orthoptic evaluation showed a tendency of poor divergence and convergence amplitude in dyslexic children.

The investigation adhered to the principles of the Declaration of Helsinki and was approved by our Institutional Human Experimentation Committee (CPP Ile de France I, Hôpital Hotel-Dieu). Written consent was obtained from the children's parents after an explanation of the experimental procedure.

#### **OCULAR MOTOR PARADIGMS**

Stimuli were presented on a PC screen of 22", its resolution was 1920 × 1080 and the refresh rate was 60 Hz. Note that even if it is well known that intermittent illumination could affect saccade accuracy and visual assessment (Kennedy et al., 1998), such a refresh rate was sufficient to assure a normal saccade performance.

The reading and visual search tasks are similar to those used by Bucci et al. (2012) and are described below.

*Reading:* A text of four lines taken from a book for children. The paragraph contained 40 words and 174 characters. The text was 29◦ wide and 6.4◦ high; mean character width was 0.5◦ and the text was written in black "courier" font on a white background. Text was different for the two different reading age of children examined. **Figures 1A,B** shows the text presented to children with a reading age of 8–9 years (extract from "*Jojo Lapin fait des farces*", Gnid Bulton, Hachette) and that presented to children with a reading age of 10–13 years (extracted from "*Bagarres à l'école*", Marc Cantin et Eric Gasté, Castro Cadet). Children were asked to read the text silently.

*Visual search*: The same text presented in the reading task was used for this task but vowels were replaced by consonants (see **Figures 1C,D**). Children were asked to count the number of "r"s occurring in the text.

In both tasks stimuli were presented without time limitation. The recording of each task stopped when the child raised one finger.

#### **EYE MOVEMENT RECORDINGS**

Eye movements were recorded with the Mobile Eyebrain Tracker (Mobile EBT®, e(ye)BRAIN<sup>1</sup> ), an eye-tracking device CE marked for medical purposes. The Mobile EBT® uses cameras that capture the movements of each eye independently. Recording frequency was set up to 300 Hz. The precision of this system is typically 0.5◦ and in controlled conditions 0.25◦ (see www.eye-brain.com for more details). There is no obstruction of the visual field with the recording system.

#### **PROCEDURE**

Children were seated on a chair in a dark room, with the head stabilized by a forehead and chin support; viewing was binocular; the viewing distance was 60 cm. Calibration was done at the beginning of eye movement recordings. The best calibration could be an haploscopic arrangement. However, it should be noted that binocular vision was normal for all children tested (see stereoacuity scores in **Table 1**), suggesting that they were fixating targets with both eyes. A previous study from Bucci et al. (2002) comparing typical and strabismic children without amblyopia confirmed that in the presence of normal visual acuity in both eyes either type of calibration (under monocular or binocular viewing) was valid.

During the calibration procedure, children were asked to fixate a grid of 13 points (diameter 0.5◦ ) mapping the screen. Each calibration point required a fixation of 250 ms to be validated. A polynomial function with five parameters was used to fit the calibration data and to determine the visual angles. After the calibration procedure, the reading or visual search tasks were presented to the child. Duration of each task was kept short (lasting a couple of minutes) allowing an accurate evaluation of eye movement recordings.

#### **DATA ANALYSIS**

Calibration factors for each eye were determined from the eye positions during the calibration procedure. The software

<sup>1</sup>www.eye-brain.com

MeyeAnalysis (provided with the eye tracker, e(ye)BRAIN, France) was used to extract saccadic eye movements from the data. It automatically determines the onset and the end of each saccade by using a built-in saccade detection algorithm. The algorithm used to detect saccades is adapted from Nyström and Holmqvist (2010). The algorithm searches for velocity peaks by identifying samples where the velocity is larger than a velocity threshold (θ > θPT). An iterative data-driven approach is proposed to finding a suitable threshold. The iterative algorithm is given an initial peak velocity detection threshold PT1, which could be in the range 100◦–300◦ /s, but the choice is not critical as long as there are saccades, with peak velocities reaching this threshold. For all samples with velocities lower than PT1, the average (µ) and standard deviation (σ) are calculated. The threshold is updated as PT*<sup>n</sup>* = µ*n*−<sup>1</sup> + 6σ*n*−<sup>1</sup> for each iteration. For each detected saccade peak (hose detected after the last iteration), the algorithm searches backward (from the leftmost peak saccade sample) and forward (from the rightmost peak saccade sample) in time for the saccade onset and offset. Saccade onset is defined as the first sample that goes below the saccade onset threshold and where θ*<sup>i</sup>* − θ*i*+<sup>1</sup> ≥ 0. Saccadic offset is defined as the first sample that goes below the saccade offset threshold and where θ*<sup>i</sup>* − θ*i*+<sup>1</sup> ≤ 0. All saccades with an amplitude superior to 1◦ were detected. All detected saccades were checked by the researcher and corrected/discarded if necessary.

The number and the amplitude of progressive saccades (prosaccades, from left to right) and regressive saccades (backward saccades, from right to left) and the duration of fixations between each saccade were analyzed. In both tasks (reading and visual search), binocular coordination was defined for each saccade and each fixation was recorded. For each saccade recorded in the two tasks (reading and visual search) we examined the amplitude of the conjugate [(left eye + right eye)/2], and the disconjugate components (left eye − right eye) during the saccade (see Bucci et al., 2012). The disconjugacy was measured as the change in vergence between the beginning and the end of each saccade. We also examined the disconjugate component of each post-saccadic fixation period over the period between two saccades ([(*x*<sup>2</sup> − *x*1)left − (*x*<sup>2</sup> − *x*1)right]; where *x*<sup>2</sup> = amplitude of the end of fixation and *x*<sup>1</sup> = amplitude of the beginning of fixation). Given that saccade disconjugacy depends on the saccade amplitude, the values of disconjugacy during and after the saccades were presented as the ratio of the disconjugacy on the saccade amplitude (in percentage).

Statistical analysis was performed by the two-way ANOVAs using the two groups of children (dyslexics and control) as intersubject factor and the two conditions (reading text and visual search) as within subject factor. The effect of a factor is significant when the *p*-value is below 0.05.

Then data were analyzed using different multiple linear regression models—the number of saccades, the amplitude of saccades (in degrees), the duration of fixations (in ms) and the duration of task (in seconds) for both groups. The predictor variable for each test was the participant's age (in year and months).

## **RESULTS**

## **EYE MOVEMENT PATTERN DURING READING AND VISUAL SEARCH (SEE** TABLE 2**)**

#### **Number of fixations**

The ANOVA showed a significant group effect (*F*(1,83) = 25.26, *p* < 0.0001) with a number of fixations in the control group significantly smaller than in the dyslexic group. We found also a significant effect of the task (*F*(1,83) = 29.23, *p* < 0.0001), meaning that the number of fixations was larger in the visual search task with respect to the reading task. Finally, a significant interaction between group and task was also reported (*F*(1,83) = 26.87, *p* < 0.0001): the control group made fewer fixations during reading than during visual search (*p* < 0.001), no difference was found between reading and visual search in dyslexic children (*p* = 0.87).

**Figure 2** shows the number of fixations assessed during reading (A) and visual search (B) as a function of age and group for each participant examined, and the regression line observed in each case. There was a significant effect of age in the reading task: the number of fixations decreased as age increased (*R* <sup>2</sup> = 0.61, *p* < 0.001) only for the control children. Effect of age in reading task was not significant for dyslexic children (*R* <sup>2</sup> = 0.0003). On the visual search task, age, there was no significant effect of age neither for the control children (*R* <sup>2</sup> = 0.07) nor for the dyslexic children (*R* <sup>2</sup> = 0.001).

#### **Duration of fixations**

In order to assess more information about fixations, we also measured the average duration of fixations, which is the time period between two saccades. The ANOVA showed a significant group effect (*F*(1,83) = 24.14, *p* < 0.0001): the duration of fixation of the control group was significantly shorter in comparison to the dyslexic group.

We found a significant effect of the task (*F*(1,83) = 18.67, *p* < 0.0001)—duration of fixations was longer in the visual search task in comparison to the reading task.

We found a significant interaction between group and task (*F*(1,83) = 38.98, *p* < 0.0001); more precisely, the control group showed shorter duration of fixations in reading task in comparison to the visual search task (*p* < 0.001). No difference was found between reading and visual search in dyslexic children (*p* = 0.18).

We found a significant effect of age on the duration of fixations (see **Figure 3A** for reading and **Figure 3B** for visual search task), which decreased with age in both tasks and both groups (reading: *R* <sup>2</sup> = 0.25, *p* < 0.0006 and *R* <sup>2</sup> = 0.32, *p* < 0.0001 respectively for dyslexic and control children; visual search: *R* <sup>2</sup> = 0.13, *p* < 0.02 and *R* <sup>2</sup> = 0.23, *p* < 0.001 respectively for dyslexic and control children).

#### **Progressive saccade**

The number of progressive saccades was significantly different between dyslexic and control groups (see **Table 2**). The ANOVA showed a significant group effect (*F*(1,83) = 25.63, *p* < 0.0001) dyslexic children made more progressive saccades than control children. We also found a significant task effect (*F*(1,83) = 15.52, *p* < 0.001), with more progressive saccades in the visual

**FIGURE 2 | Number of fixations during reading (A) and visual search (B) for both groups of subjects**. Lines represent the corresponding regressions.

search task than in the reading task, and a significant interaction (*F*(1,83) = 4.54, *p* < 0.04) with no difference between the number of progressives saccades in reading and visual search in dyslexic children (*p* = 0.18), whereas control children made less progressive saccades in reading task compared to visual search task (*p* < 0.0001).

**Figure 4** shows the number of progressive saccades assessed during reading (A) and visual search (B) tasks for each participant by age and group. There was a significant effect of age on control children: the number of progressive saccades decreased with age in the reading task (*R* <sup>2</sup> = 0.62, *p* < 0.0001) but not in the visual search task (*R* <sup>2</sup> = 0.06, *p* = 0.13). There is no effect of age on dyslexic group neither on reading (*R* <sup>2</sup> = 0.003, *p* = 0.74) nor visual search (*R* <sup>2</sup> = 0.01, *p* = 0.44) tasks.

As there is no difference between amplitudes of left eye and right eye (*F*(1,83) < 1), the mean amplitude of progressive saccades during reading and visual search task for each group of children is shown in **Table 2**. The ANOVA showed a significant group effect (*F*(1,83) = 13.46, *p* < 0.0005) with smaller saccade amplitude for the dyslexic group compared to the control group. We also found a significant task effect (*F*(1,83) = 6.88, *p* < 0.01), with smaller amplitude of progressive saccades in the visual search task than in the reading task.

The interaction between group and task was also significant (*F*(1,83) = 34.68, *p* < 0.0001). *Post hoc* comparison showed that the amplitude of progressive saccades during reading task for the control group was significantly larger than saccades in visual


**Table 2 | Ocular motor characteristic of the two groups of children examined (dyslexic and control children) during reading and visual search task**.

Mean values (standard error) of: task duration, number of fixations, duration of fixations, number of progressive saccades, amplitude (in ◦ ) of progressive saccades (mean amplitude and amplitude for each eye, LE = left, RE = right eye), number of regressive saccades, amplitude (in ◦ ) of regressive saccades (mean amplitude and for each eye), disconjugacy during the saccades (in %), disconjugacy after the saccades (in %), and percentage of "r" counting in the visual search. Asterisks indicate that value is significantly different between the two groups of children (p ≤ 0.01).

search task (*p* < 0.001) and larger than in the dyslexic group (*p* < 0.0001). The amplitude of progressive saccades during reading task for the dyslexic group was significantly smaller than saccades in visual search task (*p* < 0.03). No difference was found between dyslexic and control groups in the visual search task (*p* = 0.38).

**Figure 5** shows the mean amplitude of progressive saccades assessed during reading (A) and visual search (B) tasks for each participant by age and group. There was a significant effect of age on control children: the amplitude of progressive saccades increased with age in the reading task (*R* <sup>2</sup> = 0.62, *p* < 0.0001) but not in the visual search task (*R* <sup>2</sup> = 0.01, *p* = 0.51). There is no effect of age on dyslexic group<sup>∗</sup> on reading (*R* <sup>2</sup> = 0.05, *p* = 0.16) nor on visual search (*R* <sup>2</sup> = 0.002, *p* = 0.79) tasks.

#### **Regressive saccade**

The number of regressive saccades was significantly different between the dyslexic and the control group (see **Table 2**). The ANOVA showed a significant group effect (*F*(1,83) = 14.30, *p* < 0.0003), meaning that dyslexic children made more regressive saccades than control children. We also found a significant task effect (*F*(1,83) = 33.6, *p* < 0.001), with more regressive saccades in the visual search task than in the reading task, and a significant interaction (*F*(1,83) = 4.54, *p* < 0.04).

**Figure 6** shows the number of regressive saccades assessed during reading (A) and visual search (B) tasks for each participant by age and group. There was a significant effect of age on control children: the number of regressive saccades decreased with age in the reading task (*R* <sup>2</sup> = 0.24, *p* < 0.002) but not in the visual search

**FIGURE 5 | Amplitude of progressive saccades during reading (A) and visual search (B) for the both groups of subjects**. Lines represent the corresponding regressions.

**FIGURE 6 | Number of regressive saccades during reading (A) and visual search (B) for the both groups of subjects**. Lines represent the corresponding regressions.

task (*R* <sup>2</sup> = 0.05, *p* = 0.17). There is no effect of age on dyslexic group on reading (*R* <sup>2</sup> = 0.04, *p* = 0.19) nor on visual search (*R* <sup>2</sup> = 0.01, *p* = 0.49) tasks.

As there is no difference between amplitudes of left eye and right eye (*F*(1,83) < 1), the mean amplitude of regressive saccades during reading and visual search task for each group of children is shown in **Table 2**, ANOVA showed neither a group effect (*F*(1,78) < 1), nor a task effect (*F*(1,78) < 1), nor a significant interaction between task and group (*F*(1,78) = 2.09, *p* = 0.15). No difference was found between amplitudes of regressive saccade of dyslexic and control children in both tasks.

We did not find an effect of age on the amplitude of regressive saccades, neither in the reading task (*R* <sup>2</sup> = 0.001, *p* = 0.83 and *R* <sup>2</sup> = 0.04, *p* = 0.21 respectively for dyslexic and control children) nor in the visual search task (*R* <sup>2</sup> = 0.02, *p* = 0.42 and *R* <sup>2</sup> = 0.03, *p* = 0.31 respectively for dyslexic and control children).

#### **BINOCULAR COORDINATION DURING READING AND VISUAL SEARCH Disconjugacy during the saccades**

For the disconjugacy values reported during the saccade, the ANOVA showed a significant group effect (*F*(1,83) = 34.42, *p* < 0.0001), showing that the saccades disconjugacy of the control group was significantly smaller with respect to the dyslexic group. The ANOVA did neither show a significant task effect (*F*(1,83) = 1.96, *p* = 0.17) nor a significant interaction between group and task (*F*(1,83) < 1).

**FIGURE 7 | Disconjugacy during the saccades in the reading (A) and visual search tasks (B) for both group of subjects**. Lines represent the corresponding regressions.

We found no effect of age on the disconjugacy during the saccades (see **Figure 7**), neither in the reading task (*R* <sup>2</sup> = 0.02, *p* = 0.29 and *R* <sup>2</sup> = 0.02, *p* = 0.39 respectively for dyslexic and control children) nor in the visual search task (*R* <sup>2</sup> = 0.001, *p* = 0.81 and *R* <sup>2</sup> = 0.009, *p* = 0.99 respectively for dyslexic and control children).

#### **Disconjugacy after the saccades**

Similar statistical results were reported for the values of the disconjugacy measured after the saccade. The ANOVA showed a significant group effect (*F*(1,83) = 44.17, *p* < 0.0001) showing that the disconjugacy measured after the saccade of the control group was significantly smaller in comparison to the dyslexic group. The ANOVA failed to show either a significant task effect (*F*(1,83) = 3.23, *p* = 0.07) or a significant interaction between group and task (*F*(1,83) < 1).

We found no effect of age on the disconjugacy after the saccades (see **Figure 8**) in the reading task (*R* <sup>2</sup> = 0.03, *p* = 0.28 and *R* <sup>2</sup>=0.04, *p* = 0.19 respectively for dyslexic and control children). On visual search task there was no effect of age on dyslexic children (*R* <sup>2</sup> = 0.0007, *p* = 0.86) and only a tendency was found in control children (*R* <sup>2</sup> = 0.07, *p* = 0.08).

The performance in the visual search task was also measured (see Section Materials and Methods) by asking the child the number of "r"s read in the text. Such performance was similar in dyslexic children and control children (*F*(1,83) = 2.006, *p* = 0.16). Moreover, the total task duration was no different between dyslexic and control children (*p* = 0.26) in the visual search task, whereas the duration of the reading task was significantly higher in dyslexic children in comparison to control subject (*p* < 0.0001; interaction group<sup>∗</sup> task: (*F*(1,83) = 30.68, *p* = 0.0001)). This data suggests that all children accomplished the visual search task in a similar way but not the reading task (See mean value in **Table 1**).

Finally, we explored the presence of a correlation between subjective measures of vergence clinically assessed and ocular motor measures.

In control children, disconjugacy during the saccade was correlated to convergence values measured clinically at near distance (*r* = 0.41, *p* < 0.03); in contrast, this was not the case for dyslexic children (*r* = 0.13, *p* = 0.51). The absence of such correlation in dyslexic children could be related to their large disconjugacy values reported during reading. Such finding suggests a link between saccade performance and subjective vergence capabilities (see also Bucci et al., 2011 [26]).

### **DISCUSSION**

The main findings from this study are as follows: (i) during reading, ocular motor characteristics of dyslexic children are impaired in comparison to those reported in typical children; (ii) developmental effect during reading only for typical reader children. Effect of age in dyslexic children was observed only on fixation durations; (iii) different ocular motor behavior in the two tasks in dyslexic and in typical reader children; and (iv) disconjugacy during and after the saccade is larger in dyslexic children with respect to typical children.

Each of these findings is discussed.

#### **OCULAR MOTOR IMPAIRMENT OF DYSLEXIC CHILDREN DURING READING**

Many fixations, longer duration, and shorter amplitude saccades were found in dyslexic children in comparison to typical readers. These findings are in line with previous research done by recording one eye only in dyslexic children in several countries (see Section Introduction). The pattern of ocular motor impairment we reported also in this study by recording movements from both eyes could be due to an immaturity of visual attentional strategies, leading to reduced visual attentional span (which corresponds to the number of elements that can be processed in parallel) according to the study of Bosse et al. (2007). Such a limitation, leading to the higher number of fixations and longer fixation duration we reported, suggest that dyslexic children read the text analytically.

#### **DEVELOPMENTAL EFFECT ON OCULAR MOTOR CHARACTERISTICS DURING READING**

Our findings on the number of fixations, fixation durations and amplitude saccades, during reading are in line with findings previously reported on ocular motor behavior from McConkie et al. (1991) and Blythe et al. (2006) and more recently from Seassau and Bucci (2013) showing that typical children's reading skills develop with age. The present study showed that, with age, typical children's reading capabilities improve and they learn to read by making larger progressive saccades, fewer regressive saccades and shorter fixations. The improvement of reading skills could be due to cortical development. Luna et al. (2008) reported that the activity of some cortical areas involved in saccadic eye movements to visual stimulus by stimulating visually-guided saccades, anti-saccades and memory-guided saccades (e.g., frontal and parietal cortex) is lower in young children than in adults and increases until adolescence. Differences in the anterior left occipito-temporal cortex was recently observed between children and adults during word processing, providing evidence of developmental course of those regions (Olulade et al., 2013). Our findings in reading deal with the developmental hypothesis. However, brain imaging studies in a large population of children during reading will be needed to further explore such an issue.

In contrast, dyslexic children's reading skills are not influenced by age. Neither the number of fixations, nor the amplitudes of saccades nor disconjugacy are improved by age.

Further studies comparing poor readers and readers with dyslexia could be useful to better understand how eye movements reflect the difficulties that disabled readers are having understanding the text they're reading.

#### **DIFFERENT OCULAR MOTOR BEHAVIOR IN THE TWO TASKS IN DYSLEXIC AND IN TYPICAL READER CHILDREN**

The reading and the visual search tasks made different demands on visuo-perceptual, attentional and spatial processing. Consequently one could expect to observe different ocular motor behavior in these two tasks. Typical reader children showed significant differences between the two tasks only on the number of fixations, fixation durations and amplitude of saccades. Note that the pattern of fixation is different in the two tasks because they correspond to different cognitive demands in the case of well-reading children. In the visual search task, the child is required to identify and count a single target; he has to see all the letters in order to adequately perform the task. In contrast, in the reading task the child can skip letters because the linguistic processing is well developed. Consequently, reading is easier than visual searching for typical reader children.

In contrast, dyslexic children display similar ocular motor behaviors in both tasks. According to Prado et al. (2007) a reduced visual attentional span could have a similar impact on reading and on visual search, because visual attentional demand is similar in the two tasks. In dyslexic children, as reading capabilities are not well structured, reading and visual search tasks had similar demands in visuo-perceptual, attention and spatial processing. We suggest that dyslexic children perform both tasks in a similar way. In contrast, typical readers, who have better developed reading skills, accomplish both tasks differently, which is reflected in their ocular motor behavior.

Finally, it should be pointed out that the two tasks did not show any difference with regards to the binocular coordination of saccades, neither in typical nor dyslexic readers. Previously, we have already showed that the quality of binocular coordination during and after the saccades does not depend on the stimulus used (single word reading; fixation of LEDs or text reading; Bucci and Kapoula, 2006; Bucci et al., 2012). These results are also in line with the study of Jainta and Kapoula (2011) comparing binocular saccade coordination during reading and free exploration of painting. The present data brings new evidence on the quality of binocular coordination by showing that reading texts do not interfere with, and contrasts Heller and Radach (1999) and Kirkby et al. (2011) reports, suggesting that reading itself induces impairment in the binocular saccade control and fixation instability.

Note, however, that further studies exploring binocular coordination on linguistic and non linguistic stimuli could be useful to better understand how binocular coordination could be influenced or not by the type of visual stimuli.

#### **LARGE DISCONJUGACY DURING AND AFTER THE SACCADES IN DYSLEXIC CHILDREN**

The poor quality of binocular coordination in dyslexic children, during and after the saccades, suggests an impairment of ocular motor learning mechanisms, at central/cortical level responsible for saccade yoking. In dyslexic children the clinically assessed limited vergence capabilities (see **Table 1**) could be responsible for such a deficient interaction between saccadic and vergence movements and thus lead to disconjugate saccades.

According to Lewis et al. (1995), we can hypothesize that the fine control of binocular saccade coordination is based on an efficient relationship between the motor command of the saccades and the vergence subsystems at the premotor level. This hypothesis has been tested in different types of child populations showing poor vergence fusional capabilities (i.e., dyslexic children, children with strabismus and children with vergence insufficiency) (Bucci et al., 2012; Gaertner et al., 2013; Lions et al., 2013). Note that reading is an activity done at near distance. In order to adjust the visual axes of both eyes at the distance of the word, a correct convergence command strictly linked with the saccade command is needed for appropriate fusion of the two retinal images. All child populations with poor vergence capabilities (as those previously cited) showed poor binocular saccade control. This hypothesis, however, needs further exploration.

We could make the hypothesis that vergence training could help dyslexic children to improve the quality of their saccade coordination. This hypothesis, however, needs further exploration.

Finally, a fine binocular coordination of saccades could involve the magnocellular network and also the cerebellum according to the study of Nicolson et al. (1999). However, according to Iles et al. (2000), deficits in the magnocellular network involving the parietal cortex could be related to poor visuo-attentional capabilities already reported in dyslexic children. Further studies by combining neuroimaging techniques and visuo-attentional tasks will be necessary to test the different hypothesis on the origin of dyslexia.

#### **CONCLUSION AND FUTURE DIRECTIONS**

Deficits in ocular motor behavior reported in dyslexic children seem to be associated to the precise controlled interaction between the saccade and the vergence systems. This ocular motor deficit could be in relationship with the clinical assessment, showing poor fusional vergence capabilities in dyslexic children.

We believe that orthoptic vergence training, together with specific visual attentional training and reading tasks, could be useful tools for dyslexic children to improve visual attentional span and vergence capabilities as well as saccade yoking.

#### **FUNDING**

The authors have no support or funding to report.

#### **ACKNOWLEDGMENTS**

Authors thank the medical doctors and nurses of "Service de Psychopathologie de l'enfant et de l'adolescent", Robert Debré Hospital (Paris, France), particularly Ms. Sandrine Larger for screening dyslexic children; the directors and the teachers of the Collège Saint André (Saint Maur des Fossés) and of the elementary school Saint François (Paris) for allowing oculomotor tests in non dyslexic children; parents and children for their kind participation; and Maxine Ramsey for revising the English version of the manuscript.

#### **REFERENCES**


**Conflict of Interest Statement**: Magali Seassau declares work for the e(ye)BRAIN company and author as co-inventor of a patent (B110332FRA, 2011) for the detection of an oculomotor abnormality (parameter for binocular coordination) in a dyslexic patient by means of a visual search test, which is used in this study. Other authors have declared that no competing interests exist.

*Received: 27 February 2014; accepted: 10 October 2014; published online: 30 October 2014*.

*Citation: Seassau M, Gérard CL, Bui-Quoc E and Bucci MP (2014) Binocular saccade coordination in reading and visual search: a developmental study in typical reader and dyslexic children. Front. Integr. Neurosci. 8:85. doi: 10.3389/fnint.2014.00085 This article was submitted to the journal Frontiers in Integrative Neuroscience*.

*Copyright © 2014 Seassau, Gérard, Bui-Quoc and Bucci. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and 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*.

## LATER models of neural decision behavior in choice tasks

## *Imran Noorani\**

*Department of Neurosurgery, Wessex Neurological Centre, University Hospital Southampton, Southampton, UK*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

#### *Reviewed by:*

*Braden Alexander Purcell, New York University, USA Jing Tian, Johns Hopkins School of Medicine, USA*

*\*Correspondence: Imran Noorani, Department of Neurosurgery, Wessex Neurological Centre, University Hospital Southampton, Southampton SO16 6YD, UK e-mail: imran.noorani@cantab.net*

Reaction time has been increasingly used over the last few decades to provide information on neural decision processes: it is a direct reflection of decision time. Saccades provide an excellent paradigm for this because many of them can be made in a very short time and the underlying neural pathways are relatively well-known. LATER (linear approach to threshold with ergodic rate) is a model originally devised to explain reaction time distributions in simple decision tasks. Recently, however it is being extended to increasingly more advanced tasks, including those with decision errors and those requiring voluntary control such as the antisaccade task and those where sequential decisions are required. The strength of this modeling approach lies in its detailed, quantitative predictions of behavior, yet LATER models still retain their conceptual simplicity that made LATER initially successful in explaining reaction times in simple decision tasks.

**Keywords: saccades, decision, LATER, eye, latency, reaction time, neuron**

## **INTRODUCTION**

The question of how we choose one option over another has intrigued neuroscientists in the field of neural decision-making for many decades. Performing one action rather than another, say running toward a bus instead of walking, requires more than just sensory input and a motor output. The brain has to integrate sensory information and somehow make a decision on what is the "best" course of action based on this information, and subsequently to implement this decision as a motor response. It is now widely believed that the brain accumulates sensory information toward a threshold level, at which point the evidence has become convincing enough for a certain decision to be selected (Carpenter and Williams, 1995; Schall, 1995). Such an accumulator model approach to neural decision-making has been applied to many important features of decision behavior, yet there is still much to be explained.

Reaction time is regarded as an experimental "window" into decision processes. Reaction time, or latency, is composed of more than the simple sum of times of for sensory input and motor output, and it is this extra time that reflects the time taken for the brain to choose a response. This additional processing time can also be called neural procrastination: the brain is taking longer than it needs to if sensory input and motor output were all that is needed for a behavioral response. As sensory and motor times are relatively fixed, reaction time is therefore a useful indicator of decision time. In experimental paradigms, reaction time varies between one trial and the next, even if exactly the same experimental conditions are maintained.

Saccades are the rapid eye gaze shift movements which we make a great number of times every day. A saccade represents the output of a decision: a choice of where to look. They are especially useful for studying decision in the laboratory because they are quick so many of them can be produced in a short space of time, they have a clear sensory input (e.g., a visual stimulus), and their neuronal pathways are relatively well known. It is unsurprising, therefore, that saccades have been widely used to produce reaction time data to give us insight into decision mechanisms.

The neural control of saccades is mediated primarily by frontal and parietal cortical regions, basal ganglia and the superior colliculus. Although decision-making for generating saccades was thought to be cortical (Schall, 2001) more recent work suggests the superior colliculus may be involved in decision-making (Phongphanphanee et al., 2014) and a model of superior collicular activity can potentially explain express saccades (Trappenberg et al., 2001). Races between potential decisions have also been identified in the basal ganglia (Schmidt et al., 2013). The frontal eye fields (FEF) in the cortex have been particularly studied as regions for decision-making in eye movements (Pierrot-Deseilligny et al., 2003). Neurons in these regions project directly to brain stem structures containing ocular motor neurons so that they can directly influence the production of saccades (Segraves, 1992; Shinoda et al., 2011). Following target appearance, there is accumulation of movement-related activity in these regions toward a threshold, and once the threshold is reached a saccade is triggered (Hanes and Schall, 1996). The neuronal mechanisms for saccadic control have been reviewed in detail elsewhere (Pierrot-Deseilligny et al., 2004), and lesion studies in humans have been helpful in delineating such mechanisms (Pierrot-Deseilligny et al., 2002; Machado and Rafal, 2004; Ramat et al., 2007).

#### **THE LATER MODEL**

The variability of reaction times is an interesting phenomenon. When plotted on a histogram, the distribution of reaction times appears skewed. However, if we take the reciprocal of the latencies and plot these in a similar fashion, the resulting distribution appears Gaussian. This tells us that perhaps it is the reaction rate rather than reaction time *per se* which is a more important reflection of the underlying decision process. A Gaussian or normal distribution of reciprocal latencies implies that these reciprocals have equal variability around a mean value. If reciprocal latencies are plotted cumulatively (a reciprobit plot) then a straight lineis obtained (**Figure 1**). Such a distribution can then be explained by a very simple and elegant model—the LATER model. It is important to note that although saccadic reaction times have been studied in great detail with this type of decision modeling, manual reaction times to auditory or visual stimuli have also been found to have straight line distributions much like saccades and can therefore be modeled similarly (Carpenter, 1981; Pearson and Carpenter, 2010).

The LATER (linear approach to threshold with ergodic rate) model is an established model of neural decision that has been highly successful in explaining reaction time distributions over the last few decades. In the model, a decision signal starts from a starting point S0 and rises toward a threshold ST—once the signal reaches the threshold, the decision is made for a particular action. The rate at which the decision signal rises varies randomly from trial to trial, but the mean rate of rise is constant and denoted by the parameter μ. The standard deviation of this variation in rate of rise is given by the parameter σ. One of the main reasons LATER is so conceptually attractive for explaining decisions is that the parameters of reciprocal reaction time distributions are the parameters of the model itself—μ being the mean rate of rise of the decision signal as well as the mean reciprocal latency, and σ reflecting the variability in the latency distributions (Carpenter and Williams, 1995). There are, of course, many other different kinds of accumulator race models which have been applied to various tasks; however, these have been discussed elsewhere including some attempts at comparisons between different types of model (Usher and McClelland, 2001; Ratcliff and Smith, 2004; Bogacz et al., 2006; Ratcliff and McKoon, 2008; Heathcote and Hayes, 2012; Heathcote and Love, 2012; Bitzer et al., 2014), and here we will focus on LATER.

What does this rise to threshold of the decision signal imply about the decision process? It is thought to be an accumulation of sensory "evidence" for a hypothesis, and this accumulation occurs in a linear fashion. Once enough evidence for a certain hypothesis is accumulated, signified by the decision signal reaching its threshold, then this hypothesis is accepted and a decision to respond is made (**Figure 2**). In this way, LATER is a quasi-Bayesian model of decision-making and the decision signal itself mathematically corresponds to the log likelihood ratio of a certain choice being the correct one (Carpenter and Williams, 1995). One can easily therefore conceive many possible choices, and the final choice out of a number of options occurs when the decision signal for this choice reaches threshold before the decision signals representing the other options. Indeed, a simple task with a visual target and a visual distractor was modeled by two LATER units, one representing the target and the other the distractor, both racing against each other to threshold to determine the response (Leach and Carpenter, 2001).

Any robust neural model must be subject to empirically testable predictions, and the LATER model is no exception. When reaction time distributions are plotted on a reciprobit plot, a straight line is obtained if the reciprocal latencies follow a Gaussian distribution. If the mean rate of rise of the decision signal is increased, then one would predict that the latency distribution should shift toward shorter latencies. If we alter S0 by varying prior probability of a stimulus, then one would predict a swivel in the latency distribution. Indeed, these predictions were borne out in an elegant set of experiments. The mean rate of rise was altered by changing the rate of information provision and prior probability was altered by changing subject's expectations of stimulus locations—both manipulations produced the changes in latency distributions predicted by the LATER model (Carpenter and Williams, 1995; Reddi and Carpenter, 2000; Reddi et al., 2003). The importance of predicting complete distributions of reaction times as opposed to mean/median latencies or latency quantiles lies in the fact that there can be subtleties to a full distribution that often point to multiple decision mechanisms and these could be missed if the distribution is not plotted in full (Noorani and Carpenter, 2011).

LATER is a model of decision processes. Reaction time however is composed of sensory detection and motor implementation, as well as decision. When the visual targets are high-contrast and thus easily detectable, a LATERian approach predicts reaction

**FIGURE 2 | The LATER model of decision.** A decision signal accumulates information about a decision based on information supply and rises toward threshold in a linear fashion, triggering a response when it reaches threshold.

times very well; but when the targets are more difficult to detect, a random-walk model that integrates noisy afferent signals models behavior well. Naturally, the models can be reconciled by viewing stimulus detection and decision as two separate processes occurring sequentially, with stimulus detection occurring in a random-walk fashion followed by the decision process occurring in a LATERian manner. A model of this sort predicts reaction time distributions over a wide range of stimulus detectability (Carpenter et al., 2009).

#### **EARLY SACCADES**

On a small number of occasions, experimental subjects produce saccades whose latencies are very short and do not follow the main latency distribution. Such "early" responses form a separate component of the latency distribution on a reciprobit plot, as a straight line with a shallower gradient than the main component of the distribution (**Figure 3**). These early responses therefore cannot be explained by the same single LATER decision unit that gives rise to saccades that follow the main distribution (otherwise they would be part the same straight line on the reciprobit). Instead, it has been proposed that these early responses are the result of an "early" LATER decision unit, whose parameters differ from those of the main LATER unit in having a mean rate of rise of zero but a very large σ, such that occasionally this eccentric unit wins the race against the main unit to produce a fast response. Simulations with such a race between these two decision units do indeed produce latency distributions with an early and main component mirroring real life behavior (Noorani and Carpenter, 2011). One circumstance in which the frequency of early responses is increased is under conditions of cognitive distraction. This has been demonstrated experimentally when a subject is using a mobile phone whilst performing a simple saccadic task, in which there is a larger early component to the latency distribution. Perhaps when one is distracted there is less cortical inhibition from "higher" regions like supplementary and frontal eye fields to more primitive neural regions controlling saccades, such as the superior colliculus, allowing these maverick early responses to win the decision race more often (Halliday and Carpenter, 2010).

## **COUNTERMANDING AND GAP PARADIGMS**

The situation frequently arises when one needs to cancel an impending an action, for example stopping yourself from crossing the street as a car suddenly drives past. In scenarios like this, there must be a way of suppressing a decision signal that is itself accumulating and about to generate a response. An established paradigm for studying this behavior in the laboratory is called the countermanding task. Here, a visual stimulus appears and the subject knows they must make a saccade toward it, but sometimes after the target appears a "stop" signal also appears indicating to the subject that they must not make a saccade to the target. On these "stop" trials, the developing decision to generate a saccade must be canceled. This is called the "countermanding" task. What is the underlying neural mechanism enabling this to be achieved? It requires the involution of a new concept—the stop unit, a decision signal which can accumulate evidence needed for canceling a response when a stop signal is presented. The stop unit must race against the "go" unit, which is responsible for triggering a saccade (**Figure 4**). In this way, on some trials no saccade occurs because the stop unit wins the race, whereas on others a saccade occurs to the stimulus when the go unit reaches threshold first (Logan et al., 1984). Hanes and Carpenter then demonstrated that a linear rise to threshold for stop and go processes can successfully account for the detailed reaction time distributions in this task, with the

**FIGURE 4 | The countermanding race model, in which a stop unit competes with a go unit and the winner of the race determines the outcome.** Trials in which responses failed to be canceled despite presence of a stop signal are accounted for by the go unit reaching threshold before stop. Modified with permission from Noorani et al. (2011).

mean and variability of the rate of rise of the two types of unit will determine the precise timing and frequency of these two types of response. Such a simple model can robustly predict not just the mean latencies but also (and much more importantly) the latency distributions and incidence of stop and go responses (Hanes and Schall, 1995; Hanes and Carpenter, 1999; Boucher et al., 2007). It must be noted, however, that recent work suggests that stimulus detection or perception may play a larger role in countermanding than previous considered (Salinas and Stanford, 2013). Moreover, countermanding is distinct from task-switching in which a new instruction puts the old one out of date: in this case, this is likely to occur through a functional unit to detect the new instruction which then activates a separate LATER unit to accumulate activity for the appropriate decision (Sinha et al., 2006).

Another interesting paradigm is the gap task, in which a central fixation unit disappears to leave a short period in which there is no visual stimulus before a peripheral stimulus appears. This "gap" in between stimuli has been found to speed up reaction times, perhaps because it provides a warning effect signaling to the subject the impending appearance of a stimulus. This has also been modeled by two racing LATER units, except unlike in countermanding there is no stop unit but instead there is a "fixation" unit in addition to the main saccadic unit. The fixation unit is activated when the fixation stimulus disappears, and instead of stopping the main saccadic unit it *enhances* its decision signal allowing it to reach threshold more quickly. In this way, this model quantitatively predicts the reaction time distributions in the gap paradigm (Story and Carpenter, 2009).

#### **GO/NO-GO PARADIGM**

From this early work, LATER gained substantial popularity as a model of decision-making because of its success in explaining full reaction time distributions in simple decision tasks in such a simple conceptual manner. However, no model of decision would be complete without being able to explain how wrong decisions come about: people often make errors in their choices. To address this important problem, we needed a task that would induce subjects to make a large number of errors, such that the latencies of these responses could be studied in detail to enable us to gain insight into the underlying decision process. The go/no-go task is an experimental paradigm in which a visual stimulus is presented that signifies the subject to make a saccade toward it, but on some occasions a different stimulus is presented that the subject has been pre-warned not to respond to. Although the instructions are clear to the subject, they still respond to the latter stimulus sometimes—this is classed as an error, a wrong decision. How can we make this task produce many errors? We did this by making the two types of visual stimuli with the same shape and size, but of different color—a red dot and a blue dot, one of which is "correct" and one is an "error." In this way, color is the only distinguishing feature of the two stimuli, making it rather easy for a subject to make an error by simply responding to a novel stimulus of the "wrong" color (Noorani et al., 2011).

The resulting latency distributions from this task are rather distinctive (**Figure 5**). If the error response and correct response distributions are plotted separately on a reciprobit plot, the two distributions initially overlap, but after a further 60 ms or so, the two distributions begin to diverge with the error distribution starting to flatten off whilst the correct distribution rises further. This can be explained by the arrival of color information in the cortex after this extra delay, allowing the cortex to make a correct decision of whether to respond based on the color of the stimulus. Before this time, such information is not available to the cortex to make an informed decision and therefore the probability of making a correct response is the same as that of an error. A LATERian approach to modeling these data takes us back to the countermanding model, wherein the stop unit is of primary importance. The initial part of both correct and error distributions is modeled by an "existence" unit which rises toward threshold on presentation of any visual stimulus, regardless of its color. If this reaches threshold, a saccade will occur toward the stimulus, and this can therefore be a correct or an error response depending on the color of the stimulus. It will typically be a quick response, as the unit accumulates immediately on appearance of a stimulus. However, after another 60 ms delay when color information arrives in cortex (Thompson et al., 1996; Schall and Bichot, 1998), a separate "color" LATER unit and a stop unit activate. The stop unit has a fast mean rate of rise, enabling it to frequently cancel the existence unit (assuming it has not yet reached threshold), whilst the color unit will produce a correct saccade if it wins the race (**Figure 6**). This explains why the correct latency distribution in the go/nogo task has two steep components: the first one being generated by the existence unit, and the later part being produced by the color unit. The stop unit ensures few errors are made after color information arrives; hence the error distribution flattens off.

#### **THE ANTI-SACCADE TASK**

Another and perhaps more complex kind of saccadic decision task is the anti-saccade task. Anti-saccades are saccadic eye movements in the opposite direction to a visual stimulus. This is a much more challenging response than a typical saccade because it requires a subject to withhold a saccade to a novel stimulus and instead look away from it (**Figure 7**). Consequently, anti-saccades are often used as an experimental paradigm for studying behavioral control, and are increasingly being examined in clinical conditions such as Parkinson's disease and Alzheimer's disease as a marker of voluntary control (Munoz and Everling, 2004). Anti-saccades are typically slower than normal saccades, and in this task a prosaccade to the visual stimulus is an error (Fischer and Weber, 1992).

Given their wide relevance, it is important to understand the neural processes by which anti-saccades come about. A study in monkeys using neuronal recordings demonstrated that a vector inversion is calculated in the lateral intraparietal area after a 50 ms delay, a necessary prerequisite for producing a movement in the opposite direction to a stimulus (Gottlieb and Goldberg, 1999; Zhang and Barash, 2000). Using a similar approach as for the go/no-go task, our laboratory asked subjects to perform the antisaccade task and we plotted their data as separate anti-saccade and error distributions on a reciprobit plot. The distributions for this task are unique: the error distribution begins earlier than the anti-saccade one and levels off earlier too. The error rate varies greatly from subject to subject but typically is around 5–30%. A LATER modeling approach demonstrated that a three-unit model

fitted the data most accurately with the fewest free parameters, and this was composed of an error unit that responds to the sud-

den appearance of a novel stimulus, a stop unit which acts to cancel the developing error response, and the anti-saccade unit which begins accumulating after a vector inversion delay of 50 ms (Noorani and Carpenter, 2013; **Figures 8**, **9**).

To perform an even more stringent test of this model, subjects were asked to perform the task under varying conditions of prior probability as this is known to affect the distributions (Koval et al., 2004). For example, in one version of the task there was an 80% chance of the stimulus appearing on the left and a 20% chance of it appearing on the right. These conditions created large shifts in the latency distributions of both correct and error responses and also greatly affected the error rates. Just as for a simple saccadic task wherein prior probability is explained by changes in the distance to threshold of a LATER unit, it was hypothesized that this alteration of the appropriate LATER units would account for the changes in the observed latency distributions of the antisaccade task under varying conditions of subject expectation. Indeed this was the case: altering the distance to threshold of the decision units accurately predicted the anti-saccade and error distributions with different prior probabilities. Previous models of antisaccades have not incorporated a stop unit (Kristjansson et al., 2001; Cutsuridis et al., 2007). Crucially, alterations in the stop unit's distance to threshold were necessary to predict the large changes in error rates when prior probability is altered, highlighting the importance of the stop unit in this voluntary behavioral task (Noorani and Carpenter, 2013).

A widely recognized observation in the antisaccade task is the correctional eye movements that occur following an error: subjects often tend to make an anti-saccade after they make an error saccade in order to correct their mistakes (Mokler and Fischer, 1999). Corrections generally occur after the vast majority of errors in the anti-saccade task, and they are typically quick responses. However, how such corrections would be generated from a neural race model is not immediately obvious since the first response in the task (error or antisaccade) is the result of a decision unit having won the race and implying the race has ended. How can

another decision be made after the race has already finished? In order to answer this question, the latency distributions of these separate responses had to first be analyzed in detail. A separate study was designed to record these new correction responses, in addition to errors and anti-saccades, allowing the three distributions to be plotted separately. The correction distribution was seen to be shifted to later time points compared to errors and antisaccades, with a frequency typically slightly less than errors (after all, most errors are corrected). Two possibilities could potentially explain the correction distribution from the basic anti-saccade model:

(1) Race continues after the error unit wins. If the error unit wins and an error is thus produced, a correction can ensue if the race is allowed continue thereby allowing the accumulating anti-saccade unit to carry on rising toward threshold and generate an anti-saccade. Although this is a plausible solution, this model predicts only fast corrections, so is not able to capture the whole reaction time distribution for corrections.

(2) Race re-starts after the error unit wins. This is a novel concept for race models. Instead of the race completely ending when an error has been made, the error unit finishes the race as expected but then the antisaccade unit re-starts from scratch. It as if the brain knows an error has been made and in order to correct the mistake it re-sets the antisaccade unit, this time with no competition from the error unit. Using the same parameters as for the basic anti-saccade model, this new model accurately predicted all latency distributions in detail, including that of the correction responses (Noorani and Carpenter, 2014).

In order to re-start a race after a decision unit has won, there must be a way of monitoring the outcome of decision races. Such an idea is not itself a new one, for example there is some evidence that the supplementary eye field monitors the results of decision processes regarding eye movements (Carpenter, 2004) and this area is thought to be important in anti-saccades (Chapman and Corneil, 2014).

#### **REACTION TIME AS A CLINICAL BIOMARKER**

Saccades are increasingly being explored as potential novel biomarkers of neurological and psychiatric diseases, in particular for improving diagnostic accuracy and monitoring disease progression (Leigh and Kennard, 2004). For example, abnormalities in saccadic latency have been found in Parkinson's disease, and these patients also have higher error rates in the anti-saccade task (Anderson and MacAskill, 2013). Anti-saccade deficits have been found in patients with Huntington's disease too (Peltsch et al., 2008). LATER models of the basic saccadic step task and the increasingly complex types of saccadic task promise to be helpful in improving analysis and interpretation of eye movement deficits in such neuropsychiatric diseases (Antoniades et al., 2010). The main advantage of the models presented in this review is that they have been grounded in detailed quantitative predictions of full reaction time distributions: a very stringent test of any decision model. This promises to be useful when applied to reaction time distributions to patients with these conditions, in which mean latencies and other task parameters are abnormal, as it will allow correlation of task deficits with model parameters. Indeed, a recent international consensus has been reached regarding an optimal protocol of the anti-saccade task for use in clinical research (Antoniades et al., 2013).

#### **CONCLUSIONS AND FUTURE DIRECTIONS**

The use of saccades in neural decision research has proved a very useful approach. In particular, simulating complete reaction time distributions has been a triumph of the LATER models of decision, which in recent years has been applied to increasingly complex tasks including the go/no-go and antisaccade tasks. Given their simplicity, LATER models have proved useful for conceptualizing advanced decision processes. Clinical research is beginning to incorporate saccadic latency as a poten-

tial biomarker for neuropsychiatric disease. Recent research has excitingly begun to provide insight to how decisions with multiple options are generated, with evidence suggesting that this may involve a race between different neural pathways (Schmidt et al., 2013), rather like the way in which the LATER model has been applied to explain behavioral data in such decision tasks. Emerging work is also beginning to directly link neuronal activity with parameters of accumulator decision models (Purcell et al., 2010, 2012). Challenges of the future will be to correlate LATERian predictions with neuronal activity by direct neuronal recordings in order to demonstrate where and how such decision processes occur in the brain. We have seen how complex tasks can be modeled with multiple LATER units representing different possible response options, and it is likely that these units can be correlated with groups of neurons in the known oculomotor regions whose activity represent developing decisions.

#### **REFERENCES**


**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: 12 April 2014; accepted: 02 August 2014; published online: 22 August 2014. Citation: Noorani I (2014) LATER models of neural decision behavior in choice tasks. Front. Integr. Neurosci. 8:67. doi: 10.3389/fnint.2014.00067*

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

*Copyright © 2014 Noorani. 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.*

# PART II – NEURAL BASES OF VISUAL PERCEPTION AND BINOCULAR VISION

## Cortical and white matter mapping in the visual systemmore than meets the eye: on the importance of functional imaging to understand visual system pathologies

## *Noa Raz and Netta Levin\**

*fMRI Unit, Neurology Department, Hadassah Hebrew University Medical Center, Jerusalem, Israelm*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, The Neuropsychological Laboratory, France*

#### *Reviewed by:*

*Marcelo Fernandes Costa, Universidade de São Paulo, Brazil Hanna Heikkinen, Aalto University, Finland*

#### *\*Correspondence:*

*Netta Levin, fMRI Lab, Neurology Department, Hadassah Hebrew University Medical Center, POB 12000, Jerusalem 91120, Israel e-mail: netta@hadassah.org.il*

Information transmission within the visual system is highly organized with the ultimate goal of accomplishing higher-order, complex visuo-spatial and object identity processing. Perception is dependent on the intactness of the entire system and damage at each stage—in the eye itself, the visual pathways, or within cortical processing—might result in perception disturbance. Herein we will review several examples of lesions along the visual system, from the retina, via the optic nerve and chiasm and through the occipital cortex. We will address their clinical manifestation and their cortical substrate. The latter will be studied via functional magnetic resonance imaging (fMRI) and Diffusion Tensor Imaging (DTI), enabling cortical, and white matter mapping of the human brain. In contrast to traditional signal recording, these procedures enable simultaneous evaluation of the entire brain network engaged when subjects undertake a particular task or evaluate the entirety of associated white matter pathways. These examples provided will highlight the importance of using advanced imaging methods to better understand visual pathologies. We will argue that clinical manifestation cannot always be explained solely by structural damage and a functional view is required to understand the clinical symptom. In such cases we recommend using advanced imaging methods to better understand the neurological basis of visual phenomena.

#### **Keywords: visual cortex, visual pathways, visual pathologies, functional MRI (fMRI), Diffusion Tensor Imaging (DTI)**

In order for visual perception to occur, the physiological signal initiated by the photoreceptors in the retina must travel all the way to the visual cortex at the back of our brain. This long journey requires a set of long-range white matter tracts to communicate the signals.

The retinal ganglion cell axons leave the eye by way of the **optic nerve**, and there is a partial crossing of axons at the optic chiasm. After the chiasm, the axons (which now carry binocular information) are referred to as the **optic tract**. The optic tract wraps around the midbrain to get to the Lateral Geniculate Nucleus (LGN), where all the axons synapse. From there, the LGN axons fan out through the deep white matter of the brain as **optic radiations**, which will ultimately travel to the primary visual cortex. The **occipital-callosal fiber tract** that connects the two occipital lobes via the corpus callosum is a key pathway for the flow of information between the two hemispheres (Barr and Kiernan, 1993). The visual information is then processed in the visual cortex. The visual cortex is divided hierarchically into primary (lower in the hierarchy) areas located in the occipital lobe and late (higher in the hierarchy) areas, which are located in the temporal and parietal lobes. Mapping in primary visual areas is sensitive to the stimulus' topography, known as**retinotopic mapping**; adjacent points in the visual field are represented close to each other in the visual cortex such that the cortical representation reflects the retinal geometry (Wandell, 1999). The upper visual field is mapped within the lower lip of the calcarine sulcus and the lower visual field in its upper part; similarly, the right visual field is mapped in the left hemisphere and the left visual field in the right side. The eccentricity axis along the central to peripheral visual field is represented topographically along the calcarine sulcus from the posterior to the anterior parts. A relatively large cortical area is devoted to the fovea; large numbers of small receptive fields' neurons process information from a small region of the visual field, enabling the fine spatial resolution that characterizes foveal vision.

Mapping in higher visual areas (higher in the hierarchy) located in the occipito-parietal and occipito-temporal regions is less sensitive to stimuli topography and more involved in processing the **functional aspects** of the stimuli. This includes areas in the ventral and the dorsal streams that analyze an object's identity and object's usage, respectively (Goodale and Milner, 1992).

The possibility of mapping the functioning human visual cortex has developed during the last two decades using Functional MRI (fMRI) which utilizes the neuro-vascular coupling principle: in response to a visual task, neural activity in a specific cortical visual region will trigger change in the regional blood flow which will be captured in the MRI scan (Logothetis, 2008). *Retinotopic mapping* is obtained by presenting simple visual stimuli to the different visual fields' portions. For example, in order to map the eccentricity axis from the central to the peripheral visual field, an expanding ring stimulus is used. The subject views an expanding flashing checkerboard circular stimulus that gradually expands from the center to the periphery of the visual field. This stimulus creates a wave of neuronal activity, which moves along the calcarine sulcus from the posterior to the anterior region (Engel et al., 1997). An example of such mapping can be seen in **Figure 1**. The cortical magnification factor should be noted: a large cortical area is devoted to the processing of the 2◦ of the central visual field.

*Functional mapping* is obtained using more complex visual stimuli; targeting the cortical areas activated during the presence, as compared with the absence, of specific visual attributes. For example in order to map cortical areas specialized in object recognition, the subject is alternately presented with either pictures of objects or a scrambled version of these objects (Malach et al., 1995). While the basic parameters of the stimuli are identical in both conditions, object perception exists in the first but not in the second condition. Likewise, in order to map cortical areas that specialize in motion processing, the subject will be presented with moving and stationary objects identical in all aspects except their mobility (Tootell et al., 1995). **Figure 2** demonstrates the lateral occipital cortex (LOC), specifically activated during visual object processing and the motion sensitive region, MT.

Early usage of fMRI was dedicated to mapping the healthy human visual cortex, adding a significant layer to the data collected in animal and human lesion studies. The ability of the fMRI signal to reflect neuronal mass activity (reflecting the average activity of ∼ millions of neurons) served to explain the functioning of distributed systems, such as our visual pathways. This accumulated knowledge of the intact visual pathways provided the basis for understanding the neuronal substrate of unexplained visual dysfunctions, which cannot be addressed via standard anatomical imaging regime. fMRI has contributed, for example, to our understanding of the phenomenon of blindsight, in which a patient with cortical damage resulting in hemianopia retains some residual abilities within his blind field. fMRI studies in these patients demonstrated powerful activation in area MT, despite nearly complete destruction of the primary visual region (V1)

cortical maps. The cortical activity pattern in response to the expanding ring (top) or the rotating wedge (bottom) is overlaid on T1 slices (presented in neurological convention, i.e., right in the picture correspond to right hemisphere), **(B)** and on the inflated posterior view of the right hemisphere **(C)**. The arrow represents the wedge rotation.

(ffytche et al., 1996; Goebel et al., 2001). This provided powerful support for the hypothesis that blindsight is based on signals that travel via alternate pathways, projecting via the superior colliculus and pulvinar to area MT (Wandell and Wade, 2003).

Similarly, fMRI was useful for explaining visual hallucinations, demonstrating that hallucinations of color, faces, textures, and objects were accompanied by signal increases in visual areas whose specializations matched the induced hallucination contents (Ffytche et al., 1998) and by changes in fMRI connectivity between LGN and the visual cortex (Ffytche, 2008). fMRI has also been useful for explaining the neuronal basis of compensatory behavior in patients with no visual inputs, recruiting the visual cortex to improve non-visual functions. As was demonstrated in congenital blindness, lower, and higher-order cortical visual regions are activated during non-visual tasks. Furthermore, these activations were associated with improvement of non-visual functions (e.g., Sadato et al., 1996; Amedi et al., 2003; Raz et al., 2005).

Tracking white matter is done using Diffusion Tensor Imaging (DTI). The method is based on diffusion, the random motion of water molecules in the extra- and intra-cellular spaces. When the motion of molecules is limited by *tubular* structures such as the axons, diffusion will be mainly along the direction of the axonal bundle, rather than in the perpendicular plane (anisotropic diffusion). Using this method, information regarding localization and directionality of the fibers as well as evaluation of the quality of their cohesiveness and their ability to conduct the neuronal signal can be obtained (Basser et al., 1994; Mori et al., 1999). **Figure 3** demonstrates reconstruction of the visual pathways using DTI and fiber tractography, and shows the *optic tract* that connects the optic chiasm to the LGN, creating a rather thin and compact fiber bundle; and the *optic radiation* that connects the LGN to the occipital cortex containing its anterior segment, the Meyer's loop.

Fibers' cohesiveness may be evaluated by four diffusion properties: mean diffusivity (MD); fractional anisotropy (FA, a scalar

**FIGURE 3 | DTI visual pathways fiber groups superimposed on T1 images. Left:** axial view of the optic tracts and radiations in one control subject. The optic tracts, which connect the optic chiasm to the LGN, are colored red for the right side and blue for the left side. The right and left optic radiations, which connect the LGN to the calcarine cortex, are colored pink for right and light blue for left. **Right:** sagital view of the left optic tract and radiation.

value describing the degree of anisotropy of a diffusion process); diffusion parallel to the principal fiber direction (axial diffusivity, AD); and diffusion perpendicular to the principal fiber direction (radial diffusivity, RD) (Basser et al., 1994). Animal model studies of axon and myelin pathology have demonstrated that reduced AD is associated with axonal pathology, whereas increased RD is associated with demyelination (Song et al., 2003).

Visual perception is dependent on functioning of the visual system in its entirety and perception may be impaired by damage to any of the pathway's components, starting at the eye itself, through the visual pathways and continuing to the visual processing site, the cortex. Herein we will briefly review examples of different insults along the visual system that we have encountered in patients who have presented during recent years, and their clinical and imaging manifestations (**Figure 4**).

## **ACQUIRED EYE INSULT IN THE YOUNG (FIGURE 5)**

Case 1: MM, a 53-year-old male, lost his left eye and became blind in the right due to corneal damage at the age of 3. At age 46, following more than 40 years of blindness, MM underwent corneal and limbal stem-cell transplant in his right eye and regained his right retinal image. Nevertheless, even 7 years following surgery, his visual abilities remained severely limited, and he didn't rely on vision for his daily life functioning (poor spatial resolution, especially in high frequencies were evident).

Functional MRI, including retinotopy mapping and population receptive field (pRF) estimation revealed several differences between MM and sighted controls (Levin et al., 2010); MM's

**FIGURE 4 | Case reports localizations.** The figure localizes the sites of dysfunction in the different cases reported: case 1—the eye, case 2—the optic nerve, cases 3 and 4—optic chiasm, cases 5 and 6—the visual cortex.

left: The expanding ring stimulus used to map the visual field eccentricity axis. Red represents mapping the center of the visual field, yellow the parafoveal region and blue- green the peripheral areas. On the right: Activation along the calcarine sulcus (Ca) overlaid on posterior medial view of the right hemisphere. In contrast to the normal eccentricity mapping where the center of the visual field has large representation in the occipital cortex, MM lacks this foveal

representation. **(B)** On the right: three-dimensional rendering of the optic tract note the missing left eye. The optic tracts connect the optic chiasm and the LGN (white sphere). On the left: scatter plot of the radial and axial diffusivities for the average of the right and left optic tracts. Data are from MM (gray star), 10 normal controls (black open circles), 2 seeing monocular subjects (black asterisks) and one blind subject (black closed circle). The 2-standard deviation covariance ellipsoid (dashed) is shown.

retinotopic maps were almost entirely lacking the foveal representation. Furthermore, while posterior occipital regions typically include small pRFs, these cortical regions in MM were characterized by large pRFs. Using DTI and fiber tractography (Contrack algorithm Sherbondy et al., 2008), MM's optic tract and radiations were identified. While their anatomical appearance seemed intact, their microstructural measurements were significantly different from those of controls, with significantly impaired axial diffusivity (suggesting axonal loss) (see **Figure 5**).

MM's neuroimaging data suggest a specific cause of the limited recovery. His vision was interrupted at a time when visual neurons with small receptive fields were developing. Because these neurons are important inputs to the visual pathways of the brain, used for object recognition, MM has both poor resolution and limited ability to interpret objects.

This case serves as an example, which demonstrates that restoration of functional vision requires more than improving the retinal image contrast. To obtain functional vision, in addition to restoration of the sensing organ, the developmental trajectory of the individual and the consequences of the early deprivation on cortical circuitry need to be taken into account.

#### **INFLAMMATORY DAMAGE TO THE OPTIC NERVE**

Case 2: Optic neuritis is a demyelinating disease of the optic nerve, causing acute visual loss in young adults. Optic neuritis can be clinically isolated but more frequently appears as one of the manifestations of multiple sclerosis (MS). Early in the course of the disease the inflammatory process causes a significant reduction or even blockage in the amount of information being transferred along the optic nerve. This is expressed clinically as deficits in visual field, visual acuity, contrast sensitivity, and color perception. Later on, when the inflammation is alleviating, these functions are usually restored. Nevertheless, due to damage to the myelin surrounding the nerve fibers, there is a continuing delay in conduction along the optic nerve; which can last for several months or even years following the acute attack. Consequently, patients that are reported to have recovered according to standard visual testing continue to complain about residual visual dysfunction in their daily visual abilities. Recently, we demonstrated that static visual functions (visual acuity, visual fields, contrast sensitivity, and color perception) tend to recover shortly following the acute optic neuritis attack (weeks to months). Yet patients with optic neuritis continue to exhibit major deficiencies in dynamic visual functions (motion perception) even a year following the attack (Raz et al., 2011). We suggested that these perceptual differences between the ability to perceive static and dynamic visual functions reflect different visual processes; while static visual abilities depend solely on the amount of information being transferred along the optic nerve, the dynamic visual functions also depend on conduction velocity and therefore remain impaired for longer periods (Raz et al., 2012).

#### **THE CORTICAL REFLECTION OF MOTION PERCEPTION DEFICIT FOLLOWING OPTIC NEURITIS**

In recent years, fMRI has been used to examine the possibility that cortical mechanisms participate in the recovery process following peripheral visual insult, in addition to the afferent recovery processes. Previous fMRI studies on patients who had recovered clinically from optic neuritis showed an intact activation level in the object-related visual regions during stimulation of the affected eye. This was evident when activation in early visual areas was intact but also when it was reduced. Intact activation in higher visual areas was considered as evidence of cortical plasticity, where cortical adaptation to a persistent abnormal input contributes to the recovery process (Werring et al., 2000; Toosy et al., 2005; Levin et al., 2006; Korsholm et al., 2007; Jenkins et al., 2010).

To assess the role of cortical plasticity in static and dynamic visual recovery, we compared cortical activity in response to these stimuli in patients 12 months following the optic neuritis attack (for further details on fMRI procedure see Raz et al., 2011). **Figure 6** shows fMRI activation maps in optic neuritis patients and controls while viewing static objects and an expandingcontracting dots array. Activation is seen within the object and motion—related cortical regions (LOC and MT, respectively), which were first identified using separate functional localizers. While intact cortical activity was evident in visual areas responsible for object recognition, the activation in visual areas responsible for motion processing was reduced. Theses imaging results correspond to the behavioral data, reflecting the recovery of static but not dynamic visual processing 12 months following an optic neuritis attack.

We have proposed that the cortical activity following optic neuritis may reflect the visual percept (intact for visual acuity

**FIGURE 6 | Static and dynamic visual processing in optic neuritis patients: behavioral and cortical findings. (A)** Visual acuity (upper) and motion perception (expressed as the ability to extract motion-defined objects, lower panels) in the affected eyes of optic neuritis patients over time. (Patients' data is normalized to controls' mean). *N* = 21, 20, 18, and 14 at the acute, 1, 4, and 12 month time points, respectively. Black asterisks denote significant reduction in patients' values as compared to the controls. Gray asterisks denote significant change in patients' affected eyes' measurements between testing phases. ∗*p <* 0*.*05; ∗∗*p <* 0*.*01; ∗∗∗*p <* 0*.*001. **(B)** fMRI activation maps showing activation within the object-related region (LOC) during static object

recognition (upper) and within the motion-related region (MT) during motion processing (lower). The data are presented on a Talairach normalized inflated brain of the left hemisphere. LOC and MT are outlined on the lateral view of the cortex (LOC—purple lines, MT—green lines). Blow-ups highlight activation in the regions of interest (ROIs). Activation is seen for control subjects and affected eyes of optic neuritis patients. Color scale denotes significance levels. Bar graphs on the right denote the activation levels (beta weights) within each ROI for the two groups. Comparable cortical activation levels for the AEs and controls are found in LOC. However, a significant reduction in cortical activation during AE stimulation is evident in MT.

and impaired for motion perception) rather than demonstrating cortical plasticity, as suggested previously.

#### **THE EFFECT OF DEMYELINATIVE DAMAGE ON NEIGHBORING WHITE MATTER INTEGRITY**

In recent years DTI and fiber tractography have been used to study the effect of focal damage to neighboring pathways. As was reported in animal studies, neuronal loss following damage is often greater than might be expected from the severity of the injury to the nerve itself. We utilized the visual pathways, which comprise a well-defined system, and optic neuritis, which is usually a spatially and temporally discrete event, to study the effect of focal demyelinative lesions on neighboring white matter.

To assess this, we delineated the optic tracts and radiations in 17 optic neuritis patients (12–36 months following an optic neuritis attack) and 12 matched control subjects using DTI and fiber tractography methods and measured the directional diffusivities in those fibers. DTI data was acquired using a diffusionweighted imaging sequence (2 mm thick slices covering the whole brain; *<sup>b</sup>* <sup>=</sup> 0 and *<sup>b</sup>* <sup>=</sup> 1000 s/mm2. The high *<sup>b</sup>*-value was obtained by applying gradients along 64 different diffusion directions). Image processing was done using the open-source mrVista package (http://vistalab*.*stanford*.*edu/software). Fiber tractography was performed using the probabilistic Contrack algorithm (Sherbondy et al., 2008). Optic tracts were estimated as the most likely pathways between the right or left sides of the chiasm, and the lateral geniculate nuclei (LGN) ROIs on the corresponding hemisphere. Optic radiations were estimated as the most likely pathways between the LGNs and the calcarine sulci ROIs in the corresponding hemisphere. Chiasm, LGN and calcarine ROIs were delineated for each subject on T1 images (**Figure 7A**). Diffusion measures along the optic tract and radiation bundles were re-sampled at 30 and 50 positions, respectively, calculating FA, AD, and RD at each of these nodes. In this way, measures throughout the length of the fiber could be combined across different subjects. In order to avoid partial voluming with non-white matter (i.e., ventricles or gray matter), diffusion measurements were taken near the dense core of the fibers.

Our results demonstrated reduced AD in the optic tracts of optic neuritis patients (**Figure 7B**). Furthermore, AD correlated with the corresponding Retinal Nerve Fiber Layer (RNFL) thickness in the patients' affected eyes (linear least-squares regression with calculation of the correlation coefficient *F* = 10*.*2; *p* = 0*.*01; *r* = 0*.*71). As opposed to the optic tracts, AD within the optic radiations was similar among patients and controls (**Figure 7B**). These findings suggest that axonal loss in the optic nerves of chronic optic neuritis patients proceeded to the optic tracts, demonstrating Wallerian degeneration. However, this process did not proceed to the optic radiations, and therefore did not support anterograde trans-neuronal degeneration. Despite the intact AD along the optic radiations, reduced FA and an elevated RD were evident in the patients (FA: 0.52 and

**FIGURE 7 | Axial diffusivity in the optic tracts and radiations following optic neuritis. (A)** Fiber delineation: The optic tracts (purple) and optic radiations (blue) superimposed on T1 image in one control subject. **(B)** Axial diffusivity in the optic tracts (upper plots) and radiations (lower plots) of optic neuritis (*n* = 17) and control subjects (*n* = 12) (dark and light gray

symbols, respectively). Diffusivities are presented at each point along the tract. Inserts above plots represent differences between ON and control groups at any point along the tract. Differences are represented as *p*-values (*T* -Tests, corrected for multiple comparisons). *P-*value scale is shown at the rightmost color-bar.

0.47 for controls and patients, respectively, *p* = 0*.*003 between group. RD: 0.58 and 0.62 for controls and patients, respectively, *p* = 0*.*046 between groups). Furthermore, FA and RD levels were associated with the presence of demyelinative lesions within the optic radiations (correlations between optic radiations intra-bundle lesions' volume and FA values: *F* = 5*.*8; *p* = 0*.*03; *r* = −0*.*53. Correlation between optic radiations intra-bundle lesions' volume and RD values: *F* = 10*.*7; *p* = 0*.*005; *r* = 0*.*65), suggesting that white matter damage in these fibers could be explained by local demyelinative damage in the optic radiations (Raz et al., 2014). Our results in the visual pathways of optic neuritis patients may model normal-appearing white matter (NAWM) pathology in MS, demonstrating that a demyelinative lesion in the proximal segment of the nerve fiber may result in chronic degeneration of its distal normal-appearing portion.

#### **ACQUIRED CHIASMATIC INSULT**

Acquired chiasmal abnormalities are usually categorized as intrinsic or extrinsic; involving the substance of the optic chiasm itself (e.g., in inflammatory processes), or resulting from a mechanical compression of adjacent structures (e.g., in adenomas) (Foroozan, 2003). Bitemporal hemianopsia is the classical visualfield-defect of disorders that involve the optic chiasm (Kirkham, 1972), due to the involvement of the crossing of nasal-retinal fibers of each optic nerve (leading to inability to view the temporal visual field of each eye). Resolution of inflammation or surgical removal of the compressing lesion can improve visual field and visual acuity.

#### **THE CORTICAL REFLECTION OF TRANSIENT INFLAMMATORY BITEMPORAL HEMIANOPSIA**

Case 3: We studied the effects of intrinsic optic chiasm damage in a 36-year-old woman, who presented with acute bilateral visual loss (Raz et al., 2010). Her past history included recurrent transverse myelitis episodes and a serologically positive test for Neuro-myelitis optica (NMO) antibodies. On admission, her visual fields demonstrated bitemporal defects, which were more pronounced in the lower visual field. T1, T2, and fluid attenuation inversion recovery–weighted MRI were performed in the acute phase. Several periventricular white matter lesions were evident, with no detectable optic chiasm abnormality. To further assess her bitemporal hemianopsia, an fMRI experiment was performed. During the fMRI scan, a flickering checkerboard was projected separately to each eye in each visual field. Stimuli were projected to the lower visual fields and were designed to fall within the scotoma of the patient when projected to her blind fields. This paradigm was carried out twice: during the acute phase and after visual recovery (8 months later). During the acute phase, visual cortical areas were activated only monocularly, whereas recovery was associated with the return of normal binocular input. These results confirm the association between bitemporal hemianopsia and an ipsilateral-only projection of the retinogeniculate pathway (**Figure 8**).

**FIGURE 8 | Cortical activation patterns during acute chiasmal inflammation and following recovery. (A)** Patient's visual fields during the acute phase (left) and subsequent recovery (right). **(B)** fMRI study design: Four experimental conditions were used: (1) Stimuli projected to the LVF via the LE, marked in yellow; (2) Stimuli projected to the RVF via the LE, marked in light green; (3) Stimuli projected to the LVF via the RE, marked in red; (4) Stimuli projected to the RVF via the RE, marked in dark green. LVF—left visual field; RVF—right visual field; LE—left eye; RE—right eye. **(C)** Coronal and axial

views of the fMRI activation patterns, elicited during the four experimental conditions. Activation is shown for the patient during acute phase (left) and following recovery (right). Same colors as in **(B)** (see scale). Conv. (convergence) refers to the overlap of activation elicited by stimuli projected to both eyes. Lower panel: Average activation level (percent signal change) during stimuli presentation in V1, showing the relative contribution of all experimental conditions. Asterisks denote significance level: ∗*p <* 0*.*05; ∗∗*p <* 0*.*01.

#### **THE EFFECT OF CHIASMAL COMPRESSING MENINGIOMA ON PROCEEDING WHITE MATTER INTEGRITY**

Case 4: The effects of extrinsic optic chiasm damage were studied in a 35-year-old man presenting with bilateral hemianopsia, which he noticed 6 months prior to admission. Apart from the visual field loss, his ophthalmological examination was intact. Magnetic resonance imaging (MRI) showed a giant tuberculum sellae meningioma compressing the entire chiasm. The patient underwent right fronto-orbital craniotomy and the lesion was completely resected. Visual fields improved significantly, although some residual deficit was still evident (**Figure 9A**). To study the effect of the compression on the chiasm and optic tracts, DTI and fiber tractography (as detailed in case 2) were applied prior to and following surgical removal of the meningioma (**Figure 9B**). As seen in **Figure 9C**, the compression resulted in significant reduction of the FA as a measure of microstructure integrity (as compared to FA measured in the optic tracts of 17 agematched control subjects). Following meningioma removal, FA was enhanced, although it remained significantly impaired. This sustained reduction in diffusivity may explain the residual visual loss following meningioma removal.

#### **ACQUIRED VISUAL CORTICAL DAMAGE**

## **FUNCTIONAL MAPPING MAY EXPLAIN A SPECIFIC VISUAL DYSFUNCTION**

Case 5: A 65-year-old- woman was admitted to the rehabilitation department following diffuse occipital and occipito-parietal damage due to Posterior Reversible Encephalopathy Syndrome (PRES) (Hinchey et al., 1996). Behavioral tests suggested a specific deficit in face processing (inability to recognize famous faces despite the ability to recognize these persons following a verbal description, and a failure to recognize the pictures of her grandchildren). This deficit was seen, despite the intact processing of other object identity information (naming two-dimensional and three-dimensional visual objects, color naming). Such a behavioral deficit suggested specific damage in the brain areas responsible for face processing, while adjacent areas responsible for processing other objects remained intact. Since the distinction between damage in these different areas can only be demonstrated on the basis of function and in order to establish the specificity of face perception damage, the patient underwent an fMRI assessment (multi-slice gradient echo-planar imaging, 3 mm thickness axial slices covering the whole brain). Her examination included mapping of different areas along the hierarchy of the visual cortex, including primary visual areas and higher visual areas responsible for processing objects, motion, places, and faces. To map the object-related region (LOC), blocks of objects and scrambled version of these objects were contrasted; and to map motion-related regions (MT), blocks of expanding-contracting rings and stationary rings were contrasted. To compare activation for faces and places processing, blocks of famous faces, and places were shown. Data analysis was performed using the BrainVoyager QX software package (Brain Innovation).

The exam demonstrated a normal pattern of activation in early visual areas, as well as in areas responsible for objects and motion processing. In the parahippocampal place area (PPA), which is

**FIGURE 9 | Optic tracts' diffusivity measurement during chiasmal compression and following recovery. (A)** Patient's visual fields prior and following surgical removal of the meningioma. **(B)** Optic tracts delineation in the patient (blue fibers), superimposed on axial and coronal view. The compressing meningioma is seen in white. Red spheres represent the left and right LGNs. **(C)** Fractional anisotropy (FA) along the patient's optic tracts

(averaged across the left and right sides) prior to and following surgical removal of the meningioma (dark and light blue, respectively). Patient's data is plotted against the mean FA of 17 age-matched control subjects (white plot). Error bars represent standard deviations. X axis defines position along the tract, from the chiasm to the LGN, given in arbitrary units, 30 positions were sampled in each subject.

known to be involved in processing topographical scenes, greater activity was evident when viewing pictures of places as compared with pictures of faces (a pattern that resembles normal-sighted subjects). In contrast, no specification for face recognition was evident in the ventral-temporo-occipital fusiform face area (FFA), responsible for face processing (**Figure 10**). PPA and FFA were defined anatomically. This case clearly demonstrates the ability of functional imaging to bridge the gap between the clinical deficit and the associated cortical activity. The fMRI examination revealed an unusual pattern of activation in an area known to be responsible for face recognition, which anatomically appeared intact (via standard MRI regime).

#### **FUNCTIONAL MAPPING MAY PREDICT RECOVERY OF VISUAL DYSFUNCTION**

Case 6: A 64-year-old woman was admitted to our clinic following repeated occipital ischemic events. She reported no light perception and behaved as though she was blind. Standard MRI revealed damage in her right early visual regions. Flash visual evoked potential (VEP) testing demonstrated delayed but evident cortical responses. To reveal whether the visual cortex received and processed the visual input, an fMRI examination was performed. fMRI examination included viewing full-field flickering checkerboard to activate early visual regions; viewing visual objects (vs. scrambled images of these objects) to activate LOC; viewing pictures of faces (vs. scrambled images of these faces) to activate FFA; and viewing expanding-contracting rings (vs. stationary rings) to activate MT. In addition, resting-state fMRI was applied to assess the visual cortical network.

Viewing flickering checkerboard resulted in intact activation at early visual regions within the undamaged left hemisphere. Intact activation was also evident within higher-order visual regions: activation of LOC in response to viewing objects, activation of the FFA in response to viewing faces, and activation of MT for viewing moving stimuli.

Functional connectivity during resting-state fMRI revealed an intact visual cortical network (**Figure 11**), accompanying early and higher-order visual regions. The sum of these fMRI results suggest intact processing of visual information within the visual cortical regions. Thus, the absence of light perception may result from unawareness of visual processing rather than dysfunction of the visual cortical regions. Demonstration of the intactness of visual processing is a prerequisite step for rehabilitation.

## **CONCLUSIONS**

Vision is a perceptual experience, which includes the retinal ability to perceive inputs as well as the cortical ability to process them. Vision is a complicated process of sorting, encoding, interpreting, and understanding the meaning of the visual input. Consequently, the image created in our awareness does not always reflect the optical reality. Areas devoted to visual information processing constitute a large portion of the human cortex, indicating their complexity. The visual cortex contains an impressively large number of distinct areas and specific functional networks that

**FIGURE 10 | Cortical activation patterns during faces and places recognition in a patient with acquired visual cortical damage: Correspondence between behavioral and cortical dysfunctions.** Cortical activation in the parahippocampal place area, PPA **(A)** and in the fusiform face area, FFA **(B)** during viewing pictures of faces and places. Upper panels represent the cortical activation maps centered on the two regions of interest (ROIs) (PPA marked in green, FFA marked in blue). Lower panels represent the activation elicited in each of these ROIs for the two stimuli type. Color scale denotes significance levels. The absence of cortical specification for processing faces corresponds to the patient's inability to process the identity of faces (while ability to process the identity of other visual stimuli is intact). MRI slices are shown in neurological convention.

create the neuronal basis for perceiving different visual elements of the stimuli (Van Essen and Drury, 1997). These specifications in visual processing are manifested via clinical cases, for example the inability to recognize faces while other visual abilities are intact.

In the past two decades, the use of non-invasive functional imaging and especially fMRI has dramatically expanded our knowledge about the functional visual human brain in healthy and disease states. Intact vision requires functioning of the entire visual system, starting at the eye, continuing through the visual pathways and ending at the visual cortex. Furthermore, the visual cortex is not isolated, and functioning separately from other areas won't achieve normal vision. Therefore, we should relate to the visual system with a holistic view, integrating imaging techniques for the eye, the visual pathways and the brain, in order to assess its overall functioning.

Taking advantage of their ability to visualize the whole brain at once, DTI and fMRI are adequate tools to assess the consequences of focal lesions along the visual pathways on white matter integrity and cortical functioning in the entire system.

As in the case of optic neuritis, DTI is useful for investigating the effect of focal lesions on neighboring white matter, modeling the relationships between intact and pathological white matter structures, or, as discussed in the case of chiasmal compression, DTI may explain residual visual loss after removal of the focal lesion.

Evaluating the functional properties of the entire brain, fMRI is useful for differentiating between functioning and dysfunctioning cortical regions, and explaining specific visual dysfunctions, as was seen in the case of acquired brain damage.

Regarding the system as a whole is important when developing advanced technologies to reconstruct the sensing organ. As the first case illustrates, cutting edge studies aimed at restoring vision such as retinal prosthesis or retinal stem cell transplant, cannot only focus on reconstruction of the sensing organ. They also have to account for the intactness or rehabilitation of the entire system.

fMRI can bridge the gap between MRI-detected tissue damage and clinical manifestations, explaining behavioral phenomena even in the absence of a clear structural fingerprint (Filippi and Rocca, 2003).

In the optic neuritis studies we emphasized the association between cortical activity and visual functioning, demonstrating motion perception deficits in both. In this way, the cortical activity reflects the behavioral phenomenon. In contrast, the last case demonstrated a mismatch between the behavior and the cortical activity pattern. Intact activation within the occipital visual cortex was evident despite the patient experiencing no light perception. These findings were used to leverage the rehabilitative process since the cortical residua indicated possible recovery.

The examples reviewed above emphasize the importance of using functional imaging to understand visual system pathologies. In some cases, standard MRI scans do not have enough resolution to localize the specific pathology. The clinical picture cannot be explained by clear structural damage and an understanding of the pattern of cortical activity and connectivity between different cortical and subcortical areas is needed to explain the visual deficit. To that end we recommend using advanced imaging techniques to better understand the neuronal basis of the apparent neurological event.

## **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: 02 May 2014; accepted: 04 August 2014; published online: 27 August 2014. Citation: Raz N and Levin N (2014) Cortical and white matter mapping in the visual system- more than meets the eye: on the importance of functional imaging to understand visual system pathologies. Front. Integr. Neurosci. 8:68. doi: 10.3389/fnint. 2014.00068*

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

*Copyright © 2014 Raz and Levin. 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.*

## Origins of strabismus and loss of binocular vision

#### *Emmanuel Bui Quoc <sup>1</sup> \*† and Chantal Milleret 2 †*

*<sup>1</sup> Ophthalmology Department, Hopital Robert Debre/Assistance Publique Hopitaux de Paris, Paris, France*

*<sup>2</sup> Collège de France, Center for Interdisciplinary Research in Biology (CIRB), Spatial Navigation and Memory Team, Paris, France*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

#### *Reviewed by:*

*Catherine Elizabeth Stewart, City University London, UK Robert Hess, McGill University, Canada*

#### *\*Correspondence:*

*Emmanuel Bui Quoc, Ophthalmology Department, Hopital Robert Debre/Assistance Publique Hopitaux de Paris, 48 Boulevard Sérurier, 75019 Paris, Ile de France, France e-mail: emmanuelbui@outlook.com*

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

Strabismus is a frequent ocular disorder that develops early in life in humans. As a general rule, it is characterized by a misalignment of the visual axes which most often appears during the critical period of visual development. However other characteristics of strabismus may vary greatly among subjects, for example, being convergent or divergent, horizontal or vertical, with variable angles of deviation. Binocular vision may also vary greatly. Our main goal here is to develop the idea that such "polymorphy" reflects a wide variety in the possible origins of strabismus. We propose that strabismus must be considered as possibly resulting from abnormal genetic and/or acquired factors, anatomical and/or functional abnormalities, in the sensory and/or the motor systems, both peripherally and/or in the brain itself. We shall particularly develop the possible "central" origins of strabismus. Indeed, we are convinced that it is time now to open this "black box" in order to move forward. All of this will be developed on the basis of both presently available data in literature (including most recent data) and our own experience. Both data in biology and medicine will be referred to. Our conclusions will hopefully help ophthalmologists to better understand strabismus and to develop new therapeutic strategies in the future. Presently, physicians eliminate or limit the negative effects of such pathology both on the development of the visual system and visual perception through the use of optical correction and, in some cases, extraocular muscle surgery. To better circumscribe the problem of the origins of strabismus, including at a cerebral level, may improve its management, in particular with respect to binocular vision, through innovating tools by treating the pathology at the source.

**Keywords: children, early strabismus, binocular vision, brain development, critical period**

## **INTRODUCTION**

Visual perception is optimal in humans at adulthood, providing that all the developmental processes in relation to it have occurred properly both before and after birth, including anatomical and functional processes. As illustrated in **Figure 1**, this includes not only the correct development of the eyes themselves, but also that of eye movements through the extraocular muscles (EOMs). In parallel, all of the central structures in the brain that are related to visual perception (including those concerned with eye movements) must also develop appropriately. As a result, for example, each neuron in primary visual cortex (V1) becomes progressively able to encode the different attributes of the visual scene, such as orientation and direction of movement. Progressively, most of them also become "binocular," i.e., able to be activated through both eyes while they are initially monocular (e.g., Frégnac and Imbert, 1978; Milleret et al., 1988). In parallel, cortical maps corresponding to each of these attributes of the visual scene, including the retinotopic map encoding space, develop (e.g., Chapman et al., 1996; Crair et al., 1998; Smith and Trachtenberg, 2007; White and Fitzpatrick, 2007; Tani et al., 2012 for review). The different types of eye movements (saccades, pursuits) also mature with age (e.g., Ingster-Moati et al., 2009; Bucci and Seassau, 2012, 2014; cf. **Figure 1**). The "quality" of both the postnatal visual experience and that of the eye movements play a major role in this, in particular during the so called "critical period" (e.g., Hubel and Wiesel, 1970; Buisseret, 1995 for review). At the end of all these processes, if they have occurred properly, an optimal visual perception in terms of acuity, color vision, perception of contrasts and binocular vision (which ensures 3D perception) is acquired progressively with age; cf. **Figure 1**.

Any alteration of postnatal visual experience and/or of eye movements during the critical period (which corresponds to the period of maximum plasticity) leads to the abnormal development of various structures in the brain, both anatomically and functionally. Strabismus is among these alterations. It has been identified for centuries and is characterized by a misalignment of the eyes. It presently affects approximately 2% of the human population worldwide. When occurring early in life, strabismus induces, for example, an abnormal development of both the geniculo-cortical pathway and interhemispheric connections through the corpus callosum (CC; e.g., Innocenti and Frost, 1979; Schmidt et al., 1997; Löwel et al., 1998; Bui Quoc et al., 2012). In parallel, neurons and neuronal maps in V1, as well as those in visual areas from the dorsal and the ventral streams, develop abnormal functional properties (e.g., Chino et al., 1983; Milleret and Houzel, 2001; Schmidt et al., 2004; Bui Quoc et al., 2012; see also Von Noorden, 1978; Milleret, 1994; Wong, 2012 for reviews). Importantly here, the binocular activation of visual

cortical neurons (in V1 at least) is altered because of strabismus. Normally, these neurons receive progressively excitatory inputs from first the contralateral and then the ipsilateral eye during postnatal development (e.g., Frégnac and Imbert, 1978; Milleret et al., 1994). The neurons then potentiate each other to ensure binocular vision. Instead, after strabismus, neurons still sustain excitatory inputs from both eyes but the fixating eye neutralizes the neural response from the deviated eye through inhibition (e.g., Hubel and Wiesel, 1965: Singer et al., 1979; Chino et al., 1994; Sengpiel et al., 1994; Scholl et al., 2013). Binocularity is thus greatly disrupted. Although this phenomenon has been investigated less, neural bases for eye movements may also be abnormal, from the oculomotor muscles themselves to cortex. Altogether, this often leads to amblyopia and a loss of binocular vision (e.g., Sireteanu, 2000; Barrett et al., 2004; Birch, 2013 for reviews).

The question of the consequences of strabismus on both the neural bases of visual perception and visual perception itself has been widely investigated for decades and is still presently under investigation (e.g., Von Noorden, 1978; Milleret, 1994; Wong, 2012 for reviews). In contrast, the important question of the origins of strabismus remains poorly understood. In some cases at least, this may be extended to absence of binocular vision. Focalizing onto strabismus, it may display characteristics which vary greatly from one subject to the next. For example, it may be convergent or divergent. It may be horizontal or vertical. It may be intermittent or not. It may vary in amplitude. The age of its onset also may vary greatly (e.g., Donahue, 2007). We propose here that such "polymorphy" can only result from a multiplicity in the possible origins for strabismus. But, if so, what are these origins? Or, at least, what might they be? Our aim here is to attempt to advance the ideas surrounding that question. It is hoped that to answer such question will help in the improvement of the treatment of these pathologies in the future.

To our knowledge at least, the question of the origin of strabismus was first approached "scientifically" in the nineteenth century. Several theories were suggested. For example, Von Graefe (1854) insisted on mechanistic factors creating strabismus. Donders (1863) pointed out that refraction errors may have a role in the origin of strabismus through their links with accommodation. Duane (1869) proposed that it was an excess in vergence innervation that led to strabismus. Worth (1915) suggested that it was the absence of fusion of the images of both eyes that created strabismus, and that a "center of fusion" in the brain was implicated in this. Chavasse (1939) explained strabismus as a consequence of an excess in reflexogenic action. However, none of these mechanisms has even been proven. The same holds true regarding loss of binocular vision which may be either the consequence of strabismus or its cause. The absence of alignment itself prevents the development of normal binocular vision whereas, without binocular vision, the alignment of the eyes becomes unnecessary.

Nowadays, hypotheses on the etiology of strabismus have evolved and two major theories have thus emerged: a "sensory vs. motor" theory and a "peripheral vs. central" theory. The former theory proposes that strabismus may have a "sensory" or a "motor" origin, while the latter theory rather suggests that strabismus may have a "peripheral" or a "central" origin. The various forms of strabismus are therefore classified depending on those "sensory vs. motor" or "peripheral vs. central" oppositions.

The "Classification of Eye Movement Abnormalities and Strabismus" (CEMAS; http://www*.*nei*.*nih*.*gov/news/statements/ cemas*.*pdf) is mainly based on the "sensory-motor" opposition. It directly and clearly differentiates strabismus from "other" eye movement abnormalities that would be not considered as "pure" strabismus. In the CEMAS, first, eye movement abnormalities or strabismus are defined as resulting from an abnormal motor system, thus having: (1) abnormal full versions and ductions, abnormal fusional vergence amplitudes; (2) non-accurate and abnormal speed saccades, abnormal gain pursuit and vestibular movements; (3) pathologic oscillations or intrusions. Second, a binocular sensory system is defined as abnormal if there is no bi-fixation with normal visual acuity in each eye, strabismus, diplopia, abnormal retinal correspondence, abnormal fusional vergence amplitudes, and abnormal stereopsis. It is also subnormal if there is one or more of the following characteristics: anomalous retinal correspondence, suppression, deficient to no stereopsis, amblyopia, and decreased fusional vergence amplitudes. Finally, the system is also considered as abnormal if there is no binocular vision. The complete classification of CEMAS is then described after those statements on motor and sensory aspects of the visual system. We shall not detail it here but only recall the following divisions: (1) horizontal heterotropias, either concomitant or non-concomitant, either divergent or convergent. In this section, the early onset esotropia, the nerve palsies, the accommodative esotropia and the constant or intermittent exotropias are also included; (2) horizontal heterophorias; (3) cyclovertical heterotropias and special forms of strabismus, i.e., oblique muscles palsy or dysfunction, restrictive strabismus and neuro-myogenic strabismus. In this latter case, myasthenia gravis, chronic progressive external ophthalmoplegia, internuclear ophthalmoplegia, and skew deviation are also classified. Special forms of strabismus are also mentioned in this section, such as co-contractive retraction syndromes, Restrictive Hypotropia in Adduction (RHA) and Congenital Fibrosis of the EOMs (CFEOM); (4) cyclovertical heterophorias; (5) accommodative disorders (Paralysis, Infacility, Insufficiency, Excess); (6) nystagmus and other ocular motor oscillations.

Contrasting with the former classification, the other way to classify rationally the different types of strabismus consists of segregating strabismus according to its "peripheral" vs. "central" origins. This is the classification which is presented in a recent synthesis achieved by Péchereau (2013). In his book, the author classified strabismus with a "peripheral" origin as those resulting from abnormalities at the level of the oculomotor muscles themselves or their innervation. These include, for example, the muscular dystrophies and the palsy of the 3rd, 4th, or 6th cranial nerves. Also included are the retraction syndromes (now called CCDD disorders for "Congenital Cranial Dysinnervation Disorders"), the Basedow disease (called "Graves disease" in the United States), and, finally, oculomotor abnormalities secondary to orbital fractures. In comparison, strabismus considered as having a "central" origin have also been classified. But the precise origins of those strabismus were not specified since that question has never been approached in the literature (at least from our knowledge). Among strabismus with a central origin, the author has included different types of strabismus depending on: (a) the type of deviation (vertical or horizontal strabismus, convergent or divergent, with or without eye cyclo-torsion); (b) the age of occurrence of strabismus (early onset until 6 or 8 months, late onset after 2.5 years, intermediate); (c) whether the deviation is constant or intermittent. He has additionally pointed out the importance of the analysis of binocular status when classifying strabismus, which depends on normal (or potentially normal) or abnormal binocularity. As indicated above, strabismus is most often associated with abnormal binocularity. This occurs in case of: (a) early onset constant divergent or convergent strabismus; (b) micro-strabismus caused by a hereditary absence of fusion; (c) evolution of a micro-strabismus in strabismus; (d) secondary strabismus (i.e., caused by an anatomical abnormality that decreases the vision of an eye). However, binocularity may be present in spite of strabismus in the following cases: (a) intermittent early onset strabismus; (b) late onset strabismus, convergent or divergent, intermittent or permanent; (c) accommodative strabismus, with or without excess or convergence; (d) latent strabismus (heterophorias). The author has also pointed out that oculomotor abnormalities may exist with or without deviation of the eyes (i.e., without strabismus), such as in the case of nystagmus without strabismus, or in the case of torticollis.

The "sensory vs. motor" and the "peripheral vs. central" theories in fact complement each other. This will be obvious from our further discussions below. Nevertheless, the precise "*primum movens*" of strabismus remains vague in both theories, and the very mechanisms and pathophysiology are rarely expressed. It is our main goal to open here the "black box" dealing with the origins of strabismus, in particular its central origins, and therefore we shall use the latter classification in the present article. In that regard, we shall tentatively take into account the most relevant present knowledge about the organization of the brain and its development. This knowledge will come from both biology and medicine. Some knowledge from anatomy, physiology, as well as genetics and molecular biology, will be thus considered. "Innate" and "acquired" factors which may potentially lead to strabismus and/or the absence of binocular vision will be also examined, in addition to peripheral vs. central factors and sensory vs. motor factors. Tychsen has stated already that it is the brain that must be repaired if ophthalmologists want to treat strabismus (Tychsen, 2005). We evidently agree with that idea but much still remains to be done prior to the complete treatment of strabismus. To treat consequences of strabismus on visual perception is already relatively effective, with conventional treatments including optical treatment with glasses, monocular occlusion, and alignment of the eyes through surgery. Newly developed strategies such as binocular training and transcranial magnetic stimulations (TMS) could improve in the future the efficacy of conventional strabismus treatment since it has been shown that such strategies permit the recovery of visual acuity and binocular vision in amblyopia, even at adulthood (after alignment of the eyes); (e.g., Nyffeler et al., 2006; Hess et al., 2010a,b; Hess and Thompson, 2013). Furthermore, considering strabismus at source and dealing in particular with its central origins is currently far from effective. The same applies to loss of binocular vision with a central origin. However, as a general rule in medicine, it is always better to treat pathology at source (provided its origin and pathophysiology are precisely defined) rather than dealing with its dilatory consequences. Our article aims at assisting in this regard by treating the question of the origins of strabismus, even if practical therapeutical consequences will not be immediate.

#### **POSSIBLE ORIGINS OF STRABISMUS: FROM THE EYES TO THE BRAIN**

From the eyes to the brain, the visual system and the oculomotor system are both formed of complex neural networks which link numerous structures. Interactions exist between these structures within each system separately, as well as between both systems. Thus, all of these structures must function perfectly and in synchrony to ensure a normal visual function, i.e., the best possible acuity of each eye, a proper binocular (stereoscopic) vision, a normal alignment of the eyes and precisely shaped movements.

In the visual system, as illustrated in **Figure 2**, the retinogeniculo-cortical pathway is the main sensory route that links the retina to V1: most ganglion cells of the retina project to the dorsal lateral geniculate nucleus (dLGN) via the optic nerves and the optic tracts. Geniculate cells then project to V1 through the optic radiations. From there, most of the afferents reach "superior" visual areas that form the "dorsal" and the "ventral" streams. Other afferents may interconnect both hemispheres through the CC. Any abnormality within one of these structures alters vision (e.g., Wong, 2012; Berlucchi, 2014 for reviews). Some fibers from the optic tracts also project to extra-geniculate structures which are themselves implicated in vision, such as the Superior Colliculus (SC). This structure, among other functions, is also responsible for a precise ballistic of the eye movements and for visual attention (Krauzlis et al., 2013).

The oculomotor system must also function perfectly to provide eye movements and alignment of the eyes with optimal characteristics. This begins with the six EOMs and their pulleys, which allow movements of the eyes: the external and internal rectus, the superior and inferior rectus, as well as the inferior and superior obliques. The motor activity of these muscles is controlled directly by the IIIrd (common oculomotor), IVth (trochlear), and the VIth (abducens) cranial nerves. During these eyes movements, some receptors in the EOMs (including their tendons) are also activated, sending sensory messages related directly to eye movements all around the brain, including within V1 and the frontal cortex (Buisseret and Maffei, 1977 for visual cortex; cf. Buisseret, 1995 for review) through the ophthalmic branch of the Vth (trigeminal) cranial nerve (e.g., Batini et al., 1975). Such proprioceptive information is relayed in different sub-cortical structures located in the brainstem, and also in the cerebellum and the vestibular nuclei, finally reaching the parietal and the frontal cortex (e.g., Fillenz, 1955; Batini and Buisseret, 1974; Donaldson and Dixon, 1980; Donaldson and Long, 1980; Ashton et al., 1988, 1989).

Our hypothesis here is that strabismus can be caused by an insult at every level of both the visual and the oculomotor systems. Below we shall thus describe different possible origins of strabismus at different levels of each system. Notice that many other possible origins might be proposed. The EOMs and their innervations (up to their first relay in the brain) will be considered at the "periphery," while the eyes and the structures in the brain will be considered as "central." Considering their embryonic origin, the eyes may indeed be considered as a prolongation of the brain.

## **PERIPHERAL AND GENETIC ORIGINS OF STRABISMUS**

As a first step, abnormalities at the level of the EOMs, whether genetic in origin or not, will be discussed as potential origins of strabismus. The potential role of the oculomotor muscles themselves in this process will also be discussed, as well as their sensory afferents or motor efferents.

#### **ABNORMAL WEAKNESS OF EOMs**

#### *Muscular dystrophies, genetic myopathies, and myasthenia*

Abnormal weakness of extraocular (or oculomotor) muscles may occur for various reasons. First, it may be due to muscular dystrophies or other genetic myopathies. In that case, the muscles themselves are affected, which can cause strabismus (Shieh, 2013). The most common forms of muscular dystrophies include Duchenne muscular dystrophy, Becker muscular dystrophy, facio-scapulohumeral muscular dystrophy and Steinert myotonic dystrophy. Variable phenotypes of strabismus and abnormalities of ocular movements are often associated with such muscular dystrophies. Second, it may be the transmission of neurotransmitters at the level of the neuromuscular junction that is affected in myasthenia which may lead to variable and intermittent forms of strabismus. Moreover, EOMs, being the most fatigue-resistant muscles of the body, can therefore be less or lately affected by a muscular dysfunction.

**FIGURE 2 | Potential origins of strabismus.** Strabismus may have sensory and/or motor origins as well as peripheral and/or central origins. In periphery, one may notice, for example, abnormal vision or abnormal development of the extraocular muscles. In the latter, the extraocular muscles' proprioceptive afferents reaching the Gasser ganglion normally through the ophthalmic branch of the Vth (trigeminal) cranial nerve and/or their oculomotor nerves, i.e., the IIIrd, IVth, and the VIth cranial nerves, may be also altered. Centrally, strabismus may, for example, result from an

#### *Strabismus and pulleys*

The understanding of eye movements have been deeply modified by Demer's work, which described the role of the pulleys in the kinematics of EOMs (e.g., Demer, 2003; Demer et al., 2008). Abnormal position or function of those pulleys may lead to strabismus.

#### **ABNORMAL DEVELOPMENT OF THE INNERVATION OF EOMs IN VARIOUS FORMS OF COMPLEX INCOMITANT STRABISMUS** *Abnormal motor innervation of the EOMs*

Congenital fibrosis of the EOMs (CFEOM) leads to a first form of strabismus due to abnormal motor innervation. This affects patients with congenital restrictive ophthalmoplegia (Engle, 2006; Graeber et al., 2013). CFEOM is a misnomer for various incomitant strabismus which were described in several families, indicating a possible hereditary trait. Indeed, genetics have proven that CFEOM result from mutations in genes that are implicated in the growth of oculomotor nerves. In the absence of normal innervation, a variable atrophy of the muscles may occur. Several forms of CFEOM exist. CFEOM 1 results from KIF21A mutations and is abnormal activity in the brainstem, the Medial Reticular Formation (MRF), the Pontine Reticular Formation (PRF), the thalamus, the cerebellum or the superior colliculus. At the cortical level, visual, parietal or frontal cortex may also not function properly. Altogether, this indicates that origins of strabismus may be numerous. One may emphasize that those with a central origin likely dominate. FEF, Frontal Eye Field; SEF, Supplementary Eye Field; PEF, Parietal Eye Field; III, IV, V, VI: brainstem nuclei III, IV, V, and VI and their motoneurons.

characterized by a quasi total absence of movement of the eyes, the primary position being hypotropic with elevation being impossible, whereas the horizontal position can be either esotropic or exotropic. The heterozygous mutation of the gene is sufficient to cause the disease which results in a dominant inheritance. KIF21A is located on chromosome 12 and is responsible for the production of a developmental kinesin. Kinesins are molecular motors that interact with and transport cargo along the microtubules of axons. In CFEOM 1, abnormal development of the superior branch of the IIIrd nerve particularly affects the function of the superior rectus muscle (and the levator palpebrae superioris muscle). In CFEOM 2, the phenotype results from an insult to all branches of the IIIrd and IVth nerves. Patients with CFEOM 2 have their eye in the exotropia position with no movement possible. It is mutations in gene PHOX2A, located in 11q13 that cause the disease. PHOX2A is a paired-linked transcription factor gene and its expression is restricted to differentiating neurons in the central and peripheral nervous systems. Therefore, the pathogenesis of CFEOM 2 would be an abnormal development of both the IVth and the IIIrd nerves. In CFEOM 3, similar to CFEOM 1, it is only the development of the IIIrd nerve that is affected, although in a more severe way since all of its branches are affected in this situation and not only those branches which innervate the superior rectus muscle and the levator palpebrae superioris muscle. The gene responsible for the disease is located at 16qter.

Contrarily to CFEOM syndromes, in which development of the IIIrd nerve is affected, it is the development of the VIth nerve which is affected in the Duane syndromes. The *primum movens* of these latter syndromes is an absence of development of the VIth nerve, either unilaterally or bilaterally. Variable forms of abnormal innervation of the lateral rectus muscle by branches of the IIIrd nerve have also been observed, resulting either in esotropia or exotropia. Again, it is an abnormal development of a nerve that underlies the pathogenesis of this condition, with genes located at 8q13 and at 2q31. Duane Radial Ray Syndrome (DRRS) is a particular form of Duane Syndrome in which the ocular movement abnormality is associated with bone abnormalities in the hand, such as an absence or a malformation of the thumb which can look like a finger. Again, it is a mutation in a transcription factor gene SALL4, located at 20q13, which causes the disease by altering the normal neural development.

Other insults to the development of the VIth nerve include HGPPS (Horizontal Gaze Palsy with Progressive Scoliosis), in which it is mutations of the gene ROBO3, located at 11q23, which are responsible for the resulting phenotype. Such phenotype combines a total impossibility of horizontal gaze movements, along with a scoliosis that occurs during the first decade of life. ROBO3 is a developmental gene and is expressed in the hindbrain of the human fetus. Human ROBO3 is similar to roundabout genes that are responsible for axon guidance in other species such as mice, zebrafish or drosophile. Indeed, brainstem neurons of ROBO3−*/*<sup>−</sup> mice completely fail in crossing the midline during development (Marillat et al., 2004). In zebrafish, and in drosophile, the loss of function of ROBO3 results in aberrant midline crossing by axons (Seeger et al., 1993).

Finally, abnormal development of the VIth nerve occurs in two conditions in which mutations of the gene HOXA1 (located at 7p15 and implicated in hindbrain segmentation during fetal development) are responsible: the Bosley-Salih-Alorainy syndrome (BSAS) and the Athabascan Brainstem Dysgenesis syndrome (ABDS). In BSAS, which is a recessive condition, a bilateral Duane trait is associated with other cranial nerve dysfunctions, such as deafness due to a bilateral absence of the cochlea and misdevelopment of the VIIIth (vestibulo-cochlear) nerve. In ABDS, horizontal gaze restriction is associated with central deafness and mental retardation.

The various abnormalities of the development of the IIIrd, IVth, or VIth cranial nerves have been regrouped in a generic term: CCDD (see above). In CCDDs, the insult can cause ocular movement disorders but other conditions without strabismus can occur, such as isolated congenital ptosis which can result from a mutation in a gene located at 1p34-p32. As a general rule, genetics of CCDDs teach us that an abnormal development in general and an abnormal early routing of neurons in particular, may cause strabismus. In CCDDs, the insult occurs at the frontier between peripheral and central locations. It is the same when considering the cranial nerves that emerge from the brainstem and lead toward abnormal EOMs. The model of the CCDDs therefore emphasizes that an abnormal neural network can cause strabismus.

Palsy of the IIIrd, IVth, or VIth nerve leads to a second form of strabismus. Similar to CCDDs, it is a neural disorder although the cause is often acquired and not innate. The nerve palsy causes an atrophy of the innervated muscle. Acute palsy of the IIIrd nerve is an emergency since it can be caused by a direct compression of the nerve by a cavernous sinus thrombophlebitis, or by an aneurysm of the posterior communicant artery. A palsy of the IVth nerve is either congenital or acquired. When acquired, it is a peripheral cause that is the *primum movens.* It may however result from a direct insult after a severe cranial traumatism. A palsy of the VIth nerve may finally result from a direct compression of the nerve's fine and fragile branches. This can arise from a hypertrophic brainstem due to intracranial hypertension. It may also result from numerous other central causes such as tumors, infections, etc. In all cases however, this will induce strabismus.

#### *Strabismus and extraocular proprioception*

The outflow theory supports the idea that it is an efferent copy of the oculomotor signal from the motor centers that gives information about the position of the eyes to the brain (Von Helmholtz, 1866). By contrast, the inflow theory claims that it is the proprioceptive signals from eye muscle receptors that give such information (Sherrington, 1918). More recent experiments support one theory or the other. Thus, authors now consider that both theories are right and that efferent copy co-exists with extraocular proprioception. Proprioceptive receptors do exist in the EOMs, in particular at the level of the tendons (e.g., Cooper and Daniel, 1949; Richmond et al., 1984 for receptors in humans). These receptors are active and send sensory messages to numerous regions in the brain implicated both in visual perception and eye movements (e.g., Donaldson, 1979; Donaldson and Dixon, 1980; Milleret et al., 1987). Furthermore, they have been demonstrated by our group to strongly contribute to the maturation of visual neurons in V1 during development (cf. Buisseret, 1995 for review; see also Buisseret et al., 1978, 1988). Thus, an abnormal proprioception at the level of the EOMs is also a potential cause for strabismus, since abnormal information about the position of the eyes leads to an abnormal central and neural motor command in return.

Nevertheless, a deafferentation of EOMs has never been demonstrated to affect ocular motor control and to induce strabismus. To our knowledge at least, neither experimental research nor any medical cases have demonstrated this (the problem is a difficult one to approach). Thus, some authors have concluded that proprioceptive signals only play a role during development in calibrating the efferent copy signal, which is sufficient to provide information about eye movements and position (Lewis et al., 2001). Other authors, however, have claimed that an insult to proprioceptive receptors of the EOMs could be the cause of strabismus (e.g., Donaldson, 2000). Similarly to the CCDD (see above), abnormal development of the proprioceptive axons within the Vth cranial nerve to the Gasser ganglion may occur. It could also be hypothesized that abnormalities of the extraocular receptors could be responsible for strabismus. This is supported by investigations of some strabismic patients whose extraocular muscle receptors display abnormal morphological characteristics (Li and Shen, 2001). These changes were analyzed using transmission electron microscopy and revealed both a decrease in the number of mitochondria in axons, and the disappearance of the nerve component of the receptor. Of course, in such a study, whether the abnormalities in the proprioceptors are the cause of strabismus or its consequences cannot be distinguished. This recalls the controversy regarding whether the subtle changes at the cellular level of the muscles (especially the singly innervated orbital fibers) of strabismic patients can be the *primum movens* of strabismus, or are simply an adaptative phenomenon to the deviation (Lennerstrand, 2007). Finally, the implication of extraocular proprioception during ocular movement disorders can be emphasized by the fact that a tenotomy of all of the EOMs and their reattachment, which suppresses the proprioceptive output signals, is an effective therapy in the treatment of some forms of infantile nystagmus (Dell'Osso and Wang, 2008).

#### **GENETICS OF CONCOMITANT STRABISMUS**

Contrary to CCDDs or to nerve palsies, most strabismus such as congenital strabismus, accommodative strabismus or divergent strabismus are concomitant, meaning that the deviation is always the same whatever the gaze direction; they also likely display a central origin (see below). Also contrary to CCDDs, no single gene has been identified as the direct origin of concomitant strabismus. Nevertheless, inheritance and genetics are obvious in the development of most forms of strabismus, either incomitant or concomitant (Engle, 2007). However, additional factors, in particular those related to developmental processes, also need to be taken into account.

Hippocrates himself would have stated that: "squinters engender squinters." Physicians are evidently also aware of the influence of genetics on strabismus and usually advise strabismic parents that their children must be screened for strabismus. Indeed approximately 15% of children of strabismic parents are strabismic, compared to the 2% prevalence of strabismus in the general population (Donnelly, 2012). Ziakas showed however that this proportion may vary depending on the type of strabismus and the degree of relationship (Ziakas et al., 2002). In his study of 96 index cases with strabismus with either early onset strabismus (26 cases), accommodative esotropia (49 cases), anisometropic esotropia (15 cases), or exotropia (56 cases), he showed that the risk of having strabismus for a first degree relative is 4% for exotropia but 26.1% for accommodative esotropia. In accommodative esotropia, the risk decreases to 7.5% for second degree relatives and to 4.8% for third degree relatives. In twin studies, it has been shown that there is a specific genetic influence for esodeviation which is independent of the refractive error (Sanfilippo et al., 2012). The heritability of eso-deviation is estimated as 64% in a cohort of 1462 twin pairs with a prevalence of 8.6% of eso-deviation, the correlation being significantly greater in monozygotic twins (*r* = 0*.*65) than in dizygotic twins (*r* = 0*.*33). But, as indicated above, the genetic contribution to concomitant strabismus is not easy to circumscribe. Thus, even in the case of "simple" strabismus such as early onset esotropia, accommodative esotropia or exotropia, genetic inheritance is complex, with the possible implication of recessive genes as well as dominant genes.

The influence of the degree of development of both the eyes and the brain at birth must also be taken into account. Recalling the higher proportion of strabismus in premature infants compared to full term infants illustrates this (Torp-Pedersen et al., 2010). This emphasizes the relationship between the development of the brain (including the eyes) and the potential development of strabismus after birth. More generally, this indicates that an abnormal (or an immature) development of the neural networks, resulting from innate (i.e., genetic) or acquired factors, might, in turn, lead to strabismus.

#### **POSSIBLE CENTRAL ORIGINS OF STRABISMUS**

As indicated above, it is the anatomo-functional maturation before and after birth of multiple neural networks from the eyes to the brain that subtend the normal development of visual perception. This occurs by implicating both genetic and epigenetic factors such as postnatal visual experience. Our driving hypothesis here is that any insult to this normal process of maturation may, in turn, generate strabismus. This applies evidently to any level of the sensory and/or motor networks that are involved in the elaboration of visual perception (cf. **Figure 2**). Some examples are provided below to illustrate this. How an abnormal development of any visual path or any neural activity somewhere within the visual system may lead to strabismus are considered in succession. How an abnormal neural activity in the oculomotor system may lead to strabismus is also discussed.

#### **ABNORMAL DEVELOPMENT OF THE VISUAL PATHS**

First, we hypothesize that any insult during the normal processes of neurogenesis, axonal growth, migration of neurons, synaptogenesis, myelination, apoptosis or even elimination of juvenile exuberant axons, may potentially lead to strabismus. For example, strabismus may be the consequence of the misrouting of some paths within the visual and/or the oculomotor networks.

#### *Abnormal routing of ganglion cell axons*

Interestingly, Siamese cats spontaneously display a convergent strabismus. They also have an abnormal predominance of the crossed retino-geniculo-cortical pathway compared to normal cats (Montero and Guillery, 1978; Shatz and Levay, 1979). This results from stagnation at an early stage of development, which itself recalls the development of visual pathways during phylogenesis. We propose that such abnormal predominance of the crossed retino-geniculo-cortical pathway may also be the cause of the early onset convergent strabismus in humans, in which the early asymmetry of the optokinetic nystagmus also persists with age.

Paradoxically, in case of divergent strabismus, it could also be hypothesized that a predominance of the crossed pathways could be the *primum movens* of strabismus. During evolution, the visual system is first an "only crossed fibers" network with lateral eyes and panoramic vision. It then evolves to a balanced system with equal importance between the direct pathway and the crossed pathway, and frontal eyes allowing binocular vision. We propose here that an abnormal routing of the retinal ganglion cell axons at the level of the optic chiasm might lead to a loss of balance between crossed and direct fibers and thus lead to strabismus. This might occur by an abnormal expression of ephrins, i.e., surface molecules which are specifically implicated in guiding the retinal ganglion cell axons at the level of the optic chiasm during the developmental process (Petros et al., 2009). The axons of ipsilateral projections from temporal retina (direct fibers) express the guidance receptor ephrin B1 (but not the axons of contralateral projections from the nasal retina, i.e., crossed fibers). At the optic chiasm, radial glia cells express ephrin B2, which repulses the ephrin B1 axons from crossing the midline, unlike the contralateral fibers from the nasal retina. Expressions of ephrins and of ephrin receptors are specific and precise timing is necessary to ensure the normal and balanced development of visual pathways. If this system was altered through an abnormal expression of ephrins and/or ephrin receptors during development, an asymmetrical neural network of crossed and uncrossed fibers would develop and could result in the development of strabismus.

#### *Misrouting and abnormal retinotopy*

During development, ganglion cell axons reach progressively central visual structures by respecting "retinotopy." The visual field is thus encoded by neurons with precision from the retina up to the cortex (e.g., Tootell et al., 1998 for review), a necessary condition to ensure normal visual perception, including binocular visual perception. Guidance of axons creating retinotopy is also permitted by ephrins. Gradients of ephrins A and ephrins B, both in the retina and in the visual cortex, allow the creation of x and y coordinates (Cang et al., 2005a,b). This leads to the establishment of neuronal "retinotopic maps," which are refined with age and visual experience. Again we propose here that abnormal guidance through abnormal levels of ephrins A or B and/or their receptors during development would alter retinotopy and would cause, in turn, strabismus. In some cases at least, distortions within retinotopic maps, which may lead to abnormal retinal correspondence in early onset strabismus (e.g., Wong, 2000; Popple and Levi, 2005; Mansouri et al., 2009; Wang et al., 2012), would therefore be a cause of strabismus rather than a consequence.

The subplate, i.e., mostly temporary cells located below layer VI of Area V1, plays a major role for growing axons to reach the visual cortical plate during development (Ghosh et al., 1980; McConnell et al., 1989). Any abnormality during such process would interfere with normal development of geniculo-cortical connections, thus with normal development of retinotopic maps in V1. Again, this could potentially induce strabismus.

#### *Abnormal cortico-cortical connections*

Strabismus is now well known to disrupt the development of numerous cortico-cortical connections implicated in visual perception. These cortico-cortical connections may be "short" and located within one given area or "long," thus linking various areas which may be located very far from one to the other. Thus, for example, strabismus is known to stabilize normally transient intra-hemispheric cortico-cortical connections in V1, leading to interconnect larger cell groups driven through the same eye than in the normal case (e.g., Löwel and Singer, 1992; Schmidt and Löwel, 2008). It is also known to lead to drastic anatomo-functional changes in the organization of the interhemispheric callosal connections, which normally link reciprocally and homotopically various visual areas to "glue" both visual hemifields into a single scene (Payne, 1990, 1991; Payne and Siwek, 1991a,b; Bui Quoc et al., 2012). In particular, in the case of strabismus, it leads to the development of asymmetrical interhemispheric connections which prevent the fusion of both visual hemifields along the vertical midline (Lund and Mitchell, 1979; Milleret and Houzel, 2001; Bui Quoc et al., 2012).

Our idea here is that abnormal anatomical cortico-cortical connections within or between visual cortical areas (whatever their origin) may conversely lead in turn to strabismus. Our hypothesis may be supported first by experiments which have consisted in cutting the CC of adult cats, who rapidly displayed a misalignment of their eyes and even strabismus (Elberger, 1979; Payne et al., 1981; Elberger and Hirsch, 1982). This is further supported by studies showing the implication of the CC during eye movements (e.g., Pasik et al., 1971; Tusa and Ungerleider, 1988; Zernicki et al., 1997). Our hypothesis is also strengthened by analyzing the deficits in visual and visuo-spatial developments that are present in young children with Williams syndrome. They are interpreted as the result of a split between the ventral and dorsal stream processing of visual information (see **Figure 2**), with a generalized deficit in dorsal stream processing (Atkinson et al., 2001). Of great interest here, the authors underlined that patients with such syndrome also display a much higher incidence of strabismus, visual acuity loss, amblyopia and reduced stereopsis than the general population.

#### **ABNORMAL DEVELOPMENT OF NEURONAL ACTIVITY**

In addition to genes, axonal guidance cues and molecules, spontaneous and early visually evoked neural activity are necessary for anatomical and functional refinement of developing visual circuits (e.g., Huberman et al., 2008 for review). Appropriate synchronizations within the visual network then need to develop in order to elaborate visual perception optimally (e.g., Singer, 1999, 2013; Uhlhaas et al., 2009a,b; Menon, 2013). Any abnormality in such neural activities from the retina to the visual cortex may also lead to strabismus. Some data in the literature strengthens this idea already. The same applies to neural circuits subtending eye movements.

#### *Effects of an abnormal neuronal activity on visual system*

Let us evaluate, in succession, the potential impacts on the alignment of the eyes of: (a) abnormal prenatal retinal waves; (b) abnormalities during postnatal visual experience; (c) abnormal excitation/inhibition balance; and (d) pathological asynchrony of neural activity.

(a) **Abnormal retinal waves.** First, prenatal spontaneous neural activity in retina, discovered by Galli and Maffei (1988), must be absolutely normal to allow the visual system to organize with precision. It plays both permissive and instructive roles. Indeed, even if this activity is generated very early in life, before vision begins, it is a necessity for the proper development of functional properties of visual neurons and that of the various functional maps all along the visual system. The retinotopic organization of the retino-geniculo-cortical projections is affected first. Indeed, retinotopic maps in the SC, dLGN, and V1 all develop before photoreceptors can be driven by light. The same applies to eye-specific inputs to dLGN and ocular dominance columns in V1. Orientationselective circuits in V1 also start to form before visual experience begins. The same applies to circuits encoding spatial frequency (Tani et al., 2012). This is possible because spontaneous neural activity in the retina is highly structured, and thus allows the transmission of very precise messages to the central nervous system. This is achieved through slow wave oscillations with very specific spatial and temporal characteristics (Rochefort et al., 2009). If retinal waves display abnormality, for any reason, this entire process of development will be disrupted. The segregation of inputs from both eyes and/or the development of retinotopic maps will be abnormal (e.g., Cang et al., 2005a,b; Xu et al., 2011; Ackman et al., 2012; see also Huberman et al., 2008 for review; **Figure 1**). The orientation and/or the spatial frequency maps might also develop incorrectly. In other words, prenatal neural bases for binocular integration and/or for acuity would be altered centrally. It must also be taken into account that such alterations may in turn lead to misalignment of the eyes. For example, this may occur through incongruent interactions with the oculomotor system. As discussed below, this may also occur during development of visual perception itself. Even if it is difficult to prove, such a possibility might unavoidably correspond to the etiology of some forms of strabismus at least.


also been shown recently that it is the transformation of parvalbumin GABAergic (PV) interneurons from excitatory neurons to inhibitory ones that opens the critical period of visual development by internalizing the homeoprotein Otx2 (Sugiyama et al., 2008; Beurdeley et al., 2012). Reducing intraocular inhibition in the adult visual cortex has also been demonstrated to promote plasticity (e.g., Harauzov et al., 2010). In short, a balance between excitatory and inhibitory inputs from retina to cortex is required for elaborating correctly visual perception. Abnormality in this balance, either before or after birth, might lead to abnormal vision and/or uncorrelated eye movements.

(d) **Abnormal synchronization of neural activity.** The oscillatory pattern of neuronal responses and the synchronization of the oscillations from retina to cortex are now considered as playing a major role in elaborating visual perception (e.g., Gray et al., 1989; Engel et al., 1991; Neuenschwander and Singer, 1996; Castelo-Branco et al., 1998; Fries et al., 2002; see also Singer and Gray, 1995; Singer, 1999, 2013; Engel et al., 2001 for reviews). We propose here that any abnormality of this synchronization, at any level, may lead to strabismus (as well as binocular vision loss and/or amblyopia).

Visual cortex is a highly distributed system implicating more than forty areas distributed from the occipital lobe to the parietal and temporal lobes (**Figure 2**). To elaborate visual perception, these areas operate in parallel and interact with one another to complement each other. This is achieved through short and long cortico-cortical connections which allow synchronizing of oscillatory neuronal responses within each area and between different areas, mainly in the β and γ frequency range i.e., 20–100 Hz (Engel et al., 2001; Fries, 2005). Among other functions, this is considered as solving the "binding" problem which consists in assembling all the attributes of the visual scene (namely location in space, direction of movement, orientation, spatial frequency, disparity etc.) into a coherent form during visual perception in various contexts, attention states etc. (e.g., Singer, 1999, 2013; Fries, 2005 for reviews). Very recently, this has also been established to allow prediction of perception (Hipp et al., 2011).

Such synchronization develops with age, at least up to adolescence, in parallel to the maturation of cortico-cortical connections (including their myelinization) as well as excitatory and inhibitory circuits (Uhlhaas et al., 2009a,b). The maturation of the inhibitory PV neurons again plays a major role in this process since they serve as "pacemakers" for rhythmic neuronal activity, in particular in the γ frequency range (30–100 Hz). In other words, they assume a pivotal role in the temporal structuring and coordination of neuronal responses (Cardin et al., 2009; Sohal et al., 2009). Of interest, without going into details, all this developmental process of the brain rhythms occurs under strong genetic control (e.g., Buzsáki et al., 2013 for review). Simultaneously, visual perception also increases. Thus, for example, Csibra et al. (2000) measured γ band responses in EEG data in 6 and 8-month old infants during the perception of Kanisza squares that require the binding of contour elements into a coherent object representation. Based on prior behavioral studies that showed that infants up to 6 months of age are unable to perceive Kanisza figures, the authors hypothesized that perceptual binding in 8-month-old infants is related to the emergence of the γ band oscillations.

Not surprisingly, epigenetic factors also play a role in this. Thus, any abnormal postnatal visual experience such as the one resulting from strabismus modifies the normal development of synchronization within the visual system by altering both wiring and neural activity (e.g., Löwel and Singer, 1992; Schmidt and Löwel, 2008). Neuronal synchrony is reduced in visual cortex compared to normal (Roelfsema et al., 1994). Recent data from Hess and his group have strengthened this by establishing that interactions between cells in disparate brain regions are reduced when driven by the amblyopic eye of strabismic subjects, from dLGN to superior visual areas, via V1 (Li et al., 2011). They have also demonstrated that amblyopia (in strabismic patients) is associated to temporal synchrony deficits (Huang et al., 2012).

In turn, we postulate that any abnormality within one given visual area or between at least two visual areas, due to developmental anatomical and/or functional abnormalities somewhere in the visual system, may lead to strabismus (and amblyopia and/or binocular vision loss) by altering synchrony. Since abnormalities in synchrony may occur before or after birth (see above), this may lead to an early or a late strabismus. The same idea may be extended to the oculomotor system since it has been shown recently that it has its own dynamics (Gregoriou et al., 2012; Cordones et al., 2013) and that changes in neural synchrony also occur during development of the motor system (Kilner et al., 2000). Situations when the visual and the oculomotor systems need to interact to elaborate visual perception, i.e., during sensorimotor processing, are also affected. One may underline that our hypothesis is directly in line with increasing evidence that disturbances of synchrony in the developing brain, associated to aberrant neurodevelopment, subtend the cognitive dysfunctions associated to major brain pathologies such as schizophrenia and autism spectrum disorders (Uhlhaas and Singer, 2006; Uhlhaas et al., 2009a,b, 2011). As outlined above, genetics play a major role in that. Thus, for example, in schizophrenic patients, the GABA synthesizing enzyme GAD 65 and the calcium-binding protein parvalbumin are down-regulated in basket cells, while they are crucial for the generation of γ rhythms (Lewis et al., 2005; see also above).

#### *Effects of abnormal neural activity on oculomotor system*

Neuronal activity within the various structures implicated in the movement of the eyes also needs to be normal whatever the age. As illustrated below, any abnormality may induce strabismus.

(a) **Abnormal extraocular proprioceptive afferents from EOMs to V1.** Proprioceptive afferents from EOMs project to V1 (Buisseret and Maffei, 1977). They strongly contribute to the maturation of visual neurons in V1 during development, including their ability to perceive details. Thus, when removed in their entirety early in life, visual neurons do not develop their functional properties properly. It is as if they had never benefited from any visual experience. Also, if proprioceptive afferents in one plane are removed, a perpendicular meridian amblyopia develops (cf. Buisseret, 1995 for review). Since an amblyopia may induce strabismus (cf. above), we put forward the idea that any disequilibrium in the proprioceptive afferents from the EOMs might also induce strabismus. Note that such a process might be extended to any of the central structures which receive afferents from the EOMs, belonging to both the visual and the oculomotor systems (e.g., Donaldson, 1979; Donaldson and Dixon, 1980; Milleret et al., 1987).

(b) **Abnormal activity of the vergence neurons and abnormal cortical control.** Specific convergence neurons have been identified in the Medial Reticular Formation of the brainstem (Mays, 1984) and abnormal activity, either an excess of activity or a loss of activity, may also be responsible for a deviation of the eyes. Hyperactivity of these neurons could lead to convergent esotropia. Hyperexcitability of the neurons, enhancing the accommodation/convergence loop, may play a role in accommodative esotropia with an excess of convergence. On the other hand, a loss of activity of the neurons, premature apoptosis or a progressive degeneration of the neurons or axons may also induce exophoria and exotropia. Such a progressive insult to the system would explain the natural history of divergent strabismus, in which there is an increase in divergent deviation with time.

Similarly, divergent strabismus may result from an excess of positive inputs from the divergence neurons which have been identified in the Pontine Reticular Formation (cf. also Mays, 1984). It has been hypothesized that a lack of activity of those divergence neurons would induce esotropia.

Higher structures command eye movement and eye position. Our hypothesis here is that the genesis of strabismus may also result from abnormal inputs from those cortical structures which play a role in the triggering of ocular movements such as Frontal Eye Field, Supplementary Eye Field, and Parietal Eye Field (cf. **Figure 2**). Indeed, it has been shown by neuroimaging that, in adult strabismic patients, the gray matter volume of those cortical eye fields can be abnormal, either larger or smaller (Chan et al., 2004).

(c) **Abnormal activity in Superior Colliculus, cerebellum and vestibular pathways.** The SC is a key structure in the control of eye movements. It is another structure that may potentially contribute to inducing strabismus. For example, it has been hypothesized that an insufficiently developed neuronal coupling between both superior colliculi would be implicated in vertical dissociated deviation, which is a particular form of strabismus that is associated with early onset strabismus (Brodsky, 2011; Ten Tusscher, 2011). Cerebellum and vestibular nuclei also control eye movements in normal visual conditions. An insult to those structures could also be hypothesized to be a central cause of strabismus. This is supported by the fact that an insult to the vestibulo-ocular input, through an attempt at the level of the otoliths, can cause this particular vertical strabismus, known as a "skew deviation" (Schlenker et al., 2009). Also, a malformation of the cerebellum, such as the one found in Joubert syndrome or in rhombencephalosynapsis, is associated with strabismus (Canturk et al., 2008; Keskinbora, 2008).

## **CONCLUSION**

For the first time, at least in our knowledge, instead of treating the question of the consequences of strabismus in humans, our article highlights the question of the *origins* of strabismus. The "polymorphy" of strabismus indeed suggests multiple origins but most of them are presently unknown. Those strabismus with a peripheral origin are rather well characterized but they are only few. In contrast, the other forms of strabismus, which are considered as having a *central* origin, are poorly understood despite being the most frequent. At present it is as if these latter forms of strabismus are included in a "black box" that has never really been opened by anyone. To move forward, we have decided here to tentatively open this box. We have proposed mechanisms which show that a central abnormality may lead to the development of strabismus. Our hypothesis on the mechanisms of strabismus are based on both classical and the most recent knowledge about the development and the organization of the visual system in mammals both before and after birth. Research in that field has indeed been very active all around the world for decades and is still very active today, providing numerous "keys" which might open our black box. Some other mechanisms are based on knowledge regarding the oculo-motor system from EOMs to cortex. In that context, we have also revisited the question of the origins of binocular vision loss, with tentative new ideas on the question. Interestingly, whether the origins of strabismus or those of binocular vision loss are considered, some of the new mechanisms we propose are already supported by published data. Altogether, our findings clearly emphasize the necessity to develop and to apply as soon as possible new strategies to treat strabismus and binocular vision loss, in particular through "central" therapies, in addition to the peripheral ones.

#### **ABOUT THE ORIGINS OF STRABISMUS**

Origins of strabismus with a peripheral origin are rather well known but they represent less than 5% of strabismic patients. As discussed above, they display either rare forms of incomitant or concomitant strabismus. As illustrated in **Table 1**, it is possible to distinguish which origins may induce early strabismus (up to 8 PN months) and/or late strabismus (from 24 PN months). Thus, an abnormal weakness of EOMs, such as the ones due to muscular dystrophies, myopathies, myasthenia or abnormal muscular pulleys, may induce both forms of strabismus. In contrast, abnormal development of the innervation of the EOMs, either motor or proprioceptive, may only induce an early strabismus. Not surprisingly, innate nerve palsies and acquired ones may induce early and late strabismus respectively. Of interest, it is also already established that most forms of strabismus with a peripheral origin have a genetic origin. As summarized in **Table 2**, most genes associated to such forms of strabismus have been successfully identified. This has been facilitated by the fact that the alteration of only a few specific genes is generally associated to each specific disease leading to strabismus. Thus, for example, Duchenne muscular dystrophy leading to the weakness of EOMs implicates only the DMD gene located on Xp21.2. The abnormal innervation of the EOMs in the context of HGPPS syndrome only implicates the ROBO3 gene located on 11q23.

By contrast, other forms of strabismus (mostly concomitant) are presently poorly understood while they are the most numerous (*>*90%). About 10% of them are "congenital" strabismus (thus occurring before 8 PN months), while the remaining occur at a later stage. In this latter situation, they are characterized by either an accommodative or non-accommodative esotropia or an exophoria or an exotropia (cf. **Table 1**). They are also known to have a genetic origin (see **Table 2**). However, by contrast to strabismus with an identified (peripheral) origin, those forms of strabismus are likely related to both recessive and dominant genes, thus resulting from a complex genetic inheritance. To move forward, we have proposed different possibilities to justify the emergence of such "uncharacterized" forms of strabismus, by evoking the occurrence of abnormal development of "central" paths and abnormal development of "central" neural activity. This evidently concerns both the visual and the oculomotor systems, up to the cortex, since they are closely related. Indeed, these forms of strabismus, with unknown origins, are generally "supposed" to have a central origin. Of interest, some of the anatomo-functional abnormalities we have proposed as being responsible for strabismus may take place before birth or postnatally, with consequences that the different forms of strabismus with a central origin may have early or late onset. Thus we did not dissociate them in **Table 1**. This may help to justify, however, the occurrence of early and late strabismus. It is not necessary to underline that many other mechanisms could have been proposed as the possibilities are vast. To explain each form of strabismus supposes that a lot of mechanisms might be at source of strabismus with a central origin. Most of the mechanisms we propose are clearly hypothetical and remain to be proven. But, as indicated above, some mechanisms are already supported by precise data. For example, disrupting callosal connections alters the alignment of the eyes (Elberger, 1979; Payne et al., 1981; Elberger and Hirsch, 1982). A split between the ventral and the dorsal streams such as the one occurring in Williams syndrome most often leads to strabismus (Atkinson et al., 2001). An abnormal visual perception from one eye may also lead to strabismus (e.g., Quick et al., 1989; Niechwiej-Szwedo et al., 2010). One may also consider that disturbances of synchrony in the developing brain, associated with aberrant neurodevelopment, may also be a source of strabismus, at least in some cases, because of the growing evidence that they generally subtend cognitive dysfunction (e.g., Uhlhaas and Singer, 2006; Uhlhaas et al., 2009a,b, 2011).

#### **ABOUT THE ORIGINS OF BINOCULAR VISION LOSS**

Our article also aimed at reconsidering the question of the origins of binocular vision loss, including 3D perception loss and acuity loss. Such deficits are evidently classic consequences of strabismus, because of the abnormal visual experience they generate postnatally. But we are also convinced that their respective origins are central, at least in some cases. Logical deductions lead to such a hypothesis. As outlined above, any abnormality within the visual network (because of abnormal retinal waves prenatally or otherwise) may lead to abnormal visual perception, with a central origin. Thus, for example, an abnormal segregation of

#### **Table 1 | Peripheral and likely central origins of strabismus with early or late onset.**


*Origins at left are those that subtend strabismus occurring before 8 postnatal (PN) months while origins designated on the right are those subtending strabismus occurring beyond 2 postnatal years. Those mentioned in the middle of the table might subtend both early and late strabismus. In most cases, except in cases of palsies, whether they are situated in periphery or centrally, all the abnormalities being mentioned here have likely a genetic origin, expressing at different periods after birth (see text and Table 2). EOMs, extraocular muscles; CCDD, Congenital cranial dysinnervation disorders; CFEOM, Congenital fibrosis of the extraocular muscles; V1, primary visual cortex. Concerning the epidemiology of strabismus, see for instance: Chia et al. (2007), Greenberg et al. (2007), Mohney (2007).*

inputs from both eyes and/or any abnormality in the organization of the retina will lead to the development of abnormal ocular dominance maps and/or abnormal retinotopic maps in visual cortex. This unavoidably leads to an abnormal binocular integration. Abnormalities in the orientation and/or the spatial frequency maps in V1 (or beyond) will also unavoidably lead to amblyopia, and hence to binocular vision loss. More generally, any alteration within the M, P, or K pathways, from retina to cortex, likely with a genetic origin, will lead to such alterations. Any abnormal synchronization of neural activity at the cortical (or sub-cortical) level will also lead to amblyopia and/or abnormal 3D perception. Of interest in the present context, each may additionally lead to strabismus (cf. text for details, Section abnormal visual perception may induce strabismus).

#### **HOW TO IMPROVE TREATMENT OF STRABISMUS AND BINOCULAR VISION LOSS IN THE FUTURE**

Taking all the above developments into account, the main question is now: "How can ophthalmologists better assist strabismic patients and those with binocular vision loss in the future?" Evidently, they will have to continue applying the conventional treatments to limit the *consequences* of strabismus due to an abnormal postnatal experience. But considering what is included in the present article, more innovative treatments and new strategies would also need to be used, in particular with respect to the *origins* of strabismus. The same applies to binocular vision loss.

#### *Improving treatment of strabismus in the future Treatments being used for strabismus.*

*Conventional treatments.* As summarized in **Table 3**, evidently, the main aim of the ophthalmologist is presently to eliminate or to limit perceptive abnormalities due to strabismus in order to recover visual acuity, rectitude of the eyes and normal binocular vision as much as possible. For that, classically, the ophthalmologist currently addresses these functional disorders during the critical period and intervenes in periphery, at the level of the eyes, through refractive treatment, amblyopia treatment, binocular treatment and/or surgical treatment: (a) Glasses are prescribed after cycloplegia, which allows the correction of ametropia in order to control the influence of refractive errors and accommodative excess or lack in case of strabismus; (b) The treatment of (monocular) amblyopia is the priority since plasticity during

#### **Table 2 | Genetics of strabismus.**


#### •**Genetic abnormal development of nerves e.g.,**


#### **CENTRAL**

• **Genetic trait in accommodative esotropia and inheritance of refractive errors such as hyperopia**

• **Genetic trait in exophoria/exotropia**

Complex genetic inheritance, with the possible implication of recessive and dominant genes

*This table allows summarizing already identified genes being responsible for strabismus. In the periphery, some specific genes can be associated to some specific diseases concerning the extraocular muscles or the oculomotor nerves. Centrally, genes are also implicated in generating strabismus but their identification is more difficult, in particular because not one single gene is implicated. In addition, they might be dominant or recessive. CFEOM, Congenital fibrosis of the extraocular muscles; DRRS, Duane Radial Ray Syndrome; HGPPS, Horizontal Gaze Palsy with Progressive Scoliosis; BSAS, Bosley-Salih-Alorainy Syndrome; ABDS, Athabascan Brainstem Dysgenesis Syndrome.*

the postnatal period decreases progressively over time and therapeutic success depends on the timing of the treatment. Refractive treatment is the first step of amblyopia treatment and can be effective in mild anisometropic amblyopias. But it must be strongly emphasized that the treatment of amblyopia also requires patching of the sound eye. Some studies by the PEDIG, a group of North American Pediatric Ophthalmologists exploring different amblyopia therapies (Beck, 1998), have suggested that a "soft" treatment can be as efficacious as a "harder" one. For instance, 6 h patching vs. 2 h patching of the good eye would be sufficient (see, for example, Rees et al., 2007). However, it must be pointed out that only an improvement of vision was expected with such treatment, not a complete healing of amblyopia. To reach this latter stage full time patching is required for several weeks. Patching an eye increases the cortical input to the cortex from the amblyopic eye, and this effect is necessary to increase visual acuity. Note, however, that the efficacy of the "patchy method" greatly varies with the age of the patient since the plastic properties of the visual system evolve during the critical period (Epelbaum et al., 1993); (c) When visual acuity is recovered in strabismic patients, surgery allows the realignment of the eyes which is necessary to ensure binocular vision. Surgery has a direct effect on the EOMs, allowing the modification of the position of the eye: a recession of a muscle diminishes its effective force on the eye, whereas reinforcement is allowed by a muscular resection. However, this requires intervention on the EOMs, in particular at the level of their tendons where major muscle receptors are located. By taking into account that extraocular proprioception plays a major role in the maturation of V1, at least during the first half of the critical period, one must be aware that this may induce unfortunate consequences in the development of the brain (see above).

*New treatments.* The "monocular patchy method" has been used to treat amblyopia since major findings by Hubel and Wiesel (1965). But, over time, it has been thought to reduce binocular stimulation to the visual system. This is a very important issue, since despite what has been thought for decades, neurons in visual cortex have finally been shown to remain binocular in spite of strabismus, even if binocular interactions are abnormal (see above). This is the reason why new strategies implicating binocular stimulations are presently being developed to treat amblyopia and binocular vision loss in strabismic (and anisometropic) subjects. Hess and his colleagues are among the most active in that field, with their strategy to suppress interocular suppression in order to recover acuity by the amblyopic eye and 3D perception (Baker et al., 2007; Mansouri et al., 2008; Hess et al., 2010a,b, 2011; Zhou et al., 2012 cf. also Hess et al., 2014 for review). For that, they have developed dichoptic devices that allow a binocular stimulation with different images in each eye, the combination of which is stereoscopic. Without going into detail, during these binocular stimulations, the image with the lowest contrast is presented to the fellow fixing eye, which allows enhanced performance in the other eye. The practical management of patients using this approach is still under development but this approach for amblyopia, binocular vision loss and, more generally, consequences of strabismus on visual perception appears quite promising, more especially as it is also able to be used in adulthood (e.g., To et al., 2011; Black et al., 2012; Li et al., 2013; see also Hess and Thompson, 2013 for review). During the treatment of amblyopia (either strabismic or anisometropic), therapy sessions of this type could be carried out, in addition to patching therapy, in order to improve the treatment of the effects of strabismus. Patching would increase the monocular input from one eye to the cortex, avoiding the asymmetry of signal until a balance is obtained, whereas the binocular treatment would avoid interocular suppression and would help to restore binocular function. In the particular case of strabismus, it would also require alignment of the eyes to avoid inducing diplopia. The necessary conventional approaches through refractive treatment, amblyopia treatment and surgical treatment will need to be combined with these new approaches.

Another new and promising strategy to recover visual function in strabismic (and anisometropic) adults is also under development. This time, it consists of non-invasive transcranial brain stimulations, with the aim of modifying the balance of excitation and inhibition in the visual cortex (Thompson et al., 2010;

#### **Table 3 | Main objectives and main strategies of the ophthalmologist in case of strabismus.**

#### **Main objectives of the opthalmologist in case of strabismus**

Presently, the main aim of the ophthalmologist is to eliminate or to limit perceptive abnormalities due to strabismus in order to recover visual acuity, rectitude of the eyes and normal binocular vision as much as possible

In the future, the ophthalmologist will still have to eliminate or to limit perceptive abnormalities due to strabismus but will ALSO have to prevent strabismus and perceptive abnormalities to develop


*See text for details.*

Clavagnier et al., 2013; Spiegel et al., 2013; see also Hess and Thompson, 2013 for review). Used alone or in combination with binocular therapy (cf. see above) it may assist in the treatment of both amblyopia and binocular vision loss (e.g., Spiegel and Li, 2013). A precise knowledge of the brain abnormalities that are present in each case of strabismus is required to adjust the brain stimulation temporally and spatially, i.e., performing the treatment at the correct time during development and on the correct region of the brain to treat the consequences of strabismus. In the future it could also be possible to reshape the brain and the abnormalities that are responsible for strabismus (abnormal synchrony, lack of activity, etc.).

Indeed, the two new methods we have presented only aim now at treating consequences of strabismus. To further advance, on the basis of reliable indices, one may imagine that such methods may also be developed to treat the origins of strabismus and to prevent strabismus to develop in the future (see above for justification).

*Potential future strategies to further improve treatment of strabismus, in particular with respect to its origin.* As early as possible, ophthalmologists might improve the current management of strabismus by targeting both the *consequences* AND the *origin(s)* of strabismus. As set out above, very promising strategies are already being developed to limit or even eliminate amblyopia and binocular vision loss due to strabismus (or anisometropy), including in adulthood. Here, we suggest additional potential strategies for the future (cf. **Table 3** for summary). We propose first *to reach a better understanding* of both the origins and the consequences of strabismus than presently. Second, we are also convinced that to apply an *anticipating strategy* against strabismus, whenever possible, would be also very pertinent.

*To better understand the origins and consequences of strabismus.* Five strategies at least may be proposed to ophthalmologists in order to reach a better understanding of both the origins and the consequences of strabismus.

(a) **To acquire a better knowledge of the timing of the different phases of normal visual development in human**. Presently, ophthalmologists know a lot about the normal development of visual perception in humans, i.e., about "visual function" (cf. **Figure 1**). For example, they know perfectly that acuity increases from birth to reach its maximum between 4 and 8 PN years. They also know that binocular vision appears suddenly at about 3 PN months but continues to improve up to 9 PN years. But to move forward in the future, ophthalmologists would need to acquire a precise knowledge about the normal development of the visual system itself, from eye to cortex, before and after birth. This would constitute an additional reference for them. As illustrated above, this would require first the ability to relate genes, molecules, anatomy and neural activity to each "phase" of development of the visual system, including their timing. The same holds true concerning the timing of the critical periods of each attribute of the visual scene. As illustrated above, much is known already about all of these aspects in various mammal species, including monkeys, cats, ferrets, rats and mice. The general phases of development we are interested in here are basically the same in all mammals, except with respect to the *timings* (e.g., Huberman et al., 2008; Espinosa and Stryker, 2012 for review). Thus, finally, in relation to such an issue, what is missing principally for humans is the timing of the different phases of development. Second, ophthalmologists would have to understand how interactions between distributed brain regions mature with age to produce sophisticated cognitive functions such as visual perception (see Section Systematic Use of EEG and/or MEG and/or fMRI Recordings, in Combination with Psychological Analysis below for details). Interestingly, some recent reviews have started to treat these aspects by taking into account the time courses of neural proliferation, neural migration, apoptosis, synaptogenesis, establishment of neural circuits and myelination in the different regions of the human brain, including the visual cortex, before and after birth (e.g., Tau and Peterson, 2010; Menon, 2013). For that, neuroimaging, EEG and MEG recordings, together with traditional investigational approaches such as histological studies and cellular and molecular biology, have been used. Improving our understanding of these developmental processes in humans is likely a major key to the successful treatment of strabismus and binocular visual loss. To have this type of information with respect to the oculomotor system would be also very interesting.


question of the relationship between neural synchrony and the underlying anatomical and physiological changes that occur during normal brain development (e.g., Menon, 2013 for review). They also aim at associating a specific disruption of dynamic processes to abnormal connectivity and specific disturbances of cognitive or executive functions (see above; see also Menon, 2013 for review). Their idea is that recording dynamics of brain activity using electroencephalography (EEG) and/or magnetoencephalography (MEG) with or without new psychophysical measurements of visual perception must somehow reflect the functional architecture of cortical networks. "Because this architecture is determined by genetic factors and modified by experience, spontaneous or evoked activity patterns should contain information about evolutionary and epigenetically acquired knowledge regarding the world and serve as a convert internal model for perception and action" (e.g., Singer, 2013 for review). This indicates that recording resting-state or evoked activity from the brain would be sufficient, at least in principle, to identify whether something is wrong in the cortical networks. As suggested by Buzsáki et al. (2013), "oscillopathies or dysrhythmias could reflect malfunctioning network and, as endophenotypes, could assist in specifying diagnostics." This has not been applied as yet but it is a promising possibility for the future, in particular in the context in which we are interested here. One might speculate that to establish specific relations between "oscillopathies" or "dysrhythmias" in the brain and various types of strabismus might greatly help, both to prevent strabismus to develop and to cure dilatory consequences due to strabismus, at least in some cases. Such techniques are particularly interesting here in that they are non-invasive techniques and can be used in infants at the earliest ages (e.g., Csibra et al., 2000). The fMRI could also help in assessing abnormal connectivity, although it is not easy to perform this in young infants (e.g., Li et al., 2011).

*To anticipate against negative effects of strabismus.* The EEG and/or MEG and/or fMRI recordings, with or without psychological analysis, might be used for the diagnosis of brain "abnormalities" that are present in the case of strabismus, and that should be treated, for example by using TMS (see above). Also, it should be necessary to anticipate the negative effects of strabismus on the visual system, and one may propose to perform a genetic screening for that purpose. In the future, one may even consider repairing altered genes.

(a) **To perform systematically a genetic screening.** Genetic knowledge, whatever its complexity, may be used (see above). Gene profiles can now be established relatively easily in humans, using non-invasive methods. Some genes at least are already known to be associated with strabismus (see above). To identify more would certainly assist, whether they are implicated in strabismus or absence of binocular vision. Establishing, as early as possible after birth, that a child is at risk of developing strabismus by showing that he or she has one or more affected gene(s) would at least allow ophthalmologists to develop a *treatment plan* to limit dilatory consequences of strabismus. Indeed, as outlined above, consequences of strabismus and/or absence of binocular vision might be rather difficult to treat in particular when strabismus has occurred very early postnatally. Thus, to diagnose strabismus as early as possible would only be beneficial.

(b) **To use genetic therapy.** In the future, even if it is currently not a reality, one may also expect to use innovative technologies such as genetic therapy. It would allow the repair of altered genes susceptible to induce strabismus and/or loss of binocular vision, at least in some cases. Is it a utopia to be able to repair the brain itself in the present context? Whatever the answer, intervening as early as possible is the best strategy.

#### *How to improve treatment of binocular vision loss in the future?*

As mentioned above, some newly developed strategies such as binocular therapy and TMS stimulations are improving the treatment of binocular vision loss caused by strabismus (e.g., Hess and Thompson, 2013; Hess et al., 2014 for reviews). But it is important to emphasize that, at least presently, binocular vision cannot be restored whatever the form of strabismus. In particular, it cannot be obtained after an early onset strabismus. Congenital binocular vision loss, without strabismus, is also presently impossible to treat. In both cases, this is likely due to a major insult of the maturational process of the visual system with respect to binocularity, occurring either before or soon after birth, due to innate/genetic factors. Evidently, in these cases, neural networks in the brain would have to be "repaired" through either reshaping (if early enough after birth) or by activating functional sleepy synapses or otherwise. However, to reach such goals, the origin and the exact characteristics of the disease would need to be identified in great detail. Appropriate strategies to remove or even to prevent binocular vision loss should then be developed. Our suggestion for ophthalmologists is that they increase their knowledge concerning the different phases that characterize the normal development of the visual system.

Combining better knowledge of the origins of strabismus and loss of binocular vision with new therapies will no doubt allow the more efficient management of these pathologies.

#### **ACKNOWLEDGMENTS**

We would like to thank the LUZ Group for their financial support, Ms. Nicole Quenech'Du for the conception of the figures and Ms. Chloe Ann Barker for revising the manuscript.

#### **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 March 2014; accepted: 27 August 2014; published online: 25 September 2014.*

*Citation: Bui Quoc E and Milleret C (2014) Origins of strabismus and loss of binocular vision. Front. Integr. Neurosci. 8:71. doi: 10.3389/fnint.2014.00071*

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

*Copyright © 2014 Bui Quoc and Milleret. 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.*

## Neuroimaging of amblyopia and binocular vision: a review

## *Olivier Joly1,2 \* and Edit Frankó3,4 \**

<sup>1</sup> MRC Cognition and Brain Sciences Unit, Cambridge, UK

<sup>2</sup> Department of Experimental Psychology, University of Oxford, Oxford, UK

<sup>3</sup> Department of Neurodegenerative Disease, Institute of Neurology, University College London, London, UK

<sup>4</sup> National Prion Clinic, National Hospital for Neurology and Neurosurgery, University College London Hospitals, London, UK

#### *Edited by:*

Olivier A. Coubard, CNS-Fed, France

#### *Reviewed by:*

Kerstin Erika Schmidt, Federal University of Rio Grande do Norte, Brazil Éva M. Bankó, Hungarian Academy of Sciences – Research Center for Natural Sciences, Hungary

#### *\*Correspondence:*

Olivier Joly, MRC Cognition and Brain Sciences Unit, 15 Chaucer Road, Cambridge CB2 7EF, UK e-mail: olivier.j.joly@gmail.com; Edit Frankó, National Prion Clinic, National Hospital for Neurology and Neurosurgery, University College London Hospitals, London WC1N 3BG, UK e-mail: edit.franko@gmail.com

Amblyopia is a cerebral visual impairment considered to derive from abnormal visual experience (e.g., strabismus, anisometropia). Amblyopia, first considered as a monocular disorder, is now often seen as a primarily binocular disorder resulting in more and more studies examining the binocular deficits in the patients. The neural mechanisms of amblyopia are not completely understood even though they have been investigated with electrophysiological recordings in animal models and more recently with neuroimaging techniques in humans. In this review, we summarize the current knowledge about the brain regions that underlie the visual deficits associated with amblyopia with a focus on binocular vision using functional magnetic resonance imaging. The first studies focused on abnormal responses in the primary and secondary visual areas whereas recent evidence shows that there are also deficits at higher levels of the visual pathways within the parieto-occipital and temporal cortices. These higher level areas are part of the cortical network involved in 3D vision from binocular cues. Therefore, reduced responses in these areas could be related to the impaired binocular vision in amblyopic patients. Promising new binocular treatments might at least partially correct the activation in these areas. Future neuroimaging experiments could help to characterize the brain response changes associated with these treatments and help devise them.

**Keywords: amblyopia, binocular vision, stereopsis, visual cortex, neuroimaging**

#### **INTRODUCTION**

Amblyopia is the reduction of best-corrected visual acuity to less than 6/9 in Snellen optotype or at least two-line difference in LogMAR optotype between the eyes. This measured reduction in visual acuity cannot be directly related to structural abnormalities of the eye and cannot be corrected by spectacle glasses alone. Amblyopia is often considered as a monocular disorder. Indeed, it usually affects one eye, although rarely it can be bilateral, and it is the most common cause of monocular blindness. The prevalence of amblyopia is 2–4% in the general population (Donnelly et al., 2005; Robaei et al., 2006; Williams et al., 2008). Amblyopia is believed to be caused by abnormal visual experience during the critical period of visual development in early life (first 7–10 years). It is mainly associated with strabismus or anisometropia, more rarely with visual deprivation arising from ptosis or congenital cataract.

The three most common types of amblyopia are strabismic, anisometropic, and combined mechanism (both strabismus and anisometropia are present) amblyopia. The prevalence of these different types seems to depend on the age; in children under the age of three, amblyopia affects about 50% of the children suffering from strabismus and about 18% of the children with anisometropia (Birch and Holmes, 2010). However, this ratio seems to reverse in adults; Attebo et al. (1998) found that in 50% of the patients the cause of amblyopia was anisometropia whereas strabismus was responsible only in 19% of the cases. A possible explanation for this difference in prevalence is that anisometropia may develop later, or it may require longer duration to cause

amblyopia (Birch, 2013). The different types of amblyopia are also characterized by different patterns of visual acuity and contrast sensitivity loss. Strabismic amblyopia results in moderate acuity loss and increased contrast sensitivity at low spatial frequency, whereas anisometropic amblyopia causes moderate acuity loss and decreased contrast sensitivity. In combined mechanism amblyopia the acuity is usually very poor whereas the contrast sensitivity is normal or slightly reduced (McKee et al., 2003). It was also shown that the reduction in contrast sensitivity is disproportionally higher for high as compared to low spatial frequencies (Hess et al., 1978; Bradley and Freeman, 1981; Hess and Pointer, 1985). Importantly, visual acuity in amblyopia was also found to correlate with residual binocular function; patients with no residual binocular function generally have poorer acuity (McKee et al., 2003). The defect in stereopsis also depends on the type of amblyopia; it is more often disrupted in strabismic than in anisometropic amblyopia (McKee et al., 2003).

According to the currently accepted hypothesis, amblyopia arisesfrom the mismatch between the images to each eye; the information from one eye becomes favored while from the other eye it is suppressed (Harrad, 1996). This suppression causes reduction of visual acuity in this eye and therefore compromises binocular vision. However, it is not clear whether the visual acuity loss is the cause or the consequence of the impaired binocular function. Normal binocular vision provides a very strong visual cue for depth perception which in turn improves our ability for prehension movements : grasping and reaching (in particular the terminal reach phase) tasks (Melmoth and Grant, 2006). It has been shown that amblyopic patients indeed are impaired in planning and execution of reaching movements (Niechwiej-Szwedo et al., 2011a) and in the temporal coordination of eye-hand movements (Niechwiej-Szwedo et al., 2011b). Recently, amblyopia has been considered more as a primarily binocular disorder which motivated new approaches to treatments focusing on restoring the binocular vision.

Many studies examined the cortical network involved in the processing of depth from binocular information but only a few imaging studies have tested amblyopic patients under binocular viewing conditions. Here we review the studies focusing on the cortical processing of binocular vision and the cortical deficits in amblyopia. We highlight brain regions in which dysfunction might be related to the binocular deficits in these patients. Future work will help understand the neural plasticity mechanisms which might be involved in these brain regions in patients undergoing binocular treatments.

#### **BINOCULAR VISION**

Animals with forward facing eyes such as primates have the ability to extract depth information from the 2D retinal images. When gazing at an object, the eyes' horizontal separation induces projections onto the retinae which differ mainly in their horizontal positions. This difference in the retinal images is called horizontal binocular disparity. Detection of binocular disparity was demonstrated in human infants between 2 and 4 months of age, by comparing the visually evoked potentials (VEP) elicited by random-dot stereograms and classic black and white checkerboards (Petrig et al., 1981). Moreover, Yonas et al. (1987) demonstrated using a preferential looking procedure that 4-month-old infants sensitive to binocular disparity can also perceive the 3D shape from binocular depth cues. Despite a rather early start of binocular vision development (Fox et al., 1986), stereoacuity reaches adult level only between 6 and 9 years of age (Romano et al., 1975; Simons, 1981; Giaschi et al., 2013).

Non-human primates are very good animal models for investigating binocular vision in humans and therefore to understand its associated disorders. The main reason for this is that the monkey visual system is close to the human visual system in many aspects including its development and psychophysical properties of monocular (De Valois et al., 1974) and binocular visual processing (Cao and Schiller, 2002). Therefore, many of the studies reported hereafter were performed in non-human primates.

#### **ELECTROPHYSIOLOGICAL STUDIES**

Visual information delivered from the retina of either eye remains largely independent until it reaches the cortex. Therefore the first stage of binocular disparity processing is located in the primary visual cortex (area V1; Poggio and Fischer, 1977; Cumming and Parker, 1999). Although V1 neurons encode absolute disparity they do not encode for relative disparity (Cumming and Parker, 1999). The relative disparity, which is the difference in absolute disparities, is critical for depth-structure perception as it is independent of eye position. This suggests, disparity selective neurons in V1 are not associated with stereoscopic depth perception per se (Cumming and Parker, 1997) but perhaps more involved in vergence eye movements (Masson et al., 1997). Several

studies using single-cell recording techniques in monkeys have reported disparity selective neurons in extrastriate areas. Studies have described such neurons in the early visual areas V2 (Hubel and Livingstone, 1987; Poggio et al., 1988) and V3 (Felleman and Van Essen, 1987; Adams and Zeki, 2001), in the dorsal pathway in areas V3A (Anzai et al., 2011) and middle temporal (MT; Maunsell and Van Essen, 1983; DeAngelis and Newsome, 1999), in the ventral pathway in area V4 (Watanabe et al., 2002; Hegdé and Van Essen, 2005), and in the inferior temporal cortex particularly in the rostral lower bank of the superior temporal sulcus (STS; Janssen et al., 1999; Liu et al., 2004). In the parietal cortex, in particular in the lateral bank of the intraparietal sulcus (IPS), neurons encoding orientation in depth were reported in the caudal intraparietal area (CIP; Taira et al., 2000; Tsutsui et al., 2001), area LIP (lateral intraparietal; Gnadt and Mays, 1995), and area AIP (anterior intraparietal; Srivastava et al., 2009) where neurons were also recorded with selectivity to 3D depth profiles. Finally, in the frontal lobe, disparity-selective neurons were reported in the frontal eye field (FEF) area (Ferraina et al., 2000). In the ventral premotor cortex, a rather high proportion of disparity selective neurons was found (Theys et al., 2012). These neurons were found in area F5 known to house visuomotor neurons (Murata et al., 1997) and to receive projections from the parietal area AIP (Borra et al., 2008).

#### **BRAIN IMAGING IN HUMANS AND NON-HUMAN PRIMATES**

Several studies using functional magnetic resonance imaging (fMRI) in monkeys have either confirmed or predicted the above electrophysiological results. These imaging studies in non-human primates allow on the one hand a better integration of human fMRI results with the monkey single cell studies and on the other hand a possibility to assess the putative homologies between cortical areas in the two species. In the dorsal stream, Tsao et al. (2003) reported larger activations to non-zero than to zero disparity stimuli in area V3A and in the caudal intraparietal regions in both humans and monkeys. In humans, fMRI activations for 3D shape from disparity were reported in V3A and V7 (Backus et al., 2001; Georgieva et al., 2009) and fMRI adaptation to either relative or absolute disparities (Neri et al., 2004) was higher to absolute disparity in dorsal areas (V3A, MT/V5, V7) while ventral areas (hV4, V8/V4) showed a similar adaptation to both types of disparities. The role of the regions in the lateral bank of the monkey IPS in the processing of 3D shape from disparity was also investigated. Durand et al. (2007) found a difference between CIP and rostral part (anterior LIP and AIP) of the IPS in the different aspects of depth information in monkeys. In humans, several studies have clearly reported the involvement of the parietal cortex (Naganuma et al., 2005), DIPSM/DIPSA (dorsal IPS medial/anterior) and phAIP (putative human AIP) in processing of depth from disparity (Durand et al., 2009; Georgieva et al., 2009; Minini et al., 2010). In the ventral premotor cortex, imaging in monkeys (Joly et al., 2009) revealed responses to 3D surfaces in area F5a. This finding was later confirmed with electrophysiology and the report of disparity-selective neurons in this region (Theys et al., 2012). A similar frontal region was reported in humans using the same stimuli (Georgieva et al., 2009). In the ventral stream, a multi-voxel pattern analysis (MVPA) fMRI

study (Preston et al., 2008) has shown that the lateral occipital area (LO) codes for the sign of depth position (near vs far) while it is invariant to the magnitude of disparity. The LO complex together with area hMT+ was shown to be particularly responding to the 3D shapes either derived from the combination of binocular disparity and perspective (Welchman et al., 2005) or defined as the correlation between fMRI signal and observers'discrimination performancefor disparity-defined shape (Chandrasekaran et al., 2007). A region in the rostral part of the lower bank of the STS in monkeys (Joly et al., 2009) and the posterior inferior temporal gyrus (ITG) in humans (Georgieva et al., 2009) were also found to be sensitive for 3D stimuli. Most of these human cortical regions that define a network for depth perception from binocular disparity (illustrated in **Figure 1**) could have impaired function in amblyopia and therefore be responsible for the impaired binocular vision detected in the patients.

#### **NEURAL CORRELATES OF AMBLYOPIA**

The classic experiments of (Wiesel and Hubel, 1965) in kittens opened the way to examine the neural basis of visual loss in amblyopia. Both the feline and primate models of amblyopia failed to reveal significant anatomical and physiological abnormalities in the retina of the amblyopic eye (Cleland et al., 1980, 1982). Similarly in humans, many studies have described the retina as essentially normal in amblyopes (Hess and Baker, 1984; Repka et al., 2009; Al-Haddad et al., 2011; Walker et al., 2011; Birch, 2013). At the next level of visual processing, in the lateral geniculate nucleus (LGN), minor changes were reported in the morphology of the cells (Guillery, 1973; Sloper et al., 1988; Sloper and Collins, 1998). In spite of these morphological changes, electrophysiological studies in cats and monkeys, demonstrated that the LGN cells had normal spatial and temporal response properties following visual deprivation (Cleland et al., 1980, 1982; Crewther et al., 1985; Movshon et al., 1987; Sasaki et al., 1998).

#### **CORTEX**

Studies focusing on the cortex, reported reduction in binocularly driven neurons in the primary V1, and in the number of cells driven by the amblyopic eye (Wiesel and Hubel, 1963; Kiorpes, 2006). In infant monkeys, experimentally induced blur resulted in reduced spatial resolution and selective loss of neurons tuned to high spatial frequencies (Movshon et al., 1987; Kiorpes et al., 1998). The same authors also found that the binocular cortical connections disrupted by strabismus (Löwel and Singer, 1992) can lead to the development of fixation preference for one eye (Kiorpes et al., 1998; Kiorpes and McKee, 1999). In strabismic cats, Roelfsema et al. (1994)found similar firing rates in V1 for both eyes but reduced response coordination

**FIGURE 1 | Parcellation of different cortical regions involved in visual processing.** Some of these regions are particularly involved in binocular vision and some regions are known to show deficits in amblyopes under diverse visual stimulation. Lateral view **(A)** and ventral view **(B)** are presented. The 3D rendering (Anatomist, www.brainvisa.info) represents the cortical surface of the Conte69 human surface-based atlas (Van Essen et al., 2012). V1, V2, MT+ as defined by (Fischl et al., 2008), V3A, V3B, V4v, V7, IPS1/2/3/4 as defined by (Swisher et al., 2007), V3d, LO1, LO2, PITd, PITv,

as defined by (Kolster et al., 2010), occipitotemporal area BA37, inferior temporal area BA20 available in Caret software (www.nitrc.org/projects/ caret/, Van Essen et al., 2001). CalcS, calcarine sulcus; LOS, lateral occipital sulcus; TOS, transverse occipital sulcus; ITG, inferior temporal gyrus; ITS, inferior temporal sulcus; MTG, middle temporal gyrus; STS, superior temporal sulcus; STG, superior temporal gyrus; LF, lateral fissure; OTS, occipitotemporal sulcus; CoS, collateral sulcus; PHG, parahippocampal gyrus; PCG, postcentral gyrus; CS, central sulcus.

for responses evoked through the amblyopic eye of behaviourally tested strabismic amblyopic cats. This reduced coordinated activity between neurons driven by the amblyopic eye in V1 might be the origin of the transmission failure to higher cortical areas (Fries et al., 2002; Schröder et al., 2002). Many other studies also examined the binocular interactions withinV1 detecting increased binocular suppression (Smith et al., 1997; Zhang et al., 2005). This increase in suppression can be responsible for the detected reduction in binocularly driven neurons in V1, as it was shown previously that reducing the suppression by the GABA-receptor blocker bicuculline restored the binocular input to more than half of the cortical neurones (Duffy et al., 1976). Furthermore, Sengpiel et al. (2006) suggested that this increase in suppression might also be responsible for the loss of binocular summation seen in amblyopic patients. This hypothesis is further supported by the observation that binocular summation can occur if the signal strength to the fellow eye is reduced to compensate for the suppression of the amblyopic eye (Baker et al., 2007). Going further in the cortical visual processing, El-Shamayleh et al. (2010) found that in area MT fewer cells responded to the stimulation of the amblyopic eye as compared to the fellow eye in amblyopic macaques. In humans, many studies used VEPs to investigate the neural correlates of amblyopia. Most of them reported smaller amplitudes and/or abnormal latencies (Arden et al., 1974; Sokol, 1983; Kubová et al., 1996; McKerral et al., 1999) when the amblyopic eye was stimulated. A more recent study also demonstrated that the amblyopic deficit measured by VEPs correlated with the task performance (Bankó et al., 2013b). Moreover, using complex stimuli (faces), the same groupfound a delay of N170 relative to the early P1 component over the right hemisphere during amblyopic eye stimulation suggesting a deficit in higher visual areas involved in face perception (Bankó et al., 2013a).

#### **NEUROIMAGING IN HUMANS WITH AMBLYOPIA**

Non-invasive neuroimaging techniques allow us to investigate the neural correlates of amblyopia in humans (see **Table 1**), and compare them to the results found in animal models. Few studies focused on the subcortical structures in amblyopic patients. Using fMRI, it was shown that the LGN had reduced responses when driven by the amblyopic eye compared with the fellow eye (Miki et al., 2003; Hess et al., 2009). However, Sherman and Guillery (2002) drew attention to the fact that only 6% of the cells in LGN convey feedforward information from the retina to the cortex, the vast majority of the cells have a modulatory function. This modulation mainly originates from layer 6 of V1 (Van Horn and Sherman, 2004) and it is more susceptible to anesthesia than the feedforward input from the retina. Hess et al. (2009) used fMRI to overcome the possible effects of anesthesia, and investigated the activity in the LGN in human amblyopes. When comparing the BOLD signal change in the LGN, they found reduced averaged and peak activity when stimulating the amblyopic eye. These findings were consistent with the results of Miki et al. (2003) when examining a single amblyopic subject. This reduced activation can result from the mild morphological changes in the LGN reported previously (Wiesel and Hubel, 1963). Another possible explanation is that the modulatory feedback connections from V1 are responsible for this reduction, modifying the input of the binocular cells

in V1 already at the level of LGN. This is more consistent with the electrophysiological findings, namely that the first signs of deficit are in area V1.

Many studies therefore investigated area V1 in amblyopic patients. Early imaging studies in humans with amblyopia used positron emission tomography (Demer et al., 1988) and single photon emission computed tomography (Kabasakal et al., 1995). They reported reduced primary V1 response to the amblyopic eye compared to the fellow eye. Similarly, Choi et al. (2001) found that the amblyopic eye showed reduced activation in the calcarine sulcus using monocular presentation of black and white checkerboard patterns at different spatial and temporal frequencies. This suppression was more important for high spatial frequency in anisometropic amblyopia and for low spatial frequency in strabismic amblyopia. Lee et al. (2001) also focused on the activations in the calcarine fissure (area V1) with monocular presentation of checkerboard patterns and compared them between strabismic and anisometropic amblyopia. They found during monocular stimulation that the proportion of voxels activated by either normal or amblyopic eye was lower in the strabismic group than in the anisometropic group. The activation by higher spatial frequency stimuli is reduced in the anisometropic group, but not in the strabismic group. Goodyear et al. (2000) defined a region of interest that mainly covered area V1 and reported a reduced area (number of voxels) of activation during the stimulation of the amblyopic compared to the normal eye. In subjects with strabismic amblyopia, Barnes et al. (2001) reported reduced activation in visual areas V1 and V2. In one of the very few studies that used binocular stimulation, Algaze et al. (2002) measured in the occipital cortex the BOLD response to monocular and binocular presentation of sinusoidal gratings in amblyopic patients and compared it to the responses in controls. Monocular stimulation of the amblyopic eye induced a lower response relative to the same stimulation in the fellow eye, which is expected from the visual loss. More importantly, subjects with amblyopia showed a greater difference in activations (in terms of level and spatial extent of the activation) between binocular and monocular stimulation as compared to the control subjects, but this difference was driven by the amblyopic eye and the response to the fellow eye was close to the level of response for binocular stimulation. Similarly, Körtvélyes et al. (2012) reported that ERP responses were also statistically indistinguishable when stimulating both eyes or only the fellow eye. These results are in agreement with the increased suppression of the amblyopic eye by the fellow eye. Moreover, Farivar et al. (2011) reported delayed and reduced BOLD response in V1 for the amblyopic eye stimulation and a particularly high suppression when the fellow eye was open. More recently, Li et al. (2011b) investigated effective connectivity and reported a reduced connectivity of geniculatestriate and striate-extrastriate networks. Interestingly, the authors also found that this connectivity loss correlated with the depth of amblyopia.

Only a few studies examined the higher level visual areas in amblyopic patients. In the ventral visual stream, Muckli et al. (2006)found a reduction of responses to stimulation of the amblyopic eye in V4+/V8 and LO complex as compared to V1/V2 in both anisometropic and strabismic amblyopes. This suggests transmission failure from lower to higher visual areas.


**| Summary of the fMRI studies in humans with amblyopia.**

**Table 1**

Using more complex stimuli, Lerner et al. (2003) reported reduced activity for faces in the posterior fusiform gyrus (pFs), but normal activity for houses in the parahippocampal place area (PPA). Note thatVEP measurements were also reduced for foveally presented faces (Körtvélyes et al., 2012). In a later study, the same authors (Lerner et al., 2006) mapped activations for small and large objects. They found that during amblyopic eye stimulation, not only early visual areas but also high level visual areas showed reduced activation for foveally presented small stimuli when compared to fellow eye stimulation.

Conner et al. (2007) performed retinotopic mapping under monocular and binocular viewing conditions in amblyopes and looked at the activation in the foveal representation in V1 and in extrafoveal V1 and V2. They found a particularly high suppression at the foveal representation of the amblyopic eye when the fellow eye was open.

Very little is known about the visual areas on the dorsal pathway including the motion areas MT and MST (medial superior temporal) of amblyopic subjects. In cats, the dorsal pathway seems less affected than the ventral pathway (Schröder et al., 2002). Psychophysical studies suggest that both perception of global motion and translation of vision into movement are affected in amblyopic subjects (Simmers et al., 2003, 2005), implying deficits in the dorsal visual pathway leading to the posterior parietal cortex. A study, with attentive visual tracking of moving targets (Secen et al., 2011) reported a reduced activity in areaMT+for both eyes in amblyopic patients as compared to control subjects. This reduced activation was found for passive viewing and all of the tracking conditions. Further in the dorsal pathway, in the FEF and the anterior IPS activation from the amblyopic eye was only reduced in the condition of high attentional load (tracking several targets). Beside the classic activation studies, other MR imaging studies such as resting-state functional connectivity (Lin et al., 2012; Ding et al., 2013; Wang et al., 2014) and fMRI adaptation were used to investigate the dysfunction in amblyopia. Wang et al. (2014) have reported in amblyopic patients a reduced functional connectivity between the visual areas and parietal and frontal cortices that subserve visuomotor and visual-guided actions. This indicates that amblyopia might affect a large network beyond theV1. fMRI adaptation technique which assumes that fMRI repetition suppression reflects neuronal adaptation, has been used recently (Jurcoane et al., 2009; Li et al., 2011a). In the first study,Jurcoane et al. (2009), interocular transfer of adaptation (IOTA) was measured using orientationselective fMRI adaptation in normally sighted observers and in stereo-deficient amblyopic subjects. They found that amblyopic subjects showed consistent monoptic adaptation, but no IOTA in any striate and extrastriate cortical regions. Li et al. (2011a) reported cortical (from V1 and beyond) fMRI adaptation effects which were reduced in response to amblyopic eye stimulation.

#### **BINOCULAR TREATMENT IN AMBLYOPIA**

For long, amblyopia was considered as a disorder of monocular vision. The treatment therefore was also based on this view. Indeed, patching or pharmacological penalisation of the normal eye resulted in improved visual acuity. However, the treatment is mainly effective in children, and has a high risk for recurrence once the patching is stopped (Bhola et al., 2006). Adults who were not treated during childhood, or whose visual acuity decreased after the patching was stopped, had very limited possibilities to regain their vision. Methods using virtual reality and 3D video games were tested as possible substitute for patching (Waddingham et al., 2006a,b; Gargantini, 2011).

A recent theory looks at amblyopia as a primarily binocular disorder and suggests that the treatments should focus on restoring the binocular vision. Baker et al. (2007) demonstrated that amblyopic patients, in contrast with the previous beliefs, can experience binocular summation. This summation can occur when the suppression of the amblyopic eye is accounted for by reducing the contrast in the fellow eye. Based on this finding and the hypothesis that amblyopia is primarily a binocular disorder, Hess and colleagues (Hess et al., 2011; To et al., 2011) proposed a new binocular treatment (for a review, see Hess et al., 2014). They first used a dichoptic coherence motion discrimination task (Hess et al., 2010b). Later they adapted the method to a popular video game (Tetris, Honolulu, HI, USA) that would capture the patients' attention more resulting in better compliance with the training. The patients viewed the game dichoptically; part of the information (falling blocks) was presented only to the amblyopic eye with fixed contrast, whereas the other part (superficial ground plane blocks) was presented only to the fellow eye with decreased contrast. Only the less relevant deeper ground plane blocks were presented to both eyes in order to help binocular fusion. To play the game successfully, information from the two eyes had to be combined. By adjusting the contrast of stimulation to the fellow eye, patients could experience binocular summation, and play the game. Training nine adults with this dichoptic game that facilitated binocular summation, resulted in decreased suppression of the amblyopic eye, significantly greater improvements in visual acuity and stereopsis than with monocular training (Li et al., 2013). The decreased suppression was demonstrated as a decreased difference in stimulus contrast between the amblyopic and fellow eye that still allowed binocular summation.

This treatment overcomes many weaknesses of the previous treatment strategy using patching of the fellow eye, namely that it is effective in adults, well beyond the critical period of visual development, supports the binocular interaction between the eyes and increases the compliance with treatment when adapted to popular video games. Long-term follow up of the patients treated dichoptically will reveal whether this treatment would also decrease the rate of recurrence.

Another promising technique for treating amblyopia in adults can be brain stimulation. Thompson et al. (2008) have shown that repetitive transcranial magnetic stimulation (rTMS) of the V1 can temporarily improve contrast sensitivity in the V1 of adult amblyopic patients. When applied for 5 consecutive days (Clavagnier et al., 2013), rTMS was shown to have a long lasting effect (tested up to 78 days). A recent study (Spiegel et al., 2013) using brain stimulation (anodal transcranial direct current stimulation) and fMRI measurements in amblyopic patients indicated that the stimulation could equalize the response of the V1 to inputs from each eye. This latter study also suggests that fMRI could be used to understand the neural mechanisms and the brain regions involved in these therapies (e.g., Zhai et al., 2013).

#### **CONCLUSION**

Amblyopes suffer not only from poor visual acuity but also from deficits in binocular vision. Binocular disparity, a strong visual cue for depth perception, involves many cortical regions and some of them were shown to respond abnormally in amblyopic patients. Imaging studies in amblyopia started to use binocular stimulation, however, the cortical mechanisms of the binocular impairments remain largely unknown. Binocular treatment, a very promising alternative to patching, encourages binocular summation and might involve neural plasticity in brain regions involved in binocular vision such as the parietal cortex.

#### **ACKNOWLEDGMENT**

The authors acknowledge Diana Kyriazis and the two reviewers for their comments.

#### **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 March 2014; accepted: 12 July 2014; published online: 06 August 2014. Citation: Joly O and Frankó E (2014) Neuroimaging of amblyopia and binocular vision: a review. Front. Integr. Neurosci. 8:62. doi: 10.3389/fnint.2014. 00062*

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

*Copyright © 2014 Joly and Frankó. 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 neural bases of spatial frequency processing during scene perception

#### *Louise Kauffmann1,2 , Stephen Ramanoël 1,2 and Carole Peyrin1,2 \**

<sup>1</sup> University Grenoble Alpes, LPNC, Grenoble, France

<sup>2</sup> CNRS, LPNC, Université Pierre Mendès France, Grenoble, France

#### *Edited by:*

Olivier A. Coubard, CNS-Fed, France

#### *Reviewed by:*

Summer Sheremata, George Washington University, USA Michelle R. Greene, Stanford University, USA

#### *\*Correspondence:*

Carole Peyrin, Centre National de la Recherche Scientifique, Laboratoire de Psychologie et NeuroCognition, UMR 5105, Université Pierre Mendès France, BP 47, 38040 Grenoble Cedex 09, France e-mail: carole.peyrin@

upmf-grenoble.fr

Theories on visual perception agree that scenes are processed in terms of spatial frequencies. Low spatial frequencies (LSF) carry coarse information whereas high spatial frequencies (HSF) carry fine details of the scene. However, how and where spatial frequencies are processed within the brain remain unresolved questions. The present review addresses these issues and aims to identify the cerebral regions differentially involved in low and high spatial frequency processing, and to clarify their attributes during scene perception. Results from a number of behavioral and neuroimaging studies suggest that spatial frequency processing is lateralized in both hemispheres, with the right and left hemispheres predominantly involved in the categorization of LSF and HSF scenes, respectively. There is also evidence that spatial frequency processing is retinotopically mapped in the visual cortex. HSF scenes (as opposed to LSF) activate occipital areas in relation to foveal representations, while categorization of LSF scenes (as opposed to HSF) activates occipital areas in relation to more peripheral representations. Concomitantly, a number of studies have demonstrated that LSF information may reach high-order areas rapidly, allowing an initial coarse parsing of the visual scene, which could then be sent back through feedback into the occipito-temporal cortex to guide finer HSF-based analysis. Finally, the review addresses spatial frequency processing within scene-selective regions areas of the occipito-temporal cortex.

**Keywords: natural scene, spatial frequencies, coarse-to-fine, hemispheric specialization, retinotopy, parahippocampal place area**

#### **INTRODUCTION**

Scenes containing more realistic and more natural stimuli have increasingly become the object of scientific interest over the last 20 years, as they involve the perception of stimuli which are more complex and more realistic than simple objects or drawings. It is now widely agreed that visual recognition of scenes is a fast, automatic and reliable process. Experimental studies have shown that complex natural scenes can be categorized very rapidly (under 150 ms; Thorpe et al., 1996), indicating that a simple and efficient coding process is involved. There is considerable evidence suggesting the importance of the spatial frequency contents of images during scene recognition (Ginsburg, 1986; Field, 1987; Tolhurst et al., 1992; Hughes et al., 1996). On one hand, the primary visual cortex is mainly dominated by complex cells which respond preferentially to spatial frequencies (Poggio, 1972; De Valois et al., 1982a,b). On the other hand, findings from simulation and psychophysical experiments indicate that information from low/medium frequencies of the amplitude spectrum suffices to enable scene categorization (Torralba and Oliva, 2003; Guyader et al., 2004). Supported by convergent data from the functional neuroanatomy of magnocellular and parvocellular visual pathways (Van Essen and Deyoe, 1995), neurophysiological recordings in primates (for a review, see Bullier, 2001), and psychophysical results in humans (Ginsburg, 1986; Hughes et al., 1996), current influential models of visual

perception (Schyns and Oliva, 1994; Bullier, 2001; Bar, 2003; Hegde, 2008) suggest that the first stage of visual analysis consists of the extraction of visual elementary features at different spatial frequencies. Low spatial frequencies (LSF), conveyed by fast magnocellular pathways, provide a coarse information about a visual stimulus (e.g., the global shape and structure of a scene), whereas high spatial frequencies (HSF), conveyed more slowly by the parvocellular pathways, provide finer information about the stimulus (e.g., the edges and borders of an object in the scene).

However, exactly how and where spatial frequencies are processed within the brain remain unsettled questions. The debate on retinotopic organization and/or the existence of cerebral asymmetries in the occipital cortex in spatial frequency processing is still ongoing in the literature. A number of studies demonstrated a retinotopic mapping of spatial frequency processing in the occipital cortex and have for example showed that the perception of HSF sinusoidal gratings activated the foveal representation in all retinotopic areas of the occipital cortex, and LSF sinusoidal gratings activated more peripheral representations in the same cortical areas (Sasaki et al., 2001; Henriksson et al., 2007). However, other authors argue in favor of the hemispheric specialization for spatial frequency processing at the level of visual retinotopic areas, with the right hemisphere preferentially specialized in the processing of LSF information and the left hemisphere preferentially specialized in HSF information processing (Iidaka et al., 2004; Peyrin et al., 2004, 2006). It appears therefore important to investigate both retinotopic processing and hemispheric specialization on the same visual stimuli.

In addition, there is considerable evidence suggesting that spatial frequency processing takes place in a predominantly and default coarse-to-fine sequence (**Figure 1**) However, the cerebral circuit of the coarse-to-fine perception of scenes has never been investigated in humans. On the basis of neurophysiological recordings in nonhuman primates, Bullier (2001) suggested that during perception of a scene, LSF which are conveyed more rapidly than HSF by fast magnocellular pathways, access the occipital cortex and high-order cortical areas in the dorsal cortical stream (parietal and frontal cortices) allow a coarse perceptual parsing of the visual input, prior to their complete propagation along the ventral cortical stream (inferotemporal cortex) which ultimately mediates the input recognition. The initial low-pass analysis would serve to refine the subsequent processing of HSF, conveyed more slowly by parvocellular pathways through the ventral cortical stream. It is now essential to match data from non human primates with human data. Finally, the ventral visual stream contains a mosaic of different areas that respond selectively to different categories of visual stimuli (Haxby et al., 2001; Lerner et al., 2001; Spiridon and Kanwisher, 2002). While several studies (Epstein and Kanwisher, 1998; Bar and Aminoff, 2003; Bar, 2004; Epstein, 2005, 2008; Aminoff et al., 2007; Epstein and Higgins, 2007; Dilks et al., 2013) agree that a prominent region in the inferotemporal cortex known as the parahippocampal place area (PPA), the retrosplenial cortex (RSC) and a region around the transverse occipital sulcus called the occipital place area (OPA) all play a major role in the perception of scenes in humans, the specific functions supported by scene-selective regions during the spatial frequency processing in scenes remain unclear.

The present review addresses these issues and aims to identify the cerebral regions differentially involved in low and high spatial frequency processing and to clarify their attributes during scene perception.

#### **NEURAL CORRELATES OF SPATIAL FREQUENCY PROCESSING DURING SCENE PERCEPTION**

Many authors postulate that the two cerebral hemispheres are differently involved in spatial frequency processing, the right hemisphere predominating in the processing of LSF, and the left hemisphere predominating in the processing of HSF. Cerebral

asymmetries have been observed in behavioral studies conducted on healthy participants (Sergent, 1982, 1983; Sergent and Hellige, 1986; Kitterle et al., 1990, 1992; Chokron et al., 2003; Peyrin et al., 2003), in neurological patients (Robertson et al., 1988; Lamb et al., 1990; Robertson and Lamb, 1991; Peyrin et al., 2006; Dos Santos et al., 2013), and from functional neuroimaging studies (Fink et al., 1996, 2000; Martinez et al., 1997, 2001; Heinze et al., 1998; Kenemans et al., 2000; Mangun et al., 2000; Yamaguchi et al., 2000; Wilkinson et al., 2001; Han et al., 2002;Iidaka et al., 2004; Lux et al., 2004; Peyrin et al., 2004; Weissman and Woldorff, 2005; Musel et al., 2013). However, the hemispheric specialization for spatial frequency processing was largely inferred from studies assessing cerebral asymmetries during the processing of global and local information.

#### **PSYCHOPHYSICAL ARGUMENTS FOR HEMISPHERIC SPECIALIZATION IN SPATIAL FREQUENCY PROCESSING**

The first experimental evidence in support of hemispheric specialization for global and local processing comes from psychophysical studies using hierarchical forms as visual stimuli (i.e., in general a large global letter made up of small local letters; Navon, 1977; Kinchla and Wolfe, 1979; **Figure 2A**). Using hierarchical visual stimuli displayed in either the left visual field (projecting directly to the right hemisphere) or the right visual field (projecting directly to the left hemisphere), Sergent (1982) demonstrated that the identification of the global letter was faster when displayed in the left visual hemifield/right hemisphere, and that the identification of local letters occurred more rapidly when they were displayed in the right visual hemifield/ left hemisphere. These results suggest a right hemispheric specialization for the processing of global information, and a left hemispheric specialization for the processing of local information. Based on evidence that global information is predominantly conveyed by LSF, and that local information is predominantly conveyed by HSF (Schulman et al., 1986; Badcock et al., 1990; Lamb and Yund, 1993), the cerebral asymmetries observed during global and local processing have been interpreted as reflecting the hemispheric specialization for LSF and HSF processing, respectively (Sergent, 1982).

However, the relationship between global and local information, and LSF and HSF, respectively, is far from univocal within hierarchical visual forms (Palmer, 1993). It is for example possible that global information is conveyed by both LSF and HSF. Hemispheric specialization for spatial frequency processing was therefore subsequently tested by directly manipulating the spatial

HSF grating identification and HSF categorization; the right hemisphere (RH) is predominantly involved in the global letter identification, LSF grating identification and LSF categorization. Activations reported showed stronger activation in the left than the right occipital cortex for HSF categorization [(HSF unflip > flip) contrast] and stronger activation in the right than the left occipital cortex for LSF categorization [(LSF unflip > flip) contrast]. Figure adapted from Peyrin et al. (2004).

frequency content of visual stimuli, using either sinusoidal gratings (Kitterle et al., 1990, 1992; Kitterle and Selig, 1991; **Figure 2A**) or scene images (Peyrin et al., 2003, 2006; **Figure 2A**). It should be noted that this type of manipulation is not feasible with hierarchical forms because low-pass filtering cancels out the local form and renders the task impossible. We evaluated hemispheric asymmetry in healthy participants in a series of psychophysical studies (Peyrin et al., 2003, 2006), by making explicit changes in the spatial frequency spectrum of scene images, which were displayed in either the left or the right visual fields. In the initial study, participants were asked to recognize a target scene (a city or a highway) filtered in either LSF or HSF (Peyrin et al., 2003). Results showed more rapid recognition of LSF scenes when they were displayed in the left visual hemifield/right hemisphere than when they were presented in the right visual field/left hemisphere. Conversely, recognition of HSF scenes occurred more rapidly in the right visual hemifield/left hemisphere than the left visual hemifield/right hemisphere. This study demonstrated a right hemispheric predominance for LSF and a left hemispheric predominance for HSF processing. It should be noted that the hemispheric specialization in question has been observed in males, but not in females (Peyrin et al., 2006). These results are consistent with studies showing a lesser degree of lateralization in female functional cerebral organization compared to males (McGlone and Kertesz, 1973; Voyer, 1996). Certain factors of interference, which may affect processing speed may render detection of hemispheric specialization in healthy females more difficult. For example, the hormonal level fluctuations over the menstrual cycle has been evidenced to modulate hemispheric asymmetries in visual, attentional, and language processes (Hausmann and Güntürkün, 2000; Hausmann et al., 2002; Hausmann, 2005), and to affect interhemispheric transfer time (Hausmann et al., 2013).

#### **NEURAL CORRELATES OF HEMISPHERIC SPECIALIZATION IN SPATIAL FREQUENCY PROCESSING**

Neuropsychological and neuroimaging studies that use hierarchical visual forms provide conflicting evidence on which cortical structures present hemispheric specialization. Robertson et al. (1988) showed impairment in the performance of tasks involving the perception of hierarchical visual form in patients with unilateral damage to the temporo-parietal junction. Performance of patients with a lesion situated in the left superior temporal cortex was impaired during the identification of local elements, whereas patients suffering from lesions in the right temporo-parietal junction exhibited poor performance during the identification of the global form. These data suggests the right temporo-parietal junction specialization for global processing, and the left temporoparietal junction specialization for local processing. However, using positron emission tomography, Fink et al. (1996; see also Fink et al., 1997, 2000) reported cerebral asymmetries at a lower level of visual cortical processing, with a right lingual gyrus activation during the identification of the global form and a left inferior occipital gyrus activation during the identification of local elements. Using event-related brain potentials (ERPs), Heinze et al. (1998; see also Mangun et al., 2000) failed to show hemispheric specialization in the first-stage of the visual analysis. Instead, their results show long latency asymmetries (260–360 latency range) for global and local processing, suggesting that cerebral asymmetries was rather present at the higher-stage of the visual analysis. Some functional imaging data have, furthermore, revealed an attentional cortical mechanism located in the temporo-parietal junction which controls the attentional selection of information presented either at global or the local level depending on the visual task demands (Robertson et al., 1988; Robertson and Lamb, 1991; Fink et al., 1996; Yamaguchi et al., 2000; Wilkinson et al., 2001; Weissman and Woldorff, 2005). For example, Yamaguchi et al. (2000) recorded ERPs while participants shifted their attention to the global or local level of hierarchical visual forms (the shift direction was controlled by a cue preceding the stimulus). Cerebral asymmetries were observed during the global and local processing of hierarchical forms, but also during the time interval of attention directed toward global or local levels by the cues. ERP responses indicated greater right-hemisphere amplitudes located in the right temporo-parietal junction when attention was directed at global level, and greater left-hemisphere amplitudes located in the left temporo-parietal junction when it was directed at local level. This study provided a neural basis for a "top-down" mechanism of allocation of attention to global and local information, and revealed the asymmetrical involvement of the temporal-parietal regions.

Neuroimaging studies previously mentioned have provided conflicting results concerning hemispheric specialization for spatial frequency processing using hierarchical visual forms as stimuli. Subsequent studies, including those of our own team, which involved the direct manipulation of the spatial frequency content of stimuli, provided evidence of hemispheric specialization involving occipital areas (Iidaka et al., 2004; Peyrin et al., 2004). In an fMRI study, Peyrin et al. (2004) investigated the hemispheric specialization for spatial frequency processing during the recognition of LSF and HSF scenes (city vs. highway scenes at a visual angle of 4◦). Comparison of LSF to HSF scene recognition, revealed significant activation in regions which are known to be involved in scene processing: the right anterior temporal region which is particularly sensitive to familiar versus unfamiliar scenes (Nakamura et al., 2000), and the right parahippocampal gyrus which is known to be involved in tasks requiring the retrieval of topographical information in scenes (Maguire et al., 1998; it should be noted that right-side parahippocampal gyrus activation did not correspond to PPA activation reported by Epstein and Kanwisher,1998). These results suggest that in Peyrin et al. (2004), scene perception was based mainly on LSF extraction and analysis, and they support the models proposing the prevalence of LSF information in scene categorization (coarse-to-fine strategy; Schyns and Oliva, 1994). Significant activation also occurred in the right inferior parietal lobule near the temporo-parietal junction. This activation was interpreted as reflecting an attentional control mechanism during spatial frequency selection. Yamaguchi et al. (2000) had previously shown cerebral activity in the right temporo-parietal area for a global attention shift during the perception of hierarchical letter forms (i.e., allocation of attention to global information). Finally, LSF scene recognition (as opposed to HSF) activated the superior temporal cortex bilaterally. This particular result concerned us, because it contradicted neuropsychological studies (Robertson et al., 1988; Lamb et al., 1990; Robertson and Lamb, 1991), showing specialization of the right superior temporal cortex in the perceptual processing of global information (supposed to be preferentially conveyed by LSF), and specialization of the left superior temporal cortex in the perceptual processing of local information (supposed to be preferentially conveyed by HSF). It should be noted that HSF scene recognition (as opposed to LSF) failed to show significant activation, suggesting a processing bias toward LSF.

Based on behavioral studies in which performances between the two visual hemifields are directly compared (see our abovementioned original psychophysical experiments; Sergent, 1982; Kitterle et al., 1990, 1992; Kitterle and Selig, 1991; Peyrin et al., 2003, 2006), we suggested to directly compare activation between the two hemispheres in order to assess cerebral asymmetries in fMRI study. For this purpose, we created an fMRI method of direct inter-hemispheric comparison. Two sets of functional volumes, obtained from functional scans, are compared at individual level. One set is represented by functional volumes in accordance with neurological convention (the left hemisphere appears on the left side of images) and the other set is represented by the same functional volumes this time in accordance with radiological convention (the right hemisphere appears on the left side of images). Images from the second set are "flipped" by 180◦ in the midsagital plane, thus providing "mirror" images of the first set. Contrasts between "unflipped" and "left-right flipped" functional volumes from the same experimental condition allow to compare activity in one hemisphere with activity in homologous regions of the other hemisphere (Iidaka et al., 2004; Peyrin et al., 2004, 2005; Musel et al., 2013; see also Cousin et al., 2006 for an application of this method on language processes; **Figure 3**). This method revealed greater activation in the right than the left middle occipital gyrus for LSF scene recognition, and greater activation in the left than the right middle occipital gyrus for HSF scene

recognition (**Figure 2B**). This study provided new evidence for hemispheric specialization at the first cortical level of visual analysis. Analyzing fMRI data with a more traditional approach which contrasts spatial frequencies to one another, we observed a higher degree of activation for LSF scenes (as opposed to HSF), while the reverse contrast did not reveal any significant activation. This study suggests that the results considerably differ according to the method applied to analysis fMRI data. Inter-hemispheric comparison seems more appropriate for the investigations of cerebral asymmetries, since it allows any main effect deriving from spatial frequency bias to be canceled out.

We proceeded to investigate the role of the occipital cortex in spatial frequency processing using a neuropsychological approach (Peyrin et al., 2006). We studied the categorization of LSF and HSF scenes in a female neurological patient who suffered from a focal lesion in the right occipito-temporal cortex following the embolization of an arterioveinous malformation. This lesion had induced a left homonymous hemianopsia. Two evaluations were conducted, the first 1 week prior to surgical intervention and the second 6 months afterward. As expected, the performance of the patient was more severely impaired for LSF than HSF scene recognition following embolization. This result suggests again the right occipital cortex specialization for LSF, and on a more general level suggests that hemispheric specialization could occur in women, although this is difficult to demonstrate behaviorally in the healthy population. This finding highlights the necessity of studying males and females together and both normal and brain-damaged patients' performance in order to establish the neural correlates of visual functions.

right side.

The extent to which hemispheric asymmetries during spatial frequency processing result from perceptual or attentional processes remains to be determined. While some studies have clearly demonstrated that attentional processes exert control on hemispheric specialization in the processing of global and local information at high-level stages of visual processing (e.g., via the temporo-parietal junction; Robertson et al., 1988; Robertson and Lamb, 1991; Fink et al., 1996; Heinze et al., 1998; Yamaguchi et al., 2000; Wilkinson et al., 2001; Weissman and Woldorff, 2005), other studies have evidenced hemispheric asymmetries at lowerlevel stages, in the occipital cortex (Fink et al., 1997, 2000; Peyrin et al., 2004; Musel et al., 2013). However, activation of the occipital cortex was frequently associated with activation of the temporoparietal junction in these studies. This cortical structure may have exerted attentional influence on lower-level areas. Furthermore, a number of neuroimaging studies have evidenced attentional modulation of activity in early visual areas (Tootell et al., 1998; Watanabe et al., 1998; Brefczynski and DeYoe, 1999; Gandhi et al., 1999; Martinez et al., 1999; Sasaki et al., 2001; Silver et al., 2007; Saygin and Sereno, 2008). For example, Martinez et al. (1999) showed that attending to a target whose location was cued by an arrow enhanced the amplitude of activation in striate and extrastriate visual areas. Cerebral asymmetries observed at low-level stages of visual processing, such as the occipital cortex, may not, therefore, necessarily result from strictly perceptual processes.

However, despite the considerable body of research in favor of the hemispheric specialization for spatial frequency processing in the occipital cortex, other authors postulate that a spatialfrequency processing mapping according to the retinotopic organization of the visual cortex.

#### **RETINOTOPIC PROCESSING OF SPATIAL FREQUENCIES**

Imaging data obtained from patients with cerebral lesions (Holmes, 1918; Horton and Hoyt, 1991) and from healthy participants (Engel et al., 1994, 1997) show that the human primary visual cortex is retinotopically organized. The central (foveal) part of the visual field is represented at the very back of the visual cortex and more peripheral regions of the visual field are represented further forward (**Figure 4A**). Importantly, the distribution of retinal photoreceptors and retinal ganglion cells is nonhomogeneous throughout the retina (Curcio and Allen, 1990; Curcio et al., 1990). The density of cones and midget ganglion cells from which the parvocellular pathway originates and which are used to process HSF information, is greatest in the fovea, while the density of rods and parasol ganglion cells from which the magnocellular pathway originates and which are used to process LSF information, increases with foveal eccentricity. Therefore HSF information could be predominantly processed in the areas dedicated to foveal vision. Similarly, LSF information might well be predominantly processed in the areas devoted to peripheral vision.

Neurophysiological studies performed on cats (Everson et al., 1998; Issa et al., 2000), primates (De Valois et al., 1982a; Foster et al., 1985; Tootell et al., 1988; Gegenfurtner et al., 1997; Xu et al., 2007) and humans (Singh et al., 2000; Sasaki et al., 2001; Henriksson et al., 2007) have mapped the representation of spatial frequencies in retinotopic areas. In an fMRI study, using retinotopic encoding with achromatic sinusoidal gratings, Sasaki et al.

**FIGURE 4 | (A)** Retinotopic mapping of the visual field on the visual cortex. The central (foveal) part of the visual field is represented at the very back of the visual cortex and laterally. More peripheral regions of the visual field are represented further forward in the medial part of the visual cortex. **(B)** Retinotopic organization of spatial frequency processing during scene perception: LSF [as opposed to HSF, (LSF > HSF) contrast] scene categorization recruits areas dedicated to peripheral vision, while HSF [as opposed to LSF, (LSF > HSF) contrast] scene categorization recruits areas dedicated to foveal vision. Figure adapted from Musel et al. (2013).

(2001)showed that LSF were mapped on the peripheral visual field representation of the occipital cortex, whereas HSF were mapped on the central visual field representation. More recently, Henriksson et al. (2007) evidenced that in the retinotopic area of the occipital cortex, lower spatial frequencies selectivity was observed as eccentricity of the achromatic sinusoidal grating increased. Other studies have provided evidence of consistent cortical retinotopic mapping of more complex cognitive functions, such as visual spatial attention (Tootell et al., 1998; Watanabe et al., 1998; Brefczynski and DeYoe,1999; Gandhi et al.,1999; Martinez et al.,1999; Sasaki et al., 2001; Silver et al., 2007) and working memory (Pratte and Tong, 2014) in the early visual areas, as well as in the higher cortical areas, such as in the temporal, parietal, and frontal cortices (Silver et al., 2005; Hagler and Sereno, 2006; Wandell et al., 2007; Saygin and Sereno, 2008;Arcaro et al., 2009, 2011; Sheremata et al., 2010). Tootell et al. (1998), for example, showed that paying attention to a specific location in the visual field increased activity in the corresponding retinotopic location of the extrastriate visual areas. Attentional modulation which was similar, albeit to a lesser degree, was also observed in the primary visual cortex (V1). Saygin and Sereno (2008) subsequently investigated the independent modulation of retinotopic responses by visual stimulus properties and attention in a number of areas exhibiting retinotopic organization (in the occipital cortex, the precuneus, the motion-sensitive temporal cortex, the intraparietal sulcus, and the frontal eye fields in the frontal cortex). These authors used retinotopically rotating polar angle mapping with point-light biological motion figures as complex visual stimuli. Participants fixated and viewed a rotating pie-shaped wedge containing biological motion figures. In the background, biological motion figures were either surrounded by either scrambled figures (stimulus contrast) or similar figures (no stimulus contrast). Participants were asked to perform one of two tasks while fixating – they were asked to attend to either the wedge

(attention) or to the center of gaze (no attention). The authors demonstrated that the retinotopy of early visual areas was mainly driven by visual stimuli contrast, that the retinotopy of classical attentional control areas in the parietal and frontal cortices was mainly driven by attention, and that the retinotopy of lateral temporal regions was driven by both. In a recent study, Bressler et al. (2013) measured the effects of endogenous visual spatial attention (i.e., attention directed voluntarily by the participant) on the amplitude of retinotopic responses in the occipital and parietal cortices. Participants were asked to direct their attention toward a target of different eccentricities and to detect a target during retinotopic mapping. The authors showed that attending to the target in the visual field enhanced the amplitude of activations in corresponding retinotopic cortical locations for all the areas investigated, but that the modulation of retinotopic responses depended on target eccentricity. In occipital areas (V1, V2, V3, and hV4), directed attention elicited greater activation in cortical locations which corresponded to target eccentricities closer to the center than those which were farther out. Conversely, in parietal areas, directed attention elicited greater activation in target eccentricities which were farther away than in those which were closer. The authors suggest that endogenous attention potentially plays a role in processing the fine details of an object in central vision and in detecting relevant objects in the periphery during motor planning. Interestingly, Sasaki et al. (2001) provided direct evidence of retinotopic modulation of response resulting from global and local attentional demands in the occipital cortex. The authors used very large hierarchical arithmetic symbols (for example, a global "x" form composed of several local "+" elements). During "attention to global" periods, participants focused their attention on the global symbol (the "x") involving their peripheral vision, and during "attention to local" periods, they were instructed to focus attention on local symbols (the "+"), involving foveal vision. Results showed that when attention was directed at local (as opposed to global) level, activation occurred in the visual areas in relation to the foveal representation. When attention was directed at global (as opposed to local) level, activation was consistent with peripheral cortical representation. Since it can be assumed that global processing is mediated by low-pass spatial analysis, and local processing is mediated by high-pass spatial analysis (Schulman et al., 1986; Badcock et al., 1990; Lamb and Yund, 1993), the retinotopic organization observed in global and local attentional processing may constitute an argument in favor of a retinotopic organization for the attentional selection of spatial frequencies.

On the whole, the neuroimaging studies mentioned previously either highlight retinotopic mapping of spatial frequency processing (Sasaki et al., 2001), or reveal hemispheric specialization for spatial frequency processing (Iidaka et al., 2004; Peyrin et al., 2004). A recent fMRI study showed that spatial frequency processing could be both retinotopically mapped and lateralized between the two hemispheres (Musel et al., 2013).

#### **RETINOTOPIC AND LATERALIZED PROCESSING OF SPATIAL FREQUENCIES DURING SCENE CATEGORIZATION**

After demonstrating retinotopic organization of spatial frequency processing, Sasaki et al. (2001) concluded that neither global nor

local processing was lateralized in the occipital cortex. However, the authors compared activation elicited by global and local conditions to one another (traditional method of fMRI data analysis), rather than activation between hemispheres (direct inter-hemispheric comparison method used in Peyrin et al., 2004). Musel et al. (2013) evaluated both the retinotopy and the functional lateralization of spatial frequency processing using a categorization task of scenes (indoors vs. outdoors) filtered in HSF and LSF. They used larger scene images (with a visual angle of 24◦ × 18◦) than in Peyrin et al. (2004) in which the visual angle was 4◦ × 4◦, thus covering the same breadth of visual field as Sasaki et al. (2001). Results provided firstly evidence of retinotopic processing of spatial frequencies. At group level, the comparison between the spatial frequency content revealed that LSF scene categorization (as opposed to HSF) elicited activation in the anterior half of the calcarine fissures linked to the peripheral visual field, whereas HSF scene categorization (as opposed to LSF) elicited activation in the posterior part of the occipital lobes which are linked to the fovea, according to the retinotopic property of visual areas (**Figure 4B**). The retinotopic organization of spatial frequencies was also assessed at individual level by projecting LSF and HSF related activations onto retinotopic maps established for a number of participants. Functional activations projected onto individual retinotopic maps revealed that LSF processing is mapped in the anterior part of V1, whereas HSF processing is mapped in the posterior and ventral part of V2, V3, and V4. Furthermore, at the group level, the direct inter-hemispheric comparisons performed on the same fMRI data revealed a right-sided occipito-temporal predominance for LSF scene categorization and a left-sided temporal cortex predominance for HSF scene categorization, according to the hemispheric specialization theories. By using suitable method of fMRI analysis on the same data, as well as visual stimuli filtered in spatial frequencies covering a large part of the visual field, Musel et al. (2013) demonstrated for the first time retinotopic and lateralized spatial frequency processing in the human occipito-temporal cortex. It should be noted that hemispheric asymmetries were also highlighted within retinotopically defined parietal and frontal cortices during spatial working memory tasks (Sheremata et al., 2010; Szczepanski et al., 2010; Szczepanski and Kastner, 2013).

However, results from certain neurophysiological, computational, and behavioral studies indicate that the totality of spatial frequency information is not immediately conveyed through the brain, but that analysis follows a predominantly coarse-tofine processing sequence. If LSF extraction and analysis occurs first, followed by that of HSF, why should there be any hemispheric lateralization for the processing of LSF or HSF? Identification of the neural basis of the coarse-to-fine analysis in scene perception is the first step toward responding to this question.

#### **COARSE-TO-FINE PROCESSING DURING SCENE PERCEPTION PSYCHOPHYSICAL ARGUMENTS OF COARSE-TO-FINE PROCESSING**

Data from the functional neuroanatomy of magnocellular and parvocellular visual pathways indicate that the totality of visual information is not conveyed immediately, but that LSF reach the visual cortex before HSF (Van Essen and Deyoe, 1995; Bullier, 2001), although some controversies still remain (Merigan and Maunsell, 1993; Kaplan, 2004). A temporal precedence of LSF processing over HSF has been observed in psychophysical studies using sinusoidal gratings (Breitmeyer, 1975; Ginsburg, 1986; Hughes et al., 1996). Studies manipulating spatial frequency content of faces and scenes have provided further evidence of a coarse-to-fine processing sequence (Schyns and Oliva, 1994, 1997, 1999; Parker et al., 1996; Oliva and Schyns, 1997; Musel et al., 2012). Schyns and Oliva (1994) used hybrid images made of two superimposed scenes belonging to different categories and containing different spatial frequency bands (e.g., a city scene in LSF superimposed on a highway scene in HSF). When presentation time of hybrids was very short (30 ms), categorization of the hybrid was dominated by LSF information. However, categorization was dominated by HSF information for longer presentation times (150 ms). This suggests that LSF take precedence over HSF during scene perception. Furthermore, when the authors displayed two successive hybrids depicting simultaneously a coarse-to-fine sequence for a given scene (a LSF city in the first hybrid follows by a HSF city in the second hybrid) and a fine-to-coarse sequence for another scene (a HSF highway in the first hybrid follows by a LSF highway in the second hybrid), scene categorization was more frequently based on the coarse-to-fine than the fine-to-coarse sequence.

Although LSF information may be perceptually available before HSF, it is important to note that it does not necessarily follow that it is always used first to support visual recognition in all tasks. In Schyns and Oliva (1994), scene categorization in hybrid sequences was in fact based on a fine-to-coarse rather than a coarse-to-fine sequence in a substantial proportion of sequences (29%). Despite the apparent predominance of coarse-to-fine processing, certain flexibility in the processing sequence of spatial scale information has emerged, and it has also been seen to be sensitive to the demands of the task or the visual characteristics available in the stimuli (Parker et al., 1996; Schyns and Oliva, 1997, 1999; Morrison and Schyns, 2001; Mermillod et al., 2005; Ozgen et al., 2005, 2006; Rotshtein et al., 2010; Awasthi et al., 2013). A study by Schyns and Oliva (1999) showed that it was possible to constrain the spatial frequency band preferentially processed in hybrids by imposing a sensitization phase which implicitly "primes" visual processing in favor of a particular scale (coarse or fine). When participants were initially exposed to LSF information, subsequent categorization of hybrid images was preferentially performed following LSF cues, whereas it was biased toward HSF information after priming by HSF. The use of hybrid faces allowed Schyns and Oliva (1999) to show preferential recourse to HSF information to determine whether a face was expressive or not, and preferential recourse to LSF information to determine the nature of the emotion (e.g., happy, angry). It is therefore possible that the demands of a visual task determine which scale must be processed in hybrids (even using very short presentation). Overall, these studies suggest that all spatial frequencies are available at the beginning of categorization, and that their selection may depend on interactions between the perceptual information available and the demands of a given visual task.

Importantly, results from Schyns and Oliva (1994) studies suggest that coarse-to-fine processing constitutes a predominant and default strategy that seems advantageous for scene recognition (in the absence of task demands which constrain the use of a particular spatial frequency band). A recent study also evidenced a coarseto-fine preference in the very early stages of development, in 7- to 8-months-old infants (Otsuka et al., 2014). Furthermore, a considerable number of recent studies have provided behavioral evidence of anLSF-based processing during rapid scene recognition (Kihara and Takeda, 2010; De Cesarei and Loftus, 2011; Musel et al., 2012; Mu and Li, 2013) and object categorization (Loftus and Harley, 2004). Using dynamic scenes composed of six filtered images of the same scene, from LSF to HSF or from HSF to LSF, allowing to experimentally mimic a coarse-to-fine or a reverse fine-to-coarse sequence, Musel et al. (2012)showed that coarse-to-fine sequences were categorized more rapidly than fine-to-coarse sequences in young adults. This provided new arguments in favor of a predominantly coarse-to-fine categorization of natural scenes, and a new experimental tool which imposes a coarse-to-fine processing and allows investigations of the neural substrates of coarse-to-fine processing.

#### **NEURAL BASIS OF COARSE-TO-FINE ANALYSIS**

We do not as yet know exactly how and where in the brain LSF and HSF information is differentially analyzed and eventually merged during visual processing. Traditional models generally maintain that incoming visual cues are combined at successive stages along the cortical hierarchy (Biederman, 1995; Riesenhuber and Poggio, 1999), and suggest that LSF and HSF converge only in higherlevel visual areas of the inferior temporal cortex (such as the fusiform or parahippocampal cortex; Bar et al., 2006). However, drawing on evidence obtained from neurophysiological recordings in nonhuman primates (Hupe et al., 2001), Bullier (2001) postulated that a rapid LSF analysis takes place predominantly in the dorsal cortical stream. Information is then sent-back through feedback signals into low-level areas (e.g., the primary visual cortex, V1), where it influences subsequent slower HSF analysis and guides subsequent processing through the ventral cortical stream. The occipital cortex might therefore serve as an "active blackboard" integrating computations made by higher-order cortical areas.

Bar et al. (2006)later investigated the neural correlates and time course of spatial frequency processing during object recognition in a combined fMRI and MEG study. They found evidence that stimuli containing LSF information elicited rapid activation in the orbitofrontal cortex, 50 ms before the involvement of recognitionrelated areas in the temporal cortex (fusiform gyrus). Activation of the orbitofrontal cortex was not observed with stimuli containing only HSF information. These authors suggested that the orbitofrontal cortex – mediated by LSF information – acts as the trigger of top-down facilitation during object recognition. Using dynamic causal modeling to investigate the interaction between the orbitofrontal cortex and the fusiform gyrus during the perception of LSF and HSF objects, Kveraga et al. (2007) showed reciprocal connections between these two cortical structures, with LSF modulating feedback connections from the orbitofrontal cortex to thefusiform gyrus. LSF may therefore reach the orbitofrontal

cortex rapidly, in order to trigger plausible interpretations of any given visual input. The result of these computations would then be projected, via feedback connections, to the fusiform gyrus, and would guide subsequent analysis of HSF information. It is worth noting that in a recent study, Patai et al. (2013) presented LSF or HSF scenes as memory-cues (i.e., contextual information) and then asked participants to detect a target (e.g., an object) in the non-filtered version of the cued scene. These authors evidenced that LSF and HSF memory-cues were equally effective as triggers of contextual memory information, and facilitated target detection. This challenges Bar's proposal of LSF-based facilitation in object recognition. However, their target detection task may have involved fine-grained perception, thus favoring HSF processing.

However, to date, the neural architecture and temporal dynamics of such top-down mechanisms have never been systematically investigated via direct testing of the preferential coarse-to-fine processing sequence during visual scene perception in humans. Peyrin et al. (2010) combined fMRI and ERPs on the same participants to identify the neural substrates underlying the coarseto-fine processing sequence. To constrain the order of spatial frequency processing, the authors displayed sequences of two spatial frequency-filtered scenes in rapid succession, with either a coarse-to-fine sequence (LSF scene followed by a HSF scene), or a fine-to-coarse sequence (HSF scene followed by an LSF scene). Participants' task was to decide whether the two scenes belonged to a same category (city, beach, or indoor). FMRI examination revealed selective increased activation in early stage occipital areas, and in frontal and tempo-parietal areas for coarse-to-fine sequences (compared to fine-to-coarse sequences). ERP topography and source analyses revealed a similar cortical network, but could additionally determine the time-course of activation in these areas. Frontal and temporo-parietal areas responded more to LSF scenes when these were presented first, whereas the occipital areas responded more to HSF scenes when these were presented after LSF scenes. More specifically, results demonstrated that low-pass signals (conveyed by fast magnocellular pathways) could rapidly activate high-order areas, providing semantic information (via the left prefrontal cortex and temporal areas) and spatial information (via the frontal eye fields), as well as attentional controls (via the temporo-parietal junction), all of which may promote the ongoing categorization and perceptual organization of the scene. This lowpass or coarse analysis is perhaps refined by further processing of high-pass signals (conveyed more slowly by the parvocellular pathways). To enable this, feedback from the low-pass analysis, which take place in frontal and temporo-parietal areas, might be sent back into lower level visual areas, such as the primary visual cortex, and would then guide the high-pass analysis and assist in the selection of the relevant finer details necessary for the recognition and categorization of scenes. These results are consistent with the LSF-based top-down facilitation of recognition, as proposed by Bar et al. (2006; see also Bar, 2003) in the context of object recognition, with the exception of the cortical site for feedback projections (occipital cortex in Peyrin et al., 2010; fusiform gyrus in Bar et al., 2006).

The influential models of visual perception assume a predominantly coarse-to-fine sequence of spatial frequency processing in the whole brain, based on the functional properties of the visual pathways. However, as mentioned previously, many studies have also shown that it is possible that the two hemispheres of the human brain may complement one another in the processing of LSF and HSF. The critical issue here is how to reconcile hemispheric specialization of spatial frequency processing with coarse-to-fine analysis of scenes.

#### **CEREBRAL ASYMMETRIES FOR COARSE-TO-FINE PROCESSING**

The hemispheric specialization observed for spatial frequency processing raise the crucial question of the legitimacy of suggesting that coarse-to-fine sequencing is applied throughout brain. Peyrin et al. (2005) conducted an fMRI experiment in order to investigate whether coarse-to-fine processing predominates in only one hemisphere. They displayed sequences of two spatial frequencyfiltered scenes in rapid succession, with either a coarse-to-fine sequence (LSF scene followed by HSF scene), or a fine-to-coarse sequence (HSF scene followed by LSF scene). Participants' task was to decide whether the two scenes belonged to a same category (city, beach, or indoor). Cerebral asymmetries were identified using inter-hemispheric method of comparison (i.e., contrast between "unflipped" and "left-right flipped" functional images for each sequence). Results showed greater activation in the right than the left occipito-temporal cortex for the coarse-to-fine sequence, and greater activation in the left than the right occipito-temporal cortex for the fine-to-coarse sequence. These fMRI results suggest that the initial spatial frequency-band appearing in the sequence could determine which of the two hemispheres is preferentially involved in the sequential processing of spatial frequencies. According to input sequences or task demands, the right occipital cortex would give priority to LSF analysis for a coarse-to-fine processing and the left occipital cortex would give priority to HSF analysis for a fine-to-coarse analysis.

As far as the higher-level stages of visual scene processing are concerned, several studies have highlighted the sensitivity of sceneselective areas to low-level features, such as spatial frequencies and amplitude spectrum properties, in scenes (Andrews et al., 2010; Rajimehr et al., 2011; Zeidman et al., 2012). However, we still lack evidence of coarse-to-fine processing within the scene-selective cortical regions.

#### **SPATIAL FREQUENCY PROCESSING WITHIN SCENE-SELECTIVE AREAS**

There is considerable evidence suggesting that the occipitotemporal cortex contains a mosaic of different areas that respond selectively to different category of stimuli (Haxby et al., 2001; Lerner et al., 2001; Spiridon and Kanwisher, 2002). More specifically, three regions were evidenced as scene-selective regions: the PPA, the RSC, and the OPA. These regions are known to be involved in high-level functions such as navigation (Epstein et al., 2007; Vass and Epstein, 2013), spatial layout processing and scene recognition (Epstein and Kanwisher, 1998; Epstein et al., 1999, 2003; Epstein, 2005, 2008; Epstein and Higgins, 2007; Epstein and Ward, 2010; Dilks et al., 2013), and contextual associations (Bar and Aminoff, 2003; Bar, 2004; Aminoff et al., 2007; Bar et al., 2008a,b). However, only a few studies investigated whether these regions are sensitive to scenes low-level properties such as spatial frequencies. For example, Peyrin et al. (2004) showed that

the parahipopcampal gyrus was more strongly activated by LSF than HSF scenes. Conversely, Rajimehr et al. (2011) observed that in human and macaques, the PPA responded more strongly to HSF than LSF stimuli. This was also the main findings of Zeidman et al. (2012). In their study, they depicted three-dimensional spaces by positioning small dots following an exponential distribution and filtered them in either LSF or HSF. They showed stronger activation of the PPA when participants had to detect de disappearance of a small proportion of dots in HSF than LSF spaces. It should be noted that these studies differed in many methodological aspects such as the task demands or the duration of stimuli, that may have influenced spatial frequency selectivity within the PPA. However, whether coarse-to-fine processing of scenes occurs within scene-selective regions is still unclear.

Coarse-to-fine processing of faces in high level visual cortex was the central focus of a recent study by Goffaux et al. (2011) who showed an intriguing effect of spatial frequencies in a face-selective region, the fusiform face area (FFA; Kanwisher et al., 1997). By manipulating duration of exposure and the spatial frequency content of faces, these authors observed higher levels of FFA response to LSF when duration of exposure to faces was short, and higher levels of response to HSF for longer exposure durations. These results suggest that coarse-to-fine processing is the predominant strategy in the most prominent regions of the ventral visual stream (inferotemporal cortex). In an evoked potential study, Schettino et al. (2011) used sequences of filtered scenes (with blank screens occurring between scenes) in order to investigate the neural correlates of the accumulation of visual information during object recognition and the time course of these correlates. For this purpose, the authors used sequences in which the first scene was always in LSF and the scene was gradually revealed in six successive images by progressively adding HSF information. The authors observed that activation in the parahippocampal cortex decreases when the spatial frequency content of scenes increases, suggesting that this region is sensitive to the primary processing of LSF information, even if this study did not investigate explicit coarse-to-fine processing of scenes.

A recent fMRI study (Musel et al., 2014) tested whether such processing occurs in three scene-selective cortical regions: the PPA, the RSC, and the OPA. We measured activation in these scenepreferring regions during the categorization of dynamic scene stimuli (Musel et al., 2012). Dynamic scenes were composed of six filtered images of the same scene, from LSF to HSF or from HSF to LSF, allowing us to mimic either a coarse-to-fine or a fine-tocoarse sequence. We first identified scene-selective regions using a localizer adapted from previous studies (Epstein and Kanwisher, 1998; Epstein et al., 2003; Bar et al., 2008b; Walther et al., 2009) in which participants viewed gray-scale photographs of scenes, faces and common objects. The contrast between scenes and other categories was intended to enable localization of the regions involved in the perception of scenes. Once localized, we compared activation elicited by coarse-to-fine and fine-to-coarse dynamic scenes within the areas defined as the PPA, RSC, and OPA. Results showed greater activation of only the PPA for coarse-to-fine compared to fine-to-coarse sequences (**Figure 5**). Equivalent activations were observed for both types of sequence in the RSC and OPA. This

study therefore suggests that coarse-to-fine sequence processing constitutes the predominant strategy for scene categorization in the PPA. It should be noted that evidence of spatial frequency sensitivity within other scene-selective areas, such as the RSC and the OPA, is still lacking.

## **CONCLUSION**

The present review aimed to identify cerebral regions differentially involved in low and high spatial frequency processing and to clarify their attributes during scene perception. Several neuroimaging studies suggest that spatial frequency processing could be retinotopically mapped and lateralized in both hemispheres. Right occipital areas are more activated than the left ones during the processing of LSF scenes, while left occipital areas are more activated than the right ones during the processing of HSF scenes. Concomitantly, the processing of HSF scenes (as opposed to LSF) activates the foveal representation in retinotopic areas of the occipital cortex, and LSF scenes (as opposed to HSF) activate more peripheral representations in retinotopic areas.

The present review also studied the neural bases of coarse-tofine analysis as a default and predominant processing strategy. According to influential models (Bullier, 2001; Bar, 2003; Bar et al., 2006; Kveraga et al., 2007; Peyrin et al., 2010), LSF information may reach high-order areas rapidly, enabling coarse initial parsing of the visual scene, which can then be sent back through feedback connections into lower level visual areas to guide a finer analysis based on HSF. Studies also indicate that in scene perception, coarse-to-fine processing seems to be preferentially performed in the right hemisphere, from the occipital to the inferior temporal cortex. Overall, results from neuroimaging studies are consistent with the idea that explicit vision advances in a reverse hierarchical direction, as hypothesized by Hochstein and Ahissar (2002) and Ahissar and Hochstein (2004; see The Reverse Hierarchy Theory). According to this theory, rapid visual perception is not purely feedforward, it is also strongly mediated by topdown influences by high-level areas on lower-level areas. Finally, the present review addressed spatial frequency processing within scene-selective cortical areas. We reported results demonstrated that the coarse-to-fine strategy is a plausible modus operandi in the PPA.

Overall, these results obviously raised the question of the connectivity between the PPA and the cortical network specifically involved in coarse-to-fine processing. Baldassano et al. (2013) recently demonstrated that the PPA exhibits a gradient in connectivity with other scene-specific regions along the anterior-posterior axis in a way that suggests that the posterior part of the PPA is more closely connected to occipital areas and therefore contributes more to the processing of low level visual features (possibly to spatial frequencies and spatial envelope properties) while the anterior part of the PPA is more closely connected to the RSC and therefore contributes to the construction of a global scene representation. In Musel et al. (2014), the contrast between coarse-to-fine and fine-to-coarse processing revealed significant activation within the orbitofrontal cortex and the primary visual cortex (**Figure 6**). These two regions might play a predominant role during the coarse-to-fine categorization of scenes. Involvement of the orbitofrontal cortex was previously evidenced in rapid LSF-based categorical inferences (Bar et al., 2006; Peyrin et al., 2010) and the primary visual cortex was evidenced to be one of the cortical sites in which the first LSF computation could be "retro-injected" to guide the subsequent finer analysis of HSF (Bullier, 2001; Peyrin et al., 2010). In a proactive brain model, Bar (2007) attempts to clarify the functional role of the parahippocampal cortex (including the PPA) in object recognition. According to this model, LSF information in an object is projected from early stage visual areas to the orbitofrontal cortex. Based on the global appearance of the object, this region then triggers activation of the most probable object identities. Parallel projection of LSF information to the parahippocampal cortex and the PPA also occurs to extract the context in which this object appears and activates its contextual associations. The intersection of possible object identities (from the orbitofrontal cortex) and the objects that typically appear in such contexts (from the parahippocampal cortex) provides fast and coarse recognition of the current view of the object. This assumption is supported by studies on the macaque brain which indicate that the orbitofrontal cortex has strong and reciprocal links with the temporal cortex, notably medial regions including parahippocampal areas

(Cavada et al., 2000). In humans, studies using diffusion tensor MRI have evidenced structural connectivity between the parahippocampal cortex and orbitofrontal areas (Powell et al., 2004). Bar et al. (2006) also demonstrated strong synchrony between the orbitofrontal cortex and the temporal cortex during the recognition of LSF-filtered objects, suggesting important functional interactions between these regions. Unfortunately, to our knowledge, the functional connectivity or direct influence between the orbitofrontal cortex and PPA has not been demonstrated yet.

To conclude, the results reported in the present review provide critical support for influential models of visual perception mainly based on a spatial frequency analysis which follows a coarse-to-fine strategy (Schyns and Oliva, 1994; Bar, 2003; Hegde, 2008; Peyrin et al., 2010).

#### **ACKNOWLEDGMENTS**

Our works were supported by the National Centre for Scientific Research in France, by a doctoral fellowship from the Région Rhone-Alpes to Louise Kauffman, by the RECOR ANR Grant (ANR-12-JHS2-0002-01 RECOR) and by a research grant from the Fyssen Foundation. The authors extend warm thanks to the UMS IRMaGe 017, the "Délégation à la Recherche Clinique" of the University Hospital of Grenoble and the"Cellule de Neuroimagerie Fonctionnelle" of the Laboratory of Psychology and NeuroCognition for sponsoring their works. We thank Catherine Dal Molin for the English revision of the manuscript.

#### **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: 16 February 2014; accepted: 19 April 2014; published online: 07 May 2014. Citation: Kauffmann L, Ramanoël S and Peyrin C (2014) The neural bases of spatial frequency processing during scene perception. Front. Integr. Neurosci. 8:37. doi: 10.3389/fnint.2014.00037*

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

## Differential processing of natural scenes in posterior cortical atrophy and in Alzheimer's disease, as measured with a saccade choice task

#### **Muriel Boucart <sup>1</sup>\*, Gauthier Calais <sup>2</sup> , Quentin Lenoble<sup>1</sup> , Christine Moroni <sup>1</sup> and Florence Pasquier <sup>3</sup>**

<sup>1</sup> Laboratoire Neurosciences Fonctionnelles et Pathologies, Université Lille Nord de France, CNRS, Lille, France

<sup>2</sup> Faculté Libre de Médecine, Université Lille Nord de France, Université Catholique de Lille, Service de Neurologie du Groupement des Hôpitaux de l'Institut Catholique de Lille, Lille, France

<sup>3</sup> Centre de la Mémoire, Centre Hospitalier Universitaire de Lille, Université Lille Nord de France, Lille, France

#### **Edited by:**

Olivier A. Coubard, CNS-Fed, France

#### **Reviewed by:**

Sébastien M. Crouzet, Charité University Medecine, Germany Zoi Kapoula, Groupe IRIS Vision et Motricité Binoculaire - UMR8194, France

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

Muriel Boucart, Laboratoire Neurosciences Fonctionnelles et Pathologies, Université Lille Nord de France, CNRS, Rue Emile Laine, 59037 Lille, France e-mail: muriel.boucart@chru-lille.fr

Atrophy of the medial temporal lobe structures that support scene perception and the binding of an object to its context (i.e., the hippocampus and the parahippocampal cortex) appears early in the course of Alzheimer's disease (AD). However, few studies have investigated scene perception in people with AD. Here, we assessed the ability to find a target object within a natural scene in people with AD and in people with posterior cortical atrophy (PCA, a variant of AD). Pairs of color photographs were displayed on the left and right of a fixation cross for 1 s. In separate blocks of trials, participants were asked to categorize the target (an animal) by either moving their eyes toward the photograph containing the target (the saccadic choice task) or pressing a key corresponding to the target's location (the manual choice task). Isolated objects and objects within scenes were studied in both tasks. Participants with PCA were more impaired in detection of a target within a scene than participants with AD. The latter's performance pattern was more similar to that of age-matched controls in terms of accuracy, saccade latencies and the benefit gained from contextual information. Participants with PCA benefited less from contextual information in both the saccade and the manual choice tasks—suggesting that people with posterior brain lesions have impairments in figure/ground segregation and are more sensitive to object crowding.

#### **Keywords: Alzheimer, posterior cortical atrophy, context, scene perception, saccades**

Although memory deficits are typically the first symptoms of Alzheimer's disease (AD) to appear, a growing body of evidence suggests that many aspects of visual cognition are impaired in people with AD (Mendola et al., 1995; Valenti, 2010). Indeed, some studies indicate that visual disturbances might even precede the memory deficits (Benson et al., 1988) and may be predictive of cognitive impairments (Cronin-Golomb et al., 1995). Furthermore, impaired object perception affects instrumental activities of daily living in people with AD (Mosimann et al., 2004; Jefferson et al., 2006). Disturbances in both spatial and object recognition processes have been reported consistent with the impact of the disease on both dorsal stream areas, in posterior cortical atrophy (PCA), and atrophy of ventral stream areas (Possin, 2010) though impairments in functions of the ventral stream seem more severe in AD. In contrast, ventral stream aspects of visual cognition, such as recognizing objects and faces, tend to be less impacted than dorsal stream aspects such as mental rotation, coherent motion perception or angle discrimination in PCA than in AD (Possin, 2010). However, the pattern of visual impairments in patients with PCA is heterogeneous (Tsai et al., 2011). PCA has been defined as a "nontypical form of Alzheimer disease" or as a "visual variant" of AD. It is characterized by a relatively selective decline in visual processing and other posterior cortical functions (such as visuomotor and visuospatial abilities), whereas impairments of memory, language and other cognitive functions only occur late in the disease (Benson et al., 1988; Schmidtke et al., 2005; McMonagle et al., 2006; Lehmann et al., 2011). Neuroimaging studies shown that atrophy in PCA is more marked in posterior regions of the brain (the parietal, temporal and occipital cortices) (Tenovuo et al., 2008; Kennedy et al., 2012; Migliaccio et al., 2012), which results in a range of visual disturbances (including visual agnosia, environmental disorientation, apraxia and alexia) (Mendez et al., 2002; Crutch et al., 2012).

Object recognition experiments, whether performed with healthy subjects or patients, typically investigate objects in isolation. However, "real-world" objects rarely appear in the absence of some context. Few studies have investigated scene perception in people with AD. This is paradoxical, given (i) the early atrophy of the medial temporal lobe (Delacourte et al., 1999; Villain et al., 2008), which can start at least 3 years before AD is diagnosed (Ridha et al., 2008); and (ii) significant cell loss (Davies, 2006) in structures (the hippocampus and parahippocampal cortex) that support scene perception (Epstein and Kanwisher, 1998; Epstein, 2008) and the binding of an object to its usual context (Goh et al., 2004; Fenske et al., 2006). Lee et al. (2007) found that patients with AD were more impaired in the discrimination of scenes (either landscapes or in virtual reality rooms), than in the discrimination of faces. Shakespeare et al. (2013) examined scene perception in patients with PCA. In a description task, patients named fewer features and made more misperceptions than controls. When visual exploration was recorded, the researchers observed that a number of patients fixated on uninformative parts of the scene rather than the details that were commonly fixated by controls. This was interpreted as evidencing poor eye movement control and/or an inability to implement a successful scanning strategy.

Here, we used a saccadic choice task to investigate scene perception and, more specifically, the ability to find a target object within a natural scene in a group of people with AD and a group with PCA. The saccadic choice paradigm was initially developed by Kirchner and Thorpe (2006) to measure the speed of object categorization. In their original study, Kirchner and Thorpe presented healthy young adult participants with two lateral (left/right) photographs of natural scenes and asked them to move their eyes as quickly as possible to the scene containing an animal. The researchers found that although the median latency was 228 ms, the fastest saccades to animal targets were triggered as soon as 120–130 ms after stimulus onset. Crouzet et al. (2010) subsequently showed that these rapid saccadic responses can be initiated even more rapidly (around 100 ms post-onset) when the target is a human faces and more slowly when the target is a non-biological object (e.g., a vehicle). Crouzet and Thorpe (2011) then suggested that the underlying mechanism for such rapid responses may be based on information that is only partially processed by the occipitotemporal cortex (Cauchoix and Crouzet, 2013) (possibly in V4) (Crouzet et al., 2012).

In the present study, participants were shown two photographs (presented to the left and right of a central fixation point) and asked to saccade to the image containing an animal. To assess the patients' ability to process contextual information, we compared their performance under two different conditions: the target animal was presented either in isolation on a homogeneous gray background or in its natural setting. Given that impairments in eye movement (e.g., prolonged saccade latencies and hypometric saccades) have been reported in patients with AD (Fletcher and Sharpe, 1986; Shafiq-Antonacci et al., 2003), the saccade responses were compared with manual responses using the same stimuli and the same presentation conditions.

A growing body of literature evidence from behavioral, visual cognition (Biederman et al., 1982; Boyce et al., 1989; Boyce and Pollatsek, 1992; de Graef et al., 1990; Henderson et al., 1999; Davenport, 2007), electrophysiology (Ganis and Kutas, 2003) and brain-imaging studies (Bar and Aminoff, 2003; Goh et al., 2004; Kirk, 2008; Mudrick et al., 2010) in healthy people suggests that contextual information affects the efficiency of object searching and recognition. We therefore expected the presence of context to facilitate target detection in healthy participants. As the parahippocampal region is involved in the binding of an object to its context (Bar, 2004) and this region is affected early by cellular and structural alterations in AD (Ridha et al., 2008; Villain et al., 2008; Apostolova et al., 2012), we expected patients with AD to benefit less than healthy participants from the background. Indeed, parahippocampal atrophy is even considered as an early biomarker for AD (Echávarri et al., 2011). The impairment of basic visual skills (including visual acuity, line orientation, contour integration, figure/ground segregation, form detection and discrimination, motion discrimination and point localization) in patients with PCA (Metzler-Baddeley et al., 2010; Lehmann et al., 2011) is consistent with evidence of widespread lesions within the occipital and parietal areas in this disease. Hence, we expected patients with PCA to be more sensitive than patients with AD to (i) lateral masking of the object by the background; and (ii) crowding effects (Crutch et al., 2012) and thus more impaired than patients with AD when the target object was embedded within a scene (relative to embedding in a homogeneous background).

## **METHOD**

### **PARTICIPANTS**

Six patients (3 males) with a diagnosis of PCA, 14 patients (6 males) with a diagnosis of AD, 15 healthy elderly adults (5 males) and 10 healthy young adults (3 males) were enrolled in the study by Lille University Hospital's Memory Clinic (Lille, France). Patients taking cholinesterase inhibitors were included if the dose had not changed for at least 6 weeks prior to inclusion. Despite the absence of ophthalmologic impairments or psychiatric disorders, all the patients with PCA had progressive, insidious signs of impaired visuospatial orientation. Magnetic resonance imaging (MRI) was used to confirm the atrophy of the occipital, parietal and (in some cases) temporal cortex and the absence of hippocampal atrophy. Patients with PCA were diagnosed according to the criteria published by Tang-Wai et al. (2004).

The major initial symptom in patients with AD was a progressive memory complaint (for at least 6 months previously) whose symptoms interfered with activities of daily living. MRI showed predominant hippocampal atrophy. All patients fulfilled the International Working Group's research criteria (Dubois et al., 2010) after a comprehensive work-up including a neuropsychological assessment, MRI, CSF biomarker assays and SPECT or FDG-PET.

On average, the patients with AD were older than the patients with PCA and the elderly adult controls (mean ± SD age: 71.5 ± 10, 65.4 ± 5 and 66 ± 7, respectively) but the differences in age were not statistically significant. In the PCA group, the mean ± SD Mini Mental State Examination (MMSE; Folstein et al., 1975) and Mattis Dementia Score (DRS; Mattis, 1973) scores were 22.5 ± 3.61 and 114.5 ± 13.63, respectively. In the AD group, the mean MMSE and DRS scores were 23.3 ± 1.34 and 112.42 ± 24.55, respectively. There were no significant inter-group differences in these scores. Patients were excluded if they had evidence of vascular lesions, major depression or ophthalmologic impairments (cataract, macular degeneration or glaucoma). Patients with visual agnosia and/or hemineglect were also excluded. The control group was composed of elderly volunteers recruited from among the patients' relatives and had a mean ± SD MMSE score of 29.9 ± 0.2. Ophthalmological screening included a detailed review of current or past visual disturbances, the assessment of visual acuity, the Amsler grid (for macular degeneration) and signs of cataract. The young adult controls (mean age: 29.6 ± 8.5) were students in medicine or neuroscience and none had a history of neurological or psychiatric disease. The study was approved by the local investigational review board (CPP Nord-Ouest IV, reference 05/79) and performed in accordance with the tenets of the Declaration of Helsinki. Written, informed consent was provided by all participants.

## **STIMULI**

The stimuli were photographs of either natural scenes or isolated objects selected from commercially available libraries (Corel and Hemera Photo-Objects, respectively). Examples are shown in **Figure 1**. At a viewing distance of 57 cm, the mean angular size of the stimuli was 7◦ vertically × 5 ◦ horizontally for photographs of scenes and 5◦ × 5 ◦ for photographs of isolated objects. All pictures were displayed on a black background. Target animals included mammals, reptiles, insects, birds, crustaceans, and fish. The distractors were variously houses, monuments, means of transportation, flowers, fruits, and landscapes containing neither animals nor humans.

#### **EQUIPMENT**

Eye movements were recorded by a remote pupil-tracking system (RED-m, Senso-Motoric Instruments, Berlin, Germany) that uses infrared illumination and computer-based image processing. The tracking system records the eye position at a sampling rate of 120 Hz and compensates for slight head movements. The manufacturers report a gaze position accuracy of 0.5◦ . Images of the eye are analyzed in real-time by detecting the pupil, calculating the center and eliminating artifacts. Movement data were collected for each eye separately. The calibration stimulus was a grid containing nine white dots (2◦ × 2 ◦ degrees) displayed one at a time on a black background. During calibration, the participants were instructed to fixate the dot located in the middle of the screen and to move their eyes to the other dots as instructed. Calibration was performed twice, in order to check the stability. The experimental trials were only initiated if the eye tracker classified the calibration as "good". The participants were given the same instructions for all experimental trials: to look at the central fixation cross with their gaze as steady as possible. The pictures were displayed with Experiment Center software (Senso-Motoric Instruments). The recorded eye movement data were analyzed with BeGaze software (Senso-Motoric Instruments). Manual responses were recorded via a box fitted with two keys and connected to the computer.

#### **PROCEDURE**

Each participant was tested in two experimental sessions. The categorization tasks were composed of four separate blocks of trials determined by the response (saccade vs. manual) and the type of picture (isolated object vs. object in a scene). There were 100 trials in each block: 50 trials with the animal on the right of

the fixation cross and 50 trials with the animal on the left. To reduce priming effects, the saccade and manual response tasks were separated by at least 1 week and the stimuli were randomly selected from among 1000 scenes containing animals, 1000 scenes without animals, 200 isolated animals and 200 isolated objects. The fixation cross was displayed for 500 ms. Then, 200 ms after disappearance of the cross, a pair of photographs were simultaneously presented for 1 s (one located 7◦ left of the center of fixation and the other located 7◦ right of the center). The gap period of 200 ms between the disappearance of the fixation cross and the appearance of the photographs usually accelerates saccade initiation (Masson et al., 2000). In two blocks of 100 trials each (one with scenes as stimuli and the one with isolated objects) the participants were asked to saccade to the picture containing an animal. The left-side or right-side position of the target was randomized. Half of the participants in each group started with the isolated objects and the other half started with the scenes. In two other blocks of 100 trials each, the response was manual; the procedure was the same as in the saccade task but participants were instructed to respond by pressing the right key or the left key, depending on the target animal's location.

#### **RESULTS**

#### **THE SACCADIC CATEGORIZATION TASK**

Participants whose overall performance differed by two SDs from the mean were discarded. One patient with AD was excluded from the analysis because of performance at ceiling (i.e., a correct response rate of 96%) and one healthy elderly control was excluded because of performance at chance. Saccade latencies below 100 ms were considered to be anticipatory and were excluded from the analysis. Response accuracy and saccade latencies for correct responses were examined in analyses of variance (ANOVAs, performed with Statistica software). The target's spatial location (left/right), the group (young adult controls, elderly adult controls, people with AD, people with PCA) and the category of image (scenes/isolated objects) were included as variables (**Figure 2**). The participants in each group were the random variable.

The only difference between left and right targets was seen in the young adult controls (*t*(9) = 2.59, *p* < 0.029), who displayed greater accuracy for targets on the left (87.8%) than on the right (81%) in the scene condition. The effect of group was significant both for saccade latencies (*F*(3,39) = 11.14, *p* < 0.001) and accuracy (*F*(3,39) = 12.6, *p* < 0.001). The young adult controls and elderly adult controls did not differ significantly in terms of the saccade latency (respectively 228 and 226 ms in the isolated object condition and 223 and 229 ms in the scene condition). However, the accuracy rate was 12.4% greater in young adult controls than in older participants (*F*(1,22) = 17.4, *p* < 0.001). Patients with AD did not differ significantly from age-matched controls in terms of either latency (240 vs. 228 ms, respectively; *F*(1,25) = 0.3, *p* = 0.61) or accuracy (61.6% vs. 65.9%, respectively; (*F*(1,25) = 2.4, *p* = 0.12), except when scenes were used as stimuli (*F*(1,25) = 5.97, *p* < 0.05). Patients with PCA were slower (by 153 ms, *F*(1,18) = 28.2, *p* < 0.001) and less accurate (by 5.6%) than agematched controls. The difference in accuracy was not significant *F*(1,18) = 2.6, *p* = 0.11), except when scenes were used as stimuli

**group's median and standard error) and accuracy (with the standard error) as a function of group (young adult controls, elderly adult controls, patients with AD and patients with PCA) and the type of image (isolated targets vs. targets in scenes)**.

(*F*(1,18) = 5.9, *p* < 0.05). Patients with PCA were also slower than patients with AD (by 141 ms; *F*(1,17) = 23.5, *p* < 0.001) but not significantly less accurate (60.3% and 61.6%, respectively). When averaged over all four groups, saccade latencies were similar for targets in their natural scenes (268 ms) and for isolated targets (269 ms) but accuracy was greater for targets in scenes than for isolated targets (70.6% vs. 62.4%, respectively; *F*(1,39) = 25.7, *p* < 0.001). This difference was observed (see **Figure 2**) for all groups but was only statistically significant for controls. The difference was 4.3% for patients with PCA, 5.6% for patients with AD, 10.6% (*t*(13) = 4.1, *p* < 0.001) for elderly controls and 12.4% (*t*(9) = 7, *p* < 0.001) for young adult controls. The group × type of image interaction did not achieve statistical significance.

#### **THE MANUAL CATEGORIZATION TASK**

Participants whose overall performance differed by two SD values or more from the mean were excluded from the analysis. Two patients with AD were excluded because of slow Response times (RTs). One of the participants with PCA failed to attend the session including the manual categorization task. RTs below 100 ms were excluded. Accuracy and correct RTs were examined in ANOVAs. The target's spatial location (left/right), the group (young adult controls, elderly adult controls, people with AD, people with PCA) and the category of image (scenes/isolated

objects) were included as variables (**Figure 3**). The participants in each group were the random variable.

The target's location (left/right) did not have a significant main effect on either accuracy or the RT in any of the four groups of participants. The effect of group was significant both for accuracy (*F*(3,38) = 13.6, *p* < 0.001) and the RT (*F*(3,38) = 52.6, *p* < 0.001). Participants with PCA were slower than participants with AD (by 447 ms; *F*(1,15) = 39.2, *p* < 0.001) and less accurate (by 10.6%; *F*(1,15) = 29.1, *p* < 0.001). They were also less accurate (*F*(1,18) = 38.1, *p* < 0.001) and slower (*F*(1,18) = 77.8, *p* < 0.001) than agematched controls. As can be seen in **Figure 3**, participants with AD were slower than healthy, elderly, age-matched controls (by 127 ms; *F*(1,25) = 10.0, *p* < 0.001) but were not less accurate (97.5 and 98.6%, respectively). Young adult controls were faster than elderly controls (by 215 ms; *F*(1,23) = 25.6, *p* < 0.001) but were not less accurate (97.6% and 98.6%, respectively).

In contrast to the results for the saccade response task, accuracy in the manual categorization task was better for isolated animals (96.3%) than for animals in scenes (94% *F*(1,38) = 10.7, *p* < 0.002). RTs were shorter for isolated animals than for animals in scenes (by 18 ms, *F*(1,38) = 1.4, ns). The category of image interacted significantly with group for both the RTs (*F*(3,38) = 2.93, *p* < 0.04) and accuracy (*F*(3,38) = 9.58, *p* < 0.001). As can be seen in **Figure 3**, this was mainly due to participants with PCA, who responded more rapidly and more accurately for isolated targets than for scenes (isolated: 1009 ms and 91.9%; scenes: 1142 ms and 81.9%; accuracy (*F*(1,4) = 30.1, *p* < 0.001; RTs: *F*(1,4) = 7.5, *p* < 0.01)). There were no significant differences in accuracy or RTs when the two image categories were compared in the three other groups of participants.

When averaged across all four groups and the two types of image, accuracy was higher in the manual categorization task than in the saccade categorization task (95.2% and 66.5%, respectively; *F*(1,37) = 442, *p* < 0.001) but manual RTs were longer than saccade latencies (708 vs. 269 ms, respectively; *F*(1,37) = 635.8, *p* < 0.001).

#### **DISCUSSION**

In the present study, we investigated scene perception (and, more specifically, the use of background contextual information on object categorization) in people with AD and people with PCA (a variant of AD affecting the occipitoparietal regions of the brain).

In view of the early-onset lesions in the hippocampal area in AD and in the occipitoparietal areas in PCA, we expected both populations to be impaired when compared with age-matched controls. Indeed, our analysis of the saccade response task (during which participants had to respond quickly) showed that both patients with AD and patients with PCA were less accurate than age-matched controls when scenes were used as stimuli. In the manual response task, participants had more time to explore the images. If the first saccade went to the wrong side (i.e., to the image lacking an animal), the participant was able to shift his/her gaze to the other side before giving a manual response. Under this condition, patients with AD no longer displayed an impairment in accuracy (the correct response rate was over 95%) but were still slower than age-matched controls. Patients with PCA were always less accurate than the other groups—especially when scenes were used as stimuli. When given more time to respond (i.e., in the manual response task), the patients with PCA had markedly better performance levels when isolated objects were used as stimuli.

The lower observed accuracy in the AD group (relative to agematched controls) when scenes were used as stimuli in the saccade choice task confirms the results of previous studies (Boucart et al., 2014) in which scenes were the only stimuli used. Our results for the saccade choice task show that the three groups of older participants were particularly impaired (with a correct response rate below 60%) when isolated objects served as stimuli. Performance was better for scenes in all groups. This suggests that in a rapid choice task, the background context facilitates selection of the target. When participants are given more time to explore images (as in the manual response task), the background context has less effect: performance at ceiling was observed in all groups except for the PCA group. Worse performance for isolated objects than for scenes in the saccade task cannot be explained by the size of the target object because the animals in scenes were no larger than animals displayed in isolation. It cannot be held that isolated animals were more difficult to categorize than animals in scenes because performances for the two types of image were similar in the manual response task (except in the PCA group). It is likely that the context facilitated performance when there was limited time for providing a response. The results of studies of contextual information processing in healthy young adults suggest that background information is processed early (possibly by the fast magnocellular pathway). For instance, Bar (2004) suggested that (i) if background (contextual) information is to assist the recognition process, it has to be extracted rapidly; and (ii) this rapid extraction is mediated by general cues conveyed by low spatial frequencies in the image. This coarse information is projected rapidly from the visual cortex to the prefrontal and parahippocampal cortices (possibly via the magnocellular pathway), where it can activate a scene schema. The representation is then refined and further instantiated as specific details progressively arrive at higher spatial frequencies. Magnocellular dysfunction (as demonstrated by electroretinography and visual-evoked potentials for simple stimuli (gratings varying in spatial and temporal frequencies) has been reported in AD (Gilmore and Whitehouse, 1995; Jacobs et al., 2002; Sartucci et al., 2010). Lenoble et al. (2013) found that impairments in the magnocellular pathway affect high-level object categorization.

As can be seen in **Figures 2**, **3**, the differences between the AD group and age-matched controls in both the saccade and manual categorization tasks were smaller than the differences between the PCA group and age-matched controls. Saccade latencies and manual RTs were longer in patients with AD than in age-matched controls; this agrees with reports of greater overall processing times in AD (Shafiq-Antonacci et al., 2003). Analyses of dependent variables (accuracy and response time) in elderly controls often reveal a speed-accuracy trade-off. This has been interpreted as the use of a more cautious response strategy in order to avoid errors or to compensate for decreased cognitive control (Ratcliff et al., 2007; Endrass et al., 2012). In the present study, the AD patients' results may be due to a compensatory shift in response strategy. Nevertheless, the AD group's excellent accuracy (>90%) in discriminating a target animal presented for a limited presentation time (1 s) in the manual response task contrasts with the results of Neargarder and Cronin-Golomb (2005) study. They showed that patients with AD were impaired at detecting a change within pairs of scenes. However, Neargarder and Cronin-Golomb's patients with AD were older than those in the present study (with mean ages of 80.4 and 71, respectively) and had lower MMSE scores (19.5 and 23, respectively). Moreover, the detection of a change requires more attention to local parts of the scene than the detection of an animal within a scene; the latter task can be accomplished with low-resolution peripheral vision (at an eccentricity of 75◦ ) (Thorpe et al., 2001). Scinto et al. (1994) showed that patients with AD are more impaired in tasks that place increased demands on attention.

Our patients with PCA were slower and less accurate than age-matched controls in the saccade task (especially when scenes served as the stimuli) but, in contrast to patients with AD, they were also less accurate and slower than both age-matched controls and participants with AD in the manual response task. Another major difference between patients with PCA on one hand and

patients with AD and controls on the other is that the former exhibited better performance levels for isolated animals than for animals in scenes in the manual response task. Shakespeare et al. (2013) presented objects, faces and scenes in separate blocks of trials to patients with PCA and to age-matched controls. As in the present study, Shakespeare et al. observed a better performance for objects than for scenes in patients with PCA. However, in contrast to our present study, controls also had better performance levels for objects than for scenes. When considering Shakespeare et al.'s study in more detail, it is noteworthy that participants were shown a single image on each trial and asked to choose the corresponding name in a three-alternative, forced verbal categorization task (e.g., "Is this a forest, a desert or a beach?"). Participants may be more likely to confuse a desert and a beach than they are to fail to detect an animal in a pair of scenes. The low impact of contextual information in patients with PCA (relative to the other groups of participants) might reflect a higher sensitivity to crowding, and impaired figure/ground segregation, in this population. Consistently, Shakespeare et al. (2013) reported that patients with PCA responded more accurately and more rapidly to colored stimuli than they did to grayscale. Color is known to facilitate the segmentation of surfaces and figure/ground segregation—especially when vision is impaired experimentally (Oliva and Schyns, 2000) or by disease (Wurm et al., 1993; Boucart et al., 2008). Crowding refers to the decreased visibility of a visual target in the presence of nearby objects or structures; it impairs the ability to recognize objects in "cluttered" scenes and has a more pronounced effect on peripheral vision (Pelli et al., 2004; Levi, 2008). Consistently, Crutch and Warrington (2007) reported high sensitivity to crowding in two patients with PCA in a letter reading task in which the spatial proximity of a flanker was manipulated. In the people with PCA assessed in our present study, both crowding and difficulties in figure/ground segregation were more problematic for scenes than for isolated objects. In monkeys, figure/ground segregation is reportedly altered by lesions to the visual cortex (Supèr and Lamme, 2007).

With the exception of accuracy, there were no differences in oculomotor parameters when comparing young and elderly adult controls. Although many studies have observed effects of aging on eye movements (Irving et al., 2006; Porter et al., 2010; Paquette and Fung, 2011; Ridderinkhof and Wijnen, 2011), it seems that automatic parameters (such as latencies in pro-saccade tasks) are barely influenced by aging (Abrams et al., 1998; Munoz et al., 1998; Kaneko et al., 2004).

Conclusion: Literature studies of object and scene perception in AD have not usually distinguished between typical AD and atypical AD (i.e., forms with lesions in the posterior parts of the brain, such as PCA). Our present results revealed differences in performance in patients with PCA, who were more impaired in scene perception than patients with AD. The latter displayed performance patterns (in terms of accuracy and the benefit gained from contextual information) that were more similar to those in age-matched controls. Patients with PCA benefit less from contextual information, suggesting higher sensitivity to crowding and impaired figure/ground segregation in people with lesions in posterior areas of the brain.

## **ACKNOWLEDGMENTS**

This work was funded by a grant from the French National Agency for Research's "Neurologic and Psychiatric Diseases" program (ANR-08-MNPS-001-01).

### **REFERENCES**


of Alzheimer's disease. *Brain Struct. Funct.* 215, 265–271. doi: 10.1007/s00429- 010-0283-8


Mattis, S. (1973). *Dementia Rating Scale.* Winsor: NFER-Nelson.


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

*Received: 21 December 2013; accepted: 03 July 2014; published online: 25 July 2014*. *Citation: Boucart M, Calais G, Lenoble Q, Moroni C and Pasquier F (2014) Differential processing of natural scenes in posterior cortical atrophy and in Alzheimer's disease, as measured with a saccade choice task. Front. Integr. Neurosci. 8:60. doi: 10.3389/fnint.2014.00060*

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

*Copyright © 2014 Boucart, Calais, Lenoble, Moroni and Pasquier. 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*.

## What visual illusions teach us about schizophrenia

## *Charles-Edouard Notredame1,2\*, Delphine Pins 2, Sophie Deneve3 and Renaud Jardri 1,2,3*

*<sup>1</sup> Pediatric Psychiatry Department, University Medical Centre of Lille, Lille, France*

*<sup>2</sup> SCA-Lab, PSYCHIC Team, Université Lille Nord de France, Lille, France*

*<sup>3</sup> Group for Neural Theory, INSERM U960, Institute of Cognitive Studies, École Normale Supérieure, Paris, France*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

#### *Reviewed by:*

*Daniel Calderone, New York University Langone Medical Center, USA Ignacio Serrano-Pedraza, Complutense University of Madrid,*

#### *\*Correspondence:*

*Spain*

*Charles-Edouard Notredame, Service de Psychiatrie de l'enfant et de l'adolescent, Hôpital Fontan, CHRU de Lille, CS 70001, 59037 Lille, France e-mail: notredame.ce@gmail.com*

Illusion, namely a mismatch between the objective and perceived properties of an object present in the environment, is a common feature of visual perception, both in normal and pathological conditions. This makes illusion a valuable tool with which to explore normal perception and its impairments. Although still debated, the hypothesis of a modified, and typically diminished, susceptibility to illusions in schizophrenia patients is supported by a growing number of studies. The current paper aimed to review how illusions have been used to explore and reveal the core features of visual perception in schizophrenia from a psychophysical, neurophysiological and functional point of view. We propose an integration of these findings into a common hierarchical Bayesian inference framework. The Bayesian formalism considers perception as the optimal combination between sensory evidence and prior knowledge, thereby highlighting the interweaving of perceptions and beliefs. Notably, it offers a holistic and convincing explanation for the perceptual changes observed in schizophrenia that might be ideally tested using illusory paradigms, as well as potential paths to explore neural mechanisms. Implications for psychopathology (in terms of positive symptoms, subjective experience or behavior disruptions) are critically discussed.

**Keywords: illusions, schizophrenia, psychosis, visual perception, hallucinations, delusions, Bayesian inference, predictive coding**

#### **INTRODUCTION**

Vision has long been considered one of the main routes humans use to understand the world (Glezer, 1995). However, recent findings encourage moving beyond this presupposed primacy of vision over other senses, notably, when looking at the details of visual misperceptions. Visual arts provide eloquent illustrations of common mistakes made in vision. A first example is when artists use artifices to mislead perception and induce, for example, a paradoxical "realistic" feeling of depth or movement. Illusionists also frequently resort to human perceptual properties in their magic tricks (Martinez-Conde and Macknik, 2007) to the extent that the study of these tricks has become the object of an original sub-field in cognitive sciences called "neuro-magic."

Historically, the first scientific descriptions of misleading visual effects date back to the late 19th to early 20th century. Physiologists, such as Poggendorff, Herman, Müller-Lyer, Ponzo and Ebbinghaus, noticed that our appreciation of contrast, size or continuity can be distorted by contextual information (Zölner, 1860; Hermann, 1870; Müller-Lyer, 1889; Ponzo, 1910). Similarly, Necker and, later, Boring and Rubin described ambiguous figures that could lead to different, mutually exclusive interpretations (Necker, 1832; Boring, 1930; Rubin, 1958) (See **Figure 1**). These visual effects were named visual illusions (VIs) or optical illusions. Progressively, VIs assumed a place of increasing importance in the literature as a fertile, practical and valid way to explore the underlying mechanisms of perception in normal or pathological conditions.

There is a strong tradition of considering as illusory every image that misleads perception into instability, insolubility, distortion or fiction (for an example of synthetic classification, please refer to **Table 1**). As a consequence, an overview of the literature reveals that the stimuli traditionally considered as illusory vary greatly in terms of (a) complexity [from the simple *Three flash illusion* (Bowen, 1989) to the more ecological *Hollow-mask illusion* (Gregory, 1973)]; (b) the perceptual mechanisms or physiological pathways involved (e.g., contrast vs. motion illusions); (c) the level of integration required (e.g., contrast detection vs. bistable perception induced by an ambiguous figure); and (d) the type of subjective impression that they may induce (e.g., apparition of fictive gray points in the *Hermann's grid* vs. distorted size perception in the *Ebbinghaus illusion*). Such diversity is actually consistent with the important difficulty to consensually define illusions. On a purely theoretical level, authors such as Gregory (1997a) and Eagleman (2001) noted the pitfall of simply considering illusions as a gap with reality, a definition that might correspond to the whole process of perception. In contrast to this extensive point of view, an excessively restrictive definition would risk characterizing sub-categories of illusions rather than the general perceptual phenomenon. This would consequently impede any theorization. As a starting point for this review, we opted for the classical compromise of considering illusions as a systematic mismatch between the basic response of the sensory organs to a stimulus (related to its physical properties), and the percept this object gives rise to. Nevertheless, this working hypothesis urges a theoretical clarification, which the Bayesian framework will help us to address in Section Visual Illusions and the Bayesian Theory.

**FIGURE 1 | Main classical illusions.** In the Ebbinghaus **(A)**, Ponzo **(B)** and Müller-Lyer **(C)** illusions, same-sized patterns are misevaluated because of the context. In the Poggendorff illusion **(D),** the context disrupts the impression of continuity. Herman's grid **(E)** generates

illusory gray points at each intersection of the white lines. The Boring wife/mother-in-law **(F)**, the Necker Cube **(G)** and Rubin's Maltese Cross **(H)** are ambiguous figures that result in different interpretations.

In this perspective, illusions, hallucinations and hallucinosis are experiences that share the property of being inconsistent with the actuality of the sensory environment. As such, they all belong to the category of false percepts. Nevertheless, the present review is based upon crucial points of the definition in order to distinguish illusions from other misperceptions. VIs originate from an object already present in the environment and occur frequently in "normal" visual processing, whether naturally induced or intentionally provoked. The vulnerability to VIs does not have any pathological significance *per se*, but rather, it can be considered a common phenomenon. In contrast, hallucinations and hallucinosis consist of perceptions without any sensory substratum. Hallucinations can occur in a wide range of conditions, from nonclinical groups to full-blown psychosis. The pathological nature of these experiences basically depends on the strength of the associated beliefs and the extent to which they give rise to delusional interpretations. This point of view posits the existence of a continuum from total confidence, where the hallucinatory percept is completely integrated in the subject's life, to total distrust, where the false percept is felt as odd and interpreted as abnormal (this later end of the continuum characterizes hallucinosis).

These particular precisions are important to understand how the abnormal patterns of illusory perception potentially identified in patients who suffer from schizophrenia differ from, but can be linked to, the other false percepts (e.g., hallucinations). Schizophrenia is a severe and disabling disorder that affects approximately 1% of the general population (McGrath et al., 2008). Although no clear pathophysiological mechanism has emerged to explain the disease, complex impairments in integrative functions are thought to result from a large-scale dysconnectivity syndrome (Frith and Done, 1988; Friston, 1998, 2005a; Burns et al., 2003; Liang et al., 2006; Stephan et al., 2009; Schmitt et al., 2011; Amad et al., 2014). This disturbance would result in aberrant concept formation, delusions, and hallucinations.

In a recent paper, Silverstein and Kean remind us that visual sciences have been invested since the 1950's to improve insight into the brain functioning of schizophrenia patients (Silverstein and Keane, 2011). In line with the assumption that

#### **Table 1 | Gregory's classification.**


*The classification of visual illusions (VIs) highlights the difficult problem of criteria choice. In one of his categorizations, Gregory empirically chose to cluster VIs on the basis of an analogy between their appearance and the main language errors (e.g., ambiguity, distortion, paradox and fiction) from which he assumed they were derived (Gregory, 1997b).*

perception is closely interwoven with other high-level functions (such as belief genesis or reasoning; Fletcher and Frith, 2008), several authors have defended the idea that visual perception plays an important role in the psychopathology of schizophrenia (Butler and Javitt, 2005), and constitutes a unique way to explore the underlying mechanisms of reality construction (Silverstein and Keane, 2011). VIs were an early component of visual research in schizophrenia (see first referenced studies: Letourneau and Lavoie, 1973; Letourneau, 1974). More recently, these procedures have gained a renewed interest, with Dakin et al. (2005) as the first authors to postulate a *decreased* susceptibility to illusion in this disorder. This hypothesis relies on a simple paradigm: the participants had to compare the contrast of two patches successively presented. The first patch was isolated or surrounded by a high-contrast annulus, and the second patch was an isolated reference patch. By manipulating the reference-contrast, the authors noticed that the patients who suffered from schizophrenia were significantly less biased by the surroundings and were thus more accurate in contrast discrimination compared with the healthy and psychiatric controls. The capacity of schizophrenia patients to outperform normal subjects in tasks that involved VIs paved the way for a new approach to explore perceptual processing in psychosis. Indeed, it allows controlling for one of the main confounding factors in this population, i.e., decreased performances due to a global cognitive deficit.

In this paper, we review recent VI experiments conducted in schizophrenia patients, with the aim of explaining how apparently composite findings may offer a holistic comprehension of visual perception in schizophrenia. Furthermore, we critically discuss how probabilistic theories of perception (e.g., Bayesian) are of interest to understand the singular pattern of illusion sensitivity in schizophrenia. We also demonstrate how the computational hypotheses of psychosis benefit from VIs to provide insights in the genesis of the positive symptoms and in other psychopathological properties in this disorder.

## **VISUAL ILLUSIONS: A WAY TO PROBE THE INTEGRITY OF VISUAL PROCESSING IN SCHIZOPHRENIA?**

#### **DISCUSSING THE LIMITS OF THE STRUCTURAL APPROACH**

The physiological existence of VIs suggests that deceiving perception paradoxically requires the integrity of the different subcomponents of the visual system. At first sight, the corollary assumption appears to make sense: one could consider resistance to VIs as the expression an identifiable, specific and common disruption that affects the structures visual processing. However, trying to synthesize the studies that adopted this structural perspective led to two main limitations.

First, the structural approach conceptualizes the impairments of visual processing as potentially resulting from dysfunctions of the brain structures computing low-level sensory information. If this hypothesis is correct, this would result in poorer experimental performances of the patients compared with controls. To date, the findings based on VI paradigms in schizophrenia do not allow clear-cut conclusions. On the one hand, many authors, following Dakin, reported that patients who suffer from schizophrenia outperform controls in VI experiments. Dakin's findings were notably replicated using the same type of context-based illusion paradigms (Uhlhaas et al., 2004; Tadin et al., 2006; Chen et al., 2008), but were also extended to other VI categories, such as the Binocular Depth Inversion (Koethe et al., 2009), illusory motion and motion-induced illusions (Tschacher et al., 2006; Crawford et al., 2010). On the other hand, some studies evidenced unchanged, augmented or contrasted patterns of sensitivity to VIs in schizophrenia. A summary of the main empirical findings regarding VIs and schizophrenia is provided **Table 2**. Importantly, most of the negative or nuanced results referred to surroundsuppression illusions. This paradigm appears exposed to specific methodological issues and covariates that may be imputed in the results (see Section A Window Into the Functional Dimension of Visual Perception). Even if these data should still be interpreted with caution, the hypothesis of a diminished sensitivity to VIs in schizophrenia appears to be a most robust finding. From a methodological point of view, increased performance in patients cannot be attributed to general task difficulty, a lack of attention or motivation or to a disease-related global deficit. Rather than the classical cognitive deficit-model proposed in schizophrenia, these findings suggest a perceptual process different in nature compared to the one observed in the general population.

Second, several authors resorted to the structural approach to localize the specific disruption hypothetically responsible for the pattern of sensitivity to VIs observed in schizophrenia. However, these attempts led to apparent contradictions. This is notably the



*(Continued)*

#### **Table 2 | Continued**


#### **Table 2 | Continued**


*SCZ-T, treated schizophrenia; SCZ, untreated schizophrenia or unknown treatment status; FEP, first episode psychosis; IPS, initial prodromal state of psychosis;* -*9- THC,* -*9-Tetrahydrocannabinol; MDD, major depressive disorder; BD, bipolar disorder; CTL, controls; PANSS, positive and negative symptoms scale. The relevant clinical correlates are in italics.*

case when trying to assess, in a hierarchical perspective (from retina to high cortical areas), whether resistance to illusion is because of a high or a low-level disruption. For example, Norton et al. (2008) suggested an impairment of earlier visual processing by resorting to the *Three-flash illusion.* In this paradigm, a light pulse quickly presented twice appears as three flashes. The authors reported the illusion peaked to a longer inter-stimulus interval when presented to patients with schizophrenia compared with healthy controls. They argued that this temporal alteration may rely on a faulty sensory-integration within the very first stages of the visual stream. In contrast, by exposing participants to a battery of simple biasing-context illusions, Yang et al. (2012) and Tibber et al. (2013) found that luminance context processing was similar, whereas the strength of the illusions was inferior in patients for the other psychophysical parameters (e.g., contrast, orientation, size or motion). Thus, these authors hypothesized that the pre-cortical stages of the visual system, in which luminance is supposed to be processed, were spared, thereby contradicting Norton's conclusions.

VIs, although polymorphic, pinpoint the limits of purely anatomoclinical or linear causal models and underline, in contrast, the complexity and specificity of visual perception in schizophrenia. The data on VIs in schizophrenia patients encourage opting for a more functional and translational point of view.

#### **A WINDOW INTO THE FUNCTIONAL DIMENSION OF VISUAL PERCEPTION**

VIs are an illustration of the perceptual system's ability to bind and group visual elements into coherent patterns, which leads to a meaningful representation. According to Butler et al. (2008) this phenomenon of *perceptual organization* can be viewed as relying on two main fundamental mechanisms: *gain control* and *integration*, both of which could be impaired in schizophrenia. Interestingly, studying VIs provides crucial elements for understanding these two elementary mechanisms.

*Gain control* can be defined as an adaptive process by which the sensory system optimizes information transmission to consider the visual context. For example, adjusting neural gains can concentrate the neuron's limited dynamical range around the mean contrast or luminance of the context, thereby ensuring that their responses do not saturate with luminance or contrast. *Gain control* of neural responses is achieved through a combination of intrinsic neuronal properties, lateral interactions and feedback modulations (Must et al., 2004).

Although not always labeled as such, many VIs refer to a basic perceptual phenomenon called the *context* or *surround suppression effect,* which can be conceptualized as a form of *gain control.* This effect can be divided into facilitating or misleading effects when comparing two targets, depending on whether the context biases perception toward or away from the actual difference between the targets (see **Figure 2**). Some authors have argued, in line with the findings that stem from the study of VIs, that the impairment of *gain control* is a robust and significant property of visual processing in schizophrenia, and may even consider it a core feature of the disorder (Robol et al., 2013; Tibber et al., 2013). Nevertheless, carefully studying the VI literature encourages us to adjust and specify this hypothesis.

Despite the fact that the stimuli used in *surround suppression* VIs can easily be standardized, heterogeneous findings concerning

**FIGURE 2 | An example of a context suppression effect: the Ebbinghaus illusion.** Depending on whether the peripheral circles are large or small, the central targets appear smaller or larger, respectively. When comparing two targets, the actual difference (here, left one smaller compared with right one) **(A)** appears reduced with the misleading context **(B)** or increased with the facilitating context **(C)**.

schizophrenia have been reported in this sub-field (see **Table 2**). This noteworthy statement first raises the question of the interpretation of negative findings. Nevertheless, even if a lack of statistical power is a major cause of concern in neuroscience and may be imputed for several studies (Button et al., 2013), some clues suggest that heterogeneous findings in *surround suppression* may reflect, to a certain extent, the composite aspect of the underlying functional mechanism. (1) Some studies have probed *surround suppression* in different modalities through a battery of illusions in the same group of patients, thus limiting the methodological disparity. The authors failed to identify a steady impairment in the context effect across psychophysic characteristics. Furthermore, they could not reveal a constant inter-task correlation, which would have revealed a common uniform mechanism (Kantrowitz et al., 2009; Yang et al., 2012; Tibber et al., 2013). (2) Some evidence suggests that *surround suppression* and its impairment depends on several variables, such as the contrast (previously discussed) and the time of presentation (Calvert and Harris, 1988; Bressan and Kramer, 2013), which counters an absolute independent process. (3) The single phenomenon of contextual effect is not uniform and may be variously affected in its subcomponents as suggested by Chen et al. (2008). The authors proposed that context interaction impairment in schizophrenia may be better accounted for by alterations of surround inhibition rather than surround facilitation.

In complement to *gain control*, *integration* supports the visual system propensity to dynamically bind elements into complex perceptual constructs suitable for behavior or social skills (such as reading, face processing, and visual gnosis). From a neurophysiological point of view, *integration* relies on long-range projections of neuronal networks that connect superficial and deeper cortical layers. In VIs, the disruption in visual integration abilities becomes strikingly apparent when schizophrenia patients exhibit impaired coordinated visual skills necessary for the illusory effects to occur, such as stereopsis (Schechter et al., 2006). Recently, Schallmo et al. (2013) highlighted the abnormal integrative dimension of *perceptual organization* in schizophrenia, showing that patients' abilities for contour integration correlated with their *surround suppression* sensitivity, while these two illusory phenomena had previously only been shown to be impaired in isolation.

Unraveling such disruption of *perceptual organization* in its most basic mechanisms through VIs elucidates the connections with models that consider psychotic disorders a widespread deficit in cognitive coordination. However, testing this hypothesis and addressing how the VI literature may inform and be articulated requires a more quantitative framework, which is susceptible to coherently embrace all of the above-mentioned findings. Such computational theories should be able to integrate the results obtained in VIs and contribute new insight into the perceptual processes affected in schizophrenia. Among these theories, the Bayesian framework provides a natural framework for formalizing the interplay of integration, gain control, and feedforward and feedback processing in VIs in the general population and in schizophrenia.

#### **THE COMPUTATIONAL POINT OF VIEW**

#### **VISUAL ILLUSIONS AND THE BAYESIAN THEORY**

The links between probabilistic theories and perception can be illustrated by examining a now famous VI example: the *Hollowmask illusion* (readers can find a demonstration of the illusion on the following website: http://www.echalk.co.uk/amusements/ OpticalIllusions/hollowMask/hollowface.html). In this VI, a face is presented as depth-inverted as the result of a pseudoscopic procedure (*Binocular depth inversion*). Despite this drastic counterintuitive modification, healthy subjects still perceive the face as normally 3D-shaped, according to their knowledge of human anatomy (Gregory, 1973). This example nicely highlights how some VIs reveal the otherwise implicit interweaving between belief and perception by showing how prior expectations can overtake the actual objective properties of a given stimulus.

Starting from the Helmholtz's "unconscious inferences" theory (von Helmholtz, 1866), Bayesian models set uncertainty and belief as the core features of perception by considering sensory inputs as inherently ambiguous. In this probabilistic framework, perception is considered to result from optimal inferences concerning the world (see **Box 1**).

A re-examination, in the light of the Bayesian theory, of the stimuli that could have been traditionally considered illusory enables a conceptual refinement. Importantly, resorting to such a theoretical framework will help to re-delimit what does, and what does not fit with our definition of VIs.

In the Bayesian framework, the perceptual uncertainty that characterizes the "misleading" stimuli arises from weak or conflicting sensory evidence (Sundareswara and Schrater, 2008; Gershman et al., 2012), with 2 principal mechanisms.

(1) *A significant dissociation between a strong expectation and the sensory evidence.* For example, the *Hollow mask* results from an imbalance between lower (the 3D-inverted stimulus) and

#### **Box 1 | Bayesian theories of perception.**

The probabilistic approaches to perception consider sensory stimuli as inherently ambiguous. In order to build a coherent representation of the world, one has to combine uncertain sensory evidence with prior knowledge. Let us imagine that the task is to infer whether or not there is a tree (as summarized by a random binary variable theta). The Bayes theorem combines:


In order to compute:

• **The posterior**: the probability of the percept (is there a tree or not?) resulting from combining prior and likelihood.


#### p(θ) **Prior probability for tree**

= Probability of the parameter θ before any evidence

More generally, the Bayesian framework considers perception as a hierarchical inference process, with more abstract (higher) levels generating expectations and sending them down the cortical hierarchy (top-down process) toward sensory representation. Meanwhile, sensory evidence climbs up the hierarchy (bottom-up process) and activates these high levels representations. In this way, top-down expectations are constantly updated to account for new sensory evidence.

Several simplified framework have been proposed to describe hierarchical inference and relate it to the brain architecture.


Importantly, despite conceptual differences, these frameworks are algorithmically similar and may essentially differ in the type of variables considered (e.g., binary vs. continuous, see Jardri and Deneve, 2013).

higher (prior knowledge of anatomy) levels of abstraction. In this case, perception appears associated with a systematic bias toward prior knowledge. However, rather than a perceptual error, this bias is better understood as the consequence of inferences that optimally combine bottom-up and top-down information (see **Boxes 1**, **3**), which supports the best adaptation in a noisy environment. This corroborates, specifies and integrates in a theoretical framework the definition of illusions that we mentioned earlier (dissociation between the physical properties of the stimulus and the resulting percept). Importantly, several authors (Geisler and Kersten, 2002; Weiss et al., 2002) suggested that a whole range of illusory percepts, even those that are idiosyncratic in appearance (such as motion illusions), correspond to what an optimal Bayesian observer would perceive in similar conditions. This optimal systematic gap with sensory evidence could then also account for the illusory effect of figures for which the prior is not explicit or obvious. For example, the mechanism was proposed in different terms by Gregory when explaining the *Müller-Lyer illusion* (see **Figure 1**). According to this author, the internalized rules of perspective (i.e., prior knowledge) lead an individual to compute the line flanked with converging fins as if it were further away. Consequently, the line's shorter appearance conflicts with the actual identical size of the two lines.

(2) *Inconclusive sensory evidence.* In ambiguous figures, nicely illustrated by the *Necker Cube* or *Rubin's vase,* perception switches between two mutually exclusive interpretations, a phenomenon called *bistability*. Although Bayesian formalism can be applied to such ambiguous figures, their inclusion in the field of visual illusions is more questionable and needs closer examination (see **Box 2**).

Note that we do not claim here that all perceptual illusions, without exceptions, fall in these categories. Rather, we propose a framework that can be applied to most of the perceptual illusions considered in this review.

#### **VISUAL ILLUSIONS, BAYES AND PSYCHOTIC SYMPTOMS**

A growing field of theoretical and experimental approaches have related psychotic features, such as hallucinations and delusions, to a general disruption in the inferential process (Friston, 2005b; Fletcher and Frith, 2008; Schmack et al., 2013). By relying on the Bayesian formalism, authors typically impute the positive symptoms of schizophrenia to an imbalance in the relative weight or precision attributed to the prior and sensory evidence, which, in turn, causes false inferences. However, within this Bayesian framework, each model brings its own nuances in the way the system can be affected. The predictions sometimes clash or can be incompatible with each other. Importantly, any computational model should be sufficiently quantitative to enable the confirmation or rejection of the hypothesis based on experimental results and should be biologically plausible in terms of its mechanisms. Thus, it is important to consider in greater detail the type of Bayesian models and the possible sources of impairment that lead to schizophrenia symptoms for each of these approaches.

For example, according to the *Predictive Coding* model (see **Boxes 1**, **3**) of hallucinations and delusions, the system is altered in its metacognitive components, i.e., in the estimates of the beliefs' precision rather than in the beliefs' precision itself (Adams et al., 2013). From this point of view, a disruption in these estimates results in the allocation of an insufficient or excessive gain in prediction errors. Notably, delusions, which can be defined as false and inflexible cognitive beliefs, are thought to originate from an excessive gain, which indicates attributing too much confidence to sensory evidence compared with prior beliefs. Artificially over-trusted, sensory signals become over-salient and unpredictable. They could be transmitted up the hierarchy, but no adjustments could fully resolve the aberrancy (Fletcher and Frith, 2008). Failing to adapt, the only solution for the system to explain away the erroneous prediction errors and resulting chaotic sensory signals is to generate aberrant beliefs at the top of the hierarchy. Because they constitute the only way to make sense of lower level sensory phenomena, these beliefs would progressively become inflexible and impervious to contradictory evidence (Schmack et al., 2013). Thereby, delusions could be avatars of tenacious prior beliefs secondarily generated by the system to restore coherence in the perceptual world (Adams et al., 2013).

In contrast, the hypotheses that have been proposed for the emergence of hallucinations in reference to the *Predictive Coding* model are not as univocal. As for delusions, some authors related misperceptions to an inferential disequilibrium that favors sensory evidence. According to this assumption, the emergence of maladaptive percepts is closely linked to failures in the system of self-monitoring, i.e., the ability to correctly identify oneself as the source of one's own actions and thoughts. When one plans an action, one predicts its sensory consequences upon an "efference copy" of this plan sent to the sensory cortical areas. When the action is achieved, the actual associated sensations are compared with the expectations. If they match (i.e., no prediction error), the action is labeled as "self-generated," and the resulting sensations are attenuated. Conversely, a disrupted inferential system may fail to attenuate the sensory consequences of selfgenerated acts. Artificially overweighed, the resulting prediction error would then drive otherwise silent percepts to emerge into awareness. The only way to account for this unusual experience is then to misattribute it to an external agent. A key example is the delusional ideas regarding agency that arise from insufficient attenuation of the proprioceptive consequences of one's own movement (see Section What are the Links With Action Control and Motor Behavior?). Several authors have also related auditory hallucinations to an over-saliency of the inner speech (considered a covert motor action), which is misattributed to an alien source (McGuire et al., 1995; Allen et al., 2007; Moseley et al., 2013). However, in opposition to a theory of over salient sensory evidence, some authors have argued that excessively precise prior expectations, which correspond to a smaller-than-normal gain for prediction errors, better account for hallucinatory experiences. According to this hypothesis, perception distances itself from the sensory stimuli and becomes dependent on prior knowledge of the world (Friston, 2005b; Chambon et al., 2011; Schmack et al., 2013); thus, one would only perceive what one is expecting to perceive.

Note that the conflict between apparently contradictory hypotheses (gains of prediction errors larger than normal or

#### **Box 2 | Do visual illusions comprise ambiguous figures?**

A figure can be considered ambiguous when it provides sensory information equally supporting different interpretations. This results in a phenomenon called *bistability*: in observers, two mutually exclusive percepts stochastically switch. An archetypal example is the Necker Cube, which is alternately perceived as if it was viewed from above and from below (see **Figure 3A**).

Assessing whether ambiguous figures (considered physical stimuli) and their subsequently generated percept are dissociated, i.e., whether ambiguous figures fit with the definition of illusions that we propose, requires dissociating two conceptual levels. We will use the Necker Cube as support for our demonstration.


**FIGURE 3 | Necker Cube. (A)** The ambiguity results in the subjective impression of two interpretations switching: the phenomenon of bistability. **(B)** The shadow introduces a cue that is supposed to bias perception toward one interpretation. The fact that bistability persists despite the presence of the cue ensures this cube responds to the definition of a visual illusion, i.e., a dissociation between perception and the physical characteristics of its support.

smaller than normal, over-trusted prior or over-counted sensory evidence) has never been fully resolved, which renders the experimental testing of these theories extremely difficult. Finally, the neurophysiological processes that cause this imbalance remain unclear.

The *Circular Inference model* (see **Box 1**) attempts to overcome the contradictions by relating psychotic symptoms to a distributed excitatory-inhibitory imbalance (Jardri and Deneve, 2013). Inference in a hierarchical Bayesian system can be seen as a propagation of "bottom-up" messages (carrying sensory information) and "top down messages" (carrying top-down expectations). Posterior probabilities (and thus, percepts) are result of combining these two messages. Since long-range connections in the brain are overwhelmingly excitatory, these two types of messages would be reverberated endlessly through feedforward/feedback excitatory loops if they were not controlled, and kept in check, by the presence of equivalently strong inhibitory loops. Indeed such balance is tightly maintained in cortical networks, and appears to be affected in schizophrenia (O'Donnell, 2011). In their model, Jardri and Denève showed how excitatory/inhibitory imbalance renders the system unable to avoid circular propagation of beliefs: bottom-up sensory evidence are reverberated back down as if they were prior information (upward loops), and thus combined with themselves until weak sensory inputs or meaningless coincidences are attributed to highly trusted high-level interpretations. Vice versa, prior expectations can generate their own "fake" sensory represents, which then come back up and reinforce the prior expectations (downward loops), in the absence of any true corroborative sensory evidence.

Psychotic manifestations can be understood as resulting from such circular inferences, which cause overconfidence, surinterpretations of weak sensory data and dissociations between highlevel and low-level representations. This would be aggravated by an asymmetric impairment predominantly affecting either the upward or downward loops. Depending on which loops are mostly impaired, the model predicts that either sensory information or priors will dominate the final percept. This assumption is in line with the idea that hallucinations and delusions are two sides of a same coin (Fletcher and Frith, 2008). Even when facing weak or non-existent sensory evidence, the circular propagation generates strong perceptual beliefs: hallucinations occur where nothing relevant should have been inferred. In the same way, circular inference introduces spurious correlations between sensory (feedforward) and prior (feedback) messages that are non-existent in the real world. This leads to the learning and consolidation of "unshakable" (but false) causal relationships, resulting in delusional belief systems.

The *Circular Inference model* reconciles two eventualities that could have appeared to conflict in the *Predictive Coding*

#### **Box 3 | Predictive coding and Circular inference.**

#### **Predictive coding**

Predictive coding applies the Bayes rule while assuming that the prior and the likelihood have a Gaussian distribution. For example, if the prior has meanθ and variance σ<sup>2</sup> <sup>θ</sup> , and if the likelihood has mean x and variance <sup>σ</sup><sup>2</sup> <sup>x</sup> , then the posterior is a Gaussian distribution with mean x provided by

x =<sup>θ</sup> <sup>+</sup> *<sup>K</sup>* x −θ with *<sup>K</sup>* <sup>=</sup> <sup>σ</sup><sup>2</sup> θ σ2 <sup>θ</sup> <sup>+</sup> <sup>σ</sup><sup>2</sup> corresponding to the Kalman gain.

x Thus, the percept corresponds to the prior belief, corrected by a prediction error that corresponds to the difference between the sensation and its top-down prediction. The "salience" (Kalman gain) of the prediction error is a function of prior and sensory reliabilities. In a hierarchical network, this operation is repeated once for each layer, as schematized below (**Figure 4**):

**FIGURE 4 | Hierarchical inference with Gaussian variables.** In this toy example (left part), the inference that corresponds to the hidden cause (x) could be understood as the probability that the green color that I am observing (sensory evidence, represented by the magenta arrows) is due to the presence of a leaf, given my knowledge of the existence of a tree (prior expectation, represented by the violet arrows). Blue and yellow lines fit for the feedback and feedforward connections that enable the inferential process. Green and black circles fit for the controlling inhibitory system. The right part of the figure represents the probability distribution of each variable and the resulting posterior probability (in red).

#### **Circular inference**

To properly compute the probability of perceptual variables, the prior and likelihood must be multiplied only once. In the brain's hierarchy, top-down and bottom-up beliefs should be propagated only once in each direction (see figure above). This can be achieved if equally strong inhibitory loops exist to cancel excitatory feedforward/feedback loops (green and black units). If these inhibitory loops are impaired, beliefs are propagated multiple times, or, equivalently, the prior and likelihood are multiplied multiple times. The result is illustrated below (**Figure 5**):

**FIGURE 5 | Circular inference and relationship with predictive coding.** If both descending and climbing loops are impaired **(A)**, both sensory evidence and prior knowledge are reverberated and over counted (multiplication of the pink and violet arrows). In the *Predictive coding framework,* this results in an overconfident (narrowing of the posterior distribution) but not biased (unchanged K) inferred belief. In contrast, when the impairment only affects climbing loops **(B)**, sensory evidence, but not prior knowledge, is reverberated. The prediction error is emphasized (K is too large), and the inferred belief is biased toward sensory evidence. If, in contrast, the inhibitory disequilibrium disfavors the descending loops **(C)**, only the prior knowledge is over counted because of its reverberation. The prediction error is then minimized (K is too small), and the resulting posterior is biased toward expectations. Note that case **(B,C)**, the inferred belief is associated with an excessive degree of confidence.

*framework:* hallucinations may stem from an overweighting of either prior or sensory evidence. Thus, asymmetrically impairing the inhibitory loops biases the inferential system in two opposite ways that both may generate the same aspecific perceptual phenomenon. This raises the interesting challenge of assessing which one of the two models is the most suitable to account for the emergence of hallucinations, which may depend on the disease that is related to these symptoms, and possibly on a subjectper-subject basis. Regarding schizophrenia, the question is still not clear. The reference to VIs, however, will help to clarify the issue. Indeed, a deficit in the control of upward loops (e.g., sensory evidence that reaches a high level representation and is then misinterpreted as prior knowledge when it is reverberated back down) would cause an over-confidence in sensory evidence and a lower susceptibility to perceptual illusions. Vice-versa, a deficit in the control of downward loops (e.g., prior expectations that activate lower levels and are misinterpreted as sensory information when they are reverberated back up the hierarchy) would predict an over-confidence in the prior knowledge and a higher susceptibility to perceptual illusions. Here we review three lines of relevant findings.

We previously discussed that schizophrenia might be primarily associated with a lack of sensitivity to VIs (see Section Discussing the Limits of the Structural Approach). Several authors have empirically interpreted this phenomenon as a sign of a reduced top-down influence in perception. Some representative examples support this assumption. (1) In *context suppression* illusions, the patients, who rely more on the absolute properties of the stimulus, tend to resist the perceptual bias induced by prior belief influence. (2) Top-down expectations are thought to be primarily responsible for the *Apparent motion illusion,* in which two stationary stimuli alternatively flickering induce an impression of movement. Resistance to the illusion in schizophrenia patients was notably attributed to their incapacity to correctly use topdown processes in this situation (Sanders et al., 2013). (3) The fact that patients with schizophrenia failed to correct for the inverted *Hollow Mask* to a more plausible (or predictable) interpretation can also be explained by an underweighting of prior knowledge during perceptual inferences (Schmeider et al., 1996; Schneider et al., 2002; Koethe et al., 2009).

Aside from these behavioral findings, recent brain-imaging and electrophysiology studies have complementarily supported the assumption of an overweighting of sensory evidence in schizophrenia. In two recent papers, Dima et al. explored the neural mechanisms involved in the resistance to the *Hollow mask illusion* via an event-related functional magnetic resonance imaging procedure (fMRI; Dima et al., 2009) and an eventrelated potential procedure (Dima et al., 2010). Using dynamic causal modeling (DCM), the authors notably showed a significant between-group difference in the effective connectivity patterns measured during the VI task. While a model that places connectivity modulation from the higher-level areas to the primary visual cortex (i.e., V1) better accounted for healthy control data, the DCM revealed a reverse pattern in schizophrenia patients (i.e., the predominance of the feedforward modulation). Overall, these findings are compatible with the idea of different perceptual strategies in individuals with schizophrenia and controls when trying to minimize strong prediction errors. When facing a VI (i.e., a complex perceptual task), a tendency toward a top-down counter-balancing is observed in controls, whereas patients who suffer from schizophrenia exhibit a strengthening of the bottom-up processes that prevent them from perceiving the illusory effect.

Interestingly, using magnetic resonance spectroscopy, Yoon et al. revealed that patients who suffer from schizophrenia exhibited a reduced GABA concentration in the visual cortex compared with healthy controls (Yoon et al., 2010). Furthermore, the authors found that this reduction predicted better (less biased) performances in the *surround suppression* task by observing that the lower rates of GABA observed in patients correlated with their tendency to be more accurate in contrast discrimination despite a misleading context. More recently, Loon et al. also supported the contribution of GABA in visual perception by focusing on *bistability*. With a computational model of the assumed neural underlying interactions, the authors first predicted that a higher GABA concentration in the visual cortex would result in a slowdown of the perceptual switches. More importantly, the authors then experimentally confirmed these predictions by observing, after the systemic administration of a GABAA agonist (lorazepam), a lengthening of the percept durations and a decrease in the switch rates (Van Loon et al., 2013). Because GABA transmission reflects inhibitory neuronal modulations, these findings underpin the validity of circular probabilistic inference to account for perceptual impairments in schizophrenia by experimentally linking its assumed biological causes (excitatory-to-inhibitory imbalance) and the behavioral consequences (perceptual performances).

Overall, the heuristic value of VIs now appears clearer. Studying how patients cope with these simple stimuli provides access to underlying perceptual processes that could also account for the emergence of hallucinations and delusions (White and Shergill, 2012). In the case of schizophrenia, the evidence from the VI data converges toward the hypothesis of an asymmetrical belief formation that favors sensory evidence at the expense of prior knowledge. Among recently proposed models, the circular inference model appears particularly suitable to coherently link VIs, hallucinations, and their plausible biological causes. Indeed, schizophrenia resistance to VIs and susceptibility to hallucinations can be considered to result from the same circular inferential process in an ambiguous environment. This assumption has strong support in studies that identified negative correlations between illusion susceptibility and the presence of positive symptoms in healthy (Bressan and Kramer, 2013) and clinical populations (Keane et al., 2013; Sanders et al., 2013).

Importantly, while VIs provide a privileged access to the visual hallucinatory modality, readers should note that adult patients who suffer from schizophrenia are more concerned by auditory compared with visual hallucinations (Mueser et al., 1990; Blom, 2010). However, a recent review paper examined the clinical, phenomenological, psychological and physiological properties of visual hallucinations in schizophrenia compared with the same symptoms observed in Parkinson disease, Body-Lewy dementia or the Charles-Bonnet syndrome. The authors reported visual hallucinations with a substantial point prevalence of 27% (Waters et al., 2014). Mostly disregarded in the literature that examined hallucinations until recently, the visual modality is the subject of a renewed interest. Furthermore, we hypothesize that the demonstration we utilized for visual illusions and hallucinations could easily be transferrable to the auditory perceptual process. Moreover, this would be an interesting assumption to explore in future research.

Positive symptoms are not specific to schizophrenia. The possible occurrence of hallucinations and delusions in various psychiatric and neurological conditions or even in non-clinical populations suggests the relevance of a dimensional approach. Nevertheless, the heuristic value of VIs may open a path toward a new categorization based on the computational model that offers the best fit with particular perceptual disruptions. We effectively illustrated how an overweighting of sensory evidence may explain both hallucinations and the reduced susceptibility to VIs in schizophrenia. Interestingly, a trend to resist VIs has also been identified in autism (Happé, 1996), which suggests that an imbalance in processing toward sensory evidence could also offer a coherent and appropriate comprehensive framework for this developmental disorder (Pellicano and Burr, 2012). In contrast, some neurological disorders associated with hallucinations may exhibit an increased tendency to experience VIs. This tendency is noteworthy in the case of Parkinson's disease, in which one quarter of patients suffer from visual hallucinations and visual misperceptions (Diederich et al., 2009). Even if experimental validation is required, references to the *Circular Inference framework* (Jardri and Deneve, 2013) indicate this pattern potentially results from a reverse asymmetric impairment in the excitatoryinhibitory balance (i.e., an overweighting of prior information relative to the sensory evidence, which is caused by an insufficient inhibitory control of downward loops). This impairment could explain the concomitant increase in hallucinations and illusions in Parkinson's disease (as could the *Predictive Coding model*).

## **THEORETICAL AND CLINICAL IMPLICATIONS FOR SCHIZOPHRENIA RESEARCH**

#### **WHAT ARE THE LINKS WITH PSYCHOPATHOLOGY?**

The resistance to VIs in schizophrenia patients leads to several etiopathological implications and may drive several new experiments. For example, it appears possible to examine the correlations between this lack of susceptibility and different clinical features. Despite an abundant literature dealing with this matter, frequent methodological issues (e.g., inter-group comparability) have made the findings difficult to interpret. Three main approaches may be individualized:

• Examining the potential links between VI sensitivity and symptom severity (primarily using the PANSS scale) first provided discrepancies in the findings because these scores were computed as covariates. While some authors found no or only weak relationships between VIs and psychopathology (Koethe et al., 2006, 2009; Tibber et al., 2013; Yang et al., 2013), other authors identified an inverse correlation with either positive (Norton et al., 2008; Keane et al., 2013; Silverstein et al., 2013) or negative (Tadin et al., 2006; Silverstein et al., 2013) symptom dimensions. To draw valid conclusions, designs specifically focusing on the clinical correlates of VI insensitivity in clinical populations, with more precise and specific symptom assessments, appear necessary.


vulnerable he will be to delusional beliefs and hallucinations. Interestingly, the opposite profile of asymmetric circular belief propagation (selective impairment of upward loops) predicts opposite (increased) vulnerability to VIs but similar symptoms. Thus, the possible reverse pattern of dissociation between prior and sensory evidence leads one to consider the corollary proposal: the more weight that is attributed to prior knowledge, the more vulnerable the subject is to both VIs and hallucinations. This proposal may account for the gradual increase in susceptibility to VIs that accompanies the progressive emergence of misperceptions in some psychiatric or neurological conditions, such as Parkinson's disease (see also Section Visual Illusions, Bayes and Psychotic Symptoms).

#### **WHAT ARE THE RELATIONS WITH SUBJECTIVE EXPERIENCES?**

Considering the *Circular Inference* model, aberrant experiences in schizophrenia may be understood as an immersion in a world dominated by sensory evidence, while top-down influences lose their organizing and structuring potential. When the ambiguity is particularly strong, VIs represent an extreme perceptual task. As such, their study provides an emphasized insight into the phenomenology of perception. Although the hypothesis requires further investigation, we can assume that during the vast majority of daily life situations, patients resolve the moderate ambiguity of the environmental stimuli through strong and unambiguous percepts but with weakened links and coherence between perceptual elements (links typically implicitly driven by prior knowledge). The perceptual world would then appear as fragmented and less meaningful (Adams et al., 2013). Hallucinations would only emerge when facing highly uncertain (and normally irrelevant) sensory information. Importantly, because the necessity to adapt to highly ambiguous situations is not permanent, this may also account for the intermittent nature of hallucinations (Jardri et al., 2013).

This model may account for the cognitive deficits observed in schizophrenia, such as the patients' difficulties in correctly allocating attention and filtering out irrelevant information. This phenomenon can also be explained by the lack of prior influence on saliency. The weakening of the downward beliefs blurs the distinction between relevant and noisy items, which makes them almost equally surprising.

#### **WHAT ARE THE LINKS WITH ACTION CONTROL AND MOTOR BEHAVIOR?**

The question of whether the highlights provided by VIs help us understand behavioral features in schizophrenia requires one to consider both the complex links between perception and action and the possible common causes for their disturbances. In this regard, paradigms that test an illusory effect via visuomotor performances are thought to engage a complex cross-modal coordination and, thus, are of particular interest in schizophrenia (Pessoa et al., 2008; Chen et al., 2011).

Moreover, several lines of evidence suggest that the tendency to resist illusions observed in schizophrenia is not limited to the visual modality. Shergill et al., for example, studied the *Force matching illusion,* which consists of the systematic underestimation of a self-generated force deployed to match an externally applied target force (Shergill et al., 2005). In normal conditions, a system of self-monitoring enables the prediction of the sensory consequences of one's motor acts. In Bayesian terms, the predictions, which arise from prior knowledge, regarding the sensory outcome of one's own action permit a reduction in the weight attributed to the matching sensory evidence, and hence, attenuate the perception related to this sensation. The authors indicated that patients who suffered from schizophrenia were more accurate compared with controls when matching the externally applied force, which revealed a failure in the normal sensory attenuation mechanism. This finding outstandingly fit with the hypothesis of a false inferential process overtook by sensory evidence. Brown et al. (2013) went further in a recent paper by demonstrating how perception and action were derived from the same Bayes optimal system. Referring to the notion of *active inference*, the authors considered movement as a way to actively minimize proprioceptive prediction errors. However, this process is conceivable only if it is combined with a reduction of precision of sensory evidence to avoid conflict between action and perception. Using a probabilistic generative model, the authors predicted that a failure in attenuating sensory proprioceptive evidence led to the emergence of delusional ideas regarding agency (the ability to correctly identify oneself as the cause of one's own actions), as well as a decreased susceptibility to the *Force-matching illusion*. These findings are in accordance with previous experimental reports of resistance to VIs in schizophrenia but extend this observation to multisensory perception.

Overall, the Bayesian framework predicts that false inferences, which are biased by overweighted or insufficiently attenuated sensory evidence, may coherently (1) account for both the visual and proprioceptive perceptual changes in schizophrenia, (2) closely link these changes with action, and by extension, behavioral disruptions, (3) explain the emergence of hallucinations and delusional beliefs, and (4) provide an heuristic value to the vulnerability to illusions, which can be considered an indirect but valuable access to the global neural processing in this disorder.

#### **CONCLUSION**

Under the Bayesian scope, VIs acquire tremendous heuristic value by providing new insights into the perceptual processes that underlie misperceptions. The literature that pertains to VIs has paved the way for new hypotheses regarding psychiatric and neurological conditions (for example, overcounting of sensory evidence in schizophrenia vs. prominence of prior knowledge in Parkinson's disease). Moreover, through the probabilistic framework, VIs are an indirect but promising approach to understand several schizophrenia features as coherently emerging from the same inferential process. The research avenue may benefit from a more rigorous methodological approach, particularly by resorting to more precise classifications and conceptual definitions.

#### **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 February 2014; accepted: 23 July 2014; published online: 12 August 2014. Citation: Notredame C-E, Pins D, Deneve S and Jardri R (2014) What visual illusions teach us about schizophrenia. Front. Integr. Neurosci. 8:63. doi: 10.3389/fnint. 2014.00063*

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

*Copyright © 2014 Notredame, Pins, Deneve and Jardri. 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.*

# PART III – VISUAL TRAINING PROGRAMS IN ORGANIC DEFICITS AND THEIR NEURAL BASES

## Implications of CI therapy for visual deficit training

### *Edward Taub\*, Victor W. Mark and Gitendra Uswatte*

University of Alabama at Birmingham, Birmingham, AL, USA

#### *Edited by:*

Olivier A. Coubard, CNS-Fed, France

#### *Reviewed by:*

Annette Sterr, University of Surrey, UK Daniel-Robert Chebat, The Hebrew University of Jerusalem, Israel

#### *\*Correspondence:*

Edward Taub, University of Alabama at Birmingham, Birmingham, AL 35294, USA e-mail: etaub@uab.edu

We address here the question of whether the techniques of Constraint Induced (CI) therapy, a family of treatments that has been employed in the rehabilitation of movement and language after brain damage might apply to the rehabilitation of such visual deficits as unilateral spatial neglect and visual field deficits. CI therapy has been used successfully for the upper and lower extremities after chronic stroke, cerebral palsy (CP), multiple sclerosis (MS), other central nervous system (CNS) degenerative conditions, resection of motor areas of the brain, focal hand dystonia, and aphasia. Treatments making use of similar methods have proven efficacious for amblyopia. The CI therapy approach consists of four major components: intensive training, training by shaping, a "transfer package" to facilitate the transfer of gains from the treatment setting to everyday activities, and strong discouragement of compensatory strategies. CI therapy is said to be effective because it overcomes learned nonuse, a learned inhibition of movement that follows injury to the CNS. In addition, CI therapy produces substantial increases in the gray matter of motor areas on both sides of the brain. We propose here that these mechanisms are examples of more general processes: learned nonuse being considered parallel to sensory nonuse following damage to sensory areas of the brain, with both having in common diminished neural connections (DNCs) in the nervous system as an underlying mechanism. CI therapy would achieve its therapeutic effect by strengthening the DNCs. Use-dependent cortical reorganization is considered to be an example of the more general neuroplastic mechanism of brain structure repurposing. If the mechanisms involved in these broader categories are involved in each of the deficits being considered, then it may be the principles underlying efficacious treatment in each case may be similar. The lessons learned during CI therapy research might then prove useful for the treatment of visual deficits.

**Keywords: CI therapy, neuroplasticity, neurorehabilitation, motor deficits, visual deficits, aphasia, learned nonuse, sensory nonuse**

#### **INTRODUCTION**

In a previous article it was suggested that the principles that underlie the rehabilitation of deficits in movement and speech by Constraint Induced (CI) therapy training are similar to the improvement by training of visual deficits in amblyopia (Taub, 2010). Both types of remediation probably have one mechanism in common, overcoming learned nonuse, and perhaps another, neuroplastic change. In the present article, the possibility will be raised that similar principles may underlie efficacious rehabilitation training to improve other visual deficits and perhaps deficits in other sensory systems, such as somatic sensation.

In the first half of the article the origin, methods, and representative results of CI movement therapy will be described, as well as the extension of this rehabilitation method from motor deficit after stroke to motor deficit after traumatic brain injury (TBI), cerebral palsy, and multiple sclerosis (MS). The extension of CI therapy from stroke to these other conditions was based, actually predicted, by the learned nonuse formulation. The application of the CI therapy approach, again based on the learned nonuse formulation, to a non-motor neurological disorder, poststroke aphasia, will be presented. This will be followed by an analysis of amblyopia training, outlining the detailed similarity

of methodology and type of results obtained with this efficacious treatment for a sensory deficit to those of CI therapy. Next the research on training techniques used for the rehabilitation of two common visual deficits resulting from brain damage, unilateral spatial neglect, and visual field defects, will be summarized briefly. Given the prior effectiveness of CI therapy with two non-motor deficits, aphasia and particularly the visual deficit associated with amblyopia, it will be suggested that the application of a CI therapy approach to unilateral neglect and visual field defects might enhance the impressive effects that have been achieved in their treatment to date.

In the past, evidence has been presented to indicate that at least two mechanisms underlie the operation of CI therapy. One mechanism, use-dependent neuroplastic alteration in cortical territorial function, would presumably remain the same for both CI motor therapy/CI aphasia therapy and the rehabilitation of visual deficits. The concept of learned nonuse, however, would have to be broadened to include "sensory nonuse," both learned motor nonuse and sensory nonuse having in common the development of diminished neural connections (DNCs) after damage to the brain, or nonuse of a function, or both; their rehabilitation then could be referred to as DNC strengthening. Importantly, the concept of DNC strengthening provides a bridge connecting

overcoming learned nonuse/sensory nonuse on the one hand and use-dependent cortical reorganization on the other.

The plan of the article then is to first review CI therapy and then to suggest the potential value of applying the principles and general procedures of the CI therapy protocol to visual deficits. Considerable space is spent on describing CI therapy since that is what is known. It is hoped that the detail of this account can provide a platform upon which appropriate experiments can be designed for developing new treatments for the rehabilitation of visual deficits.

## **CI THERAPY**

#### **ORIGIN – PRIMATE STUDIES ON SOMATOSENSORY DEAFFERENTATION**

Constraint Induced therapy is derived from basic behavioral neuroscience research. It was an early observation in neuroscience research that when a single forelimb in monkeys is surgically deprived of somatic sensation by serial dorsal rhizotomy, the animal does not make use of the deafferented extremity in the free situation. This phenomenon was observed repeatedly in the history of neuroscience (Mott and Sherrington, 1895; Lassek and Moyer, 1953; Twitchell, 1954) and the nonuse was long thought to be irremediable. However, we found that monkeys can be induced to use their deafferented extremity by one of two behavioral techniques (summarized in Taub, 1977, 1980). One technique is training; the type of training that was found to be particularly effective is termed shaping. Shaping is an operant conditioning training method in which a desired behavioral or other functional objective is approached in small steps, by "successive approximations," so that the improvement required for successful performance at any one point in the training is small (Skinner, 1938, 1968; Taub et al., 1994). The actions shaped in the primate deafferentation experiment included (a) pointing at visual targets (Taub et al., 1975a) and (b) prehension in juveniles deafferented on day-of-birth (Taub et al., 1973) and prenatally (Taub et al., 1975b) who had never exhibited any prehension previously. The second behavioral technique was restricting movement of the intact limb. The monkey was thereby forced to either use the deafferented extremity or be rendered virtually helpless, unable to ambulate, climb, or grasp food to feed itself. The monkey may not have used the affected extremity for several years, but the application of either of these two techniques resulted in a striking conversion of a useless forelimb into a limb that was used for a wide variety of purposes. The movements were not normal; they were clumsy since somatic sensation had been abolished, but they were extensive and effective. When the experimental manipulations were more than just transient, the reversal was permanent, persisting for the remainder of the animal's life.

The nonuse of a single deafferented limb had been used by Sherrington, based on his experiment in 1895 with Mott (Mott and Sherrington, 1895), as one of the primary foundations of his influential reflexological position in which reflex mechanisms were viewed as the basic building blocks underlying purposive movement (Sherrington, 1931). Since the monkeys in the above-noted experiments could be induced to make extensive and effective purposive movements with a limb from which all same segment reflex arcs had been abolished, the reflexological explanation of

movement could not be correct. To explain the nonuse of an extremity that followed its deafferentation and the subsequent overcoming of that nonuse by one of the two techniques that we employed, the learned nonuse formulation was developed (Taub, 1977, 1980; Taub et al., 2006b). In brief, the monkeys were said to not use the deafferented limb because they learned not to use it in the early postoperative period. The behavioral "contingencies of reinforcement" that established the nonuse condition persisted throughout the lifespan, and thus so did the nonuse of the limb, but this could be overcome by a technique which consistently increased the motivation to use the limb, such as either of the two techniques that we employed. (For a fuller account of this formulation, see "Learned Nonuse" section below). A fundamental aspect of the formulation was that learned nonuse should occur after any substantial damage to the central nervous system (CNS). The formulation was developed to explain the results after single forelimb deafferentation in monkeys and its first application in humans was to stroke. However, from the outset the formulation was meant to apply to deficits that followed any type of damage to the CNS which, depending on the location of the CNS damage, could be motor, visual, or possibly some other sensory system (and to some peripheral somatic injuries as well, e.g., fractured hip, arthritis).

#### **APPLICATION OF THE REHABILITATION PROTOCOL DEVELOPED WITH DEAFFERENTED MONKEYS TO HUMANS AFTER STROKE**

The conversion of a useless extremity to a limb that was used extensively on a permanent basis clearly constituted a strong form of rehabilitation, though the term rehabilitation was not used in relation to animals at the time. When considering the primate results in this light, the application of the techniques employed with primates to the rehabilitation of function after CNS damage in humans seemed plausible. As noted, the first application was to stroke. The original two components used with monkeys were kept basically intact, but the human environment and motivation structure differs substantially from that of primates in an animal colony. Therefore, in making the translation from monkeys to humans two new elements were added. The four components of the treatment that have been used in all of its applications with humans are as follows (Taub,2004; Morris et al.,2006; Taub and Uswatte,2006; Taub et al., 2006a): (1) intensive training of the more-affected arm for multiple days; (2) training with the behavioral technique termed shaping; (3) the transfer package (TP), a set of behavioral techniques designed to facilitate transfer of therapeutic gains from the treatment setting to daily life; and (4) discouraging compensatory activities that are employed to avoid using the impaired function. For unilateral upper extremity motor deficit after stroke or other hemiparetic illnesses (as well as somatosensory deafferentation of a limb in monkeys) the intact limb is prevented from doing the work of both arms by restraining it. For aphasia after stroke, communication by gesture or any other non-verbal means is strongly discouraged; there is no physical restraint. There is also no physical restraint used for lower extremity CI therapy.

The training method of shaping involves approaching a behavioral objective in small steps by "successive approximations" (i.e., a task is gradually made more difficult with respect to a participant's motor capabilities). Its principles were explicitly formulated by Skinner (1938, 1968) and they have been applied to the rehabilitation of movement in this laboratory (Taub et al., 1993, 1994). For rehabilitation, shaping involves (a) providing immediate and very frequent feedback concerning improvements in the speed and quality of movement or speech, (b) selecting tasks that are tailored to address the deficits of individual participants, (c) modeling, prompting, and cuing of task performance, (d) systematically increasing the task difficulty in small steps when improvement is present for a period of time, and (e) motivating the participant to improve performance by what might informally be called "cheerleading." In this laboratory shaping has two distinct levels: (1) Improving the speed and quality of movement from trial to trial within a task with frequent feedback and encouragement being given. (2) Introducing a new task that is similar to but more difficult than the one being used when motor performance improves to the point where the therapist feels that the new task can be accomplished by the participant. The procedure employed here involves use of both levels of shaping but focuses more attention on improving within-level task performance.

The TP consists of a set of techniques commonly used in the behavior analysis field (Taub, 2012, 2013) to treat a variety of conditions for such problems as medication adherence, adherence to an at-home exercise regimen for low back pain, drug addiction treatment, addiction relapse prevention, and alteration of autism spectrum behaviors; but they have not been used systematically in physical rehabilitation. The TP techniques used here are: behavioral contracts, daily home diary, tracking amount and quality of use of the more-affected arm in 30 important activities of daily living (ADL) in the life situation by daily administration of the Motor Activity Log (MAL), problem solving to overcome perceived barriers to more-affected arm or language use in ADL performance, written assignment during treatment of practice to be carried out at home of use of the more-affected arm in specified ADL (home skill assignment), post-treatment home skill practice assignments, weekly telephone calls for the first month after laboratory treatment in which the MAL is given and problem solving carried out. For further description of the TP techniques and a demonstration of its efficacy see Taub et al. (2013b) and its online supplement.

Taub and coworkers (Taub et al., 1993, 2006a) applied this protocol to the rehabilitation of persons with a chronic upper extremity hemiparesis in two randomized controlled trials (RCTs) that employed attention-placebo control groups and emphasized transfer of therapeutic gains in the laboratory to the life situation. Patients with chronic, rather than acute, stroke were targeted as subjects for this study because in the primate deafferentation research substantial motor rehabilitation was possible well into the chronic phase. In addition, according to the research literature at the time (Twitchell, 1951; Bard and Hirschberg, 1965; Parker et al., 1986), and almost universal clinical experience, spontaneous motor recovery was thought to plateau within 1 year after stroke. There was no evidence that any treatment could produce further recovery of function more than 1 year after stroke. Therefore, any marked improvement in the motor function of individuals with chronic stroke would be of particular therapeutic significance.

After a long-standing plateau, the probability would be very low that an abrupt, large improvement in motor ability could be due to spontaneous recovery.

#### **LARGE DIFFERENTIAL EFFECT OF THE TREATMENT ON (1) IMPAIRMENT AND MOTOR CAPACITY TESTED IN THE LABORATORY AND (2) ACTUAL ADL PERFORMANCE IN THE LIFE SITUATION**

Over 1000 adult and pediatric patients with chronic stroke with mild/moderate motor deficits (grade 2; an estimated 25% of the chronic stroke population) have been given upper extremity CI therapy in this laboratory and clinic. On the level of impairment, active range of motion (AROM) was assessed for 26 movements in adults distributed among each of the arm joints. Of the joints whose movements were outside of normal limits prior to treatment, 44% improved more than 5◦. The mean improvement for these joints was 22% relative to their pretreatment joint angle value. Motor capacity was measured by the Wolf Motor Function Test (WMFT; Wolf et al., 1989, 2005; Taub et al., 1993; Morris et al., 2001), a laboratory motor function test in which the tester requests that the subjects make the best movements of which they are capable in 15 timed tasks. The effect size (ES) of the mean preto post-treatment change was *d* = 0.9 for this measure. Actual performance of ADL in the life situation was measured by the MAL (Taub et al., 1993, 2006a) a structured, scripted interview with established reliability and validity (van der Lee et al., 2004; Uswatte et al., 2005b, 2006b). The mean ES (*d* ) for the MAL was 3.3. The treatment change for ADL in the life situation from one RCT (Taub et al., 2006a) is shown in **Figure 1**.

The much larger ES for the MAL than for the WMFT indicates that CI therapy has its greatest effect on increasing the actual amount of use of a more-affected upper extremity in the real-world

**FIGURE 1 | Mean MAL arm use scores from CI therapy (***n* **= 21) and placebo control (***n* **= 20) patients with chronic stroke.** CI therapy subjects showed a very large improvement in arm use outside the laboratory from pretreatment to post-treatment (1.8 ± 0.6; P < 0.0001; d' = 3.0). Before treatment the data indicate that these patients were using the more affected arm 14% as much as before stroke, while after 2 weeks of treatment it was 52%, an almost four times increase. Controls showed little change. CI therapy subjects retained all of their immediate treatment gains 4 weeks after therapy and showed only a 23% decrease after 2 years from post-treatment levels of real-world arm use. At 1-year post-treatment the loss in retention was 14%. Reprinted from Taub et al. (2006a).

setting, though the improvement in maximal motor capacity as indexed by the WMFT is still substantial. In the meta-analysis literature, an ES (*d* ) of 0.2 is considered small, a 0.4–0.6 *d* is moderate, while *d* s of 0.8 and above are large (Cohen, 1988). Thus, the ES of CI therapy for improvement in motor function on a laboratory test where a best effort is requested is large, but for real-world outcome in patients with chronic stroke the ES is extremely large. This differential effect would appear to be due to the ability of CI therapy to overcome the "learned nonuse" that frequently depresses the spontaneous use of a more-affected arm after CNS damage. It should be emphasized, though, that even though CI therapy has its largest effect on ADL function in the real world, it also has a large effect on both the impairment level and on performance on a laboratory motor function test. This will be an important consideration for the applicability of the CI therapy protocol to visual deficits.

#### **NAMING THE TREATMENT: CONSTRAINT-INDUCED MOVEMENT THERAPY (CI THERAPY)**

The movement-restriction and training situations of CI therapy share a common feature. They both are powerful means of inducing use of the more-affected arm. One procedure physically *restrains* the less-affected arm so that the individual must use the more-affected extremity to avoid being rendered more dependent or, in the case of the unilaterally deafferented monkeys, virtually helpless. The other method, training, induces use of the more-affected arm by structuring a situation so that the limb must be used in order to achieve success or avoid failure. Thus, both procedures constitute *constraints* that promote use of the more-affected arm by a major alteration of environmental conditions. Though the name of the treatment is accurate, the use of the term "constraint" in the title of the therapy has turned out to be confusing. The most salient aspect of CI therapy for the upper extremity to a casual observer is that the less-affected arm is *restrained*. Moreover, the rehabilitation field was not used to thinking of training as imposing a constraint on behavior. Instead, the large majority of professionals interpreted the focal word in the name of the therapy as being an alternate way of saying "restraint." Thus, the general impression arose that restraint of the less-affected arm was the central and most important feature of the therapy. That is very far from being true; physical restraint of the less-affected arm can be dispensed with entirely in achieving a maximal result if the training conditions are arranged appropriately.

Variants of upper extremity CI therapy that do not involve physical restraint of the less affected arm have been found to be as efficacious as the initial protocol (Taub et al., 1996, 1998; Taub and Uswatte, 2000; Uswatte et al., 2006a). These include (1) placement of a non-restrictive half-glove (with fingers cut off) on the lessaffected arm as a reminder not to use it and shaping of the paretic arm, and (2) shaping of the paretic arm only (Uswatte et al.,2006a). The half-glove intervention was designed so that CI therapy could be employed with patients who have balance problems and might be at risk for falls when wearing a sling; this intervention expanded the population of stroke patients amenable to CI therapy threefold. Currently, a padded or protective safety mitt is used instead of an arm sling. This restraint leaves the less-affected arm free so that it can be used for defense in case of a fall, but prevents use of the hand and fingers in ADL. Thus, there is nothing talismanic about use of a restraint device. Any rehabilitation technique that requires that the more-affected arm be used extensively should be efficacious.

#### **IMPORTANCE OF TRANSFERRING TREATMENT GAINS FROM CLINIC TO LIFE SITUATION: ROLE OF THE TRANSFER PACKAGE**

In most rehabilitation regimens, participants carry out exercises guided by a therapist during treatment sessions, and then the treatment usually stops. The TP of techniques makes patients more active participants in their own improvement, not only during the treatment sessions but also at home. The TP provides a systematic means of specifying explicitly what the participant is expected to do when outside the treatment setting, monitors what in fact is done, and provides a structure within which to solve apparent barriers to carrying out treatment goals. Thus, the TP immerses participants in a therapeutic environment for a meaningful portion of their day. Therapy is not confined to the limited period that the current health care system with its reimbursement limits permit.

A 2 × 2 factorial experimental components analysis of CI therapy was carried out to assess the relative contribution made by the TP and shaping to the magnitude of the treatment effect (Taub et al., 2013b). Two levels, presence versus absence, of each of these treatment factors, TP and shaping, were tested. Thus, there were four groups: (1) the full CI therapy package, which includes the TP and shaping, (2) the CI therapy package minus the shaping component only, (3) the CI therapy package minus the TP component only, and (4) the CI therapy package minus the shaping and TP components. In other words, all four groups received the same amount and intensity of training and wore a padded mitt preventing use of the more-affected hand during training in the laboratory, but they varied in whether they received training on upper-extremity tasks in the laboratory with shaping or with practice on tasks at the same level of difficulty throughout treatment; they also varied in whether they received the TP or not. Spontaneous use of the more-affected arm in daily life and maximum motor capacity of that arm in the laboratory were assessed with the MAL and the WMFT, respectively.

Use of the TP, regardless of the type of training received, resulted in MAL gains that were 2.4 times as large as the gains in its absence (*P* < 0.01; **Figure 2**). The MAL gains were retained without loss 1 year post-treatment. An additional substudy (*N* = 10) showed that a single component of the TP, weekly telephone contact with participants for 1 month after treatment, doubled MAL scores at 6-month follow-up, closing half the quantitative gap in real-world spontaneous movement between the two groups. Thus, the TP would appear to be a method for strongly enhancing spontaneous use of a more-affected arm in the life situation. After treatment, voxel based morphometry (VBM) analysis of MRI scans indicated that the two TP groups also exhibited a profuse increase in gray matter in the sensorimotor cortices, more anterior motor areas, and the hippocampus in both hemispheres (**Figure 3A**; Gauthier et al., 2008). The groups not receiving the TP showed no change in amount of gray matter after treatment. Additional results from this experiment

indicated that shaping improved maximal motor performance made on request in the WMFT, but while it also improved real world spontaneous use of the more-affected limb significantly, it did so only a fraction as much as when the TP was used in addition.

The question might arise as to whether the TP increases treatment effect by simply increasing the amount of practice of more-affected arm use, since more-affected arm use is strongly encouraged in life situation ADL as well as in the clinic. Alternatively, it is possible that the TP promotes integration of therapeutic gains achieved in the laboratory into real-world activities so that more-affected arm use becomes habitual. These two possibilities are not mutually exclusive. Addressing this question in future research would be of mechanistic and theoretical interest; however from the point of view of practical therapeutics, the resolution of this important issue does not really matter. The TP appears to be a means of increasing real-world treatment outcome that does not involve increasing costly therapist time; this would be of considerable value whatever the mechanism by which the TP achieved its effect.

#### **CI THERAPY IN OTHER LABORATORIES**

At UAB, over 1000 patients with stroke have been given one variant or another of CI therapy and all but four of these patients have demonstrated substantial improvement in motor ability (i.e., improvement greater than a minimum clinically important difference; Lum et al., 2004; van der Lee et al., 2004; Uswatte and Taub, 2005; Lang et al., 2008). There have also been approximately 600 papers from other laboratories on adult and pediatric CI therapy published to date. To our knowledge all but two of the studies have reported positive results. In particular, CI therapy was the subject of a multi-site RCT (Wolf et al., 2006); the results were strongly positive.

Some of the papers on CI therapy from elsewhere report outcomes as large as those obtained in this and other laboratories; however, most of these studies report results that are significant, but only one third to one half as large as those obtained here. The likely reasons for the reduced treatment effect in these laboratories are twofold: (1) there was incomplete or complete lack of use of the procedures of the TP, which, though reported in the papers from this laboratory, had been largely ignored. As noted above, we have replicated the reduced treatment effect obtained by others

by duplicating everything that is normally done in treatment here except implementation of the TP (Gauthier et al., 2008; Taub et al., 2013b). (2) Another probable reason for obtaining reduced effects is that a protocol with attenuated intensity (tasks or movements per unit time) was used, such as in a study by van der Lee et al. (1999).

This laboratory's results have been replicated with patients with chronic stroke in published studies from four laboratories where therapists were trained at UAB: the laboratories of Miltner and Bauder (Miltner et al., 1999), Flor and Kunkel (Kunkel et al., 1999), Elbert and Sterr (Sterr et al.,2002), and Dettmers andWeiler (Dettmers et al., 2005); in the first three studies, CI therapy was set up with the collaboration of one of us (E.T.) and then monitored twice yearly. In all four of these studies some but not all elements of the TP were employed. However, in each case attention was focused on the transfer of therapeutic gains in the laboratory to spontaneous use of the more-affected arm in the life situation and some TP elements were used.

The reason for going into this detail in describing the TP and the range of results obtained in other laboratories with what is designated by those authors as CI therapy, is that the same issues probably pertain to the treatment of sensory deficits by training as they do to CI movement therapy. The same consideration applies to some of the sections that follow.

#### **APPLICATIONS OF CI THERAPY** *Severity and chronicity of motor deficit*

Most of the patients treated at UAB could be characterized as having deficits that were mild/moderate. Experiments have also been carried out with patients with moderate and moderately severe deficits (grades 3 and 4; Taub et al., 1999). Their treatment change was somewhat less than for higher functioning patients, e.g., increases of approximately 400% and 350% for patients with moderate and moderately severe deficits, respectively, compared

to approximately 500% for patients with mild/moderate deficits, but the treatment changes were nevertheless very large. Recently, work has been carried out with patients with useless, plegic hands that were initially fisted (Uswatte et al., 2008; Taub et al., 2013a). At the end of treatment, the patients exhibited a 186% improvement in the real-world use of the more-affected arm. It had been converted into a useful "helper" in the life situation. We estimate that CI therapy is applicable to at least 50% of the chronic stroke population with motor deficit, perhaps more.

There does not appear to be any upper limit to the length of time since stroke and the benefit that is obtained from CI therapy. There is also no correlation between chronicity and magnitude of treatment effect. The patient with the longest time period between the event and treatment was 50 years; stroke occurred at 5 years of age and treatment was given at 55. The improvement in moreaffected arm movement was in the middle of this laboratory's range. There is also no correlation between age and treatment outcome. Several patients have been treated in their 90s, and each one had a substantial improvement in motor function.

### *Lower extremity*

CI therapy techniques have been applied to the more-affected lower extremity of stroke patients (Taub et al., 1999). The treatment (Lower Extremity-CI therapy or LE-CI therapy) consists of massed or repetitive practice of lower extremity tasks (e.g., over-ground walking, treadmill walking with and without a partial body weight support harness, sit-to-stand, lie-to-sit, stair climbing, walking over obstacles, various balance and support exercises). Task performance is shaped as in the upper extremity protocol. No restraining device is placed on the less-affected leg. The lower-extremity procedure is considered to be a form of CI therapy because of the use of the TP, the strong massed practice/shaping element, and because the reward of adaptive patterns of ambulation over maladaptive patterns in our training procedure constitutes a significant general form of constraint. This is another form of CI therapy where constraint is imposed by the training paradigms but there is no physical restraint.

#### *Retention*

The treatment effect is long-lasting. For the upper extremity, its persistence is related to the severity of the initial impairment. For patients with mild/moderate initial impairment – grade 2 (but who entered treatment because of markedly reduced use of the more-affected arm in the real-world environment; i.e., MAL score <2.5), retention at 1 year after treatment varied in different experiments around the 90% level; at 2 years retention was approximately 80%. For patients with moderate initial impairment (grade 3), retention at 1 year was 75%; for patients with moderately severe impairment (grade 4), retention after 1 year was 60%; and for patients with severe deficit (grade 5) with initially plegic hands, retention across two experiments was 46%. The situation for the legs is quite different. Retention after 1 year is approximately 100% for all grades of motor deficit studied, from mild/moderate to patients who are minimally ambulatory. Some patients do exhibit a decrement in performance at 1 year, but others continue to improve above their immediate post-treatment level. One might speculate that to ambulate and for most lower extremity functions, both legs must be used, the more-impaired as well as the less-impaired extremity. Thus, use of the more-affected leg is maintained, and retention of the treatment effect is high. For the arms, however, just one extremity can be used for many tasks, and since use of the more-impaired arm remains effortful, a part of the conditions that initially gave rise to learned nonuse remain in effect; thus, reduced use of the more-affected extremity gradually returns. The greater the effort to use the more-affected arm (i.e., the greater the impairment from grade 2 through grade 5), the greater the reduction in real-world arm use over time; so that a smaller amount of the treatment effect is retained. The same retention results as for the upper extremities with respect to retention of treatment effect are likely to apply to visual deficit training.

#### *The broad range of applications to other motor disorders following CNS injury*

The finding that after unilateral forelimb deafferentation in monkeys, training of the deafferented limb or restraint of the intact limb enabled extensive purposive movement in a previously useless upper extremity undermined the reflexological interpretation of movement. The need to explain the initial nonuse of the deafferented limb even though motor pathways were still intact, and the way this nonuse could later be overcome gave rise to the learned nonuse formulation. This formulation is predicated on the fact that any substantial damage to the CNS generally results in reduced excitability of neural structures with extensive connections to the directly injured area. This loss of excitability in turn results in a deficit in or complete loss of the function that the defacilitated neural tissue had previously supported. However, levels of excitability slowly recover spontaneously (Swayne et al., 2008; Takechi et al., 2014), and this is reflected in the process of spontaneous recovery of function. While monkeys with deafferented forelimbs were never observed to spontaneously recover the ability to use the affected forelimb purposively, the two techniques already described enabled extensive use of the deafferented extremity. Though these two techniques had to be applied for a period of days in the case of restraint of the intact limb or over multiple sessions with training for purposive movement to become long-term, the fact that purposive movement could be evoked in very short periods of time, hours in the case of physical restraint and several 1-h sessions in the case of training, suggested that the excitability of neural tissues had been restored previous to these interventions, after the initial period of reduced CNS excitability had dissipated. Why then was there no return of the motor function that should have been possible? To account for this situation, the learned nonuse formulation was developed (Taub, 1977, 1980; Taub et al., 2006b). However, though developed to explain this specific set of experimental facts, learned nonuse was viewed as a general phenomenon independent of specific neural structures. It would not matter from the point of view of the formulation, if a motor deficit were due to the abolition of an extremity's afferent supply, or to a lesion in any of a number of different locations in the brain or spinal cord. These would presumably include lesions in: motor cortex, descending tracts in the spinal cord, basal ganglia, sensorimotor cortex, or more

anterior motor areas. For that matter, injury could be to skeletal or other tissue, as long as there is a change in the organism that makes attempted use of a function punishing. Learned nonuse was thus considered to be the basis of what was termed "excess motor disability," or disability in excess of what appears to be warranted by the organic damage sustained, a frequently observed phenomenon whose mechanism had been unexplained. According to the formulation, the initial reduction in neural excitability and loss of function was said to set up the conditions for the development of learned nonuse, which could convert the temporary early deficit in function produced by the initial neural defacilitation into a permanent condition unless appropriate techniques were applied. For the extremities the appropriate techniques for lifting the learned nonuse would presumably be the same whatever the location of the precipitating lesion. Thus, the learned nonuse formulation predicted that the approach employed with deafferented monkeys would also be applicable to humans after any of the above types of neural damage and to many different pathological conditions. The work described above showed that this prediction was amply confirmed for stroke. Given this promising start, an attempt was made to apply the general procedure that had been employed to motor deficits produced by other types of CNS damage and other diagnostic categories. On the basis of this formulation, these applications of CI therapy were straightforward. This prediction has been confirmed to date by the application of the CI therapy protocol with the same positive results to the upper extremity after TBI (Shaw et al., 2003), MS (Mark et al., 2008), cerebral palsy, and other pediatric motor disorders of neurological origin across the full range of age from 1 year old through adolescence (Taub et al., 2004, 2007, 2011; see **Figure 4**).

The pediatric motor deficits treated included those resulting from TBI, brachial plexus injury, congenital brain malformations, and hemispherectomy (Taub et al., 2009). An adaptation of CI therapy for the lower extremity has been carried out in adults not only after stroke, but also after spinal cord injury and fractured hip (Taub et al., 1999) and MS (Mark et al., 2013). Both in our adult and pediatric clinics we have also worked with numerous cases of brain resection and obtained results comparable to those with stroke when the initial motor deficits were similar (Taub et al., 2007).

#### **CI THERAPY AND NEUROPLASTIC CHANGE**

It has been found that CI therapy-type interventions involving training of extremity use after a CNS injury results in both improved extremity function and reorganization of brain activity. Nudo and co-workers demonstrated this first in new world monkeys (Nudo et al., 1996), showing that the area surrounding a motor cortex infarct that would not normally be involved in control of the hand came to participate in that function at the same time that performance on an experimental task involving manual dexterity improved. In humans whose upper extremity function had been enhanced by CI therapy, Liepert, Taub, and co-workers (Liepert et al., 1998, 2000b) used focal transcranial magnetic stimulation to show that the cortical representation of an important muscle of the hand (abductor pollicis brevis) was greatly enlarged. The results are presented in **Figure 5**.

**FIGURE 4 | Spontaneous use of the more impaired arm (Pediatric Motor Activity Log score) of young (2–6 years) children with hemiparetic CP receiving Constraint-Induced Movement therapy or standard occupational therapy.** Data for the Constraint-Induced Movement therapy group are shown before and during treatment and 1, 6, and 12 months after treatment. Data for the control subjects are shown at corresponding times for 6 months after treatment, at which time they were crossed over to Constraint-Induced Movement therapy. After crossover data are shown for treatment and 1 and 6 months afterward. The data are similar to those for adults shown in **Figure 1**. For both the children given CI therapy first and those given the intervention after crossover, the amount of spontaneous use of the more affected arm in the ADL in the real-world environment increased from approximately 15% compared to use of the less affected arm to approximately 65%. Reprinted from Taub et al. (2011).

**FIGURE 5 | (A)** Pretreatment cortical map of the excitable area for the contralateral abductor pollicis brevis in a group of stroke patients determined by transcranial magnetic stimulation (TMS), superimposed on an unlesioned post mortem brain to indicate approximate size and location. **(B)** Post-treatment TMS map in the same group of patients. Reprinted from Mark et al. (2006).

The finding that CI therapy is associated with substantial changes in brain activity was confirmed in other early studies in which one of us (E.T.) also collaborated involving the Bereitschafts potential (Bauder et al., 1999), positron emission tomography (Wittenberg et al., 2003), and EEG source-imaging (Kopp et al., 1999). To date, there have been more than 20 studies, many involving functional magnetic resonance imaging, that have obtained similar results (summarized until 2006 by Mark et al., 2006). These studies employed functional brain imaging and brain mapping techniques. The question remained whether CI therapy could measurably alter brain structure in humans. Starting approximately 15 years ago it was shown that increased use of a function or body part could result in an increase in the

amount of regional gray matter in the part of the brain associated with that function. For example, it was shown that the cortical representations of the left hand of string players were larger than those of control subjects (see **Figure 6**). It was also found that experienced taxi drivers have significantly expanded hippocampi (Maguire et al., 2000), and jugglers acquire significantly increased temporal lobe gray matter density (Draganski et al., 2004), among many other examples. Conversely, thalamic gray matter density significantly declines after limb amputation (Draganski et al., 2006). Moreover, in an animal model of stroke, CI therapy combined with other exercise reduced tissue loss associated with stroke (DeBow et al., 2003). Thus, it became appropriate to ask whether there are anatomical changes following the administration of CI therapy and whether these are correlated with clinical improvements.

Longitudinal VBM was performed on subjects enrolled in our study of the contribution made by the TP to CI therapy outcome (Gauthier et al., 2008). It was found that structural brain changes paralleled changes in amount of use of the impaired extremity for ADL. Groups receiving the TP showed profuse increases in gray matter tissue in sensorimotor cortices and more anterior motor areas both contralateral and ipsilateral to the more-affected arm, as well as in bilateral hippocampi (see **Figure 3A**). The increases in gray matter were significantly correlated with increases on the MAL for the sensorimotor clusters on both sides of the brain and the predefined hippocampus region of interest (*r*s > 0.45). Thus, this change in the brain's morphology is directly related to administration of the TP which in turn substantially increases the amount of real-world use of the affected arm. In contrast, the groups that did not receive the TP showed significant but relatively small improvements in real-world arm use and failed to demonstrate gray matter increases. The fact that the anatomical change is directly related to the TP lends increased credibility to the importance of the TP.

In another study (Sterling et al., 2013), children with hemiparetic cerebral palsy also showed increases in gray matter in the bilateral sensorimotor cortices (see **Figure 3B**). These changes showed a strong correlation with improvements in spontaneous real-world arm use as recorded on the pediatric version of the MAL. In more recent work, patients with progressive MS were found to benefit from CI therapy to the same extent as patients with stroke. In addition they also exhibited a substantial increase in gray matter in sensorimotor areas of the brain (Mark et al., 2014).

#### **APPLICATION OF CI THERAPY TO NON-PARALYTIC DISORDERS** *Speech*

In the context of the present article in which the extension of the CI Movement therapy protocol to visual deficits is considered, an application of CI therapy of particular interest is its use with post-stroke aphasia, since this involves a non-motor deficit. In a substantial number of stroke patients, because of halting and slow verbal production and incomplete understanding, speech becomes very effortful and often embarrassing. The person compensates by greatly reducing attempts to speak or remaining silent entirely and by using gestures and other non-verbal means of communication. The demonstrations described above that learned nonuse associated with motor deficits is modifiable in chronic

**FIGURE 6 | (A)** Equivalent current dipoles elicited by stimulation of the thumb (D1) and fifth finger (D5) of the left hand are superimposed onto an magnetic resonance imaging (MRI) reconstruction of the cerebral cortex of a control, who was selected to provide anatomical landmarks for the interpretation of the MEG-based localization. The arrows represent the location and orientation of the ECD vector for each of the two digits' averaged across musicians (black) and controls (shaded). The length of the arrows represents the mean magnitude of the dipole moment for the two digits in each group. The average locations of D5 and Dl are shifted medially for the string players compared to controls; the shift is larger for

stroke raised the possibility that verbal impairment could also be rehabilitated by an appropriate modification of the CI therapy protocol. The LNU formulation predicted that this was a strong possibility. In the initial study, by Pulvermüller, Taub, and coworkers (Pulvermüller et al., 2001; Taub, 2002), aphasic patients with chronic stroke who had previously received extensive conventional speech therapy and had reached an apparent maximum in recovery of language were induced to talk and improve their verbal skills by engaging them in a card game that required frequent and detailed spoken requests and replies for 3 h each weekday over a 2-week period. The intervention was termed Constraint-Induced Aphasia therapy (CIAT I). Constraint was imposed by the requirements of the training/shaping paradigm that was used; there was no physical restraint, though as noted, physical restraint is not necessary to obtain a good result with CI Movement therapy for the upper extremity and it is not used at all in this laboratory for the lower extremity. This study has since been replicated (e.g., Bhogal et al., 2003; Meinzer et al., 2004, 2007; Maher et al., 2006; Kirmess and Maher, 2010). While the results of the CIAT I protocol have been positive, the intervention was only an incomplete translation of CI Movement therapy. CIMT produces an improvement of approximately 500% in real-world use of the more-affected extremity of chronic stroke patients with mild to moderate motor deficit in the UAB laboratory (Taub et al., 2006a). Aphasic patients given CIAT I improved by 30% in real-world verbal behavior. This is a large treatment effect compared to conventional speech language therapies, but it is very small compared to the results produced by CIMT. Consequently, to determine whether the large difference resulted from an incomplete translation of the CI therapy protocol employed in the UAB laboratory with motor deficits to the treatment of language impairment, the initial aphasia treatment protocol (CIAT I) was modified to more closely resemble the CIMT protocol.

In the restructured and enhanced protocol (CIAT II; Johnson et al.,2014), revisions involved addition of new exercises, including D5 than for Dl. The dipole moment is also larger for the musicians' D5, as indicated by the greater magnitude of the upper arrow. **(B)** The magnitude of the dipole moment as a function of the age of inception of musical practice; string players are indicated by filled circles, control subjects by hatched circles. Note the larger dipole moment for individuals beginning musical practice before the age of 12. **(C)** Scatterplot of the Euclidean distances (in centimeters) between the cortical representations of Dl and D5. This distance for the musicians' left hands was greater than that in controls, but this difference is not statistically significant. Reprinted from Elbert et al. (1995).

a final exercise, considered to be the most important, in which everyday verbal interactions were simulated and modeled. In addition, a TP parallel to that used in CIMT was introduced, there was increased emphasis on the shaping of responses, and the primary caregiver was trained as an alternate therapist with their training beginning in the laboratory but focused largely on the at-home practice of verbal behavior. To date, six patients have been treated with the new protocol (four reported in Johnson et al., 2014). Their results have far exceeded those obtained with CIAT I and are more comparable to the results obtained with CIMT. With CIAT I, as noted, there was a 30% improvement in real-world verbal behavior; for the recent patients, the mean improvement was approximately 300%. Of additional interest is the fact that in the 6 months following the completion of treatment verbal behavior scores increased substantially (see **Figure 7**). This increase would appear to be attributable to the continuation of training by the caregivers in the real-world environment and other aspects of the TP.

#### *Focal hand dystonia*

Focal hand dystonia involves manual incoordination that occurs in individuals, including musicians, who engage in extensive and forceful use of the digits. It involves loss of ability to make use of one or more digits independent of movement of another digit. Using magnetic source imaging, we found that musicians with focal hand dystonia exhibit a use-dependent overlap or smearing of the representational zones of the digits of the dystonic hand in the somatosensory cortex (Elbert et al., 1998, 2000). M. Hallett's laboratory obtained similar results (Bara-Jimenez et al., 1998).

From the vantage of this article, this application of CI therapy is of interest because it was used to treat a deficit that was not produced by destruction of CNS tissue, and though the deficit treated was motor in nature, it involved disordered cortical representation zones in a sensory area of the brain. Digital overuse

the Verbal Activity Log (VAL) amount of use scale was significant at posttreatment. After treatment, there was an additional significant performance gain. Data in the graph are ipsitized—that is, pretreatment scores are set to 0 for each participant, and subsequent scores are reported as changes from pretreatment. S1–S6 refer to subject numbers. Modified from Johnson et al. (2014).

had previously been found to produce a similar phenomenon in monkeys in the laboratory of M. Merzenich. Since behavioral mechanisms apparently underlie both the cortical disorder and the involuntary incoordination of movement, we hypothesized that a behavioral intervention could reduce or eliminate both of these correlated abnormalities. Learned nonuse was not hypothesized to be involved in the disorder, but it will be remembered that CI therapy when used to treat patients with stroke, TBI, MS, and other conditions not only overcame learned nonuse, but produced a large improvement in impairment and movements in the laboratory when participants were requested to perform at a maximal level.

Eight professional musicians (six pianists and two guitarists) with long-standing symptoms were studied (Candia et al., 1999, 2002). Our therapy involved immobilization by splint(s) of one or more of the digits other than the focal dystonic finger. The musicians were required to carry out repetitive exercises with the focal dystonic finger in coordination with one or more of the other digits for 1.5–2.5 h daily (depending on patient fatigue) over a period of eight consecutive days (14 days in one case) under therapist supervision. The practice was thus massed; practice of this intensity and duration was very taxing and was at the limit of the patients' capacity. The movements in the laboratory were continuously shaped, the patients were given daily home practice exercises, and a derivative of the MAL was administered each morning before therapy in which problem solving was carried out. After the end of an initial period of treatment, the patients continued practicing the exercises with the splint for 1 h every day or every other day at home in combination with progressively longer periods of repertoire practice without the splint.

All patients showed significant and substantial improvements without the splint at the end of treatment in the smoothness of finger movement, as determined by a device that measured finger displacement, and self-reported dystonia symptoms. The improvement persisted for the 2 years of follow-up in all the patients but one who did not comply with the home practice regimen prescribed. Half of the subjects returned to the normal or almost normal range of digit function in music performance. The treatment is characterized as a form of CI therapy because it has all its main components: massed practice, the main elements of the TP, frequent feedback during exercises, shaping of improved finger movements, and restraint of a body part.

As noted above, focal hand dystonia does not appear to involve learned nonuse. That is, there does not seem to be an advantage in not being able to make independent use of the digits, nor is there any apparent period of CNS defacilitation when an inappropriate pattern of coordination could be learned and "locked in" by being partially successful. Nevertheless, a CI therapyapproach improved function. This may be related to the fact, discussed above, that CI therapy not only overcomes learned nonuse, thereby substantially improving function in the life situation, but also improves impairment. This becomes relevant when considering the potential application of a CI therapy approach to the rehabilitation of visual and other sensory systems after CNS damage.

#### *Amblyopia*

In a previous article (Taub, 2010) we have noted the striking similarities in the procedures and magnitude of treatment results between CI therapy and the type of amblyopia treatment developed by Polat, Levi, and their respective co-workers (summarized in Levi and Polat, 1996). The parallels between the two treatments are reviewed here to establish the possibility that the CI therapy method might produce similar results with other types of visual impairments.

Levi and Polat treat amblyopia by intensive training while occluding the non-amblyopic eye during training. The treatment effects reported are very large. In both amblyopia training and CI movement therapy for the upper extremity the unimpaired part of the body is not permitted to participate in accomplishing the training task, in amblyopia training by patching the sound eye during the training period and in CI therapy by use of a restraint device or sometimes more simply by having the therapist verbally discourage use of the less affected arm during the training period (Uswatte et al., 2006a). Similarly in CI Aphasia therapy non-verbal modes of communication are strongly discouraged, thereby imposing a strong constraint on behavior; physical restraint is not used. Training in each of these types of treatment starts with easily accomplished tasks at or just a little beyond present ability, and after mastery is demonstrated, the individual is challenged with more difficult tasks. In behavioral psychology this process, as noted above, is termed shaping, in which tasks are made more difficult progressively, typically in small steps (Skinner, 1938, 1968; Taub et al., 1994). In both amblyopia training and all forms of CI therapy with adults, immediate trialby-trial feedback of results is provided and in all cases this is felt to be an important factor in patient improvement. In addition, in the work of Levi and Polat, the nature of the discrimination tasks employed in the training is tailored to the specific deficits of individual patients.

Both CI therapy and amblyopia training rely heavily on generalization of improved ability from the trained task to other functions. The basic measure for efficacy in amblyopia is improvement in visual acuity as measured in standard fashion by Snellen charts at a 3 m distance. In different experiments by Polat, Levi and others with adult amblyopes, the training tasks employed have included a contrast sensitivity task with flankers (e.g., Polat et al., 2004; Zhou et al., 2006; Huang et al., 2008), vernier acuity, and position discrimination (Levi and Li, 2009). Despite evidence of specificity (e.g., no improvement at untrained orientations), the procedures in all experiments led to improvement in Snellen acuity. Moreover, other degraded visual functions which were not trained nevertheless improved substantially, such as stereoacuity and visual counting (Li and Levi, 2004; Li et al., 2007). Similarly, the training in the laboratory/clinic carried out in CI therapy which includes few full ADL, nevertheless transfers very extensively to the life situation, as measured by the MAL which, as noted, obtains information about 30 ADL important in life functioning, in such areas as eating, grooming, and dressing (Taub et al., 1993, 1999; Uswatte et al., 2000, 2005a,b, 2006b), and confirmed by objective accelerometry (Uswatte et al., 2000, 2005a,b, 2006b).

A recent advance in the treatment of amblyopia has come from training binocular viewing rather than forced use of the amblyopic eye by patching the good eye (Li et al., 2013; Mansouri et al., 2014). Patients are trained on a computer game, with the contrast levels for each eye adjusted to improve stimulus detection by the weaker eye, and requiring binocular viewing to score points. Preliminary findings indicate greater improvement in visual acuity with dichoptic stimulation over monocular patching, along with improved stereopsis. This result is not dissimilar from a form of rehabilitation termed bimanual training that has recently emerged in the stroke motor rehabilitation field for which good results have been reported. However, since the therapy involves using tasks that can only be accomplished effectively by using both arms, the participant is constrained to use the more-affected arm to carry out the task, this may be viewed as a form of CI Movement therapy, which may explain its relative efficacy. The same result from visual deficit training is achieved in the recent papers involving training with binocular viewing by adjusting the contrast levels for the two eyes so that the participant was constrained to place greater reliance on the amblyopic eye than under normal viewing conditions. Normalized visual acuity has not so far been obtained with binocular training (Li et al., 2013). Incorporating CI therapy methods with binocular training could potentially produce improved results and should be attempted.

In summary, both CI therapy and amblyopia training prominently use constraints on function which force or increase motivation to use the impaired portion of the body and set up the conditionsfor positive reinforcement when this results in improvements, however small, in the target function. The constraints include physical restraints: a padded mitt which prevents use of the less affected hand after damage to the nervous system in CI therapy and either patching the sound eye or equalizing stimulus strengths between the eyes by adjusting contrast levels during amblyopia training. It also involves more general methods for inducing use

of the impaired body part. For example, in both types of treatment, training especially by shaping requires an individual to keep improving performance with the affected portion of the body. This is viewed as a constraint on behavior rather than a restraint, but it has the same effect. (For the most recent discussion of the role of constraints in CI therapy, see Taub et al., 2006b; Taub, 2012).

In one study using the amblyopia training procedure, adults whose defect was not too severe were trained in half-hour sessions given four to six times a week (Polat, 2008). Asymptote in treatment effect was often reached in 30–40 sessions, or 15– 20 h of treatment (Levi and Li, 2009). In contrast, occlusion-only treatment usually continues for 100–400 h (Cleary, 2000; Stewart et al., 2004). The situation is very similar for CI therapy. An early attempt at applying the CI therapy procedure with adults (Wolf et al., 1989) involved use of just one-half of the suggested protocol (Taub, 1980). The less affected arm of chronic stroke patients with mild/moderate motor deficit was restrained in a sling for 90% of waking hours, but no training of the more affected arm was employed. Treatment outcome was measured on a laboratory motor function test (WMFT). The results were reliable, but small (ES *d* = 0.2). Similar results were obtained in another laboratory (Ploughman and Corbett, 2004). However, when training of the more affected arm was added to the regimen, the effect recorded on the WMFT was four times as great (Taub et al., 1993, 2006a); the ES was 0.8–1.0. (As was noted above, the ES for real-world arm use as measured by the MAL was greater still.) Thus, as in the case of amblyopia treatment when only restriction of the unimpaired eye is used without training, the outcome is very much reduced. Training the affected part of the body greatly enhances the treatment effect.

The greatest gains in amblyopia training occur within the first eight sessions; the rate of improvement then slows and asymptote is reached in a mean of 30–40 0.5 h sessions (Li et al., 2007). This describes the negatively accelerated curve typical of many learning situations. The same pattern is exhibited in CI therapy for the upper extremity with both children (Taub et al., 2004) and adults (Taub et al., 1993, 2006a).

The mean visual acuity in the amblyopic eye before the beginning of training in one study was 20/70 to 20/80 (Levi and Li, 2009). At the end of a mean of 35 sessions (17.5 h) when the average patient reached an asymptote, Snellen acuity typically ranged from 20/20 to 20/40; that is, normal or very nearly. This treatment effect was certainly dramatic, but the initial deficit was only mild or at most moderate. Data from several hundred adult CI therapy participants indicate that patients with initial grade 2 (mild/moderate) motor deficit start treatment using their more affected arm spontaneously in the life situation approximately 9% of the amount they used it before stroke. After 2 weeks of treatment that amount increases to a mean of 52% (e.g., Taub et al., 2006a). This is an approximate five times increase, but it is not by any means a "cure." Patients are still using their more affected upper extremity spontaneously in the life situation only half as much as they did before stroke. With lower functioning patients, the treatment change (as contrasted with the absolute level of function observed) is not quite as great, but it is similar. Severe amblyopes require approximately 50 h of training to reach a performance asymptote. There can be as much as a

fivefold improvement over the course of treatment. The absolute level reached, though, can frequently be very much less than in the studies with participants with more moderate deficit (Li and Levi, 2007; Li et al., 2007). More generally, it is difficult to compare magnitude of treatment effects across such very different domains as visual acuity and motor deficit in the extremities. It may be that the improvement in visual acuity resulting from the perceptual training protocol in amblyopia is somewhat greater than for extremity movement in stroke or TBI patients following CI therapy. However, this is by no means certain, and an attempt at detailed comparison may not be profitable at this early stage. The clearest conclusion that can be reached is that very large improvements in function can be produced by use of the appropriate technique in both amblyopia and extremity motor deficit after stroke or brain damage resulting from a variety of causes.

In one study improvement in position discrimination remained stable after monocular training for the 3–12 months tested (Li and Levi, 2004). In other studies, a high level of retention is reported for monocular training 12 months after treatment (Polat et al., 2004; Zhou et al., 2006). For CI therapy, we have observed a clear, linear relationship between persistence of the treatment effect and severity of initial deficit for grade 2 (mild/moderate deficit) to grade 4 (moderately severe deficit) patients. For grade 2 patients at the end of 1 year there is an approximate 15% decrease at the end of 1 year. For lower functioning patients retention is less (Taub et al., 1999). However, in the EXCITE multi-site RCT, the retention in upper extremity function was 100% for grade 2 and 3 patients (Wolf et al., 2008). In our laboratory for lower extremity CI therapy retention is greater than for the upper extremity. Similarly, at 1 year after CI Aphasia therapy II (CIAT II) there was not only no loss in retention, but verbal ability improved significantly (19%). The improvement following the end of treatment is probably due to a strong emphasis on TP procedures (Johnson et al., 2014).

Prior to the work on amblyopia training, the treatment of amblyopia was not generally undertaken after the age of 9 and rarely beyond adolescence (Levi and Li, 2009). Though there were scattered, largely anecdotal, reports of success with sound–eye occlusion in adults, these were largely ignored. The assumption was that once the "critical period" for visual development had been passed, or was too long in the past, the potential for major modification of visual acuity was no longer possible. The amblyopia training studies just noted have unequivocally demonstrated the incorrectness of this traditional, almost axiomatic, belief. This is a major finding. Polat (2008) reports that among 44 patients ranging in age from 9 to 55 years the correlation between amount of improvement and age was not significant. The important comparison of the treatment effect for amblyopia training for children younger than 9, where traditional occlusion-only therapy has been concentrated, and persons in older age ranges has apparently not been made. Further research in this area would be of value.

In parallel research with CI therapy, the initial studies showed that this intervention produces a large and reliable improvement in motor function of the more affected arm of adult chronic stroke patients. More recent work shows that individuals from 18 months to 92 years who have been treated with CI therapy show no correlation between age and amount of spontaneous use of the more affected arm in the life situation.

## **REHABILITATION OF FUNCTION IN OTHER SENSORY SYSTEMS AFTER CNS DAMAGE**

There has been substantial work on improving deficient sensory function by training after CNS damage and as a result of congenital disorders. This work has typically used a part or parts of a CI therapy-type approach but not all of it. The CI therapy protocol involves, as already noted: (1) massed practice, (2) shaping, (3) a TP of techniques to bridge the gap between treatment setting and life situation, and (4) restraint/constraint of compensatory dominance of a more-intact portion of the body. This work will be reviewed briefly here. The results have been generally positive. However, papers reporting results for each separate modality tend to have a narrow focus and there is little reference to similar rehabilitation or remediation research done in other modalities, to say nothing of results of research directed toward improving deficits in motor systems. The question that this article asks is whether the results to date could be increased by adapting a comprehensive CI therapy-type approach to these other types of deficits.

#### **SOMATIC SENSATION**

Given that motor deficit can be substantially reduced after CNS injury by, for example, a procedure such as CI therapy, it might not be unexpected that the sensory system most intimately connected to movement, somatic sensation, would also be amenable to enhancement by training after CNS injury. As Sherrington famously noted, somatic sensation follows movement like a shadow (unless, of course, that intimate connection is artificially abrogated, as by the surgical section of all dorsal roots innervating a body part while leaving the motor outflow over the ventral roots intact). Moreover, a large subliterature amply demonstrates that the cortical somatosensory representations of different portions of the body sustain a high degree of neuroplastic change following the alteration of peripheral afferent inflow (summarized in Buonomano and Merzenich, 1998; Taub et al., 2014). The expectation that impaired somatic sensation after CNS damage can be improved by training is borne out by the experimental evidence. In the early literature Ruch et al. (1938) reported that training could improve sensory discrimination in primates and man after parietal lesions. Similar results were obtained in later work (Forster and Shields, 1959;Vinograd et al., 1962;Van Deusen Fox, 1964; De Jersey, 1979; Yekutiel and Guttman, 1993; Knecht et al., 2001). Though, as Smania et al. (2003) point out, flaws in individual studies prevent clear interpretation of the results in detail, the main finding that somatosensory deficits can be reduced after CNS damage by training emerges clearly. In one study of interest it was found that training in proprioception or position sense at the elbow improved discrimination in that submodality, but the improvement did not generalize to tactile sensation. When training was given in more than one somatosensory submodality, each of those sensory modalities showed improvement (Carey et al., 1993; Sormani et al., 2014). In a recent study with two patients with somatosensory deficit after stroke, multimodal somatosensory training was given for 4 h/day for 10 consecutive

weekdays, a schedule of training similar to that employed in CI therapy. The participant with the more severe deficit showed no improvement, however, the participant with the lesser deficit significantly improved in tactile perception, proprioception, and blind match to samples of palpated novel objects (Borstad et al., 2013). These studies in general, then, demonstrate that impaired somatosensory function after stroke can be improved by training. However, none of these studies employed two of the important components of CI therapy; shaping to progressively increasing levels of performance proficiency and TP techniques. It is possible that the reported improvements in impaired somatosensory discrimination could have been improved by adding these procedures.

#### **HEARING**

A forced use training approach to acute unilateral sensorineural hearing loss was recently attempted, inspired by CI movement therapy (Okamoto et al., 2014). Patients were randomized either to usual corticosteroids or to corticosteroids plus listening to classical music through a headphone with only the impaired ear for 6 h per day for 9–10 days, while the other ear's auditory canal was plugged. The sound levels and frequency distributions were self-adjusted to approximate pre-illness experiences. The investigators observed that the experimental approach was associated with significantly better pure tone audiograms relative to steroid treatment alone and were maintained at 2–3 months follow-up. The study also found improved laterality indices of auditory evoked responses as measured by magnetoencephalography following the experimental intervention relative to baseline. Although a complete CI therapy approach was not used, the findings are encouraging and suggest that more of a CI therapy modification could further benefit treatment outcomes.

#### **VISION**

#### *Unilateral spatial neglect*

Unilateral spatial neglect is a common consequence of stroke occurring in about half of all survivors in the acute phase. It refers to a failure to report, respond to, or orient to stimuli primarily in one part of space, most often opposite to the side of the brain injury (Mark, 2003; Barrett et al., 2012). It results in the disturbance of basic functional activities such as feeding, grooming, locomotion, orientation of the body in space, activities involving eye-hand coordination and other types of "motor aiming." Neglect is not purely a disorder of visual input. It can be manifest in the dark (Karnath et al., 1998), in lingually clearing one's own mouth (André et al., 2000), and when describing the spatial layout of distant familiar landmarks while in the laboratory (Guariglia et al., 2013). However, because it is most often observed during visual tasks, we will discuss it as a visual disturbance here. Although the condition is termed "neglect" because of underattention toward one direction, it can also manifest hyperattention in the opposite direction (Mark et al., 1988), and thus should best be considered an imbalance in the spatial distribution of attention.

Unilateral neglect frequently resolves spontaneously during the first year after brain injury, yet many patients remain chronically impaired during daily living activities (Kinsella and Ford, 1985). Treatments for unilateral neglect have been largely unsuccessful or impractical, or they have not been evaluated in controlled studies (Mark, 2003). A therapeutic approach that is commonly used is termed visual scanning training (Lawson, 1962; Weinberg et al., 1977; Toglia, 1992; Quintana, 2002). Two strategies are to place a highly distinctive stimulus such as a colored stripe in contralateral space ("anchoring") to overcome scanning biases, or teaching the patient to use their finger to guide themselves contralaterally. Some patients have been observed to benefit from this type of intervention. However, not all patients benefit, and systematic studies on the efficacy of this approach have not been conducted (Mark, 2003; Barrett et al., 2012). One intervention that has received considerable attention in the last decade is the use of adaptation to the lateral displacement of vision in the direction opposite to the neglected field of vision. Since most unilateral neglect occurs after a stroke affecting the right hemisphere leading to neglect of the left visual hemi-field, prismatic lenses that displace the visual field to the right are used in the therapeutic intervention. Initially the subject, including patients with left neglect, make errors in pointing to a visual target in a rightwards direction. If subjects are allowed to see their executive hand, pointing is rapidly corrected leftwards so that pointing becomes accurate. If the goggles containing the prisms are then removed, the subject makes an error in pointing, termed the aftereffect, in the opposite direction, that is leftwards. For patients with left neglect this in part corrects for part of their deficit, although transfer to real-world daily living activities has seldom been evaluated. The improvement has been found to persist in some patients (Barrett et al., 2012). For other patients the improvement is short-lived. In general the results have been mixed (Barrett et al., 2012).

An important parcellation of the unilateral spatial neglect phenomenon has been made between a visuo-motor aiming deficit and a disturbance in "spatial where" perception (Adair and Barrett, 2008; Eskes and Barrett, 2009; Heilman et al., 2011). The two undoubtedly have an overlap in brain regions in terms of the causative damage (Goedert et al., 2014). Empirically they are separate issues, and this is probably true also for methods of achieving therapeutic remediation. Goedert et al. (2012) using the Catherine Bergego scale (Azouvi et al., 2003) found differential perceptual where and motor aiming deficits. It has also been found that "visual where" perception does not improve after prism adaptation therapy while motor aiming behavior usually does (Striemer and Danckert, 2010; Goedert et al., 2014). Moreover, the opposite is also true. Some patients with apparently intact ability to perceive and represent the visual environment nevertheless make persistent motor aiming errors (Laplane and Degos, 1983; Coslett et al., 1990; Triggs et al., 1994; Barrett et al., 1999).

The dissociability of remediation of "visual where" perception and motor aiming deficits may be based on differences in their underlying corrective mechanisms. In the older literature there was an experimental debate on whether adaptation of pointing at visual targets to lateral displacement of vision by wedge prisms involved a recalibration of vision (e.g., Held and Hein, 1958; Held, 1965) or, perhaps counterintuitively, a recalibration of position sense of the executive arm (Harris, 1963, 1965; Hochberg, 1963; Pick et al., 1963; Hamilton, 1964). It appeared that a critical

method for evaluating these positions would be to determine how the surgical abolition of the theoretically relevant proprioception from the executive arm would affect the ability to compensate for prismatic displacement of vision. Deafferentation was achieved by the serial section of dorsal roots from the second cervical to the third thoracic segment. Monkeys with a deafferented forelimb were first trained to point with reasonable accuracy at a visual target without view of the body or limbs, receiving feedback on the terminal position of the pointing finger only via food reward (Taub et al., 1975a). After adaptation to displacement of vision occurred, the helmets with prism lenses were removed and the course of the aftereffect was tracked (Taub and Goldberg, 1975). For normal animals the prism aftereffect was 39% of full prism displacement; for the deafferented animals, it was 100%. It thus seemed clear that the presence of proprioception inhibits adaptation to laterally displacing prisms.

Earlier research (Taub and Berman, 1963, 1968; Taub et al., 1973) presented evidence that position sense consists of two independent, redundant components: one peripheral (usually termed proprioception) and one central (variably termed central feedback loops or central efferent monitoring). In monkeys with a deafferented neck and forelimbs, the peripheral component is absent; there is thus less resistance to recalibration of the arm during prism adaptation and, accordingly the process proceeds to completion more rapidly and is more stable than in normal subjects. One might say that the less there is of position sense, the easier it is to recalibrate, and the more resistant to alteration the recalibration is (as evidenced by the larger aftereffect in deafferented monkeys). The recalibration of body parts can be rather specific. For example, in a visuo-motor pointing experiment, one normally gets a recalibration of the pointing extremity only with no intermanual transfer except under specially arranged circumstances (e.g., appropriate spacing of pointing practice trials; Taub and Goldberg, 1973). Thus, if prism adaptation involves a recalibration of position sense and not vision, this would explain why prism adaptation therapy for unilateral spatial neglect after stroke would correct the deficit in motor aiming, but have a reduced effect on the deficit in visual where perception.

This does not mean that vision cannot be recalibrated. For example, Taylor and Papert (1955) noted that some of their subjects who wore spectacles that inverted the visual field eventually reported after a period of days that the perceived world no longer appeared to be inverted. A similar phenomenon had previously been reported by (Stratton, 1897). There has been controversy as to whether Stratton did in fact experience an upright world (Carr, 1935) or whether he did not (Ewert, 1930, 1937;Woodworth, 1934; Higginson,1937; Peterson and Peterson,1938). Snyder and Pronko (1952) repeated Stratton's experiment but could not give a conclusive answer to the question. There is clearly a verbal ambiguity in the question of whether the visual field appears upright. While this question and its analysis are of conceptual interest, in terms of practical import there is no question that humans can adapt to massive transformations of the visual field. The recalibration of vision (or perhaps the recalibration of direction of gaze) is probably responsible for the partial success in some patients of visual scanning therapy (Weinberg et al., 1977), training of visual

awareness (Tham et al., 2001), therapist-coached use of mental imagery (Niemeier et al., 2001), and review with a therapist of videotaped feedback of task performance (Tham and Tegnér, 1997). The fact that the success of these techniques is limited is not surprising in view of the report of Taylor and Papert (1955) and other investigators (Stratton, 1897; Ewert, 1930; Kohler, 1951; Kottenhoff, 1957; Ardiago, 1886; as reported in Giannitrapani, 1958) that adaptation to displacement, reversal, and inversion of vision requires continuous experience with the transformed vision over a period of many days. The study by Antonucci et al. (1995) is relevant in this regard. These authors reported that comprehensive scanning training and prolonged practice with reading, copying, and describing pictures resulted in significant improvement on standard and functional aspects of unilateral neglect that had not been directly trained. The authors suggested that the basis for their successful behavioral intervention was "massive stimulation." Training sessions were 1 h daily, 5 days a week, for eight consecutive weeks, or 40 h total. This is considerably more than is allotted for the therapies for unilateral post-stroke neglect noted above. This report is consistent with the extended experience required for adaptation to inversion and reversal of the visual field in healthy adults. It is also consistent with the amount of training that is given in CI therapy for motor deficit after stroke.

There are other parallels with some phenomena associated with unilateral neglect and one of the mechanisms that underlie a portion of the motor deficit after stroke that is improved by CI therapy. Neglect patients commonly fail to protect the paretic limb during transfer from bed to chair or during mat mobility (Mark, 2003). Even when the paresis is mild (as demonstrated by limb movement made at the request of an examiner), neglect patients frequently fail to use the limb to assist themselves when balancing or during other activities. This failure of spontaneous limb use in the face of a demonstrated ability to carry out specific behaviors when they are requested by an examiner is termed learned nonuse in the context of CI therapy. In the neglect literature, this is commonly termed "motor neglect." The term is used because the deficit is viewed to arise as a unilateral deficit of purposive limb activation (Sampanis and Riddoch, 2013). However, it is noteworthy that (1) the neglect literature seldom contrasts this phenomenon with learned nonuse, and (2) accordingly the possibility of the deficit arising as a result of behavioral conditioning, as opposed to being an unlearned disorder, is not routinely entertained.

One of the primary reasons for the efficacy of CI therapy is that it overcomes learned nonuse. The apparent presence of this phenomenon in unilateral neglect is an additional reason to suggest that a CI therapy approach to the treatment of unilateral neglect is worth attempting, even if the initial deficient unilateral activity may not have been learned. The possibility of treating motor neglect with CI therapy has been previously raised, although not in depth (Punt and Riddoch, 2006). The approach might involve intensive massed practice of both a variety of motor aiming behaviors, as well as visual scanning exercises and other strategies for the remediation of the visual where perception problem. The training, as in CI movement therapy, would also include shaping and many elements of the behavioral TP. Insurance reimbursement would not be available for 36–40 h of therapy for unilateral spatial neglect. However, the initial question is whether this approach is efficacious. Questions of pragmatic import are of course of significance, but they do not arise until efficacy is demonstrated. At that point, it becomes worth determining how treatment can be modified to make it available for reimbursement.

There have been at least three studies in which conventional CI movement therapy has been used to reduce unilateral spatial neglect, two in which CI movement was used alone (Bollea et al., 2007; Welfringer et al., 2013), and one in which conventional CI therapy was used in conjunction with patching of the eye opposite to the neglected visual field (Wu et al., 2013). The results reported were good. This, however, is not the procedure being advocated here. The proposed objective would be to overcome both the visual aiming and "perceptual where" deficit. Conventional CI Movement therapy would have a significant role in the suggested therapeutic approach, but so would prism adaptation training and one or more procedures to improve "visual where" perception. Optimally, extended practice, shaping, and a visually appropriate TP would be embedded in the combined approach.

#### *Visual field defects due to cortical injury*

Stroke results in visual field deficits in approximately 25% of cases (Kerkhoff et al., 1994). The consequent cortical blindness often involves a homonymous hemianopia, in which half the visual field is affected in both eyes. This kind of visual deficit also occurs after TBI, brain resection, and other types of post-chiasmic damage. It often results in serious impairments in the activities of daily life, such as in reading, driving, and visual exploration (Zihl and Von Cramon, 1985); this can have pervasive disabling consequences for everyday activities and importantly in being able to maintain appropriate employment. The extent of a patient's awareness of the visual field deficit can be shown on formal testing to vary with the amount of attention directed at the visual environment along with the kinds of stimuli that are presented (e.g., faces; Williams and Gassel, 1962; Gassel and Williams, 1963).

Over the last several decades a substantial number of studies have been carried out demonstrating that this type of deficit can be reduced by appropriate training techniques. Four different general strategies have been used, not only with cases of partial cortical blindness, but also in cases of retinal and optic nerve damage.

(1) Saccadic eye movement training. In this method patients are trained to make large eye movements into the blind hemifield (Kerkhoff et al., 1992, 1994; Zihl, 1995; Pambakian et al., 2004). In the procedure employed by Kerkhoff and coworkers, training was given at least once daily for 30 min, 5 days per week for 4–12 weeks. This strategy led to considerable success in the performance of visual ADL including an impressive return to part-time work in 20 of 22 patients (Kerkhoff et al., 1994). In that study actual restitution of part of the scotomatous visual field occurred in only 54% of the patients. However, the patient's mean visual search field size increased a mean of approximately 20◦, an effect that persisted for at least 3 months and is presumably the basis for the substantial improvement in visual ADL. Surprisingly, this mode of rehabilitation training has not been employed as much as seems warranted.


In summary, the literature unequivocally indicates that visual deficit after brain damage can be reduced by the application of appropriate techniques. The successes achieved in the rehabilitation of visual function are certainly impressive. However, the literature also suggests that the amount of remediation that is currently being achieved can be substantially enhanced.

By far the largest amount of work to date has been carried out with various forms of vision restoration therapy, where the size of the blind field is reduced by perimetric stimulation of the border area. The current status of this area resembles where the field of motor rehabilitation after CNS damage was in the 1990s. The most frequently employed therapies at that time, such as neurodevelopmental therapy (NDT) and proprioceptive neurofacilitation (NPF), were successful in producing changes in movement in the clinic, but the focus of the therapeutic work was on producing the best movement the participant was capable of on demand in the presence of the therapist; for example, how far the patient could raise a more affected arm above shoulder level or maintain balance when subjected to a perturbing force. There was a reduced focus on how movements and movement adjustments could be used in functional activities (e.g., folding a towel, picking up a glass of water, bringing it to the mouth and drinking) when this might be requested by a therapist, and little or no attention was paid to whether the improved ability to make the trained movements generalized to the life situation. While it is incontrovertible that improving the quality of impaired movement or of making the performance of previously abolished movements again possible is important, it is also true that if the improvement on the impairment level does not lead to improved capability of carrying out functional activities in the clinic and then translation of this improved capacity to ADL in everyday life, the improvement in function has no pragmatic import. Unless this final step is taken to enhance independence and quality of life, one might argue that the improvement observed in the clinic is primarily of academic interest. The field of vision restoration training has progressed past that point. It has been shown that this therapeutic approach improves health-related quality of life and visual ADLs (Mueller et al., 2003; Sabel et al., 2004; Gall et al., 2008; Roth et al., 2009b). However, the improvements do not take place in all visual ADL areas and while the correlations between changes in field size and improvements in certain types of visually guided real-world behavior are significant, they are small. Moreover, the beneficial effects of this treatment on behavior have not been universally accepted (Horton, 2005a,b; Plant, 2005). For example, in a RCT, saccadic eye movement training was shown to be superior to a form of visual restoration therapy for improving exploration toward the blind hemifield (Roth et al., 2009a,b). In addition, fully one-third of patients given vision restoration training show no enlargement of the visual field. In comparison, CI movement therapy produces a clinically significant increase in the use of a more affected arm in the life situation. In addition, approximately 97% of patients improve a clinically significant amount (>10% of the full scale of the MAL: van der Lee et al., 2004; Taub et al., 2005; Dromerick et al., 2009). The only participants not reaching this level of success were those who explicitly rejected the therapeutic procedures after study enrollment.

The success of the visual rehabilitation of cortical blindness has been real and important. The question remains, however, whether the magnitude of the restitution could be greater, especially in terms of generalization to everyday activities in the life situation. Areas that could be explored in this regard are as follows:


Given the impressive gains that have been made in visual rehabilitation to date, it could be that attention to these factors could yield a robust increase in the real-world relevance of the existing methods. In addition, it might be of considerable value to expand the training sessions somewhat so that two or more of the techniques described above could be combined. An objection might be that a combined approach would not be clinically feasible or acceptable to patients. However, the prior question might be whether a combined approach confers any objective improvement over use of a single technique. If that is the case, work could later be carried out to streamline a protocol; less important elements could be eliminated so that a more therapeutically potent procedure could be designed that could both be carried out in an acceptable period of time and that was accessible to an appropriately large number of patients.

#### **GENERAL MECHANISMS UNDERLYING EFFICACIOUS REHABILITATION OF FUNCTION AFTER BRAIN INJURY OVERCOMING LEARNED NONUSE REVISITED: SENSORY NONUSE AND DIMINISHED NEURAL CONNECTION STRENGTHENING**

As described in an earlier section of this paper, the development of CI Movement therapy began with work with monkeys who had received a surgical abolition of sensation from one forelimb. Subsequently, the deafferented monkeys made no purposive use of that extremity though motor innervation remained intact. Sherrington explained this puzzling phenomenon on the basis of the interruption of all same-segment reflex arcs from the affected limb. This experiment formed an important basis of Sherringtonian reflexology, a dominant position in neuroscience for the first three-quarters of the twentieth century. It was based on the belief that spinal reflex arcs constituted the basic building blocks upon which purposive or voluntary movement was based. However, work from approximately 1960–1980 showed that a deafferented primate forelimb could be converted into a limb that, while not normal, could be used for a very wide variety of purposes by training of the deafferented extremity or restraint of the intact forelimb

(Taub, 1977, 1980). In ambulation, for example, it was not easy for a casual observer to discriminate between an intact and a deafferented forelimb. The reflexological or peripheralist position could therefore not be correct. As a replacement, the learned nonuse (LNU) mechanism was suggested (Taub, 1977, 1980). It was held that during the early post-injury period, spinal shock rendered the segmental motor apparatus fully or partly inoperative so that the affected limb really could not be used. During this period the monkey had to learn compensatory motor patterns to carry out the basic primate ADL. The compensatory activities, primarily the use of the intact forelimb to do the work of both, involved a degraded, inefficient pattern of coordination, but since it was at least partially successful, it was reinforced, strengthened, and through repeated use overlearned. Subsequently, the spinal shock resolved spontaneously and use of the deafferented limb would have become possible, but because of the overlearning, termed learned nonuse, the newly acquired ability to use the deafferented limb was masked. Moreover, even in the chronic phase the absence of sensation in the limb reduced the afferent drive of the segmental motor apparatus, rendering movement of the limb more difficult, and thereby providing a contemporaneous basis for maintaining the LNU. Training, especially by the behavioral technique of shaping, or prolonged restraint of the intact limb forced use of the deafferented limb by increasing the motivation to use it, thereby overcoming the LNU. However, if simple repeated practice of a learned movement were carried out as in a conditioned response situation (Knapp et al., 1963; Taub and Berman, 1963; Taub et al., 1965, 1966) rather than by administering training by a shaping sequence or if the restraint was removed too soon after the appearance of purposive movement of the deafferented limb, the animal immediately reverted to nonuse of the affected extremity. The strongly learned habit of nonuse of the deafferented limb simply overcame the insufficiently or weakly learned habit of using that limb produced by the repeated repetition of a simple movement in a conditioned response situation or by an insufficiently long period of intact limb restraint. Using the affected limb still required considerable effort due to the incompletely recovered afferent drive. The LNU formulation was confirmed in two experiments (Taub, 1977, 1980). A long-term reversal of LNU could be achieved by shaping or prolonged restraint. In effect, a weakened habit or alternately a weakened neural drive had been strengthened sufficiently to make spontaneous use of the deafferented limb in the animal's life situation possible on a long-term basis. One could also say that the animal's ability to attend to use of the deafferented extremity had been increased or alternately weak or DNCs had been strengthened.

The first application of the CI Movement therapy protocol to humans was to patients with chronic stroke. However, as noted earlier, the LNU formulation predicted, in effect required, that the CI therapy approach apply to other types of damage to the CNS. This was subsequently confirmed for the upper extremity in work with patients with TBI, MS, cerebral palsy, other childhood motor disorders with a variety of etiologies (brain resection including hemispherectomy, TBI, different congenital brain malformations), and for the lower extremity after the same types of CNS damage and fractured hip. A CI therapy approach has also been shown to be effective for aphasia after stroke. In the present

context of the rehabilitation of visual deficits, the latter is of particular interest since in aphasia it is not the motor component of speech that is affected (dysarthria), but rather the linguistic aspect. This then is a clear example of the CI therapy training approach being effective for the rehabilitation of a non-motor function after brain damage. It was also noted above that the work of Levi and Polat and their respective coworkers showed that a CI therapy-type approach was successful in the treatment of amblyopia. The content of the methods applied to each of these different functional deficits vary substantially, but the present contention is that the principles of effective rehabilitation might be the same for all of them. It is proposed herein that a CI therapy approach would also be effective for reducing unilateral visual neglect, somatosensory, and auditory deficit after stroke or other brain injury and for improving vision in partial or even complete cortical blindness.

The question arises as to why all of these disparate functional deficits should be responsive to the same rehabilitation approach. On the level of mechanism, is there some commonality that would explain why such different aspects of impaired organismic function should be responsive to the same general principles of treatment, if that turns out to be the case. CI movement therapy is efficacious for motor deficits in part because it overcomes learned nonuse. This was characterized above as involving a strengthening of habits weakened by injury, which reduces on a physiological level to a strengthening of weakened or DNCs (e.g., Liepert et al., 2000a). Another way of stating the general principle is that the organizing concept is the manipulation of attention by appropriate methods of training. The individual is taught to attend to previously ignored or weak aspects of the motor repertoire or sensation. In each case the effective elements would be repetitive practice, training by shaping, discouraging inefficient compensatory patterns of motor coordination or perception, and TP techniques to integrate the newly developed motor or sensory/perceptual ability into the activities of everyday life. The result is that weak or DNCs are strengthened or increased. It is proposed that this would be as much the case for sensory function after CNS damage as it is for motor function. In the case of sensory deficits one might speak of sensory nonuse rather than learned nonuse. Both sensory nonuse and learned nonuse would be related to the more general phenomenon of the presence of DNCs compared to the stronger or more numerous neural connections that existed before brain damage, or nonuse of a function, or both. The rehabilitation of function in either sensory or motor systems could then be considered to involve diminished neural connection strengthening (DNCS). The term sensory nonuse is not meant to convey any connotation related to conscious choice or conscious awareness; that is, that a person has access to a given type of sensory input but chooses to ignore it. It is meant rather to suggest that the information associated with an afferent input is present in some form in the nervous system, but it is not responded to by an individual either because the signal is too weak to rise above a given threshold or because the signal has fallen below a threshold that would enable it to balance a countervailing sensory input that captures attention and response tendencies. Learning would be important in setting those thresholds. The development of learned nonuse and sensory nonuse and the process by

which they can be overcome is schematized in **Figures 8** and **9**, respectively.

There are clearly several mechanisms that could lead to diminished or weakened neural connections. One that would be most prominent in the development of learned nonuse would be based on the loss of neural excitability following damage to the brain; this would lead to nonuse of an affected extremity which through learning would persist even when the excitability of the neural connections returned through spontaneous recovery, at least, partially, so that use of the affected limb was potentially possible. Another mechanism would involve loss of a portion of the neural connections that normally support a function due to brain damage, rendering the reduced number of connections that remain incapable of supporting that function. Input from the remaining connections would then simply be ineffective or disregarded unless the remaining connections were strengthened or supplemented by the application of an appropriate rehabilitation technique. Future research might test these two and other possible mechanisms.

#### **CORTICAL REORGANIZATION REVISITED: BRAIN STRUCTURE REPURPOSING**

In the seminal original research of M. Merzenich and coworkers, structural brain remapping in new world monkeys was studied after amputation of a single digit. The enlargement of the cortical representation zones of the neighboring still-intact digits was small and was thought to be the result of the increased density of existing dendritic arborizations; this gave rise to the 2 mm rule which proposed that cortical reorganization could take place over a limit of 2 mm of cortical territory. Later Pons et al., working with the deafferented monkey of one of us (E.T.) found that

the entire representation zone of the now-deafferented extremity had been "invaded" by extensions of the innervation of the still-intact face. The invaded zone was 10–14 mm in extension. Merzenich agreed that the 2 mm rule no longer represented a limitation of the extent possible for neuroplastic territorial cortical reorganization. The earlier view was seen to be an artifact of the experimental model studied, since the cortical representation zone of a single new world monkey digit is small. However, aspects of the original formulation continue to exert an influence and may limit the perceived horizon of what the limits of rehabilitation might be. In the study by Pons et al., using monkeys with a complete limb deafferentation, the authors coined the term "massive cortical reorganization" to describe their results. However, the process observed involved an intact cortical region spreading its connections over a contiguous region on the same side of the brain. Since then structural reorganization that is even more massive involving non-contiguous portions of the brain has been observed. In this laboratory, we have found a profuse structural remapping taking place after CI therapy in patients after stroke over all or a large part of ipsilesional motor areas of the brain, but also in non-contiguous contralesional cortex that normally contributes only a small portion of the innervation of the trained arm affected by stroke (Gauthier et al., 2008), cerebral palsy (Sterling et al., 2013), and MS (Mark et al., 2014). We have also obtained preliminary data strongly suggesting that after CIAT II there is a large increase of gray matter in the areas in the right hemisphere homologous to the language areas in the stroke-affected left hemisphere. In each of these cases, there had probably been an elaboration of a small but already existing source of functional innervation. However, even larger reorganizations

of cortical territory have been found to occur. In individuals who have been blind since birth, both auditory (Kujala et al., 1992, 1995a,b, 1997; Alho et al., 1993; Weeks et al., 2000), and tactile (Rösler et al., 1993; Uhl et al., 1993; Kujala et al., 1995a; Röder et al., 1996, 1997; Cohen et al., 1997) stimuli come to be processed in the visual cortex. This remarkable case of territorial neuroplasticity and the profuse increase in gray matter in unexpected areas of the brain in correlation with the large improvements in motor function in response to CI therapy and the phenomena just cited for sensory areas of the brain would seem to warrant a new term descriptive of the extraordinary capacity of the brain to make territorial adjustments after brain damage to compensate for lost function. This marked capacity for territorial neuroplasticity, then, might be referred to as brain structure repurposing (BSR). The magnitude of what is possible with BSR or use-dependent cortical reorganization suggests that we do not yet know the limits of rehabilitation. The application of appropriate techniques such as the CI therapy approach applied in new areas of damaged function after brain injury and the combination of multiple techniques in this pursuit may give promise of the emergence of new vistas for rehabilitation.

To summarize, in this article the concept of learned nonuse, demonstrated to occur after single limb deafferentation in monkeys and movement and speech after stroke and other types of brain damage, has been elaborated to include the sensory nonuse that occurs in amblyopia, unilateral visual neglect, and visual field defects. According to the new formulation both learned nonuse and sensory nonuse involve a weakening of neural connections due to damage to the brain, the loss of connections, the reduced excitability of connections due to nonuse of a function, or a "shock-like" phenomenon in the early post-injury phase, or all of the above. CI therapy overcomes learned nonuse and it is proposed

that it also has the potential for overcoming sensory nonuse by strengthening DNCs. Thus, DNC strengthening provides a bridge between learned nonuse/sensory nonuse on the one hand and use-dependent cortical reorganization on the other, the two mechanisms that presumably underlie the efficacy of CI therapy, which until now have been viewed as separate entities.

## **CONCLUDING CONSIDERATIONS**

In the two preceding sections several mechanisms are postulated as constituting the basis of CI therapy which has been shown to be efficacious for motor and speech deficits, and the potential extension of its principles to the domain of sensory deficits after brain damage or abnormal development. The mechanisms suggested are presented in general terms and are offered as the basis for hypothesis formation and further testing. The specific anatomical and physiological details of the mechanisms are intentionally left largely unspecified. They can be supplied by future research with the recognition that additional processes may be involved or that alternative processes may be operative. The future development of this line of conceptualization may be similar to the progressive process by which the learned nonuse formulation developed. This concept, not yet named, was proposed in 1968 (Taub and Berman, 1968). Over the next decade the operation of the proposed mechanism was confirmed experimentally in work with monkeys (Taub, 1977, 1980), and it subsequently provided a basis for the extension of the CI therapy methodology from monkeys to the upper extremity of humans after stroke (Taub et al., 1993), and subsequently to the many applications of CI therapy to other pathological conditions, the lower extremities, and speech. Learned nonuse started as a formulation that adequately explained the available data, and gradually acquired a more substantive standing as experiments confirmed its existence and numerous predictions that it generated were confirmed. This paper proposes that neural connection insufficiency or weakness lies at the basis of both learned nonuse in motor systems and sensory nonuse in sensory systems and that both can be overcome by a rehabilitation methodology that modifies the neuroplastic potential of the brain, a potential that persists throughout the lifespan and that enables BSR. It is suggested that this formulation explains much of the data currently available relating to efficacious treatments for motor and sensory deficits. However, the new formulation is tentative in terms of specific details and requires future experimental confirmation. The formulation is testable and it stands now at the level of development that the learned nonuse formulation originally had when it was first proposed.

#### **REFERENCES**


by different motor therapies after stroke. *Stroke* 39, 1520–1525. doi: 10.1161/strokeaha.107.502229


for persons with traumatic brain injury. *J. Neurol. Phys. Ther.* 27, 180. doi: 10.1682/JRRD.2005.06.0094

Sherrington, C. S. (1931). Quantitative management of contraction in lowest level coordination. *Brain* 54, 1–28.

Skinner, B. (1938). *The Behavior of Organisms*. New York: Appleton-Century-Crofts.


in monkeys and man: the role of shaping. *J. Exp. Anal. Behav.* 61, 281–293. doi: 10.1901/jeab.1994.61-281


Zihl, J. (1995). Eye movement patterns in hemianopic dyslexia. *Brain* 118, 891–912. Zihl, J., and Von Cramon, D. (1979). Restitution of visual function in patients with cerebral blindness. *J. Neurol. Neurosurg. Psychiatry* 42, 312–322.

Zihl, J., and Von Cramon, D. (1985). Visual field recovery from scotoma in patients with postgeniculate damage: a review of 55 cases. *Brain* 108, 335–365.

**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: 09 June 2014; accepted: 17 September 2014; published online: 09 October 2014.*

*Citation: Taub E, Mark VW and Uswatte G (2014) Implications of CI therapy for visual deficit training. Front. Integr. Neurosci. 8:78. doi: 10.3389/fnint.2014.00078 This article was submitted to the journal Frontiers in Integrative Neuroscience.*

*Copyright © 2014 Taub, Mark and Uswatte. 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.*

## Rehabilitation of homonymous hemianopia: insight into blindsight

## **Céline Perez <sup>1</sup> and Sylvie Chokron1,2\***

<sup>1</sup> Neurology, Unité Fonctionnelle Vision et Cognition, Fondation Ophtalmologique Rothschild, Paris, France <sup>2</sup> Laboratoire de Psychologie de la Perception, Université Paris-Descartes, UMR 8242 CNRS, Paris, France

#### **Edited by:**

Olivier A. Coubard, CNS-Fed, France

**Reviewed by:** Arash Sahraie, University of Aberdeen, UK Marco Tamietto, Tilburg University, Netherlands

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

Sylvie Chokron, Neurology, Unité Fonctionnelle Vision et Cognition, Fondation Ophtalmologique Rothschild, 25, Rue Manin, 75019 Paris, France e-mail: sylvie.chokron@gmail.com Strong evidence of considerable plasticity in primary sensory areas in the adult cortex, and of dramatic cross-modal reorganization in visual areas, after short- or long-term visual deprivation has recently been reported. In the context of patient rehabilitation, this scientifically challenging topic takes on urgent clinical relevance, especially given the lack of information about the role of such reorganization on spared or newly emerged visual performance. Amongst the most common visual field defects found upon unilateral occipital damage of the primary visual cortex is homonymous hemianopia (HH), a perfectly symmetric loss of vision in both eyes. Traditionally, geniculostriate lesions were considered to result in complete and permanent visual loss in the topographically related area of the visual field (Huber, 1992). However, numerous studies in monkeys, and later, in humans, have demonstrated that despite destruction of the striate cortex, or even following a hemispherectomy, some patients retain a certain degree of unconscious visual function, known as blindsight. Accordingly, there have recently been attempts to restore visual function in patients by stimulating unconscious preserved blindsight capacities. Herein we review different visual rehabilitation techniques designed for brain-damaged patients with visual field loss. We discuss the hypothesis that explicit (conscious) visual detection can be restored in the blind visual field by harnessing implicit (unconscious) visual capacities. The results that we summarize here underline the need for early diagnosis of cortical visual impairment (CVI), and the urgency in rehabilitating such deficits, in these patients. Based on the research precedent, we explore the link between implicit (unconscious) vision and conscious perception and discuss possible mechanisms of adaptation and plasticity in the visual cortex.

**Keywords: rehabilitation, homonymous hemianopia, blindsight, plasticity, brain reorganization, cortical visual impairment, post-chiasmatic damage**

## **INTRODUCTION**

Although cortical visual impairments (CVI) are frequently encountered after brain damage, they are unfortunately rarely considered in neuro-rehabilitation programs. Whereas traditionally, the treatment of speech, language, and motor disorders is systematic, no visual training is usually proposed to patients with CVI.

The most frequent CVI in brain-damaged patients is homonymous hemianopia (HH), a total loss of vision in the contralesional hemifield of both eyes (Zhang et al., 2006a). However, despite the frequent occurrence of HH in these patients, it is rarely diagnosed and treated. This paucity can be explained by the fact that HH is often accompanied by more obvious neuropsychological disorders (e.g., aphasia, alexia and unilateral spatial neglect), and by a common assumption on the part of health professionals that objective recovery from visual field loss is impossible. Moreover, patients suffering from visual field defects might present with consequent anosognosia or incorrectly assume that their deficit is the consequence of an ophthalmologic lesion. Consequently, these patients are either totally unaware of their deficit, as frequently observed in cortical blindness (Chokron, 2014) or repeatedly see ophthalmologists, who are rather powerless in terms of neurovisual rehabilitation.

Surprisingly, albeit visual-field rehabilitation has been neglected, a growing number of studies on brain-damaged patients have focused on the dissociation between severely impaired explicit (conscious) vision and preserved implicit (unconscious) vision. Indeed, numerous studies in monkeys or humans with retrochiasmatic lesions have shown that some visual functions can be preserved (Humphrey and Weiskrantz, 1967). These studies have mainly promoted research in the field of perception and consciousness, but have also led to development of new visual rehabilitation programs based on the hypothesis that impaired conscious vision could be restored by training residual unconscious visual capacities.

In the review presented here, we had three objectives. Firstly, we sought to provide an overview of hemianopia and of its deleterious consequences on perception and daily life activities. Secondly, we devised new compensation and restoration techniques to treat visual field defects, emphasizing that remaining unconscious visual capacities are invaluable for restoring the visual field. Finally, we explored the possible cortical mechanisms behind restoration of visual function, by briefly overviewing neuroimaging studies on cortical plasticity in patients that had suffered from visual-field defects.

## **HOMONYMOUS HEMIANOPIA**

#### **DEFINITION**

The loss of vision in HH cannot be explained by injury to the eye itself (Hécaen, 1972): the lesion usually occurs in occipital regions that include to the primary visual cortex of the right or left hemisphere. HH can be total or partial (e.g., quadrantopia and scotoma) and have or lack macular sparing (if the region devoted to the central visual field is impaired), as illustrated in **Figure 1** (Chokron, 1996; Danckert and Goodale, 2000). Usually, these field deficits are *totally homonymous*, meaning that the blind portion of each eye can be superimposed.

#### **ETIOLOGY AND LESION LOCALIZATION**

The most frequent etiology of CVI is stroke (either ischemic or hemorrhagic). According to Marshall et al. (2010), CVI, and especially visual field defects, affect more than 15% of brain-damaged patients. Moreover, in several studies researchers have proposed that 50% of the hospitalizations in neurology departments and rehabilitation centers in England are consecutive to a stroke,

**24-2, SITA-FAST program)**. **Top:** Right homonymous hemianopia without macular sparing. Visual fields of the left and the right eye. **Bottom:** Right homonymous hemianopia with macular sparing. Visual fields of the left and the right eye.

with 30% of these patients suffering from HH (Pambakian and Kennard, 1997; Kerkhoff, 2000; Sand et al., 2013). In one striking report, the authors indicated that among 323 stroke patients, only 8% did not show any visual impairment and 49% presented visual field loss (Rowe and VIS Group, 2009). According to a recent study, 60.5% of stroke patients might present a CVI with HH affecting 35% of stroke patients in this database of 11900 cases (Ali et al., 2013).

The aforementioned findings are testament to the urgency of visual field rehabilitation in public health. HH and other visual field defects can also result from brain tumors, cerebral hypoxia, along postchiasmatic visual pathways, occipital lobectomies, trauma, progressive multifocal leukoencephalopathy (Diller and Thompson, 2007), or even degenerative diseases (Levine et al., 1993; Chokron, 1996; Kerkoff, 1999; Zihl, 2000; Meek et al., 2013). However, it is the topography and size of the lesion, rather than the type, that determine the extent and severity of the visual field defect (Tant et al., 2002; Atchison et al., 2006). Among HH, 40% imply lesions of the occipital cortex; 30% result from parietal damage; 25%, from temporal damage; and 5%, from lesions of the optic tract or the lateral geniculate nucleus (LGN; Fujino et al., 1986; Huber, 1992).

#### **CLINICAL MANIFESTATIONS OF HH**

In addition to not being able to detect visual stimuli in their contralesional visual field, HH patients suffer from other clinical manifestations, including impaired visual search/orientation in 2D and 3D space, reading difficulties (see below), and slowed and inaccurate performance in functional visual activities (Pambakian et al., 2005; Leff et al., 2006; McDonald et al., 2006). Furthermore, numerous HH patients report locomotion disabilities, especially when outdoors: for instance, they bump into other pedestrians or obstacles in their blind hemifield. Accordingly, these patients are usually not allowed to drive. They also express difficulties in building a global representation of their visual environment (Pambakian and Kennard, 1997). Reading is considerably affected by the visual field defect: patients suffer from omission of letters (in right HH) or lines (in left HH) (Zihl, 2000). Another problem that has been described in these patients is impaired visuospatial exploration: in a study of HH and patients that were asked to generate a saccade towards the blind hemifield, the HH patients exhibited less consistent oculomotor behavior and longer fixation times (Zihl, 1995). Consequently, hemianopes suffer from disorganized, and inefficient visual search strategy that requires a lot of attention to be useful. Moreover, subjective visual assessment indicates that these patients have a reduced quality of life (Papageorgiou et al., 2007). However, and surprisingly, Papageorgiou et al. (2007) found that subjective impairment does not correlate to visual-field assessment, especially when testing the correlation between the extent of the macular spare, and the subjective evaluation of the impairment in everyday activities. This finding suggests that patients with HH should benefit from a complete evaluation including objective visual-field perimetry, evaluation of locomotion and exploration capacities as well as subjective interviews.

## **RESIDUAL CAPACITIES AND BLINDSIGHT**

Although HH patients exhibit partial or complete apparent visual loss in their contralesional visual field in objective perimetric examinations, some of them present unexpected visual capacities in their blind visual field. For example, certain patients can guide one of their hands towards a small line, according to its orientation, even though they claim that they are unaware of the stimulus (Weiskrantz et al., 1974; Perenin and Jeannerod, 1975). These capacities were referred to as *blindsight* by Weiskrantz et al. (1974). As we explain below, blindsight refers to the ability to respond to visual stimuli in the blind visual field without visual consciousness.

#### **BLINDSIGHT CHARACTERISTICS**

Blindsight phenomenon has been observed in numerous tasks. Numerous experiments performed over the past several decades have shown various residual capacities in the blind field of HH patients. Using forced-choice procedures, researchers have highlighted the capacities of these patients to detect a visual stimulus placed in their blind field (Fendrich et al., 1992); to localize a visual stimulus by an eye jerk (Zihl and von Cramon, 1980) or by manual checking (Perenin and Jeannerod, 1975); to detect stimuli in movement (Riddoch, 1917) or of changing luminous intensity (Barbur et al., 1980); to discriminate among shapes (Weiskrantz, 1986); and to distinguish facial expressions (Pegna et al., 2005). For example, researchers have extensively described an HH patient named *GY* that was able to compare stimuli according to color attributes, or detect movement in his blind visual field, despite not having conscious vision of the stimuli, but that was unable to compare two degrees of luminance (Ffytche et al., 1996; Morland et al., 1999). Similar behavior has been reported in hemianopic monkeys in numerous studies by Stoerig and Cowey (1989) and Cowey and Stoerig (1995, 1997, 1999). For instance, in some studies hemianopic monkeys were able to distinguish different orientations, wavelengths and colors (for a review, see: Stoerig and Cowey, 1997; Stoerig et al., 2002). Furthermore, in hemianopic monkeys, response times towards stimuli presented in their healthy visual field can be facilitated by first displaying the stimuli in the blind field (Cowey et al., 1998).

Blindsight was not only reported for simple visual tasks such as grasping but also for tasks requiring more complex visual processing. As a matter of fact, some patients were found to be able to analyse the visual stimulus in order to perform category discrimination (Trevethan et al., 2007; Van den Stock et al., 2013, 2014) whereas de Gelder et al. (2008) were able to show that TN, a patient with cortical blindness, was able navigate and avoid obstacles although not being to report their presence. In addition, blindsight was also shown for emotional stimuli including facial and bodily expressions (e.g., Tamietto et al., 2009; Van den Stock et al., 2011) as well as for social cues such as gaze direction (Burra et al., 2013).

In a very recent study on hemianopic patients, Fayel et al. (2014) reported that these patients retained the ability to direct a saccade toward their contralesional hemifield, despite their hemifield defect, but that their verbal detection reports were at chance level. However, saccade parameters (latency and amplitude) were altered by the defect. Saccades to the contralesional hemifield in the patients exhibited longer latencies and shorter amplitudes than did those in the corresponding hemifield in a cohort of healthy subjects. Their findings confirmed previous studies on the direction of saccades in the blind field (for a review and discussion, see Cowey, 2010).

Blindsight has also been described in children. For instance, Tinelli et al. (2013) recently measured sensitivity to several visual tasks in a group of four children with congenital unilateral brain lesions that had left their optic radiations severely damaged, and in a group of three children with similar lesions that they had acquired during childhood. Using functional magnetic resonance imaging (fMRI), the authors measured blood oxygenation leveldependent (BOLD) activity in response to stimulation of each visual field quadrant. They found residual unconscious processing of position, orientation and motion of visual stimuli displayed in the scotoma in the children with the congenital lesions, but not in those with the acquired lesions.

#### **NEURAL SUBSTRATES OF BLINDSIGHT**

Visual pathways predominantly arise to the primary visual cortex (V1). Considering the global neuronal workspace (GNW) framework (Sergent and Dehaene, 2004), conscious perception would arise through the subsequent activation of several neurons, with a self-amplifying process that leads to consciousness threshold. However, if the system fails to reach said threshold, then conscious vision is not possible. Stoerig and Cowey (1995) proposed that activation of neurons in the primary visual cortex is essential for conscious vision. This premise would explain why hemianopes cannot consciously process visual information. However, visual pathways also project onto several other areas of the brain. Thus, any loss of neurons in V1 could be compensated for by the activity of other visual areas that remain stimulated by visual input. Such compensation would explain the persistence of visual capacities in the absence of conscious vision. Indeed, this idea is the central hypothesis of blindsight literature (de Gelder et al., 1999; Morland et al., 1999; Danckert and Goodale, 2000; Danckert and Rossetti, 2005; Pegna et al., 2005). Thus, researchers proposed that the underlying mechanism behind blindsight is that subcortical pathways bypass V1 to directly project onto secondary visual areas such as V5 (for motion detection), the thalamus, the brain stem, the hypothalamus, and/or the amygdala (for emotional response). In fact, this hypothesis was confirmed based on anatomical data acquired from fMRI studies. Goebel et al. (2001) identified extrastriate activations in the damaged hemisphere of the aforementioned hemianopic patient G.Y., during a forced-choice task known to elicit blindsight. Additionally, bilateral extrastriate cortex activations have been observed in several patients (Nelles et al., 2007). Unfortunately, current neuroimaging techniques are not sufficiently precise to temporally track the temporal course of visual information through subcortical pathways; consequently, the relationship between the activation of subcortical structures and blindsight is presently difficult to establish in humans (Sahraie et al., 1997).

Various hypotheses have been proposed to explain blindsight, including the presence of spared islands in V1 and the projection of visual information from the superior colliculus (SC) or LGN to preserved visual areas. We discuss these mechanisms in the following section.

#### **Blindsight enabled by spared islands in V1**

Given that blindsight is not observed in all hemianopic patients, some researchers have suggested that the residual visual capacities observed in some patients are enabled by spared islands in the primary visual cortex (i.e., areas that retain their function after the lesion) (Fendrich et al., 2001). However, this hypothesis has been partially disproved by reports of blindsight in patients lacking V1: in fact, patients that have undergone complete ablation of V1 can demonstrate blindsight (Perenin and Jeannerod, 1978). Moreover, in recent work based on diffusion tensor imaging (DTI), Leh et al. (2006) studied blindsight in hemispherectomy patients, ruling out the possibility of any spared islands in the primary visual cortex. They found ipsilateral and contralateral projections between the superior colliculi and primary or secondary visual areas, and frontal eye field projections, in patients exhibiting "attention blindsight", but not in patients that did not exhibit blindsight. Moreover, in some fMRI studies, the authors did not report residual activity in the primary visual cortex in hemianopic patients (Stoerig et al., 1998). Finally, in other studies (Ptito et al., 1996), the authors observed some residual visual abilities in patients that had suffered V1 lesions, but not in patients with SC lesions. This latter finding suggests the existence of a secondary visual pathway, one that would bypass V1 to directly transmit visual information through the superior colliculi and/or the LGN. We explore the possibility of such a pathway in the next section.

#### **Blindsight enabled by a secondary visual pathway that bypasses V1**

The proposed secondary visual pathway would represent an alternative to the major retino-geniculo-striate pathway and entail transfer of visual information to extrastriate cortical areas through the SC or the LGN. The basis for this hypothesis is the principle observation of Riddoch (1917) that hemianopic patients can perceive and/or feel movement in their blind hemifield. He observed that patients suffering from V1 lesions could process moving stimuli but could not perceive static ones. This phenomenon, known as the *Riddoch phenomenon*, can be explained by the presence of numerous projections that extend from the superior colliculi and the pulvinar nuclei to the visual extrastriate cortex (Rodman et al., 1989). The ability of patients to plan motor actions towards visual targets that they do not consciously detect in their blind visual field corroborates the implication of such a retino-geniculo-extrastriate pathway (Weiskrantz et al., 1974; Milner, 1995). Studies on monkeys and in humans (Humphrey and Weiskrantz, 1967; Girard et al., 1992; Danckert et al., 2003), in which the dorsal pathway was found to be functioning even after destruction of V1, together indicate that dorsal structures could be involved in blindsight. Accordingly, these dorsal structures would receive afferents from subcortical areas such as the superior colliculi and the pulvinar nuclei. Indeed, Schmid et al. (2010) by testing a monkey with a damage at the level of V1 confirmed that LGN is involved in blindsight phenomenon. As a matter of fact, this study showed that the good performance in perception of high-contrasted stimuli in the blind visual field of the monkey was acompanied by a significant activation at the level of the extra-striates areas: V2, V3, V4, V5 V5/ (MT), sulcus temporal superior (FST) and the lateral parietal area (LIP), thus bypassing V1. However, following a temporary inactivation of the LGN in the lesioned hemisphere the monkey could not detect anymore the visual stimuli in its blind visual field. These results show that the direct projections from the LGN to the extra-striate cortex strongly contribute to blindsight phenomenon.

Some studies on facial categorization and judgment of facial expressions corroborate the implication of a subcortical pathway. Vuilleumier et al. (2003) reported that when subjects are shown images of faces expressing fear, their amygdala is activated via their SC. However, they observed activation of the SC and the pulvinar nuclei only when the faces expressing fear were presented at low spatial frequencies known to activate the SC. Based on this observation, the authors suggested that this subcortical pathway might provide inputs to the amygdala. In related work, Pegna et al. (2005), upon studying a patient that suffered from cortical blindness following a bi-occipital lesion, observed that the patient was able to recognize feelings on faces. This phenomenon is known as *affective blindsight* and is associated to activation of the right amygdala. Tamietto and de Gelder (2008), proposed that affective blindsight could be enabled when visual information is transmitted by the subcortical pathway to the colliculi, and further on to the amygdalae, all the while bypassing V1. Moreover, Morris et al. (2001), using fMRI, confirmed that affective blindsight can be elicited through the same subcortical pathway involving the SC, the thalamus and the amygdalae (spared by V1 damage) (for a review and discussion, see: Tamietto and de Gelder, 2010). Finally, Tamietto et al. (2010) also have provided evidence that the collicular-extrastriate pathway has a crucial role in non-conscious visuomotor integration: they showed that, in the absence of V1, the SC is essential for translating visual signals that cannot be consciously perceived, into motor outputs. They presented a gray stimulus in the blind field of patients with a unilateral lesion of V1 and observed that, although the patients did not consciously see the stimulus, it nevertheless influences their behavioral and pupillary responses to stimuli consciously seen in their intact field. The authors called this phenomena *implicit bilateral summation* because the unseen stimulus could affect the response to the seen stimulus. However, it should be noted that this effect was accompanied by activation in the SC and in occipito-temporal extrastriate areas. Interestingly, when the authors instead presented their subjects with purple stimuli (which predominantly entail S-cones and consequently, are not detected by the SC), they did not observe any evidence of implicit visuomotor integration and found a massive decline in activations in the SC. Based on these findings, the authors suggest that the SC bridges cerebral sensory and motor processing, thereby contributing to visually-guided behavior that is functionally and anatomically separate from the geniculo-striate pathway and entirely external to conscious vision. Such a scenario would partly explain blindsight.

Despite the aforementioned results, in other neuroimaging studies, the authors reported an absence of relationship between blindsight, and activation of subcortical structures. The lack of evidence highlights the limits of conventional fMRI to study blindsight. This might explain why Leh et al. (2006) used DTI, rather than conventional fMRI, in order to study the neural substrate of blindsight in hemianopic patients. They tested hemispherectomy patients with a visual field defect, seeking to exclude the presence of spared islands in the visual cortex. They observed ipsilateral and contralateral projections from the SC to the primary and extrastriate visual areas in patients with type I blindsight (attention blindsight), but not in patients that did not exhibit attention blindsight or in control subjects.

More recently, Bridge et al. (2008) described new evidence for three anatomical connections that could underlie blindsight. Firstly, control subjects and the patient GY showed a tract that bypassed V1 and connected the LGN to the ipsilateral visual motion area MT+/V5 as reported by Sincich et al. (2004) in the macaque monkey. Secondly, ipsilateral pathways between MT+/V5 and LGN were found in GY lesioned and intact hemispheres as controls. Finally, they found two other pathways in GY but not in the controls: the first one crossed white-matter tracts that connect the LGN to the contralateral MT+/V5 (i.e., contralateral tracts between the LGN in one hemisphere and MT+/V5 in the other), through the splenium; and the second one was a transcallosal connection between the MT+/V5 areas in each hemisphere. These specific connections found in GY are consistent with a contralateral pathway from right LGN to left MT+/V5, and with the increased inter-hemispheric transfer of information found in several studies (Goebel et al., 2001; Silvanto et al., 2007; Bridge et al., 2008). However, whether the emergence of new connections results from strengthening of existing pathways, or from development of new pathways, remains to be determined. Moreover, since GY was 8 years old at the moment of his lesion, he had a greater likelihood of regeneration of connections than would an adult patient. Very recently, also using DTI, Tamietto et al. (2012) suggested that selective changes may occur in patients with CVI in the SC, the pulvinar and the amygdala. The authors suggested that these changes may explain the residual sensitivity to emotions or social signals in blindsight (Tamietto et al., 2012).

In summary, blindsight might be enabled by retinotectal projections that bypass V1 and could result from connections specific to it. However, its dynamics remain scarcely understood. Recently, Ioannides et al. (2012) used magnetoencephalography to study the spatiotemporal profiles of visual processing and the causal contribution of V1 in three neurologically intact participants and in GY in whom residual visual functions mediated by the extrageniculostriate pathways have been reported. Whereas normally perceived stimuli in the left hemifield of GY elicited a spatiotemporal profile in the intact right hemisphere that closely matched that of healthy subjects, stimuli presented in his contralesional hemifield produced no detectable response during the first phase of processing. The authors reported that in contrast to responses in the intact hemisphere, the back-propagated activity in the early visual cortex did not exhibit the classic retinotopic organization and did not have well-defined response peaks thus suggesting a modification in the spatiotemporal profiles of visual processing after a unilateral destruction of V1.

Below we present the spontaneous recovery of HH, and then examine the various rehabilitation techniques that have been proposed to compensate for and/or reduce this related visual-field defect.

## **SPONTANEOUS RECOVERY OF HH**

Patients can spontaneously recover from HH, but the probability of such recovery is proportional to the time that has elapsed since the lesion occurred. Reported recovery rates range from 7% to 86% (for a review, see: Sabel and Kasten, 2000). Furthermore, a large area of residual vision is a better predictor of spontaneous recovery following stroke than is a small area. In a 15-year longitudinal study, Zhang et al. (2006b) analyzed spontaneous recovery in hemianopia patients. They observed recovery approximately 38.4% of the cases within the commonly accepted period of 6 months (after which, the HH becomes chronic). The chance of recovery diminished with increasing time since injury: at 1 month, the rate was > 50%, whereas at 6 months, it was only 20%. Moreover, in most cases, the recovery was very limited: the patients had only regained a few degrees of vision; indeed, only 5.3% of the patients had completely recovered from their visual field defect. Some HH patients adopt spontaneous compensations, especially in terms of exploration strategies, that are quite different from those in non-hemianopic subjects (Pambakian et al., 2004). Sabel and Kasten (2000) considered that after 3–6 months post-lesion, partially recovered patients can only achieve further improvement through visual capacities training.

## **REHABILITATION OF HH PATIENTS**

Although HH does not seem to be as debilitating as spatial neglect, it can seriously affect daily activities such as driving, walking in crowded areas, crossing the street, and reading. Problems with these and other activities can pose serious problems for recovering HH patients upon their return to work. Given the poor rate of spontaneous recovery in HH, several training programs have been proposed to help patients recover. These programs can be classified into three categories according to their objectives for the visual-field deficit: *substitution, compensation* and *recovery*.

#### **SUBSTITUTION THERAPIES**

An early but now-defunct substitution technique entailed the use of optical aids (e.g., mirrors or Fresnel prisms) to shift visual information from the blind visual field to the central or the ipsilesional, preserved visual field. However, there were only anecdotal reports of the positive effects of this approach; furthermore, such tools were reported to cause diminished acuity, confusion or even diplopia in patients (for a review, see: Pambakian and Kennard, 1997; Grunda et al., 2013). Therefore, this technique is no longer used in rehabilitation of HH patients.

Currently, compensatory techniques for HH are used principally to enlarge and reinforce visual search, by training patients in oculomotor strategies. Indeed, there are extensive reports that hemianopic patients have difficulties in visual scanning for object detection, as well as in identifying people. These problems can lead to omission of important parts of a scene and consequently, to poor comprehension and to social misunderstanding. Parafoveal visual-field defects also affect reading (*hemianopic alexia*), due to a reduction in the "perceptual window" involved in letter identification and in saccades planification and thus compromizing guidance of eye movements along text (Poppelreuter, 1971[1990]; Zihl, 1995; McDonald et al., 2006). These deficits can selectively be trained in hemianopic patients (Schuett et al., 2012).

#### **COMPENSATION THERAPIES**

Compensation therapies can be proposed regarding the fact that recovery from a very severe perceptual deficit can be difficult to obtain. Therefore, they typically involve using and modifying the patient's preserved capacities to sidestep the impairment or render it less disabling. Accordingly, compensation strategies for HH require the use of the ipsilesional hemifield or of the central visual field to compensate for the blind area in the contralesional hemifield.

In classical oculomotor rehabilitation, patients are trained to search for a stimulus projected into their blind hemifield, and then respond to it as quickly as possible. The target can be presented alone or amongst distracters (Zihl and Werth, 1984). Researchers usually record reaction times (RTs) and error rate, given that an inefficient search will lead to longer RTs. These techniques are generally efficient, with shorter RTs, but sometimes lead to longer RTs (Pambakian et al., 2004). Although longer RTs could be considered as being detrimental, in some cases, they have been described as an improvement: the authors of these studies affirm that longer RTs might reflect underlying compensatory mechanisms (i.e., the development of a new strategy) that probably need more time to progress and reach maximum efficacy. These types of compensation techniques are principally based on *top-down* mechanisms, because they train patients to focus their attention on the blind hemifield.

Another compensation technique involves a *bottom-up* strategy based on multisensory stimulation and integration (Bolognini et al., 2005). According to these authors, this therapy does not require the patient's attention, which can be impaired in some patients and therefore, enables more interesting perspectives for rehabilitation. They based their rehabilitation technique on the existence of neurons that encode information coming from different sensory modalities, in the superior colliculi and other parts of the brain. As the superior colliculi are involved in gaze orientation, and the visual modality is impaired in these patients, Bolognini et al. proposed that a non-visual modality (e.g., auditory) could be harnessed to reinforce gaze orientation towards the blind hemifield. They presented their patients with audiovisual stimulation to help them find a subsequently presented visual target. The authors observed a greater increase in visual oculomotor exploration upon the addition of an auditory cue together with the visual stimulus than without it. Thus, multisensory stimulation enabled the capacities of a fully functioning sense (hearing) to be transferred to a deficient sense (vision).

Finally, other researchers have focused on the reading impairment caused by HH. For instance, Spitzyna et al. (2007) induced an optokinetic nystagmus in order to facilitate reading in hemianopic patients. They presented patients with right-to-left moving text that the patients had to read. The right-to-left visual movement induced a left-to-right nystagmus that increased reading speed.

Although the aim of these compensatory techniques is not to restore *per se* the impaired visual field, they nevertheless improve the quality of life of patients. Indeed, overall, in these studies the authors report subjective improvement in everyday life, according to subjective questionnaires (Bolognini et al., 2005). Therefore, compensation therapies seem to be a first step that should be systematically proposed for hemianopia treatment when restoration therapy is unavailable. However, and as we previously mentioned, since these techniques are compensatory in nature, the visual impairment remains. In addition, there are reports that the ipsilesional hemifield of hemianopic patients, when thought to be intact, can actually also be impaired—not in terms of visual-field loss, but in terms of quality of vision (Paramei and Sabel, 2008; Bola et al., 2013). Thus, relying exclusively on the "intact" visual field may not be the best option in rehabilitation techniques. Furthermore, special care for visual-field improvement, and restoration therapies of the blind hemifield, are needed whenever possible. In the following section we discuss restoration therapies.

### **RESTORATION THERAPIES**

Attempts to enlarge the visual field appear to be a more encouraging way to help hemianopic patients. Unfortunately, as we explained in the previous section, restoration of visual function may seem impossible in HH patients. This might be linked to the fact that, even if a few studies demonstrated in particular condition (see above) that visual experience is possible even in the absence of V1, generally V1 is considered to be crucial for visual consciousness (Weiskrantz et al., 1974; Perenin and Jeannerod, 1975; Tamietto et al., 2010).

Thus, recovery of explicit conscious visual detection in the absence of V1 would imply a degree of neural plasticity that is known to be unattainable after a normal rearrangement during the first 3 months post-lesion. Moreover, whether any reorganization that did occur would be powerful enough to activate conscious vision is unknown. However, studies in animals and humans have shown that perceptual learning is possible in hemianopia (i.e., training can improve visual perception) (Fahle and Poggio, 2002). A few researchers have hypothesized that the size of the visual field could be increased to enable recovery from blindness (Chokron et al., 2008). In the next section we describe oculomotor rehabilitation techniques, followed by restorative techniques using blindsight.

#### **Oculomotor rehabilitation techniques**

Most of these studies derive from the use of compensatory techniques: the authors found that training HH patients to direct saccades towards the border zone of their blind field could partially increase the size of their visual field (van der Wildt and Bergsma, 1997). These results led to the development of a special training platform, called Visual Restoration Therapy (VRT; Sabel and Kasten, 2000), to enlarge the size of the isolated patches of residual vision in the blind hemifield. Patients were trained at home, on their TV screen, with a computerized program that is adapted to their visual capacities and evolves according to their progresses. They were asked to fix their gaze in the center of the screen, and trained to detect a target placed in the border zone of their hemianopic field, by pressing a button whenever they saw it (Sabel et al., 2005; Kasten et al., 2006). Partial improvement was found in some of the studies. For example, in one study, 15 patients trained with VRT for 3 months showed only a 3.8% increment in stimulus detection (Kasten et al., 2006). In a recent, larger and longerterm study, in which 302 patients were trained with VRT for 6 months (Mueller et al., 2007), 17.2% of the patients exhibited increased detection of supra threshold stimuli, although 29.1% did not show any improvement. However, the field extension averaged 4.9 ± 0.41◦ . Subjective assessment showed improved visual confidence in 75.4% of the patients. Overall these authors have been criticized for a fastidious program that would induce only weak improvement (Glisson and Galetta, 2007). Horton (2005) explained that patients showed subjective improvement, and emphasized that this improvement was only measured with the device that was used for therapy. Indeed, when tested on classical automatic perimeter, no objective enlargement of the visual field was recorded.

A training device named the Lubeck Reaction Perimeter was created by Schmielau (1996), and then evaluated by Schmielau and Wong (2007) in a study of 20 patients that were trained twice a week for an average of 8.2 months. The patients were seated in front of a large size hemispheric half-bowl filled with LEDs that would light up according to their visual capacities, as in the VRT. The patients were asked to fixate on the red central LED during testing. The results were interesting: 17 out of the 20 patients exhibited an improvement in detection rate in their impaired field (average rate: 11.3◦ ± 8.1). However, among the drawbacks of the Lubeck Reaction Perimeter are its large size and the fact that it is not amenable to home use (Sabel, 2008).

A completely different approach to that used in the aforementioned devices is one based on that hypothesis that conscious visual detection could be restored by training unconscious visual processing capacities (i.e., blindsight) in hemianopia.

#### **From blindsight to sight**

As we mentioned above, some patients can actually perform visual tasks in their impaired hemifield, despite claiming that they cannot see or feel anything (Weiskrantz et al., 1974). Although Ducarne and Barbeau (1981) and Ducarne de Ribaucourt and Barbeau (1993) did not relate their visual training to blindsight, they were the first authors to report the need to stimulate the blind visual field by using tasks of localization and detection of salient visual targets associated with prehension. Indeed, Danckert et al. (2003) subsequently proposed that the driving action directed in the defective visual field enables strengthening of visual perception. Thus, asking patients to perceive, judge, recognize, locate or grasp stimuli in their hemianopic visual field could help them to "relearn" how to see.

Although blindsight had been extensively studied from the theoretical and experimental perspectives over the past few decades, only in the past 15 years did a few authors begin to hypothesize that blindsight capacities could be improved through training (Sahraie et al., 2006) and therefore, that unconscious vision might be transformable into conscious vision (Ro and Rafal, 2006). Firstly, Zihl (2000) observed that patients that had been trained to locate targets in their blind hemifield gained subjective improvements in sight for everyday life activities. More recently, Sahraie et al. (2006) claimed that visual sensitivity in the blind hemifield could be improved without the patient's awareness, and that this could be done even in the very depth of the impaired hemifield (and not only in the border zone of blindness, as suggested in previous studies). The authors "trained" 12

patients to discriminate grating stimuli from non-grating ones in their blind hemifield over 3 months. Before and after the treatment, they tested the patients for detection of various levels of contrasts and spatial frequencies, clinical perimetry, and a subjective estimate of the visual-field defect. Overall, the patients exhibited improved sensitivity in target detection at various contrast sensitivity levels and spatial frequencies, and objective improvements at clinical perimetry tests. Therefore, authors concluded that training blindsight could be a way to improve visual field defect in hemianopia. In related work, Raninen et al. (2007) trained patients to detect flickering stimuli and to discriminate letters at various eccentricities within the blind hemifield. The patients were tested twice weekly for roughly 1 year, ultimately improving in these tasks, but not in Goldman perimetry.

Considering that objective improvement has been reported in some studies on blindsight training, and that the use of tasks that are more ecological might help patients to recover, Chokron et al. (2008) trained nine brain damaged patients on various forced-choice tasks in their blind hemifield, including visualtarget pointing, letter recognition, comparison of two stimuli presented in both hemifields and target location. Such tasks are usually proposed to test for blindsight. Patients took pre- and post-tests (letter identification and target detection), and were evaluated with an automated perimetry exam. They all improved in behavioral tasks, and eight of the nine patients exhibited a significant enlargement of their visual field (as determined by classical perimetry examination). These findings suggest that stimulating blindsight can help the patients to be aware of unconscious perceptions that typically remain unknown and unused by the patient if not trained. Furthermore, Sahraie et al. (2013) recently tested five HH patients on a forced-choice detection task, observing improvement in four of the patients. With repeated stimulation, detection in the hemianopic field improved.

Interestingly, therapies visual or blindsight stimulation in the contralesional visual field are presently being evaluated in children with visual-field defects, consecutive to perinatal asphyxia, or traumatic brain injury (Dutton and Bax, 2010; Pawletko et al., 2014). However, no clinical trials have yet been performed in this area.

The possibility of visual-field restoration raises the question of whether some type of cortical reorganization occurs after a unilateral occipital lesion. As we explain below, further neuroimaging studies on patients before and after blindsight stimulation are required in order to understand how vision might be regained.

## **CORTICAL REORGANIZATION AFTER AN OCCIPITAL LESION CORTICAL REORGANIZATION IN HEMIANOPIC PATIENTS**

Plasticity is a fundamental mechanism for the brain to adjust to sensory changes in the surroundings, to improve perception and to recover from damage to the visual system. Studies on neural plasticity have shown that the brain can react to environmental inputs after a lesion, in infants, children and even seniors. Thus, researchers have demonstrated that part of the areas adjacent to a lesion can replace the function of the affected one, as can occur after sensory-motor defects (Liepert et al., 2000), or that the healthy (undamaged) hemisphere can reorganized itself to take over part of the damaged hemisphere's functions (Netz et al., 1997; Johansen-Berg et al., 2002; Rossini and Dal Forno, 2004). In related work, Safran and Landis (1999) suggested that even in adult brains, cortical maps are not fixed.

Studying blindsight enables researchers to determine, in the absence of the primary visual system, whether the spared or recovered visual ability stems from an existing alternative pathway or from newly formed pathway. As described above, Bridge et al. (2008), using DTI, and Silvanto et al. (2009), using transcranial magnetic stimulation (TMS), each observed pathways in patients with visual-field defects that they did not observed in control patients. Their findings suggest that after a brain lesion, specific connections can be created. Nelles et al. (2007) conducted a related study in thirteen hemianopic patients that had suffered a unilateral ischemic stroke in the striate cortex. Stimulation in the blind visual field led to bilateral activation in the extrastriate areas, and this activation was stronger in the ipsilateral (healthy) hemisphere. This result suggests that activation had been transferred from the damaged hemisphere to the healthy one.

Interestingly, in most of the published studies concerned with cortical reorganization in hemianopic patients, the authors did not address the side of the lesion. This issue was tackled by Perez et al. (2013), who sought to determine the effects of the lesion side on cortical reorganization in brain-damaged patients. They demonstrated that the pattern of cortical activation during a visual detection task and a visual categorization task depends on the side of the occipital lesion. Indeed, RHH patients showed activation predominantly in the right (intact) hemisphere (occipital lobe and posterior temporal areas), whereas LHH patients showed more bilateral activation (in the occipital lobes). Thus, lateralization of an occipital lesion is not without consequence for the subsequent cortical reorganization. Along those lines, the impact of the lesion lateralization might be crucial to the cortical reorganization and consequently, to the choice of rehabilitation scheme for left or right hemianopic patients.

The main question when studying brain reorganization after recovery is: *Does recovery involve recruitment of existing pathways, establishment of new neural connections or both?* We address this question in the following section.

#### **CORTICAL REORGANIZATION IN HEMIANOPIC PATIENTS AFTER VISUAL RECOVERY**

As explained above, various techniques are now available to compensate for damage, or restore vision, in the hemianopic visual field. Regarding the functional basis of these rehabilitation techniques, some studies have suggested that visual training can induce neural modifications in V1 (Furmanski et al., 2004; Maertens and Pollmann, 2005). For example, Raninen et al. (2007) demonstrated that intensive visual training can improve abilities in the blind visual field. After training, their patients achieved a level of performance in the blind visual field similar to that in the healthy visual field. Furthermore, this gain was underlined by changes in cortical activation. Indeed, after training, the patients exhibited an ipsilateral response to blind visualfield stimulation. In a related fMRI study, Marshall et al. (2008) were the first to report changes in cerebral responses to stimuli after visual training in hemianopic patients. They trained the patients with repetitive stimulation of the border zone. After 1

month of training, the patients exhibited increased BOLD activity for border-zone detection compared to detection in the healthy part of the visual field. After visual training, the most activated areas were the right inferior and lateral temporal areas, the right dorsolateral frontal areas, the bilateral anterior cingulate cortices and the bilateral basal ganglia. Moreover, these authors observed a correlation with behavioral data obtained out-of-scanner, which revealed an improvement in response times for detecting stimuli in the border zone. The authors concluded that visual training induced a shift of attention from the non-trained seeing field to the trained border zone, and stated that the effect seemed to be mediated by frontal regions and other higher-order visual areas. In related work, Ro and Rafal (2006) trained patients, and then observed an ipsilateral response to blind visual-field stimulation. They proposed that the visual pathway used in blindsight—in this case, the sub-cortical pathway—is probably the same one that underpins recovery after training.

Henriksson et al. (2007) trained patients, and then studied them by fMRI, observing that visual information in both hemifields was processed in the intact hemisphere. Thus, training patients for visual detection in the hemianopic field induced cortical representation of this visual field in their ipsilateral, healthy hemisphere. As such, after training, the visual areas of the healthy hemisphere (in particular, V5) might have represented not only the contralateral field, but also the ipsilateral visual field. This premise is consistent with the findings of Nelles et al. (2007), in their study of hemianopic-field stimulation in patients with occipital strokes: they found that stimulation induced bilateral activation within the extrastriate cortex, and this activation was stronger in the ipsilateral (contralesional) hemisphere. Furthermore, the activation pattern in the patients was different to that observed in normal subjects, who exhibited contralateral activation of the striate and extrastriate regions (as similarly observed by Silvanto et al., 2007, using TMS).

Recently, Plow et al. (2012) studied a 3-month regime of VRT in two patients suffering from a left occipital lesion (Right HH). They coupled sessions of visual training with TMS of the occipital area. They observed better recovery in the patients (who had received TMS) compared to control patients (who had received sham stimulations). Using fMRI, they also recorded activation around the lesioned area and bilateral activation of the associative visual areas.

Taken together, the aforementioned studies suggest that visual training in the blind visual field, regardless of whether it is coupled to TMS, can induce cortical reorganization. Nevertheless, more studies are needed in order to standardize these rehabilitation programs as well as to understand the neural bases of recovery. Furthermore, psychophysical studies are required to determine the extent to which the recovered vision in the hemianopic visual field resembles the vision in the ipsilesional field.

## **CONCLUSION**

Since hemianopia involves impairment in early perceptual processes, recovery from it was once considered impossible. However, recent findings on blindsight offer new perspectives for recovery of visual function in patients with a post-chiasmatic damage. From a theoretical point of view, we still need to understand how training unconscious visual capacities can lead to a restoration of conscious visual capacities. According to the GNW model (Dehaene and Changeux, 2011) conscious access occurs when incoming information is made globally available to multiple brain systems through a network of neurons with long-range axons densely distributed in prefrontal, parieto-temporal, and cingulate cortices. We still need to understand how stimulating blindsight can promote conscious perception. However, one can hypothesize that responding to a stimulus, although without being conscious of it, may in turns, activate parts of this network and in this way induce conscious access. As a matter of fact, Sergent et al. (2013) recently tested the influence of postcued attention on perception, using a single visual stimulus (Gabor patch) at threshold contrast in healthy participants. The authors showed that postcued attention can retrospectively trigger the conscious perception of a stimulus that would otherwise have escaped consciousness. One can thus hypothesize that acting on a stimulus which was not consciously detected might act as a postcue and thus produce conscious perception in the blind field of hemianopic patients. Along those lines, stimulating blindsight could help restoring conscious perception.

Nevertheless, further testing is required in order to confirm the findings that we have summarized here as well as to explain how they help vision recovery. Furthermore, rehabilitation programs should be generalized and standardized in order to facilitate identification of those patients that are the best candidates for treatment and to enable better treatment. Therefore, and as has been proposed for rehabilitation of spatial neglect, we propose that rehabilitation of hemianopia should include a combination of compensatory and restorative techniques tailored to each patient. Indeed, the majority of studies have shown that the use of both types of techniques should help patients to make improvements in everyday life activities.

Future research on blindsight should include neuroimaging studies. One interesting line of research would be to study the link between neural activation and post-training recovery in HH patients. Another interesting study would be to use fMRI to compare HH patients during the acute phase and after training, once they have recovered their field of vision, in order to understand the neural substrates of visual recovery. Finally, researchers should also compare HH patients that have recovered vision to those that have not, in order to determine if the two groups exhibit the same cortical reorganization.

#### **REFERENCES**


Fahle, M., and Poggio, T. (2002). *Perceptual Learning.* Cambridge, MA: MIT Press.


**Conflict of Interest Statement**: The Guest Associate Editor Olivier A. Coubard declares that, despite having collaborated in the past with authors Céline Perez and Sylvie Chokron, the review process was handled objectively and no conflict of interest exists. 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: 06 June 2014; accepted: 30 September 2014; published online: 22 October 2014*.

*Citation: Perez C and Chokron S (2014) Rehabilitation of homonymous hemianopia: insight into blindsight. Front. Integr. Neurosci. 8:82. doi: 10.3389/fnint.2014.00082 This article was submitted to the journal Frontiers in Integrative Neuroscience*.

*Copyright © 2014 Perez and Chokron. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and 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*.

## Visualizing the blind brain: brain imaging of visual field defects from early recovery to rehabilitation techniques

## **Marika Urbanski 1,2,3,4,5\*, Olivier A. Coubard6,7 and Clémence Bourlon<sup>8</sup>**

<sup>1</sup> Service de Médecine et de Réadaptation Gériatrique et Neurologique, Hôpitaux de Saint-Maurice, Saint-Maurice, France


#### **Edited by:**

John J. Foxe, Albert Einstein College of Medicine, USA

#### **Reviewed by:**

Georgios A. Keliris, Max Planck Institute for Biological Cybernetics, Germany Marco Tamietto, Tilburg University, Netherlands Simon Grant, City University London, UK

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

Marika Urbanski, Service de Médecine et de Réadaptation Gériatrique et Neurologique, Hôpitaux de Saint-Maurice, 12-14 Rue du Val d'Osne, 94410 Saint-Maurice, France e-mail: marika.urbanski@gmail.com Visual field defects (VFDs) are one of the most common consequences observed after brain injury, especially after a stroke in the posterior cerebral artery territory. Less frequently, tumors, traumatic brain injury, brain surgery or demyelination can also determine various visual disabilities, from a decrease in visual acuity to cerebral blindness. Visual field defects is a factor of bad functional prognosis as it compromises many daily life activities (e.g., obstacle avoidance, driving, and reading) and therefore the patient's quality of life. Spontaneous recovery seems to be limited and restricted to the first 6 months, with the best chance of improvement at 1 month. The possible mechanisms at work could be partly due to cortical reorganization in the visual areas (plasticity) and/or partly to the use of intact alternative visual routes, first identified in animal studies and possibly underlying the phenomenon of blindsight. Despite processes of early recovery, which is rarely complete, and learning of compensatory strategies, the patient's autonomy may still be compromised at more chronic stages. Therefore, various rehabilitation therapies based on neuroanatomical knowledge have been developed to improve VFDs. These use eye-movement training techniques (e.g., visual search, saccadic eye movements), reading training, visual field restitution (the Vision Restoration Therapy, VRT), or perceptual learning. In this review, we will focus on studies of human adults with acquired VFDs, which have used different imaging techniques (Positron Emission Tomography, PET; Diffusion Tensor Imaging, DTI; functional Magnetic Resonance Imaging, fMRI; Magneto Encephalography, MEG) or neurostimulation techniques (Transcranial Magnetic Stimulation, TMS; transcranial Direct Current Stimulation, tDCS) to show brain activations in the course of spontaneous recovery or after specific rehabilitation techniques.

**Keywords: visual field defect, plasticity, cortical reorganization, rehabilitation, restoration, neuroimaging studies**

#### **INTRODUCTION**

Most studies interested in visual field defects (VFDs) have concentrated on the more prevalent ones. Complete homonymous hemianopia (HH) represents 70–75% of VFDs (Duquette and Baril, 2009), incomplete hemianopia (e.g., quadrantanopia) 29% of VFDs (Zhang et al., 2006a), and cerebral blindness—which is rare because it usually follows bilateral lesions—represents less than 10% of VFDs given only vascular context (Aldrich et al., 1987; Brandt et al., 2000; Niimi et al., 2008).

Principal etiologies of HH are strokes in the posterior cerebral artery territory (PCA), traumatic brain injury (TBI), and tumors (see **Table 1**, which displays the different etiologies reported in the literature and the percentage of associated VFDs).

After a stroke, 45% of the lesions involved the occipital lobes and 32.2% the optic radiations (Zhang et al., 2006a); after TBI 12.5% of lesions involved the occipital lobes and 23.2% in association with a lesion of the optic radiations (Bruce et al., 2006). Most VFDs occurred after a lesion in the primary visual cortex (V1) although a lesion in the early extrastriate areas has been exceptionally reported to give rise to VFDs (e.g., patient with a lesion in ventral V3 and V4 presenting a right upper homonymous quadrantanopia, Slotnick and Moo, 2003; patients with a lesion in MT+/V5 presenting motion blindness or "akinetopsia", Zeki, 1991; Zihl et al., 1991; Vaina et al., 2001).

In France, there have been very few studies on the recovery of VFDs, or if they exist, they rely on relatively small samples. In the UK, a recent review reported visual loss in 45–67% of patients in the acute phase of the stroke, and in the long term for 8–25% of patients following adjustment for recovery of visual field (Rowe et al., 2013). Concerning HH, about 90,000 to 120,000 new cases

<sup>2</sup> Inserm, U 1127, ICM FrontLab, Paris, France


#### **Table 1 | Etiologies of HH and bilateral cortical blindness (bilateral CB) reported in the literature**.

N, number of patients included in the studies; VFDs are given in percentages in each study; –, not reported in the study.

per year both in the US and in Europe are reported by Sahraie (2007). Ajina and Kennard (2012) reported 30% of patients at the acute phase of the stroke (Haerer, 1973) and 8–26% of patients left with persistent HH (Gray et al., 1989; Gilhotra et al., 2002). About 6% of patients are left with cerebral blindness at a chronic stage (Zihl, 2000).

In this review, we will focus on studies in human adults with acquired VFDs, in which a neuroimaging technique (Positron Emission Tomography, PET; Diffusion Tensor Imaging, DTI; functional Magnetic Resonance Imaging, fMRI; MagnetoEncephalography, MEG) or a neurostimulation technique (Transcranial Magnetic Stimulation, TMS; transcranial Direct Current Stimulation, tDCS) has been employed to document brain changes. We will first present neuroimaging studies documenting the changes in the brain in the context of spontaneous recovery and then other studies documenting the changes in the brain after the use of a rehabilitation technique. For the purpose of this review, we only included VFDs in the context of postgeniculate lesions. Eye diseases or optic nerve pathologies were excluded. Studies in lesioned animals, in children or employing only perimetries or questionnaires to document the improvement of VFDs were also not included in this review.

#### **EARLY SPONTANEOUS RECOVERY**

Full spontaneous recovery of VFDs is rare: 5% (Duquette and Baril, 2009); 10% patients with hemianopia within the first 2 weeks (Gray et al., 1989; Pambakian and Kennard, 1997). However, many studies have shown that a partial and early spontaneous recovery may occur after brain injury although its quickness is still being debated.

Zhang et al. (2006b) have shown that in 254 patients with hemianopia followed over a 15-year period, 50–60% chances of recovery occurred at 1 month, 20% at 6 months, whereas no patient improved after 6 months. Similarly, Perez et al. (2009) reported that in 101 patients with hemianopia, 40% had spontaneous recovery within 3 months though the recovery only concerned a part of the visual field, consistent with other prospective studies showing improvement only in the peripheral zones of the lower quadrants (Celebisoy et al., 2011) or limited to 3–7 degrees of visual angle depending on the extent of sparing in the affected hemifield (Zihl, 2000).

Recovery seems to be limited to the first 6 months after the injury and may also depend on the lesion site. Bosley et al. (1987) showed that the impairment of glucose metabolism in the striate cortex measured with PET did not change over time in three patients with HH after a lesion involving V1 whereas it improved over time in two patients with HH after lesion involving extrastriate areas sparing V1. This improvement in striate metabolism was associated with a recovery of their VFDs, contrary to the former three patients. Thus recovery is restrictive in time and in space (Zhang et al., 2006b) because the spontaneous plasticity following V1 damage may be related to changes in the properties of neural circuits beyond the lesion and to a decreased inflammation around the lesion site during the first few weeks after damage (Huxlin, 2008).

Neuroimaging studies have attempted to document the neural changes during spontaneous early recovery. Raposo et al. (2011) studied eight patients with an infarct in the posterior cerebral artery territory presenting VFDs, who were examined within the first month of the stroke, 1 month later and 3 months later. Five of eight patients had restricted V1 ventral lesion and three of eight had V1 ventral and dorsal damage. Patients underwent a neurological, neuropsychological and ophthalmological examination, visual behavioral study (conscious color and motion perception) and fMRI (visual stimuli targeting motion and color presented separately within each hemifield). The authors reported that color and motion vision recovery were complete or subcomplete 1 month after the onset of the stroke. At the acute phase, there was no ipsilesional V1 activation for color or motion stimuli, while it appeared in color perception at follow-up. There was also nonspecific bilateral V4 activation in color task and nonspecific contralesional MT+/V5 activation in motion task. With time, activations in MT+/V5 and V4 bilaterally became more specific and correlated with performance.

Polonara et al. (2011) reported the case of a 24-year-old woman with left hemianopia who underwent fMRI and DTI in the acute phase and 1 month after an ischemic stroke involving

<sup>1</sup>This article review reported results from 4 series of PCA stroke leading to a total of 322 patients. Percentage ranges correspond to the minimal and maximal percentages of VFDs reported by Brandt et al. (2000).

the right calcarine cortex. At the acute phase, ipsilesional V1 did not show any activation when peripheral stimulation was presented in the left hemifield. The mean fractional anisotropy (FA) measured by DTI in the ipsilesional optic radiations was reduced compared with the left hemisphere. At 1-month followup, both right and left V1 elicited comparable activations in response to stimulation in the contralateral hemifield. Mean FA in the optic radiations was more similar in both hemispheres. Similarly, Yoshida et al. (2006) reported the case of a 68-year-old man presenting right hemianopia after an infarction of the left extrastriate areas who underwent fMRI and DTI over the ensuing 12 months. Functional magnetic resonance imaging was acquired 2 days, 9 days, 30 days and 1 year after the onset of the stroke. The results showed that larger areas of left cortical activation were activated progressively and that the asymmetry between the activations of both hemispheres decreased. Diffusion tensor imaging was acquired 2 days, 9 days and 1 year after the onset of the stroke and a tractography of the optic radiations was performed. At 2 days after onset, fiber tracking was completely interrupted in the left side due to the cortical lesion, whereas 1 year later left optic radiations could be reconstructed by fiber tracking. The authors concluded that the initial larger recruitment of cortical areas in the intact hemisphere decreased with recovery along with a progressive increasing activation in the lesioned hemisphere.

In summary, spontaneous recovery of VFDs occurs during the first 6 months following brain damage (mostly due to stroke and TBI) with a peak in recovery after 1 month. Consistently, functional brain imaging using PET, fMRI or DTI shows progressive activation of cortical and subcortical areas with increasing recovery, which correlates with psychophysical performance. The lesion side may have differential effects on recovery and prognosis.

#### **RESIDUAL VISION IN THE CHRONIC PHASE**

At a more chronic phase after brain injury (usually after 6 months after the lesion), VFDs become more stable and one can observe some phenomena of residual vision in the affected visual field. Studies in monkeys and later in humans have demonstrated visual function persistence even when the primary visual cortex had been destroyed (Cowey and Stoerig, 1995; Weiskrantz, 2004; Stoerig, 2006). Contrary to anosognosia where patients are unaware of their deficits (e.g., Anton's syndrome, for review see Bisiach and Geminiani, 1991), patients with blindsight are aware of their deficits but unaware of their intact functions. For example, patients could perform visual discrimination in the blind hemifield though they persisted in saying that they could not see anything. This phenomenon, called "blindsight", has been first described in patient DB in whom right primary visual cortex was removed and the left visual field (LVF) was defected (Weiskrantz, 1986, 1996). According to Weiskrantz, blindsight may be defined as "a visual discrimination in the absence of awareness" and may be separated in two sub-types: (1) Type I for an unconscious version: patient seemed to be able to detect visual target aspects without any conscious awareness of the stimulus presented and (2) Type II for a residual vision accompanied by a level of awareness: patients reported a feeling that something happened or moved without real visual experience.

According to the nature of the task performed, a different classification of blindsight capacities was proposed (Danckert and Rossetti, 2005). Action-blindsight was used to refer to patients who were able to localize a stimulus not consciously perceived in blind field by pointing or making saccades; attentionblindsight was used to refer to patients who were able to discriminate the direction of motion of a stimulus without action and who have the feeling that something happened in the visual field. Finally, 'Agnosopia' defined abilities to discriminate forms or colors without conscious awareness (Zeki and Ffytche, 1998).

Evidence for basic residual visual motion (Barbur et al., 1980; Weiskrantz, 1986; Schoenfeld et al., 2002; Morland et al., 2004), shape (Barbur et al., 1993; Stoerig and Cowey, 1997; Goebel et al., 2001), and color in the blind field (Stoerig, 1987; Barbur et al., 1998; for a review see Huxlin, 2008) has been reported in the literature consistently.

More recently, evidence for residual vision has been demonstrated for non basic visual properties of stimuli such as category discrimination (patient DB in Trevethan et al., 2007; patient TN with bilateral CB following two consecutive occipital strokes less than 2 months interval in Van den Stock et al., 2013) or navigation skills (patient TN in de Gelder et al., 2008).

Another type of blindsight has been described and has led to many interesting studies: the affective blindsight, in which patients are able to process emotional cues presenting in their blind hemifield (de Gelder et al., 1999). Pegna et al. (2005) studied patient TN using fMRI and showed that he was able to guess the valence of the facial expressions (positive/negative) on photographs presented to him. This ability correlated with the activity of his right amygdala, consistent with the results of Morris et al. (2001) and of Tamietto and de Gelder (2008) in patient GY, who is a right hemianopic patient extensively studied. Affective blindsight has been shown in studies using fMRI for the perception of dynamic whole-body emotional expressions (patient GY in Van den Stock et al., 2011), for perception of body and facial emotional expressions (patient TN in Van den Stock et al., 2014) and for perception of gaze direction (patient TN in Burra et al., 2013).

However blindsight is not present in all patients with VFDs and depends on particular neurophysiological properties of subcortical structures (the superior colliculus and the pulvinar which receive visual information from the magnocellular pathway). Indeed, blindsight is sensitive to the temporal characteristics of the stimuli (5–20 Hz in Sahraie et al., 1997; 10–33 Hz inTrevethan and Sahraie, 2003), to the spatial channel of the stimuli (low spatial frequency <3.5 cycles/degree in Sahraie et al., 1997, 2010; Trevethan and Sahraie, 2003), facilitated by an increase in stimulus size (Sahraie et al., 1997) or for high contrast stimuli (Ffytche et al., 1995), but insensitive to short wavelength (Tamietto et al., 2010). Therefore, different underlying anatomical mechanisms have been proposed to explain blindsight and many studies have concentrated on the brain activations associated with this phenomenon.

#### **SPARED CORTICAL ISLANDS OF V1**

Some authors have postulated a direct relation between the preserved portion of striate areas and blindsight in the corresponding visual field (Fendrich et al., 2001; Morland et al., 2004).

Specifically Morland et al. (2004) used fMRI with static, moving and flickering stimuli in seven hemianopes and one patient with a clearly spared region of the visual field in an otherwise blind hemifield. Their findings support the existence of small, spared active regions of V1 that mediate residual vision in some patients, consistent with the results of Raposo et al. (2011). However, some patients with V1 damage showed no activation in the striate cortex in fMRI but could still present a blindsight phenomenon, which does not support the hypothesis of the existence of spared islands in V1 to account for blindsight (Kentridge et al., 1997; Sahraie et al., 1997; Stoerig et al., 1998; Zeki and Ffytche, 1998; Ptito et al., 1999; Goebel et al., 2001; Morland et al., 2004). The existence of spared islands of V1 has been considered to be responsible for conscious visual perception (Celesia et al., 1991; Fendrich et al., 1992; Stoerig and Cowey, 1995) but many studies have later demonstrated that visual awareness could be present in the absence of a healthy V1 (Barbur et al., 1993; Ffytche et al., 1996; Zeki and Ffytche, 1998; Morland et al., 1999; Kleiser et al., 2001; Ffytche and Zeki, 2011).

#### **EXTRA-GENICULOSTRIATE PATHWAYS (OR SUBCORTICAL ROUTES)**

Another finding in Morland et al. (2004) is the existence of residual motion direction discrimination in patients in whom a lesion ruled out the hypothesis of spared cortical islands of V1. For these patients, the existence of other pathways bypassing V1 has been implicated to account for residual motion direction consistent with results of Celesia et al. (1991) and Ptito et al. (1999).

The existence of two sub-cortical pathways, one associated with the dorsal visual stream and the other with the ventral visual stream (Goodale and Milner, 1992), bypassing V1 and reaching directly the extrastriate cortex have been implicated in different blindsight classifications (see **Figure 1**). Retinal projections to the superior colliculi (SC) and the pulvinar bypass both V1 and dorsal lateral geniculate nucleus (LGN) to project to area MT+/V5, which is part of the dorsal visual stream (Bittar et al., 1999; Ptito et al., 1999; Schoenfeld et al., 2002). This colliculo-pulvinar pathway has been involved in accounting for residual motion discrimination and in the patients' ability to make accurate saccadic eye movements to localize stimuli. Superior colliculi allows to drive visually-guided behavior without awareness and would be involved in the reflexive orienting of attention (Rafal et al., 1988), the unconscious visual processing and attentional orienting (Kentridge et al., 1997). Action-blindsight may be underlied by the projections from the pulvinar to MT+/V5 whereas attention-blindsight may be underlied by the projections from the colliculo-pulvinar pathway to the posterior parietal cortex. However, both types seem to be closely related and may rely on the same neural pathways associated with the dorsal stream (Danckert and Rossetti, 2005).

Since collicular neurons do not have color opponency (Stoerig and Cowey, 1989, 1991; Ro and Rafal, 2006), a second pathway projecting directly from the LGN to V4 and MT+/V5 (especially demonstrated in monkeys in Sincich et al., 2004; for a review see Huxlin, 2008) has been implied to account for the residual color discrimination (Stoerig, 1987; Barbur et al., 1998; Bridge et al., 2010) and form discrimination in patients (Barbur et al., 1993; Stoerig and Cowey, 1997; Goebel et al., 2001). This route projecting to V4 and associated with the ventral visual stream has been referred to account for agnosopia (Zeki and Ffytche, 1998).

Another subcortical pathway has been implied to account for affective blindsight. Morris et al. (2001) have shown that the activation of the amygdala of patient GY for unseen faces correlated with activity in the SC and the pulvinar, suggesting the existence of a colliculo-pulvino-amygdalar pathway, whose existence has been demonstrated by Tamietto et al. (2012) using DTI tractography.

#### **INTERHEMISPHERIC CONNECTIONS (CALLOSAL AND NON CALLOSAL)**

Residual vision can not only be mediated by sub-cortical pathways but also by the reorganization in the ipsilesional and/or the contralesional hemisphere (Baseler et al., 1999; Bittar et al., 1999; Goebel et al., 2001; Bridge et al., 2008, 2010), allowing for the processing of visual information either ipsilaterally or bilaterally.

Neuroimaging studies using DTI have suggested that this redistribution of cerebral activations may rely on interhemispheric connections (Silvanto et al., 2007; Bridge et al., 2008). Specifically, Bridge et al. (2008) studied patient GY, five healthy controls and one age-matched male who underwent diffusion weighted-MRI and fMRI. GY exhibited similar to the controls bilateral tracts between LGN and MT+/V5. However, in GY, the ipsilateral pathways between LGN to V1 were bilaterally smaller than the pathway between LGN to extrastriate areas (see **Figure 2**, in blue and orange). The authors found two differences in GY's brain compared to controls: the presence of prominent bilateral tracts from the splenium to MT+/V5 (suggesting an increased cortico-cortical connectivity between these extrastriate areas through callosal connections) and the presence of a contralateral pathway between right LGN and left extrastriate areas.

These differences in connectivity pattern in GY have also been documented by Tamietto et al. (2012) using DTI tractography to reconstruct three pathways previously identified in healthy controls (see **Figure 2**). The colliculo-pulvinar pathway in GY's damaged hemisphere (in green **Figure 2**) was reduced in strength and failed to extend to frontal areas, especially to area 46 which has been implicated in conscious perception (Sahraie et al., 1997), consistent with the absence of conscious emotion perception in GY (Tamietto and de Gelder, 2010). Moreover, the connections from contralateral posterior areas in GY's intact hemisphere were strengthened, consistent with an increase of interhemispheric connections after V1 damage. Connections of the pulvino-amygdalar pathway in the damaged GY's hemisphere (in pink **Figure 2**) also extended more posteriorly to visual areas than to frontal regions compared to controls. The connections of the colliculo-pulvino-amygdalar pathway were strengthened in the damaged hemisphere of GY (in maroon **Figure 2**), consistent with the role of this pathway in affective blindsight, for which information did not depend directly from V1.

Bridge et al. (2010) studied patient SBR with bilateral damage to the gray matter of V1 sparing the adjacent white matter and surrounding visual areas. Using a "motion" task in fMRI contrasting moving dots with stationary dots, they found a bilateral activation of areas MT+/V5 despite no significant activation of V1. The tracts between LGN and V1 appear to show some degeneration while tracts between LGN and V5 did not differ from controls. The authors had previously reported similar findings in patient GY (Bridge et al., 2008) but in patient SBR, the very specific lesion suggested that ipsilateral connection between LGN and MT+/V5 may be particularly important for residual function.

Using fMRI, Perez et al. (2013) presented images of natural scenes filtered (in high and in low frequencies or non filtered) to right and left hemianopes who where asked to perform a detection and a categorization tasks. They showed a different pattern of reorganization depending on the lesion side. The right hemianopes (left occipital lesion) seemed to have a predominant intra-hemispheric reorganization whereas the left hemianopes (right occipital lesion) a predominant inter-hemispheric reorganization, suggesting that hemispheric specialization (visuospatial abilities for the right hemisphere and language abilities for the left hemisphere) could be present at this early level.

Neurostimulation studies using TMS have also highlighted the importance of interhemispheric connections in VFDs and their recovery. Silvanto et al. (2007) used TMS and reported that GY experienced visual sensation of phosphenes in his blind field only when bilateral stimulation were applied over MT+/V5. According to the authors, this conscious sensation can only be conveyed by the contribution from GY's intact hemisphere to explain why stimulation of the damaged hemisphere can reach awareness. These findings also suggested the presence of an increased connectivity via transcallosal connections, consistent with the tractography results obtained by Bridge et al. (2008). In a subsequent study in TMS performed on GY, Silvanto et al. (2009) showed that TMS applied over the area MT+/V5 in the damaged hemisphere modulated the appearance of phosphenes induced from V1 in the intact hemisphere (contrary to control subjects whose TMS over area MT+/V5 never influence the phosphenes induced from V1 in the other hemisphere). This finding was consistent with the abnormal functional connectivity between GY's both hemispheres. However in their previous study (Silvanto et al., 2007), GY experienced bilateral phosphenes, consistent with a role of interhemispheric connections between extrastriate areas in both hemispheres and with the increased anatomical connectivity documented in Bridge et al. (2008). In this study (Silvanto et al., 2009), GY never perceived bilateral phosphenes

because the combined stimulation of the left V5/MT+ and the right V1 did not induce phosphene in his blind field. Actually, GY had atrophy of callosal fibers in the forceps major which is part of the splenium (Silvanto et al., 2007), but it has been argued that non callosal pathway can mediate this interhemispheric transfer, although more slowly (Ffytche et al., 2000).

The existence of interhemispheric non callosal connections such as the intercollicular commissure and the ipsilateral and contralateral projections from the SC to various brain areas may explain why patients with bilateral occipital lesions (involving the splenium of the corpus callosum) or with a unilateral lesion impairing the splenium may nonetheless present blindsight (see **Figure 2**).

To summarize, residual vision is present after the first 6 months following brain damage. Blindsight refers to the ability of VFDs' patients to perform well in tasks involving eye movements, pointing, reaching, prehension, discrimination, identification, emotional processing, though they have no consciousness of their performance. Functional brain imaging shows that the retino-collicular pathway is more likely to account for blindsight than cortical islands in V1. Interhemispheric connections (either callosal between homologous visual areas, either non callosal connections via subcortical structures) may also play an important role in blindsight phenomena and variability, which is compatible with the differential effect of the lesion side on residual vision due to distinct underlying plasticity mechanisms.

#### **CORTICAL REORGANIZATION/PLASTICITY**

Some studies have demonstrated that the retinotopic organization of V1 could be preserved even if the visual cortex is damaged (Baseler et al., 1999; Ho et al., 2009; Reitsma et al., 2013). However, in some cases, patients can also present an atypical organization of their visual cortex after brain injury (Reitsma et al., 2013). In 27 patients with clear anatomical evidence of damage involving visual cortex and/or underlying white matter, Reitsma et al. (2013) presented three patients with an expanded ipsilateral field representation compared with healthy controls, whereas 22/27 patients had a typical retinotopic organization. For the authors, this atypical organization could rely on the unmasking of the interhemispheric suppression from the intact visual cortex due to the lesion that would in turn unmask retinogeniculate afferents representing the vertical meridian and ipsilateral visual field. They acknowledged other plausible mechanisms such as axonal sprouting and synaptogenesis in the deafferented visual cortex in association with the strengthening of long-range excitatory connections.

Some neuroimaging studies have also reported cases to document reorganization and plasticity in the visual system. Plasticity could be viewed as a recruitment of the contralesional hemisphere (and therefore interhemispheric connections), which has been demonstrated in the early stage of recovery (Raposo et al., 2011) and in more chronic stages (Nelles et al., 2002) where it could also mediate blindsight (Zeki and Ffytche, 1998; Ffytche et al., 2000; Silvanto et al., 2007; Bridge et al., 2008).

Plasticity could also be viewed as a recruitment of activity by nearby healthy cortex (Baseler et al., 1999), a disinhibition of preexisting long-range horizontal connections within V1, sprouting of new horizontal connections in V1, which have been demonstrated in animals studies (Darian-Smith and Gilbert, 1994, 1995; Das and Gilbert, 1995) or changes in the functional interactions between higher-level visual cortical areas and V1 (Huxlin, 2008). Processes of plasticity may have different time courses, which overlap, from synaptic gain in the short term to axonal sprouting and new circuits properties in the long term (Wandell and Smirnakis, 2009).

Goebel et al. (2001) used fMRI and retinotopic mapping in two patients with long-standing left VFDs history (FS and GY) to compare the responsiveness of dorsal and ventral stream areas after stimulation of both hemifields. They found that GY's ipsilesional extrastriate areas responded to stimulation to either hemifield. The authors proposed that these findings were consistent with a kind of plastic changes of the system compensating for the loss of V1 (which is normally the major source of MT+/V5 input) in GY.

Baseler et al. (1999) performed an fMRI study in GY. They found that the foveal stimulation in the lesioned occipital lobe exhibited normal retinotopic organization as GY's lesion spared the foveal representation. The stimulation of the blind VF exhibited a different topography of GY's extrastriate areas depending on the stimulus configuration (full or annular wedge), which now responded to positions restricted near the lower vertical meridian. Their findings suggested the involvement of subcortical projections to extrastriate cortex, transcallosal projections (consistent with the restricted activity around the lower vertical meridian) and residual inputs from V1 near the margin of the lesion. They assumed that because V2 neurons in GY's lesioned occipital lobe were deprived of their V1 input, they were colonized by other neurons in neighboring cortex. The colonization could be mediated by "strengthening or disinhibition of long-range connections or by the creation of new connections" (Das and Gilbert, 1995) through plastic reorganization.

Ioannides et al. (2012) studied patient GY and three healthy controls with MEG using a distributed source model to estimate the spatiotemporal properties of neural activity following the presentation of checkerboard pattern stimuli in different portion of the visual field. In control subjects, activity started in the first 100 ms in V1 and spread through dorsal and ventral streams in the next 100 ms towards extrastriate areas. In GY's damaged hemisphere no activity was detected before 130 ms. The first activity detected was in the ipsilesional extrastriate cortex (around the middle occipital gyrus, the middle temporal gyrus and the superior temporal sulcus) and spread towards higher level areas and backward to early retinotopic visual areas. Moreover, the back-propagated activity did not follow the retinotopic organization and did not have well-defined response peaks. Again, these findings in GY may be due to plastic reorganization following long-term lesion.

Dilks et al. (2007) reported the case of a patient with a stroke sparing V1 but affecting the right inferior optic radiations (which normally provide information to V1 from the upper field). The patient was blinded in the upper quadrant of the LVF but he also exhibited a distorted perception of the intact lower visual field (stimuli appeared vertically elongated). Six months after the onset of the stroke, the patient underwent behavioral testing and retinotopic mapping-fMRI, which revealed that this perceptual distortion was mirrored by a distorted visual field map in V1. They found that the regions normally dedicated to the representation of the upper LVF were now activated by lower LVF stimuli due to the upper quadrantanopia. Thus the regions of V1 representing the lower quadrant of the LVF were expanded, leading to an expanded representation of the left horizontal meridian and to the perceptual distortion. Huxlin (2008) proposed that the perceptual plasticity exhibited by the patient reported by Dilks et al. (2007) could include "dis-inhibition of pre-existing long-range horizontal connections within V1, sprouting of new connections in V1 or changes in the functional interactions between higher-level visual cortical areas and V1" (Huxlin, 2008).

Schoenfeld et al. (2002) have studied color changes and motion direction discrimination to target integrity of the ventral and the dorsal visual streams in a patient suffering from a left hemorrhagic PCA stroke, who underwent both fMRI and MEG recordings. They found activation following motion and colorchange stimuli in the hemianopic field in several extrastriate areas of the lesioned hemisphere. The MEG recordings provided evidence for activation first in V5 in the lesioned hemisphere with other extrastriate areas being activated later. In the intact hemisphere, V1/V2/V3 activity preceded V4/V5/V8 activity whereas in the lesioned hemisphere, motion and color stimuli activated first V4/V5/V8 regions. The authors proposed that the cortical reorganization after a V1 lesion may involve a change of connectivity between extrastriate areas (V4/V8 and V5), and a change in the dominant direction of flow of visual information between areas spared by the lesion, with V5 playing a key role in distributing subcortical signals to other extrastriate regions via feedback and feedforward connections already in place (Hupe et al., 1998). Indeed, the existence of recurrent loops between higher and early visual areas has been demonstrated in animals and in human studies (Hupe et al., 1998; Goebel et al., 2001; Schoenfeld et al., 2002; Silvanto et al., 2005). These loops have been shown to amplify and focus activity of neurons in lowerorder areas (Hupe et al., 1998) and are supposed to organize neuronal activity into stable resonant states and could be a neural correlate of conscious vision (e.g., Tononi and Edelman, 1998; Engel and Singer, 2001; Goebel et al., 2001; Silvanto et al., 2005).

However there are controversies about plasticity and cortical reorganization *per se*. Indeed, in most of these studies reported previously (except that of Ho et al., 2009, in which the patient was studied 1 year post stroke; that of Dilks et al., 2007, in which the patient suffered from a stroke 6 months before the study; and that of Reitsma et al., 2013, in which 2/3 patients with abnormal retinotopic organization had acquired VFDs during adulthood), all the patients had a long-standing history of VFDs, because the injury had occurred earlier in their lives, thus probably allowing for a better reorganization (Goebel et al., 2001; Haak et al., 2014). Alternative explanations have been proposed to account for cortical reorganization.

Rather than cortical reorganization or remapping, the findings may be more accurately accounted for by properties of neuronal receptive fields and modulatory feedback signals from extrastriate areas (Haak et al., 2012, 2014). For example, according to Haak et al. (2014) the term "reorganization" implies the presence of long term anatomical changes (Wandell and Smirnakis, 2009). Thus some cases of abnormal activity, such as that observed by Dilks et al. (2007), could be explained on the basis of intrinsic neuronal properties that surface only when the normal input signal is absent. When neurons are deprived from their original input, the feedback signals from the far periphery of the VF become visible as a distorsion of the visual field map and thus affect perception (Haak et al., 2014). Papanikolaou et al. (2014) used fMRI to measure the population receptive field (pRF) properties in area V1 in five patients with partial or complete quadrantanopia after a lesion in V1 or in optic radiations. They showed that in two patients, some pRF centers shifted their location near the border of the scotoma. Moreover, the pRF size in the spared V1 cortex of patients was increased in both the damaged and the healthy hemispheres, suggesting the recruitment of area nearby the lesion and a reorganization of the subcortical inputs from subcortical structures (LGN, pulvinar).

Different hypotheses have been advanced to account for cortical/subcortical reorganization and underlying plasticity mechanisms: nearby cortex recruitment, disinhibition of long-range connections, sprouting of new connections, interactions between V1 and higher level visual areas, or modulatory feedback signals. Far from being exclusive, we suggest that these different mechanisms are likely to co-occur and their effects to interact resulting in complex and variable solutions within the damaged brain. Visual field defects are associated with bad functional prognosis because they impair daily activities (driving, reading, social activities, leisure or work, fall risk, accidental risk and quality of life) (Trauzettel-Klosinski, 2011). To counteract VFDs' deleterious impact, different rehabilitation programs have been proposed.

#### **REORGANIZATION AFTER REHABILITATION**

Plasticity in the central nervous system after brain injury may play a key role in restitution and may be enhanced through training of visual functions (Pöppel et al., 1978; Zihl and von Cramon, 1985). Based on experiments in primates with VFD after brain injury (Cowey, 1967; Mohler and Wurtz, 1977), visual training aims at restoring or at compensating visual blind field. Though some authors have postulated that lost visual functions cannot be recovered (Horton, 2005b), others have emphasized the critical role of rehabilitation techniques such as visual retraining or field stimulation in the recovery of VFDs (Popelreuter, 1917; Preobrazenskaya cited in Luria, 1963; Zihl and von Cramon, 1982). Three main approaches may be considered for rehabilitation. The first employs compensatory techniques by using intact visual abilities to improve natural adaptation strategies (i.e., eye movements training, see Bolognini et al., 2005; Roth et al., 2009; Schuett et al., 2009). The second concerns optical aids, using relocation of the visual field with monocular or binocular prisms (Peli, 2000; Bowers et al., 2008; Ross et al., 2012). Finally restorative therapy (reported in this review) has been proposed, using training stimulation programs to increase the blind visual field directly.

#### **REHABILITATION BY BORDERZONE STIMULATION**

The first promising studies of training (Zihl and von Cramon, 1979, 1985; Zihl, 1981) proposed external stimulation between normal and impaired visual field (partially defective area called transition zone or borderzone) in patients with damage in the geniculostriatal visual system. Results highlighted an increase in seeing visual field size in patients with cerebral blindness. Kasten and Sabel (1995) proposed a standardized and automatic program called "Visual Restoration Therapy" (VRT, NovaVision©). Rehabilitation consisted of a home-based program to be performed on a computer (Kasten and Sabel, 1995). The region between the intact and damaged visual fields along the vertical meridian (borderzone) was also targeted by the therapy. Patient with VFD had to maintain a central fixation on the screen device and respond by pressing a key whenever a light stimulus appeared. These training exercises were performed twice daily for half an hour for 6 months.

However VRT remained criticized because of the variability of results reported in different studies. In most cases reported, visual expansion did not exceed 5% (Kasten and Sabel, 1995; Kasten et al., 1998b) and was less pronounced in others studies (from 1 ◦ to 6.7◦ ) (Pommerenke and Markowitsch, 1989; Kerkhoff et al., 1992, 1994) or even absent (Balliet et al., 1985). Some authors suggested that visual field improvements could be an artefact of eye movements (Horton, 2005a; Reinhard et al., 2005) and that controversial results could be related to (1) the perimeter techniques employed; (2) lesion localization; or (3) explanatory mechanisms (Pouget et al., 2012).

Behavioral VRT studies have described a visual field expansion after this rehabilitation and proposed an explanatory mechanism based on the reactivation of residual neuronal activity in the ischemic transition zone through the expansion of the receptive field of small spared neural structures (Kasten et al., 1998a). The borderzone has been characterized as an area of suboptimal visual perception corresponding to surviving neurons (Kasten et al., 1998a; Sabel et al., 2011). Thus Pleger et al. (2003) have shown an increase of BOLD signal in perilesional primary visual cortex after 6 months of rehabilitation in three patients with cortical blindness. Julkunen et al. (2003) explored rehabilitation at three times (before and after visual field training and after a follow up of 3 months) one patient with right homonymous upper quadrantanopia after left occipito-temporal lesion. Perimetry, subjective evaluation and visual evoked potential to right hemifield stimulation improved during the training period and were maintained at the follow-up (specifically with a 5–10◦ increase in visual field). Positive correlation between changes in rCBF and changes in perimetry results was found in the contralesional occipital area. Similarly, a cohort of six chronic right hemianopic patients with a left temporal or occipital lesion underwent fMRI before and 1 month after beginning VRT (Marshall et al., 2008). Results reported a modification of brain activity correlated with a relative improvement in response time for detection of the stimuli in the borderzone after therapy, in secondary and associated visual areas, right dorsolateral prefrontal cortex, bilateral anterior cingulate cortex, and bilateral basal ganglia. Positive correlation was also observed and the authors concluded that VRT could induce a modification of brain activity associated with a process of shift in spatial attention (from the seeing location toward the borderzone location). More recently, Raemaekers et al. (2011) have explored the properties of visual cortex (V1, V2, and V3) before and after VRT by using fMRI and perimetry in eight chronic patients with visual field defect. Vision restoration therapy induced visual field recovery as measured with perimetry (mean increase of 3.94◦ ) and fMRI results showed a shift of receptive fields to a higher eccentricity and some growth of receptive field size. However, no evidence for extensive representation of regained visual field was found.

Moreover, based on motor recovery after training rehabilitation combined with tDCS, (Plow et al., 2011, 2012) proposed a visual rehabilitation (VRT) associated with tDCS. They tested two patients in chronic phase with left occipital lesion and right VFD (Plow et al., 2011). One patient received VRT combined with active tDCS and the other received VRT combined with sham tDCS (no stimulation). The anode electrode was placed with the intention to stimulate the occipital cortex bilaterally. Objective perimetry pre- and post-treatment showed a greater expansion in the visual field border after rehabilitation in patients with VRT combined with active tDCS (expansion of 3.55◦ in the central visual field and a 4◦ shift inward from the periphery). These results suggested that the stimulation of the occipital cortex with tDCS during VRT promoted visual rehabilitation. Functional magnetic resonance imaging data were also obtained in this patient following rehabilitation and revealed activation in perilesional and in bilateral higher area (V2/V3 and MT+/V5), consistent with reactivation surviving visual area hypothesis. However, it seems difficult at this time to disentangle the contributory effect of VRT from tDCS. Plow et al. (2012) confirmed these previous results by proposing the same protocol to 12 patients and suggested that tDCS associated with VRT accelerate the recovery in accuracy detection (Halko et al., 2011; Plow et al., 2012). The exact underlying mechanism of tDCS in humans is not known. It is supposed that tDCS may enhance VRT effects by modulating excitability of surviving visual networks including perilesional area but also bilateral higher visual areas (for a review see Valero-Cabré et al., 2011).

#### **REHABILITATION BY BLINDSIGHT**

The ability to perform visual discrimination in the absence of awareness (Weiskrantz et al., 1974) opened up new horizons for neuro-visual rehabilitation (Ro and Rafal, 2006). As described in Section Residual vision in the chronic phase, blindsight may suggest the existence of a residual visual treatment process after striate pathway injury whereby visual information may travel through LGN directly to extrastriate cortical areas (Sincich et al., 2004; Bowers et al., 2008; Bridge et al., 2010). Thus Vanni et al. (2001) trained one patient (MR) with a right posterior medial cerebral infarct and left hemianopia to detect flickering luminance patterns (disk and letter). Magneto encephalography results showed a right attenuated transient occipital response and a prominent response in the right superior temporal cortex. The authors concluded that the input of the superior temporal cortex might come through the SC and pulvinar and compensate for the impaired input of the primary visual cortex. Consistent with these results, Raninen et al. (2007) reported the case of two patients (KS and IT) with a left occipital lesion, who were trained with a detection task of a flickering light stimulus and a letters identification task. Neuromagnetic responses (at 1–2 months intervals) of patient KS showed the strongest response in the ipsilesional posterior superior temporal area. In contrast, the strongest response of the other patient (IT) was in the controlesional occipital area. Henriksson et al. (2007) showed with an fMRI mapping that both visual hemifields were represented in IT's intact hemisphere and more specifically in the MT+/V5 area, in a region around the superior temporal sulcus and in visual areas V1, V2, V3 and V3a. Implication of interhemispheric connections appeared required. However, the minor activation observed in the left hemisphere excluded a pathway through callosal connections as if a disconnection of the left occipital regions from visual processing occurred. Inter-commissural connection of the SC, connecting left hemisphere activity to the right extrastriate visual areas via the pulvinar seemed to be a more plausible explanation.

To summarize, depending on whether theoretical and physiological frameworks have pointed out the role of either cortical or subcortical pathways in residual vision, rehabilitation techniques in VFDs have privileged either cortical borderzone stimulation or blindsight training. In both cases, significant improvements in visual performance have been observed. However, the underlying mechanisms are not well understood, which could be disentangled by more systematic functional brain imaging studies testing more patients.

#### **CONCLUSION**

Visual field defects following post-geniculate lesion, among which 70–75% are HH and 29% quadrantanopia, are common neurological disorders after stroke or traumatic brain injury, with high deleterious impact on activities of daily life: walking, driving, reading, etc. Spontaneous recovery only occurs within the 6 months following brain damage, and most of the recovery is done after 1 month. After 6 months, almost no spontaneous recovery may occur and rehabilitation techniques have been developed to improve residual vision. Depending on theoretical frameworks and physiological considerations underlying visual capabilities, strategies have alternatively given importance to either borderzone stimulation (the stimulation through behavioral training or transcranial stimulation of cortical areas in the neighborhood of damaged primary visual cortex) or blindsight (the stimulation of spared subcortical routes). Visual improvements have been documented in both cases. However, from a phenomenological point of view, there is a great variability regarding the success of the rehabilitation according to the patients. Some patients seem to "see" in their blind field and feel more confident in their perception (Chokron et al., 2008), leading to a better quality of life. Subcortical processes, which have been involved in mediating blindsight, have been shown to be non conscious but they might interact with cortical processes via an integration of their activity with activity in cortical structures, which in turn exert a feedback to subcortical structures (Tamietto and de Gelder, 2010). Nonetheless, subcortical processes seem to be distinct from cortical processes in terms of sensory threshold, time-scale and in terms of main connections to cortical areas implicated in conscious perception.

Most of rehabilitation techniques are specific to the type of stimuli used in training and therefore non-transferable to other stimuli present in real life. Moreover, they require repetitive training over an extended period (Ajina and Kennard, 2012). As example, Rowe et al. (2013) have shown that in 479 patients with visual loss treated with different options (visual search training, visual awareness, typoscopes, substitutive prisms, low vision aids, refraction, and occlusive patches), only 7.5% had full recovery, 39% had improvement and 52% did not recover at follow-up.

Underlying physiological mechanisms of spontaneous recovery and/or rehabilitation interventions as well as their interactions are not well understood in humans. In this context, functional brain imaging examining gray and white matter in brain-damaged patients is the only objective tool to understand further plasticity and compensatory mechanisms of visual loss, recovery and rehabilitation. Learning plays a crucial role in plasticity because practice has a role in facilitating recovery and reorganization (Levin, 2006). For example, studies using perceptual learning in control subjects have demonstrated changes in receptive field properties within early visual cortex and increases in activation (Bao et al., 2010; Frank et al., 2014). Brain imaging evidence of perceptual learning has shown cortical and white matter changes, which took place quickly and efficiently. Gray matter changes may rely on dendritic spine growth and synapse turnover (Barnes and Finnerty, 2010; May, 2011) whereas white matter changes may be based on larger-scaled axonal remodelling and increased myelination (Johansen-Berg et al., 2012; Ditye et al., 2013; Lövdén et al., 2013). As cortical reorganization or plasticity might rely on this type of changes in the functional architecture and RF properties in visual areas (Gilbert and Li, 2012), promising studies should perhaps use perceptual learning in patients (see Huxlin et al., 2009, who have trained seven patients with HH to discriminate complex motion direction in their intact and blind field). The success or the failure may depend on the pattern of cortical damage and the involvement of damage to subcortical and interhemispheric connections (Reitsma et al., 2013).

#### **ACKNOWLEDGMENTS**

Marika Urbanski was supported by the French Agence Nationale de la Recherche (project CAFO-RPFC, No: ANR-09-RPDOC-004-01).

#### **REFERENCES**


primary visual cortex. *Neuroimage* 63, 1464–1477. doi: 10.1016/j.neuroimage. 2012.07.058


visual cortex complements perimetry in patients with homonymous visual field defects. *Proc. Natl. Acad. Sci. U S A* 111, E1656–E1665. doi: 10.1073/pnas. 1317074111


field defects? A fundus controlled study. *Br. J. Ophthalmol.* 89, 30–35. doi: 10. 1136/bjo.2003.040543


Zihl, J., von Cramon, D., Mai, N., and Schmid, C. (1991). Disturbance of movement vision after bilateral posterior brain damage. Further evidence and follow up observations. *Brain* 114, 2235–2252. doi: 10.1093/brain/114.5.2235

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

#### *Received: 08 April 2014; accepted: 03 September 2014; published online: 30 September 2014*.

*Citation: Urbanski M, Coubard OA and Bourlon C (2014) Visualizing the blind brain: brain imaging of visual field defects from early recovery to rehabilitation techniques. Front. Integr. Neurosci. 8:74. doi: 10.3389/fnint.2014.00074*

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

*Copyright © 2014 Urbanski, Coubard and Bourlon. 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*.

## Functional activity within the frontal eye fields, posterior parietal cortex, and cerebellar vermis significantly correlates to symmetrical vergence peak velocity: an ROI-based, fMRI study of vergence training

#### *Tara L. Alvarez\*, Raj Jaswal , Suril Gohel and Bharat B. Biswal*

*Department of Biomedical Engineering, New Jersey Institute of Technology, Newark, NJ, USA*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

#### *Reviewed by:*

*Stephanie Jainta, Leibniz Research Centre for Working Environment and Human Factors, Germany Kai Markus Schreiber, Max-Planck Institute for Neurological Research, Germany*

#### *\*Correspondence:*

*Tara L. Alvarez, Department of Biomedical Engineering, New Jersey Institute of Technology, 323 Martin Luther King Boulevard, University Heights, Newark, NJ 07102, USA e-mail: tara.l.alvarez@njit.edu*

Convergence insufficiency (CI) is a prevalent binocular vision disorder with symptoms that include double/blurred vision, eyestrain, and headaches when engaged in reading or other near work. Randomized clinical trials support that Office-Based Vergence and Accommodative Therapy with home reinforcement leads to a sustained reduction in patient symptoms. However, the underlying neurophysiological basis for treatment is unknown. Functional activity and vergence eye movements were quantified from seven binocularly normal controls (BNC) and four CI patients before and after 18 h of vergence training. An fMRI conventional block design of sustained fixation vs. vergence eye movements stimulated activity in the frontal eye fields (FEF), the posterior parietal cortex (PPC), and the cerebellar vermis (CV). Comparing the CI patients' baseline measurements to the post-vergence training data sets with a paired *t*-test revealed the following: (1) the percent change in the BOLD signal in the FEF, PPC, and CV significantly increased (*p* < 0.02), (2) the peak velocity from 4◦ symmetrical convergence step responses increased (*p* < 0.01) and (3) patient symptoms assessed using the CI Symptom Survey (CISS) improved (*p* < 0.05). CI patient measurements after vergence training were more similar to levels observed within BNC. A regression analysis revealed the peak velocity from BNC and CI subjects before and after vergence training was significantly correlated to the percent BOLD signal change within the FEF, PPC, and CV (*r* = 0.6; *p* < 0.05). Results have clinical implications for understanding the behavioral and neurophysiological changes after vergence training in patients with CI, which may lead to the sustained reduction in visual symptoms.

**Keywords: vergence, frontal eye fields, posterior parietal cortex, cerebellar vermis, vision therapy, vergence training, convergence insufficiency, Convergence Insufficiency Symptom Survey**

#### **INTRODUCTION**

Throughout the day, the visual system mediates vergence eye movements to acquire visual information located at different spatial depths from the retina using the medial and lateral recti muscles. The inward rotation of the eyes is known as convergence. Convergence insufficiency (CI), a prevalent binocular vision disorder in adults (Porcar and Martinez-Palomera, 1997) and children (Rouse et al., 1999), is typically characterized by reduced fusional convergence amplitude, receded near point of convergence (NPC), greater exophoria at near than at far and visual symptoms. The visual symptoms commonly experienced by CI patients include the following: double/blurred vision, eyestrain, and headaches when engaged in reading or other near work, thus interfering with activities of daily living (Scheiman et al., 2011; Lee et al., 2014). Exophoria is the outward deviation of the eye when binocular fusion is disrupted (i.e., one eye is occluded while the other eye is fixating on a target). Scheiman and others hypothesize that CI patients are symptomatic because of the excessive convergence needed to compensate for a larger exophoria at near (40 cm) compared to far (6 m) (Cooper et al., 1998; Scheiman and Wick, 2008; Scheiman et al., 2011; Cooper and Jamal, 2012). Double vision is a common symptom of CI patients which could be explained by a reduced speed of convergence eye movements. Thus, their convergence responses require more time to attain fusion leading to double vision. Several investigations also report that vergence peak velocities elicited from abrupt changes in disparity are reduced in those with CI compared to binocularly normal controls (BNC) and improves to levels more similar to controls after repetitive vergence training (Van Leeuwen et al., 1999; Alvarez et al., 2010b; Thiagarajan et al., 2011; Alvarez and Kim, 2013).

Randomized clinical trials report that Office-Based Vergence and Accommodative Therapy with home reinforcement (OBVAT) reduces the visual symptoms in CI patients (Scheiman et al., 2009, 2011). The reduction of visual symptoms is sustained 1 year posttherapy in most subjects (CITT, 2009). Clinicians commonly prescribe vergence training (also known as vision therapy, vision rehabilitation, or orthoptic exercises) to reduce visual symptoms (Scheiman et al., 2011; Cooper and Jamal, 2012). However, the true neural mechanism by which vergence training leads to a reduction in visual symptoms is currently unknown (Scheiman et al., 2011).

There are many single cell and lesion studies on primates, as well as human case reports and fMRI studies that form the basis of our understanding of the neural substrates used to mediate a convergence response. Studies on primates report that the convergence circuit does involve cortical areas within the posterior parietal cortex (PPC) where cells have been identified that have a preferred direction for targets closer or farther away (Gnadt and Mays, 1995; Sakata et al., 1999; Taira et al., 2000; Genovesio and Ferraina, 2004). Single cell recordings from primates reveal a distinct area within the frontal eye fields (FEF) that is allocated for step convergence responses and is located more anterior compared to the cells responsible for generating saccadic responses (Gamlin and Yoon, 2000). Our team reports differentiation within FEF between saccadic and step convergence responses using fMRI (Alvarez et al., 2010a; Alkan et al., 2011a). Other investigators support that smooth convergence tracking is also encoded within FEF using single cell recording from primates (Kurkin et al., 2003; Akao et al., 2005) and using fMRI (Petit and Haxby, 1999). Many studies support the cerebellum is active during convergence responses (Miles et al., 1980; Gamlin and Clarke, 1995; Zhang and Gamlin, 1998; Gamlin, 2002; Takagi et al., 2003; Nitta et al., 2008a,b). In addition, lesions to the cerebellar vermis (CV) VI/VII in primates (Takagi et al., 2003) and humans (Sander et al., 2009) result in a convergence dysfunction. Many single cell studies also show neural activity within the midbrain (Mays et al., 1986; Zhang et al., 1991, 1992). In summary, the PPC, FEF, cerebellum, and midbrain are utilized to generate a convergence response.

Functional imaging investigations non-invasively quantify the metabolic demand generated through an experimental task by studying the blood oxygen level dependent (BOLD) response. For this study, we will measure the paramagnetic properties of blood during sustained fixation compared to active vergence eye tracking. The portions of the brain that are more metabolically active during vergence eye tracking will be assumed to be responsible for generating a vergence response. Functional imaging signals from the cerebrum and cerebellum are easier to obtain compared to signals within the brainstem. The brainstem is more susceptible to breathing and swallowing motion artifacts compared the cerebrum and cerebellum and hence is beyond the scope of this present study. Vergence training leads to a sustained reduction in symptoms suggesting neuroplasticity. Based upon the aforementioned studies about the neurophysiology of the vergence circuit, this investigation analyzed the PPC, FEF, and CV before and after vergence training in patients with CI compared to BNC. BNC did not participate in vergence training since they did not have visual symptoms. This research takes a critical step in understanding the neural basis of how vergence training leads to a sustained reduction of vision symptoms in patients with CI. Such knowledge may ultimately lead to new vergence training protocols to further improve vision function.

This study investigated convergence responses and functional activity of the vergence neural substrates before and after repetitive vergence training in symptomatic CI patients compared to BNC subjects. The following hypotheses were tested. First, reduced convergence peak velocity elicited from symmetrical convergence step stimuli and reduced functional activity within the FEF, PPC, and CV would be observed in those with CI before repetitive vergence training compared to BNC subjects. Second, after repetitive vergence training, the peak velocity of symmetrical convergence step responses would increase along with the percent change in functional activity of the vergence neural substrates from CI patients compared to their baseline measurements. Third, the peak velocity of convergence responses would significantly correlate to the functional activity of the FEF, PPC, and CV neural substrates quantified as the BOLD percent signal change.

## **MATERIALS AND METHODS**

## **SUBJECTS AND VISION PARAMETERS**

Four CI (four females) and seven BNC subjects (three females) participated in this study. CI was diagnosed by an optometrist using methods described in our prior study, which included a receded NPC and reduced positive fusional amplitudes (Alvarez et al., 2010b). The diagnosis criteria comply with conventional clinical methods (Cooper et al., 1998). At the beginning of the study, the CI subjects had the following parameters denoted as the average with one standard deviation: near point convergence of 13.4 ± 5.6 cm, positive [base out (BO)] vergence amplitude of 14 ± 4.5-, and a dissociated near (measured at 40 cm from midline) phoria of 9.0 ± 1.4 exophoria. NPC was assessed by measuring the distance from the orbit to the location where a high acuity target was perceived as diplopic along the subject's midline (Von Noorden and Campos, 2002). Stereopsis was assessed by the Randot Stereopsis Test (Bernell Corp., South Bend, IN, USA). Normal binocular vision was defined as having a NPC of less than 8 cm. The inclusion criteria were as follows: normal stereopsis, no ocular surgeries and corrected to normal acuity. In addition, if an eye spectacle prescription was required then subjects who required a prescription greater than 2D or less than -3D were excluded to reduce potential confounding variables. All subjects had a stereopsis of ≤50 s of arc. Three of the CI and five of the BNC did not need spectacles to read clearly at near. All subjects had no history of brain disorders and were between the ages of 18 and 35 years. All subjects signed written informed consent forms approved by the University of Medicine and Dentistry of New Jersey (UMDNJ) and New Jersey Institute of Technology (NJIT) Institution Review Boards (IRB) in accordance with the Declaration of Helsinki.

#### **CONVERGENCE INSUFFICIENCY SUBJECT SYMPTOMS**

Symptoms were quantified using the Convergence Insufficiency Symptom Survey (CISS), which contains 15 questions (CITT, 2008). All questions were in regard to the subject's ability to read or perform near work. Each symptom was scored between zero and four where zero represents the symptom never occurs and four represents the symptom occurs very often. A prior investigation compared visual symptoms to the clinical diagnosis of CI defined as an exophoria at near at least 4 greater than at far, failure of Sheard's criteria or a minimum normal positive fusional vergence (break < 15-), and a receded (≥6 cm) NPC (Rouse et al., 2004). The responses of CISS were summed where a score of 21 or higher had a sensitivity of 98% and specificity of 87% using the aforementioned diagnostic criteria of CI in young adults between the ages of 18 and 35 years of age (Rouse et al., 2004). Hence, the CISS symptom survey was used within this present investigation to assess visual symptoms of the CI and BNC subjects who participated in this study.

#### **OVERALL EXPERIMENTAL PROTOCOL GOALS**

The CI subjects participated in 18 h of vergence training as described below. The CI data were compared to BNC data where BNC subjects did not participate in vergence training since they did not have visual symptoms. The primary measurements compared were the peak velocity from convergence step responses and the percent change in the BOLD signal within the FEF, PPC, and CV. Secondary measurements included the NPC, positive fusional range, near dissociated phoria and the CISS score. A group level analysis compared the following: (1) the BNC data to the CI baseline data and (2) the CI pre and post-vergence training measurements.

#### **VERGENCE TRAINING PROTOCOL FOR CONVERGENCE INSUFFICIENCY SUBJECTS**

Repetitive vergence training was designed to provoke changes in the neural substrates (FEF, PPC, and CV) that stimulate vergence ocular motor responses. The CI subjects participated in a total of 18 h of vergence training, 6 h at home, and 12 h in the laboratory. Home training was monitored by having each CI subject record the amount of time spent on training and entailed two 10-min sessions (morning and evening) 3 days per week for 6 weeks. Laboratory training was composed of 1-h sessions, twice per week for 6 weeks. Within a single day, a subject participated in either laboratory or home training but not both.

For the laboratory step training, 2, 4, and 6◦ disparity convergence steps and 4◦ disparity divergence steps within the range of a 2◦ vergence angle to 8◦ vergence angle were presented after a randomized delay of 0.5–2.0 s. The randomized delay reduces anticipatory cues that are known to alter the latency and peak velocity of vergence responses (Alvarez et al., 2002, 2005, 2010a; Kumar et al., 2002a,b). The ramp training consisted of 1 and 2◦/s ramps starting at an initial vergence angle of 2◦ producing a convergence ramp response to the vergence angle of 8◦ and then stimulating a divergence ramp response to a vergence angle of 2◦. The laboratory and home training consisted of step and ramp stimuli similar to methods used clinically (Griffin, 1988; Scheiman and Wick, 2008).

#### **EYE MOVEMENT ACQUISITION AND ANALYSIS**

Eye movements were recorded using a Skalar Iris (model 6500, Delft, Netherlands) infrared (λ = 950 nm) limbus tracking system. The manufacturer reports that the linear range of the system was ± 25◦ where all responses of this study were within the linear range of the device. Prior research confirms a high degree of linearity, within 3% between 5◦ horizontally for this system (Horng et al., 1998). A 12-bit acquisition hardware card (National Instruments 6024 E series, Austin, TX, USA) digitized the individual left-eye and right-eye movements with a sampling rate of 200 Hz. The visual stimuli utilized green light emitting diodes (LEDs) (Stanley model MU07 part 5101, London, OH, USA) 2 mm wide by 25 mm in height with a wavelength of 555 nm. Subjects were situated in a head and chin rest assembly to reduce any influence from the vestibular system (Khojasteh and Galiana, 2007). The subject initiated each experimental trial by pressing a button, which allowed the subject to blink between experimental trials. Potential subject fatigue was also reduced by allowing the subject to initiate the experimental trial (Yuan and Semmlow, 2000).

A custom MATLAB™ (Waltham, MA, USA) program was used for all eye movement analyses. Left-eye and right-eye movement data were converted from voltage values into degrees using the individual calibration data. Eye movements were calibrated using 2, 4, 6, and 8◦ vergence angles. Vergence was calculated as the difference between the right-eye and the left-eye position to yield a net vergence response. Convergence responses were plotted as positive. Blinks were easily identified based upon manual inspection of the left-eye and right-eye movement response. Responses with blinks at any point during the movement were omitted (up to 2.1% of the data depending upon the subject). Only convergence responses were analyzed since convergence responses were reported to have reduced peak velocities in CI subjects compared to BNC (Alvarez et al., 2010b).

Peak velocity generated from a 4◦ symmetrical convergence step stimulus was a primary measurement within this study. Velocity was computed by taking the derivative of the position response using a two-point central difference algorithm (Bahill et al., 1982). Each individual left-eye and right-eye convergence movement response was manually inspected for the presence of a saccade, which was easily identified because saccade velocities are an order of magnitude greater than vergence. A phase plot (vergence velocity as a function of vergence amplitude) for the left-eye and the right-eye movement was used to determine whether the saccades obscured the peak velocity of the vergence response to the symmetrical stimulus. Only when saccades obstructed the convergence peak velocity was the response omitted from the peak velocity analysis, which occurred in less than 10% of the responses depending on the subject as shown in our prior investigations (Alvarez et al., 1998; Lee et al., 2008; Kim and Alvarez, 2012). The peak velocity of the combined vergence response was quantified as the maximum value.

#### **IMAGING INSTRUMENTATION AND ACQUISITION**

A 3-Tesla Siemens Allegra Magnetron MRI Scanner with a standard single channel head coil (Erlangen, Germany) was used to perform the fMRI scans during the experimental tasks. The fMRI imaging parameters used during the acquisition included: repetition (*TR*) = 2000 ms, echo time (*TE*) = 27 ms, matrix size = 64 × 64, field of view (FOV) = 220 mm, and flip angle = 90◦. A total of 32 slices were collected (axial orientation) with a slice thickness of 5 mm. The voxel resolution was 3.4 × 3.4 × 5.0 mm. High resolution anatomical volumes acquired using a magnetization-prepared rapid acquisition with gradient echo (MPRAGE) were collected after all functional tasks. The MPRAGE imaging parameters included the following attributes: *TR* = 7.2 ms, *TE* = 4.38 ms, *T*1 = 900 ms, flip angle = 8◦, matrix size = 256 × 256 with a total of 80 acquired slices. The voxel resolution was 0.9 × 0.9 × 2.0 mm. Subjects were instructed to limit head motion and foam padding was used to facilitate the restriction of physical movement. All subjects were positioned supine on the gantry of the scanner with their heads situated along the midline of the coil.

#### **FUNCTIONAL MRI VISUAL STIMULUS EXPERIMENTAL DESIGN**

The visual stimulus (see **Figure 1A**) was carefully aligned with the subject's midline to stimulate symmetrical vergence eye movements to test the hypotheses of this study. Subjects could see the targets with the aid of a mirror. Visual stimuli were a set of nonferrous LED targets that formed a line 5 cm in height by 2 mm in width secured with polyvinyl chloride (PVC) tubing. The LED stimulus targets were located at the following three full vergence angle demands: 2, 3, and 4◦. The target positions were chosen because smaller vergence movements have been shown to elicit fewer saccadic responses compared to larger vergence movements (Coubard and Kapoula, 2008; Semmlow et al., 2008, 2009; Chen et al., 2010).

The experiment utilized a conventional block design of sustained fixation for the "off" stimulus compared to vergence eye movements for the "on" stimulus as shown in **Figure 1B**. Anticipatory cues are known to decrease the latency and increase peak velocity of convergence responses (Alvarez et al., 2002, 2005, 2010a; Kumar et al., 2002a,b). Hence, to reduce anticipatory or predictive cues, this experiment utilized a series of vergence eye movements where each target was illuminated for a random duration between 3 and 5 s. LED targets were never simultaneously illuminated. The eye movement sequence illuminated one of the following three stimuli: the near (4◦), middle (3◦), or far (2◦) target where the subject could not anticipate when the next target would illuminate or which of the three targets would be illuminated. Each phase lasted 20 s and was repeated for 3.5 cycles. Hence, the total experiment time was 2 min 20 s. The experiment was repeated three times per subject.

#### **IMAGING ANALYSIS**

#### *Image preprocessing*

The AFNI (Cox, 1996) and FSL (Jenkinson and Smith, 2001; Jenkinson et al., 2002) software suites were used to process and analyze the raw data retrieved from the MRI scanner. The first five images of each trial dataset were removed to mitigate the effect of transient scanner artifact (Biswal et al., 2010).

The AFNI motion correction utilizes the application of a sixparameter, rigid-body, least-squares alignment routine. Three parameters calculate the amount (mm) of movement within each plane (anterior to posterior, right to left, and inferior to superior) and three parameters calculate the amount of rotation (◦) between planes (yaw, pitch, and roll). These six motion regressors are used within the linear regression model to minimize motion effects of the acquired BOLD signal. A detailed motion analysis of all subjects using a frame displacement method which calculates the absolute value of movement was conducted (Satterthwaite et al., 2013). The average frame displacements with one standard deviation for the degree of rotation were 0.18 ± 0.07◦, 0.16 ± 0.09◦, and 0.20 ± 0.08◦ for yaw, pitch, and roll, respectively. The average frame displacement analyzing all subjects within each plane, with one standard deviation, were 0.36 ± 0.13 mm, 0.42 ± 0.11 mm, and 0.37 ± 0.08 mm for the anterior to posterior, left to right, and inferior to superior planes, respectively. No

significant differences in motion artifacts were observed between the post and pre-vergence training data sets of the CI patients assessed using a paired *t*-test (*p* > 0.9). No significant difference in head motion was observed between the two groups (CI compared to BNC) (*p* > 0.9). Hence, head motion was not considered problematic within this dataset.

The CompCor data-driven method was used to further reduce effects of noises in the BOLD signal, as described below (Behzadi et al., 2007). FSL's BET (Brain Extraction Tool) (Smith, 2002) function removed non-brain tissue from the anatomical image dataset. FSLs FAST (FMRIBs Automated Segmentation Tool) (Zhang et al., 2001) stratifies the skull-stripped anatomical dataset into three different segments. The whole brain probability maps of CSF, WM, or gray matter (GM) were derived. The segmented anatomical CSF and WM probability images were transformed into functional space using FSL's FLIRT function (Beckmann and Smith, 2004, 2005). To create CSF and WM regressors, all voxels of the CSF and WM probability images were first thresholded using levels of 99 and 97% probability, respectively. Time-series from all the voxels surviving the threshold were extracted. The probability levels of this study are more conservative compared to those used previously, which used a threshold level of 80% (Biswal et al., 2010). Then, the first five principle components relating to CSF and WM timeseries were calculated. FSL's FEAT command was used to perform the voxel-wise linear regression analysis on all datasets using the 16 aforementioned regressors (six motion parameters, five principle components of CSF, and five principle components of WM). The residuals of the regressed datasets (removal of the 16 artifacts) were then filtered in AFNI using a band pass filter [full width at half maximum (FWHM) Gaussian filter with cut off frequencies of 0.01 and 0.15 Hz]. The band pass filter was used to remove DC offset and high frequency signals that were probably not neuro-physiological in nature. Following band-pass filtering, a general linear model (GLM) analysis was performed to derive functionally active regions during the task.

#### *General linear model*

A GLM using a reference time series representation of the block design experimental stimulus convolved with the hemodynamic response function (HRF) was used. Correlation maps were created using a threshold of *r* ≥ 0.4 (*p* < 0.05) to show active brain regions. Mask identification was facilitated by observing the active brain regions coupled with the anatomical locations described above for the FEF, PPC, and CV. Broca's region was the control region of interest (ROI) and was identified strictly using anatomical markers. Since the datasets were not transformed into a standardized space such as the Montreal Neurological Institute (MNI) space, some variance was also observed for the mask of Broca's region. Broca's region served as a control ROI (unrelated to the hypotheses of this study). Language was not manipulated within the experimental protocol. Prior investigations show Broca's region was stimulated during experiments that study language (Geschwind, 1970; Kim et al., 1997) but is not stimulated within vergence eye movement experiments (Alkan et al., 2011a,b).

## *Cortical and subcortical regions of interest (ROIs) within the fMRI experiment*

The ROIs were defined using anatomical markers coupled with a model-driven method to identify functional activity near the anatomical markers. Neurophysiology studies on primates support the following ROIs are involved in vergence eye movements: FEF, PPC, and CV (Gamlin et al., 1996; Gamlin and Yoon, 2000). This experiment sought to stimulate the cortical and cerebellar regions required to mediate vergence eye movements.

The following ROIs were drawn in native space using anatomical markers and functional activity derived using a GLM: FEF, PPC, and CV. The bilateral FEFs were defined as the area within the intersection between the precentral sulcus and superior frontal sulcus. The PPC was around the intraparietal sulcus as shown in **Figure 2**. The CV regions VI and VII were defined on the mid-sagittal plane. Broca's region served as a control ROI because it was not stimulated in prior fMRI vergence studies (Alvarez et al., 2010a; Alkan et al., 2011a,b). The mask for Broca's region was created using only anatomical markers that are defined near the inferior frontal gyrus anterior to the motor strip as shown in **Figure 2**. **Figure 2** depicts the ROIs within a series of axial slices. The average with one standard deviation for the masks studied (measured in mm3) were 880 <sup>±</sup> 118, 901 ± 126, 1430 ± 275, 1194 ± 382, 1006 ± 131, 499 ± 58, 541 ± 47 for the FEF-L, FEF-R, PPC-L, PPC-R, CV, Broca-L, and Broca-R, respectively. The average and one standard deviation of each ROI are shown in **Figure 2** using an MNI template. As **Figure 2** shows, none of the masks overlap to avoid any partial volume effects. The centroid of the mask listed as left (positive) or right (negative), anterior (positive) or posterior (negative), and superior (positive) or inferior (negative) are (30, 12, 42), (−30, 12, 42), (52, 10, 12), (−52, 10, 12), (−2, −74, −28), (26, −54, 48), and (−26, −54, and 48) for the FEF-L, FEF-R, PPC-L, PPC-R, CV, Broca-L, and Broca-R, respectively.

#### *Analysis of percent change of BOLD signal to quantify functional activity*

All data were kept in native space (i.e., data were not transformed into Talairach and Tournoux or MNI space) to avoid any warping artifacts. The time series located within the vicinity of the anatomical markers, which had a Pearson correlation coefficient of *r* ≥ 0.4 (*p* < 0.05) with the hemodynamic model described above, were pooled. While the percent signal change will be threshold dependent, this study is longitudinal comparing the data after vergence training to the baseline measurements before vergence training for the CI subjects. Since the threshold used is the same for both pre and post-vergence training analysis, we assume that any potential differences observed within the data sets are mostly due to vergence training. The BOLD percent signal change for each ROI per subject comparing elevated activation observed during the vergence task to the baseline of sustained fixation was computed from the time series. The individual-subject percent signal change values were pooled to conduct the group-level statistics described below.

**FIGURE 2 | Series of axial images showing the average masks with one standard deviation used within the analysis.** The cerebellar vermis (yellow), Broca-Right (red), Broca-Left (dark blue), Posterior Parietal Cortex–Right (black), Posterior Parietal Cortex–Left

(pink), Frontal Eye Field-Right (green), and Frontal Eye Field–Left (medium blue) are shown. The masks did not overlap. Broca's Region served as a control ROI to study the variability within a non-stimulated ROI.

### **STATISTICAL ANALYSES**

The subject data were stratified into the following three groups: BNC, CI subjects before vergence training, and CI subjects after vergence training. An unpaired *t*-test was used to determine whether significant differences were observed between BNC and CI subjects before vergence training when analyzing (1) the peak velocity of convergence responses stimulated from 4◦ symmetrical convergence step stimuli and (2) the percent signal change of the BOLD fMRI signal within an ROI. A paired *t*-test determined whether the CI subjects exhibited significant changes comparing the pre and post vergence training measurements for the following parameters: (1) peak velocity of convergence responses stimulated from 4◦ symmetrical convergence steps, (2) percent signal change of the fMRI BOLD signal within an ROI, (3) CISS score, (4) NPC, (5) positive vergence amplitude ranges, and (6) near dissociated phoria. A linear regression analysis was conducted between the peak velocity of convergence responses stimulated from 4◦ symmetrical convergence steps and the BOLD percent signal change for the following ROIs: FEF, PPC, CV and Broca's region. Statistics were calculated using NCSS2004 (Kaysville, UT, USA). Significance was defined as a *p*-value < 0.05. Bonferroni correction for multiple parameters was not applied because of the limited number of subjects within the study. Figures were generated using MATLAB (Mathworks, MA).

#### **RESULTS**

#### **CONVERGENCE EYE MOVEMENTS FROM SYMMETRICAL CONVERGENCE STEP STIMULI**

Peak convergence velocity was one of the primary measurements within this study. **Figure 3** plots multiple eye movements. Each colored line in **Figure 3** is a convergence eye movement response evoked from a symmetrical 4◦ convergence step stimulus. **Figure 3A** is convergence responses from a BNC. The BNC subject attains fusion of the new target within the first half second. **Figures 3B,C** are from the same CI subject before and after vergence training, respectively. The CI subject before vergence training has more variability between responses compared to the BNC and can take up to 2 s to fuse the new target. After vergence training, the CI subject's responses are still slower than the BNC (comparing **Figures 3A,C**), but considerably faster than the subject's baseline responses (**Figure 3B**).

#### **TIME-SERIES OF THE BOLD SIGNAL FROM THE ROIs STUDIED**

**Figure 4** shows data from two subjects, one BNC (**Figure 4A**), one CI subject before vergence training (**Figure 4B**), and the same CI subject after vergence training (**Figure 4C**). **Figure 4** shows the average time series from the FEF-L (red lines), PPC-L (green lines), Broca-L (purple lines), and CV (blu shows data from two subjects, one e lines). **Figure 4** plots the BOLD percent signal

**FIGURE 3 | Convergence eye movement responses stimulated from 4◦ symmetrical convergence step stimuli from a BNC (A), a CI subject before vergence training (B), and the same CI subject**

**after 18 h of vergence training (C).** Each colored trace is a single eye movement response denoted in degrees of rotation as a function of time (s).

change as a function of volumes collected (70 volumes equating to 140 s in duration). The BNC shows an FEF time series which is more correlated (*r* = 0.66; *p* < 0.001) with the experimental block design (white and gray boxes for the 3.5 cycles of the experiment) compared to the CI subject before vergence training (*r* = 0.33; *p* < 0.01). After vergence training, this subject's FEF correlation with the block design increases (*r* = 0.73; *p* < 0.001). Similar trends are observed for the PPC and the CV. As expected, the time series from Broca's region (control ROI to study variability of a non-stimulated region) does not correlate with the experimental block design for the BNC and the CI before or after vergence training (*r* = 0.15 ± 0.05; *p* > 0.1).

#### **GROUP-LEVEL ANALYSES**

The peak velocities elicited from 4◦ symmetrical convergence steps were averaged for the seven BNC and the four CI subjects before and after repetitive vergence training. **Figure 5A** plots the average (bar) with one standard deviation (error bar) of the peak velocity (◦/s) from CI subjects before vergence training (blue bar), the same CI subjects after vergence training (red bar), and from

BNC subjects (green bar). When comparing the BNC group with the CI before vergence training group, an unpaired *t*-test revealed that significant differences were observed (*T* = 2.92; *p* < 0.02). The CI subjects had significantly slower peak velocities evoked from 4◦ symmetrical convergence steps compared to BNC subjects. The CI subjects also exhibited significant changes in peak velocities to symmetrical 4◦ convergence steps when comparing the responses after vergence training to the baseline before vergence training responses, when using a paired *t*-test (*T* = 6.93; *p* < 0.02).

**Figure 5B** shows the average with one standard deviation for the group-level analysis of the percent change in the BOLD signal per ROI for the following groups: CI subjects before vergence training (blue bar), CI subjects after vergence training (red bar), and BNCs (green bar). When comparing the BNC to the CI data before vergence training using an unpaired *t*-test, significant differences were observed within the FEF, PPC, and CV (*t* > 2.3; *p* < 0.05). No significant differences were observed within Broca's region between the BNC and either the before or after vergence training CI datasets (*t* > 1.1; *p* > 0.3). A paired *t*-test showed the percent change in the BOLD signal in the FEF, PPC, and CV of the four CI subjects who participated in vergence training were significantly greater after training compared to the baseline values (*t* > 2.6; *p* < 0.001). No statistical difference was observed in Broca's region (*t* = 1.2; *p* > 0.3) when comparing the baseline and after vergence training data.

A linear regression analysis was conducted of the average convergence peak velocity evoked from 4◦ symmetrical convergence steps as a function of the BOLD percent signal change for the FEF (**Figure 6A**), PPC (**Figure 6B**), Broca's Region (**Figure 6C**), and CV (**Figure 6D**). The left and right ROIs were averaged for **Figure 6**. A regression analysis revealed the convergence peak velocity of 4◦ symmetrical convergence steps from BNC and CI patients before and after vergence training was significantly correlated to the percent BOLD signal change within the FEF (*r* = 0.5; *p* < 0.05), PPC (*r* = 0.7; *p* < 0.01), and CV (*r* = 0.6; *p* < 0.01). Conversely, convergence peak velocity of 4◦ symmetrical convergence steps from BNC and CI patients before and after vergence training was not significantly correlated to the percent BOLD signal change within Broca's regions, which was the control ROI (*r* = −0.0059; *p* > 0.98).

#### **CLINICAL VISION PARAMETERS**

A paired *t*-test revealed a significant difference comparing the baseline (before vergence training) parameters and the after vergence training parameters for the following measurements: the NPC (*t* = 4.9; *p* = 0.04), BO positive fusional vergence range (*t* = 9.5; *p* = 0.01), near dissociated phoria (*t* = 11; *p* = 0.008), and CISS (*t* = 3.6; *p* = 0.05). All significant changes are improvements to each parameter studied.

#### **DISCUSSION**

The data support the hypotheses that were tested. Reduced convergence peak velocity from convergence step stimuli and functional activity within the FEF, PPC, and CV were observed in **(**

Cerebellar Vermis.

percent BOLD signal changes were averaged. The blue diamonds are from the CI subjects before vergence training, the red diamonds are the same CI

NPC and positive vergence amplitude (Scheiman et al., 2009).

those with CI before repetitive vergence training compared to BNC subjects. Both convergence peak velocity and functional activity significantly improved after vergence training in the CI subjects when comparing the pre and post-vergence training measurements. The average peak velocity of convergence responses was significantly correlated to the BOLD percent signal change within the functional activity of the FEF, PPC, and CV neural substrates. The results of this study will be compared to those in the literature.

#### **CLINICAL IMPLICATIONS OF LONG-TERM ADAPTATION EVOKED THROUGH VERGENCE TRAINING**

Understanding the relationship between the functional activity within the FEF, PPC, and CV and convergence eye movement responses has both basic science and clinical applications. Although the majority of humans perform vergence movements with ease, the dysfunction known as CI is reported to be present within 4.2–7.7% of the population (Hokoda, 1985; Scheiman et al., 1996; Porcar and Martinez-Palomera, 1997; Rouse et al., 1998, 1999). CI is an eye co-ordination and alignment problem, which can result in visual symptoms when engaged in reading or performing other near work (Scheiman et al., 2011).

The randomized clinical trial, the Convergence Insufficiency Treatment Trial (CITT), showed that OBVAT was successful in 73% of children, resulting in significantly improved symptoms, Clinical signs and symptoms were sustained 1 year post-therapy for most subjects (CITT, 2009). OBVAT is composed of symmetrical, horizontal convergence movements. Hence, although the stimulus used within this current study may not occur often in natural viewing conditions, it is the basis of therapeutic interventions to treat patients with CI (Cooper et al., 1998; Scheiman and Wick, 2008; Scheiman et al., 2011).

The Dual-Mode model of vergence describes vergence as a two component system composed of a fusion initiating and a fusion sustaining component (Hung et al., 1986; Lee et al., 2012). The transient fusion initiating component is modeled as a preprogrammed control system mainly contributing to the vergence system's speed but is not necessarily very accurate. The fusion sustaining component is feedback controlled and facilitates the accuracy of the vergence system. The present data suggest that the fusion initiating component, which mainly contributes to the vergence peak velocity, is modified after vergence training. The CI subjects had reduced peak velocity before training, which increased after training. The results of this study suggest that vergence training protocols may concentrate on the stimulation of the preprogrammed portion of vergence system.

Investigations identifying the neural substrates responsible for vergence oculomotor learning are scare. Several saccade oculomotor studies suggest the oculomotor vermis is responsible for oculomotor learning within the saccadic system (Iwamoto and Kaku, 2010; Prsa and Thier, 2011). Evidence also suggests that the FEF can be modified through adaptation when studying saccades (Lee et al., 2011). This present study provides a critical step in understanding the brain-behavior relationship of how vergence training is inducing changes to the functional activity within the FEF, PPC, and CV, which in part mediates convergence oculomotor responses. Future research can study neurological differences between various vergence training protocols to determine how changes within neural substrates facilitate an improvement in visual comfort while performing near work such as reading. Such knowledge could ultimately lead to an improvement in vergence training protocols.

#### **BOLD PERCENT SIGNAL CHANGE IN RELATION TO OTHER BODIES OF LITERATURE**

Non-human primate single cell electrophysiology studies have investigated the influence of disparity in the FEF using symmetrical step stimuli (Gamlin and Yoon, 2000), near and far targets (Ferraina et al., 2000), and smooth sinusoidal tracking stimuli (Fukushima et al., 2002; Akao et al., 2005). The FEF and PPC have also been shown to be involved in predictive oculomotor learning (Tseng et al., 2013). The PPC encodes for different binocular distances defined by different vergence angles studying primates using single cell recordings (Genovesio and Ferraina, 2004; Ferraina et al., 2009; Breveglieri et al., 2012) and humans using transcranial magnetic stimulation (Kapoula et al., 2001, 2004, 2005; Yang and Kapoula, 2004) and fMRI (Quinlan and Culham, 2007; Alvarez et al., 2010a; Alkan et al., 2011a,b). Primate single cell studies have also shown that the CV is used to mediate a vergence response (Gamlin et al., 1996; Nitta et al., 2008a,b). Patients, particularly those with lesions to the cerebellar vermal regions, exhibit a decrease in slow tracking vergence (Sander et al., 2009).

This present study further confirms that the FEF, PPC, and CV are metabolically active during a vergence task. The novelty of this study's results is that the functional activity of the FEF, PPC and CV are: (1) reduced in CI subjects at baseline compared to BNC subjects, (2) significantly increased after 18 h of vergence training to levels more similar to those exhibited by the BNC subjects, and (3) significantly correlated to the convergence peak velocity elicited from 4◦ symmetrical convergence stimuli. The results support the hypothesis that subjects with CI have reduced functional activity within the vergence neural substrates and reduced peak velocity of convergence responses compared to BNC. Results further support that after vergence training; the functional activity improves to levels more similar to those observed in BNC subjects.

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

The reduced strength in fMRI activity and convergence peak velocity measurements observed from CI subjects compared to BNC is recommended for investigation in a masked randomized clinical trial where both CI and BNC participate in vergence training. Such a study could determine whether these differences between BNC and CI subjects generalize in a larger population and hence may, in part, explain asthenopia in CI patients.

The techniques used within the present study could also be applied to study the brain-behavior relationships of other oculomotor and vision dysfunctions. For example, Bucci and colleagues studied vergence insufficiency patients whose symptoms included headache and vertigo before and after orthoptic vergence training (Bucci et al., 2004, 2006a,b, 2011; Jainta et al., 2011). The techniques presented here could be used to better understand the mechanisms underlying vergence training for those with other visual and vestibular dysfunctions.

## **CONCLUSIONS**

The data collected within this study support that CI subjects have significantly reduced convergence peak velocity to 4◦ symmetrical convergence steps and BOLD percent signal change within the FEF, PPC, and CV compared to BNC subjects. Both convergence peak velocity and BOLD percent signal changes within the FEF, PPC, and CV significantly improved post-vergence training in CI subjects compared to their baseline measurements. The convergence peak velocity was significantly correlated to the BOLD percent signal change in the FEF, PPC, and CV. Results suggest that repetitive vergence training leads to an increase in the functional activity of the FEF, PPC, and CV which may in part lead to the increase in convergence peak velocity to symmetrical step stimuli.

## **ACKNOWLEDGMENT**

This research was supported in part by NSF MRI CBET1228254 and NIH EY023261 to Tara L. Alvarez and NIH AG032088 to Bharat B. Biswal.

## **REFERENCES**


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

*Received: 01 April 2014; accepted: 27 May 2014; published online: 17 June 2014. Citation: Alvarez TL, Jaswal R, Gohel S and Biswal BB (2014) Functional activity within the frontal eye fields, posterior parietal cortex, and cerebellar vermis significantly correlates to symmetrical vergence peak velocity: an ROI-based, fMRI study of vergence training. Front. Integr. Neurosci. 8:50. doi: 10.3389/fnint.2014.00050 This article was submitted to the journal Frontiers in Integrative Neuroscience. Copyright © 2014 Alvarez, Jaswal, Gohel and Biswal. 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.*

#### *Marine Vernet <sup>1</sup> \*, Romain Quentin1, Lorena Chanes 1, Andres Mitsumasu1 and Antoni Valero-Cabré1,2,3*

*<sup>1</sup> Centre de Recherche de l'Institut du Cerveau et de la Moelle Epinière, CNRS UMR 7225, INSERM UMRS 975 and Université Pierre et Marie Curie, Paris, France <sup>2</sup> Laboratory for Cerebral Dynamics Plasticity and Rehabilitation, School of Medicine, Boston University, Boston, MA, USA*

*<sup>3</sup> Cognitive Neuroscience and Information Technology Research Program, Open University of Catalonia, Barcelona, Spain*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

## *Reviewed by:*

*Albino J. Oliveira-Maia, Champalimaud Foundation, Portugal Thomas Nyffeler, Bern University Hospital and University of Bern, Switzerland*

#### *\*Correspondence:*

*Marine Vernet, Groupe de Dynamiques Cérébrales, Plasticité et Rééducation, Institut du Cerveau et de la Moelle Epinière, Pitié-Salpêtrière, 47 Boulevard de l'Hôpital, 75013 Paris, France e-mail: marine.vernet@gmail.com*

The planning, control and execution of eye movements in 3D space relies on a distributed system of cortical and subcortical brain regions. Within this network, the Eye Fields have been described in animals as cortical regions in which electrical stimulation is able to trigger eye movements and influence their latency or accuracy. This review focuses on the Frontal Eye Field (FEF) a "hub" region located in Humans in the vicinity of the pre-central sulcus and the dorsal-most portion of the superior frontal sulcus. The straightforward localization of the FEF through electrical stimulation in animals is difficult to translate to the healthy human brain, particularly with non-invasive neuroimaging techniques. Hence, in the *first* part of this review, we describe attempts made to characterize the anatomical localization of this area in the human brain. The outcome of functional Magnetic Resonance Imaging (fMRI), Magneto-encephalography (MEG) and particularly, non-invasive mapping methods such a Transcranial Magnetic Stimulation (TMS) are described and the variability of FEF localization across individuals and mapping techniques are discussed. In the *second* part of this review, we will address the role of the FEF. We explore its involvement both in the physiology of fixation, saccade, pursuit, and vergence movements and in associated cognitive processes such as attentional orienting, visual awareness and perceptual modulation. Finally in the *third* part, we review recent evidence suggesting the high level of malleability and plasticity of these regions and associated networks to non-invasive stimulation. The exploratory, diagnostic, and therapeutic interest of such interventions for the modulation and improvement of perception in 3D space are discussed.

**Keywords: FEF, brain mapping, transcranial magnetic stimulation, visual performance, visuo-spatial attention, 3D vision, visual awareness, plasticity rehabilitation**

#### **INTRODUCTION: FEF, A CROSSROADS FOR EYE MOVEMENTS AND VISUO-SPATIAL COGNITION**

The frontal eye field (FEF) is an area of the frontal cortex in animals over which electrical stimulation is able to trigger eye movements. Electrophysiological studies in the monkey defined the FEF as an area containing visual, motor, and visuo-motor cells (Bruce and Goldberg, 1985) essential for the preparation and triggering of eye movements. This site operates as a crucial site of networks integrating other regions located in widespread locations. In humans for example, such gaze control systems include in the frontal lobe the supplementary eye field (SEF), the pre-supplementary eye field (pre-SEF), the dorsolateral prefrontal cortex (DLPFC), the cingulate eye field (CEF) within the anterior cingulate cortex and the dorso-medial frontal cortex, and in the parietal lobe, the parietal eye field (PEF) and areas of the posterior parietal cortex (PPC). Finally, subcortical structures, such as the superior colliculus (SC) in the midbrain are also considered essential to trigger eye movements. All these areas operate cooperatively, nonetheless some of them contribute to the triggering of eye movements under specific situations: the PEF for example has a role in reflexive saccades, the FEF participates in voluntary saccades, the SEF contributes to the development of more complex motor programs involving gaze (Pierrot-Deseilligny et al., 2002). Other areas, such as the CEF and the DLPFC, are more generally dedicated to cognitive aspects (e.g., motivation, memory) of oculomotor control (Gaymard et al., 1998b).

The anatomy of input and output projections within nodes of this network has been particularly well characterized in the monkey brain, and has revealed itself as a highly complex constellation of widespread interactions. The predominant neural

**Abbreviations:** CEF, cingulate eye field; DLPFC, dorsolateral prefrontal cortex; EEG, electroencephalography; FEF, frontal eye field; fMRI, functional magnetic resonance imaging; IPS, intraparietal sulcus; LIP, lateral intraparietal area; MEG, magneto-encephalography; MST, medial superior temporal; PEF, parietal eye field; PET, positron emission tomography; PPC, posterior parietal cortex; rTMS, repetitive transcranial magnetic stimulation; SC, superior colliculus; SEF, supplementary eye field; tACS, transcranial alternate current stimulation; tDCS, transcranial direct current stimulation; TMS, transcranial magnetic stimulation.

inputs to the FEF originate in other cortical eye fields, including the SEF, the PEF, the middle superior temporal area, and the *principal sulcus* region (Schall et al., 1993; Tian and Lynch, 1996). The FEF also receives weak connections from the middle temporal area (MT), which may act as a relay between the striate / extrastriate cortices and the parietal cortex and FEF (Tian and Lynch, 1996). The FEF projects to many areas within the frontal cortex (Stanton et al., 1993), the occipital and parietal cortices such as V2/V3/V4, the middle temporal area (MT), the medial superior temporal area (MST) and the superior temporal visual area (Stanton et al., 1995). Finally, important reciprocal connections have been demonstrated between the FEF and the lateral intraparietal area (LIP) and more generally with the parietal cortex (Huerta et al., 1987; Cavada and Goldman-Rakic, 1989; Stanton et al., 1995; Tian and Lynch, 1996). Subcortically, the FEF projects directly to the brainstem (pons) (Leichnetz et al., 1984; Segraves, 1992). It also sends afferents to the SC (Schlag-Rey et al., 1992), either directly (Segraves and Goldberg, 1987) or indirectly *via* the basal ganglia (Stanton et al., 1988), and to other subcortical nuclei within the thalamus, subthalamus and tegmentum (Stanton et al., 1988). The FEF receives inputs from subcortical sites, including the substantia nigra, the SC. Finally, the cerebellum projects to thalamic regions innervating the FEF (Lynch et al., 1994).

Most of the earlier knowledge about the FEF was built-up on the basis of non-human primates experiments. A major emphasis has been put on the role of the FEF in the preparation and execution of saccades (Bizzi, 1968; Bruce and Goldberg, 1985). However, FEF also participates in the control of all the other types of eye movements, such as smooth pursuit or optokinetic nystagmus (OKN) (Bizzi, 1968; MacAvoy et al., 1991) and fixation (Izawa et al., 2004a,b, 2009). The intracortical stimulation of several subareas within the FEF is also able to trigger vergence movements (changes of the depth of the gaze) (Crosby et al., 1952, cited by Robinson and Fuchs, 1969). More recently, Gamlin and Yoon (2000) showed that a region within the pre-arcuate cortex in rhesus monkeys, immediately rostral to the saccade-related region in the anterior bank of the arcuate sulcus, is involved in vergence, accommodation and the sensorimotor transformations required for these movements. Moreover, Ferraina et al. (2000) showed that most neurons in a region of the anterior bank of the arcuate sulcus where saccades could be evoked with low current stimulation were also sensitive to disparity. The caudal portion of the FEF that contains smooth pursuit neurons also carries binocular signals related to vergence movement (Kurkin et al., 2003) and the majority of FEF pursuit neurons would respond to both frontal pursuit and pursuit in depth (Fukushima et al., 2002). There is also evidence that near and far spaces are differentially encoded in the frontal cortex including the FEF (Pigarev et al., 1979; Rizzolatti et al., 1983). Thus, the FEF appears to be involved in every sort of eye movements in 3D space.

It is expected that human FEF will be recruited, as in animals, for all types of ocular behavior: saccades, fixation, smooth pursuit, OKN, vergence. However, within each type, the specific experimental set-ups conditioning different categories of eye movements (e.g., reflexive, voluntary) will modulate the involvement of the FEF. Indeed, different cortical oculomotor areas are differentially recruited according to the category of eye movements (Gaymard et al., 1998a). The fact that the cognitive context is modulating the involvement of the FEF is reminiscent of the other roles played by the FEF in visuo-spatial attention, visual awareness, and perceptual modulation.

Before entering into the details on the various roles of the FEF (part II) and how its activity can be modulated for clinical purposes (part III), we will describe the efforts made to localize this area in both non-human and human primates (part I). As expected from an area contributing to numerous functions, the exact localization will strongly depend on the methods and specific paradigms used to assess it.

#### **LOCALIZATION OF FEF**

The primate FEF is defined physiologically as the portion of the dorsolateral prefrontal cortex from which low-intensity intracortical stimulation is able to elicit rapid eye movements. Using this invasive approach, the monkey FEF has been located by some studies in the frontal lobe along the anterior border of the arcuate fissure, which would correspond to Brodmann's area 8, or overlapping with both areas 8 and 6 (or, using Walker's nomenclature, with areas 8A and 45) (for a review, see Tehovnik et al., 2000). According to the results of neuroimaging studies, the human FEF is mostly thought to be located in the superior pre-central sulcus near the caudal end of the superior frontal sulcus, which corresponds to Brodmann's area 6. However, as will be described in the second part of this review, the FEF contributes not only to several aspects of eye movements but also to different cognitive domains, and the exact location of the FEF will strongly depend not only on the methods (e.g., stimulation vs. neuroimaging) but also on the tasks (e.g., type of eye movements and type of control conditions, see e.g., Paus, 1996) and activation criteria (e.g., intensity of stimulation, see Blanke et al., 2000 for a discussion) used. Overall, it is still not entirely clear whether the reported inter-species differences in FEF location can be related to genuine anatomical differences between non-human primates and humans, caused by the use of different mapping methods or they simply reflect interindividual differences, which have not always been systematically studied in large cohorts of animals and human participants.

In the next pages, we will review some of the numerous studies that have attempted to determine the anatomical location of the FEF employing: microstimulation, intracranial recordings, functional magnetic resonance imaging (fMRI), positron emission tomography (PET), magnetoencephalography (MEG) and transcranial magnetic stimulation (TMS). A summary of the localizations reported in these studies can be found in **Table 1**.

#### **MICROSTIMULATION AND RECORDINGS IN NON-HUMAN PRIMATES: THE ORIGINAL DEFINITION**

In 1874, Ferrier summarized stimulation studies performed on several animal species including cats, dogs and rabbits as follows: "*In the superior frontal convolution, in advance of the centre for certain forward movements of the arm, as well as in the corresponding part of the middle frontal convolution, are areas, stimulation of which causes lateral (crossed) movements of the head and eyes and dilatation of the pupils."* (Ferrier, 1874).

#### **Table 1 | Localization of FEF across studies, techniques and species.**


More than one hundred years later, other microstimulation studies evoking eye movements (Robinson and Fuchs, 1969; MacAvoy et al., 1991; Gottlieb et al., 1993; Izawa et al., 2004a,b, 2009), electrophysiological recordings during visual stimulation and/or eye movements (Bizzi, 1968; Wurtz and Mohler, 1976; Bruce and Goldberg, 1985; Segraves and Goldberg, 1987) and studies comparing cells discharge patterns during behavior or its alteration during the stimulation of these same neuronal populations (Bruce et al., 1985; Gottlieb et al., 1994) confirmed the existence of an FEF located in the posterior part of the prearcuate sulcus. They distinguished visual (modulated by functional significance), motor and visuo-motor neural populations for saccade, pursuit and fixation/saccade suppression, somehow spatially segregated and with different stimulation thresholds, which depended on the activation state of the monkey at the time of the stimulation. Each sub-region showed its specific organization. For saccades for instance, stimulation of ventro-lateral regions evoked small amplitude saccade whereas stimulation of dorso-medial regions induced large saccades; moreover the direction of the saccades varied as a function of the depth of stimulation in the arcuate sulcus (Tehovnik et al., 2000). Interestingly there is also evidence that in some primate species (e.g., owl monkeys), the stimulation of the dorsal premotor area, posterior to the usually defined FEF, can also evoke saccades, suggesting that such posterior area, potentially closer to the human FEF, could also belong to the non-human primates FEF (Preuss et al., 1996).

Although microstimulation is considered a gold standard technique to reveal a causal relation between a region and a brain function, it has potential limitations (see Amiez and Petrides, 2009; for review). First, the extent and number of responding areas depends on stimulation intensity, whose traditional threshold level (50µV) is set up arbitrarily. Second, within the same study or across studies and depending on the experimental design chosen, some cortical areas have been less systematically sampled than others, a fact that could have biased output maps overemphasizing the role of certain locations while undermining the contribution of others. Third and last, intracortical stimulation can evoke eye movements from direct FEF activation, but also by activating intracortical white matter pathways connecting the FEF to other areas (Luna et al., 1998), a phenomenon that could easily blur the borders of cortical representations and lead to mislocalizations.

#### **MICROSTIMULATION IN HUMANS**

Microstimulation procedures have not been solely restricted to a use in animal models. They have also been occasionally performed in epileptic patients, either per-operatively or outside of surgery rooms in more ecological conditions via chronically implanted subdural electrodes in fully awake patients. Using the first procedure, Foerster (1936, cited by Blanke et al., 2000) induced eye movements only from the posterior part of the middle frontal gyrus, whereas Rasmussen and Penfield (1948 cited by Blanke et al., 2000) report to have induced similar effects from all frontal gyri and the pre-central gyrus. With implanted subdural electrodes at the level of and in front of the motor representation, Godoy et al. (1990) evoked contralateral conjugate eye movements (mostly saccades), and sometimes accompanying head version following eye deviation. Blanke et al. (2000) investigated systematically the current intensity needed to elicit unilateral eye movements and found, consistently with monkey studies, that the eye fields inducing saccades and smooth eye movements are located in the posterior part of the middle frontal gyrus and neighboring portions of the superior frontal gyrus but not in the inferior frontal gyrus or in the pre-central sulcus.

Thus, microstimulation in well-controlled settings in human patients can yield results equivalent to those demonstrated in non-human primates with similar interventions. As also mentioned above for the animal, the intensity used for intracortical stimulation in humans arbitrarily determines the number and the size of the cortical clusters that activated directly or indirectly by connectivity are ultimately causally associated to the FEF. In addition, such studies are also constrained by the spatial location, distribution, and coverage of the implanted electrodes, which are strictly guided on the basis of clinical and not scientific criteria, and limited by the scarcity of time available for testing and the lack of large cohort of similarly implanted patients available to provide statistical evidence. Moreover, for ethical reasons, such procedures are only performed in human patients who have undergone developmental or acquired anatomical and functional alterations and do not necessarily provide accurate information on the healthy brain. In view of such limitations, non-invasive neuroimaging techniques, such as PET, fMRI, MEG and also non-invasive neurostimulation by TMS have become particularly popular in cognitive neuroanatomy and have been employed in the quest to locate the FEF in both humans and to a lesser extent in animals.

#### **NEUROIMAGING**

The spreading of neuroimaging techniques such as PET and more recently fMRI has allowed the evaluation of FEF location and function in healthy human brains. The gradual increase of spatial resolution has permitted defining progressively smaller and better-delimited regions corresponding to the FEF. Within the large FEF region characterized by means of PET, several subareas associated with eye movements have been revealed using fMRI.

The variability of FEF location and function, found across different PET studies, has been reviewed by Paus (1996). Pioneering explorations using PET reported large activations in the human lateral frontal cortex during saccade execution. Most of these studies defined the FEF as part of the pre-central sulcus in the frontal lobe (Petit et al., 1995, 1996). Nonetheless this region has been sometimes localized in the anterior portion of the precentral gyrus around the pre-central sulcus (Fox et al., 1985; Anderson et al., 1994; Law et al., 1997), or within the posterior portion of the pre-central gyrus around the central sulcus (Sweeney et al., 1996). A large range of eye movement types have shown to activate the FEF: fixation (Petit et al., 1995), reflexive or memory saccades (Anderson et al., 1994), saccades with or without visual cues (Fox et al., 1985), suppressed or imagined saccades (Law et al., 1997), anti-saccades (O'Driscoll et al., 1995; Sweeney et al., 1996), predictive saccades and gaze pursuit (O'Driscoll et al., 2000). Some of these studies showed that the intensity of FEF activation was neither influenced by target presence, cue type, task complexity (Fox et al., 1985) nor by whether the saccades were voluntary or previously learned (Petit et al., 1996), whereas other studies showed, on the contrary, a modulation of FEF activation from fixation to reflexive or volitional saccades (O'Driscoll et al., 1995; Sweeney et al., 1996).

The higher spatial resolution of fMRI recordings in humans has allowed researchers to restrict the site hosting the FEF along the pre-central sulcus (Darby et al., 1996; Muri et al., 1996). It also permitted to identify within this sulcus, several sub-areas subtending potentially distinct functions related to saccadic activity. Petit and Haxby (1999) and Petit et al. (1997) reported the FEF as located at the junction of the pre-central sulcus and the superior frontal sulcus extending laterally to the pre-central gyrus. They described a saccade-related FEF and a smaller, more inferior, and more lateral gaze pursuit-related FEF, which according to another study could overlap (Berman et al., 1999). Rosano et al. (2002) found a restricted area within the pre-central sulcus, integrating the saccade area, as located mainly on the rostral bank close to the cortical surface, and the pursuit area situated deeper in the sulcus, suggesting similar superficial/deep activation as the one characterizing non-human primates. Activation restricted to the pre-central sulcus was also shown in individual subjects in the study from Luna et al. (1998) contrasting simple visually-guided saccades to fixation. They described a consistent activation of the superior portion of the pre-central sulcus and a less consistent activation of the inferior portion of the pre-central sulcus. Similarly, different clusters of activation within the precentral sulcus were found in other studies (Corbetta et al., 1998; Beauchamp et al., 2001). A meta-analysis performed on PET and fMRI datasets confirmed that for both visually and voluntarilytriggered saccades, the FEF lies in the pre-central sulcus close to its intersection with the superior frontal sulcus, potentially extending onto the superior and inferior subregions of the superficial portions of the pre-central gyrus (Grosbras et al., 2005). It should be noted, however, that some recent studies localized the superior FEF within the ventral portion of the superior pre-central sulcus, either at the end or at the most posterior region of the middle frontal gyrus, instead of at the level of its intersection with the superior frontal sulcus (Amiez et al., 2006).

Neuroimaging approaches show some limitations as compared to neurostimulation methods to determine the brain regions involved in a given saccadic behavior. First, neuroimaging methods are less sensitive than neurostimulation approaches in the detection of small saccade-related areas (Luna et al., 1998); second, group-averaging strategies employed in neuroimaging approaches to increase statistical power may come at the risk of shifting activation sites in case of strong interindividual anatomical differences (Luna et al., 1998; Amiez and Petrides, 2009); third, whereas brain stimulation mostly reveals contralaterallyevoked saccades, fMRI studies are built on protocols embedding bilateral and repetitive eye movements conditions and compared to a gaze fixation baseline. Hence differences in region size and shifted FEF localizations (Blanke et al., 2000) could be well caused by either the influence in contrast analyses from cells within the FEF involved in fixation and/or the mix up of activity related to different saccade directions within the same analyses. Another important concern raised by Tehovnik et al. (2000) and Amiez and Petrides (2009) is that in neuroimaging protocols, no instruction is given regarding blinking and return-to-center saccades in between trials (which is often accompanied by blinks). This could also explain rather posterior mislocalizations of the FEF, which would mistakenly encompass activity from regions within the motor strip involved in eyelid motion. In favor of this possibility, a PET protocol with multiple saccades, inducing comparable blinks frequency in the saccade and the control condition, found activity within the middle frontal gyrus (Kawashima et al., 1998). This observation is consistent with a more anterior location for the FEF in the frontal lobe (Tehovnik et al., 2000) and argues in favor of important blinking-related biases in prior PET and fMRI explorations.

In spite of the above-mentioned problems, neuroimaging studies still have the advantage of providing normalized coordinates corresponding to group mean activation peaks (see **Figure 1**) that can be easily compared across studies and used as targets for subsequent non-invasive brain stimulation approaches on search of causality. In that vein, the meta-analysis of 8 PET studies involving 62 healthy participants designed by (Paus, 1996) suggested a reference location in Talairach coordinates (**Table 2**). Subsequent fMRI have contributed Tailarach coordinates reflecting similar or more posterior loci for the main (e.g. the superior) FEF site during the active performance of saccades or, on the contrary, more anterior location when blinks were avoided (**Table 2**).

#### **RECONCILING NON-HUMAN PRIMATES' AND HUMANS' LOCATIONS FOR THE FEF?**

In brief, the non-human primate FEF, localized mainly thanks to microstimulation studies, lies in a more rostral location

**FIGURE 1 | Localization of FEF according to several studies on the MNI (Montreal Neurological Institute) brain template viewed from top (A), front (B), right (C) and left (D).** Color codes as follows. Green: meta-analysis of PET studies from Paus (1996); Blue: fMRI study of Luna et al. (1998); Red: fMRI study of Petit and Haxby (1999); Yellow: MEG study of Ioannides et al. (2004); Purple: coordinates estimated by Tehovnik et al. (2000) based on the PET study of Kawashima et al. (1998). A sphere of 1 cm radius is positioned at the center of FEF activation from each study. SPM (Statistical Parametric Mapping, http://www*.*fil*.*ion*.*ucl*.*ac*.*uk/spm/) with MarsBar toolbox was used to design the spheres and MRIcroGL software (http://www*.*mccauslandcenter*.*sc*.*edu/mricrogl/) was used for glass brain illustration.


**Table 2 | Coordinates of left and right FEF from a few neuroimaging studies.**

(Brodmann's area 8) than the human FEF, localized mainly thanks to neuroimaging studies (Brodmann's area 6). A suggestion to reconcile such discrepancies between monkey and human reports is that the more posterior FEF location in humans has been erroneously attributed to Brodmann's area 6. Following that line, a study focused on the delimitation of cytoarchitectonics areas in post-mortem human brains containing the superior element of the pre-central sulcus and the caudal end of the superior frontal sulcus (Rosano et al., 2003). This study suggested that the precentral sulcus might represent a transitional area between the rostral granular cortex and the caudal agranular cortex. Thus, the FEF would be located within a region that appears to have a similar chemoarchitecture (Stanton et al., 1989; Rosano et al., 2003) in both species, even if lying in a more caudal location in humans.

Other studies have suggested that discrepancies between monkeys and humans in FEF location arise from methodological differences rather than from a genuine inter-specie divergence. We already mentioned that microstimulation in humans can yield results equivalent to those demonstrated in non-human primates with similar interventions (see Section Microstimulation in Humans). Do monkey fMRI recordings also reveal similar activations than the ones shown in humans with this same mapping technique? Koyama et al. (2004) conducted an fMRI study in macaque monkeys and revealed three saccade-related foci of activation. One was located in the bank of the arcuate sulcus, approximately in Brodmann's area 8, which corresponds to the classical non-human primate FEF, whereas the remaining two laid in premotor areas, and more precisely, in the inferior and superior precentral sulci within Brodmann's area 6. Thus, monkey fMRI studies reveal indeed activations similar to those found in humans. Further studies are needed to conclude on whether the discrepancy between non-human primates and humans results mainly arises from different cytoarchitectonics areas in different species or from the use of different methods. Probably, a deeper exploration of the multiple foci associated with the FEF will help to clarify its role and localization across species.

Finally, the use of a third methodology can shed a new light on the interpretation of results arising from monkeys' microstimulation and humans' fMRI studies. Taking advantage of the exquisite temporal resolution of MEG and the possibility of localizing source signals with a reasonable spatial resolution Ioannides et al. (2004, 2005) suggested an anterior location similar to the one found in microstimulation studies (e.g., in Ioannides et al., 2004 in 3 subjects, see **Figure 1** and **Table 2** for Talairach coordinates; however, note the high inter-individual variability of the Y coordinate between 24 and −3 for the right FEF). According to a MEG single subject study of this same group, the activity associated to the FEF could shift along a rostro-caudal axis, from the rostral site identified in microstimulation studies to the caudal region reported in fMRI studies, during the saccade preparation time (Ioannides et al., 2010), suggesting an unexpected confounding role of this variable. Most importantly, this study suggested that both the rostral (usually described for the non-human primates) and the caudal (usually described for the humans) sites can be identified in humans at different timing.

#### **TMS: IN SEARCH OF A CAUSAL FUNCTIONAL LOCALIZER IN HEALTHY HUMANS**

In order to overcome the limitations of invasive human microstimulation but still benefit from its causation power, some researchers have turned to TMS as a causal brain mapping technique. TMS is based on a non-invasive induction of small currents intracortically in order to modulate brain activity at specific cortical areas with a relatively good spatial resolution, in the order of 1.2–3.5 mm radius (Wagner et al., 2007; Bijsterbosch et al., 2012). Depending on variables such as the stimulated area, magnetic pulse intensity, pre or post event time window chosen for pulse delivery, or the temporal distribution of individual pulses employed either in short bursts or long stimulation patterns, TMS can have an immediate (i.e., the so-called *online*) or lasting (so-called *offline*) facilitatory or disruptive impact on neurophysiological activity and consequently on the performance driven by the targeted cortical region and its associated network of areas (Valero-Cabre et al., 2005, 2007). Thanks to these properties, this technique is used to explore the causal contribution of different cortical areas and associated anatomical systems to human behavior in healthy individuals, whereas in clinical applications TMS has been employed to manipulate patterns of activity and drive therapeutically interesting outcomes for neurological or neuropsychiatric conditions (Valero-Cabre et al., 2011).

As TMS operates by using a magnetic field to non-invasively induce electrical current within the cortex, it has been hypothesized that, as intracranial electrical stimulation does, magnetic stimulation should also be able to trigger eye movements. However, TMS delivered systematically into frontal locations where the FEF is located has surprisingly proven unable to trigger eye movements (Muri et al., 1991; Wessel and Kompf, 1991), to disturb central fixation (Zangemeister et al., 1995) or to modify saccade or smooth pursuit movement in flight (Wessel and Kompf, 1991). Only under facilitating conditions, e.g., during the performance of a double-step saccade task, has rTMS been reported to be able to induce multistep short-latency eye movements in a few subjects (Li et al., 1997). This result strongly suggests that the organization of the systems within the FEF devoted to eye movement is different than that characterizing the primary motor cortex for limb movements. Indeed, the latter projects directly to spinal motor neurons and can thus easily trigger hand movements when the primary motor cortex is stimulated with TMS. In contrast, circuits leading to gaze movements include intermediate synaptic chains and structures and hence might not be that easy to activate with the same technique. Additionally, it has been also argued that such differences in activation could also be attributed to the fact that the TMS-induced currents may have been either insufficiently high or too poorly focalized to effectively activate polysynaptic chains down to saccadic motor neurons (Muri et al., 1991; Wessel and Kompf, 1991).

Although TMS cannot directly induce eye movements in healthy humans, it can effectively interfere with the processing of visually and non-visually guided saccades. Such modulatory phenomena have been employed to design new causal methods to localize the FEF in healthy humans. In such procedures, a TMS coil is moved around the approximate location of the FEF. Pulses are delivered with intensity generally at or slightly above the resting motor threshold (RMT), i.e., the intensity at which they induce overt evoked hand muscle activations in half of the trials when stimulating the primary motor cortex. Like in microstimulation studies, the choice of intensity is somehow arbitrary. Indeed, it is likely that simulating at 100 or 120% of RMT will lead to different results. More importantly, stimulating at an intensity based on the RMT does not warrant consistent results across participants as it is known that, except under certain circumstances (Deblieck et al., 2008) the TMS-measured excitability of one area is poorly predicting the TMS-measured excitability of another area (Stewart et al., 2001; Boroojerdi et al., 2002; Antal et al., 2004; Kahkonen et al., 2005). Notwithstanding this limitation, TMS procedures allow identifying the FEF as the area in which stimulation significantly modifies some saccadic outcome parameters, generally the latency of a specific type of saccade.

Using such methods, the greatest delays in saccade latencies have been obtained when targeting an area on or 2 cm anterior to the inter-aural line, approximately 6 cm lateral to the vertex, situated between areas over which TMS could generate motorevoked potentials in hand's and face's muscles (Thickbroom et al., 1996). The authors of these reports did not exclude that the FEF could also extend more rostrally, and that such projections cannot be easily assessed either because rostral stimulation would cause blinks, or because the anterior portions of FEF are involved in other aspects of saccade programming. Other studies localized the FEF within areas situated 2 cm (Ro et al., 1999) or 1.5 cm (Ro et al., 2002) rostral to the motor hand area. However, such site, probably belonging to the middle-frontal gyrus and close to the pre-central sulcus, could not be localized in every tested participant. Moreover, this localization suffers from important interindividual differences, mostly within the coronal or dorsal to medial plane, consistent with reports from neuroimaging studies (Paus, 1996). Studies by O'Shea et al. (2004) and Silvanto et al. (2006) targeting the FEF based on anatomical landmarks within the middle frontal gyrus, just rostral from the junction of the pre-central and the superior frontal sulci, reported that such area corresponds to about 3–4 cm rostral to the individual motor hand area representation. In spite of its rostral location, the reported mean Talairach coordinates locate very close to the coordinates reported by Paus (1996) in their meta-analysis (**Table 2**). Further work to causally define the FEF location in individual participants by means of TMS employing individualized MRI guidance and studies directly comparing TMS and fMRI FEF localizers within the same population of subjects remain to be performed.

#### **CONSEQUENCES FOR STUDYING MOTOR, VISUAL OR COGNITIVE PROPERTIES OF THE FEF**

**Table 1** summarizes the findings for the above-cited literature concerning the search for FEF localization. Variability across species, methods, hemispheres, and individuals in the number of foci associated with FEF and their exact localization raises concerns about how we can explore its role in eye movements or cognitive function.

In TMS studies exploring the causal contributions of FEF in eye movements or cognitive processes such as attentional orienting, consciousness or decision making (see Section on the Role of the FEF), the gold standard would be to use a similar mapping methodology to identify the exact location of this region prior to its manipulation. Based on this notion, for instance, Olk et al. (2006) took the time to identify an area around its *a priori* anatomical location on which TMS induced longer latencies for contralateral than ipsilateral saccades. However, in order to limit the duration of the experiments, other studies employed relative coordinates leading to the average location, expressed as the distance in cm from the motor hand area (which can be easily identified with TMS) and successfully reported significant effects on quantitative measures of eye movements (Wipfli et al., 2001; Nyffeler et al., 2006a,b; Nagel et al., 2008; van Donkelaar et al., 2009). Similarly, in another study, the FEF was localized by probing a series of frontal cortical sites rostral to the motor hand area until evoked hand motor responses disappeared (Leff et al., 2001). Nevertheless, one of the most commonly used strategies consisted in targeting those locations identified in anatomical MRIs by means of on sulci/gyri configurations (O'Shea et al., 2004), or on the basis of normalized coordinates from neuroimaging studies or meta-analyses (Grosbras and Paus, 2002), or employing individual functional localizers based on fMRI acquisitions performed during eye movements tasks (Gagnon et al., 2006).

In conclusion, potential conflicting results across studies concerning the function of the FEF might be related, among other factors, to variability in the way it is localized. This observation has to be kept in mind when interpreting the results that will be presented in the following part.

#### **ROLE OF FEF**

This section will review the role of the FEF in eye movements and in visuo-spatial attention, visual awareness, and perceptual modulation.

#### **ROLE OF FEF IN OCULOMOTOR TASKS**

In humans, knowledge on the role of FEF in several types of eye movements (summarized in **Table 3**) has been mainly derived from clinical cases in which the FEF has been damaged or from applying TMS on the FEF of healthy persons. These studies are reported in **Tables 4**, **5** and the conclusions derived from them are reported below.

#### *Lesions studies*

Most of lesion studies describing the role of the FEF have been focusing on oculomotor deficits that are reported in **Table 4**. The general pictures emerging from this literature is that FEF lesions very mildly affect the most reflexive saccades but might delay eye movements for which a voluntary component is introduced, for instance concerning fixation disengagement. Thus, although the triggering of reflexive saccades is more likely under the control of the PPC (Pierrot-Deseilligny et al., 2004; Muri and Nyffeler, 2008), the FEF could still play a role, revealed under specific cognitive conditions. In that vein, the FEF has been hypothesized to play a context-dependent modulatory influence over different cortical and subcortical structures involved in different categories of reflexive saccades. Such role could be revealed by switching cost or benefit when alternating between gap and overlap pro-saccades (Vernet et al., 2009). The role of FEF in reflexive saccade inhibition remains controversial, the DLPFC being a more likely candidate to control such inhibition (Pierrot-Deseilligny et al., 2004; Muri and Nyffeler, 2008). Finally, the FEF (together with the DLPFC and other subcortical structures) is more commonly thought as a controller for voluntary saccades such as predictive, memory-guided and anti-saccades

**Table 3 | Types of eye movements and experimental paradigms to elicit them.**


#### **Table 4 | Effects of FEF lesions on eye movements.**




*(Continued)*

**Table 5 | Continued**


(Pierrot-Deseilligny et al., 2004; Muri and Nyffeler, 2008). In addition, the FEF is involved in the computation of the amplitude of all types of eye movements.

Despite their undeniable value, several aspects limit the strength of the conclusions that can be drawn from lesion studies. First, lesions are rarely limited to the FEF, making it difficult to isolate the specific involvement of the FEF in the observed deficits. Second, different deficits might be observed during the acute and chronic phase following the lesions. Transient hypoperfusion of areas connected to the damaged area, a phenomenon known as diaschisis or, on the contrary, complex plastic reorganization within the impaired network, render the role of the FEF difficult to isolate from the role of the entire network. Other cortical and subcortical areas, or the contralesional FEF, seem to play an important role in developing compensatory mechanisms (for a review see Muri and Nyffeler, 2008). In monkey studies, in which more spatially precise transient inactivation or lesions can be performed, acutely observed deficits (Sommer and Tehovnik, 1997; Dias and Segraves, 1999) rapidly disappeared, except for complex tasks such as memory-guided saccades or saccades toward flashed targets, or if lesions to the FEF were combined with lesions to other areas (for a review see Tehovnik et al., 2000; Muri and Nyffeler, 2008).

#### *TMS studies*

The most commonly reported effect of TMS over the FEF during a saccadic task is a modulation of its preparation latency. Because of the alerting effect linked to the clicking sound and taping sensation associated with the coil discharge, it is known that TMS can have unspecific (i.e., not related to the effects of the electrical currents induced on brain tissue) effects on reaction times and eye movement latencies. Thus, shorter latencies could be related to crossmodal facilitation, whereas longer latencies could result from the participants waiting for TMS discharge as for a "go" signal. Thus, it is important to ensure that the effects on latencies are either stronger or in the reverse direction than the effects obtained in a control condition, such as sham stimulation or the active stimulation of a control brain area unrelated to saccadic control or execution. Using such cautionary measures, TMS over FEF has been shown to modulate the latency of different types of saccades.

TMS studies exploring the role of FEF in eye movements are reported in **Table 5**. As with patients' studies, whether TMS over FEF can delay reflexive saccades toward suddenly appearing visual targets remains unclear and most of the effects on latency modulations have been shown on pro-saccades involving some degree of voluntary or intentional component. In anti-saccade modulations, whether TMS stimulation of the FEF disturbs the suppression of the reflexive pro-saccade or the preparation of the voluntary anti-saccade or both is not entirely clear. In general, TMS is believed to interfere with several stages in the execution of saccades, including the perceptual analysis of the cues or targets and the motor preparation (burst signal). Occasional facilitatory effects on saccade latency have been attributed to suppression of fixation activity (within the FEF or the FEF projections to the SC). While most reports demonstrated, in accordance with microstimulation studies, effects on contralateral saccades, some studies demonstrate ipsilateral or bilateral effects that could be related to a modulation of fixation cells activity or to transcallosal modulation of both FEFs. Interestingly, TMS can modulate the latency of several eye movements performed in 3D space. Finally, the FEF is not only involved in fixation, fixation release and the triggering of voluntary eye movements but also contributes to the computation of eye movements dynamics (gain, velocity).

In conclusion, non-invasive neurostimulation studies employing TMS largely confirmed, in healthy humans and with higher spatial and temporal resolution, the insights drawn from patient studies. The unquestioned role of the FEF in the triggering of voluntary eye movements as well as the still controversial role of this region in reflexive movement inhibition and initiation is reminiscent of the blurred frontiers between reflexive and voluntary movements and of the importance of entire oculomotor networks for the control of eye movements, in which the relative contribution of each node is modulated by the cognitive context. The rest of this Section on the Role of FEF will explore how the FEF is involved in a very diverse set of higher cognitive functions (see also **Table 6** reporting TMS studies on these topics).

#### **VISUAL ACTIVITY AND SALIENCY MAP WITHIN THE FEF**

The FEF encodes visual signals and is believed to participate in the visuo-motor transformation for the preparation of eye movements, as suggested by the influence of FEF on the accuracy of eye movements (see effects of lesions and TMS on other eye movement's parameters in **Tables 4**, **5**). Beyond this contribution, the FEF can be considered as a visual area in itself, with early visual-evoked responses reported in anesthetized animals, peaking even before activity reaches V2 or V4 (Schmolesky et al., 1998). Moreover, the projections from the FEF to V4 could be characterized as feed-forward connections, i.e., going from lower to higher hierarchical levels (Barone et al., 2000). Wurtz and Mohler (1976) reported that some of the visual cells within the FEF displayed an enhanced response to a visual stimulus when a saccade was made toward the receptive field rather than away from it. Such selective enhancement would demonstrate the ability of the FEF (and similarly, also that of the SC) to evaluate stimulus significance and use such information for saccade preparation. Although there is a clear relationship between visual and movement properties of the FEF in terms of spatial selection, there is also some degree of dissociation. Bruce and Goldberg (1985) described in the FEF a continuum of visuo-motor cells, from purely visual to purely motor cells, the latter cells being less sharply tuned to direction and amplitude than the former cells, and with visuo-motor cells showing intermediate tuning. In humans, Blanke et al. (1999), recording with intracranial electrodes visual-evoked potentials in epileptic patients, showed strong visual responses for contralateral visual stimuli (consistent with the direction of the electrically-elicited eye movements) but also responses of lower amplitude after ipsilateral visual stimulation.

The visual activity encoded within the FEF has been primarily related to the computing of a saliency map, where neural activity codes for the location of a behaviorally relevant target displayed among distractors during a typical visual search task (for a review see Schall and Bichot, 1998; Thompson and Bichot, 2005). There would be a gradual suppression of distractor-related activity paralleled by an enhancement of target-related activity. Saccades are generally performed toward the "winner" of this saliency map. However, similar computations are performed even when no saccades are required or when a saccade should be performed away from the ultimately selected "winner" target. Indeed, in a go/no go visual search task, although visual response within the FEF is enhanced when the saccade is executed (go trials), discrimination of the target occurs within similar timing in both go and no go trials (Thompson et al., 1997). There would be an early (around 50 ms) non-discriminative visual response within the FEF followed by a later (around 100–150 ms) discriminative selection of the target among distractor regardless of its visual features (Thompson et al., 1996; Thompson and Bichot, 2005), but even the early response can show discriminative properties in experienced animals (Bichot et al., 1996). When saccades are triggered toward the target, the variability in saccade latencies is poorly correlated with the speed of discrimination of the FEF cells and seems to be rather related to distinct motor preparation stages (Thompson et al., 1996).

Walker et al. (2009) brought direct causal evidence in humans that the FEF might be participating in the elaboration of a saliency map for the selection of a target of an upcoming saccade. Indeed, when a competing visual distractor appeared in the same direction as the saccade goal but at unpredictable locations, saccade trajectories deviated away from the distractor. The magnitude of this distractor-related deviation of saccade trajectory was increased by single-pulse TMS over the right FEF. The interpretation is that stimulation of the FEF might have disrupted the process of enhancing target salience or could have increased the inhibition associated with the distractor.

#### **THE FEF AT THE HEART OF THE COUPLING BETWEEN ATTENTION AND EYE MOVEMENTS?**

The saliency map described above could reflect the deployment of visuo-spatial attention. The premotor theory of attention postulates that orientation of spatial attention derives from the same mechanisms dedicated to action: attention is oriented to a given point in space when the oculomotor program for moving the eyes to that point is ready to be executed (Rizzolatti et al., 1987). In this perspective, FEF "visual" activation could be attributed to the preparation of saccade programs, which may or may not be overtly executed, rather than to the visual analytic processes in the FEF.

Many behavioral and neurophysiological studies support this theory, according to which covert attention shifts without eye movements, conceived as a specific and distinct process with a mechanism of its own, might simply be an artificial separation of otherwise unified underlying processes. Among the many behavioral pieces of evidences in accordance with the premotor theory of attention, one could cite the enhanced visual discrimination


#### **Table 6 | Effects of TMS over the FEF on visuo-spatial attention, visual awareness and perceptual modulation.**

*(Continued)*

**Table 6 | Continued**


performance when a discrimination stimulus and a saccade target converge to the same object whereas it declines steeply when they refer to items at different locations, arguing against the ability to direct visual attention to one location while simultaneously preparing a saccade toward another location (Deubel and Schneider, 1996). Neuroimaging studies often find similar activations, including in the FEF, for eye movements and attentional shifts, and a remarkable level of overlap of the underlying circuits of these operations, as summarized in a meta-analysis on PET and fMRI studies (Grosbras et al., 2005). Interestingly, the involvement of the FEF in contralateral attention shifts would be particularly marked when participants have to overtly respond to a target, for instance with a manual response (Corbetta et al., 1993) or when the attentional task is particularly demanding (Donner et al., 2000).

However, there is also evidence against a strict interpretation of the premotor theory of attention. For instance, TMS during saccade preparation was able to modulate discrimination performance at the target location: while TMS over the intraparietal sulcus (IPS) ipsilateral to the saccade's direction increased general performance, non-invasive stimulation over the FEF contralateral to the saccade's direction specifically decreased or enhanced discrimination on the target location depending on the exact stimulation parameters (Neggers et al., 2007; Van Ettinger-Veenstra et al., 2009). Thus, the FEF plays a role in mediating the coupling between visuo-spatial attention and eye movements and such coupling can be modulated by TMS (Neggers et al., 2007; Van Ettinger-Veenstra et al., 2009). Other arguments against a motor preparation toward the target location to which attention is oriented can be found in microstimulation experiments with monkeys (Juan et al., 2004) or TMS experiments in humans (Juan et al., 2008). In the first study (Juan et al., 2004), monkeys had to perform a visual search and a saccade toward (pro-saccade) or away from (anti-saccade) a visual target depending on its orientation. Microstimulation of the FEF at variable timings after target onset evoked, in anti-saccade trials, saccades progressively toward the endpoint of the correct saccades but never toward the visual target. Using a similar task in humans, Juan et al. (2008) showed that double-pulse TMS over the right FEF can delay saccade latencies in two distinct time-windows: an early window (40–80 ms after target onset) in which the delay in pro-saccades was interpreted as a disruption of the visual stimulus processing and also a late window (200–160 ms before the expected saccades) in which a delay in pro- and anti-saccades was interpreted as a disruption of saccade preparation.

#### **CONTEXT-DEPENDENT ROLE OF FEF DURING VISUAL SEARCH**

Whether or not eye movement preparation is strictly linked to attention orientation does not question the involvement of the FEF in visual discrimination performance, either directly or indirectly though its massive set of anatomical projections toward the visual cortex. Several TMS studies in humans have been designed to accurately describe the role of the left and right FEF in covert voluntary attentional orienting and visual discrimination performance. For instance, Muggleton et al. (2003) showed that rTMS at 10 Hz for 500 ms over the right FEF during the presentation of a search array disrupted visual search. These authors showed that a decrease of the visual sensitivity explained by a higher number of false positives (i.e., incorrect detections reported by participants when the target was absent) and attributed to a reduced ability to process the items. Interestingly, only specific subtypes of visual search impaired by the stimulation, such as conjunction search was impaired (i.e., when the target shares the same color than about half of the distractors and the same orientation than the remaining distractors) and, to a lesser extent, interleaved feature search (i.e., when the color of target and distractors is randomly attributed at each trial). On the contrary, rTMS had no effect on constant feature search (i.e., when the target and distractors always look the same across trials). The authors concluded that the right FEF is particularly important for visual search when the visual target is neither salient nor predictable. Using double-pulse TMS paradigms, such findings were confirmed for an early time window of up to 80 ms after search array onset, i.e., much earlier than the involvement of the PPC in visual search (O'Shea et al., 2004; Kalla et al., 2008).

Using similar visual search paradigms, the role of FEF in visual priming (form of implicit memory that facilitates the detection of a target that shares common features with a recently inspected search target) or, on the contrary in switch detection has also been addressed in TMS approaches. Indeed, fMRI experiments reported a suppression of BOLD response in fronto-parietal networks, including the FEF, during simultaneous color and location repetition (Kristjansson et al., 2007). Non-invasive brain stimulation studies showed that 10 Hz rTMS patterns for 500 ms over the left (but not the right) FEF disrupted spatial priming, as measured by increased reaction times, when applied during the presentation of the search array (O'Shea et al., 2007) or during the inter-trial interval (Campana et al., 2007). This result suggested that the memory trace is probably distributed through visual and oculomotor networks typically required for those behaviors and that the FEF would be an area of convergence and integration during the preparation of an overt response (O'Shea et al., 2007). Finally, the left FEF would also be involved in the ability to detect a color switch (or select a new target) as identical rTMS patterns delivered to the left FEF applied in-between trials increased switching costs by slowing down the response time for switch trials (Muggleton et al., 2010).

However, the right/left hemisphere frontal asymmetries described above are questioned by other studies showing that rTMS at 10 Hz for 500 ms, over the right but not the left FEF, from the beginning of an array onset, disrupts spatial priming and that similar rTMS over both right and left FEF increases reaction time when the target position is random (Lane et al., 2012). Interestingly, the same team also demonstrated that such involvement of the right FEF is independent of the depth (near vs. far space) at which the task is performed, whereas the right PPC would be involved in near space and right ventral occipital cortex in far space (Lane et al., 2013).

Finally, it is possible that the FEF is more directly involved in spatial memory, in particular in trans-saccadic memory. Indeed, Prime et al. (2010) showed a decrease of the number of items participants could remember when left or right FEF were disrupted around saccadic time. Such effect could however be related to the stimulation of a spatial working memory area that has been identified just rostral to the FEF (Courtney et al., 1998).

#### **TOP-DOWN MODULATION OF VISUAL AREAS**

Several studies suggested that the contributions of the FEF to discrimination performance are mediated by its output projections to the visual cortex. Indeed, electrophysiological evidence in both animals and humans demonstrated a relation between activity within the FEF and excitability of occipital brain areas. Moore and Armstrong (2003) showed that the intracortical stimulation of the FEF in monkeys at current intensities below those required to evoke saccades (i.e. subthreshold stimulation), enhanced visual responses in visual area V4. Such enhancements were retinotopically specific. If the endpoint of the saccade evoked by suprathreshold stimulation of the FEF overlapped with the receptive field of a V4 cell, subthreshold FEF stimulation enhanced this V4 cell's visual responses. This type of top-down modulation of visual cortex excitability could explain earlier findings in non-human primates consisting in enhanced perception (decreased threshold for detecting a luminance change) of peripheral visual stimulus after subthreshold FEF stimulation, only when the visual stimulus was displayed within the "motor field" of the stimulated FEF (Moore and Fallah, 2001).

Although TMS cannot reach the spatial resolution required to target neural populations within the FEF subtending specific visual or motor fields, several studies in healthy humans, combining TMS with EEG (Taylor et al., 2007), TMS with fMRI (Ruff et al., 2006) or employing double coil TMS and psychophysics, showed similar top-down influence of the FEF on visual areas and visual performance. Short 5-pulse trains of 10 Hz rTMS applied over the right FEF during a cueing period of a covert orienting task modulated attention-related ongoing EEG posterior potentials before visual stimulation, as well as the potentials evoked by the visual stimulus (Taylor et al., 2007). Similar short 5-pulse trains of 9 Hz TMS over the right FEF modulated the BOLD activity recorded with fMRI within visual areas V1-V4 led to activity increases for retinotopic representations of the peripheral visual fields combined with activity decreases of central retinotopic locations (Ruff et al., 2006). A follow up experiment showed that TMS over the right FEF enhanced perceived contrast for peripheral relative to central visual stimuli (Ruff et al., 2006), hence proving that such activity modulation was behaviorally relevant. Finally, the double-coil TMS technique can be used to simultaneously induce activity within the left or right FEF and measure the excitability of MT/V5. Stimulation of the FEF 20–40 ms prior to stimulation of MT/V5 decreased the intensity of MT/V5 stimulation required to elicit phosphenes, demonstrating that the FEF has a direct modulatory effect on the excitability of this motion visual area (Silvanto et al., 2006).

#### **MODULATION OF VISUAL PERFORMANCE AND AWARENESS**

In line with animal and human studies showing respectively, enhanced perception and increased activity in visual areas following FEF stimulation, several reports have also shown that TMS over the left and right FEF was able to speed-up discrimination and/or increase detection and visual awareness (Grosbras and Paus, 2002, 2003; Chanes et al., 2012). Grosbras and Paus (2002) reported that a TMS pulse delivered to the left or right FEF 53 ms prior to target onset could decrease reaction time in a forced-choice discrimination task. These same authors showed that a TMS pulse delivered to the left or right FEF 40 ms before the onset of a masked target could also increase sensitivity in a visual detection task (Grosbras and Paus, 2003). Similarly, Chanes et al. (2012) provided evidence showing that a TMS pulse delivered to the right FEF 80 ms before a low-contrast target could increase visual perceptual sensitivity in a detection task. These studies have shown that right FEF stimulation generally leads to bilateral effects whereas left FEF stimulation results in an increase of performance solely for stimuli presented in the contralateral visual hemifield (Grosbras and Paus, 2002, 2003; Chanes et al., 2012). In addition, interactions between TMS effects and the manipulation of visuo-spatial attentional orienting in space before target presentation have been found in studies that combined a strategy to modulate attentional processes by means of spatially informative visual cues and by means of noninvasive neurostimulation delivered to specific cortical regions. More specifically, the increase of performance after right FEF stimulation reported in the above-mentioned studies occurred specifically for validly cued (or attended) locations, and also following spatially neutral cues, but not for unattended locations following invalid cueing (Grosbras and Paus, 2002; Chanes et al., 2012).

Enhancement of perception may result from a global increase of background activity, drifting closer to a perceptual threshold, hence allowing any incoming weak signal to reach it more easily; in addition to this global injection of activity, TMS outcomes are also highly dependent on the state of the targeted regions and their mixed populations of neurons. Accordingly, TMS may selectively boost specific clusters of neurons according to their level of activity (O'Shea and Walsh, 2004). In this context, prior reports have suggested that the visual performance and awareness enhancement occurred directly by changing activity in FEF (manipulating genuine processes purported by this frontal region) or indirectly *via* connections between the FEF and visual regions modulating the input gain of incoming visual signals (Grosbras and Paus, 2003). Interestingly, interindividual differences in the direction and magnitude of the TMS driven facilitatory effects reported in Chanes et al. (2012) correlated significantly with the probability of anatomical connection between the FEF and the SC estimated by means of white matter probabilistic tractography. Such result suggests a key role for white matter connectivity between the stimulated area and other key brain structures to explain at the network level the strength of TMS modulatory influences on visual performance (Quentin et al., 2013).

It was also suggested that TMS effects could actually enhance perception in a way similar to what occurs naturally during cover shifts of attention (Grosbras and Paus, 2003). Following that line of thought, Chanes et al. (2013) conceived an experiment in which TMS was used as a way to emulate activity that would mimic neurophysiological spatio-temporal patterns signaling processes of attentional orienting. This study was based on a prior report by Buschman and Miller (2007) demonstrating, in non-human primates, high-beta (∼30 Hz) and gamma (∼50 Hz) fronto-parietal synchronizations subtending top-down and bottom-up attentional processes, respectively. By using short trains of stimulation at those same frequencies, Chanes et al. (2013) observed an increase of perceptual sensitivity during a low-contrast target detection task following right-FEF stimulation at 30 Hz, which would be consistent with an increase of endogenous attention and/or facilitated access to visual consciousness. Interestingly, the strength of the individual TMS improvements correlated significantly with the volume of the first branch of the superior longitudinal fasciculus, which links the stimulated FEF with areas of the posterior parietal cortex within and in the vicinity of the IPS, supporting the idea of a frequency-specific fronto-parietal synchronization induced by rhythmic TMS subtending visual performance ameliorations (Quentin et al., 2014). Moreover, stimulation of the right FEF at 50 Hz induced a relaxation of response criterion, as if sensory evidence in favor of target presence were increased whether or not the target was actually present (Chanes et al., 2013). This result provides support for a frequency-based multiplexing of two distinct processes, such as visual sensitivity and response criterion with bearing on visual performance, emerging from neuronal resources within the same area. Whereas short trains of rTMS have been widely used to drive stronger behavioral effects (see e.g., Smith et al., 2005), they can also be applied, as in Chanes et al. (2013), as a novel way to manipulate rhythmic brain activity, in line with evidence of oscillatory entrainment with such technique (Thut et al., 2011).

To reconcile these studies showing improvements in visual perception and awareness with TMS of the FEF with previously cited reports showing impairment of visual discrimination during visual search tasks (e.g., Muggleton et al., 2003; O'Shea et al., 2004; Kalla et al., 2008), it should be reminded that TMS lacks the spatial resolution to selectively enhance perception in one particular area of the perceptual space. Thus in visual search paradigms, distractors might benefit as much as targets from TMS-driven visual enhancement, decreasing the relative benefit for the latter and leading to perceptual impairments instead of enhancements.

Along the same lines, in addition to increased visual performance at cued locations, TMS over the FEF should disrupt the inhibition of stimulus processing at unattended locations. This hypothesis has been confirmed by the study by Smith et al. (2005), that in agreement with this notion, showed that in a visual detection task, short trains of rTMS at 20 Hz for 200 ms starting 50 ms before cue onset (and not around the timing of target onset as in the previously cited studies) over the left FEF were able to decrease the reaction time cost of invalid cueing before contralateral targets. Such disruption of the inhibition for unattended locations could also explain the results reported by Ro et al. (2003) who showed that single TMS pulses, delivered over the right FEF, showing that single 600 ms after the cue and 150 ms prior to target onset, decreased the inhibition of return phenomenon. This well-known attentional process consists in a worsening of visual performance at locations that had been cued a certain interval of time preceding target onset. It is probably caused by a disengagement of attention and is thought to prevent the re-exploration of an already scrutinized region of the space. The above-mentioned studies support the notion that increased visual detection performance at unattended spatial locations could result from TMS interfering with active mechanisms of inhibition and exploration suppression of unattended spatial locations subtended by the FEF.

To summarize, the FEF cannot be only conceived as an area important for preparing and triggering eye movements but also as an essential region contributing to cognitive processes such as attentional orienting, visual awareness, conscious access, perceptual performance, and decision making. However, as mentioned before, these processes are probably mediated by activity within largely distributed cortico-cortical and cortico-subcortical networks. In particular, fronto-parietal systems are particularly relevant, and effects similar to those of the FEF reviewed in this paper, have been found in specific PPC regions such as the IPS (Chica et al., 2011; Bourgeois et al., 2013a,b). If there is full agreement on the fact that dorsal frontal and posterior parietal areas operate commonly and in synchrony in attentional and visual performance modulation processes, other studies emphasize the differences between the contributions of these two regions. For instance, whereas the dominance of the right PPC in attentional orienting tasks is well known, inter-hemispheric asymmetry is less evident with regards to the contributions of the FEF (Gitelman et al., 1999), and such aspect might prove highly task dependent. Moreover, the FEF could be more involved than the IPS in intentional behavior when overt responses are required (Corbetta et al., 1993). Future studies will be necessary to further understand the common and distinctive role of these two highly interconnected areas.

Finally, it should be mentioned that although the FEF is involved in both eye movements and visual cognition, few studies have explicitly explored simultaneously the combined role of the FEF in both types of function (with the exception of the studies reviewed in a prior section addressing the role of the FEF in the coupling between attention and eye movements). The cognitive context modulates the role of the FEF in eye movements but direct report of conscious perception, for instance, is not performed in eye movements studies. Conversely, studies on cognition rarely explore eye movements (even if correct visual fixation is often assessed with eye-trackers). In future studies, exploration of microsaccades and other fixation eye movements (Martinez-Conde et al., 2013) might shed new light on the relevance of the experimental dissociation between eye movements and cognition.

#### **IMPROVING VISUAL PERCEPTION AND AWARENESS**

In the last part of this review, we will briefly address some links between FEF activity, eye movements and visuo-spatial awareness. Such evidence will allow us to elaborate on the rational behind the potential use of non-invasive brain stimulation and eye movement training in 3D space for the treatment and rehabilitation of visuo-spatial disorders. As an example, we will focus on the case of hemi-spatial visual neglect, often suffered by patients after right hemisphere damage.

#### **DEFICITS OF VISUO-SPATIAL AWARENESS: THE CASE OF HEMISPATIAL NEGLECT**

As reviewed above, the FEF is not only a key node contributing to the planning and execution of eye movements but it is also involved in attentional orienting and several aspects of visual cognition. Surprisingly however, a large majority of the FEF lesion studies in non-human and human primates focused on the consequences of frontal damage on oculomotor deficits, neglecting the exploration of other behavioral consequences. Nevertheless, it has been shown that lesions damaging right attentional networks can often induce attentional orienting and visual awareness disorders such as visuo-spatial neglect. This condition is a highly impairing syndrome consisting in an inability to orient attention to regions of the contralesional space and thus become aware of sensory stimuli presented herein. It is common after stroke lesions impacting cortical or subcortical regions, particularly in the right hemisphere. In a multicenter study developed in a cohort of 1281 acute stroke patients, signs of visuo-spatial neglect occurred in 43% of right brain-injured patients and also in 20% of left brain-damaged patients. Apparent spontaneous recovery of these deficits seems often to occur, but does not necessarily eliminate all signs and deficits, particularly visual extinction, a fact that becomes evident when more challenging and robust-to-learning computer-based tasks are employed instead of paper and pencil tasks to evaluate patient status. Indeed, at 3 months, signs of moderate neglect are still present in 17% of right brain-injured patients and 5% of left brain-injured patients (Ringman et al., 2004). Neglect interferes with the rehabilitation of deficits in other domains, such as motor and sensory, that can also be present in such patients and, if it endures, it can lead to poor clinical recovery outcomes and preclude a reintegration to normal or adapted life. However, despite considerable therapeutic advances in behavioral, sensorial and pharmacological treatments, many patients remain enduringly impaired after rehabilitation (Fierro et al., 2006).

One of the most influential hypotheses to understand visuospatial neglect suggests the existence of an impairment in the balance between the orienting attentional bias of each hemisphere toward the contralateral hemispace (Kinsbourne, 1970). This explanation has received support from animal studies (Sprague, 1966; Rizzolatti et al., 1983). In humans, a revealing single patient case revealed how a second stroke involving the left FEF a few days after a right parietal stroke was able to fully compensate the clinical signs of severe neglect induced by the first lesion (Vuilleumier et al., 1996).

#### **NON-INVASIVE BRAIN STIMULATION TO IMPROVE VISUO-SPATIAL AWARENESS**

In the context of a hemispheric imbalance, the use of non-invasive brain stimulation, namely rTMS and transcranial direct current stimulation (tDCS) has proven particularly promising. Indeed, both techniques have demonstrated efficacy in modulating transiently brain regional excitability (Nitsche et al., 2008; Rossi et al., 2009). Inhibitory stimulation over the intact hemisphere (mostly at the level of parietal areas) has shown promise in decreasing the excessive inhibition over viable regions of the damaged hemisphere and by virtue of this effect relieving neglect symptoms (e.g., Oliveri et al., 2001; Sparing et al., 2009; Cazzoli et al., 2012; Koch et al., 2012).

However, the ability of FEF-TMS to increase visual perception and awareness, at least in healthy subjects on a trial-by-trial basis (Grosbras and Paus, 2002, 2003; Chanes et al., 2012) could revive the interest in targeting this node of the attentional network for rehabilitation purposes. Indeed, neglect is a "network" impairment: there is a striking variability of areas whose lesions results in neglect. Temporal, parietal or frontal cortical lesions or subcortical lesions, or combined lesions of these areas are the most likely to induce enduring neglect 3 months post-stroke (Ringman et al., 2004). Aside from the impairments resulting from the lesions themselves, neglect symptoms could also arise from diaschisis (i.e., abolition of neural activity in areas that are distant but anatomically connected to the lesioned area) and disconnections/hypoperfusion affecting the whole fronto-parietal network. Recent evidence from an fMRI study suggests that functional connectivity in damage and also intact fronto-parietal attentional orienting networks is impaired in neglect patients (He et al., 2007) and the disconnection theory should be strongly considered to explain the pathophysiology of neglect (Doricchi et al., 2008). A recent explanatory model of visuospatial neglect (Corbetta and Shulman, 2011) suggests that neglect primarily arises from damage to the right hemisphere-dominant non-spatial ventral attention network. This ventral network would be interacting with the dorsal fronto-parietal network (encompassing the FEF and the IPS) that controls spatial attention. Thus, structural damage to the right ventral network would result in functional resting and task-related activity asymmetries in the dorsal network, leading to the typical lateralized attention deficit of neglect syndrome.

The existence of fronto-parietal synchronization at specific frequencies associated with attentional processes (Buschman and Miller, 2007) and the possibility to inject such rhythms with TMS over the FEF to increase visual awareness (Chanes et al., 2013) also open new perspectives for treatment. In addition, MEG experiments in 5 right stroke neglect patients revealed that target omissions correlated with a build-up of low beta activity in left frontal locations before target presentation (Rastelli et al., 2013). Thus, neglect occurrence seems to arise from abnormal oscillatory activity and consequently could be manipulated with rhythmic TMS. Whereas traditional rTMS protocols have shown complex effects on brain oscillations (Thut and Pascual-Leone, 2010; Vernet et al., 2013), the use of rhythmic TMS or Transcranial Alternate Current Stimulation (tACS) combined with EEG demonstrated a possibility to entrain physiologically relevant oscillations at a chosen frequency (Thut et al., 2011; Helfrich et al., 2014). Similarly, the use of double-coil (bifocal) stimulation is a promising tool to modulate synchronization between distant brain areas (Plewnia et al., 2008). Future studies are needed to explore the possibility to induce such modulations in clinical populations lasting long enough to be clinically relevant.

#### **PERFORMING 3D EYE MOVEMENTS TO IMPROVE VISUO-SPATIAL AWARENESS**

Another topic of interest for the treatment of visual awareness disorders would consist in further exploring the link between eye movements and conscious perception. Indeed, eye movements training can be natural way to activate the FEF and other areas of attentional networks. The resulting plasticity might in turn improve visual awareness. Vergence is particularly fragile and subject to aging, fatigue and neurological insults (Scheiman et al., 2005a,b,c; Yang et al., 2010) and hence reeducation of eye movements in depth might promote better visual navigation in space and as a consequence improve conscious perception.

The link between exploratory movements performed in 3D space and the gathering of visuo-spatial information goes beyond a simple sharing of brain resources. Indeed, efferent copies of vergence movements, proprioception on convergence state (Priot et al., 2012), or simply disparity indices on which vergence movement can be programmed (Ziegler and Hess, 1997) are all candidates to be involved in our ability to assess depth. Conversely, our understanding of the 3D space can trigger movements in accordance with our perception. However, there are striking examples of dissociations between our perception and action as for instance the famous Müller-Lyer illusion, in which erroneous judgments about an object size and correct manual seizing movement may coexist. This type of experience led to the traditional dissociation between vision for perception and vision for action (Milner and Goodale, 2008). However, such dissociation is controversial and might be an experimental artifact. Indeed, without any visual feedback, hand and eye movements (and consequently the "vision for action") can also be affected by the illusion (Bruno and Franz, 2009; Bruno et al., 2010). Similarly, in illusions where the perceived depth is different from the actual depth, the vergence movements are sometimes subject to the depth cues of the physical world (Wismeijer et al., 2008), and sometimes to the illusory percept (Sheliga and Miles, 2003). An interesting interpretation is that there are two types of convergence: a fast one to serve action and a slower one, which would allow the construction of a conscious percept (Wagner et al., 2009).

Besides the deficit of awareness in contralesional space, neglect patients seem to also suffer from problems that are specific to the depth at which tasks are performed: neglect symptoms might be sometimes more severe in the near space or in the far space (see Aimola et al., 2012 for a review). Probably, the evaluation paradigms and tasks, the effector used, and the location and extent of the right hemisphere damage could explain the differences found by different studies with regards to this issue (see Aimola et al., 2012 for a discussion). Experimental evidence supports the notion that the representations of peripersonal and extrapersonal spaces are subtended respectively, by rather dorsal or ventral regions within fronto-parietal systems (Aimola et al., 2012). Hence, performance dissociation between tasks performed in near and far space could reveal the specificity of the disconnection patterns between areas required for the task and areas devoted to monitor specific portions of the space (Weiss et al., 2000). In any case, it is possible that neglect patients show deficits in visual navigation in depth, which could aggravate their difficulties in exploring fronto-parallel space for certain depths. Oculomotor training focusing on vergence movements in depth (Jainta et al., 2011) could be another interesting path to explore for a more effective rehabilitation of awareness disorders.

## **CONCLUSION**

In this review, we focused on the role and localization of the human FEF, a cortical area which is part of highly distributed saccadic and visuo-spatial networks with important bearing on the control of eye movements in 3D space and contributing importantly to several aspects of attentional and visual cognition. A particular emphasis was placed on TMS studies, which have allowed a successful causal exploration of the contributions of this frontal region. We provided evidence that the results of such studies with regards to their ability to map FEF cortical location are not necessarily similar to those of other mapping techniques employed in neuroscience, such as lesion studies, microstimulation, intracranial recordings, PET, fMRI, and EEG/MEG in both humans and non-human primates. Similarities and discrepancies across the results provided by different techniques, the use of different paradigms and/or experimental models were presented and discussed. Finally, we speculated on the posibility to manipulate FEF activity with non-invasive neurostimulation and oculomotor training in order to improve visuo-spatial awareness in 3D space in healthy population and to promote functional recovery in stroke patients.

## **ACKNOWLEDGMENTS**

Marine Vernet and Romain Quentin were supported by the "*Fondation pour la Recherche Médicale.*" The activities of the laboratory of Dr. Valero-Cabré are supported by grants from eraNET NEURON Beyondvis by ANR, FP6, and IHU-A-Translationnel.

## **REFERENCES**


right hemisphere damage. *Cortex* 49, 1989–1996. doi: 10.1016/j.cortex.2013. 04.001


usefulness in cognitive and clinical neuroscience. *Brain Topogr.* 22, 219–232. doi: 10.1007/s10548-009-0115-4


**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 May 2014; accepted: 01 August 2014; published online: 22 August 2014. Citation: Vernet M, Quentin R, Chanes L, Mitsumasu A and Valero-Cabré A (2014) Frontal eye field, where art thou? Anatomy, function, and non-invasive manipulation of frontal regions involved in eye movements and associated cognitive operations. Front. Integr. Neurosci. 8:66. doi: 10.3389/fnint.2014.00066*

*This article was submitted to the journal Frontiers in Integrative Neuroscience. Copyright © 2014 Vernet, Quentin, Chanes, Mitsumasu and Valero-Cabré. 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.*

## Corrigendum: Frontal eye field, where art thou? Anatomy, function, and non-invasive manipulation of frontal regions involved in eye movements and associated cognitive operations

## *Marine Vernet\*, Romain Quentin , Lorena Chanes , Andres Mitsumasu and Antoni Valero-Cabre*

*Centre National de la Recherche Scientifique, Institut du Cerveau et de la Moelle Epinière, Paris, France \*Correspondence: marine.vernet@gmail.com*

#### *Edited by:*

*Olivier A. Coubard, CNS-Fed, France*

#### *Reviewed by:*

*Albino J. Oliveira-Maia, Champalimaud Foundation, Portugal Thomas Nyffeler, Bern University Hospital and University of Bern, Switzerland*

**Keywords: FEF, brain mapping, transcranial magnetic stimulation, visual performance, visuo-spatial attention, 3D vision, visual awareness, plasticity rehabilitation**

#### **A corrigendum on**

**Frontal eye field, where art thou? Anatomy, function, and non-invasive manipulation of frontal regions involved in eye movements and associated cognitive operations**

*by Vernet, M., Quentin, R., Chanes, L., Mitsumasu, A., and Valero-Cabré, A. (2014). Front. Integr. Neurosci. 8:66. doi: 10.3389/fnint.2014.00066*

A few errors were introduced during the proof reading process, which the authors wish to correct with this corrigendum.

**A few sentences may lead to erroneous interpretations and should be reformulated**

#### **Page 5, line 3**


#### **Page 10, column "Interpretation," square 10**


*and the execution of saccades (including perceptual analysis of the go signal)*

#### **Page 14, column "Effects," square 2**

Incorrect: *contralaterally (i.e., for left targets) after left FEF stimulation* Correct: *contralaterally (i.e., for right targets) after left FEF stimulation*

**Page 15, column 2, second paragraph, line 3**

	- Correct: *several studies in healthy humans, combining psychophysics with TMS and EEG (Taylor et al., 2007), TMS and fMRI (Ruff et al., 2006) or double-coil TMS (Silvanto et al., 2006)*

#### **Page 15, last paragraph, line 1:**

Incorrect: *In line with animal and human studies showing respectively, enhanced perception and increased activity in visual areas following FEF stimulation*

Correct: *In line with animal and human studies showing enhanced perception and increased activity in visual areas following FEF stimulation*

#### **A few typos, grammatical mistakes and erroneous extra-words should be removed from the manuscript**

#### **Page 2, Part "Localization of FEF," line 20:**

Incorrect: *Overall, it is still not entirely clear whether the reported inter-species differences in FEF location can be related to genuine anatomical differences between non-human primates and humans, caused by the use of different mapping methods or they simply reflect interindividual differences*

Correct: *Overall, it is still not entirely clear whether the reported interspecies differences in FEF location can be related to genuine anatomical differences between non-human primates and humans, or caused by the use of different mapping methods, or whether they simply reflect interindividual differences*

#### **Page 8, column 2, line 1**

Incorrect: *The general pictures emerging from this literature is [. . . ]*


#### **Page 11, last paragraph, lines 4 and 6**


#### **Page 13, column "Effects," square 7:**

Incorrect: *TMS stimulation* Correct: *TMS*

#### **Page 14, 3 lines before the end:**


#### **Page 15, column 2, paragraph 2, line 11**


### **Page 16, last paragraph, line 3**


#### **Page 16, last paragraph, line 10**


#### **REFERENCES**

Ro, T., Farne, A., and Chang, E. (2003). Inhibition of return and the human frontal eye fields. *Exp. Brain Res*. 150, 290–296. doi: 10.1007/s00221-003- 1470-0


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

*Received: 17 September 2014; accepted: 17 October 2014; published online: 11 November 2014.*

*Citation: Vernet M, Quentin R, Chanes L, Mitsumasu A and Valero-Cabre A (2014) Corrigendum: Frontal eye field, where art thou? Anatomy, function, and noninvasive manipulation of frontal regions involved in eye movements and associated cognitive operations. Front. Integr. Neurosci. 8:88. doi: 10.3389/fnint.2014.00088 This article was submitted to the journal Frontiers in Integrative Neuroscience.*

*Copyright © 2014 Vernet, Quentin, Chanes, Mitsumasu and Valero-Cabre. 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.*

## What saccadic eye movements tell us about TMS-induced neuromodulation of the DLPFC and mood changes: a pilot study in bipolar disorders

#### **Lysianne Beynel <sup>1</sup>\*, Alan Chauvin<sup>1</sup> , Nathalie Guyader <sup>2</sup> , Sylvain Harquel 1,3 , David Szekely<sup>4</sup> , Thierry Bougerol <sup>4</sup> and Christian Marendaz <sup>1</sup>\***

<sup>1</sup> Department of Psychology, Laboratory of Psychology and Neurocognition, Grenoble Alpes University, Université Pierre Mendes France, Grenoble, France

<sup>2</sup> Department of Images and Signal, Grenoble Image Parole et Signal Automatique-Lab, Grenoble Alpes University, St Martin d'Héres, Grenoble, France

<sup>3</sup> Department of Psychology, IRMaGe, Grenoble Alpes University, Grenoble, France

<sup>4</sup> Department of Psychiatry and Neurology, Hospital of Grenoble, Grenoble Alpes University, La Tronche, France

#### **Edited by:**

Olivier A. Coubard, CNS-Fed, France

#### **Reviewed by:**

Hao Zhang, Pfizer Neuroscience, USA Gregory Kroliczak, Adam Mickiewicz University in Poznan, Poland

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

Lysianne Beynel and Christian Marendaz, Department of Psychology, Laboratory of Psychology and Neurocognition, Grenoble Alpes University, Université Pierre Mendes France, BP 47, F-38040 Grenoble Cedex 9, Grenoble, France e-mail: Lysianne.beynel@ upmf-grenoble.fr; Christian.Marendaz@upmfgrenoble.fr

The study assumed that the antisaccade (AS) task is a relevant psychophysical tool to assess (i) short-term neuromodulation of the dorsolateral prefrontal cortex (DLPFC) induced by intermittent theta burst stimulation (iTBS); and (ii) mood change occurring during the course of the treatment. Saccadic inhibition is known to strongly involve the DLPFC, whose neuromodulation with iTBS requires less stimulation time and lower stimulation intensity, as well as results in longer aftereffects than the conventional repetitive transcranial magnetic stimulation (rTMS). Active or sham iTBS was applied every day for 3 weeks over the left DLPFC of 12 drug-resistant bipolar depressed patients. To assess the iTBS-induced short-term neuromodulation, the saccadic task was performed just before (S1) and just after (S2) the iTBS session, the first day of each week. Mood was evaluated through Montgomery and Asberg Depression Rating Scale (MADRS) scores and the difference in scores between the beginning and the end of treatment was correlated with AS performance change between these two periods. As expected, only patients from the active group improved their performance from S1 to S2 and mood improvement was significantly correlated with AS performance improvement. In addition, the AS task also discriminated depressive bipolar patients from healthy control subjects. Therefore, the AS task could be a relevant and useful tool for clinicians to assess if the Transcranial magnetic stimulation (TMS)-induced short-term neuromodulation of the DLPFC occurs as well as a "trait vs. state" objective marker of depressive mood disorder.

**Keywords: antisaccades, DLPFC, rTMS iTBS, bipolar disorder, short-term neuromodulation, long-term neuromodulation**

#### **INTRODUCTION**

#### **rTMS TREATMENT OF DEPRESSION: NECESSITY TO ASSESS THE INDUCED NEUROMODULATION**

Transcranial magnetic stimulation (TMS) is a non-invasive technique that induces a magnetic field on the skull which changes rapidly enough to induce electrical currents in underlying cortical tissue and thus to induce a neuromodulation effect (Hallet, 2000). Repeated TMS (rTMS) has been used as a therapeutic tool for the treatment of drug-resistant mood disorders since the 1990s; patients receive a daily dose of rTMS over frontal regions for several weeks. The rationale for this treatment is that whereas a single rTMS session induces an early long-term potentiation of the targeted cortical area (short-term neuromodulation), cumulative rTMS sessions induce widespread late long-term potentiation across multiple neural circuits (Noda et al., 2013).

Meta-analyses of rTMS as a treatment for depression have been recently published (Janicak et al., 2010; Slotema et al., 2010; George and Post, 2011; Berlim et al., 2014). All these studies agree that the therapeutic efficacy of rTMS is statistically significant but affects few patients. For example, Berlim et al. (2014) showed that about 30% of depressed patients receiving active excitatory rTMS responded to the treatment compared to 10% of patients who received sham treatment (analysis based on 29 studies using randomized, double-blind and sham-controlled trials).

To improve the therapeutic efficacy of rTMS, a new protocol theta burst stimulation (TBS)—(Huang et al., 2005) has been developed. TBS has been shown not to differ from rTMS in terms of strength and direction of aftereffects (Thut and Pascual-Leone, 2010; Di Lazzaro et al., 2011) but to exert longer-lasting poststimulation effects (Thut and Pascual-Leone, 2010) with less stimulation time and lower stimulation intensity (Hinder et al., 2014). For these reasons TBS is of particular interest for clinicians, and recently some studies investigated the therapeutic effectiveness of TBS for the treatment of mood disorders, using intermittent theta burst stimulation (iTBS) that produces long-term potentiationlike effects, or continuous TBS (cTBS) that produces long-term depression-like effects. Li et al. (2014) compared cTBS, iTBS and sham stimulation. They found that depressive mood improved in all groups, with a better antidepressant effect for iTBS (40% responders) than for cTBS (25% responders) and sham (13% responders). Plewnia et al. (2014) compared TBS to sham stimulation. They found that 56% of patients receiving active TBS responded to the treatment compared to 25% of patients who received sham treatment. As a whole, these results indicate that whatever the TMS protocol, the therapeutic efficacy of rTMS is statistically significant but remains limited, and the rTMS clinical relevance is still debated (Padberg and George, 2009).

Apart from idiopathic reasons, several technical/neurophysiologic factors might account for the lack of TMS efficacy. For example, the stimulation parameters (intensity, frequency) might not be adjusted properly for every patient, the dorsolateral prefrontal cortex (DLPFC) might be identified or targeted incorrectly (Fox et al., 2012), or the TMS treatment might fail to induce short-term neuromodulation of the DLPFC in some patients, which prevents the long-term potentiation. To move forward on these issues, it is crucial to ensure that TMS-induced neuromodulation of the DLPFC effectively occurs. This question requires the development of an instrument that can objectively assess this neuromodulation. To be helpful for clinicians, this instrument has to be non-invasive and easy to use. We assumed that an oculometric task such as the antisaccade (AS) task may provide such a tool (Crevits et al., 2005; Malsert et al., 2012a,b, 2013).

### **RELEVANCE OF AS TASK TO ASSESS TMS-INDUCED NEUROMODULATION OF THE DLPFC AND OF MOOD IMPROVEMENT**

To perform an AS requires the inhibition of a reflexive saccade toward a target and the generation of a voluntary saccade in the opposite direction (Everling and Fischer, 1998). Several lines of evidence suggest that the DLPFC is involved in a cortical network underlying the inhibition process required to perform a correct AS. Some studies showed that lesions of the DLPFC in humans lead to an increase in inhibition errors during AS tasks (Pierrot-Deseilligny et al., 2003; Ploner et al., 2005). Electrophysiological and TMS studies confirmed the involvement of the DLPFC in saccadic inhibition, suggesting a lateralized inhibitory control of the DLPFC (Müri et al., 1999; Johnston and Everling, 2006; Wegener et al., 2008; Müri and Nyffeler, 2008; but Nyffeler et al., 2007). As a consequence, studying AS performance during rTMS treatment applied over the DLPFC, might inform about the rTMS-induced neuromodulation of this cortical region. Moreover, some studies showed that a clinical improvement could result in cognitive improvement. Biringer et al. (2005) showed that recovery from depression is associated with a recovery of many aspects of executive functions to a normal level. Moreover, using rTMS over the left DLPFC on treatmentresistant depressed patients, Kedzior et al. (2012) found, after 20 days of rTMS treatment, a cognitive improvement for these patients as well as a mood improvement. As a consequence, we expected that the saccadic task could also be a marker of mood improvement.

## **PILOT STUDY**

To test the relevance of AS task to assess TMS-induced neuromodulation of the DLPFC and of mood improvement, we conducted a pilot study in 12 drug-resistant bipolar depressed patients receiving either active or sham iTBS over the left DLPFC for 1–3 weeks. Short-term TMS-induced neuromodulation was tested by comparing performances to a saccadic task performed just before and just after the iTBS session. It was expected that AS performances would be better just after the iTBS session than before for the iTBS active group and not for the sham group; and that changes in AS performances would be stronger in the contralateral hemifield of the stimulated DLPFC, i.e., here, the right hemifield. The relevance of the AS task to assess patients' mood improvement was analyzed by computing the correlation between mood improvement and AS performance improvement. Besides this, we examined the difference between depressed bipolar patients and healthy controls; it was expected that before treatment, patients' AS performances would be impaired compared to those of healthy subjects.

## **METHODS**

### **EXPERIMENTAL DESIGN**

Active or sham iTBS treatment was applied twice a day, 5 days a week for 1–3 weeks. Two sessions of saccadic task were conducted on the first day of each week (D0, D7, and D14), one session (S1) before the iTBS and a second (S2) immediately following the iTBS (**Figure 1**). To ensure that any improvement in the AS performances could not be accounted for by a simple learning effect, two training sessions were administered 4 days before (D-3) the beginning of the iTBS treatment to minimize the practice-related effects. Two sessions of the saccadic task were also performed at the end of the protocol (D18).

## **SUBJECTS**

#### **Patient group**

Twelve patients (6 females, 6 males; mean age 51.6 ± 11.7 years) with drug-resistant bipolar disorder (Types I, II, or III) participated in this randomized double-blind placebo-controlled iTBS study. The study was approved by the regional Ethics Committee (Committee for the Protection of Persons in biomedical researches: CPP Sud-Est VI) and allowed by the ANSM (French National Agency for Medicines and Health Products Safety) (Authorization: 2010-A01085-34).

Patients were randomly assigned to the active or to the sham group. This randomization was performed using a randomization table and assessed 3 by 3 to have the same proportion of patients in each group. As allowed by the protocol of the study, the TMS operator did the un-blinding for the purpose of this study, independently from the clinical research team, who remained fully blind of each patient's treatment status.

The inclusion criteria were having a drug-resistant major depressive episode defined according to DSM IV-TR. The criterion of severity was a score of over 20 with a maximum of 60 on the *Montgomery and Asberg Depression Rating Scale* (MADRS: Montgomery and Asberg, 1979) (see **Table 1** for details). Drug resistance was defined as the absence of a response to any antidepressant treatment over at least a 4-week period

**Table 1 | Individual MADRS scores arranged by Group (active vs. sham) and day of treatment (beginning vs. end) and intergroup comparisons**.


of time. Patients with a history of substance abuse and patients who did not meet inclusion criteria for use of TMS and MRI (no pacemaker, no history of epilepsy or other neurological disorders) were excluded. Mood-stabilizers were allowed during the treatment period only if the patients were stable for at least 4 weeks before the rTMS treatment. Only anxiolytic drugs (cyamemazine or hydroxyzine) at low doses could be administered if necessary during the rTMS treatment period.

#### **Control group**

Twelve control subjects (7 females, 5 males; mean age 50.6 ± 10.9 years) also participated in the study. These subjects had no psychiatric history and were not taking any medication. This group only performed one saccadic task session. Every participant provided written, informed consent.

### **INTERMITTENT THETA BURST STIMULATION (iTBS)**

The iTBS was applied using a figure-of-eight coil (MCF-B65- H0) and an air-cooled stimulator (MagProX100, MagVenture). For each patient the left DLPFC was delimited using threedimensional magnetic resonance imaging (3D-MRI). The TMS coil was monitored throughout stimulations using a neuronavigation device (TMS Navigator, Localite). We used an iTBS protocol in which a 2 s train of bursts containing three pulses at 50 Hz was repeated at 200 ms (i.e., 5 Hz) every 10 s (Huang et al., 2005), these parameters mimicked the theta rhythm in EEG nomenclature. The TMS operators applied the iTBS twice per day, with a minimum inter-session interval of 3 h every day for 1–3 weeks depending of clinical relevance. For each rTMS session 990 pulses were administered to give a total of 5 min 30 s of stimulation, thus, patients received 1980 pulses per day. Patients were stimulated with either an active coil (active group: *n* = 5) or a sham coil (sham group: *n* = 7). The sham coil made the same "clicking" sound than the active coil, and produced a weak magnetic field on the scalp for reproducing the same skin sensation than the active coil.

We individually set the stimulation intensity at 80% of the patient's resting motor threshold (RMT), which we determined 3 days before the beginning of the iTBS treatment phase (D-3). We began by placing three electrodes over the patient's first dorsal interosseous muscle (FDI) in a belly-tendon montage. Electromyograms were amplified (1–10 K), band-pass filtered (1–6 KHz), and sampled at 12 KHz using a Dantec Keypoint portable system (Natus Medical Incorporated). We placed the coil on the "hotspot": the position on the motor cortex that elicited the greatest motor-evoked potential (MEP) in the contralateral FDI. We defined the RMT as the minimum stimulation intensity needed to evoke a MEP greater than 50 µV on at least 5 out of 10 consecutive trials (Rossini et al., 1994).

#### **SACCADIC TASK**

We used an EyeLink 1000 video-based eye-tracking system (SR Research) with a temporal resolution of 500 Hz. The eye-tracker detects saccades automatically, using three thresholds: velocity (30◦ /s), acceleration (8000◦ /s2), and saccadic motion (0.15◦ ). Stimuli were displayed on a computer screen located 57 cm from the participants. The computer screen resolution was 1024 × 768 pixels and the screen refresh rate was 85 Hz. Participants were seated in a darkened room and their heads were stabilized using chin rests. We used a SPAN task (Saccade: Pro, Anti and No) which mixed three types of saccades: antisaccades (AS), nosaccades (NS) and prosaccades (PS). PS was used to verify that the bipolar patients did not suffer from a general deficit in saccadic function. Additionally, compared to a traditional AS task, this mixed saccadic paradigm increased the cognitive load in terms of executive functions, and thus reinforced the implication of the DLPFC (Smith and Jonides, 1999).

Each trial began with a 500 ms presentation of a white central fixation dot, and then the central fixation dot became red, blue, or green for 2 s. Participants were told to make a PS if it was green, an AS if it was red, or an NS response if it was blue (**Figure 2**). After this time, a blank screen was displayed for a 200 ms gap, and a "cue", the number "0", was flashed for 50 ms at 10◦ peripherally (randomly on the right or left side of the screen). During AS trials, patients had to look towards the opposite side from the cue as quickly as possible in order to identify a numeric target ("6" or "9"), which was presented for 1 s beginning as soon as they looked at the correct location (gaze-contingent display) or after a 2-s delay. During PS trials, patients had to look as quickly as possible towards the side of the cue to identify the numeric target. During NS trials, they had to keep their gaze fixed on the center of the screen. There was a break of 1 s between two successive trials. During the first SPAN session, both patients and healthy controls received 20 practice trials and 80 test trials (16 NS trials, 32 PS trials, and 32 AS trials). In the following SPAN sessions, patients performed 80 test trials. We assessed the performances using the inhibition error rate, i.e., the proportion of saccades towards the cue for AS and NS trials and the latency of correct saccades for PS and AS trials.

We analyzed the oculometric performance, i.e., the inhibition error rates and the saccadic reaction times using Matlab (MATLAB, R2009b, The MathWorks Inc., Natick, MA, 2009) and Statistica (Statistica 10, Statsoft Inc., 1984).

#### **MOOD EVALUATION**

We analyzed MADRS scores at the beginning and at the end of the experiment in order to assess mood changes over the time course of the experiment, as well as to assess responses to the iTBS treatment. We defined "response to the treatment" when there was a 50 % improvement of the MADRS score and "remission" when the MADRS score became lower than 8.

## **RESULTS**

In the following results we only present the inhibition error rates in AS trials as we did not find any differences for PS, NS and latencies of AS trials.

#### **COMPARISON OF "HEALTHY SUBJECTS VS. DEPRESSIVE BIPOLAR PATIENTS"**

We performed an ANOVA on inhibition error rates with the Group (bipolar patients (*n* = 12) and healthy controls (*n* = 12)) as the between-subjects factor, and the Cue Position (left or right) as the within-subjects factor. This comparison was performed on the learning session only (D-3) for both groups. The ANOVA revealed a main effect of the Group (*F*(1,22) = 4.8; *p* = 0.04). Patients committed significantly more errors than did controls (26.9% vs. 13.8%). We did not find any effect of the Cue Position, nor of the interaction Group × Cue Position (*F*(1,22) < 1).

#### **SHORT-TERM iTBS NEUROMODULATION**

We performed another ANOVA on inhibition error rates with the Group" (active (*n* = 5) and sham (*n* = 7)) as the betweensubjects factor, and the Session (Session 1 and Session 2) as the within-subjects factor. The ANOVA did not reveal any effect of

the Group (*F*(1,10) < 1). However, we found a main effect of the Session (*F*(1,10) = 6.29; *p* = 0.03). Performances were improved in session 2 compared to session 1 (24.2% vs. 19.9%). We also found a significant interaction Group × Session (*F*(1,10) = 5.29; *p* = 0.04) (**Figure 3**). Patients in the active group committed fewer errors in session 2 than in session 1 (23.2% vs. 14.9%) (*F*(1,10) = 9.9; *p* = 0.01), while the performances of the patients in the sham group did not show any improvements (25.1% vs. 24.8%) (*F*(1,10) < 1) (**Figure 3**). When the cue was presented in the right hemifield (contralateral field of iTBS neuromodulation), 100% of patients in the active group improved their performances (vs. 57% of the sham group).

#### **EFFECTS OF MOOD IMPROVEMENT**

The MADRS scores analyses did not reveal any differences between the active and sham groups before the treatment (*p* = 0.51). After the treatment, four out of five patients in the active group and four out of seven in the sham group responded to the treatment i.e., showed an improvement of more than 50% on the MADRS scores. The *t*-tests did not reveal any differences between groups, neither on the MADRS scores nor for mood improvement (**Table 1**). To assess the relevance of the AS task as a marker of this mood improvement, we calculated the Pearson's correlation coefficient between the improvement in the MADRS scores and the difference in inhibition error rates between the end and the beginning of the treatment. We found a significant and positive linear association (*r* = 0.65; *p* = 0.02; *R* <sup>2</sup> = 0.42): better mood is associated with better performance in AS (**Figure 4**).

#### **DISCUSSION**

This study investigated the AS task as a psychophysical tool to assess the short-term neuromodulation that has been hypothesized to be induced by daily iTBS delivered over the left DLPFC. We also examined the ability of the AS task to discriminate depressive bipolar patients from healthy subjects, and to be a marker of mood improvement.

#### **RELEVANCE OF AS TASK TO DISCRIMINATE DEPRESSED BIPOLAR PATIENTS FROM HEALTHY SUBJECTS**

Over the past three decades, there has been an increase in the number of neuropsychophysical studies of saccadic performance

in psychiatric patient groups (Gooding and Basso, 2008). Some authors suggested that performance on the AS task could be used as a psychophysical marker for mood disorders (García-Blanco et al., 2013; Malsert et al., 2013). Our results confirmed that the AS task could discriminate depressed bipolar patients from healthy subjects. We did not find any group differences on PS performance, which means that depressive bipolar patients did not suffer from a general impairment in saccadic function.

Maybe due to the small number of patients, our study did not show group differences associated with the cue position, i.e., no cerebral asymmetry. Whether depressed bipolar patients display a cerebral asymmetry in the inhibitory functions of the DLPFC is still being discussed in the literature (Clark et al., 2006; Savitz and Drevets, 2009); it is a crucial issue since rTMS therapy is often based on the premise that left DLPFC is hypoactive in depression.

#### **RELEVANCE OF AS TASK TO ASSESS SHORT-TERM iTBS-INDUCED NEUROMODULATION OF THE DLPFC**

Until now only one study investigated the AS performance during rTMS treatment applied over the DLPFC to investigate the rTMS neuromodulation effect of this cortical region (Crevits et al., 2005). In their study, the left DLPFC of 11 depressed patients was stimulated with 1000 pulses per day of facilitatory rTMS (10 Hz). The treatment lasted for at least 10 sessions, with no more than one session a day, over a maximum period of 3 weeks. The AS task was only performed twice: before the first rTMS session and after the last rTMS session. They found a significant decrease in AS latencies at the end of the treatment. However, the absence of a control group of patients receiving sham rTMS prevented the authors from drawing any conclusions about the long-term effects of rTMS. Additionally, the latency reduction in AS might have reflected a practice effect as the AS task was being repeated at the end of the rTMS treatment. Chauvin et al. (2011) studied the evolution of performance across several sessions of an AS task and found that performance only improved (with decreases in AS latency and inhibition error rate) over the two first sessions. Finally, mood improvement was able to explain the AS performance improvement (Salvadore et al., 2011).

In our knowledge, our study was the first research investigating by oculometry the short-term iTBS-induced neuromodulation of the DLPFC, using a neuronavigation system. As expected, we observed an aftereffect of the iTBS sessions: only the patients in the active group improved their capacity to inhibit the reflexive saccades immediately after the iTBS sessions. This improvement was consistently observed when the cue was presented in the right hemifield, i.e., processed by the left DLPFC. These results provide evidence that iTBS induces short-term neuromodulation of the targeted cortical area. This means that the AS task could be a useful instrument to ascertain whether short-term neuromodulation induced by iTBS occurs.

#### **RELEVANCE OF AS TASK AS A POTENTIAL MARKER OF MOOD IMPROVEMENT**

We found that the active and sham groups showed a similar mood improvement. This improvement in the sham group and possibly in the active group too was due to a placebo effect (see Mayberg et al., 2002). Some studies showed that mood improvement could result in cognitive improvement. Biringer et al. (2005)showed that recovery from major unipolar depression is associated with a recovery of many aspects of executive functioning, improving executive functioning to a normal level. In accordance with these studies, we found a significant correlation between mood improvement and AS improvement. This indicates that the saccadic task might not only be a useful marker of the short-term neuromodulation, but a marker of mood changes too.

#### **CONCLUSION**

This pilot study investigated the iTBS-induced short-term neuromodulation of the DLPFC. This is a crucial issue since little is known about the aftereffect of TBS over the DLPFC while being used in clinical research, in particular, with psychiatric disorders (Soekadar et al., 2009; Chistyakov et al., 2010; Holzer and Padberg, 2010; Plewnia et al., 2014). Exploring the rTMS/iTBS aftereffects requires the development of an instrument to enable one to objectively measure the short-term TBS-induced neuromodulation, which is the *sine qua non* condition to long-term neuromodulation taking place (Pascual-Leone et al., 1994; Valero-Cabré et al., 2011). Our study demonstrates that an AS task could be used to assess it. Moreover, we confirmed that AS performance could discriminate depressive bipolar patients from healthy subjects and be used as a marker of mood variation (response to treatment or relapse into illness) (Malsert et al., 2012b; Aleman, 2013).

However, due to the small sample size, our findings should be replicated using a larger cohort of patients. Moreover, the oculometric task could be improved by adding an emotional component to increase the load imposed on the DLPFC inhibitory control. Indeed, in humans, the existence of connections between the DLPFC and the limbic regions is well established, although the anatomical details of the connections remain unclear in humans (Fox et al., 2012) and monkeys (Petrides and Pandya, 1999). Adding emotional cues should improve the psychometric relevance of the oculometric sessions (García-Blanco et al., 2013). Using an implicit emotional oculometric paradigm every day with depressive bipolar patients should enable a finer-grained analysis of short-term neuromodulation induced by rTMS over the DLPFC and could be a way to optimize and customize the TMS treatment by adjusting rTMS parameters of each patient according to the obtained post-effect.

## **AUTHOR AND CONTRIBUTORS**

Alan Chauvin, Nathalie Guyader, and Christian Marendaz designed the research; Thierry Bougerol and David Szekely analyzed and interpreted clinical data; Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel and Christian Marendaz performed research; Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel and Christian Marendaz analyzed and interpreted data; Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel, Christian Marendaz and Thierry Bougerol wrote the paper; Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel, Christian Marendaz and Thierry Bougerol gave the final approval of the version to be published.

Lysianne Beynel, Alan Chauvin, Nathalie Guyader, Sylvain Harquel, Christian Marendaz, David Szekely and Thierry Bougerol agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

#### **ACKNOWLEDGMENTS**

This work was supported by Grenoble-Alpes University, the National Center for Scientific Research (CNRS), Rhône-Alpes Region, Health and Society research association. The authors want to thank Marcia Bécu for her experimental help; Vincent Meille and Benoit Trojak for the DLPFC localization.

#### **REFERENCES**


antidepressant response to partial sleep deprivation in major depression? *Psychiatry Res.* 146, 43–51. doi: 10.1016/j.pscychresns.2005.09.007


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

*Received: 04 February 2014; accepted: 31 July 2014; published online: 19 August 2014*.

*Citation: Beynel L, Chauvin A, Guyader N, Harquel S, Szekely D, Bougerol T and Marendaz C (2014) What saccadic eye movements tell us about TMS-induced neuromodulation of the DLPFC and mood changes: a pilot study in bipolar disorders. Front. Integr. Neurosci. 8:65. doi: 10.3389/fnint.2014.00065*

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

*Copyright © 2014 Beynel, Chauvin, Guyader, Harquel, Szekely, Bougerol and Marendaz. 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*.

## ADVANTAGES OF PUBLISHING IN FRONTIERS

FAST PUBLICATION Average 90 days from submission to publication

COLLABORATIVE PEER-REVIEW

Designed to be rigorous – yet also collaborative, fair and constructive

RESEARCH NETWORK Our network increases readership for your article

#### OPEN ACCESS

Articles are free to read, for greatest visibility

#### TRANSPARENT

Editors and reviewers acknowledged by name on published articles

GLOBAL SPREAD Six million monthly page views worldwide

COPYRIGHT TO AUTHORS

No limit to article distribution and re-use

IMPACT METRICS Advanced metrics track your article's impact

SUPPORT By our Swiss-based editorial team

EPFL Innovation Park · Building I · 1015 Lausanne · Switzerland T +41 21 510 17 00 · info@frontiersin.org · frontiersin.org