Vergence Eye Movements Redefined: the Neural Control of Fast Versus Slow Vergence

Vergence eye movements redefined: The neural control of fast versus slow vergence

Marion R. Van Horn

Aerospace Medical Research Unit Department of Physiology McGill University Montreal, Quebec, Canada

August 2010

A thesis submitted to the faculty of Graduate Studies and Research in partial fulfilment of the degree of Doctorate in Philosophy

ABSTRACT…………………………..…………………………..………………... xiii RÉSUMÉ…………………………..…………………………..…………………… xv ACKNOWLEDGEMENTS…………………………..………………………… ..xviii CONTRIBUTIONS OF AUTHORS…………………………..………………….. xx LIST OF ABBREVIATIONS…………………………………..……………….. xxii

CHAPTER 1: GENERAL INTRODUCTION & LITERATURE REVIEW 1.1 Overview and conceptual framework………...……………………………… 1 1.2 The extraocular eye muscles and their innervations…………….…………… 4 1.3 Classification of eye movements………………….…………….…………… 8 1.4 The neural control of disconjugate saccades: Hering versus Helmoholtz... 11 1.4.1 Evidence supporting Hering’s Law …………………………….... 11 1.5 The premotor circuitry of conjugate saccades ………………………….... 14 1.5.1 Saccadic burst neurons…………….…………………………….... 15 1.5.2 Omnipause neurons………………..…………………………….... 16 1.5.3 Oculomotor integration ………………………………………….... 17 1.6 The premotor circuitry of saccade-free vergence ……………………….... 19 1.7 Evidence against “Hering’s Law” ……………………………………….... 20 1.8 Research Goals ……………………………….………………………….... 22

CHAPTER 2: DYNAMIC CHARACTERIZATION OF OCULOMOTONEURONS DURING CONJUGATE AND DISCONJUGATE EYE MOVEMENTS 2.1 ABSTRACT ………………………………………………………………. 25 2.2 INTRODUCTION ………………………………………………………... 26 2.3 METHODS .………………………………………………………………. 28 2.3.1 Animal and surgical procedures……………………………………. 28 2.3.2 Behavioral paradigms ………………………………………………. 29 2.3.3 Data acquisition procedures ……………………………………….. 30

2.3.4 Data Analysis ………………….…………………………………… 31 2.3.5 Quantification of ocular preference ….…………………………….. 34 2.4 RESULTS ...……..………………………….……………………………... 35 2.4.1 Dynamic analysis during conjugate saccades ………...……………. 35 2.4.2 Dynamic responses estimated during OFF-directed conjugate saccades ……………………………………………….. 38 2.4.3 Example OMN with a preference for the ipsilateral eye ….………. . 38 2.4.4 Response of OFF-directed disconjugate saccades …………………. 43 2.4.5 Summary of ocular preferences ……….……………………..…….. 46 2.4.6 Comparison of ocular preference during disconjugate saccades and disconjugate fixation …………………………………………….... 48 2.4.7 Model parameters estimated across oculomotor behaviors: fixation, microsaccades, and saccades …………………………………….... 50

2.5 DISCUSSION ……………………………………………………………... 53 2.5.1 Dynamic discharge of OMNs during conjugate saccades …………. 55 2.5.2 Responses of oculomotoneurons during disconjugate fixation and saccades …………………………………………………………… 56 2.5.3 Comparison of the motor drive of agonist medial and lateral rectus motoneurons during disconjugate saccades…...…………………... 59 2.5.4 Consideration of the antagonist muscle when modeling across oculomotor behaviors ……………………………………………... 60

CHAPTER 3: DYNAMIC CHARACTERIZATION OF SACCADIC BURST NEURONS DURING CONJUGATE AND DISCONJUGATE EYE MOVEMENTS 3.1 ABSTRACT………………………………………………………………... 63 3.2 INTRODUCTION ………………………………………………………… 64 3.3 METHODS ………………………………………………….…….………. 65 3.3.1 Animals and surgical preparations …………………………………. 65 3.3.2 Behavioral paradigms ………………………………………………. 66 3.3.3 Data acquisition procedures…..…………………………………….. 69

3.3.4 Data analysis………………………………………………………….70 3.3.4a Dynamic analysis of BN firing rate …………………………71 3.3.4b Metric analysis of BN discharges …………………………………..72 3.3.5 Quantification of ocular preference……...……………….….………73 3.3.6 Simulation design………………………...……………….….……... 75 3.4 RESULTS………………………………………………………………….....78 3.4.1 SBN discharge timing is appropriate to facilitate vergence movements ………………………...……………………..………. 79 3.4.2 Testing the null hypothesis: SBNs only encode conjugate eye movement dynamics ..………………………………………….…. 79 3.4.3 Rejecting the null hypothesis: The conjugate prediction fails ….…. 84 3.4.4 Comparison across neuron types ….………….………….………… 90 3.4.5 Calculation of the net premotor drive…………………….………… 92 3.4.6 Metric analysis of conjugate and disconjugate saccades .…………. 92 3.4.7 Comparison of dynamic and metric analyses .…………………….. 96 3.4.8 Simulation design…………………………….…………………….. 96 3.4.9 Analysis of simulation weights ……………….…………………… 99 3.5 DISCUSSION…………………………………………………………….. 100 3.5.1 Comparison with previous reports: conjugate saccades …….……. 102 3.5.2 The timing and dynamics of SBN burst activity are appropriate to facilitate vergence during disconjugate saccades ………………… . 102 3.5.3 The role of the saccadic burst generator during saccade-vergence interactions …………………………………………………...…... 104 3.5.4 Source of vergence-related signals ………………………………… 108

CHAPTER 4: VERTICAL FACILITATED VERGENCE BY PREMOTOR SACCADIC BURST NEURONS 4.1 ABSTRACT……………………………………………………………….. 114 4.2 INTRODUCTION .………………….…………………………………… 115 4.3 METHODS ……………………………………………………………….. 118 4.3.1 Animals and surgical preparations…………………………………. 118

iii

4.3.2 Behavioral paradigms………………………………………………. 118 4.3.3 Data acquisition …………………………………………………….. 121 4.3.4 Definitions and conventions ……..………………………………… 122 4.3.5 Data analysis ……………………………………………………….. 123 4.3.6 Quantification of ocular preference………………………………… 125 4.3.5 Statistical analysis………….……………………………………….. 125 4.4 RESULTS…………………………………………………………………... 126 4.4.1 Characterization of vergence facilitated by vertical saccades .……. 126 4.4.2 Temporal alignment of peak vertical and vergence velocities …….. 128 4.4.3 Test of the hypothesis: vergence is facilitated by the classical saccadic pathway during disconjugate saccades …………………….……... 131 4.4.4 Estimation of the vergence-related signal encoded by horizontal saccadic burst neurons during vertical disconjugate saccades……. 139 4.4.5 Ocular sensitivities across of the population of SBNs……………... 142 4.4.6 Comparison of ocular preference during horizontal and vertical disconjugate saccades…………………………………………….. 144 4.5 DISCUSSION……………………………………………………………… 148 4.5.1 Vergence velocity is facilitated during vertical disconjugate saccades ……………………………...……………………………. 149 4.5.2 Vergence and vertical velocities are temporally aligned during vertical saccades in monkeys …………………...…………………. 150 4.5.3 The dynamics of SBNs during horizontal and vertical conjugate saccades ……………………………………...... 151 4.5.4 SBNs contribute to increasing vergence velocities during disconjugate saccades ……………………………..……………………………… 152 4.5.5 Premotor circuits for the control of changes in vergence angle….… 154 4.6 Supplemental Material ……………………………………………….….. 155

CHAPTER 5: IDENTIFICATION AND DYNAMIC CHARACTERIZATION OF VERGENCE NEURONS IN THE ROSTRAL SUPERIOR COLLICULUS

5.1 ABSTRACT……………………………………………………………...... 165 5.2 INTRODUCTION.………………………………………………………... 166 5.3 METHODS…………………………………………………………………. 167 5.3.1 Surgical procedures ………………………………………………... 167 5.3.2 Behavioral paradigms………………………………………………. 168 5.3.2.1 Conjugate paradigms ……………………………………… 168 5.3.2.2 Vergence paradigms……………………………………….. 169 5.3.3 Data acquisition procedures...... ………………………………… 169 5.3.3.1 Extracellulular single unit recordings………….………… 170 5.3.3.2 Microstimulation……………………………….………… 171 5.3.4 Data Analysis……………….………………………………………. 171 5.3.4.1 Definitions and Conventions………..………….………… 171 5.3.4.2 Metric analysis………………………...……….………… 172 5.3.4.3 Dynamic analysis……………………………….………… 172 5.3.5 Histology………………..….………………………………………. 173 5.3.6 Statistical analysis…….…….………………………………………. 173 5.4 RESULTS………………………………………………………………….. 173 5.4.1 Neurons in rostral SC dynamically encode vergence during vergence tracking………………………………………………….. ………. 179 5.4.2 Convergence and divergence neurons in the rostral SC encode Slow but not fast vergence…………………………..……………. 182 5.5 DISCUSSION………………………………………………………..……... 184 5.5.1 Microstimulation of the rostral SC…………………………………. 187 5.5.2 The neural control of fast versus slow vergence..……..…………… 190

CHAPTER 6: GENERAL DISCUSSION 6.1 Signals encoded by extraocular motoneurons……………..……………….. 198 6.2 Premotor coding in 3D……………………...……………..……………….. 200 6.3 Superior colliculus ensures binocular realignment of gaze..……………….. 202 6.4 Clinical applications……………..…………………………………………. 205 6.5 Conclusions and future directions…………..……………..……………….. 205

6.5.1 Are small motoneurons more specialized in encoding vergence movements?...... 206 6.5.2 What are the upstream sources of vergence?...... 206

APPENDIX A: LOCAL NEURAL PROCESSING AND THE GENERATION OF DYNAMIC MOTOR COMMANDS WITHIN THE PREMOTOR NETWORK A.1 Abstract ……………………………………………………………………. 210 A.2 Introduction ……………………………………………………………….. 211 A.3 Methods ……………………………………………………………………. 212 A.3.1 Surgical procedures ……………………………………………….. 212 A.3.2 Experimental paradigms and data acquisition…...………………… 214 A.3.3 Data analysis……… ……..………………………………………... 215 A.3.3.1 Metric analysis……………………………………………. 216 A.3.3.2 Dynamic analysis…………………………………………. 216 A.3.3.3 Eye velocity reconstruction……………………………….. 218 A.3.3.4 Spike triggered average and spike field coherence……… 220 A.3.3.5 Spectrogram analysis.……………………………………... 221 A.3.3.5 LFP and spike tuning curves.……………………………… 221 A.4 Results …………………………………………………….…………….. 222 A.4.1 LFPS: a reflection of intracellular activity……... ………………… 222 A.4.2 LFP response timing is consistent with sequential processing within saccadic network …………………………….…………………… 224 A.4.3 The temporal dynamics of LFPs predict motor behavior………….. 228 A.4.4 Temporal and spatial relationships between spike trains and LFPs.. 234 A.5 Discussion …………………………………………………………………. 244 A.5.1 LFPs display similar properties to intracellular recordings ………. 245 A.5.2 LFPs provide precise timing information and predict motor responses for movements in the non-preferred as well as preferred directions ………………………………………………. 240 A.5.3 From local field potentials to spike trains: temporal and spatial relationships……………………………………………………….. 246

A.5.4 Functional implications of LFPs as related to the analysis of neural circuits…………………………………………………………….. 248

References 257

vii

LIST OF FIGURES

1.1 Schematic illustration of the cortex and brainstem…………………………. 2

1.2 Model of the syringe-spring system as compared to the eye in the orbit..... 7

1.3 Example conjugate saccade and saccade-free vergence... ………………… 9

1.4 Theoretical frameworks of binocular control……………………………….. 12

1.5 Pathways that control saccades and saccade-free vergence…………………. 18

2.1 Discharge patterns of 2 typical oculomotoneurons…………. …………….. 37

2.2 Example of model fits of a typical OMN during conjugate saccades …….. 39

2.3 Example of model fits of a ipsilateral OMN during ON disconjugate

saccades …………………………………………………………………... 42

2.4 Example of model fits of a contralateral OMN during OFF disconjugate

saccades …………………………………………………………………... 45

2.5 Distribution of ratios indexes for OMNs …………………………………… 47

2.6 Distribution of Ratio indexes for disconjugate fixation…………………...... 49

2.7 Dependence of eye velocity coefficients, eye position coefficients and biases

on peak velocity ………………………………………………………….... 52

2.8 Simulation of population drive during conjugate and disconjugate saccades 58

3.1 Example conjugate and discharge saccade and saccade-free vergence…….. 67

3.2 Neural activity of typical IBN and OPN during saccade-free vergence..….. 80

3.3 Neural activity of IBN during disconjugate saccade..……………………... 81

viii

3.4 Example of model fits of an example EBN and IBN during disconjugate

saccades …………………………………………………………………... 83

3.5 Example monocular EBN during disconjugate saccades…………………… 86

3.6 Example conjugate IBN during disconjugate saccades …………………...... 87

3.7 Distribution of Ratiodyn indexes....………………………………………..... 89

3.8 Distribution of RatioNOS and Rationdyn for EBNs and IBNs...... 95

3.9 Schema of circuitry used for simulations…………………….…………….. 97

3.10 Estimated weights of premotor neurons…………………………….…….. 98

3.11 Simulation results…………………………………………………………. 101

3.12 Proposed neural circuitry involved in generating disconjugate saccades... 107

S3.1 Estimated weights of eight premotor neurons as a function of BT weight. 112

4.1 Schema of paradigms used to elicit vertical vergence movements…………. 120

4.2 Typical examples of vergence facilitated by vertical saccades.…………….. 127

4.3 Average vergence velocities ………………….……………………………. 129

4.4 Individual examples of behavior and distributions of temporal

temporal dissociations and vergence velocities……….……………….... 132

4.5 Neural responses and polar plot of SBN during conjugate saccades …….… 135

4.6 Neural responses and model fits of SBN during vertical

disconjugate saccades...... ……………………………………………….. 136

4.7 Distribution of vertical sensitivities…………………………..…………….. 138

4.8 Distribution of Rationdyn indexes…………………………..……………….. 143

4.9 Neuronal responses and model fits of SBN during horizontal

disconjugate saccades…………………………………………………… 145

4.10 Coherence of ocular preference…………………………..……………….. 147

S4.1 Example EBN with broad tuning………………………..………………… 160

S4.2 Neuronal responses and model fits for example EBN with broad tuning

during conjugate saccades………………………………………………. 161

S4.3 Neuronal responses and model fits for example EBN with broad tuning

during disconjugate saccades……………………………………….…… 162

5.1 Example sites of microstimulation of rostral SC………………..…………. 175

5.2 Example divergence neuron………………………………….…………….. 177

5.3 Example convergence site.………………….…………….………………. 180

5.4 Discharge rates of vergence neurons during smooth pursuit…………….... 181

5.5 Comparison of vergence sensitivities ……………………………..…….… 183

5.6 Divergence neuron during disconjugate eye movement...... ………. 185

5.7 Convergence neuron during disconjugate eye movement…..…………….. 186

5.8 Schematic diagram of premotor control of vergence……..……………….. 193

5S.1 Example of parafoveal visual neuron…. …………………………………. 195

5.S2 Example of rostral SC neuron.……………………. ……….…………….. 195

A.1 Diagram of saccadic brainstem circuitry. …………………………………. 213

A.2 Characteristics of LFPs recorded from typical OPN. ……….…………….. 224

A.3 Characteristics of LFPs recorded from typical SBN………………………. 226

A.4 Dynamic analysis of LFP and single unit responses recorded from an example

SBN, MN and OPN. ………………………………….……………….... 230

A.5 Stimulus reconstruction of the eye velocity ………………………….….… 233

A.6 Average population coding fractions...... …………………………………… 235

A.7 Spike triggered average and spike field coherence ………... .…………….. 237

A.8 Spatial relationship of SBN LFP responses.……………..…….………….. 239

A.9 Spatial relationship of OPN LFP responses. ……………………………… 241

A.10 Effect of electrode distance on LFP response. ..………..……………….. 243

A.S1 Average correlation coefficients……………… ..………..……………….. 251

A.S2 Example of LFPs for OPNs and SBNs………….………..……………….. 252

A.S3 Distribution of saccade amplitudes……………..………..……………….. 253

List of Tables

1.1 Classes of eye movements…………………………………....……………. 10

2.1 Average VAFs across population of OMNs………………….……………. 41

2.2 Average VAFs and BICs estimated using binocular versus reduced models.. 44

2.3 Mean model parameters estimated across oculomotor behaviors.…………. 54

3.1 Categories of ocular preference……………………………….……………. 74

3.2 Model parameters used in simulation………………………………………. 77

3.3 Model predictions and estimations during disconjugate saccades…………. 91

3.4 Model parameters during conjugate saccades………………………………. 93

3.S1 Model parameters during conjugate saccades………...…...... ……………. 110

3.S2 Model parameters used in simulation……………………...... ……………. 111

4.1 Temporal alignment of vertical and vergence peak velocities...……………. 130

4.2 Average predictions, VAFs and ΔBICs ……………………....……………. 140

4.S1 Estimated of horizontal and vertical sensitivities for SBNs...……………. 163

xii

Abstract

Seeing in 3D relies on the fact that our two eyes are spaced slightly apart. As a result, when our eyes are aligned on an object each eye has its own, slightly different view of the world.

Information about relative depth is sent from visual cortex to the motor control centers in the brainstem, which are responsible for generating appropriate motor commands to move the eyes.

Surprisingly, how the neurons in the brainstem use the depth information supplied by visual cortex to precisely aim each eye on a visual target remains highly controversial. This thesis investigates how individual neurons control the movement of each eye when we look between objects located at different depths.

I first studied the signals encoded by individual oculomotoneurons during conjugate and disconjugate saccades and evaluated the role the antagonist muscle plays in accurate binocular positioning. I found that a first-order model (containing eye position and velocity terms) provided an adequate description of the neural discharges of oculomotoneurons during conjugate saccades.

However, during disconjugate saccades I found the conjugate sensitivities could not be used to predict the responses during disconjugate saccades. Instead, the majority of the neurons preferentially encoded the movement of the ipsilateral eye.

I next studied the signals carried by saccadic burst neurons during conjugate and disconjugate saccades. I first showed that the majority of the saccadic burst neurons dynamically encode the movement of the ipsilateral eye during the fast saccadic component of a movement.

Moreover, using a neural simulation I found that the saccadic burst generator carries all the vergence drive necessary to shape the activity of the abducens motoneurons to which it projects.

These results are consistent with the proposal that classically assumed “conjugate” saccadic structures in the oculomotor brainstem underlie vergence facilitation by providing monocular

xiii saccadic commands to the abducens during disconjugate saccades and a separate vergence pathway is used to adjust ocular alignment following the saccadic component of the movement.

Subsequently, I tested the prediction that if the monocular commands issued by saccadic burst neurons are important for facilitating vergence velocities during horizontal saccades, they should also contribute to facilitating vergence associated with a vertical saccade when the conjugate component of the movement is negligible. As predicted, I found that saccadic burst neurons are also well suited for facilitating vergence during vertical saccadic movements. In particular, saccadic burst neurons contribute to generating increased vergence velocities by dynamically encoding the movement of an individual eye rather than the conjugate component of the movement.

Finally, I used single unit recording and microstimulation techniques to investigate the role the superior colliculus plays in generating vergence eye movements. I provide evidence that individual neurons within the rostral SC encode slow changes in vergence angle. These results suggest that there exists a distinct grouping of neurons that encode slow vergence within the rostral SC and indicate that activation of the rostral SC underlies the ability to accurately position each of the two eyes when fixating targets in 3-dimensional space to ensure stereopsis.

Taken together, my findings provide important new insight into how the brain controls 3- dimensional gaze shifts. I provide evidence that distinct neural pathways control fast and slow vergence. These results contradict the traditional view that the brain is circuited with independent pathways for conjugate and vergence control and suggest the need to update textbooks and review articles to emphasize the physiological differences between the neural circuitry controlling fast and slow vergence.

xiv

Résumé

Voir en 3D se fonde sur le fait que nos deux yeux sont espacés légèrement l’un de l’autre.

En conséquence, chaque œil a sa propre vision différente du monde lorsque nos yeux s’alignent

sur un objet. L’information recueillie sur la profondeur relative est envoyé du cortex visuel aux

centres moteurs du tronc cérébral qui sont responsables des commandes motrices appropriées

pour le mouvement des yeux. Cependant, la manière avec laquelle les neurones du tronc

cérébral emploient l'information de profondeur fournie par le cortex visuel pour viser chaque

œil sur un objectif visuel avec précision reste fortement controversée. Cette thèse étudie

comment les différents neurones commandent le mouvement de chaque œil lorsque nous

regardons des objets localisés à profondeurs différentes.

Pour commencer, j’ai étudié les signaux produits par différents neurones oculomoteurs lors

des saccades conjuguées et disconjuguées et j’ai évalué le rôle des muscles antagonistes sur la

précision du positionnement binoculaire. J’ai constaté que le modèle de premier ordre décrit

adéquatement les décharges neuronales des neurones oculomoteurs lors des saccades

conjuguées. Cependant, j’ai trouvé que les sensibilités conjuguées ne peuvent pas être

employées pour prévoir les réponses des saccades disconjuguées lors de ces dernières. Au lieu

de cela, la majorité des neurones préfèrent le codage de la direction de mouvement de l'œil

ipsilatéral.

Par la suite, j’ai étudié les signaux portés par les neurones phasiques lors des saccades conjuguées et disconjuguées. J’ai montré que la majorité des neurones phasiques codent dynamiquement le mouvement de l'œil ipsilatéral. En plus, j’ai constaté que le générateur de la saccade déchargeant en bouffée porte toute la commande de vergence nécessaire pour former l'activité des neurones moteurs de l’abducens auxquels ils projettent. Ces résultats correspondent

xv au modèle qui a classiquement supposé que les structures de saccades « conjuguées » du tronc cérébral oculomoteur sont à la base de la facilitation de vergence en fournissant les commandes monoculaires de saccades aux noyaux d'abducens lors des saccades, alors qu’une voie séparée est employée pour l’ajustement de l’alignement oculaire suite aux composantes saccadiques du mouvement.

Par la suite, j’ai examiné la prévision qui si les commandes monoculaires initiées par les neurones phasiques sont importantes pour faciliter les vitesses de vergence lors des saccades horizontaux, alors qu’ils devraient également contribuer à faciliter la vergence lié aux saccades verticales lorsque les composantes conjuguées du mouvement sont négligeables. J’ai constaté que les neurones phasiques sont également bien adaptés pour faciliter la vergence pendant les mouvements oculaire de saccades verticales.

En dernier, j’ai employé la configuration d’électrode simple pour l’enregistrement et des techniques de microstimulations pour étudier le rôle du colliculus supérieur (SC) dans les mouvements oculaires de vergence. J’ai fourni l'évidence que les différents neurones dans le SC rostral codent pour les changements d'angle de vergence. Ces résultats suggèrent qu'il existe au niveau du SC rostral, un groupe de neurones qui encodent les mouvements de vergence lent.

L'activation du SC rostral montre la capacité à positionner précisément chacun des yeux lors d'une fixation de cible dans l'espace en 3-dimensions dans le but d'assurer la stéréoscopie.

Dans l'ensemble, je démontre qu'il existe des circuits neuronaux distinctifs pour le contrôle des mouvements de vergence rapides et lents. Ces résultats sont contradictoires avec la vision traditionnelle du "cablage" du cerveau avec des réseaux indépendants pour le contrôle des mouvements conjugués et de vergence. Ils suggèrent aussi la nécessité d'une remise-à-jour des

xvi manuels scientifiques afin de souligner les différences physiologiques entre les circuits neuronaux contrôlant les mouvements de vergence rapides et lents.

xvii

Acknowledgements

Throughout my studies at McGill I have had the opportunity to work and learn from a number of wonderful and extremely talented individuals. I am thankful to each and every member of the Cullen lab, past and present. I consider so many of my lab mates not only as colleagues but also as such dear friends. I thank everyone for their continued help and support.

In particular, Jessica Brooks, who proves to me time and time again that she can overcome, tackle – and most importantly, solder! - anything life throws at her; Moshen Jamali, proofreading master and stock market guru; Diana Mitchell, my LFP confrere; my French postdocs: Mathieu

Beraneck, Jerome Carriot, and Corentin Massot, who have all proven to be such unique and motivated individuals; Alexis Dale and Adam Schneider, the enthusiastic new members of the

Cullen lab; Ariana Andrei, who continues to inspire and challenge my creative spirit; and of course, Soroush Sadeghi, the father of single unit recording, who showed me the ropes, helped with all the knots and taught me what it means to be a real scientist. I would also like to thank all the members and students in the Department of Physiology, in particular, Ana, Tara and Christine for their unrelenting help and support.

It goes without saying that I could have never made it this far without the continuous support and scientific guidance of my supervisor Dr. Kathleen Cullen. She took me into the lab when I was a very young and naive undergraduate physiology student and provided me with thoughtful and daily guidance. She has been particularly encouraging throughout my studies and has given me the opportunity to present my work at numerous international conferences. Through her generosity, passion and dedication to the sciences, she moulded me into the scientist that I am today.

xviii

I would like to thank my committee members: Drs. Mimi Galiana, Dan Guitton and Doug

Watt for their insight, encouragement, and advice. I am also very fortunate to have collaborated with Dr. David Waitzman, a neurologist and exceptionally dedicated and talented scientist, who introduced me to the mysteries of the midbrain and helped bridge the gap between “the bench” and the clinic.

The wellness of the animals used in research is the result of the hard work and compassion of Steve Nuara and all of the animal technicians at the McGill Animal Care Center. The success of our experiments is due to the unrelenting and extraordinary efforts of Walter Kucharski. The skills of “Uncle Wally’s” go way beyond building and mending the equipment in the lab – he makes himself available to help and support each and every student with whom he works.

Finally, I would like to thank all of my dear friends and family: My husband David, Mom,

Dad, Barbara, Morty, Cecily, Vincent, Lindsay, Denise, Amanda, Ana and Lyn (who not only supported and encouraged me throughout this journey, but also helped entertain baby Mila so that

I could assemble my research into a complete and cohesive document!). And I have to thank my little girl Mila, who has no idea why I spend so many hours staring at a computer, but was patient and understanding nonetheless.

And, of course, I thank all the reviewers of this thesis, and also the reviewers of previous published manuscripts for their feedback and insight. And I thank my animals, past, present and future, whose lives made this research possible. I hope that their sacrifice helps inspires us, and helps us all better understand the mysteriousness of the infamous “black box”.

Marion

xix

Contribution of Authors

Chapter 1 was adapted from a book chapter that is in press: The Handbook on Eye

Movements, Chapter 5, in press. I prepared the first draft of the manuscript (text and figures). My thesis supervisor, Dr. Kathleen Cullen provided guidance and instruction throughout the preparation of the final manuscript.

Chapter 2 contains a published manuscript: Van Horn MR and Cullen KE. Dynamic

Characterization of Agonist and Antagonist Oculomotoneurons. Journal of Neurophysiology 102:

28-40, 2009. For this study, I trained the monkeys to perform the necessary tasks, collected the neural recordings, analyzed the data, and prepared the first draft of the manuscript (text and figures). My thesis supervisor, Dr. Kathleen Cullen provided guidance and instruction throughout the different stages of the study, including the experimental design, surgeries, data analysis and the preparation of the final manuscript.

Chapter 3 contains a published manuscript: Van Horn MR, Sylvestre PA and Cullen KE,

The Brain Stem Saccadic Burst Generator Encodes Gaze in Three-Dimensional Space. Journal of

Neurophysiology 99: 2606-2616, 2008. For this study, I trained monkeys to perform the required tasks, collected and analyzed the final data used in this study, and collaborated in preparing the final manuscript (text and figures). Dr. P.A. Sylvestre collected some of the neural recordings, designed the neural simulation and collaborated in preparing the first draft of the manuscript. My thesis supervisor, Dr. Kathleen Cullen provided guidance and instruction throughout this study, including the initial set-up, surgeries, the neural recordings, the analysis and the preparation of the final manuscript.

Chapter 4 contains a published manuscript: Van Horn MR and Cullen KE. Dynamic

Coding of Vertical Facilitated Vergence by Premotor Saccadic Burst Neurons. Journal of

Neurophysiology 100:1967-1982, 2008. For this study, I designed and implemented the target display, trained the monkeys to perform the necessary tasks, collected the neural recordings, analyzed the data, and prepared the first draft of the manuscript (text and figures). My thesis supervisor, Dr. Kathleen Cullen provided guidance and instruction throughout the different stages of the study, including the experimental design, the surgeries, data collection, data interpretation and the preparation of the final manuscript.

Chapter 5 contains a manuscript in preparation for submission: Van Horn, M.R.,

Waitzman, D.M., and Cullen, K.E., Neurons in the Rostral Superior Colliculus Encode Vergence

Eye Movements. For this study, I trained the monkeys, collected the single unit and stimulation data, analyzed the data and prepared the first draft of the manuscript (text and figures). Dr. David

Waitzman participated in designing the experiment, collecting neural and stimulation data. My thesis supervisor, Dr. Kathleen Cullen provided guidance and instruction throughout the different stages of the study, including the experimental design, the surgeries, data collection, data interpretation and the preparation of the final manuscript.

Appendix A contains a published manuscript: Van Horn, M.R., Mitchell, D.E., Massot, C., and Cullen, K.E., Local Neural Processing and the Generation of Dynamic Motor Commands within the Saccadic Premotor Network. J. Neuroscience 30(32):10905-17, 2010. For this study I trained the monkeys, collected the neural data and wrote the first draft of the manuscript (text and figures). Diana Mitchell participated in collected some of the data and collaborated in analyzing the data and making figures. Dr. Corentin Massot performed the stimulus reconstruction described in Fig. 5-7. As my thesis supervisor, Dr. Kathleen Cullen played an instrumental role in helping design the experiment, perform the surgeries, collect, analyze and interpret the data, as well as prepare the final manuscript.

xxi

List of abbreviations

III Oculomotor nucleus

IV Trochlear nucleus

VI Abducens nucleus

ABN Abducens nucleus

AMN Abducens motoneuron

INN Internuclear neuron

AIN Abducens internuclear neuron

INN Internuclear neuron

OMN Oculomotoneuron

MLF Medial longitudinal fasciculus

PPRF Paramedian pontine reticular formation

SBN Saccadic burst neuron

EBN Excitatory burst neuron

IBN Inhibitory burst neuron

BT Burst-tonic neuron

EH Eye-head neuron

PVP Position-vestibular-pause neuron cMRF central Mesencephalic Reticular Formation

SC Superior Colliculus rSC rostral Superior Colliculus

SOA Supraoculomotor area

FEF Frontal eye fields

xxii

Chapter 1

General Introduction & Literature Review

1.1 Overview and Conceptual Framework

In the last century, research on the brain has provided valuable information about how we see the world clearly in three dimensions. Seeing in 3D relies largely on the fact that our eyes are spaced slightly apart. As a result, when our eyes are aligned on an object each eye has its own, slightly different view of the world. To prove this to yourself, try closing each eye in turn. As you switch your vision from one eye to the next you will notice that the world will appear to move from side to side. Remarkably, during normal viewing, the brain takes the image from each eye and combines them so that only one clear image is perceived.

During normal viewing, light first enters the pupil of each eye and is focused by the cornea and lens onto the retina – a group of cells connected by synapses at the back of the eye.

Light sensitive cells within each retina convert light into an electrical signal that can be read by the brain. Information about the image is then transmitted to the brain via the optic nerve, where it is processed to build a visual representation of the world (Fig. 1.1). Initially, information from

Fig. 1.1: Schematic diagram of how visual information is transmitted from the eye to cortex and eventually to the brainstem where it is converted to an appropriate motor output, which is sent back to the eye. Excitatory burst neurons (EBNs) for horizontal saccades lie in the paramedian pontine reticular formation (PPRF), and for vertical and torsional lie in the rostral interstitial nucleus of the medial longitudinal fasciculus (riMLF). Oculomotoneurons lie in the abducens nucleus (ABN; VI), the oculomotor nucleus (OMN; III) and the trochlear nucleus (IV). INC: interstitial nucleus of Cajal; NRTP: nucleus reticularis tegmentis pontis; nPH: nucleus prepositus hypoglossi. VN: vestibular nucleus. LGN: Lateral geniculate nucleus 2

each eye is processed by the brain separately. In order to calculate relative distances, however, information from the two eyes must come together. This convergence eventually occurs at the primary visual cortex, a grouping of neurons at the back of the brain in the occipital lobe. For example, some neurons respond optimally to images that appear in front of a fixation point, while others respond optimally to images that appear behind a fixation point (Barlow et al.

1967).

Information about relative depth is then sent to the motor control centers in the brainstem, which are responsible for generating appropriate motor commands to move the eyes (Fig. 1.1).

Surprisingly, how the neurons in the brainstem, which are responsible for moving the eyes, use the depth information supplied by visual cortex to precisely aim each eye on a visual target remains highly controversial. Accordingly, the overall goal of this thesis was to understand how individual neurons realign gaze to targets located throughout 3-dimensional space.

Over a century ago, a largely theoretical and philosophical debate arose between two

German physician-scientists, Ewald Hering and Hermann von Helmholtz. On the one hand,

Helmholtz argued that we learn to precisely aim both eyes at a visual target (von Helmholtz

1962). To fully appreciate Helmholtz‟s argument it is helpful to consider how chameleons move their eyes. A chameleon has the unique ability to point each eye in different directions.

Essentially Helmholtz‟s proposal was that the brain is wired so that we, like chameleons, can move each eye independently, but over time we have learned to move them together so that both eyes are accurately aimed at the same point, thus resulting in clear 3-dimensional viewing. On the other hand, Hering argued that we are born with the ability to move our eyes in a coordinated fashion (Hering 1977). A common analogy used to explain this argument is that our eyes are like the reins of a horse – if you pull on one side the other side moves by the same amount, in the

same direction. In other words, Hering‟s proposal suggested that the eyes should be seen as a single organ rather than two separate entities.

Based on these initial arguments neurophysiologists have made more precise predictions about how the brain controls binocular eye movements (Maxwell and King 1992; Mays 1984;

Ono et al. 1978; Rashbass and Westheimer 1961b; Sylvestre et al. 2003; Zhou and King 1998).

To better appreciate the framework of these predictions, the anatomy and mechanical properties of the eye and orbit, as well as a description of the neural pathways that control eye movements, are first briefly reviewed. The evidence that initially supported Hering‟s hypothesis is then considered and this is followed by a review of recent work that has subsequently challenged this view. Finally, the experiments in this thesis, which further probe the validity of Hering‟s hypothesis, are introduced.

1.2 The extraocular eye muscles and their innervations

The oculomotor system is one of the better understood motor systems. In contrast to locomotion or limb control, the structural features of the oculomotor system are mechanically simple. The precise position of each eye in its orbit is controlled by six extraocular muscles.

Horizontal eye movements are controlled by the medial and lateral rectus muscles, while vertical and torsional movements are controlled by the superior and inferior rectus muscles and superior and inferior oblique muscles. Together, these three pairs of muscles permit the eye to rotate with three degrees of freedom. The extraocular muscles are innervated by three groups of motoneurons located in the brainstem (Fig. 1.1). The lateral rectus is innervated by the abducens nerve (cranial nerve VI), the medial, inferior and superior recti are innervated by the oculomotor nerve (cranial nerve III) and the inferior and superior obliques are innervated by the trochlear

nerve (cranial nerve IV). Approximately half of the neurons within the abducens and oculomotor nucleus are internuclear neurons, which project via the medial longitudinal fasciculus (MLF) to the contralateral oculomotor or abducens nucleus, respectively (Buttner-Ennever and Akert

1981; Maciewicz et al. 1975).

A novel feature of the oculomotor system is that each extraocular muscle is comprised of two distinct layers, a global layer and an orbital layer. These layers contain two types of muscle fibers: twitch and non-twitch fibers (Buttner-Ennever et al. 2001; Spencer and Porter 1988).

Recent retrograde studies have shown that twitch fibers receive innervations from large motoneurons, which lie within the oculomotor nuclei, whereas as non-twitch fibers receive innervations from small motoneurons, which tend to lie separately around the periphery of the nucleus (Buttner-Ennever et al. 2001; Ugolini et al. 2006). In turn, larger motoneurons receive strong innervations from premotor sources involved in fast eye movements (e.g., the saccadic burst generator) while recent studies of the abducens nucleus suggest small motoneurons may receive innervations from premotor sources involved in executing slow eye movements (e.g., vestibular nucleus, prepositus hypoglossi and supraoculomotor nucleus) (Ugolini et al. 2006).

In general, to better understand how the eye moves in the orbit it is helpful to think of a mechanical spring and a syringe. The eye is surrounded by muscles, which have elastic properties and, similar to a spring, a force is needed to actively extend the muscles to move the eye to a new position. Furthermore, a force is required to keep the eye at the new position since the elastic properties of the muscle will naturally cause the eye to recoil, or spring-back to its primary position. The force generated by the eye muscles must also compensate for the passive restraining forces of the oculomotor plant (e.g., the eyeball and the surrounding orbital tissue). If you consider the plunger of a syringe, a force is needed to move the plunger and the required

force will increase as a function of the viscosity of the fluid within the syringe. For example, it is harder to push honey through the syringe than water (Fig. 1.2; left panel).

Engineering has shown that the mechanics of a “syringe-spring” (visco-elastic) system can be modeled using a relatively simple first order equation:

F = kL + r(dL/dt) where the force (F) generated to move the spring a specific length (L), and the syringe a certain velocity (dL/dt), is related to the stiffness (k) of the spring and the viscosity (r) of the syringe.

By recording the neural activity of extraocular motoneurons during eye movements scientists found their activity is appropriate for overcoming the viscous-elastic properties of the oculomotor plant. Accordingly, the relationship between firing rate of an extraocular motoneuron and eye motion can be described using a simple, first order model similar to that which describes the syringe-spring system. In particular, the firing rate of a neuron is related to a given neuron's sensitivity to eye position and velocity:

 FR(t)  b  k E(t td )  r E(t td ) where FR is instantaneous firing rate, b, k and r are constants which represent the bias, and the

 neuron‟s eye position and velocity sensitivities, respectively. E(t) and E(t) refer to the instantaneous position and velocity of the eye and td is the neuron‟s dynamic lead time (Fig. 1.2; right panel).

This model has been shown to adequately describe the discharge characteristics of motoneurons in the abducens nucleus (Sylvestre and Cullen 2002). However, an assumption that has been made is that the agonist drive, sent to the contracting muscle, dictates the dynamics of a

Fig. 1.2: Illustration of how a syringe-spring system is a helpful for understanding how the eye moves in the orbit. Left: The syringe-spring (i.e., visco-elastic) system can be modeled using a first order equation where the force generated to move the spring a specific length (L), and the syringe a certain velocity (dL/dt), is related to the stiffness (k) of the spring and the viscosity (r) of the syringe. Right: The activity of a neuron needed to move the eye a certain amplitude (E) and velocity (dE/dt) is related to a neuron's sensitivity to eye position (k) and velocity (r).

given eye movement. In reality, it is the ratio of the agonist and antagonist neuronal activity that precisely positions the eyes. Thus, the first goal of this thesis was to create a realistic model of eye movements by considering the contribution of the agonist as well as the antagonist muscle.

To fully assess the drive to the medial rectus muscle I recorded the activity of medial rectus motoneurons when the muscle was contracting and relaxing. The results of this study are presented in Chapter 2 of this thesis.

1.3 Classification of eye movements

In the early 1900‟s Raymond Dodge described five distinct classes of eye movements used to redirect or stabilize a visual image on the fovea (Dodge, 1903). Three classes of eye movements namely, (i) saccades, (ii) smooth pursuit, and (iii) vergence, are voluntarily initiated to redirect gaze to a particular object in the visual field. Saccades, the fastest and most common type of eye movement, are constantly used to redirect gaze between stationary objects and bring images onto the fovea. Typically saccades have been studied when looking between objects that are located at the same relative depth. During these conditions the eyes rotate in the same direction and by the same angle and are known as “conjugate” saccades (Fig. 1.3; left panel).

Smooth pursuit eye movements are considerably slower movements (<100deg/sec) that are used to track targets moving across a stationary visual background. Vergence eye movements rotate the two eyes in opposite directions to fixate on targets located at different depths (Fig. 1.3; right panel). The remaining two classes of eye movements, the (iv) vestibulo-ocular reflex and the (v) optokinetic reflex, are reflexive and function to hold images stationary on the retina. The properties of these five classes of eye movements are summarized in Table 1.1.

Fig. 1.3: Schematic illustration of eye position during an example conjugate saccade (left) and slow symmetrical vergence eye movement (right).

1.4 The neural control of disconjugate saccades: Hering versus Helmholtz

During most natural, voluntary orienting and tracking behaviours two or more types of eye movements often work together to control gaze. For example, to quickly and accurately redirect gaze between near and far targets we typically combine saccadic and vergence eye movements. During these eye movements, termed disconjugate saccades, the eyes rotate by different amounts and with different trajectories. How the brain generates precise movements of the eyes during disconjugate saccades remains a controversial topic of debate. The basic schemas showing the theoretical frameworks of binocular control, as originally inspired by Hering and

Helmoholtz, are illustrated in Fig. 1.4. Based on Hering‟s theory, neurons within a “saccadic” pathway should be exclusive to conjugate movements and a separate “vergence” pathway should exist to ensure binocular positions between targets located at different depths. In this schema, a summation of these two signals by motoneurons would result in accurate binocular positioning

(Fig. 1.4). In contrast, Helmholtz‟s theory suggested that the movement of each eye could be programmed independently; rather than having separate control systems for conjugate and vergence eye movements, the brainstem would contain two motor control pathways comprised of neurons whose electrical activity would explicitly command movement of the left or right eye

(right panel, Fig 1.4).

1.4.1 Evidence supporting “Hering’s Law”

In the 1950‟s the development of sophisticated experimental techniques allowed for the predictions Hering and Helmholtz‟s theories to be more directly tested. Single unit recording methods, where thin microelectrodes are inserted into the brain to measure the electrical activity of neurons, became instrumental in decoding how neurons communicate. In particular, by 11

Fig. 1.4: Theoretical frameworks of binocular control as inspired by Hering (left) and Helmholtz (right).

studying the spiking activity of single neurons (i.e., action potentials) in monkeys and cats, while simultaneously monitoring eye movements, scientists have been able to map out exactly what brain structures are involved in generating specific eye movements. Notably, there has recently been an increased interest in investigating the information that is carried by low frequency electrical activity (i.e., local field potentials; LFPs). While spiking activity represents the action potentials produced by a neuron, LFPs are generally thought to reflect the summed activity of synaptic potentials, dendritic spikes and spike afterpotentials occurring around the tip of the recording electrode (Mitzdorf 1985; 1991; 1987). Accordingly, LFPs are considered to reflect the input to a given brain area whereas spiking activity represents the output that is sent to other parts of the brain. In Appendix A of this thesis I explore how simultaneous recordings of spiking activity and LFPs can provide an effective means for evaluating the local computations that take place to produce accurate eye movement commands.

Initially, neurological evidence accumulated in favour of Hering‟s hypothesis. Conjugate saccades and vergence eye movements were studied in isolation and this subsequently led to the identification of neurons appropriate for encoding conjugate saccades and a separate population of neurons encoding changes in vergence position. Overall, these initial findings supported

Hering‟s hypothesis that there exist separate neural pathways for conjugate and vergence eye movements. Below, the premotor pathways that control conjugate saccades and vergence eye movements, when preformed in isolation, are first reviewed. This is followed by a review of work that has challenged this traditional view that conjugate saccades and vergence are generated by independent neural subsystems. To date, whether individual neurons, which have been shown to be essential for generating conjugate saccades, contribute to generating disconjugate saccades still remains a controversial debate. Thus, the second goal of this thesis

was to investigate how individual saccadic neurons contribute to generating unequal movements of the eyes during disconjugate saccades. Indeed, the results presented in Chapters 2-4 provide significant insight into how individual neurons in the brainstem precisely align gaze to targets located throughout 3-dimensional space.

1.5 The premotor circuitry of conjugate saccades

Saccades are the fastest, most common type of eye movements that are made. They are constantly used to redirect gaze between stationary objects to bring images onto the fovea (Fig

1.3; left panel). Accurate saccades can be made in less than 50ms. As described above, to produce such rapid eye movements a motoneuron must generate a burst (or “pulse”) of action potentials to overcome the viscous drag on the eye as it moves in the orbit. Firing rate bursts can reach frequencies as high as 500spikes/s. This is in striking contrast to other types of motoneurons, such as spinal motoneurons, which reach firing rates of only 10-30spikes/s. Once the eye reaches its final position at the end of a saccade, it is then held steady by a sustained motoneuron firing that produces a fixed contraction of the extraocular muscle.

The difference between the neuron‟s sustained discharge rate at the initial and final eye positions is referred to as the “step”. Thus, to a first approximation, the neural control of saccades requires the generation of a “pulse-step” command signal. As reviewed below, neural correlates for the pulse and step command signals have been identified and well studied during horizontal conjugate saccades (i.e., when targets are located at a distance that produces equal amplitude rotations of the two eyes, see Fig. 1.3 left panel). In the section 1.5 of this chapter I describe the limited number of studies that have begun to investigate these same neurons during

disconjugate saccades (i.e., when targets are located at different depths and produce non- identical rotations of the eyes).

1.5.1 Saccadic Burst Neurons

The “pulse” or burst signal required to drive horizontal conjugate saccades is generated by neurons located in the paramedian pontine reticular formation (PPRF). Physiological and anatomical tracing studies have demonstrated that excitatory burst neurons (EBNs), in the rostral portion of the PPRF are the source of the high frequency burst of discharges responsible for driving the lateral rectus muscle (Sasaki and Shimazu 1981; Strassman et al. 1986a).

EBNs work together with a second group of inhibitory burst neurons (IBNs) in the caudal pontine reticular formation, which silence the antagonist motoneurons (i.e., contralateral abducens) (Hikosaka et al. 1978; Hikosaka and Kawakami 1977; Strassman et al. 1986b;

Yoshida et al. 1982) Yoshida et al., 1982; Strassman et al., 1986b)[reviewed in: (Scudder et al.

2002; Shinoda et al. 2008)] (Fig. 1.5; left panel). Typically, EBNs and IBNs [collectively referred to as saccadic burst neurons (SBNs)] burst most vigorously for ipsilaterally directed saccades, preceding the onset of a saccade by ~10-20ms (Cullen and Guitton 1997; Scudder et al.

1988). Far fewer spikes are observed during contralateral, oblique and vertical saccades (Cullen and Guitton 1997; Scudder et al. 1988; Van Gisbergen et al. 1981).

Another class of saccade related burst neurons termed, long-lead burst neuron (LLBN), like EBNs fire a burst of spikes prior to saccades, however their burst is preceded by a longer prelude of activity (e.g., latency 16-200ms). Unlike EBNs, LLBNs can also display visual responses, saccade amplitude sensitivity, and non-visual or motor discharge (Crandall and Keller

1985; Kaneko 2006). A detailed analysis of LLBNs has revealed that these neurons can be

further divided into three classes of neurons based on discharge characteristics and cell location, namely; excitatory LLBNs (eLLBN), dorsal LLBNs (dLLBNs) and nucleus reticularis tegmentis pontis LLBNs (nrtp LLBNs) (Crandall and Keller 1985; Kaneko 2006).

Using traditional metric-based analysis approaches, it was shown that the duration of a

SBN‟s burst was tightly correlated to the duration of the corresponding saccade; the number of spikes in a burst is related to saccade amplitude and peak burst discharge is correlated with peak saccade velocity (Cullen and Guitton 1997; 1996; Hepp and Henn 1983; Keller 1974; Luschei and Fuchs 1972; Strassman et al. 1986a; b; Van Gisbergen et al. 1981). Furthermore, an analysis of the dynamic relationship between SBN discharges and eye velocity showed that SBNs dynamically encode saccade trajectories in their spike trains (Cullen and Guitton 1997). In particular, the discharge of IBNs during saccades can be well described by the model:

 FR(t td )  b  r E(t)

where FR is instantaneous firing rate, b is the estimated bias and r is the estimated

 velocity sensitivity, E is the velocity of the eye during saccades and td is the neuron‟s dynamic lead time. However, as noted above these initial experiments were done during conjugate saccadic eye movements – i.e., when the movement of the two eyes was identical and thus it was impossible to determine if the neuronal activity was preferentially related to the movement of one eye or the other.

1.5.2 Omnipause neurons

The generation of saccades relies on an additional class of brain stem neurons called omnipause neurons (OPNs). These neurons, which lie around the midline of the caudal pontine

reticular formation in the nucleus raphe interpositus (RIP), are thought to act as an inhibitory gate for saccades (Fig. 1.5; left panel). Using glycine as their primary neurotransmitter (Horn et al. 1994), OPNs tonically inhibit both horizontal and vertical SBNs(Buttner-Ennever et al. 1988;

Furuya and Markham 1982; Horn et al. 1994; Langer and Kaneko 1990; Nakao et al. 1980;

Ohgaki et al. 1989; Strassman et al. 1987). OPNs discharge at a constant rate when gaze is steady and pause their firing during saccades in all directions. Additionally, the duration of the pause is well correlated to saccade duration (Keller 1974; Luschei and Fuchs 1972). The role of OPNs in controlling fixation, and inhibiting saccadic eye movements is well supported by the results of studies, which have shown that microstimulation of the OPN region of the RIP cause complete cessation of eye movements (Keller 1974).

Intracellular recordings have revealed tight relationships between the durations of OPN hyperpolarizations and saccades as well as the amplitudes of OPN hyperpolarizations and saccade velocities. Interestingly, the dynamic time course of OPN hyperpolarization also resembles the velocity profile of the corresponding eye movement (Yoshida et al. 1999). These findings suggest that OPN pauses are initiated by inputs carrying an eye velocity signal. Likely candidates are contralateral SBNs in the PPRF and ipsilateral superior colliculus both of which have been shown to have projections to OPNs (Scudder et al. 1996; Shinoda et al. 2008;

Takahashi et al. 2005).

1.5.3 Oculomotor integration

In order to hold the eyes steady at the end of a saccade the pulse command generated in the PPRF and MRF must be transformed into an appropriately scaled tonic discharge. The SBNs in the PPRF project to neurons in the nucleus prepositus hypoglossus (nPH), which serves, in

Fig. 1.5: Schematic of pathways that mediate saccades and vergence. Left: Saccadic eye movements can reach speeds of up to 900 deg/s (top).The bottom panel illustrates the premotor pathway for producing saccades. Command signals, issued in the deep layers of the superior colliculus, are delivered to burst neurons (BN) in the PPRF. Right: Symmetric vergence eye movements, which are characteristically much slower than saccades , move the eyes in opposite directions to fixate objects lying at varying depths (top). The bottom panel illustrates how projections from near response neurons (NR) in the mesencephalic reticular formation (MRF) are thought to contribute to the generation of slow vergence eye movements. Cerebellum, cortex and the superior colliculus are likely sources of vergence information. 18

part, as a neural integrator for horizontal saccades (Cannon and Robinson 1987)[reviewed in

(Fukushima et al. 1992; McCrea 1988)](Fig. 1.5; left panel). Neurons in the nPH and interstitial nucleus of Cajal (INC) encode robust position signals and can also burst during saccades. These neurons are commonly referred to as burst-tonic neurons (BTs). Velocity integration is leaky or imperfect and as a result, in the dark, the eyes gradually drift back to center from eccentric positions with a time constant of ~25s. A lesion of either the nPH or INC results in an inability to hold the eyes at a new position after a saccade. Taken together, the results of anatomical, lesion, and single-unit recordings in nPH or INC are consistent with their proposed role in integrating the pulse command (Cannon and Robinson, 1987).

1.6 The premotor circuitry of saccade-free vergence

When redirecting gaze to objects along the mid-saggital plane the eyes rotate by the same angle, in opposite directions. This type of eye movement is known as symmetrical, or saccade- free, vergence (Fig 1.3; right panel). The dynamics of saccade-free vergence have been shown to be much more sluggish than saccades. For example, maximum vergence velocities are generally

<60deg/s. At the brainstem level, a group of neurons, called near-response neurons, have been located in the midbrain reticular formation (MRF) whose discharge is proportional to vergence angle when tracking visual targets located along the midline (Judge and Cumming 1986; Mays

1984; Zhang et al. 1992) (Fig. 1.5; right panel). Some neurons have been found to increase their activity during symmetric convergence (i.e., convergence neuron) while others decrease their activity during symmetric convergence (i.e., divergence neuron) however neither group responded during conjugate saccades. Notably, vergence eye movements are accompanied by changes in accommodation, which changes the shape of the lens and helps the eyes to focus on 19

near objects. To determine if near-response neurons were related to vergence, accommodation, or a combination of the two, experiments were designed to partially dissociate the two responses.

It was found that the majority of near-response neurons were related to both vergence and accommodation (Gamlin 1999; Judge and Cumming 1986).

Anatomical studies have shown that near response neurons receives input from two deep cerebellar nuclei, the posterior interposed and fastigal nucleus. Moreover, near-response neurons were found to be antidromically activated from the medial rectus subdivision of the oculomotor nucleus (Zhang et al. 1991; Zhang et al. 1992). However to date, no vergence related neurons have been found to be antidromically activated from the abducens nucleus. Studies also identified a second class of neurons in the MRF, which have activity related to vergence velocity

(Mays et al. 1986). Both convergence and divergence neurons have been described, although far fewer divergence neurons than convergence neurons have been reported. Vergence burst neurons were identified in two regions of the MRF. One group was located in proximity to the neurons that encode vergence angle and a second group was located in a more ventral area (Mays et al.

1986).

Overall the initial neurological evidence supported Hering‟s hypothesis that there exist separate neural pathways for conjugate and vergence eye movements. However, while this appeared to be an elegant solution to the problem of binocular control, the findings of recent experiments have provided reasons to question this view.

1.7 Evidence against “Hering’s Law”

First, behavioural studies have shown that during disconjugate saccades, vergence velocities reach far greater values than would be expected. Notably, when vergence eye

movements are made alone, eye velocities reach maximum values of ~60deg/sec whereas vergence velocities during disconjugate saccades can reach values of >200deg/sec – even when the required change in vergence angle is identical (Busettini and Mays 2005a; Collewijn et al.

1997; Enright 1984; Enright 1992; Maxwell and King 1992; Ono et al. 1978; Oohira 1993; van

Leeuwen et al. 1998; Zee et al. 1992). These behavioral findings challenge the traditional

“Hering” view that there exists separate vergence and conjugate oculomotor subsystems and suggests that conjugate and vergence commands are integrated upstream of the level of the motoneurons.

Second, to further probe the validity of Hering‟s hypothesis more recent research has investigated how individual neurons, in the commonly assumed “conjugate” saccadic pathway, respond during disconjugate saccades when the eyes are shifted to targets located at different depths and eccentricities. In particular, by simultaneously minimizing the movement of one eye while monitoring brain activity, researchers have been able to assess whether the electrical activity of an individual neuron was related to the movement of one eye or the other. These experiments were particularly novel since previous studies, as described in the above sections, had only recorded the neural activity of eye movement neurons during conjugate saccades.

In 1998, King and colleagues made the first observation that the number of action potentials produced by the SBNs of the PPRF during a disconjugate saccade was better correlated to the movement of an individual eye, than to the conjugate component of the movement. More recently, Sylvestre and Cullen (2002, 2003) extended these findings to address the key question of what eye movement command signals are dynamically encoded by other premotor and motoneurons (e.g., BTs and ABNs) during conjugate saccades, versus disconjugate saccades. To determine whether neuronal commands were more consistent with a Hering (i.e.,

conjugate / vergence commands) or Helmholtz (i.e., right eye / left eye commands) inspired framework, Sylvestre and Cullen (2002) first determined whether the same simple mathematical model of a neuron‟s response during conjugate saccades (see eq 1.2) was also applicable during disconjugate saccades. In the case that Hering‟s hypothesis were true, a model describing the relationship between a neuron‟s response and conjugate component of the movement [e.g., (left eye velocity + right eye velocity)/2] should be the same for both types of eye movements.

Instead, they found that this prediction generally failed. Specifically, they found that the neuronal responses of ABNs and BTs reliably encoded the movement of an individual eye rather than the conjugate component of each eye movement.

1.8 Research Goals

There is now accumulating evidence that suggests that neurons within the previously assumed “conjugate” saccadic premotor pathway may actually encode the movement of an individual eye (Cova and Galiana 1995; King and Zhou 2002; Sylvestre et al. 2003; Sylvestre and Cullen 2002; Zhou and King 1996; 1998). However, these findings cannot rule out the alternative possibility that the overall contribution of the saccadic circuitry is relatively unimportant compared to that of the vergence subsystem. In fact, the most recent model of saccade-vergence interactions suggests that SBNs exclusively encode conjugate saccadic dynamics that interact with the vergence subsystem (Busettini and Mays 2005b). In this thesis, I present a series of experiments that further investigate how individual saccadic neurons contribute to generating unequal movements of the eyes during disconjugate saccades.

Specifically, the main goal of the studies presented in this thesis was to determine whether neurons within the traditionally defined conjugate premotor pathway are in fact strictly encoding

solely conjugate commands and whether they significantly contribute to the generation of disconjugate saccades. To perform these studies an experimental setup was designed to monitor brain activity during 3D eye movements.

In Chapter 2, I provide the first dynamic characterization of medial rectus motoneurons during conjugate and disconjugate saccades and evaluate the role the antagonist muscle plays in accurate binocular positioning. In Chapter 3, I describe the discharge dynamics of EBNs and

IBNs during a task where monkeys were required to look at targets aligned with one eye such that the resulting conjugate movement was minimized. I evaluate whether a binocular expansion of the model used to describe SBNs during conjugate saccades is needed to accurately estimate

SBN discharges during disconjugate saccades. Moreover, using a neural simulation I further evaluate whether this command signal from the premotor saccadic circuitry is in fact sufficient to drive the target extraocular motoneurons during disconjugate saccades. In particular, this simulation explicitly tests whether additional input (i.e., from a separate vergence subsystem) is required to shape the activity of abducens motoneurons during disconjugate saccades. In Chapter

4, I test the prediction that if the monocular commands issued by the saccadic burst neurons are important for facilitating vergence velocities during horizontal saccades they should also contribute to facilitating vergence velocities associated with a vertical saccade made between near and far targets for which the conjugate component of the movement is negligible.

Specifically, I describe the discharge dynamics of EBNs and IBNs during a novel task that evokes a saccadic vergence command but no horizontal conjugate saccade command. Finally, in

Chapter 5, I investigate whether the superior colliculus contributes to the development of neural signals that are suitable for controlling disconjugate saccades or whether their discharge is strictly related to conjugate control.

Chapter 2

Dynamic characterization of oculomotoneurons during conjugate and disconjugate eye movements

In this chapter, I describe the discharge dynamics of medial rectus oculomotoneurons during conjugate and disconjugate saccades and fixation. Prior studies have shown that a first-order linear model provides an adequate description of the discharge dynamics of lateral rectus motoneurons. I provide the first detailed characterization of the dynamics of individual medial rectus motoneurons during conjugate and disconjugate saccades. The responses of medial rectus motoneurons are compared to those of lateral rectus motoneurons to get a better understanding of how the agonist and antagonist muscles are working together to ensure accurate three- dimensional viewing. This chapter was adapted from: Van Horn MR and Cullen KE. Dynamic

Characterization of Agonist and Antagonist Oculomotoneurons. Journal of Neurophysiology

102: 28-40, 2009.

2.1 ABSTRACT

In this report we provide the first quantitative characterization of the relationship between the spike train dynamics of medial rectus oculomotoneurons (OMNs) and eye movements during conjugate and disconjugate saccades. We show that a simple, first-order model (i.e., containing eye position and velocity terms) provided an adequate model of neural discharges during both

ON and OFF-directed conjugate saccades, while a second order model, which included a decaying slide term, significantly improved the ability to fit neuronal responses by ~10% (p<0.05). To understand how the same neurons drove disconjugate eye movements, we evaluated whether sensitivities estimated during conjugate saccades could be used to predict responses during disconjugate saccades. For the majority of neurons (68%) a conjugate-based model failed and instead neurons preferentially encoded the position and velocity of the ipsilateral eye. Similar to our previous results with ABNs, we also found that position and velocity sensitivities of OMNs decreased with increasing velocity and the simulated population drive of OMNs during disconjugate saccades was less (~10%) than during conjugate saccades. Taken together, our results provide evidence that the activation of the antagonist, as well as agonist motoneuron pools must be considered to understand the neural control of horizontal eye movements across different oculomotor behaviors. Moreover, we propose that the under-sampling of smaller motoneurons (e.g., non-twitch) was likely to account for the missing drive observed during disconjugate saccades; these cells are thought to be more specialized for vergence movements, and thus could provide the additional input required to command disconjugate eye movements.

2.2 INTRODUCTION

In order to precisely accomplish a desired eye movement, the appropriate neural command must be sent to the extraocular muscles. During saccades extraocular motoneurons generate a burst of activity (i.e., pulse) that compensates for the resistance of the muscles and orbital tissues. Then, at the end of the saccade, motoneurons generate a higher tonic discharge

(i.e., step) in order to resist against the natural elastic recoil of the orbital tissues and hold the eye at a new desired position. Robinson (1964) proposed that the relationship between motoneuron activity and muscle force could be approximated using a fourth-order linear model. Subsequent analysis of the “pulse-step” nature of medial and lateral rectus motoneurons led to the prediction that a first-order model, including the neuron‟s discharge when the eye is in the center of the orbit, as well eye velocity and position sensitivities, could provide a simpler, yet useful model for approximating the firing rate of extraoculor motoneurons during conjugate saccades (Robinson

1970; Robinson and Keller 1972).

To verify the original predictions made by Robinson and colleagues a detailed analysis of abducens motoneurons (ABNs) has since been conducted. Van Gisbergen and colleagues (1981) proposed that including an accelerating term would provide a better description of ABN discharges during saccades. However, a major limitation of the analysis used in this study was that the relative contribution of terms could not be objectively determined. More recently,

Sylvestre and Cullen (1999) evaluated the importance of each term by directly fitting the neuronal responses of ABNs during saccades. They found that the addition of both an acceleration term and a slide term, which is used to explain the exponential decay of neuron‟s firing rate, notably improved their ability to describe the discharge dynamics during saccades.

Overall, however they concluded that a first-order linear model provided an adequate description 26

of the discharge dynamics of ABNs during saccades, smooth pursuit and vestibular nystagmus.

Sylvestre and Cullen (2002) further demonstrated that this same model could be used to describe the discharge dynamics of ABNs during disconjugate saccades (i.e., saccades with changes in viewing distance and eccentricity). Notably, this model needed to be expanded to include the movement of each eye (i.e., ipsilateral and contralateral). They found the majority of ABNs preferentially encoded the position and velocity of the ipsilateral eye and that the remaining neurons encoded the motion of both eyes to various degrees (Sylvestre and Cullen 2002).

First-order models have proven valuable for describing how extraocular motoneurons control eye movements. However, an assumption that has been made in previous studies is that the agonist drive, sent to the contracting muscle, dictates the dynamics of the movement. In reality, it is the ratio of agonist and antagonist motoneurons activity that positions the eye.

Previous studies, which have compared the discharge rates of motoneurons during fixation at different depths, have revealed that for a given position of the eye in the orbit the majority of abducens neurons fire at higher rates during convergence than when gaze is relaxed (Gamlin et al. 1989; Mays and Porter 1984). Furthermore, eye velocity and position sensitivities have been found to invariably decrease as a function of the generated eye velocity (Fuchs et al. 1988;

Sylvestre and Cullen 1999). For example, slower movements (e.g., pursuit) were found to have higher sensitivities than faster movements (e.g., saccades) (Sylvestre and Cullen 1999).

Interestingly, these potentially surprising relationships can be explained if the activity of the antagonist muscle were considered. For example, during slow movements (e.g., pursuit) the majority of antagonist motoneurons continue to discharge (Sylvestre and Cullen 1999) and hence the contribution of the antagonist muscle could be providing an additional active force that opposes the agonist muscle.

In order to create a realistic model of eye movements, it is thus necessary to consider the contribution of the antagonist muscle when evaluating motoneuron activity across multiple oculomotor behaviors. However to date, studies have only provided a complete description of the discharge dynamics of lateral rectus motoneurons (i.e., ABNs) in their ON-direction. Here we provide the first detailed characterization of the dynamics of individual medial rectus neurons in the oculomotor nucleus (OMNs) during conjugate and disconjugate saccades. To fully assess the drive to the medial rectus muscle, the activity of OMNs is characterized when the medial rectus is contracting and relaxing (i.e., “ON” and “OFF” directions). During conjugate saccades, we fit the dynamic discharge of individual OMNs using an approach previously used to describe ABNs and saccadic burst neurons (Cullen and Guitton 1997; Sylvestre and Cullen 2002; 1999). We then determine whether we could use the parameters estimated during conjugate saccades to predict the activity of OMNs during disconjugate saccades. Finally, we compare the responses of

OMN to ABNs to get a better understanding of how the agonist and antagonist muscles are working together to ensure accurate three-dimensional viewing.

2.3 METHODS

2.3.1 Animal and surgical procedures

Recordings were made in two rhesus monkeys (Macaca mulatta). The monkeys were prepared for chronic extracellular recording using the aseptic surgical procedures described previously (Sylvestre and Cullen 1999). In brief, a stainless steel post was attached to the animal's skull with stainless steel screws and dental acrylic permitting complete immobilization of the animal's head. Two stainless steel recording chambers, oriented stereotaxically toward the oculomotor nucleus on the right and left side of the brainstem, were also secured to the implant. 28

To record binocular eye position an eye coil (3 loops of Teflon coated stainless steel wire, 18-20 mm diam) was implanted in each eye (Judge et al. 1980). All procedures were approved by the

McGill University Animal Care Committee and complied with the guidelines of the Canadian

Council on Animal Care.

2.3.2 Behavioral paradigms

During the experiments monkeys were head-fixed and seated in a primate chair in the dark. Monkeys were trained to follow a red HeNe laser target projected onto a cylindrical screen located 55 cm away from the monkey's eyes (isovergent, 3.5 deg convergence) and red light emitting diodes (LEDs), with intensities comparable to that of the laser target, positioned between the screen and the monkey. The timing and location of target illumination, data acquisition and on-line data displays were controlled using REX, a UNIX-based real-time acquisition system (Hayes et al. 1982).

Neuronal responses were recorded during horizontal conjugate and disconjugate saccades and fixation. Ipsilaterally and contralaterally directed conjugate saccades were elicited by stepping the laser target between horizontal positions (5-30 deg), in 5 deg increments, in predictable and unpredictable sequences. A horizontal array of 16 LEDs, which were positioned between the screen and the monkey, were used in combination with the laser target to elicit horizontal disconjugate saccades (Sylvestre et al. 2003; Sylvestre and Cullen 2002; Van Horn et al. 2008; Waitzman et al. 2008). In particular, an illuminated target changed from one of four mid-sagittal LEDs to an eccentric (i.e., right or left of the midsagittal plane) laser target. During this paradigm, monkeys made saccades with horizontal components of 5-30 deg in amplitude in both directions, and vergence components with amplitudes of ~4-13 deg. In order to elicit 29

disconjugate saccades, in which the movement of the right eye or left eye movement was minimized, four LEDs were aligned with the left eye or right eye at an angle of 45deg to the left or right of the mid-sagittal plane. To increase the variety of disconjugate saccades in our data set, trials where the LEDs and laser targets were randomly presented were also included.

2.3.3 Data acquisition procedures

During the experiments the monkeys sat in a primate chair in the center of a 1-m3, magnetic eye coil system (CNC Engineering). Horizontal and vertical eye position signals were measured using the magnetic search coil technique (Fuchs and Robinson 1966a; Judge et al.

1980). Each eye coil signal was calibrated independently by having the monkey fixate, with one eye masked, a variety of targets at different horizontal and vertical eccentricities and different depths. Position signals were low-pass filtered at 250 Hz (analog 8 pole Bessel filter) and sampled at 1 kHz. Since ocular saccades include very little power above 50Hz (Cullen et al.

1996b; Van Opstal et al. 1985; Zuber et al. 1968) eye position signals were further digitally filtered (with a 51st order finite-impulse-response filter with a Hamming window and a cut-off at

125 Hz), before being differentiated to obtain eye velocity signals (using zero-phase forward and reverse digital filtering to prevent phase distortion).

Extracellular single unit activity was recorded using enamel insulated tungsten microelectrodes [2-10 M impedance, Frederick Haer; for details, see (Sylvestre and Cullen

1999)]. Neurons in the oculomotonucleus were identified on-line based on their stereotypical discharge properties during eye movements (Robinson 1970). The existence of reciprocal connections between oculomotor internuclear neurons (OINs) to the abducens has been well described (Carpenter et al. 1963; Highstein and Baker 1978; Highstein et al. 1982; Maciewicz et

al. 1975; Steiger and Buttner-Ennever 1978; Ugolini et al. 2006). In a previous study, in which

OINs were identified using antidromic identification and collision testing, <5% of the neurons were identified as OINs (Clendaniel and Mays 1994). Accordingly, the majority of the neurons recorded in this study were most likely motoneurons.

When a neuron was properly isolated, unit activity, horizontal and vertical positions of the right and left eyes, and target position were recorded on a digital audio tape (DAT). The isolation of each neuron was reassessed, offline during playback. An oculomotoneuron was considered to be adequately isolated only when individual action potential waveforms could be discriminated using a windowing circuit (BAK) during saccades (e.g., see Fig. 1 in Sylvestre and

Cullen 1999), and during fixation. Subsequent analysis was performed using custom algorithms

(Matlab, The MathWorks).

2.3.4 Data Analysis

The eyes are referred to as either ipsilateral or contralateral based on their location relative to the recording site. Positive and negative values indicate eye positions that are to the right and left of the central position (i.e., straight ahead), respectively. The movement of the eyes is also reported in terms of conjugate [conjugate = (left eye + right eye)/2] and vergence

[vergence = left eye – right eye] coordinates (where the left eye and right eye inputs could be either position or velocity signals). Note vergence velocity signals are positive during convergence and negative during divergence.

The onset and offset of all saccades was determined using a 20 deg/s saccade velocity criterion. Saccades were categorized as conjugate if the change in vergence angle was < 2.5deg.

Disconjugate saccades during which both eyes moved either ipsilateral or contralateral to the recording site and for which one eye moved at least twice as much as the other were included in the analysis. Notably, an equal number of converging and diverging saccades were included in the disconjugate dataset to prevent biasing the parameter estimates. Also, since previous studies have found that eye velocity and position coefficients are larger during slow velocity movements

(Sylvestre and Cullen 1999), the movements included in the conjugate and disconjugate datasets were in the same range of velocity.

For a subset of neurons (N=10), velocity and position sensitivities were estimated and compared across saccades with different velocities. For this analysis microsaccades and saccades with slow velocity dynamics were included. Microsaccades were defined as short movements with velocities <150deg/sec, slow saccades were defined eye movements with velocities ranging from 150-300deg/sec and fast saccades were defined as eye movements with velocities ranging from >300deg/sec.

The linear optimization techniques used to quantify the dynamic sensitivity of a neuron to eye movements, during conjugate saccades (Cullen and Guitton 1997; 1996; Sylvestre and

Cullen 1999) and disconjugate saccades (Sylvestre et al. 2003; Sylvestre et al. 2002), have been described previously. The specific linear regression models used in the present study are described in RESULTS. The goodness-of-fit of a given model to the data were quantified using the

Variance-Accounted-For {VAF =1 - [var (mod - fr)/var (fr)]}, where mod represents the modeled firing rate and fr represents the actual firing rate}. When estimating linear models, the VAF is mathematically equivalent to the correlation coefficient R2. A VAF value of 1 indicates a perfect fit to the data, and a value of 0 indicates a fit that is equivalent to a mean value. Note that the

VAF can be used for the direct comparison of the goodness-of-fit of model estimations and

predictions. The dynamic lead time of individual neurons (td) was determined during conjugate saccades as described in Sylvestre and Cullen (1999).

For each model parameter in the analysis of disconjugate saccades, we computed 95% confidence intervals using a non-parametric bootstrap approach (Carpenter and Bithell 2000).

This analysis method has been described in detail previously and is particularly well suited for small samples with unknown probability distributions (Carpenter and Bithell 2000; Sylvestre et al. 2003; Sylvestre and Cullen 2002; Van Horn et al. 2008; Van Horn and Cullen 2008) and can be used to identify non-significant or identical model parameters. Briefly, model parameters in a binocular model (see RESULTS) were estimated from an original dataset of N (usually >40) disconjugate saccades. Notably, N/2 saccades were divergent and N/2 saccades were convergent.

1,999 “new data sets” of N saccades were obtained by randomly re-sampling with replacement from the original data set. Therefore every new data set differed from the original data set. After obtaining the new data sets, parameters values were computed for the 1,999 iterations and the

95% confidence intervals were obtained for each model parameter. Parameter values with 95% confidence intervals that overlapped with zero were non-significant and removed from the model

Parameters values that overlapped with each other were statistically identical and were replaced with one conjugate parameter. The reduced models were then re-run.

Since adding extra terms to a model invariably improves its goodness-of-fit, we also calculated the Bayesian information criterion (BIC). The BIC, which served as a “cost index”, was calculated for each model estimation to quantitatively determine whether removing the term was justified. If the change in BIC was very small (<0.05), this indicated that the new model described the data as well as the more complex model thereby justifying the removal of the term.

2.3.5 Quantification of ocular preference

Quantification of ocular preference has been described previously (Sylvestre et al. 2003;

Sylvestre and Cullen 2002; Van Horn et al. 2008). Briefly, for any given neuron the sensitivity to the position and velocity of each eye was used to compute a Ratio index:

Ratio = (smaller estimated parameter value) / (larger estimated parameter value). to indicate which eye provided the larger parameter value (i.e., the neuron's "preferred eye"), each Ratio index was assigned an "i" or a "c", for the ipsilateral or contralateral eye, respectively.

Using the ratio indexes, neurons were assigned to one of five categories, namely; monocular ipsilateral, monocular contralateral, binocular ipsilateral, binocular contralateral or conjugate

[see Table 1 in (Van Horn et al. 2008) for specific criteria for each category].

Mean values in the text are described as means ± standard deviations. A student‟s t-test was used to determine whether the average of two measured parameters differed significantly from each other.

2.4 RESULTS

Our analysis approach was the following: first, we characterized the dynamic responses of medial rectus neurons in the oculomotor nucleus (OMNs) during contralateral (ON-direction) and ipsilateral (OFF-direction) directed conjugate saccades. Second, we assessed whether we could predict the dynamic discharge of OMNs during disconjugate saccades based on their estimated responses during conjugate saccades. Third, we estimated the sensitivity of individual neurons to the velocity and position of the ipsilateral and contralateral eye. We then examined the firing rates of individual OMNs across different oculomotor behaviors. In particular, we determined whether the parameters estimated during conjugate saccades could be used to predict the activity of OMNs during fixation, microsaccades, and slower saccades. Moreover, to get a better understanding of how the agonist and antagonist muscles work together we compared the responses of OMN to abducens motoneurons (ABNs).

2.4.1 Dynamic analysis during conjugate saccades

A total of 34 isolated OMNs were analyzed during conjugate saccades. As has been described previously, all neurons increased their firing rates during contralateral (i.e., ON- direction) saccades (Gamlin and Mays 1992; Keller 1973; Robinson 1970; Schiller 1970). Fig.

2.1A1 and 2.1A2 illustrates two typical example OMNs during contralateral saccades. We first estimated a neuron‟s sensitivity to eye movements during ON-directed conjugate saccades using the following dynamic model, which has previously been shown to adequately describe the neuronal discharge of ABNs during conjugate saccades (Sylvestre and Cullen 2002; 1999):

 FR(t)  bCJ  kCJCJ (t  td )  rCJ CJ (t  td )

where FR(t) is the neuron‟s instantaneous firing rate, bCJ and kCJ and rCJ are constants that represent the bias and the neuron‟s horizontal eye position and eye velocity sensitivity estimated

 during conjugate saccades, respectively. td refers to the dynamic lead time and CJ(t) and CJ(t) refer to the instantaneous horizontal conjugate eye position and velocity, respectively. Model fits are superimposed on the firing rates for the two example neurons shown in Fig. 2.1A1 and 2.1A2.

Similar to ABNs, this first-order model provided a very good fit of OMNs‟ firing rate (mean population VAF=0.60±0.1, bias=116.5±60.346sp/s, k=4.44±2.4(sp/s)/deg, r=0.55±0.32(sp/s)/(deg/s), td=10.4±2.5ms). Notably, the addition of an acceleration term did not significantly improve our ability to estimate the neuron‟s firing rate (mean population

VAF=0.60±0.11, p=0.46; ΔBIC<0.05) while the addition of both an acceleration term and a slide

 term [proportional to the derivative of the neuron‟s firing rate (c MN )] significantly increased the fit by ~10% (VAF=0.70±0.09; p<0.05; cOMN = 23ms±18). Note the parameter estimates in the present study were similar to those calculated previously for ABNs (cABN= 15ms±16) (Sylvestre and Cullen 1999).

Previous studies have shown that the mean firing rate of an OMN is related to eye position during steady periods of fixation (Gamlin and Mays 1992; Keller 1973; Robinson 1970;

Robinson and Keller 1972; Schiller 1970). We verified this relationship in our sample of OMNs.

During fixation, mean firing rate was correlated to eye position (mean y-intercept = 80±97sp/s,

2 mean slope = 6.3±4.8(sp/s)/deg, mean R = 0.66±0.22). Fig 1B1 and 1B2 shows the results of this fixation analysis for the two example neurons. As will be discussed in more detail below, the mean position sensitivity estimated during fixation was significantly larger than that estimated during conjugate saccades (6.4±4.9((sp/s)/deg) versus 4.4±2.4(sp/s)/deg); p<0.05).

Fig. 2.1: (A) Discharge patterns of two typical oculomotor neurons. Two contralateral (i.e., ON- direction) conjugate saccades are shown for each neuron (A1, A2). Gray shaded area in the top row represents the neurons‟ firing rate. Superimposed on the firing rate in black is the estimated model fit obtained using a first-order model, which included the neuron‟s discharge when the eye is in the center of the orbit as well eye velocity and position sensitivities. Below the firing rate are the ipsilateral (ipsi), contralateral (contra), conjugate (conj) and vergence (verg) velocities and positions. Note that during conjugate saccades the ipsilateral and contralateral eyes move the same amount and have the same dynamics. Horizontal dotted lines represent zero velocity and zero firing rate and vertical dotted lines denote saccade onset and offset (20deg/sec criterion). (B1, B2) Mean neuronal firing rate is plotted as a function of mean eye position during fixation for the same two neurons shown in (A).

2.4.2 Dynamic responses estimated during OFF-directed conjugate saccades

To fully assess the drive to the extraocular muscles, the activity of OMNs was also characterized during OFF-directed (i.e., ipsilateral) conjugate saccades. Notably, similar to

ABNs, the majority of the neurons (75%) were driven into inhibitory cut-off for very large saccades (>15deg) in the OFF-direction. We estimated the neuron‟s sensitivity using the same dynamic model used during conjugate saccades for movements that were not driven into inhibitory cut-off. This first-order model also provided a very good fit of OMNs‟ firing rate during the OFF-direction (mean population VAF=0.54±0.13, bias=97.88±36.3sp/s, k=4.8±2.7(sp/s)/deg, r=0.16±0.09(sp/s)/(deg/s). Model fits estimated during OFF-directed conjugate saccades for the two example neurons shown in Fig. 2.1 are illustrated for in Fig.

2.2A1 and 2.2A2.

2.4.3 Example OMN with a preference for the ipsilateral eye

We next determined if the conjugate parameters estimated during conjugate saccades could predict the neuron‟s activity during disconjugate saccades. Sufficient disconjugate behavior was recorded in 22 of the 34 neurons. Fig. 2.3 shows the activity of the example OMN shown in Fig. 2.1 and Fig. 2.2 (unit v3) during converging (panel A) and diverging (panel B) disconjugate saccades, Note the large differences in dynamics for the two eyes during these movements: in the converging case (panel A) the ipsilateral eye moved while the contralateral eye was relatively stationary, whereas in the diverging case (panel B) the contralateral eye moved while the ipsilateral eye was relatively stationary. Notably, although the conjugate component of the movements was comparable in the two behaviors the corresponding firing rates were strikingly different.

Fig. 2.2: (A, B) Discharge patterns of the same two neurons shown in Fig. 2.1. Two ipsilateral (i.e., OFF-direction) conjugate saccades are shown for each neuron. Model fits using a first-order model, which included the neuron‟s discharge when the eye is in the center of the orbit as well eye velocity and position sensitivities, are superimposed on the firing rate.

Interestingly, for this example neuron, as well as the majority of SBNs (68%) in our study, a conjugate-based prediction tended to undershoot the firing rate when the ipsilateral eye moved more (i.e., during the converging movements for this example neuron, panel A) and to overshoot when the ipsilateral eye moved less (panel B). In fact, the neural activity was best predicted when ipsilateral (Fig. 2.3; superimposed blue trace; VAFpred-ipsi = 0.42), rather than conjugate or contralateral eye positions and velocities (superimposed black and red traces;

VAFpred-conj = 0.37 and VAFpred-contra = 0.24), were the model inputs. The mean VAFs for the population of neurons are presented in Table 2.1.

The results of the prediction-based analysis suggest that the majority of the OMNs preferentially encode the movement of an individual eye. We next investigated whether a binocular expansion of the conjugate model might provide an improved description of neuronal discharges during disconjugate saccades:

  FR(t)  b  ki IE(t td )  kc CE(t td )  ri IE(t td )  rc CE(t td ) where b, ki, kc, ri and rc are the bias, ipsilateral and contralateral eye position and velocity sensitivities of the neuron, respectively (subscripts i and c refer to the ipsilateral and contralateral

  eyes relative to the recording site, respectively), and IE(t), CE(t), IE(t) and CE(t) are instantaneous ipsilateral and contralateral position and eye velocities, respectively. When the parameters of binocular expansion model were freely estimated, a very good description of the example OMN's discharge patterns was obtained (Fig. 2.3, VAFest-bino = 0.47, second row, thick black curve).

Fig. 2.3: Discharge patterns of an example monocular OMN during (A) converging and (B) diverging disconjugate saccades in the neuron‟s ON-direction. Predicted model fits using ipsilateral, conjugate and contralateral eye velocities are superimposed on firing rate in the top row in blue, black and red, respectively (VAFipsi=0.42, VAFconj = 0.37, VAFcontra=0.24). Estimated model fits using the binocular model (black trace) and reduced ipsilateral model (dashed blue trace) are shown in the second row. (C) Results of bootstrap analysis. Histograms represent the distribution of parameter values obtained with the bootstrapping analysis using the binocular model for the contralateral (red) and ipsilateral (blue) eye. Black vertical lines indicate the mean value for each parameter and the thick horizontal bars below the histograms indicate the 95% confidence intervals associated with each parameter.

To determine if all parameters of the binocular expansion model were required to accurately describe the neuron‟s firing rate, 95% bootstrap confidence intervals were used to reduce the model to its simplest form (see METHODS). The 95% bootstrap confidence intervals revealed that only the ipsilateral eye position (ki) and ipsilateral velocity sensitivity term (ri) and bias were significantly different from zero (Fig. 2.3C). Removing the contralateral eye position

(kc) and contralateral eye velocity sensitivity term (rc) had a negligible impact on our ability to fit this neuron's discharge (blue dashed curve, second row, Fig. 2.3; VAFest-ipsi = 0.46, ∆BIC<0.05).

We therefore conclude that this neuron is monocular with a preference for the ipsilateral eye. The average estimated VAFs and differences in BIC (i.e., ΔBIC) provided by the complete binocular versus reduced models for all neurons is summarized in Table 2.2.

2.4.4 Response of OFF-directed disconjugate saccades

To fully assess the drive to the extraocular muscles, the activity of OMNs was also characterized during OFF-directed disconjugate saccades. Fig. 2.4 shows the activity of the same example OMN illustrated in Figs. 2.1, 2.2 and 2.3 during converging (panel A) versus diverging disconjugate saccades (panel B) in the OFF-direction. Notably, when the ipsilateral eye was stationary the firing rate of the neuron stayed relatively constant (panel A). In comparison, when the ipsilateral eye moved more the neuron‟s firing rate decreased (panel B). Thus, as for ON- directed disconjugate saccades, this neuron, and the majority of the neurons recorded in the study, the neural activity was best predicted when ipsilateral (Fig. 2.4; superimposed blue trace;

VAFpred-ipsi = 0.33), rather than conjugate or contralateral eye positions and velocities

(superimposed black and red trace; VAFpred-conj = 0.09 and VAF pred-contra = -0.22), were the model inputs. 43

Fig. 2.4: Discharge patterns of an example monocular neuron during (A) converging and (B) diverging disconjugate saccades in the neuron‟s OFF-direction. Predicted model fits using ipsilateral, conjuagte and contralateral eye velocities are shown in the top row in blue, black and red, respectively (VAFipsi=0.33, VAFconj = 0.09, VAFcontra=-0.22). Estimated model fits using the binocular model (black trace) and reduced ipsilateral model (dashed blue trace) are shown in the second row. (C) Bootstrap histograms and 95% confidence intervals (thick horizontal bars) for this neuron.

As described in detail above, we next investigated whether a binocular expansion of the conjugate model would provide an improved description of neuronal discharges during OFF- directed disconjugate saccades. When the parameters of binocular expansion model were freely estimated, a very good description of the example OMN's discharge patterns was obtained (Fig.

2.4, VAFest-bino = 0.51, second row, thick black curve). As was true for ON-directed disconjugate saccades, only the ipsilateral eye position (ki) and ipsilateral velocity sensitivity (ri) and bias terms were significantly different from zero (Fig. 2.4C). Removing the contralateral eye position

(kc) and contralateral eye velocity sensitivity term (rc) had a negligible impact on our ability to fit this neuron's discharge (blue dashed curve, second row, Fig. 2.4; VAFest-ipsi = 0.50, ∆BIC<0.01).

We therefore conclude that this neuron is monocular with a preference for the ipsilateral eye.

2.4.5 Summary of ocular preferences

Average predicted and estimated VAFs and differences in BIC provided by the complete binocular versus reduced models during the ON-direction are summarized in Table 2.1. As described previously (Sylvestre and Cullen 2002) a ratio index was used to assign each OMN to one of five ocular categories, namely; monocular ipsilateral, monocular contralateral, binocular ipsilateral, binocular contralateral or conjugate. Briefly, a ratio of ipsilateral and contralateral eye velocity (Ratiovel) and a ratio of ipsilateral and contralateral eye position (Ratiopos) were computed based on the estimated parameters of the expanded binocular model (see METHODS):

Ratio = (smaller estimated parameter value) / (larger estimated parameter value).

Fig. 2.5: Distribution of Ratio indexes for OMNs. For each neuron, a Ratio index was calculated for velocity and position sensitivities using: [smaller parameter value] / [larger parameter value], where the smaller and larger parameter values are yielded by the non-preferred and preferred eyes, respectively in both the (A) ON-direction and the (B) OFF-direction. For comparison, the distribution obtained for abducens neurons (Sylvestre and Cullen 2002) is shown as an inset in (A). Similar to ABN neurons, the majority of the OMNs encoded the velocity and position of the ipsilateral eye (blue bars).

Thus for monocular units, where one of the sensitivities was equivalent to zero, the Ratio of the sensitivities is equal to zero. Conjugate units had Ratio value of one since both sensitivities had equal values.

The distributions of Ratiovel and Ratiopos obtained using this method for OMNs is shown in Fig. 2.5 for ON and OFF directions. With respect to the eye velocity sensitivities, the majority of the neurons (ON: 74%; OFF: 80%) in our sample had monocular velocity sensitivities (i.e.,

Ratiovel=0; red and blue columns, Fig 2.5A1). Of the monocular units, the majority of the neurons

(ON: 76%; OFF: 89%) encoded the movement of the ipsilateral eye. The distribution of Ratiopos was similar to that of Ratiovel (Fig 5B). The main difference between Ratiovel and Ratiopos was that in the ON-direction more neurons (10% vs 27%) encoded the motion of both eyes with respect to their position sensitivities (i.e., conjugate; Ratio=1; black columns). For comparison, the distribution obtained for abducens neurons (Sylvestre and Cullen 2002) is shown as an inset.

2.4.6 Comparison of ocular preference during disconjugate saccades and disconjugate fixation

We next addressed whether individual OMNs have the same ocular preference during disconjugate saccades and disconjugate fixation. For each neuron we estimated the mean firing rate during steady periods of disconjugate fixation. We then estimated the mean firing rate as a function of the average ipsilateral and contralateral eye. A Ratiofix value was calculated using the same procedure as described above for Ratiopos and Ratiovel (e.g., smaller parameter value/larger parameter value). Fig. 2.6 illustrates the distribution of Ratiofix during disconjugate fixation.

During disconjugate fixation the majority of the neurons (53%) preferred the ipsilateral eye. This

Fig. 2.6: Distribution of Ratio indexes during disconjugate fixation. A Ratio index was calculated for position sensitivities using: [smaller parameter value] / [larger parameter value], where the smaller and larger parameter values are yielded by the non-preferred and preferred eyes, respectively.

distribution is similar to that obtained during disconjugate saccades (compare Figs. 2.5A2 and

2.6). Notably, for the majority of the neurons, the preferred eye estimated during disconjugate fixation was same as the preferred eye estimated during disconjugate saccades. In other words,

OMNs have similar ocular preferences during fixation and saccadic behaviors.

2.4.7 Model parameters estimated across oculomotor behaviors: fixation, microsaccades, and saccades

A second goal of this study was to evaluate whether the same linear model relating eye motion to OMN discharge could be applied across different oculomotor behaviors. Velocity and position sensitivities were estimated and compared across 3 conjugate oculomotor behaviors, namely: fixation, microsaccades, and saccades. Saccades were grouped according to velocity such that faster saccades (>300deg/sec) could be compared to slower saccades (150-300deg/sec).

As described previously (Van Gisbergen et al. 1981), we found that OMNs display a small increase in firing rate during miscrosaccades in the neuron‟s preferred direction. We estimated a neuron‟s sensitivity to velocity and position during miscrosaccades using the same dynamic model that was shown to accurately describe the neuronal discharge of OMNs during disconjugate and conjugate saccades. This model structure also provided an accurate description of the neuronal discharge during microsaccades (mean population VAF=0.50±0.18, bias=100.6±46.34spk/s, r=0.63±0.43(spk/s)/(deg/s), k=6.0±2.9(spk/s)/deg).

Coefficient values estimated across the oculomotor behaviors were averaged across neurons and plotted as a function of the mean peak velocity generated during each behavior.

Figure 7 highlights the observed trends. Overall, it was found that coefficients values estimated 50

during slower movements differed from faster movements. The eye velocity (r) and eye position

(k) coefficients estimated during fast saccades (i.e., >300deg/sec) were significantly smaller than those estimated during fixation and microsaccades (p<0.05). For comparison, the trends observed for ABN neurons during comparable oculomotor behaviors are also shown (grey dashed trace).

We also applied the same model to a data set that included all of the behaviors (e.g., fixation microsaccades, slow saccades and fast saccades). As expected, the VAF and estimated sensitivities were smaller, albeit not significantly different (p>0.05), from those estimated during the large saccades.

The trends shown in Fig. 2.7 suggest a non-linear relationship between eye movement and firing rate. Accordingly, we tested two simple non-linear models to see whether we could improve our ability to fit neuronal responses. In the first, two higher-order velocity terms were

   2 3 included (e.g., FR(t)  b  k E(t  td )  r1 E(t  td ) + r2 E (t - td) + r3 E (t - td)). The addition of these terms only slightly increased our ability to fit the complete data set (VAF = 0.58±0.09 versus 0.60±0.09), and the accompanying reduction in the BIC was minimal (ΔBIC<0.05) suggesting the addition of these terms was not warranted. In the second model, a term to account

for an interaction between position and velocity (e.g., r2 E ) was included. Again, addition of this term resulted in only a small increase in VAF (0.58±0.09 versus 0.59±0.09), and minimal change in BIC (ΔBIC <0.05). Accordingly, we conclude that the nature of the non-linearity is more complex and most likely reflects the mechanics of the extraocular muscle and/or agonist/antagonist interaction. This point is addressed below and in the discussion.

It has been proposed that the consideration of the activation of the antagonist muscle could account for the increase in sensitivities that is associated with decreasing velocities

Fig. 2.7: Eye velocity coefficients (r), eye position coefficients (k) and biases (b) plotted as a function of peak velocity during fixation, microsaccades, slow saccades (150-300deg/sec) and fast saccades (>300deg/sec). Data are shown for the subset of neurons for which we had adequate behavior during microsaccades (N=10). For comparison, the trends observed for ABN neurons during comparable oculomotor behaviors are shown in dashed grey. During larger movements the majority of the neurons were driven into inhibitory cut-off therefore velocity and position sensitivities were not estimated (asterisks).

(Sylvestre and Cullen 1999). For example, during slower movements the contribution of the antagonist muscle could be providing an additional active force that opposes the agonist muscle.

Here, we tested this prediction. In particular, we estimated the sensitivities of OMNs during the neuron‟s OFF-direction (i.e., antagonist motoneurons) for the same oculomotor behaviors and compared these responses to those previously published for ABNs (i.e., agonist motoneurons) during comparable oculomotor behaviors (Sylvestre and Cullen 1999). During microsaccades and slow saccades the neurons did not cease firing and thus we estimated velocity and position sensitivities during these movements. The estimated sensitivities are shown in Fig. 2.7 (grey trace). During larger movements the majority of the neurons were driven into inhibitory cut-off therefore velocity and position sensitivities were not estimated (asterisks Fig. 2.7). Taken together, these findings suggest that the antagonist muscle force is not negligible during slower movements and could account for the larger sensitivities reported for agonist motoneurons during slow movements. Average sensitivities for each of the oculomotor behaviors, in the ON and OFF-direction (i.e., agonist and antagonist, respectively), are summarized in Table 2.3.

2.5 DISCUSSION

In this study we provide the first detailed characterization of oculomotoneurons (OMNs) in the oculomotor nucleus during conjugate and disconjugate saccades. In particular, we determined whether a first-order linear model, which has been previously verified as an accurate description of abducens neurons during conjugate saccades, is also appropriate for describing the neuronal discharge of OMNs. We then determined whether sensitivities estimated during conjugate saccades could be used to accurately predict OMNs discharges during disconjugate

saccades. Our main finding was that the discharge dynamics of OMNs are most accurately described using the position and velocity of the ipsilateral eye. Finally, to assess how agonist and antagonist extraocular muscles work together to ensure accurate viewing, the responses of

OMNs, across a range of velocities, were compared to the responses of abducens neurons

(ABNs) during comparable oculomotor behaviors.

2.5.1 Dynamic discharge of OMNs during conjugate saccades

Overall, our results clearly demonstrate that a first-order linear model, which includes a bias term, an eye position term and an eye velocity term, provides an accurate description of the discharge dynamics of OMNs during ON and OFF directed conjugate saccades. These results are in agreement with prior characterizations of the mechanical properties of the oculomotor plant

(Robinson 1964), as well as a detailed characterization of ABNs discharges during conjugate saccades (Sylvestre and Cullen 1999). Notably, including both a slide term and an acceleration term improved the fit by approximately 10%, similar to the improvement previously described for ABNs (~7%, Sylvestre and Cullen, 1999). Taken together, these findings confirm that first- order linear plant models are a very useful and simple way for describing discharge dynamics of motoneurons during saccades (Robinson 1970; Sylvestre and Cullen 2002; 1999; Van Gisbergen et al. 1981).

2.5.2 Responses of oculomotoneurons during disconjugate fixation and saccades

Previous studies that have evaluated the neural activity of OMNs during vergence have been limited in the respect that they solely evaluated neural activity during static or slow changes in vergence angle (Clendaniel and Mays 1994; Gamlin and Mays 1992; Keller 1973; King et al. 55

1994; Mays and Porter 1984). It has been shown that OMNs have increased levels of tonic activity with adduction of the ipsilateral eye during both conjugate and convergence movements.

OMN discharges were found to be directly proportional to the position of the eye in the orbit.

Moreover, this relationship was consistent during both converged and relaxed gaze. In the present study, we confirm that OMN discharges are also directly proportional to the position of the ipsilateral eye in the orbit during disconjugate fixation. Moreover, we provide the first detailed analysis of the discharge dynamics of OMNs during disconjugate saccades. We found that the majority of the neurons (68%) were best described using the movement of the ipsilateral eye. When we investigated whether a binocular expansion of the conjugate model provided an improved description of neuronal discharges during disconjugate saccades we found that the majority of the neurons were most accurately described using the velocity and position of the ipsilateral eye. Notably, the majority of OMNs had the same ocular preference during disconjugate fixation and saccades.

2.5.3 Comparison of the motor drive of agonist medial and lateral rectus motoneurons during disconjugate saccades

The majority of the neurons in the abducens, which are responsible for driving the lateral rectus muscle, have also been shown to be tuned to the movement of the ipsilateral eye

(Sylvestre and Cullen 2002; Zhou and King 1998). A recent neural simulation, of the population drive generated by ABNs during conjugate and disconjugate saccades, revealed that the drive could nearly account for the movement of the ipsilateral eye during disconjugate saccades.

However, during disconjugate saccades the simulated neural activity was 15% smaller than that

computed during conjugate saccades (see Fig. 13 Sylvestre and Cullen 2002). In the present study, using a similar analysis, we found that the simulated population drive generated during disconjugate saccades was also smaller (approximately 10%) than that generated during conjugate saccades (Fig 2.8A, “ON”).

The fact that the neural activity generated during disconjugate saccades is slightly less than that during conjugate saccades is not that surprising if we consider the distribution of ocular preferences of both OMNs and ABNs (see inset Fig 2.5). For example, a proportion of both

ABNs and OMNs are sensitive to the movement of the contralateral eye. Notably, this finding is also consistent with what is known about the premotor inputs to OMNs and ABNs. While the majority of ABN premotor neurons, such as neurons in the saccade burst generator (Van Horn et al. 2008; Van Horn and Cullen 2008; Zhou and King 1998) and the neural integrator (Sylvestre et al. 2003), are tuned to the movement of the ipsilateral eye, a significant percentage (~40%) are also tuned to the movement of both eyes (i.e., conjugate). Moreover, OMNs receive converging premotor inputs from a number of different neurons which encode vergence angle (e.g., near response neurons) (Judge and Cumming 1986; Zhang et al. 1992) and/or the movement of each eye (e.g., abducens internuclear neurons, and central mesencephalic reticular neurons) (Sylvestre and Cullen 2002; Waitzman et al. 2008).

It is also possible that the difference in neural activity of OMNs and ABNs during conjugate versus disconjugate saccades is offset by the contribution of the antagonist muscle. For example, since the firing rate of OMNs is lower during disconjugate saccades one might expect the discharge of the antagonist motoneurons (i.e., lateral rectus motoneurons) to also be lower during disconjugate saccades in order to drive the eye to the same position. However previous studies, which have evaluated the firing rate of ABNs during disconjugate fixation, suggest that

Fig. 2.8: (A) A simulation of oculomotor population drive during a conjugate and disconjugate saccade in both the “ON” and “OFF” direction. Dynamic models estimated on the actual data were used to reconstruct the firing rate that each neuron in our sample would have generated a typical saccade. The resulting N firing rates were then averaged to provide an estimate of the population drive (FRN). A comparable simulation was then performed for a disconjugate saccade that was derived from the example conjugate saccade where the contralateral eye position and velocity was scaled down to 25% of its original amplitude. Notably this was done to ensure that the movement of the ipsilateral eye was identical in the two saccades. The average firing rates obtained is shown on the far right for saccades in the ON (top curves) and OFF (bottom curves) direction. Solid curve is for the conjugate saccade and dotted curve is for the disconjugate saccade. (B) Schematic diagram of muscle fiber innervations. (C) Schematic diagram of how the contribution of antagonist motoneurons could account for differences in sensitivities observed across oculomotor behaviors.

this is not the case. In particular, studies have shown that for a given position of the eye in the orbit, the majority of ABNs actually fire at higher rates during convergence than when gaze is relaxed (Gamlin et al. 1989; Mays and Porter 1984). In the present study, we furthered our understanding of agonist/antagonist interactions by evaluating the dynamic responses of OMNs in the OFF-direction (Fig. 2.8). We found that the population drive generated during disconjugate saccades is the same as that generated during conjugate saccades (Fig 2.8A, “OFF”) suggesting a decreased antagonist drive cannot account for differences observed at the level of the abducens when it is the agonist muscle. Taken together, these results suggest that other mechanisms, such as selective weighting, or a sampling bias may be responsible for the apparent missing motoneuron drive during disconjugate saccades (Sylvestre and Cullen 2002).

Indeed, there is evidence for the possibility that different motoneurons may contribute more to certain oculomotor behaviors than others. Extraocular muscle, which has two types of muscle fibers: twitch and non-twitch fibers (Buttner-Ennever et al. 2001; Spencer and Porter

1988), receives innervations from different types of motoneurons. Recent retrograde studies have shown that twitch fibers receive innervations from large motoneurons, which lie within the oculomotor nuclei, whereas non-twitch fibers received innervations from small motoneurons, which tend to lie separately around the periphery of the nucleus (Buttner-Ennever 2006; Ugolini et al. 2006).

Moreover, non-twitch motoneurons, in the abducens have been found to receive innervations from premotor sources involved in executing slow eye movements (e.g., vestibular nucleus, prepositus hypoglossi and supraoculomotor nucleus), whereas twitch motoneurons receive premotor signals from the saccadic burst generator (Ugolini et al. 2006; Wasicky et al.

2004) (Fig 8B). In the present study, as well as in previous studies, the neurons recorded were

most likely larger motoneurons within the motor nucleus (e.g., twitch fibers motoneurons).

Accordingly, an under sampling of neurons that are more specialized for vergence movements

(e.g., non-twitch motoneurons) could account for the observed missing drive during disconjugate saccades.

2.5.4 Consideration of the antagonist muscle when modeling across oculomotor behaviours

Previous studies have shown that a single linear model equation cannot be used to describe the discharge dynamics of motoneurons across different oculomotor behaviors (Fuchs et al. 1988; Gamlin and Mays 1992; Sylvestre and Cullen 1999). For example, an analysis of eye velocity and position sensitivities across oculomotor behaviors with different eye velocities have shown they tend to decrease as peak and mean velocity increases (Fuchs et al. 1988; Gamlin and

Mays 1992; Sylvestre and Cullen 1999). Here, when we compared eye velocity and position sensitivities of OMNs during fixation, microsaccades, slow and fast saccades we also found that both parameters invariably decreased with increasing velocity. These results are consistent with the proposal that the viscosity of the plant is non-linear (Collins 1971; Collins et al. 1975; Miller and Robins 1992). For example, Collins (1971) showed that the viscosity of the extraocular muscle varied non-linearly as a function of the muscle‟s stretch velocity and that the viscosity of passive orbital tissues remains relatively constant across eye velocities. Accordingly, if faster movements have less viscous resistance than slower movements they would require less force and hence velocity and position coefficients would be lower (see DISCUSSION Sylvestre and

Cullen 1999).

It has also been proposed that the relative contribution of antagonist motoneurons could account for these differences in sensitivities observed across oculomotor behaviors. For example, during slow movements (e.g., pursuit) the majority of antagonist motoneurons continue to discharge (Sylvestre and Cullen 1999). Thus, the contribution of the antagonist muscle is not negligible but rather it provides an additional active force to oppose that produced by the agonist muscle. Consequently, a given motoneuron might generate the same discharge in two conditions, but depending on the activation of the antagonist muscle the movement of the eye could differ

(Fig 8C). In the present study, we tested the prediction that the antagonist muscle is not necessarily negligible during slower eye movements. We found that during microsaccades and slow saccades in the OFF-direction the motoneurons controlling the antagonist were not completely silent. This finding is consistent with the reports that eye velocity and position sensitivities are higher for slower movements since the agonist muscle has to oppose the contribution of the antagonist muscle.

Taken together, our findings suggest that to model different oculomotor behaviors it is necessary to consider how populations of neurons work together to ensure accurate three- dimensional viewing. Specific behaviors may recruit different populations of neurons, as well as different ratios of agonist versus antagonist motoneurons. Accordingly, accounting for the contribution of the antagonist neurons, as well as individual neuronal properties, is essential for fully understanding how particular eye movements are generated.

Chapter 3

Dynamic characterization of saccadic burst neurons during conjugate and disconjugate saccades

This chapter describes the discharge dynamics of brainstem saccadic burst neurons during conjugate and disconjugate saccades. A combination of experimental and modeling approaches were used to investigate whether these neurons are strictly “conjugate” or whether they encode sufficient vergence-related information to drive disconjugate saccades. This chapter has been adapted from Van Horn MR, Sylvestre PA, and Cullen KE. Dynamic characterization of saccadic burst neurons during conjugate and disconjugate saccades. J Neurophysiol 99:2602-

2616, 2008.

3.1 ABSTRACT

When we look between objects located at different depths the horizontal movement of each eye is different from the other, yet temporally synchronized. Traditionally, a vergence- specific neuronal subsystem, independent from other oculomotor subsystems, has been thought to generate all eye movements in depth. However, recent studies have challenged this view by unmasking interactions between vergence and saccadic eye movements during disconjugate saccades. Here, we combined experimental and modeling approaches to address whether the premotor command to generate disconjugate saccades originates exclusively in “vergence centers”. We found that the brainstem burst generator, which is commonly assumed to drive only the conjugate component of eye movements, carries substantial vergence-related information during disconjugate saccades. Notably, facilitated vergence velocities during disconjugate saccades were synchronized with the burst onset of excitatory and inhibitory brainstem saccadic burst neurons (SBNs). Furthermore, the time-varying discharge properties of the majority of

SBNs (>70%) preferentially encoded the dynamics of an individual eye during disconjugate saccades. When these experimental results were implemented into a computer-based simulation, to further evaluate the contribution of the brainstem saccadic burst generator in generating disconjugate saccades, we found that it carries all the vergence drive that is necessary to shape the activity of the abducens motoneurons to which it projects. Taken together, our results provide evidence that the premotor commands from the brainstem saccadic circuitry, to its target motoneurons, is sufficient to ensure the accurate control shifts of gaze in three dimensions.

3.2 INTRODUCTION

Precisely coordinating the movements of our eyes is critical for achieving an accurate visual perception in a three-dimensional world. In particular, unequal yet tightly controlled rotations of the eyes must be programmed whenever the point of fixation is shifted between objects located at different depths. The difference between the rotations of the eyes is referred to as a vergence eye movement. Traditionally saccadic and vergence eye movements are considered as two distinct subclasses of eye movements generated by largely distinct neuronal circuitries. However, numerous studies have provided results that argue against this view.

Vergence velocities are greater than what would be predicted by a linear summation of a conjugate saccade with a saccade-free vergence movement, while conjugate velocities are decreased (Busettini and Mays 2003; 2005a; Collewijn et al. 1997; 1995; Enright 1984; Enright

1992; Erkelens et al. 1989; Kenyon et al. 1980; Kumar et al. 2005a; Maxwell and King 1992;

Ono et al. 1978; Oohira 1993; Zee et al. 1992). Moreover, the amount of vergence facilitation is dependent on peak saccadic velocity (Busettini and Mays 2005a).

How the brain facilitates vergence eye movements during disconjugate saccades remains a topic of debate. On the one hand, it had been proposed that vergence facilitation occurs because inhibitory omnipause neurons in the dorsal raphe nucleus simultaneously gate activity of distinct saccadic and vergence pathways (Mays and Gamlin 1995; Zee et al. 1992). On the other hand, a series of studies have provided evidence that the saccadic premotor pathway plays a role in facilitating vergence shifts by encoding integrated conjugate and vergence premotor commands.

For example, while stimulation of the superior colliculus generally elicits conjugate saccades

(Schiller and Stryker 1972), it can disrupt vergence movements when applied midflight during a disconjugate saccade (Chaturvedi and Van Gisbergen 1999; 2000). In addition, premotor 64

saccadic burst neurons (SBNs) show monocular tuning (i.e., a combination of conjugate and vergence signals) during disconjugate saccades (Zhou and King 1998). While these findings are consistent with the hypothesis that the saccadic circuitry plays a role in facilitating vergence shifts, they cannot rule out the alternative possibility that the overall contribution of the saccadic circuitry is relatively unimportant compared to that of the vergence subsystem. Notably, the most recent proposal is that SBNs exclusively encode conjugate saccadic dynamics that interact with the vergence subsystem (Busettini and Mays 2005b).

Here we investigated whether the saccadic burst generator is strictly a “conjugate” premotor control pathway to resolve whether the saccadic pathway encodes sufficient vergence- related information to drive disconjugate saccades. We recorded from single neurons in the brainstem saccadic burst generator during disconjugate saccades and found that they dynamically encode vergence-related information. Using simulation, we further show that this command signal from the premotor saccadic circuitry is in fact sufficient to drive the target extraocular motoneurons during disconjugate saccades. Accordingly, a separate vergence subsystem is not required to control the abducens nucleus during disconjugate saccades. Overall, our results strongly suggest that the drive from the saccadic burst generator is essential for the control of gaze in three dimensions.

3.3 MATERIALS AND METHODS

3.3.1 Animals and surgical procedures.

Two rhesus monkeys (Macaca mulatta) were prepared for chronic extracellular recording under aseptic conditions. All procedures were approved by the McGill University Animal Care

Committee and were in compliance with the guidelines of the Canadian Council on Animal Care.

The surgical preparation has been described previously (Sylvestre and Cullen 1999). Briefly, using aseptic techniques and isofluorane anesthesia (2-3%, to effect), we implanted several stainless steel screws into the skull and attached a stainless steel recording chamber and a post for head restraint to these screws with dental cement. In the same procedure, a 17-18 mm diameter eye coil, consisting of 3 loops of Teflon coated stainless steel wire, was implanted in each eye beneath the conjunctiva (Judge et al. 1980). Following the surgery, the animals were administered buprenorphine (0.01 mg/kg IM) for post-operative analgesia, and the antibiotic

Cephazolin (Ancef®; 25 mg/kg IM, for 5 days). Animals were given at least two weeks to recover from the surgery before experiments began.

3.3.2 Behavioral paradigms

Head-restrained monkeys were seated in a primate chair that rested on a vestibular turntable, and were trained to fixate targets in a dimly lit room for a juice reward. Monkeys were required to fixate light targets for 1-3 seconds to receive a reward. The timing and location of target illumination, data acquisition and on-line data displays were controlled using REX, a

UNIX-based real-time acquisition system (Hayes et al. 1982). Neuronal responses were recorded during three types of eye movements: 1) Conjugate saccades, 2) disconjugate saccades and 3) smooth, saccade-free vergence. Fig. 3.1 illustrates an example of each of these three types of eye movements where the right eye moved approximately the same amount, in the same direction.

Fig. 3.1 Example of (A) a conjugate saccade (B) a disconjugate saccade and (C) smooth saccade-free vergence. Arrows in (B) denote when the onset of the saccade and the onset of vergence facilitation. Note that in each case the right eye is moving approximately the same amplitude (5 degrees to the left) and that the final vergence amplitude in (B) and (C) is same. The final vergence amplitude is reached much faster when a disconjugate saccade is made.

First, to elicit conjugate eye movements (i.e., Vergence < 2deg; Fig. 2.1A) , a red HeNe laser was projected via a system of two galvanometer controlled mirrors onto a cylindrical screen

(approximately isovergent) located 55cm away from the monkey‟s head (4 deg convergence).

Ipsilaterally and contralaterally directed conjugate saccades were elicited by stepping the target between horizontal positions (5-30 deg), in 5deg increments, in predictable and unpredictable sequences. In addition, conjugate smooth pursuit eye movements were obtained using a sinusoidally moving laser target (40 deg/s peak velocity, 0.5 Hz).

Second, a horizontal array of 16 computer controlled red light emitting diodes (LEDs) with intensities comparable to that of the laser target (see (Sylvestre and Cullen 1999) was used in combination with the laser target to elicit disconjugate saccades (Fig. 2.1B). Disconjugate saccades were generated when the illuminated target changed from one of the close midsagittal

LEDs to an eccentric (i.e., right or left of the midsagittal plane) laser target. During this paradigm, monkeys made disconjugate saccades with conjugate components 5-30 deg in amplitude in both directions, and vergence components with amplitudes 4-13 deg. In addition, some LEDs were positioned in a configuration similar to the Müller paradigm in order to generate „monocular‟ saccades in which the movement of one eye is largely substantially reduced (see (Ramat et al. 1999) for comparable examples). Finally, saccade-free vergence eye movements were also evoked by having monkeys look between four LEDs (convergence angles:

17, 12, 8 and 6 deg) and a laser target that were aligned with the monkey‟s midsagittal plane

(Fig. 2.1C). Neurons that burst during conjugate saccades, but also responded to conjugate sinusoidal smooth pursuit (40 deg/s peak velocity, 0.5 Hz), or cancellation of the vestibuloocular reflex (VORc; 40 deg/s peak velocity, 0.5 Hz), were not included in our sample.

3.3.3 Data acquisition procedures

Extracellular single unit activity was recorded using high impedance enamel insulated tungsten microelectrodes (2-10 M impedance; FHC Inc., Bowdoinham, ME). Saccadic burst neurons (N = 74) were identified on-line based on their stereotypical discharge properties during eye movements (Cullen and Guitton 1997). Excitatory and inhibitory burst neurons (EBNs and

IBNs , respectively) were distinguished based on their recording location relative to the abducens nucleus. EBNs were recorded in a small region extending 1-2mm rostral to the abducens nucleus, and 0.5-1.5mm from the midline. IBNs were recorded in a region extending 0-2mm caudal to the abducens nucleus, and 0.5-1.5mm from the midline. Both areas correspond to previous anatomical characterizations (Strassman et al. 1986a; b). A small sample of omnipause neurons

(OPNs, N = 10) was also recorded. A neuron was considered to be sufficiently isolated only when individual action potential waveforms could be discriminated during saccades using a windowing circuit (BAK Electronics Inc., Mount Airy, MD; see Fig. 1 in Sylvestre and Cullen

1999).

The magnetic search coil technique was used to record the horizontal and vertical positions of both eyes (Fuchs and Robinson 1966b; Judge et al. 1980). Each eye coil signal was calibrated independently by having the monkey fixate, with one eye masked, a variety of targets at different horizontal eccentricities and depths. Position signals were low-pass filtered at 250 Hz

(analog 8 pole Bessel filter) and sampled at 1 kHz. Because ocular saccades include very little power above 50Hz (e.g., (Cullen et al. 1996b; Van Opstal et al. 1985; Zuber et al. 1968) eye position signals were further digitally filtered (with a 51st order finite-impulse-response filter with a Hamming window and a cut-off at 125 Hz), before being differentiated to obtain eye

velocity signals (using zero-phase forward and reverse digital filtering to prevent phase distortion). Targets, rewards, on-line data displays, and data acquisition were controlled using custom designed algorithms developed in the REX environment (Real-time Experimentation

System, Hayes et al. 1982). Off-line analysis was performed in the Matlab programming environment (The MathWorks Inc, Natick, MA).

3.3.4 Data analysis

In this report, the eyes are referred to as ipsilateral or contralateral based on their location relative to the recording site. Positive and negative values correspond to positions right and left of the sagittal plane, respectively. We also describe eye movements in terms of conjugate [conjugate = (left eye + right eye)/2] and vergence (vergence = left eye - right eye) coordinates. Note that vergence positions are always positive, but vergence velocities can be either positive (convergence) or negative (divergence). For all saccades, the onset and offset was determined using a 20 deg/s conjugate velocity criterion. Analysis was limited to horizontal saccades, which were defined as movements having changes in vertical eye position < 10% of the change in horizontal position. Conjugate saccades were defined as having changes in vergence angles < 2deg. Disconjugate saccades were selected for which one eye moved more than the other, generating vergence velocities > 100 deg/s and mean intra-saccadic vergence shifts of 6.5±1.1 deg. Moreover, only saccades for which both eyes moved in the same direction were used in order to limit the analysis to ON-direction responses. Conjugate and disconjugate datasets contained >40 saccades (average Nconj=45.8±6.9; average Ndisconj=44.1±5.9 saccades).

An equal number of converging and diverging saccades were included in the disconjugate dataset to prevent biasing the parameter estimates. Many disconjugate saccades were 70

accompanied by periods of slow vergence preceding or following the onset of the saccade. The onset and offset of these slower movements was determined using a 10 deg/s vergence velocity criterion. For our data set of pure vergence movements, analysis was restricted to periods of saccade-free vergence movements and smooth vergence responses for which changes in conjugate horizontal or vertical component < 10% of the change in vergence.

3.3.4a Dynamic analysis of BN firing rate

The linear optimization techniques used to quantify the dynamic sensitivity of a neuron to eye movements during conjugate saccades (Cullen and Guitton 1997; 1996; Sylvestre and

Cullen 1999) and disconjugate saccades (Sylvestre et al. 2003; Sylvestre et al. 2002), have been extensively described. Neuronal discharges were represented as a spike density function in which a Gaussian function (standard deviation of 5 ms) was convolved with the spike train (Cullen and

Guitton 1997; 1996; Sylvestre and Cullen 2002; 1999). This Gaussian width effectively low- pass filtered burst neuron discharges so that the frequency content is comparable to the associated saccadic movement (Cullen et al. 1996a). A neuron's saccadic lead time was determined using both a first-spike and a dynamic lead time (td) approach (Cullen and Guitton

1996).

The specific model structures used are reported in RESULTS. The goodness-of-fit of the data to each model was quantified using the Variance-Accounted-For (VAF = 1 - [var(mod-fr) / var(fr)], where mod represents the modeled firing rate and fr represents the actual firing rate).

The VAF in linear models is equivalent to the square of the correlation coefficient (R2) such that

a model with a VAF of 0.64 provides as good a fit to the data as a linear regression analysis that yields a correlation coefficient of 0.80 (Cullen et al., 1996).

For each model parameter in the analysis of disconjugate saccades, we computed 95% confidence intervals using a non-parametric bootstrap approach (Carpenter and Bithell 2000;

Sokal and Rohlf 1995) and used these confidence intervals to identify non-significant or identical model parameters (Sylvestre et al. 2003; Sylvestre et al. 2002). If a confidence interval overlapped with zero the model was re-run with the non-significant term removed. The Bayesian information criterion (BIC: Schwarz 1978), which served as a “cost index”, was calculated for each model estimation to quantitatively determine whether removing the term was justified. If the BIC did not change this indicated that new model described the data as well as the more complex model thereby justifying the removal of the term.

3.3.4b Metric analysis of BN discharges

To compare our sample of EBNs and IBNs with those previously described in the literature, saccade-related burst activity was also characterized using classical metric-based analyses. The number of spikes (NOS) was defined as the total number of action potentials that a neuron produced during an associated saccade and burst duration was defined as the time between the onset and offset of the burst. For each neuron, standard linear regression techniques were used to describe the relationships between 1) saccade duration and burst duration, 2) total vergence duration and burst duration, 3) saccade amplitude and NOS, and 4) peak saccade velocity and firing rate.

During disconjugate saccades, the classic metric approaches were adapted to account for the movements of both eyes:

NOS  b  zi IE  zcCE (3.1)

  FRmax  b  pi IE max  pc CE max (3.2)

where b is the bias, zi and zc, are the regression coefficients relating the number of spikes to ipsilateral (ΔIE ) or contralateral saccade amplitude (ΔCE), respectively, pi and pc are the

 regression coefficients relating the peak firing rate to the peak velocity of the ipsilateral ( IE ) and

 contralateral eye ( CE ), respectively.

3.3.5 Quantification of ocular preference

First, to quantify the ocular preference (see Table 3.1), Ratio indices were defined as follows:

Ratio = (smaller estimated parameter value) / (larger estimated parameter value).

For each neuron three ratio indices were calculated. The ratio of a given neuron‟s sensitivity to the velocity of each eye (estimated using the dynamic analysis) was used to compute Ratiodyn. Ratios of the regression coefficients estimated using the metric-based relationships between NOS and movement amplitude (eq. 3.1), and peak firing rate and velocity

(eq. 3.2) were used to compute RatioNOS and RatioFRmax, respectively.

Second, in order to facilitate comparison between our sample of neurons to those of Zhou and King (1998), who used a NOS-based approach identical to that shown in Eq.(3.1), we computed an ocular index (OI) that was converted to a RatioNOS index:

zi  zc OI 1 zi OI  , that we reorganized to   Ratio NOS zi  zc 1OI zc

where zi and zc are the regression coefficients relating the number of spikes to ipsilateral or contralateral saccade amplitude respectively.

Parameter values were compared across neuron types (short and long lead EBNs and

IBNs, see RESULTS), using a one-way ANOVA followed by a standard post-hoc multiple comparison test. Briefly, this latter test computes a Student's t-test for all k permutations among the groups included in the ANOVA, and uses the Dunn-Šidák method [where the significance level (‟) of each k t-test is defined as ‟ = 1 - (1 - )1/k;  = 0.05] to correct for the effect of multiple comparisons (Sokal and Rohlf 1995). To compare the frequency of distribution of different neuron types across categories of ocular preferences, we performed a 2-test on a contingency table (Wackerly et al. 1996). Note that the unequal binocular neurons were pooled in two categories for this analysis: unequal binocular with a preference for the ipsilateral eye, or unequal binocular with a preference for the contralateral eye.

3.3.6 Simulation design

A computer-based simulation was implemented in which the discharges of saccadic burst neurons (present study) were combined with those of other saccade-related premotor neurons

with known projections to the abducens nucleus (McConville et al. 1994; Roy and Cullen 2003;

Roy and Cullen 2002; Scudder and Fuchs 1992; Sylvestre et al. 2003; Sylvestre and Cullen

1999). The discharge dynamics of abducens motoneurons were predicted based on the weighted sum of inputs from different premotor neuron types (n=8). The following criteria were used in the simulation's design. First, all premotor neurons included were shown experimentally to project to the abducens nucleus (Langer et al. 1986; McFarland and Fuchs 1992; Scudder and

Fuchs 1992; Strassman et al. 1986a; b). Second, the nature of each connection (i.e., excitatory or inhibitory) was determined experimentally (Scudder and Fuchs 1992; Scudder et al. 2002).

Third, for each neuron type (including abducens nucleus neurons), mathematical descriptions obtained experimentally from populations of neurons were available to reconstruct average population discharges during different oculomotor behaviors (Table 3.2).

While the sign and general behavior (e.g., SBN input was zero during fixation) of each premotor neuron projection was fixed based on physiological knowledge (Langer et al. 1986; Scudder and

Fuchs 1992; Scudder et al. 2002) its weight was optimized. This optimization was performed on a dataset of conjugate eye movements that included: 1) fixation (–30 to +30 deg in 5 deg increments), 2) sinusoidal smooth pursuit (0.5Hz, 40 deg/s peak velocity); 3) passive sinusoidal whole-body rotation with an earth-fixed target (0.5Hz, 40 deg/s peak velocity); 4) passive sinusoidal whole-body rotation with a head-fixed target (0.5Hz, 40 deg/s peak velocity); 5) saccades (amplitudes 3-20deg). Each paradigm provided 2000 data points to the algorithm.

Optimization using a rich dataset of conjugate eye movements was important to set realistic weights that account for the relative contributions of each neuron type across various oculomotor behaviors (Cullen et al. 1993b; Cullen and McCrea 1993; Hazel et al. 2002; Roy and Cullen

2002). Weights (wi) for all premotor neuron types (n = 8) were obtained using a least-square optimization procedure:

FR(t) ABN  w1 FR(t)EH _ contra  w2 FR(t)PVP _ contra ... w8 FR(t)EH _ ipsi

When freely estimated, the weight of ipsilateral burst-tonic neurons became unrealistically large, while the other weights were generally reduced to negligible values (see Fig. 3.10A, rightmost endpoints). This is not surprising given that burst-tonic and abducens neurons have highly similar discharge patterns across oculomotor behaviors (Sylvestre et al. 2003). To estimate more physiologically realistic values, the weight of ipsilateral BT was fixed to different values <0.8

(the weight obtained when freely estimated), and the remaining weights were estimated. All weight values obtained using this approach showed dependence on the ipsilateral BT weight.

Notably, all weight sets yielded relatively similar goodness-of-fits (0.93 ≤ VAF ≤ 0.96; measured simultaneously over all five paradigms) to reconstructed ABN discharges (See Fig. 3.10B).

To determine the „optimal‟ weight set, the weight values estimated on the conjugate dataset were used to predict the discharges of ABNs on a dataset of disconjugate saccades that were not used to estimate the weights. The „optimal‟ weight set was defined as the weight set that accounted for most of the variance in the average ABN firing rates during disconjugate saccades (both converging and diverging disconjugate saccades when both eyes were moving in the neuron‟s “on” i.e., ipsilateral, direction used in the simulations). Note that for disconjugate saccades, coefficients were taken from this and prior studies when possible (i.e.(Sylvestre et al.

2003; Sylvestre and Cullen 2002) and inferred for contra PVP neurons based on published data

(sensitivities and bias taken from (McConville et al. 1994; Roy and Cullen 2002) since most

PVPs ramp up during contra saccades without a burst or pause). For neurons that were conjugate,

and/or for which only conjugate estimations have been done (e.g., eye head neurons), the coefficients of each eye are equivalent to the conjugate estimate /2 (by definition).

3.4 RESULTS

The primary goal of this study was to determine if the modulation of SBNs could account for the increased vergence velocities during disconjugate saccades. The approach chosen to address this problem was two-fold, and involved 1) a comprehensive analysis of the timing and dynamic properties of SBNs‟ discharges during disconjugate saccades, and 2) a quantitative simulation of the command generated by brainstem premotor neurons during conjugate versus disconjugate saccades.

Excitatory burst neurons (EBNs; N = 30) and inhibitory burst neurons (IBNs; N = 44) were recorded in the paramedian pontine reticular formation (PPRF). The neurons in this study were further categorized as short or long lead neurons depending on whether the mean period between the onset of the first spike and the onset of eye velocity was 15 ms, or >15 ms, respectively (Cullen and Guitton 1997; Scudder et al. 1988). Comparable numbers of short and long lead EBNs (N= 16 and 14, respectively) and IBNs (N= 23 and 21, respectively) were recorded. A thorough comparison of short and long lead EBNs and IBNs was completed for each aspect of the study. Where no differences were found between the discharge patterns of short and long lead IBNs and EBNs (with the obvious exception of the burst lead times), the two populations are discussed as a pooled population.

3.4.1 SBN discharge timing is appropriate to facilitate vergence movements

An example of a short lead IBN is shown in Fig. 3.2 and 3.3. This neuron was typical in that it discharged a compact burst of action potentials during ipsilaterally-directed conjugate saccades. Although no action potentials were observed during saccade-free vergence (Fig. 3.2A), the example neuron was active during the saccade component of the disconjugate saccades (Fig.

3.3A). Importantly, no action potentials were observed before or following this interval, despite the presence of a significant but much slower vergence velocity (grey shaded areas in Fig. 3.3A).

Fig. 3.3B illustrates the relationship between burst duration and saccade duration for this example burst neuron during conjugate saccades and disconjugate saccades (gray and black filled circles, respectively). Burst duration was well related with saccade duration during both conjugate (mean R2=0.89, range 0.4-1.00) and disconjugate (mean R2=0.89, range 0.46-1.00) saccades. In contrast, burst duration was less correlated (P<0.05) with the total duration of the vergence movement during disconjugate saccades (Fig. 3.3C, mean R2=0.48, range 0.03-0.93), where total vergence duration included the combined duration of both the saccade as well as any slow vergence movement that preceded or followed it. Similar findings were obtained for the analysis of OPNs during the same paradigms (Fig. 3.2B). The implication of these results are further considered in the DISCUSSION.

3.4.2 Testing the null hypothesis: SBNs only encode conjugate eye movement dynamics

The timing of SBNs discharge is coincident with the facilitation of vergence movements during disconjugate saccades. We next quantitatively characterized the signals that are dynamically encoded by the brainstem saccadic burst generator during conjugate and disconjugate saccades using the following approach: First, we used systems identification

Fig. 3.2: Example neural activity for a short lead IBN (A) and omnipause neuron (OPN) (B) during smooth saccade-free vergence. The gray shaded areas in the top rows represent the firing rate of the neuron and the unit activity is shown in the second rows. Also shown are the conjugate (CJ) and vergence (VG) velocity traces as well as the velocity traces of each eye (i.e. the eye ipsilateral (IE) and contralateral (CE) to the recording site). The light gray shaded areas in vergence velocity traces in (A) and (B) highlight the areas of smooth saccade-free vergence. Horizontal dotted lines denote zero velocity.

Fig 3.3: Example neural activity for a short lead IBN (A) during facilitated vergence. Converging saccades are on the left and diverging saccades are on the right. The light gray shaded areas in the vergence velocity traces in (A) highlight the areas of smooth saccade-free vergence. The example IBN (A) did not fire any action potentials saccade-free vergence but started firing at the onset of the saccade. The onset and offset of the saccade was determined using a typical 20deg/sec velocity criterion, which is marked by vertical dotted lines. (B) The relationship between the burst duration of a typical IBN and saccade duration in conjugate saccades and in disconjugate saccades immediately preceded by a period of slow vergence (>10ms, see also Busettini and Mays 2005a). (C) The burst duration, of the same IBN shown in (A), and total vergence duration (including the saccade-free vergence).

techniques to provide a movement-based description of the discharge dynamics of each neuron during ipsilaterally directed conjugate saccades. Second, we assessed whether we could predict the discharge of the same neuron during disconjugate saccades based on its responses during conjugate saccades. Third, we directly estimated the sensitivity of individual neurons to the movements of the right / left eyes or the conjugate / vergence profiles on the same data set of disconjugate saccades (see METHODS). Based on their ocular preference during disconjugate saccades, neurons were sorted into five categories (see Table 3.1). Finally, to fully assess the premotor drive to the extraocular motoneurons during saccades we also characterized SBN activity during OFF-direction saccades (i.e., contralaterally directed saccades).

The original analysis of Cullen and Guitton (1997) determined the simplest firing rate model that can be used to describe IBNs‟ discharge properties during conjugate saccades is:

 FR(t  td )  b  r CJ (t) (Conjugate-est model) (3.3) where FR is instantaneous firing rate, b is the estimated bias and r is the estimated velocity

 sensitivity, CJ(t) is the velocity of the eye during conjugate saccades and td is the neuron‟s dynamic lead time. Here this finding was confirmed for IBNs and extended to EBNs (see Table

3.2 for population averages of parameters and VAFs, td = 14.13.8ms and 13.82.8ms, EBNs and IBNs, respectively). Fig. 3.4 shows model fits for a representative EBN and IBN during two conjugate saccades (top row; VAF = 0.57±0.16 and 0.57±0.16, EBNs and IBNs; notably VAF values indicated here, and for the subsequent figures, were calculated when fitting the entire data set and not only the example movements that are shown in the figures). Adding an eye position or acceleration term (Fig. 3.4, second row, thick black and gray lines, respectively) to Eq.(3.3) did not markedly improve the model fits (EBNs, VAF = 0.59±0.15 and 0.57±0.15; IBN, VAF =

Fig. 3.4 Example discharges from a short lead (A) EBN and (B) IBN during two conjugate saccades. The top row represents the neurons‟ firing rates (gray shaded areas), and the model fits (thick black curve) obtained with eq.(3). The second row shows the same firing rate traces duplicated for clarity, but with the model fits obtained using eq.(3) + eye position (thick black curve), and eq.(3) + eye acceleration (thick gray curve). Also shown are ipsilateral eye (IE), contralateral eye (CE) and conjugate (CJ) velocity traces (third row) and position traces (fifth row), and vergence (VG) velocity traces (fourth row) and position traces (bottom row). Vertical dotted lines denote saccade onsets and offsets, and horizontal dotted lines represent zero velocity.

0.59±0.13 and 0.58±0.13; position and acceleration terms, respectively). Indeed, the position and acceleration parameter values did not differ from zero across neuron types (P=0.40, P=0.38, respectively, one-way ANOVA, see METHODS).

3.4.3 Rejecting the null hypothesis: The conjugate prediction fails

The discharge dynamics of the EBNs and IBNs were next analyzed during disconjugate saccades. For each neuron, we first determined whether the simple model estimated above during conjugate saccades [eq.(3.3)] could predict its activity during disconjugate saccades. Fig. 3.5 shows the activity of an example short lead EBN during converging versus diverging disconjugate saccades (i.e., panels A and B, respectively) elicited during “monocular” saccades

(i.e., Müller paradigm, see Methods). Note the large differences in dynamics for the two eyes during these movements: in the converging case (panel A) the contralateral eye moved while the ipsilateral eye was relatively stationary, whereas in the diverging case (panel B) the ipsilateral eye moved while the contralateral eye was relatively stationary. Notably, the conjugate component of the movements was comparable in the two conditions. For the example neuron, as well as the majority of SBNs (54%) in our study, the conjugate-based prediction tended to overshoot the firing rate when the preferred eye moved less (i.e., during the diverging movements for this example neuron, panel A) and to undershoot when the preferred eye moved more (panel B). Specifically, this neuron‟s activity was best predicted when ipsilateral

(superimposed blue trace; VAFpred-ipsi = 0.62) rather than conjugate or contralateral eye velocities

(superimposed black and red trace; VAFpred-conj = 0.49 and VAF pred-contra = 0.13) were the model inputs.

The results of the prediction-based analysis suggest that the majority of the neurons preferentially encode the movement of one eye rather than the conjugate eye velocity. We next investigated whether estimating a more complex model, namely a binocular expansion of the conjugate model (eq.(3.3)), might provide an improved description of neuronal discharges during disconjugate saccades:

  FR(t td )  b  ri IE(t)  rc CE(t) (Binocular-est model) (3.4) where b, ri and rc are the bias, ipsilateral and contralateral eye velocity sensitivities of the neuron, respectively (subscripts i and c refer to the ipsilateral and contralateral eyes relative to the

  recording site, respectively), and IE(t) and CE(t) are instantaneous ipsilateral and contralateral eye velocities, respectively. For each parameter in eq.(3.4), bootstrap 95% confidence intervals were used to reduce the model to its simplest form.

When the parameters of eq.(3.4) were freely estimated, a very good description of the example EBN's discharge patterns was obtained (Fig. 3.5, VAFest-bino = 0.78, second row, thick black curve). The 95% bootstrap confidence intervals revealed that only the ipsilateral eye velocity sensitivity term (ri) and bias were significantly different from zero (Fig. 3.5C). Thus, removing the contralateral eye velocity sensitivity term (rc) from eq.(3.4) had a negligible impact on our ability to fit this neuron's discharge (gray curve, second row, Fig. 5; VAFest-bino = 0.78,

VAFest-ipsi = 0.77, ∆BIC=0). We therefore conclude that this neuron is monocular with a preference for the ipsilateral eye. Overall, we found that most SBNs (>70%) preferentially encoded the velocity of an individual eye (average bias, ipsilateral and contralateral eye velocity sensitivities of the SBNs were 150±58, 0.32±0.28 and 0.43±0.30).

Fig. 3.5: Example discharges from an example monocular EBN during (A) converging and (B) diverging disconjugate saccades. Predicted model fits using conjugate, ipsilateral and contralateral eye velocities are shown in the top row in black, blue and red respectively. Estimated model fits using the binocular model [eq.(2.4)] are shown in the second row (thick black curve). Estimated model fits using the reduced ipsilateral model are also shown (dashed gray curve). (C) Bootstrap histograms and 95% confidence intervals (thick horizontal bars) for this neuron.

Fig. 3.6: Example discharges from an example conjugate IBN during (A) converging disconjugate saccades, and (B) diverging disconjugate saccades. Note the good fits obtained when the conjugate parameters are used to predict the firing rate of this neuron (black trace on firing rate) compared to when the ipsilateral and contralateral eye velocity are used (blue and red traces respectively). The second row shows the estimated model fits using the binocular model (thick black curve). Estimated model fits using the reduced conjugate model are also shown (dashed gray curve). (C) Bootstrap histograms and 95% confidence intervals. 87

Notably, a minority of the SBNs in our population had no monocular tuning. Fig. 3.6 shows the discharge patterns of such an example „conjugate‟ SBN during disconjugate saccades.

The neural activity of this neuron was best predicted when conjugate (Fig. 3.6, superimposed black trace, VAFpred-conj=0.71) rather than ipsilateral or contralateral eye velocities (superimposed blue and red trace; VAFpred-ipsi = 0.5 and VAF pred-contra = 0.41) were the model inputs. The goodness-of-fit provided by the conjugate prediction was nearly as good as that provided by eq.(3.4) when its parameters were freely estimated (VAFpred-conj = 0.71 vs VAFest-bino = 0.73).

Furthermore, the estimated ipsilateral and contralateral eye velocity sensitivities of this neuron were statistically identical (Fig. 3.6C). Since, by definition, a neuron that has equal sensitivities to both eyes' movements has no vergence sensitivity (recall, vergence = left eye - right eye), this neuron similarly encoded conjugate eye movement dynamics during both conjugate and disconjugate saccades. Thus, as expected, when the ri and rc parameters in eq.(3.4) were replaced by a conjugate velocity sensitivity, the obtained model fit was nearly the same as that obtained with the full binocular model (Fig. 3.6, second panel, gray curve, VAFest-bino = 0.73,

VAF est-conj =0.71; ΔBIC=0).

For each EBN and IBN, a Ratiodyn index was computed based on the estimated parameters of eq. 3.4 (see METHODS) in order to objectively assign each neuron to one of five ocular categories (Table 2.1). The distributions of Ratiodyn obtained using this method for EBNs and IBNs are shown in Fig. 3.7A and 3.7B. Comparison of the results of our prediction and estimation-based analyses are shown in Fig. 3.7C. Each symbol represents an individual neuron.

The firing rates of neurons that were classified as monocular based on their Ratiodyn value, were best predicted when the preferred eye velocity was the input (i.e., red and blue circles above the line of unity in Fig. 3.7C), whereas as the firing rates of neurons that were classified as conjugate

Fig. 3.7 Distribution of RatioDyn indexes for (A) EBNs, and (B) IBNs. Columns marked with an asterisk indicate a greater number of binocular EBNs with a negative Ratiodyn (A) and a greater number of IBNs with a positive Ratiodyn (B). (C) Comparison of the results of the prediction and estimation-based analyses. Each symbol represents an individual neuron.

The firing rates of neurons that were classified as monocular based on their Ratiodyn value, were best predicted when the preferred eye velocity was the input (red and blue circles), whereas as the firing rates of neurons that were classified as conjugate were better predicted when conjugate eye velocity was the input to the model.

89 were better predicted when conjugate eye velocity was the input to the model (black squares below the line of unity in Fig. 3.7C). The average VAFs and difference in BIC provided by the complete binocular versus reduced models are summarized for each of the five categories of neurons in Table 3.3.

The above findings were confirmed when the analysis of our samples of EBNs and IBNs was performed again using the following model:

  FR(t  td )  b  rcj CJ (t)  rvg VG(t) (Vergence-est-model) (3.5)

  where CJ(t) and VG (t) are instantaneous conjugate and vergence velocities, respectively. We found that 75% of EBNs, and 70% of IBNs encode significant vergence velocity sensitivities

(rvg) during disconjugate saccades. Moreover, both eq.(3.4) and eq.(3.5) yielded similar conclusions on a neuron-by-neuron basis.

3.4.4 Comparison across neuron types

Statistical analyses were performed to compare the discharge properties of short and long lead EBNs and IBNs. During conjugate and disconjugate saccades, most parameters evaluated were statistically identical across all neuron types (P > 0.05, with the exception that biases were slightly less for short lead EBNs compared to long lead IBNs). While the distributions of long and short lead EBNs and IBNs across the five categories of ocular preference were not identical, the differences were highly non-significant (P = 0.98; 2-test on a 4 by 5 contingency table; see

METHODS). Moreover, for the metric-based analyses, there were no significant differences in the distribution of the four neuron types across the five categories of ocular preference (P = 0.36

and 0.76, NOS and peak analyzes, respectively; 2-test on a 4 by 5 contingency table). Hence, except for the prelude discharges of long lead neurons, all four neuron types tested had similar discharge properties during both conjugate and disconjugate saccades.

3.4.5 Calculation of the net premotor drive

The activity of SBNs was also characterized during OFF-direction (i.e., contralaterally directed) saccades to fully assess the premotor drive to the extraocular motoneurons. Overall, we found that the OFF-direction discharges were relatively minor during both conjugate and disconjugate saccades and thus had only a negligible impact on eye velocity. During conjugate saccades, the majority of EBNs (26/30) were completely silent, while remaining 4 neurons had very small bursts that were poorly related to saccade dynamics (mean VAF = 0.22±0.06;

Supplementary Table 3.4). Similarly, during disconjugate saccades, the same EBNs, as well as two additional EBNs, produced minor responses (mean VAF = 0.20±0.06). The majority of IBNs were also silent (30/44), and the remaining neurons produced a few spikes during OFF-direction conjugate (N = 14/44) and disconjugate (N = 16/44) saccades.

3.4.6 Metric analysis of conjugate and disconjugate saccades

To compare our sample of EBNs and IBNs with those previously described in the literature (Cullen and Guitton 1997; Kaneko et al. 1981; Scudder 1988; Strassman et al. 1986a; b; Yoshida et al. 1982), saccade-related burst activity was also characterized using classical metric-based analyses. During conjugate saccades, the number of spikes in a neuron‟s saccadic burst (NOS) was linearly related to the conjugate amplitude (CJ) of the eye movement (NOS = 92

b + zCJ; see Supplementary Table 3.2). In addition, the peak saccadic firing rates of neurons were also linearly correlated to peak conjugate velocity, although this relationship was generally

 more noisy (FRmax = b + p CJ max; see Table 3.4).

During disconjugate saccades, this approach was modified to account for the movements of both eyes (see METHODS eq(3.1) and eq(3.2)). The parameter values estimated with these models were then used to compute RatioNOS and RatioFRmax indices (see METHODS) that were used to objectively assign each neuron to one of the five categories (Table 3.1). The distributions of RatioNOS and RatioFRmax indexes are shown in Fig. 3.8 (left and right panels, respectively) for

EBNs and IBNs (A and B, respectively). The two analysis approaches yielded similar population results for both types of neurons. In general, most EBNs and IBNs were monocular (blue and red bars).

These results can be directly compared to those of Zhou and King (1998) which performed the only other characterization of EBN discharge during disconjugate saccades using a NOS-based approach identical to that shown in Eq.(3.1). For each neuron in their sample, Zhou and King computed an ocular index (OI), which can be converted to a Ratio index to facilitate comparison (see METHODS). The distribution of RatioNOS indexes that was estimated from

Zhou and King (1998) using this approach is shown in Fig. 3.8A (inset, top left panel). No significant differences in the distribution of neurons across the five categories of ocular preference (Table 3.1) were observed between the two studies (P = 0.53; 2-test on a 2 by 5 contingency table).

Fig. 3.8: Distribution of (A) RatioNOS indexes, and (B) RatioFRmax indexes for EBNs (top row) and IBNs (bottom row). Inset: distribution of RatioNOS for the sample of EBNs in Zhou and King (1998), estimated from their Fig. 2.3B.

3.4.7 Comparison of dynamic and metric analyses

It is important to note that on a neuron-by-neuron basis, similar conclusions were obtained using both the dynamic and NOS-based metric analyses (refer back to Fig. 3.5 for dynamic analysis results). Overall, 61% of the neurons were classified in the same category of ocular preference (Table 3.1) using both the dynamic and metric analyses. In the DISCUSSION section, we consider the limitations of the metric-based approach, which does not take advantage of the information encoded in the temporal pattern of action potentials.

3.4.8 Simulation design

Can the brainstem saccade generator provide the vergence drive required to drive the abducens motoneurons during disconjugate saccades? To address this question, a computer- based simulation was implemented (see methods) in which the discharges of saccadic burst neurons (present study) was combined with those of other saccade-related premotor neurons

(previous studies, Table 3.2) with known projections to the abducens nucleus. This allowed us to determine whether the SBNs encode sufficient information to shape motoneuron discharge during disconjugate as well as conjugate saccades, or whether premotor input from a separate vergence subsystem (see discussion of Busettini and Mays 2005b) is needed. The sign of projection for each premotor neuron was fixed based on physiological knowledge (Langer et al.

1986; Scudder and Fuchs 1992; Scudder et al. 2002) and their weights were optimized.

Importantly, pure vergence-related inputs (e.g., from vergence-velocity neurons) were deliberately omitted from the simulation. Example reconstructed average population drives during a conjugate saccade are shown in Fig. 3.9.

Fig. 3.9: Schema of the circuitry used for simulations. Eight premotor neuron types were selected. The nature of each connection is indicated in the box labeled ABNs as a (+) for excitatory connections, or a (-) for inhibitory connections. The shading of neurons‟ cell bodies indicates whether they increase (dark gray) or decrease (light gray) their discharges for a rightward saccade. „Optimal‟ weights are superimposed on each connection. The thickness of arrows indicate the relative contribution of each neuron type during saccades, using the shown weights. Reconstructed population discharges are shown for each neuron type during an example conjugate saccade (bottom row, gray shaded areas). Weights were optimized on a conjugate dataset that included: 1) fixation 2) sinusoidal smooth pursuit; 3) passive sinusoidal whole-body rotation with an earth-fixed target; 4) passive sinusoidal whole-body rotation with a head-fixed target; 5) saccades.

Fig. 3.10: (A) Plot of the eight premotor neurons‟ estimated weights as a function of the ipsilateral BTs weight. Note that weights are plotted as absolute values to facilitate comparison between excitatory and inhibitory neurons. The arrow indicates the „optimal‟ weight set found during disconjugate saccades (see Fig. 3.9A). (B) The variance-accounted-for (VAF) obtained for each weight set used to reconstruct ABN discharges during conjugate saccades.

The weight values estimated on the conjugate dataset (Fig. 3.10) were used to predict the discharge of ABNs on a dataset of disconjugate saccades (i.e., a dataset different from that originally used to estimate the weights). The „optimal‟ weight set (defined as the weight set that yielded the largest VAF during the prediction) accounted for 98% of the variance contained in average ABN firing rates during disconjugate saccades (Fig. 3.11A). As is shown in Fig. 3.11B, the goodness-of-fit for the prediction was equivalent to that obtained when the weights were estimated during conjugate saccades. Notably, this general result held for all weight sets (i.e., regardless of if the BT weight was set relatively small or large). Thus this result supports the proposal that additional vergence-related premotor signals are not required to shape the activity of abducens motoneurons.

3.4.9 Analysis of simulation weights

The weight sets were analyzed based on the contribution of each neuron type in shaping

ABN discharges (where we define contribution = wi* FR i(t) /  ABN(t); “i” represents one of eight premotor neuron types). Fig. 3.11C plots the contribution of the eight premotor neuron types as a function of the weight of ipsilateral BT neurons during saccades.

Two main observations can be made from the „optimal‟ weight set. First, the estimated weights of ipsilateral EBNs and contralateral IBNs were approximately the same. This is consistent with both neuron types projecting with similar densities to the abducens nucleus

(Strassman et al. 1986a; b), and with IBNs inhibiting most ( > 80%) contralateral ABNs during ipsilateral saccades (Sylvestre and Cullen 1999). Second, additional excitatory drives to the

abducens nucleus during saccades originate from ipsilateral BTs and from contralateral EHs and type I PVPs. In addition, as expected contralateral BTs and IBNs, and ipsilateral EHs and PVPs, provide inhibitory inputs to ABN motoneurons during saccades. Note that EHs and PVPs provide equal excitatory contributions, which are half that provided by BTs. This ratio is consistent with that previously deduced to predict ABN discharges during conjugate smooth pursuit, VOR in the dark, and cancellation of the VOR (Cullen et al. 1993b).

3.5 DISCUSSION

The present results conclusively demonstrate that: 1) The brainstem saccadic generator, which is commonly assumed to drive only the conjugate component of eye movements, carries substantial vergence-related information that is temporally and dynamically related to the dynamics of disconjugate saccades and 2) the resulting premotor command is, in fact, sufficient to drive the agonist extraocular motoneurons during disconjugate saccades. Thus, the premotor command to generate diverging saccades is present in what had been mistakenly assumed was the “conjugate” saccade generator. Overall, our experimental and theoretical results strongly support the hypothesis that the brainstem saccadic circuitry shapes the activity of abducens nucleus neurons by encoding integrated conjugate and vergence premotor commands (Cova and

Galiana 1996; King and Zhou 2002; 2000; Sylvestre et al. 2003).

100

Fig. 3.11: (A) Prediction VAFs for the dataset of disconjugate saccades plotted as a function of the ipsilateral BTs weight used to determine the “optimal” weight set. (B) Comparison of ABN firing rates reconstructed using models estimated on the data (gray shaded areas) and using the simulation (white shaded areas) during three conjugate (left) and three disconjugate saccades (right). (C) Contribution of the eight premotor neuron types to shaping ABN discharges, plotted as a function of the ipsilateral BTs weight. Positive and negative values indicate excitation and inhibition, respectively.

101

3.5.1 Comparison with previous reports: conjugate saccades

The responses of SBNs during conjugate saccades were first characterized using traditional metric-based approaches and results highly comparable to those that have been previously described were obtained (Cullen and Guitton 1997; Hepp and Henn 1983; Keller

1974; Luschei and Fuchs 1972; Scudder 1988; Strassman et al. 1986a; b; Van Gisbergen et al.

1981). Next, an analysis of the dynamic relationship between the IBN discharges and eye velocity yielded results equivalent to those of prior studies (Cullen and Guitton 1997). Finally, we found that EBNs encode saccade dynamics during conjugate saccades with the same accuracy as do IBNs. This latter finding confirms the proposal that EBNs, like IBNs, encode saccade trajectories in their spike trains (Eckmiller et al. 1980; Hepp and Henn 1983; Robinson

1973; Van Gisbergen et al. 1981).

3.5.2 The timing and dynamics of SBN burst activity are appropriate to facilitate vergence during disconjugate saccades

While numerous studies have shown that the timing of SBNs‟ bursts are appropriate to drive conjugate saccades, this study is the first to have systemically evaluated the burst timing during saccade-free slow vergence and disconjugate saccades. During disconjugate saccades burst onsets and durations were tightly linked to saccade onsets and durations, respectively. In contrast, SBNs did not discharge during periods of saccade-free vergence that preceded or followed disconjugate saccades. Since the onset of saccade-facilitated vergence is coincident with saccade onset (Busettini and Mays 2005a), our results show that the premotor drive from the saccadic burst generator is appropriately timed to facilitate vergence velocity during

102

disconjugate saccades. Moreover, these results compliment those of Busettini and Mays (2003) showing that the OPN pause is similarly linked to the saccadic component of disconjugate saccades. Taken together these findings provide strong evidence that the vergence-facilitation observed during disconjugate saccades only occurs when the saccadic burst generator (i.e.,

SBNs) is active.

To determine whether SBNs encode similar signals during conjugate and disconjugate saccades, we analyzed neuronal discharges during disconjugate saccades using metric-based and dynamic-based approaches. Only one previous study had evaluated SBNs during disconjugate saccades (Zhou and King 1998). The study used a metric-based approach and its overall conclusion was that the burst activity of EBNs was better correlated with the movement of an individual eye than the conjugate eye movement during disconjugate saccades.

In the present study, results highly comparable to those of Zhou and King (1998) were obtained for IBNs as well as EBNs. However, the NOS-based approach is limited because it ignores important information that is encoded within neuronal discharge dynamics (Cullen and

Guitton 1997b; Eckmiller et al. 1980; Eckmiller and Mackeben 1980; Hepp and Henn 1983;

Robinson 1973; Van Gisbergen et al. 1981). As a result, these findings are consistent with the hypothesis that the premotor saccadic circuitry plays a role in facilitating vergence shifts, but cannot rule out the alternative possibility that the overall contribution of the premotor saccadic circuitry is relatively unimportant compared to that of the vergence subsystem (e.g., see discussion of (Busettini and Mays 2005b). To determine whether SBNs play a significant role in facilitating vergence velocity during disconjugate saccades we used a dynamic-based approach to explicitly describe the relationship between the temporal pattern of neuronal firing and the

103

velocity of each eye. Our results show that vergence-related signals are dynamically encoded in the burst of most SBNs during disconjugate saccades.

3.5.3 The role of the saccadic burst generator during saccade-vergence interactions

Several models have been proposed to account for how vergence is facilitated when it is combined with a saccade (Busettini and Mays 2005a; Cova and Galiana 1996; 1995; King and

Zhou 2002; Kumar et al. 2006; Mays and Gamlin 1995; Zee et al. 1992). The results of recent behavioural studies, which have shown that saccade dynamics are an important determinant of the dynamics of vergence facilitation, have led to the rejection of the two most influential models, namely the “Multiply Model” and the “Saccade-Related Vergence Burst Neuron Model”

(Busettini and Mays 2005b; Zee et al. 1992; Zhang et al. 1992). Accordingly, to account for the correlation between conjugate and vergence movements, these investigators have proposed that

SBNs exclusively encode conjugate saccadic dynamics, and that projections from these conjugate SBNs to the vergence premotor pathway underlie the vergence facilitation during disconjugate saccades.

Our findings do not support this proposal. Firstly, if the drive from the vergence burst neurons encoded all of the vergence command to drive the vergence part of the saccadic eye movement during disconjugate saccades (i.e. suggested by Busettini and Mays 2005b) this would imply that the SBNs neurons should then code only conjugate eye movements. We do not see this in our data; the results of both the metric-based and dynamic-based approaches demonstrated that SBNs do not solely encode conjugate saccade dynamics. Alternatively, one could consider an intermediate case in which vergence burst neurons encoded half of the vergence command to drive the vergence part of the saccadic eye movement during disconjugate saccade. However, in

104

this scenario our analysis would have estimated larger gains for the preferred eye during disconjugate saccades relative to the gain estimated during conjugate saccades. Again, this was not observed. Indeed, we found that discharge of the majority of neurons during disconjugate saccades could be predicted based on the gains estimated during conjugate saccades (see Fig.

3.5).

Our computer-based simulations further support the proposal that the vergence-related information encoded by the premotor saccadic circuitry can largely account for vergence facilitation during disconjugate saccades. Notably, motoneuron discharges were well predicted despite the highly conservative assumptions that were made: 1) weight sets were estimated to reproduce abducens nucleus neuron discharges during five different conjugate behaviors (Cullen et al. 1993; Hazel et al. 2002), and were then validated on disconjugate saccades that were not included in the estimation process and 2) simulating average population discharges rather than individual neuron discharges restricted the number of weights to optimize, and thus the simulation's flexibility. Moreover, this conclusion was very robust in that it was not sensitive to the specific weight set used (see Fig. 3.12A). Taken together, our results are consistent with a model in which integrated control at the level of the brainstem saccadic burst generator drives saccades in three-dimensional space since the vergence-information encoded by this premotor pathway is largely sufficient to drive abducens motoneurons during disconjugate saccades (Fig

3.12A).

Notably, as shown in Fig. 3.2 and Fig. 3.3, SBNs are silent during the slow component of disconjugate gaze shifts and pure vergence movements, respectively. Overall, this suggests that while the SBNs function to rapidly drive the eyes to a new position, an additional command (i.e., from premotor neurons that project directly to the abducens nucleus) is required to ultimately

105

align the eyes on target. While no such input has been identified to date, a preliminary report has described neurons encoding slow vergence information near the abducens nucleus (Gnadt et al.

1988). Our study was not designed to address whether such an additional input is driven by a shared controller or separate subsystem. Nevertheless, it is likely that both groups of premotor neurons would be under the control of an integrated controller where for example, the BNs might have a higher threshold for activation (in terms of motor error); such an organization would ensure accurate redirection of the gaze axes.

To model the neural control of horizontal eye movements it is important to also consider the activation of the medial rectus muscle. Lesion studies (Gamlin et al. 1989) and anatomical studies (Highstein and Baker 1978) suggest that the abducens internuclear neurons (AINs) provide a primary input to the medial rectus subdivision of the contralateral oculomotor nucleus for the control of saccades. Both AINs and motoneurons discharge similarly during slow eye movements and conjugate saccades (Gamlin et al. 1989; Mays and Porter 1984; Sylvestre and

Cullen 2002). Moreover, while King and Zhou (2002) reported that the majority of AINs had a preference for the contralateral eye during pursuit (note, the method of identification was not explicitly stated), Sylvestre and Cullen (2002) found that AIN and motoneurons appear to have comparable ocular tuning during disconjugate saccades. If AINs and AMNs do have similar discharge patterns, then an additional command would be required at the level of the medial rectus motoneurons (Cova and Galiana 1996; Gamlin et al. 1989; King and Zhou 2002) to appropriately activate the medial rectus muscle.

Inputs from vergence specific neurons located in the midbrain are likely candidates to provide at least some of this input (Fig. 3.12B); these neurons carry vergence-related information and project to the medial rectus motoneurons in the oculomotor nucleus (e.g., near response

106

Fig. 3.12: Neural circuitry involved in generating disconjugate saccades. (A) When the lateral rectus is the agonist muscle (e.g. during diverging saccades), the SBNs provide the abducens motoneurons with an integrated vergence/conjugate command to drive the movement of the eye. (B) Framework for the control of disconjugate saccades when the medial rectus is the agonist muscle. When the eye moves medially (i.e. during converging saccades), inputs from AINs provide an important input for driving the saccade. However, since abducens motoneurons and abducens internuclear neurons discharge similarly during conjugate and disconjugate saccades, an additional vergence input is required. For simplicity, the burst tonic neurons (BTs) have been excluded from this schematic diagram. Notably, these neurons also have direct inputs to the ABN and also encode vergence related information (Sylvestre et al. 2003). OIN, oculomotor internuclear neurons; AMN, abducens motoneurons; AIN, abducens internuclear neurons; OMN, oculomotor motoneurons; III and VI, oculomotor and abducens nuclei; LR and MR, lateral and medial recti.

107

neurons) (Mays et al. 1986; Zhang et al. 1992). However, it is unclear whether the discharge from these vergence neurons would be sufficient to account for the saccadic vergence related activity of the OMNs. For example, it has been hypothesized that the addition of vergence-related signals encode a difference in eye position signals encoded by monocular integrators (i.e., nucleus prepositus or vestibular neurons) that would ultimately reflect a binocular alignment signal (King and Zhou 2002). The predictions from these circuits (Fig. 3.12

A and B) are also consistent with clinical studies of internuclear ophthalmoplegia (Leigh and Zee

1999) showing that interruption or inactivation of the INN pathway impairs adductions (i.e., movements controlled but the medial rectus) but preserves abduction.

3.5.4 Source of vergence-related signals

Although the source of the vergence-related signals to SBNs remains unknown, there are several putative sites. Two likely candidates are the central mesencephalic reticular formation

(cMRF) and the superior colliculus (SC). Both of these structures receive inputs from disparity sensitive cortical regions including the frontal eye fields (Ferraina et al. 2000; Gamlin and Yoon

2000) and lateral intraparietal area (Gnadt and Beyer 1998). Furthermore, stimulation of the SC and cMRF have clear affects on vergence movements (Chaturvedi and Van Gisbergen 1999;

2000; Luque et al. 2006; Suzuki et al. 2004; Waitzman et al. 2007). Consistent with these findings are the recent reports that cMRF neurons dynamically encode the movement of an individual eye (Waitzman et al. 2007) and SC neuronal activity is altered when saccades are accompanied by vergence (Walton and Mays 2003). Although, since the modulation of primate

108

SC neurons was observed to be more robust for purely conjugate than disconjugate saccades, it has been suggested that the SC is not tuned in three dimensions (Walton and Mays 2003).

However, given that 1) the SC has strong anatomical connections to saccadic neurons in the

PPRF and cMRF (Moschovakis et al. 1988) and 2) neurons in both of these regions have been found to dynamically encode the movement of an individual eye (present study, (Waitzman et al.

2007) suggests that neurons in the SC should be re-examined for evidence of an individual eye command.

Another possible source of vergence inputs to SBNs is the fastigial nucleus of the cerebellum.

This nucleus projects directly to the SBN region (Noda et al. 1990), and contains neurons with saccade-related activity (Fuchs et al. 1993). While it is not known whether these neurons encode vergence modulations during disconjugate saccades, neurons in the adjacent interposed nucleus can carry vergence and disparity information (Zhang and Gamlin 1998). Finally, it is also possible that brainstem vergence-velocity or near-response neurons could project to the SBN region. However, there is currently no evidence available to support this projection (Gamlin

1999). Future studies of vergence/sacccade interactions in other brain areas are required for further testing of this framework.

109

110

111

Supplementary Fig. 3.1. Simulation results. (A) Plot of the eight premotor neurons‟ estimated weights as a function of the ipsilateral BTs weight. Note that weights are plotted as absolute values to facilitate comparison between excitatory and inhibitory neurons. The arrow indicates the „optimal‟ weight set found during disconjugate saccades (see Fig. 9A). (B) The variance-accounted-for (VAF) obtained for each weight set used to reconstruct ABN discharges during conjugate saccades.

112

Chapter 4

Vertical facilitated vergence by premotor saccadic burst neurons

Chapter 3 provided evidence that premotor burst neurons do not encode conjugate commands but rather dynamically encode the movement of an individual eye. In this chapter I directly test this finding using a combination of complementary behavioral and recording approaches. I first determined whether vertical saccades – like horizontal saccades – are effective in facilitating vergence velocities. I then recorded the neural activity of SBNs during this task, which required a vergence but no horizontal conjugate component, to determine if SBNs dynamically encode monocular signals that are appropriate for generating the observed facilitated vergence velocities.

This chapter has been adapted from Van Horn MR and Cullen KE. Dynamic coding of vertical facilitated vergence by premotor saccadic burst neurons. J Neurophysiol 100:1967-1982, 2008.

113

4.1 ABSTRACT

To redirect our gaze in three–dimensional space we frequently combine saccades and vergence. These eye movements, known as disconjugate saccades, are characterized by eyes rotating by different amounts, with markedly different dynamics, and occur whenever gaze is shifted between near and far objects. How the brain ensures the precise control of binocular positioning remains controversial. It has been proposed that the traditionally assumed

“conjugate” saccadic premotor pathway does not encode conjugate commands but rather encodes monocular commands for the right or left eye during saccades. Here, we directly test this proposal by recording from the premotor neurons of the horizontal saccade generator during a dissociation task that required a vergence but no horizontal conjugate saccadic command.

Specifically, saccadic burst neurons (SBNs) in the paramedian pontine reticular formation were recorded while rhesus monkeys made vertical saccades made between near and far targets.

During this task, we first show that peak vergence velocities were enhanced to saccade-like speeds (e.g., > 150deg/sec versus < 100deg/sec during saccade-free movements for comparable changes in vergence angle). We then quantified the discharge dynamics of SBNs during these movements and found that the majority of the neurons preferentially encode the velocity of the ipsilateral eye. Notably, a given neuron typically encoded the movement of the same eye during horizontal saccades that were made in depth. Taken together our findings demonstrate that the brainstem saccadic burst generator encodes integrated conjugate and vergence commands, and thus provide strong evidence for the proposal that the classic saccadic premotor pathway controls gaze in three dimensional space.

114

4.2 INTRODUCTION

In order to quickly and accurately redirect our gaze between near and far targets, we typically combine saccadic and vergence eye movements. During such eye movements, termed disconjugate saccades, the eyes rotate by different angles and with different trajectories to precisely realign the two visual axes on the new target of interest. Traditionally, disconjugate saccades were thought to be controlled by linear summation of premotor commands from two distinct neural control pathways which separately encode the conjugate and vergence components of eye motion: i) a conjugate saccadic subsystem, which commands a rapid but yoked movement of the two eyes in a given direction, and ii) a separate vergence subsystem, which rotates the eyes in opposite directions to ensure accurate binocular positioning (Hering

1977; Mays 1998; 1984). Accordingly, the premotor circuitry involved in generating horizontal saccades (e.g., the saccadic burst neurons of the paramedian pontine reticular formation (PPRF)) was generally assumed to provide the command to drive the horizontal conjugate component of such movements, whereas a specific subpopulation of neurons in the mesencephalic reticular formation (MRF), which encode a signal proportional to viewing distance, were thought to produce the required vergence command (Busettini and Mays 2005b; Gamlin 2002; Gamlin et al.

1989; Mays et al. 1986; Zhang et al. 1992).

The summation of commands from two distinct premotor pathways, however, cannot account for a number of observations that have been made regarding disconjugate saccades.

Notably, vergence velocities reach values greater during disconjugate saccades than would be predicted by the linear summation of commands from separate saccadic and vergence premotor pathways (Busettini and Mays 2005b; Collewijn et al. 1997; Enright 1984; Enright 1992;

Maxwell and King 1992; Ono et al. 1978; Oohira 1993; van Leeuwen et al. 1998; Zee et al. 115

1992). Thus, it is now generally recognized that the commands driving conjugate saccades and vergence eye movements are not generated by strictly independent neural subsystems. To date, however, the mechanism responsible for the facilitation of vergence during disconjugate saccades remains controversial.

Two general classes of models have been proposed to account for the facilitation of vergence during disconjugate saccades (Busettini and Mays 2005b; Gamlin 2002; King and Zhou

2002; Kumar et al. 2006; Mays 1998; Mays and Gamlin 1995b; Scudder et al. 2002; Van Horn et al. 2008; Zhou and King 1998). In one model, the premotor saccadic pathway drives the conjugate component of the saccade, while changes in vergence are exclusively driven by a premotor command from vergence neurons. In this view, projections from the “conjugate” saccadic pathway to the “vergence” pathway play a pivotal role in enhancing the premotor vergence command during disconjugate saccades (Busettini and Mays 2005b; Kumar et al.

2005a). Alternatively, it has been proposed that classically assumed “conjugate” saccadic structures in the oculomotor brainstem underlie vergence facilitation by providing monocular saccades commands to the abducens nuclei during saccades (Cova and Galiana 1996; King and

Zhou 2002; 2000). In this view, the vergence pathway is used to adjust ocular alignment following the saccadic component of the movement. Consistent with this latter proposal, we and others have reported that the premotor burst neurons in the PPRF that drive horizontal saccades do not encode conjugate commands (King and Zhou 2000; McConville et al. 1994; Sylvestre et al. 2003; Zhou and King 1996; 1998). Indeed, we have recently shown that the vergence-related information dynamically encoded by the premotor brainstem saccadic circuitry alone is sufficient to shape the activity of the abducens motoneurons during horizontal disconjugate saccades (Van

Horn et al. 2008).

116

In the present study we tested the proposal that vergence is facilitated by the classical horizontal saccadic pathway using a combination of complementary behavioral and recording approaches in rhesus monkeys. While prior single unit studies had exclusively focused on neural correlates during horizontal disconjugate saccades (Van Horn et al. 2008; Zhou and King 1998), there is evidence suggesting that vergence might be similarly facilitated during vertical saccades

(Busettini and Mays 2005a; Enright 1984; Kumar et al. 2005; Maxwell and King 1992; van

Leeuwen et al. 1998; Zee et al. 1992). Understanding how the brain drives vertical saccades between near and far targets is particularly interesting in terms of the current debate regarding the premotor control of vergence during saccades. While these saccades require a vertical conjugate command, which would originate from the vertical burst neurons of the rostral interstitial nucleus of the medial longitudinal fasciculus (riMLF) (Buttner et al. 1977; Crawford and Vilis 1991; 1992; King and Fuchs 1979; Missal et al. 2000; Moschovakis et al. 1991a;

Moschovakis et al. 1991b) they do not require the simultaneous production of a horizontal conjugate command. Instead, a command to generate horizontal movements of the two eyes in equal and opposite directions (i.e., vergence) is needed. Thus, by recording the discharges of horizontal saccadic burst neurons (SBNs) during this dissociation task, we were able to address whether neuronal commands from the horizontal saccadic pathway dynamically encode the movement of an individual eye even when no horizontal conjugate saccade command is required.

Our results provide firm evidence that vergence is facilitated during vertical saccades, and that integrated conjugate-vergence information encoded by the classical horizontal saccadic pathway is appropriate to drive the observed facilitation.

117

4.3. METHODS

4.3.1 Animals and surgical preparations

The neurons in this study were obtained from three rhesus monkeys (Macaca mulatta).

The monkeys were prepared for chronic extracellular recording using the aseptic surgical procedures described previously (Sylvestre and Cullen 1999). Briefly, a stainless steel post was attached to the animal's skull with stainless steel screws and dental acrylic permitting complete immobilization of the animal's head. Two stainless steel recording chambers, oriented stereotaxically toward the abducens nucleus on the right and left side of the brainstem, were also secured to the implant. To record binocular eye position an eye coil (3 loops of Teflon coated stainless steel wire, 18-20 mm diam) was implanted in each eye (Judge et al. 1980). All procedures were approved by the McGill University Animal Care Committee and complied with the guidelines of the Canadian Council on Animal Care.

4.3.2 Behavioral paradigms

Monkeys were trained to fixate targets for a juice reward. The timing and location of target illumination, data acquisition and on-line data displays were controlled using REX, a

UNIX-based real-time acquisition system (Hayes et al. 1982). Neuronal responses were recorded during 1) horizontal and vertical conjugate saccades, 2) oblique saccades, 3) saccade-free symmetric vergence 4) vertical saccades combined with vergence and 5) horizontal saccades combined with vergence.

First, to elicit conjugate movements a red HeNe laser target was projected onto a cylindrical screen located 55 cm away from the monkey's eyes (isovergent, 3.5 deg

118

convergence). Ipsilaterally and contralaterally directed conjugate saccades were elicited by stepping the laser target between horizontal positions (5-30 deg), in 5 deg increments, in predictable and unpredictable sequences and vertical saccades were elicited by stepping the laser target between vertical positions (5-30 deg). Oblique saccades were generated by stepping the laser target between a central target to a sequence of targets that had varying vertical and horizontal components within this same range.

Next, to elicit changes in vergence a horizontal array of 16 red light emitting diodes

(LEDs), with intensities comparable to that of the laser target, was positioned between the screen and the monkey. Symmetric vergence was elicited by sequentially illuminating LEDs located along the midline (convergence angles: 17, 12, 8 and 6 deg). To generate vertical saccades with vergence two specific paradigms were used, similar to those previously described in humans

(Kumar et al. 2005a; van Leeuwen et al. 1998):

(A) A far laser target was located higher than a central near LED, i.e., Up

(B) A far laser target was located below a central near LED, i.e., Down

The near LED target was located 8 cm in front of the monkey‟s nose and the upper and lower targets were located 20 deg above and below the near target, respectively (Fig. 4.1). These paradigms generated vertical saccades with a horizontal vergence amplitude of ~10 deg. For example, when the monkey looked from a “far-up” target to a “near” LED the monkey made a downward converging movement (Converge-Down) whereas when the monkey looked from a

“near” LED to “far-up” laser target the monkey made an upward diverging movement (Diverge-

Up). Notably, the two eyes moved in opposite directions such that the change in the horizontal conjugate [i.e., (left eye + right eye)/2] component was negligible. 119

Fig. 4.1: Schematic representation of paradigms used to generate combined vertical-vergence movements. LEDs located between the far screen and monkey‟s nose are used to generate vergence movements. Specifically, LEDs lit up along the midline were used to generate pure symmetric vergence (convergence angles: 17, 12, 8 and 6 deg). To elicit combined vertical vergence movements a near target was located 8cm in front of the monkey‟s nose and the upper and lower targets (Up and Down, respectively) were located 20° above and below the near target. Abbreviations: An eye movement from made from an “Up” target to a central near LED is denoted (Converge-Down) and a movement from a near LED to “Up” laser target would be (Diverge-Up).

120

Finally, to compare our sample of SBNs to those previously described (Sylvestre and

Cullen 2002; Van Horn et al. 2008) neurons were also characterized during horizontal saccades combined with vergence (i.e., horizontal disconjugate saccades). The paradigms to elicit horizontal disconjugate saccades have been described previously (Sylvestre et al. 2003; Sylvestre and Cullen 2002; Van Horn et al. 2008). Briefly, an illuminated target changed from one of the close midsagittal LEDs to an eccentric (i.e., right or left of the midsagittal plane) laser target.

During this paradigm, monkeys made saccades with horizontal components of 5-30 deg in amplitude in both directions, and vergence components with amplitudes ~4-13 deg. In addition, some LEDs were positioned in a configuration similar to the Müller paradigm [see (Ramat et al.

1999) for examples] in order to elicit disconjugate saccades in which the movement of the right eye or left eye was minimized.

4.3.3 Data acquisition

During experiments monkeys were seated in primate chair located within the center of a

1-m3, magnetic eye coil system (CNC Engineering). Horizontal and vertical eye position signals were measured using the magnetic search coil technique (Fuchs and Robinson 1966a; Judge et al. 1980). Each eye coil signal was calibrated independently by having the monkey fixate, with one eye masked, a variety of targets at different horizontal and vertical eccentricities and different depths. Position signals were low-pass filtered at 250 Hz (analog 8 pole Bessel filter) and sampled at 1 kHz. Since ocular saccades include very little power above 50Hz (Cullen et al.

1996b; Van Opstal et al. 1985; Zuber et al. 1968) eye position signals were further digitally filtered (with a 51st order finite-impulse-response filter with a Hamming window and a cut-off at

121

125 Hz), before being differentiated to obtain eye velocity signals (using zero-phase forward and reverse digital filtering to prevent phase distortion).

Extracellular single unit activity was recorded using enamel insulated tungsten microelectrodes [2-10 M impedance, Frederick Haer; for details, see (Sylvestre and Cullen

1999)]. Saccadic burst neurons (N=57) were identified on-line by their stereotypical discharge properties during eye movements (Cullen and Guitton 1997). Excitatory and inhibitory burst neurons (EBNs and IBNs, respectively) were distinguished based on their recording location relative to the abducens nucleus. EBNs were recorded in a small region extending 1-2mm rostral to the abducens nucleus, and 0.5-1.5mm from the midline. IBNs were recorded in a region extending 0-2mm caudal to the abducens nucleus, and 0.5-1.5mm from the midline. Both areas correspond to previous anatomical characterizations (Strassman et al. 1986a; b). When a neuron was isolated, unit activity, horizontal and vertical positions of the right and left eyes, and target position were recorded on a digital audio tape (DAT). The isolation of each neuron was reassessed offline during playback. A burst neuron was considered to be adequately isolated only when individual action potential waveforms could be discriminated using a windowing circuit

(BAK) during saccades (e.g., see Fig. 1 in Sylvestre and Cullen 1999), and during fixation.

Subsequent analysis was performed using custom algorithms (Matlab, The MathWorks).

4.3.4 Definitions and Conventions.

122

velocity signals denote convergence and divergence, respectively. In addition, we report the movements of each eye as either ipsilateral or contralateral based on their location relative to the recording site. Note, positive and negative values indicate eye positions that are to the right and left, or up and down of the central position (i.e., straight ahead), respectively.

4.3.5 Data Analysis

The onset and offset of all saccades was determined using a 20 deg/s saccade velocity

(i.e., horizontal or vertical) criterion. Horizontal saccades were defined as movements for which changes in vertical eye position were < 10% of the change in horizontal position, vertical saccades were defined as movements for which changes in horizontal eye position were < 10% of the change in vertical position. Saccades were categorized as conjugate if the change in vergence angle was < 3.0 deg and further categorized as vertical, horizontal or oblique. The onset and offset of slow, saccade-free vergence was determined using 10 deg/s vergence velocity criteria. Symmetrical vergence was defined as movements with a change in vergence > 2.5 deg that were not accompanied by saccades.

The preferred direction for each neuron was determined by fitting a Gaussian function to the relationship between the number of spikes (NOS) in a unit‟s discharge and saccade direction for saccades ranging in amplitude from 20 to 25 deg. A spike density function, in which a

Gaussian function was convolved with the spike train (SD of 5 ms), was used to estimate neuronal firing rate (Cullen et al. 1996a; Sylvestre and Cullen 1999). Linear optimization techniques were used to quantify each neuron‟s dynamic sensitivity to eye movements, during conjugate saccades (Cullen and Guitton 1997; 1996; Sylvestre and Cullen 1999) and

123

disconjugate saccades (Sylvestre et al. 2003; Sylvestre et al. 2002; Van Horn et al. 2008), as has been described previously. Briefly, for each neuron we estimated the sensitivity to ipsilaterally directed conjugate saccades using the following dynamic model which we have previously shown is an accurate description of both EBNs and IBNs (Cullen and Guitton 1997; Sylvestre and Cullen 2006; Van Horn et al. 2008):

 FR(t)  b  rhori CJ (t  td ) Conjugate-Estimation model (4.1) where FR(t) is the neuron‟s instantaneous firing rate, b and rhori are constants that represent the bias and the neuron‟s horizontal eye velocity sensitivity respectively. td refers to the dynamic

 lead time and CJ(t)refers to the instantaneous horizontal conjugate eye velocity.

The specific linear regression models used for the analysis of neural responses during disconjugate saccades are elaborated in the RESULTS. The goodness-of-fit of a given model to neuronal data was quantified using the Variance-Accounted-For {VAF =1 - [var (mod - fr)/var

(fr)]}, where mod represents the modeled firing rate and fr represents the actual firing rate}.

Note, that when estimating linear models, the VAF is mathematically equivalent to the correlation coefficient R2. Accordingly, a VAF value of 1 indicates a perfect fit to the data, whereas a value of 0 indicates a fit that is equivalent to the mean value of the firing rate models.

The dynamic lead time of individual neurons (td) was computed during conjugate saccades as previously described by Sylvestre and Cullen (1999).

For each model parameter, which was estimated in our analysis of neuronal firing rates during disconjugate saccades, we computed 95% confidence intervals using a non-parametric bootstrap approach (Carpenter and Bithell 2000) and used these confidence intervals to identify non-significant or identical model parameters (Sylvestre et al. 2003; Sylvestre and Cullen 2002;

124

Van Horn et al. 2008). Notably, an equal number of converging and diverging saccades were included in the disconjugate dataset to prevent biasing the parameter estimates. If a confidence interval overlapped with zero the model was re-run with the non-significant term removed. The

Bayesian information criterion (BIC), which served as a “cost index”, was calculated for each model estimation to quantitatively determine whether removing the term was justified (Schwartz

1978).

4.3.6 Quantification of ocular preference

The ocular preference of each neuron was quantified as previously described (Sylvestre et al. 2003; Sylvestre and Cullen 2002; Van Horn et al. 2008). Briefly, for a given neuron the velocity sensitivity of each eye was used to compute a Ratio index:

Ratiodyn = (smaller estimated parameter value) / (larger estimated parameter value).

Then, to indicate which eye provided the larger parameter value (i.e., the neuron's "preferred eye"), each Ratio index was assigned an "i" or a "c", for the ipsilateral or contralateral eye, respectively. Using this approach, neurons were assigned to one of five categories, namely; monocular ipsilateral, monocular contralateral, binocular ipsilateral, binocular contralateral or conjugate [see Table 1 in Van Horn et al. (2008) for specific criteria for each category].

4.3.7 Statistical analysis

125

Data presented in the results are described as means ±standard deviations. A one way ANOVA followed by a Tukey-Kramer Multiple-Comparison test was used to compare results across behavior and neuron types.

4.4 RESULTS

The goals of this study were two-fold. In order to establish that vertical saccades - like horizontal saccades - are effective in facilitating vergence in monkeys, we first compared vergence velocities during saccades and slow movements which required a comparable change in vergence angle. In addition, we determined whether, during such vertical saccades, peak vertical saccadic velocity and vergence velocities were temporally aligned. Our second goal was to characterize SBN firing rates during combined vertical-vergence movements to determine whether they dynamically encode vergence-related information (i.e., monocular signals) in a manner appropriate to facilitate vergence during vertical saccades. We then compared the command provided by a given SBN when vergence is facilitated during vertical versus horizontal saccades.

4.4.1 Characterization of vergence facilitated by vertical saccades

Fig. 4.2 shows average traces of the four vertical-vergence movements studied. Position and velocity traces are shown for monkey D making saccades from a central near target to a higher far target (A: Diverge-Up), from the higher far target to the lower central near target (B:

Converge-Down), from a lower far target to the higher near target (C: Converge-Up) and from the near target to a lower far target (D: Diverge-Down). In each example, fixation of the LED

126

Fig. 4.2: Typical examples of behavior during disconjugate saccades made from a far up vertical target to a central near LED (Converge-Down), from a central near LED to a far up vertical target (Diverge-Up), from a far down vertical target to a central near LED (Converge-Up) and from a central near LED to a far down vertical target (Diverge-Down). Average individual eye velocity (i.e., ipsilateral (IE) and contralateral (CE), blue and green traces, respectively) and vertical (VT), conjugate (CJ) and vergence (VG) velocities (grey, black and red traces, respectively) are plotted as a function of time for each paradigm. Average traces are superimposed on standard deviations. Corresponding vertical and vergence positions are shown in the insets for each paradigm. Closed and open arrows are used to mark peak vertical and vergence velocities, respectively. 127

that was positioned 8cm from the monkey‟s nose required a vergence angle of ~10°. Note that since the velocities of the two eyes were equal and opposite in all four conditions, the horizontal conjugate velocity associated with each eye movement was virtually zero. During each of the four paradigms vergence velocities were i) substantially greater than expected for saccade-free vergence movements and ii) peak vergence velocities were relatively synchronized with peak vertical velocities (see arrows in Fig. 4.2). Below, we further quantify each of these findings.

We first compared the peak vergence velocities generated during each paradigm with those produced during pure symmetric vergence tasks requiring comparable changes in vergence angle. As illustrated in Fig. 4.3 vergence velocities were much faster when combined with a vertical saccade (P < 0.05) in all four of the paradigms and in all three monkeys. Average peak vergence velocities for each of the paradigms are quantitatively compared in Table 4.1. In addition, we found that vergence velocities during downward convergence (i.e., FU/N paradigm) were significantly slower in all three monkeys (P < 0.05) when compared to the other vertical- vergence paradigms.

4.4.2 Temporal alignment of peak vertical and vergence velocities

Next, we evaluated the temporal alignment of peak vertical and vergence velocities. Fig.

4.4A1 and 4.4A2 illustrate typical examples of eye movements in which vertical and convergence movements were combined (Converge-Down and Converge-Up, respectively). As shown for these example movements, vergence and vertical velocities peaked at approximately the same time. This result was typical for all four behavioral conditions with only small differences in temporal dissociation between monkeys (see Table 4.1). On average, vergence velocities peaked

~3-4 msec after peak vertical velocity. Recently, Kumar et al. (2005) observed that significant 128

Fig. 4.3 Average horizontal peak vergence velocities (º/sec) during converging (top) and diverging (bottom) for symmetric vergence and vertical disconjugate saccades in monkeys M, R and D. Stars indicate average vergence velocities that are significantly greater than saccade- free symmetrical vergence (P < 0.05)

129

130

temporal dissociations (up to 320 msec) could occur when humans made self-paced shifts between far targets and higher near targets (i.e., Converge-Up). A similar dissociation pattern was not observed in the monkeys evaluated in this study. As illustrated in Fig. 4.4B1, during the

Converge-Up condition the temporal dissociation was less than 40 msec in the majority of the trials (> 95%). On the very rare occasion that peak vergence velocity was markedly delayed compared to peak vertical velocity in this condition (closed star in Fig 4.4B1) the peak vergence velocity associated with the overall eye movement was relatively small (~50º/sec). Indeed, the dynamics of the vergence components of these eye movements were similar to those of saccade- free vergence. As illustrated in Fig. 4.4B2 peak vergence velocities were generally greater than

100deg/sec when combined with a vertical saccade (Fig. 3.4B2, black bars). However, when peak vergence velocity was delayed, the resulting vergence velocity was in the range of vergence velocities observed during symmetric vergence (4.4B2, dark gray versus light gray, respectively).

These behavioral findings show that when monkeys make vertical saccades between near and far targets, vergence velocities are significantly facilitated. Furthermore, a robust temporal alignment of peak vertical and vergence velocities was observed suggesting that the synchronization of saccadic and vergence velocities is an important determinant of vergence facilitation.

4.4.3 Test of the hypothesis: vergence is facilitated by the classical saccadic pathway during disconjugate saccades

The second goal of this study was to determine whether the premotor burst neurons, which drive horizontal saccades, discharge in a manner appropriate to facilitate vergence velocities during vertical saccades between near and far targets. To address this question, we

131

Fig. 4.4: (A) Individual examples of typical behavior observed when a vertical saccade is combined with a convergence movement. Closed arrows indicate peak vertical velocity and open arrows indicate peak vergence velocity. (B1) Distribution of temporal dissociations (i.e., time of peak vertical velocity – time of peak vergence velocity) calculated during the Converge-Up paradigm. The majority of the behavior resulted in velocities that have a temporal dissociation of less than 40msec (black histograms) whereas an atypical example, resulted in a dissociation of up to 200msec (grey histogram). (B2) Distributions of vergence velocities recorded during Converge-Up (black and grey bars) and symmetric vergence (light grey). Open (*) and closed stars (*) correspond to vergence velocities and dissociations calculated in the typical Converge-Up example (A2) and an atypical example (data not shown), respectively.

132

characterized the command signal that was dynamically encoded by the horizontal SBNs during vertical saccade facilitated vergence. Notably, these saccades would also require a vertical conjugate command, which would originate from the vertical burst neurons of the riMLF

(Buttner et al. 1977; Crawford and Vilis 1991; 1992; King and Fuchs 1979; Missal et al. 2000;

Moschovakis et al. 1991a; Moschovakis et al. 1991b).

A total of 57 SBNs were recorded in the paramedian pontine reticular formation (PPRF), the majority of which (N=38) were recorded with sufficient behavior during vertical facilitated vergence to determine the neuron‟s ocular sensitivity (see METHODS). Neurons were classified as excitatory (EBNs; N=22) or inhibitory (IBNs; N=35) based on their anatomical location (see

METHODS) and further categorized as short or long lead neurons depending on whether the mean period between the onset of the first spike and the onset of eye velocity was  15 ms, or > 15 ms, respectively during conjugate saccades (25 long lead IBNs; 10 short lead IBNs; 9 long lead

EBNs; 13 short lead EBNs) (Cullen and Guitton 1997; Scudder et al. 1988). As shown in a recent comparison of short and long lead EBNs and IBNs during disconjugate saccades (Van

Horn et al. 2008), we found no major differences (with the obvious exception of the burst lead times) between the two groups of neurons. Thus, for simplicity, short and long lead EBNs and

IBNs are discussed as a pooled population and are referred to as SBNs.

We first tested whether the command signal encoded by SBNs correlates to the increase in vergence velocity that is observed when vertical saccades are made between far and near targets (i.e. Fig. 4.3). Our logic was the following: If SBNs provide a saccadic monocular command to the extraocular motoneurons, their responses should preferentially encode the movement of an individual eye during vertical as well as horizontal disconjugate saccades.

Alternatively, if SBNs provide only a conjugate command to the extraocular motoneurons, then

133

their firing patterns should be unaffected when compared to that generated when vertical saccades are made between two far targets. Overall, our results support the former proposal.

Specifically, the movement of an individual eye was required to accurately describe the burst activity of the majority (84%) of the SBNs when vertical saccades were made between near and far targets. An example SBN illustrating this main finding is presented in Fig. 4.5 and 4.6.

Fig. 4.5 shows the discharge of a typical neuron during four conjugate saccades. Notably, as further illustrated in the polar plot representing the average discharge for many vertical, oblique and horizontal saccades, this neuron burst rigorously for ipsilaterally directed conjugate saccades but was completely silent for contralateral and vertical saccades made between two far targets. In contrast, when a vertical saccade was made simultaneously with a vergence movement to shift gaze between near and far targets, this same neuron was not silent (Fig. 4.6A). This is a striking result, given that the accompanying conjugate component of the horizontal eye movement was close to zero. Indeed, we found that this neuron only fired action potentials when the ipsilateral eye moved in the “on” (i.e., ipsilateral) direction consistent with the proposal that

SBNs preferentially encode the movement of a single eye during saccades. Notably, no neural activity was associated with vergence velocities that were not accompanied by saccades. For example, SBNs were silent during symmetric vergence and during the Converge-Up paradigm when vergence velocities were temporally delayed compared to the vertical saccade.

We next tested whether we could predict the firing rate of the neuron based on its sensitivity to eye movements during horizontal conjugate saccades using the following dynamic model:

 FR(t)  b  resthori E(t  td ) Horizontal-prediction (4.2)

134

Fig. 4.5: Neural responses of a typical SBN during four 10° saccades in the contralateral (contra), up, ipsilateral (ipsi) and downward direction. A polar plot representing average discharge for this neuron during numerous ipsilateral, oblique and vertical saccades is shown in the middle of the four movements. Note that this neuron was completely silent (i.e., no burst) during saccades in the contralateral direction and during vertical saccades.

135

Fig. 4.6: (A) Neuronal responses and model fits for the same neuron shown in Fig. 4.5 when a vertical saccade is combined with a vergence movement (Diverging-Down). The left and right columns illustrate two example movements. The firing rate of the neuron is shown as the gray shaded area (top row, and reproduced in second row for clarity). Predicted model fits using ipsilateral, conjuagte and contralateral eye velocities are shown in the top row in blue, black and red, respectively (VAFipsi=0.60, VAFconj = 0.53, VAFcontra=0.33). Estimated model fits using the binocular model and reduced ipsilateral model are shown in the second row. Dotted vertical lines represent vertical saccade onsets and offsets of 20°/sec. (B) Bootstrap histograms and 95% confidence intervals (thick horizontal bars) for this neuron. Note the 95% confidence interval for the contralateral eye (red bar) overlaps with zero. 136

where FR(t) is the predicted firing rate of the neuron, b is the bias, rest-hori is the neuron‟s horizontal eye velocity sensitivity estimated during horizontal conjugate saccades respectively, td

  is the dynamic lead time, and E(t) represents either 1) conjugate velocity [ CJ ], 2) ipsilateral eye

  velocity [ IE ] or 3) contralateral eye velocity [ CE ] (see also Van Horn et al. 2008, for a comparable approach in the analysis of horizontal disconjugate saccades). The distribution of horizontal velocity sensitivities for the population of SBNs are shown in Fig. 4.7A and average parameters estimated during conjugate saccades are provided in Table 4.2. For the example neuron illustrated in Fig. 4.6, we found that neuronal discharge dynamics were most accurately predicted when we used the movement of the ipsilateral eye to estimate firing rate (Fig. 4.6, blue trace superimposed on firing rate). When conjugate or contralateral eye velocity was used in the estimation, we obtained much poorer predictions of neuronal firing (black and red traces superimposed on the firing rate, respectively).

Consistent with previous studies (Cullen and Guitton 1997; Hepp and Henn 1983;

Kaneko 2006; Scudder et al. 1988; Strassman et al. 1986a; b), approximately half of the SBNs in our population were not completely silent during vertical saccades made between two far targets but instead produced a small discharge of action potentials. For these neurons (N= 22/38) we also characterized the firing rates during vertical saccades using a vertical-eye velocity based version of eq. (4.1) and found that they were dynamically related to vertical eye velocity (mean

VAF = 0.39±0.15 and 0.28±0.25 obtained for vertical up and down saccades, respectively; see supplementary Table 1 for average parameters estimated during vertical saccades). Furthermore, we could well predict the firing rate of a given neuron in this subpopulation during oblique saccades by simply accounting for its sensitivity to horizontal and vertical velocity during

137

Fig. 4.7: (A) Distribution of horizontal sensitivities of SBNs estimated during conjugate saccades. (B) A neuron-by-neuron comparison of horizontal and vertical sensitivities estimated during ipsilateral and vertical saccades, respectively. Neurons which were silent (i.e., no neuronal discharge) during vertical saccades are represented by filled circles and neurons which burst during vertical saccades are represented by grey squares.

138

horizontal and vertical saccades, respectively (mean VAF = 0.51±0.17). Note this quality of fit is comparable to that obtained during conjugate saccades (see supplemental Table 1). The tuning and discharge of an example neuron during oblique saccades are shown in supplementary Figs.

4.1 and 4.2, respectively. Thus, in order to fully describe the discharge dynamics of these neurons when vertical saccades were made between near and far targets (i.e. vertical saccades combined with vergence) we used a model that also accounted for each neuron‟s sensitivity to horizontal as well as vertical eye velocity during saccades (supplementary Fig. 4.3 shows this approach applied to the same example neuron that was shown in supplementary Fig. 4.1 and

4.2).

A neuron-by-neuron comparison of horizontal and vertical sensitivities estimated during ipsilateral versus vertical saccades is shown in Fig. 4.7B. Because we found no major differences

(other than obvious exception of their tuning to saccade direction) between neurons with or without vertical sensitivities, they are discussed as a pooled population below. Overall, we found that the discharges of the majority of the neurons in our population (27/38) were best predicted by the horizontal velocity of an individual eye, rather than the conjugate velocity, during vertical facilitated vergence (~25% average relative improvement in VAF). Table 4.2 summarizes the average VAFs for conjugate versus individual eye predictions for the population of SBNs.

4.4.4 Estimation of the vergence-related signal encoded by horizontal saccadic burst neurons during vertical disconjugate saccades

We next investigated whether estimating a more complex model, specifically a binocular expansion of the conjugate model, might provide an improved description of neuronal discharges during vertical saccades that are made between near and far targets using the formulation: 139

140

  FR(t)  b  rIE IE(t  td )  rCE CE(t  td ) Binocular- Estimation (4.3) where FR(t) is the estimated firing rate, and rIE and rCE are constants that represent the neuron‟s

  ipsilateral and contralateral velocity sensitivity, respectively, and IE(t) and CE(t) refer to the instantaneous velocity of the ipsilateral and contralateral eye, respectively. Note, that for neurons which were not completely silent during vertical saccades made between two far targets, an addition term was included in [eq. (4.2)] in order to account for their sensitivity to vertical saccadic eye motion. Model fits using this expanded binocular model [eq. (4.2)] for the example neuron are shown in the second row of Fig. 4.6 (black trace superimposed on firing rate). To

determine if both eye velocity parameters in the binocular model (i.e., rIE  rCE ) were necessary to describe the firing rate of this neuron, we estimated the 95% confidence intervals using a bootstrapping technique described in METHODS and in previous studies (Sylvestre et al. 2003;

Sylvestre and Cullen 2002; Van Horn et al. 2008). Fig. 4.6B shows the values of the original parameter estimates (vertical arrows) as well as the distributions of the estimates obtained using the bootstrapping approach (histograms). For each of the two parameter estimates, the 95% confidence interval is denoted by the heavy horizontal line below the distribution. Two important observations can be made from this figure: 1) the confidence intervals of the two eyes were statistically different (i.e., they do not overlap with each other and could therefore not be replaced with a single term proportional to conjugate eye velocity) and 2) the confidence interval for the sensitivity of the contralateral eye velocity was not statistically different from zero (i.e., the confidence interval overlaps with zero).

Accordingly, we removed the contralateral eye velocity term from the estimation model such that only ipsilateral eye velocity was used to estimate the firing rate using the formulation:

141

 FR(t)  b  rIE IE(t  td ) Ipsilateral-Estimation (4.4)

As expected, the resulting fit was nearly identical to that of the full binocular model (Fig 3.6. second row, dotted blue trace superimposed on firing rate, ∆BIC=0) confirming that the contralateral eye velocity parameter played no significant role in describing the discharge of this neuron.

Similar results were obtained for the majority of neurons in our population of SBNs.

Overall, the 95% confidence intervals for most neurons (N=34/38) did not overlap with each other confirming that individual eye velocity terms should not be replaced with a single conjugate term. Moreover, in ~2/3 (N=24/38) of the neurons the 95% confidence interval for one eye velocity parameter overlapped with zero and the fits resulting from the full binocular model

[eq. (3.3)] was nearly identical to the fits using the preferred individual eye [eq. (4.4)] (i.e., the eye for which the eye velocity parameter did not overlap with zero, mean population VAFbinocular

= 0.47±0.07 vs. VAFreduced = 0.44±0.07). Thus, these results suggest that horizontal SBNs facilitate vergence during vertical saccades between near and far targets by preferentially encoding the horizontal movement of an individual eye.

4.4.5 Ocular sensitivities across of the population of SBNs

For each SBN, a Ratiodyn index was computed based on the parameters estimated for the binocular model to objectively assign each neuron to one of five ocular categories (see METHODS for details, as well as Van Horn et al. 2008). The distributions obtained for Ratiodyn during vertical saccades between near and far targets for all SBNs (N=38) are shown in Fig. 4.8. We found that the majority of the SBNs could preferentially encode the movement of either eye (e.g., monocular ipsilateral N = 15; monocular contralateral N = 9). Table 4.2 summarizes the average 142

Fig. 4.8: Distribution of Ratiodyn indexes for SBNs. For each neuron, a Ratio index was calculated using: [smaller parameter value] / [larger parameter value], where the smaller and larger parameter values are yielded by the non-preferred and preferred eyes, respectively. The majority of the SBNs encoded the velocity of the ipsilateral eye.

143

VAFs and changes in BIC provided by the complete binocular versus reduced models for each of the five categories during vertical facilitated vergence. This distribution is comparable to that previously described for a separate population of SBNs during horizontal disconjugate saccades

(Van Horn et al. 2008).

4.4.6 Comparison of ocular preference during horizontal and vertical disconjugate saccades

To more directly relate our present results with those previously described (Van Horn et al. 2008) a subset of SBNs (N=16) were also recorded during horizontal disconjugate saccades.

Approximately an equal number of SBNs with and without vertical sensitivities (N = 9 and 7, respectively) were tested. A comparable analysis to that described above was then used to determine if, on a neuron-by-neuron basis, SBNs had similar ocular preferences when vergence was facilitated by either horizontal or vertical saccades. Fig. 4.9 illustrates the neuronal discharge of the same example neuron shown in Fig. 4.5 and 4.6 during four example horizontal disconjugate saccades (converging and diverging are shown in panels A1 and A2, respectively).

Note the large differences in dynamics for the two eyes during these movements: in the converging case (panel A1) the contralateral eye moved while the ipsilateral eye was relatively stationary, whereas in the diverging case (panel A2) the ipsilateral eye moved while the contralateral eye was relatively stationary. The conjugate component of the movements was comparable in the two conditions. Consistent with the results found during vertical facilitated vergence, we found that the neuron‟s activity preferentially encoded the velocity of the ipsilateral eye (Fig. 4.9, blue trace superimposed on firing rate). In particular, an ipsilateral-based prediction most accurately predicted the firing rate (blue trace superimposed on firing rate) 144

Fig. 4.9: Neuronal responses and model fits for the same neuron shown in Fig. 3.5 and 3.6 when a horizontal saccade is combined with a vergence movement (A1: converging disconjugate saccade when the contralateral eye moves more A2: diverging disconjugate saccades when ipsilateral eye moves more). Predicted model fits using ipsilateral, conjugate and contralateral eye velocities are shown in the top row in blue, black and red, respectively (VAFipsi=0.46, VAFconj=0.39, VAFcontra=0.15) . Estimated model fits using the binocular model and reduced ipsilateral model are shown in the second row. (B) Bootstrap histograms and 95% confidence intervals (thick horizontal bars) for this neuron. Note the 95% confidence interval for the contralateral eye (red bar) overlaps with zero.

145

whereas conjugate and contralateral-based predictions tended to overshoot the firing rate when the ipsilateral eye moved less (i.e., during the diverging movements for this example neuron, panel A1) and to undershoot when the ipsilateral eye moved more (panel A2; black and red traces superimposed on the firing rate). Confidence intervals obtained using the bootstrapping technique, further confirmed that the neuron‟s sensitivity to contralateral eye velocity was not statistically different from zero (Fig. 4.9B). Furthermore, when we removed the contralateral eye velocity term from our dynamic model such that only ipsilateral eye velocity was used to estimate the firing rate, the resulting fit was nearly identical to that of the full binocular model

(Fig. 4.9, second row, dotted blue trace superimposed on firing rate, ∆BIC=0).

To directly address whether the command provided by a given SBN similarly facilitates shifts of vergence during both horizontal and vertical saccades, we calculated the Ratiodyn during horizontal saccades between near and far targets for this subset of neurons (N=16) that were fully characterized during horizontal disconjugate saccades (e.g., Fig. 4.9). The ocular preference of a given neuron was typically the same when vergence was facilitated during either vertical or horizontal saccades. This result is shown in Fig. 4.10 where the preferred eye during vertical facilitated vergence is plotted against the preferred eye during horizontal facilitated vergence.

For the majority of the neurons in our population (~75%), ocular sensitivities were identical during horizontal and vertical disconjugate saccades (black, red and blue columns). Taken together, our analysis of vertical and horizontal disconjugate saccades suggest that the premotor burst neurons of the brainstem saccade burst generator functions to facilitate vergence when saccades are made between near and far targets by preferentially encoding the horizontal movement of a specific eye. The implications of this finding are further discussed in the

DISCUSSION.

146

Fig. 4.10: Coherence of the preferred eyes for vertical and horizontal facilitated vergence. The x and y axes plot the 3 preferred eye categories (Ipsi, ipsilateral; Contra, contralateral; and Conj, conjugate) and the z axis plots the fraction of neurons. Solid blue, red and black columns indicate coherence between the two behaviors and gray bars indicate absence of coherence.

147

4.5 DISCUSSION

The saccadic burst neurons of the paramedian pontine reticular formation project to the extraocular motoneurons and provide the primary command that drives horizontal saccades. The results of recent studies suggest that these neurons do not encode conjugate commands during saccades but instead preferentially encode the movement of an individual eye (King and Zhou

2000; McConville et al. 1994; Sylvestre et al. 2003; Zhou and King 1996; 1998), thereby mediating the facilitation of vergence as well as horizontal conjugate velocity during horizontal disconjugate saccades (Van Horn et al. 2008). Here we investigated whether the monocular commands coded by saccadic burst neurons are suitable for facilitating vergence when vertical saccades are made between near and far targets. Notably, such saccades require the generation of commands to make vertical conjugate saccade, which would originate from vertical saccadic burst neurons in the riMLF (Buttner et al. 1977; Crawford and Vilis 1991; 1992; King and Fuchs

1979; Missal et al. 2000; Moschovakis et al. 1991a; Moschovakis et al. 1991b) and vergence (but not horizontal conjugate) eye movements. Thus, we could more directly test the hypothesis that the vergence information encoded by the classical horizontal saccadic pathway is consistent with the vergence facilitation that occurs during saccades.

We first established that vergence velocities are significantly facilitated when vertical saccades are made between targets located at different distances. Next, we tested whether horizontal SBNs encode monocular commands that are appropriate to account for the facilitation of vergence during these vertical saccades, even though the commanded saccade had no significant horizontal conjugate component. We find that the majority of the SBNs preferentially encode the movement of a specific eye during both vertical and horizontal saccades that are

148

made in depth. Thus, our results are consistent with the proposal that the monocular command generated by the brainstem saccadic burst generator is for facilitating all shifts of vergence.

4.5.1 Vergence velocity is facilitated during vertical disconjugate saccades

In everyday life, we typically combine saccades and vergence to look between near and distant objects in three–dimensional space. During these orienting eye movements, there is general agreement that vergence velocity is facilitated (i.e., vergence velocities are greater than expected relative to saccade free vergence of comparable amplitudes), while conjugate saccade velocities are slowed. While most previous studies have focused on the facilitation that occurs when vergence movements are made in combination with horizontal saccades (Busettini and

Mays 2005a; Collewijn et al. 1997; Enright 1984; Enright 1992; Maxwell and King 1992; Ono et al. 1978; Oohira 1993; Van Horn et al. 2008; van Leeuwen et al. 1998; Zee et al. 1992), there had been some evidence for the facilitation vergence velocity during vertical saccades (Busettini and Mays 2005a; Enright 1984; Kumar et al. 2005a; Maxwell and King 1992; van Leeuwen et al.

1998). However, because these studies characterized vertical saccades with non-negligible horizontal components (Busettini and Mays 2005a), or did not methodically examine changes in horizontal conjugate position (Enright 1984; Ono et al. 1978; van Leeuwen et al. 1998), the possibility that the observed facilitation might be, at least in part, due to horizontal saccadic interactions, had been left open.

In the present study we limited our analysis to saccades for which vertical component of the movement was ≥90% of total movement amplitude and found that during vertical saccades vergence velocities reached values as large as 300 deg/sec as compared to saccade-free movements where velocities were generally less than 100 deg/sec. Our findings were consistent

149

with those of prior reports; although in our study downward convergence was slower than upward convergence while van Leeuwen et al. (1998) found the opposite tendency in humans.

4.5.2 Vergence and vertical velocities are temporally aligned during vertical saccades in monkeys

Horizontal gaze shifts between targets located at different depths are characterized by the synchronized occurrence of saccadic and vergence movements [(Busettini and Mays 2005a;

Enright 1984; Kumar et al. 2005a; Maxwell and King 1992; van Leeuwen et al. 1998; Zee et al.

1992)]. This finding has been used as evidence for the proposal that the neural circuitries commanding horizontal saccades and vergence interact (Collewijn et al. 1997). In the present study we further show that when vertical saccades are made between far and near targets, the peak vergence and vertical velocities are generally temporally aligned in monkey. Specifically, peak vergence velocity peaked roughly 3-4ms after peak vertical velocity. This finding was consistent regardless of whether gaze was shifted to or from a lower or higher more distant target.

Similar temporal alignments have also been recently reported for humans during comparable tasks (Kumar et al. 2005a; van Leeuwen et al. 1998; Zee et al. 1992). For example, most recently, Kumar et al. (2005) reported dissociation intervals that were on the order of ~4-20 ms for the majority of the trials. Notably, however Kumar and colleagues (2005) did observe that in trials where human subjects shifted their gaze from a lower distant target to near higher target

(i.e., Converge-Up), the occurrence of peak convergence velocity could be considerably delayed relative to peak vertical velocity (delays generally > 40 and as large as 320 ms). In the present study, we rarely (e.g., < 5%) observed such delays. Moreover, the few trials that had larger

150

dissociation intervals were characterized by vergence movements that began after the onset of the vertical saccade and had velocity profiles resembling those of saccade-free vergence velocity

(~50º/sec). Thus, the results indicate that the synchronized occurrence of a saccade is important for the facilitation of vergence velocities in monkey.

4.5.3 The dynamics of SBNs during horizontal and vertical conjugate saccades

Previous studies, which have used system identification techniques to characterize SBN discharge during horizontal conjugate and disconjugate saccades, have demonstrated that SBNs encode saccade trajectories in their spike trains. In particular, a clear relationship between EBN and IBN firing rates and eye movement dynamics has been described (Cullen and Guitton 1997;

Sylvestre and Cullen 2006; Van Horn et al. 2008). Here, we demonstrate that this approach can be extended to describe the responses of SBNs during vertical and oblique saccades.

While all neurons recorded in this study discharged primarily for ipsilaterally directed saccades the directional tuning differed between neurons. In particular, approximately half of the neurons in our sample were broadly tuned and were not completely silent during vertical saccades. This result is in agreement with many previous studies that have also reported SBNs with broad tuning curves (Cullen and Guitton 1997; Hepp and Henn 1983; Kaneko 2006;

Scudder et al. 1988; Strassman et al. 1986a; b). Although a metric-based analysis preformed by

Scudder et al. (1988) did report that the number of spikes generally increased for larger vertical movements this is the first study to describe the dynamic relationship between SBN discharge and eye velocity during vertical and/or oblique saccades. We found that for SBNs, that were not completely silent during vertical saccades, the firing rates during vertical saccades were dynamically related to vertical eye velocity (see Supplemental Table 1). Furthermore, the

151

discharge characteristics during horizontal and vertical saccades could be used to accurately predict the firing rate of all SBNs during oblique saccades that had varying vertical and horizontal components.

4.5.4 SBNs contribute to increasing vergence velocities during disconjugate saccades

We have previously shown that the saccadic burst generator in the PPRF, which was commonly thought to encode horizontal conjugate saccades (Busettini and Mays 2005b), in fact provides temporally appropriate vergence information to facilitate vergence during horizontal disconjugate saccades (Van Horn et al. 2008). In particular, the information was encoded in terms of an individual eye, where the majority of the neurons encoded the movement of the ipsilateral eye. In the present study, we tested the prediction that if the monocular commands issued by the saccadic burst neurons (SBNs) are important for facilitating vergence during horizontal saccades they should also contribute to facilitating vergence associated with a vertical saccade when the conjugate component of the movement is negligible. As predicted, we found that SBNs are also well suited for facilitating vergence during a vertical saccadic eye movement.

In particular, SBNs contribute to generating increased vergence velocities by dynamically encoding the movement of an individual eye rather than the conjugate component of the movement.

To directly compare our present results with those previously described (Van Horn et al.

2008) we recorded a subset of SBNs during both horizontal and vertical facilitated vergence. On a neuron-by-neuron basis, we found that SBNs have similar ocular preferences during both conditions. For example, a neuron that was found to dynamically encode the movement of the

152

ipsilateral eye during horizontal disconjugate saccades was also found to encode the movement of the ipsilateral eye during vertical facilitated vergence. Taken together with previous findings

(Van Horn et al. 2008), this implies that the command provided by a given SBN is appropriate for facilitating shifts of vergence during disconjugate saccades by encoding integrated conjugate and vergence commands.

Interestingly, a number of neurons in this study were found to encode the movement of the “wrong” eye (e.g., the contralateral eye) or a combination of both eyes (e.g., binocular cells).

This finding is consistent with the results of previous studies that have evaluated the responses of neurons in other premotor and motor nuclei. For instance, individual neurons in both the nucleus prepositus and abducens nucleus can preferentially encode the movement of either the contralateral or ipsilateral eye (McConville et al. 1994; Sylvestre et al. 2003; Sylvestre and

Cullen 2002; Van Horn et al. 2008; Zhou and King 1998). While at first glance this observation might appear surprising, it can be easily reconciled with the existing circuitry. Firstly, the abducens nucleus is comprised of internuclear as well as motoneurons and neurons. Thus, a premotor neuron (such as an EBN) that preferentially encodes information about the contralateral eye may in fact project to the appropriate eye through internuclear neurons. Secondly, while neurons are generally assumed to have equal synaptic weights, unequal weighting of the projections must certainly exist. For example, premotor neurons that encode the movement of the ipsilateral eye may provide stronger synaptic inputs to the motoneurons. Finally, inappropriate signals that are sent to abducens nuclei could be both cancelled out by additional premotor inputs such that the final drive to the lateral rectus muscle is correct and/or offset by a co-contraction of the antagonist muscle (i.e., medial rectus; see DISCUSSION of (Sylvestre and Cullen 2002).

153

4.5.5 Premotor circuits for the control of changes in vergence angle

While SBNs have been shown to carry vergence-related (e.g., monocular) information during disconjugate saccades two important questions remain when considering the neural control of gaze in three dimensional space: 1) what is the source of the vergence-related information to the horizontal SBNs, and 2) what additional vergence commands are required to drive non-saccadic vergence movements (i.e., when the saccadic burst generator is silent)?

First, although it is well recognized that information about an individual eye is available throughout most of the visual and visual-motor cortex the source of monocular information to the

SBNs remains unknown (Gnadt and Beyer 1998; Hubel and Wiesel 1962; 1970). Two possible candidates are the mesencephalic reticular formation (MRF) and superior colliculus (SC)

(Ferraina et al. 2000; Genovesio and Ferraina 2004; Gnadt and Beyer 1998; Gnadt and Mays

1995; Mimeault et al. 2004). Both of these structures receive inputs from disparity sensitive cortical and subcortical regions and stimulation of both the MRF [goldfish: (Luque et al. 2006); monkey: (Waitzman et al. 2008)] and SC [monkey: (Chaturvedi and Van Gisbergen 1999; 2000;

Suzuki et al. 2004) cat: (Suzuki et al. 2004)] have clear effects on vergence. Moreover, neurons in the SC [cat: (Jiang et al. 1996), monkey: (Walton and Mays 2003)] and the MRF (Gamlin et al. 1994; Judge and Cumming 1986; Mays et al. 1986; Waitzman et al. 2008) are modulated during vergence eye movements. Notably, a recent report has shown that saccade-related burst neurons in the central MRF (cMRF) dynamically encode the movement of an individual eye during disjunctive saccades (Waitzman et al. 2008). These findings further support the proposal that inputs from the MRF and SC to the saccadic premotor neurons function in parallel with the cortico-pontine-cerbellar-midbrain loop which has traditionally been viewed as the main pathway for the control of vergence (reviewed in Gamlin 1999). Although the traditional view

154

describes the vergence and saccadic pathways as two distinct neural systems, our results, taken together with those of other recent studies, support the proposal that changes in vergence angle are controlled by means of a more distributed network.

Second, while we have clearly demonstrated that SBNs carry important vergence-related information to control saccades in three-dimensional space, SBNs are silent during slow vergence movements. For example, we have previously shown that the saccadic burst generator does not fire any action potentials during symmetric vergence as well as during periods of slow vergence that precede or follow disconjugate saccades (Van Horn et al. 2008). Thus, this suggests that while the SBNs function to rapidly drive the eyes to a new position, an additional vergence command is required to ensure accurate binocular realignment of gaze (King and Zhou

2000). Notably, neurons encoding slow vergence, which have been described near the abducens nucleus (Gnadt et al. 1988) and oculomotor nucleus (Judge and Cumming 1986; Mays 1984;

Zhang et al. 1991; Zhang et al. 1992), are likely candidates to drive such eye movements. Further work will be required to understand how neural circuits in the brain stem and cerebellum interact with descending control pathways to ensure the accurate control of gaze in three-dimensions.

4.6 SUPPLEMENTARY MATERIAL

3.6.1 SBNs with broad tuning also facilitate vergence during vertical saccades made between near and far targets

Some SBNs (N=22, neurons represented by grey squares in Fig. S4.1) produced a small burst for vertical saccades made between two far targets (i.e. vertical saccades with no vergence component). In order to fully describe the discharge dynamics of these neurons during saccades 155

we used a strategy similar to that used to analyze neurons with no vertical sensitivity. However, in order to describe the responses of this subpopulation of neurons we expected that we would need a model that also accounted for instantaneous vertical eye velocity

  FR(t)  b  r E(t  t )  r VT (t  t ) (4.5) esthori d estvert d where FR(t) is the predicted firing rate of the neuron, rest-hori and rest-vert are constants that represent the neuron‟s horizontal and vertical eye velocity sensitivities estimated during

  horizontal and vertical conjugate saccades, respectively, and E(t)and VT (t)refer to the instantaneous horizontal and vertical eye velocity, respectively (See Supplementary Table S4.1 for average parameters estimated during vertical saccades).

We first accessed whether we could identify a general description of SBNs that could represent discharge dynamics when either horizontal, vertical or oblique saccades were made between two far targets (i.e., when vergence is zero). To do this we tested the hypothesis that the discharge dynamics of SBNs during oblique saccades could be predicted based on their sensitivities to horizontal and vertical saccades using the formulation:

(4.6) where FR(t) is the predicted firing rate of the neuron, rest-hori and rest-vert are constants that represent the neuron‟s horizontal and vertical eye velocity sensitivities estimated during horizontal and vertical conjugate saccades, respectively, and and refer to the instantaneous horizontal and vertical eye velocity, respectively.

An example neuron that discharged during vertical saccades is illustrated in

Supplemental Fig. S4.1-4.3. Supplemental Fig. S4.1 demonstrates that while the neuron discharges most vigorously during ipsilaterally directed saccades, the neuron also produced a

156

small burst of activity during vertical saccades. As illustrated in Supplemental Fig. S4.2, during oblique saccades we found that the discharge dynamics could be well predicted using the estimated horizontal and vertical parameters (black trace superimposed on the firing rate,

VAFpred = 0.6, mean population VAFpred = 0.50±0.18).

Given that we could predict the firing rate of these neuron during oblique saccades (i.e., when a vertical saccade was combined with a horizontal saccade with no vergence requirement) we next asked whether, during the vertical saccades made between near and far targets (i.e., vertical saccades in which vergence was facilitated), we could predict neuronal firing rates using the same approach. Accordingly, we used a strategy similar to that used above to analyze neurons with no vertical sensitivity (see Fig. 4.6). However, in order to describe the responses of this subpopulation of neurons we expected that we would need a model which also accounted for instantaneous vertical eye velocity. Supplemental Fig. S4.3 illustrates the results obtained for the same example neuron shown in Supplemental Figs. S41-4S.2. Consistent with our expectation, we found the dynamic discharge could be most accurately predicted with a model that included terms representing sensitivity of the neuron to horizontal and vertical saccades. Notably, neuronal sensitivity to horizontal eye movements during saccades predominately reflected the movement of the ipsilateral eye (Supplemental Fig. S3A, blue trace superimposed on firing rate).

In contrast, predictions based on either instantaneous conjugate or contralateral eye velocity provided substantially poorer predictions (compare black and red traces superimposed on the firing rate, respectively). Overall, the neuronal discharges of the majority (68%) of the SBNs in our population, which fired during vertical saccades between distant targets, were better predicted by a model based on the velocity of an individual eye than on the conjugate component of the saccade (P < 0.05).

157

To more directly quantify each neuron‟s sensitivity during vertical facilitated vergence we next described its firing rate using a modified form of the expanded binocular model used in the main manuscript [eq. (3)]. Notably, since these neurons were sensitive to vertical velocities we also included a vertical velocity term to obtain the following formulation:

   FR(t)  b  r IE(t  t )  r CE(t  t )  r VT (t  t ) (4.7) IE d CE d VT d where FR(t) is the estimated firing rate, rIE, rCE and rVT are constants that represent the neuron‟s

  ipsilateral, contralateral and vertical eye velocity sensitivity respectively and IE(t) ,CE(t) and

 VT (t) refer to the instantaneous velocity of the ipsilateral, contralateral and vertical eye, respectively.

Model fits using this expanded binocular model [eq. (4.7)] are shown in the second row of Fig. S4.3 (black trace superimposed on firing rate in second row) for the example neuron. Fig.

S4.3B shows the corresponding parameter estimations (vertical line) and distributions obtained using the bootstrapping approach (histograms) as well as the 95% confidence intervals (heavy horizontal lines beneath each distribution). Similar to the example neuron with no vertical sensitivity (e.g., Fig. 4.6), we found 1) the confidence intervals of the two eyes were statistically different (i.e., they did not overlap with each other and could therefore not be replaced with one conjugate term, note) and 2) the confidence interval for the contralateral eye velocity was not statistically different from zero (i.e., the confidence interval overlaps with zero). Furthermore, the 95% confidence interval for the vertical eye velocity sensitivity did not overlap with zero confirming it was a significant parameter (data not shown). Accordingly, we completed our analysis by removing the contralateral eye velocity term from the expanded binocular model [eq.

(4.7)]. We found that a model structure based on ipsilateral and vertical eye velocity alone

158

provided a fit that was nearly identical to that of the full binocular model (Supplemental Fig. S4.

4, second row, blue dotted trace superimposed on firing rate, ∆BIC=0), confirming that the contralateral eye velocity parameter played no significant role in describing the discharge of this neuron.

Similar results were obtained for the majority of the subpopulation of neurons which produced a small burst for vertical saccades between two far targets. Notably, for more than 2/3 of the neurons, the 95% confidence intervals for the individual eye parameters did not overlap with each other, confirming that a single conjugate term could not be used to replace terms representing sensitivities to the velocity of an individual eye. Moreover, for 1/2 of the neurons the 95% confidence interval for one eye velocity parameter overlapped with zero and the fits resulting from the full binocular model [eq. (4.7)] was similar to the fits using the preferred individual eye (i.e., the eye for which the eye velocity parameter did not overlap with zero, mean population VAFbinocular = 0.48±0.19 vs. VAFreduced = 0.47±0.19). Thus, these results support that proposal that the majority of neurons (>75%), which are not silent during vertical saccades between far targets, do not encode the conjugate component of a saccade but rather preferentially encode the movement of an individual eye.

159

Fig.S4.1: Neural responses of an EBN during four 10° saccades in the left, up, right and downward direction. A polar plot representing average discharge for this neuron during numerous ipsilateral, oblique and vertical saccades is shown in the middle of the four movements. This neuron was not completely off (i.e., burst) during the contralateral direction or during pure vertical saccades.

160

Fig. S4.2: Neuronal responses and model fits for the same neuron shown in supplemental Fig. S3.1 during oblique saccades. Predictions using horizontal and vertical sensitivities estimated during pure conjugate horizontal and vertical saccades are shown superimposed on firing rate (thick black, VAF=0.6) and estimations using vertical and horizontal eye velocities (dashed grey traces, VAF=0.7).

161

Fig. S4.3: (A) Neuronal responses and model fits for the same neuron shown in Fig. S4.1 and S4.2 during combined vertical-vergence movements (Diverging-Down). Predicted model fits using vertical sensitivities as well as either conjugate, ipsilateral or contralateral eye velocities are shown in the top row in black, blue and red, respectively (VAFipsi=0.70, VAFconj=0.64, VAFcontra=0.6). Estimated model fits using the binocular model and reduced ipsilateral model are shown in the second row. Ipsilateral and contralateral eye velocities as well as conjugate, vergence and vertical velocities are also shown (bottow row). Dotted vertical lines represent vertical saccade onsets and offsets of 20°/sec. (B) Bootstrap histograms and 95% confidence intervals (thick horizontal bars) for this neuron.

162

163

Chapter 5

Identification and dynamic characterization of vergence neurons in the rostral superior colliculus

This chapter describes a study that has provided exciting results regarding the role of the superior colliculus (SC) in generating vergence eye movements. The goal of this study was to establish whether neurons in the rostral SC contribute to the development of neural signals that are suitable for controlling vergence eye movement or whether their discharge is strictly for conjugate control. The results provide evidence that SC neurons encode vergence information that is necessary for shaping the discharge dynamics of upstream neurons and suggests that there exist distinct groupings of neurons that encode slow and fast vergence within the rostral SC. This chapter is in preparation for submission.

164

5.1 ABSTRACT

The rostral SC encodes small position errors for multiple eye movements including microsaccades, small saccades, smooth pursuit and fixation. Here we address whether the rostral

SC contributes to the development of neural signals that are suitable for controlling vergence eye movements. We use both single unit recording and microstimulation techniques to answer this question. We found that both converging and diverging eye movements can be evoked using microstimulation in the rostral SC. Moreover, amongst the classically described neurons in rostral SC, we recorded a novel population of neurons that either increased (i.e., convergence neurons) or decreased (i.e., divergence neurons) their activity during convergence eye movements. In particular, these neurons dynamically encoded changes in vergence angle during vergence tracking, fixation in 3-dimensional space and binocular realignment that occurs after

3D saccades, but were completely unresponsive during conjugate or disconjugate saccades and conjugate smooth pursuit. Taken together the results suggest that there exist distinct groupings of neurons that encode slow versus fast vergence within the SC and indicate that activation of the rostral SC underlies the ability to accurately position each of the two eyes when fixating targets in 3-dimensional space to ensure stereopsis.

165

5.2 INTRODUCTION

The superior colliculus (SC) plays an essential role in the sensory-motor mechanisms that underlie eye movements. When gaze is shifted to an object of interest, neural activity across the

SC encodes the distance between the current eye position and the final goal. At the rostral tip of the SC, neuronal activity represents small position errors for multiple eye movements including microsaccades, small saccades, smooth pursuit and fixation (Bergeron and Guitton 2000; Choi and Guitton 2006; Hafed et al. 2009; Hafed and Krauzlis 2008; Krauzlis et al. 1997). These previous studies have focused on the commands produced by the SC during conjugate eye movements, yet during normal viewing conditions we typically look between objects located at different depths and eccentricities. Accordingly, despite the fact that the SC is considered one of the most thoroughly studied structures in the primate brain, an examination of motor responses of SC neurons during 3-dimensional viewing is virtually non-existent.

There is mounting evidence to suggest that the SC could play a critical role in the neural control of vergence eye movements. This area receives inputs from disparity-sensitive cortical

[frontal eye field: (Ferraina et al. 2000); lateral intraparietal area (LIP): (Genovesio and Ferraina

2004; Gnadt and Beyer 1998; Gnadt and Mays 1995)] and subcortical regions [e.g., superficial layers of the SC (Mimeault et al. 2004))] and provides strong descending projections to premotor neurons, which encode vergence-related information (Ohtsuka and Nagasaka 1999; Scudder et al. 1996). Microstimulation of the rostral SC in the cat has been shown to evoke convergence eye movements (Suzuki et al. 2004) and bilateral lesions of the rostral SC in humans have been reported to result in convergence palsy (Ohtsuka et al. 2002). Although there is extensive evidence that visual information in 3-dimensions converges upon the SC, the only study to have examined the relationship of neurons in the deep layers of the caudal SC of the monkey during 3- 166

dimensional viewing found that the shifts in the movement fields were not appropriate for driving vergence eye movements and concluded that the SC is primarily involved in driving conjugate movements (Walton and Mays 2003). Notably, no studies to date have explored whether the rostral SC is involved in realigning gaze between near and far targets.

Thus, the goal of this study was to establish whether neurons in the rostral SC contribute to the development of the neural command signals required to control eye movement in 3- dimensional space, or whether their discharge strictly controls only the conjugate component of eye motion. To answer this question, we used both single unit recording and microstimulation techniques. For the first time, we describe a distinct group of neurons within the rostral SC which are unresponsive during traditional conjugate eye movement paradigms but are significantly modulated during smooth vergence tracking and 3D fixation as well as the binocular realignment that follows saccades made in three dimensional space. Our results implicate a novel role for the rostral SC and reveal that it is required to binocularly position the eyes to ensure accurate stereopsis.

5.3 METHODS

5.3.1 Surgical procedures

Two rhesus monkeys (Macaca mulatta) were prepared for chronic extracellar recordings using the aseptic surgical procedures described previously (Sylvestre and Cullen 1999). Briefly, a stainless steel post was attached to the animal's skull with stainless steel screws and dental acrylic permitting complete immobilization of the animal's head. Two stainless steel recording chambers were also secured to the implant. In one monkey, the recording chambers where

167

oriented stereotaxically toward the oculomotor nucleus on the right and left side of the brainstem.

In the second monkey, the recording chamber was tilted backward by 37deg. To record binocular eye position an eye coil (3 loops of Teflon coated stainless steel wire, 18-20 mm diam) was implanted in each eye (Judge et al. 1980). All procedures were approved by the McGill

University Animal Care Committee and University of Connecticut Health Center and complied with the guidelines of the Canadian Council on Animal Care and NIH.

5.3.2 Behavioral paradigms

Monkeys were trained to fixate targets for a juice reward. The timing and location of target illumination, data acquisition and on-line data displays were controlled using REX, a

UNIX-based real-time acquisition system (Hayes et al. 1982).

5.3.2.1 Conjugate paradigms.

To elicit conjugate movements a red HeNe laser target was projected onto a cylindrical screen located 55 cm away from the monkey's eyes (isovergent, 3.5 deg convergence).

Ipsilaterally and contralaterally directed conjugate saccades were elicited by stepping the laser target between horizontal positions (5-30 deg), in 5 deg increments, in predictable and unpredictable sequences and vertical saccades were elicited by stepping the laser target between vertical positions (5-30 deg). Oblique saccades were generated by stepping the laser target between a central target to a sequence of targets that had varying vertical and horizontal components within this same range. Fixational intervals were obtained by keeping the target stationary for 1-3sec. A blink paradigm was also used to test if a neuron showed visual sensitivity. Briefly, while the monkey fixated a prolonged visual target, the target was “blinked” 168

off for 100-500ms. The monkey was required to continue fixating throughout the blink period to receive the juice reward. A neuron was considered visual if the neuronal firing rate decreased to less than 5spk/sec during the blink period (Munoz and Wurtz 1993). Conjugate smooth pursuit eye movements were generated by moving the laser target sinusoidally (20deg/sec peak velocity,

0.4Hz ).

5.3.2.1 Vergence paradigms.

Changes in vergence angle were elicited using a horizontal array of 16 red light emitting diodes (LEDs), with intensities comparable to that of the laser target, which was positioned between the screen and the monkey. Symmetric vergence was elicited by sequentially illuminating LEDs located along the midline (convergence angles: 17, 12, 8 and 6 deg).

Fixational intervals were obtained by keeping the target stationary for 1-3sec. Disconjugate saccades were generated by stepping the target from one of the 16 LEDs to an eccentric laser target position on the screen. This approach yielded a rich variety of disconjugate saccades.

Monkeys were also trained during a vergence smooth pursuit task where the laser target was projected onto a board positioned at eye level, between the screen and the monkey. The laser target moved sinusoidally at the same frequency and velocity used during conjugate smooth pursuit (20deg/sec peak velocity, 0.4Hz).

5.3.3 Data acquisition procedures.

During experiments monkeys were seated in primate chair located within the center of a

1-m3, magnetic eye coil system (CNC Engineering). Horizontal and vertical eye position signals 169

were measured using the magnetic search coil technique (Fuchs and Robinson 1966a; Judge et al. 1980). Each eye coil signal was calibrated independently by having the monkey fixate, with one eye masked, a variety of targets at different horizontal and vertical eccentricities and different depths. Position signals were low-pass filtered at 250 Hz (analog 8 pole Bessel filter) and sampled at 1 kHz. Since ocular saccades include very little power above 50Hz (Cullen et al.

1996b; Van Opstal et al. 1985; Zuber et al. 1968) eye position signals were further digitally filtered (with a 51st order finite-impulse-response filter with a Hamming window and a cut-off at

125 Hz), before being differentiated to obtain eye velocity signals (using zero-phase forward and reverse digital filtering to prevent phase distortion).

5.3.3.1 Extracellular single unit recordings

Extracellular single unit activity was recorded using enamel insulated tungsten microelectrodes [2-10 M impedance, Frederick Haer; for details, see (Sylvestre and Cullen

1999)]. The topographic map of the SC was determined by calculating the “preferred” saccade amplitude and direction of neurons encountered in the intermediate layers of the SC. When a neuron was isolated, unit activity, horizontal and vertical positions of the right and left eyes, and target position were recorded on a digital audio tape (DAT). The isolation of each neuron was reassessed offline during playback. A neuron was considered to be adequately isolated only when individual action potential waveforms could be discriminated using a windowing circuit (BAK) during saccades (e.g., see Fig. 1 in Sylvestre and Cullen 1999), and during fixation. Subsequent analysis was performed using custom algorithms (Matlab, The MathWorks).

A number of criteria were used to verify that recordings were indeed made in the intermediate/deep layers of the rostral SC. First, upon entering the rostral SC we recorded visual

170

neurons that were tonically active during the fixation of a central visual target but whose neural activity decreased when the target was momentarily blinked off (see Supplemental Fig. 5.S1).

Second, at the same site that we recorded our novel population of vergence neurons, we also recorded neurons that have the same characteristics as rostral SC neurons that have been previously described. Notably, these neurons burst for very small saccades and increased activity during contralateral smooth pursuit (see Supplemental Fig. 5.S2).

5.3.3.2 Microstimulation

In both monkeys, site specific eye movements were elicited in the left SC using microstimulation. Saccades were elicited using stimulation parameters, which have been previously used to reliably evoke saccadic eye movements (15-50µA, 333Hz) (Freedman et al.

1996; Gandhi and Keller 1999; Melis and van Gisbergen 1996; Paul and Gnadt 2006; Stanford et al. 1996). A stimulation threshold was determined at each site by gradually increasing the current strength. The stimulation threshold was defined as the current strength where at least 75% of all stimulations led to a saccadic response while the monkey fixated a far target. To ensure a reliable response during experiments, we used twice the stimulation threshold. For each stimulation, the latency between the onset of stimulation and the onset of the first saccade and first vergence eye movement was calculated. Both saccade and vergence onset were respectively defined as the time at which conjugate and vergence velocity was greater than 15deg/sec.

5.3.4 Data Analysis.

5.3.4.1 Definitions and Conventions.

171

Eye movements are described in terms of conjugate [conjugate = (left eye + right eye)/2] and vergence [vergence = left eye – right eye] coordinates (where the left eye and right eye inputs could be either position or velocity signals), such that positive and negative vergence velocity signals denote convergence and divergence, respectively. In addition, we report the movement of each eye as either ipsilateral or contralateral based on their location relative to the recording site. Note, positive and negative values indicate eye positions that are to the right and left, or up and down of the central position (i.e., straight ahead), respectively.

5.3.4.2 Metric analysis.

A neuron‟s vergence sensitivity was measured as the slope of the relationship between the mean vergence position and the mean firing rate measured during fixation. Periods of fixation were defined as time intervals ≥200ms during which peak conjugate and vergence velocities were <5deg/sec.

5.3.4.3 Dynamic analysis.

The linear optimatization techniques used to quantify the dynamic sensitivity of a neuron to eye movements have been described in detail elsewhere (Cullen et al. 1996a; Sylvestre and

Cullen 1999). A spike density function, computed by convolving a Gaussian function (SD of

10 ms) with the spike train, was used to estimate neuronal firing rate (Cullen et al. 1996a;

Sylvestre and Cullen 1999). The specific model structure used here is reported in RESULTS.

The goodness-of-fit of the data to each model was quantified using the Variance-Accounted-For

(VAF = 1 - [var(mod-fr) / var(fr)], where mod represents the modeled firing rate and fr

172

represents the actual firing rate). The VAF in linear models is equivalent to the square of the correlation coefficient (R2) such that a model with a VAF of 0.64 provides as good a fit to the data as a linear regression analysis that yields a correlation coefficient of 0.80 (Cullen et al.,

1996).

5.3.5 Histology

Histology, done in one monkey, confirmed that penetrations were in the rostral pole of the left SC.

5.3.6 Statistical analysis

Unless specifically stated otherwise, data presented in the results are described as means

±standard deviations. A Student‟s t-test was used to determine whether the average of two measured parameters differed significantly from each other.

5.4 RESULTS

In this study, we used microstimulation and single unit recording techniques to determine the role the rostral SC plays in generating vergence eye movements. We recorded a total of 82 neurons in the rSC of two monkeys and identified a significant proportion of neurons (N=19) that could not be classified using previously described visual, saccade and fixation tasks

(Krauzlis et al. 2000; Munoz and Wurtz 1993). A thorough examination of the response of these unclassified neurons, during vergence eye movements, revealed that the rSC contains a unique

173

population of neurons involved in reorienting gaze between near and far. An analysis of the discharge patterns of surrounding neurons during traditional visual, saccade and fixation tasks, as well as reconstructions of the recording tracks, suggest that these identified vergence-related neurons were dorsal to the parafoveal visual neurons and uniformly distributed amongst rSC neurons as previously described (Hafed et al. 2009; Hafed and Krauzlis 2008; Munoz and Wurtz

1993; Reyes-Puerta et al.).

At the rostral pole of the SC, we found microstimulation elicited very small, conjugate staircase saccades when the monkey fixated a far target (Fig. 5.1B; top). A very striking, and reproducible result was observed when the same microstimulation parameters were used while the monkey fixated a near target. Specifically, if microstimulation occurred when the eyes were initially converged, the resulting series of staircase saccades was also accompanied by a series of diverging eye movements, which was sustained throughout the duration of the stimulation (Fig.

5.1B; bottom).

For each microstimulation site (N=15) that elicited diverging eye movements when the monkey fixated a near target, the latency between the onset of microstimulation and the onset of either the first saccade or vergence eye movement was calculated. Fig. 5.1C plots the distribution of latencies during near (right panel) and far (left panel) viewing. During far and near viewing the average latency between the onset of microstimulation and the onset of the first saccade were not significantly different (22.5±6.1ms and 21.9±6.7ms, respectively; p>0.05 Fig

5.1C, black bars). Moreover, during near viewing the onset of vergence (Fig. 5.1C, right panel, red bars) was not significantly different than the onset of the first saccade (21.9±6.7 versus

20.63±5.3ms). This latter result is important since it suggests that the elicited vergence

174

Fig. 5.1: Example of vergence eye movements evoked by electrical stimulation of the rostral SC. (A) Schematic representation of the topographic map of the SC. * denotes the location an example site of stimulation in the rostral SC that evoked vergence eye movements. (B) Two examples of a series of staircase saccades evoked by stimulation (dark bar) during fixation of a far (above) and near (below) target. Stimulation during fixation of a far target evoked very small conjugate staircase saccades with no change in vergence angle. Conversely, stimulation during fixation of a near target (convergence ~10deg) evoked small staircase saccades with a significant diverging eye movement. (C) Distribution of latencies calculated from the onset of stimulation to the onset of either a conjugate saccade (black bars) or the vergence movement 175 (red bars).

movement was evoked by the microstimulation and was not just a result of the eyes naturally relaxing to an unconverged state.

We next recorded the neural activity of neurons at the same site where microstimulation elicited diverging eye movements. Fig. 5.2 illustrates a typical example neuron that fired tonically during conjugate saccadic eye movements and was completely unresponsive during a visual “blink” paradigm (Fig. 5.2B). This neuron was typical in that it continued to fire during fixation even after the target light had been extinguished. Interestingly however, when the monkey looked from a far target to a near target (i.e., during convergence) the neuron‟s activity decreased. The neuron‟s response decreased as a function of increasing vergence angle such that the neuron was completely silent when the monkey fixated a very near target (Fig. 5.2C).

Quantification of this relationship showed that neural discharge decreased linearly as a function of increasing convergence angle (Fig. 5.2D; slope=-3.7; R=-0.75). Since the activity of this neuron was higher when the eyes were diverged, we refer to this neuron as a „divergence‟ neuron.

In addition to identifying a number of sites at which microstimulation elicited diverging eye movements, we identified (N=3) neighboring sites where the same stimulation parameters elicited converging eye movements. Examples of movements evoked at these sites are illustrated in Fig. 5.3A. Notably, converging eye movements were evoked when the eyes were initially diverged (Fig. 5.3A, left panel) or converged (Fig. 5.3A, right panel). The range of latencies calculated from the onset of stimulation to the onset of either a saccade or convergence eye movement are illustrated in Fig. 5.3B. The average latency to the onset of convergence eye movements was slightly longer than the latency of the saccades (47.2±10.4ms versus

36.9±10.0ms, respectively; p<0.05). Nevertheless, the average latency to the onset of

176

Fig. 5.2: Example of a typical divergence neuron recorded at the same site as the stimulation shown in Fig. 1, which evoked diverging eye movements. Discharge rate of the neuron (A) during very small conjugate saccades, aligned on saccade onset (B) during a blink paradigm and (C) during converging eye movements, aligned on convergence onset. (D) Linear regression between vergence angle and firing rate.

177

convergence was far less than one would expect from a spontaneously initiated convergence eye movement (e.g., <150ms) (Erkelens and Collewijn 1991; Rashbass and Westheimer 1961a;

Westheimer and Mitchell 1956).

At the same sites where microstimulation elicited convergence we recorded neurons that increased their activity during converging eye movements. Fig. 5.3C illustrates one example neuron. Remarkably, the neuron was completely silent while the monkey looked at a distant target suggesting that previous studies, which identified neurons in the rostral SC during tasks that required only far viewing, would not have recorded these neurons. It was not until the monkey looked to a near target that the neuron began to discharge. A linear regression of the neural response plotted against vergence angle shows that this neuron‟s activity was linearly related to vergence angle (Fig. 5.3D; slope=2.7; R=0.79). In contrast, a linear regression between conjugate position and firing rate confirmed that the neuron was unrelated to changes in conjugate position (inset Fig. 5.3D; R=0.07). Since the activity of this neuron was higher during convergence we refer to it as a “convergence” neuron.

Overall, we obtained a population of divergence and convergence neurons in two animals that were linearly related to vergence angle (N=8, slope=-2.5±1.03, R=-0.71±0.17; N=11, slope=3.2±2.3; R=0.61±0.25, for divergence and convergence neurons, respectively). Notably, at all sites in the rostral SC microstimulation elicited either diverging or converging saccades.

Although we could reliably elicit diverging eye movements at all sites that we recorded a divergence neuron, it was not the case that we could always elicit converging eye movements at sites that we recorded convergence neurons. Approximately half of the convergence neurons were recorded at sites where we elicited diverging eye movements when the monkey fixated a near target. Interestingly, we often recorded convergence and divergence neurons within a few

178

µm from each other, on the same recording track. Overall, this suggests that microstimulation could be effectively recruiting the two populations of vergence neurons and thus the resulting converging eye movements would be minimized.

5.4.1 Neurons in rostral SC dynamically encode vergence during vergence tracking

We further tested the responses of diverging and converging neurons during a number of tracking paradigms. We found that when the monkey pursued a distant target (i.e., conjugate pursuit) both neuron types were unresponsive (Fig. 5.4A). However, during sinusoidal, smooth tracking of a target in depth (i.e., vergence pursuit) the neurons were clearly modulated (Fig.

5.4B). In particular, as illustrated in Fig. 5.4B, the firing rate of a divergence neuron (Fig. 5.4B1; same neuron shown in Fig. 5.2) decreased as the eyes tracked an approaching target (i.e., during convergence). In contrast, the firing rate of a convergence neuron (Fig. 5.4B2; same neuron as in

Fig. 5.3) increased as the eyes tracked the approaching target. We found that we could estimate the firing rate of each neuron using a simple linear model:

 FR(t)  b  kVG(t)  rVG(t) where FR is the instantaneous firing rate, b, k and r are constants that represent the bias, the

 estimated position sensitivity and estimated velocity sensitivity, respectively. VG and VG refer to instantaneous vergence position and velocity during vergence tracking. Model fits (in blue) are superimposed on the firing rates (black) for the two example neurons shown in Fig. 5.4B. For our population of neurons, this model well described discharge modulation during vergence pursuit (VAF=0.34±0.19). In addition, as illustrated in Fig. 5.5, a comparison of the bias (Fig. 179

Fig. 5.3: Example convergence eye movements evoked by electrical stimulation and single unit activity of a neuron at the same site. (A) Examples of stimulation in the rostral SC that evoked converging eye movements when the monkey looked either far (left) or near (right). (B) Distribution of latencies calculated from the onset of the stimulation to the onset of either the saccade (black bars) or the vergence eye movement (red bar). (C) Discharge rate of a neuron recorded at the same site as the stimulation during converging eye movements. (D) Linear regression between vergence angle and firing rate.

180

Fig. 5.4: Discharge rates of vergence neurons during smooth pursuit. (A) Discharge rate of a divergence (left) and convergence neuron (right) during conjugate smooth pursuit. (B) Discharge rate of a divergence (left) and convergence neuron (right) during vergence smooth pursuit. Model fits, estimated using eq 1., are superimposed in blue on the firing rate.

181

5.5A) and position sensitivities (Fig. 5.5B) estimated during vergence pursuit were comparable to those estimated using the linear regression during fixation. This result suggests a tight linkage between neuronal responses and motor output during disconjugate eye movements.

5.4.2 Convergence and divergence neurons in the rostral SC encode slow but not fast vergence

We have shown that vergence neurons in the rSC dynamically encode slow changes in vergence during slow smooth pursuit of a target in depth, as well as static changes in vergence angle. We next analyzed whether these same vergence neurons might play a role in commanding eye motion when gaze is rapidly redirected between near and far targets. Notably, 3-dimensional reorienting eye movements are typically characterized by a period of fast saccadic vergence, which quickly redirects the eyes by unequal amounts (i.e., disconjugate saccade), in addition to periods of initial and late slow vergence, which are needed to binocularly position the two eyes and ensure stereopsis. Previous studies have shown that neurons within the brainstem saccadic network dynamically encode the movement of an individual eye during disconjugate saccades but are unresponsive during initial and late periods of slow vergence (King and Zhou 2002; Van

Horn et al. 2008). Accordingly, this left open the question of what premotor neurons were required to binocularly reposition the two eyes. Interestingly, we found that the activity of divergence and convergence neurons in the rostral SC was appropriate for driving the slow vergence preceding and following a disconjugate saccade.

182

Fig. 5.5: Comparison of (A) bias and (B) position sensitivities estimated during vergence pursuit and fixation on a neuron-by-neuron basis.

183

Specifically, as illustrated in Fig. 5.6A and Fig. 5.7A, the recommencement of neuronal responses coincided with the onset of the slow vergence velocity that occurred just after the rapid saccadic component of these disconjugate movements. As is shown in Fig. 5.6B and 7.B the onset of activity, of diverging and converging neurons respectively, coincided with the offset of the fast saccadic vergence (Fig. 5.6B and 5.7B; middle panel) rather than the onset of the saccade

(Fig. 5.6B and 5.7B; left panel) or the onset of steady fixation (Fig. 5.6C, R2=0.76 vs 0.33; and

7C, R2=0.64 vs 0.14). Further quantitative analysis of neuronal discharge dynamics demonstrated that a neuron‟s activity during these slow vergence epochs could be predicted based on the sensitivities estimated during slow vergence tracking. On average, this model predicted 75% of the variance in discharge dynamics that could be accounted for by neuronal sensitivities to eye position and velocity. Thus, the discharge of vergence neurons in the rostral SC not only encode vergence during pursuit in depth but also dynamically encode slow changes in vergence that are suitable for ensuring precise binocular alignment.

5.5 DISCUSSION

Neurons in the rostral SC have been shown to represent small positions errors for multiple eye movements including very small saccades, smooth pursuit and fixation (Bergeron and Guitton, 2000; Choi and Guitton, 2006; Krauzlis et al., 1997; Hafed and Krauzlis 2008). In this study, we extend the functional role of the rostral SC and demonstrate it also plays essential role in vergence eye movements. In particular, using microstimulation we find that we can elicit both converging and diverging eye movements. Moreover, we recorded a distinct class of neurons whose firing rate modulated as a function of vergence angle. These „vergence‟ neurons 184

Fig. 5.6: Divergence neuron during disconjugate eye movement. (A) The neural activity of an example divergence neuron during two typical disconjugate eye movements with periods of slow vergence preceding the onset of a fast disconjugate saccade (dotted vertical line). (B) Raster plots aligned on the onset of disconjugate saccades (left), onset of slow vergence (middle) and onset of fixation (i.e., slow vergence offset) (right). (C) Linear regressions comparing the time of the first spike to the time of saccade offset (i.e., slow vergence onset; left) and time to fixation onset (right).

185

Fig. 5.7: Convergence neuron during disconjugate eye movement. (A) The neural activity of an example rostral SC during two typical disconjugate eye movements with periods of slow vergence preceding the onset of a fast disconjugate saccade (dotted vertical line). (B) Raster plots aligned on the onset of disconjugate saccades (left), onset of slow vergence (middle) and onset of fixation (i.e., slow vergence offset) (right). (C) Linear regressions comparing the time of the first spike to the time of saccades offset (i.e., slow vergence onset; left) and time to fixation onset (right).

186

were identified below visual neurons with foveal and parafoveal visual receptive fields in the intermediate to deep layers of the rostral SC and were recorded amongst rostral SC neurons that responded during visual fixation, smooth pursuit and small saccades. Importantly, identified vergence neurons were unresponsive during conjugate movements, including small saccades and pursuit, but were found to dynamically encode slow changes in vergence angle during pursuit as well as during slow vergence that precedes and follows a 3D saccade to ensure accurate binocular alignment. Overall, our findings provide important new insight into how the brain controls 3-dimensional gaze shifts and suggest that distinct neural pathways control fast and slow vergence.

5.5.1 Microstimulation of the rostral SC

Several lines of evidence suggest that the SC is involved in the neural control of vergence eye movements. In humans, a bilateral lesion of the rostral SC has resulted in a patient being unable to maintain proper binocular eye alignment as they look from a distant to near target (i.e., convergence palsy (Ohtsuka et al. 2002)). Monkeys with large SC lesions have impairments in binocular stereopsis and are less accurate when viewing binocularly as compared to when one eye is covered (i.e., monocular viewing) (Lawler and Cowey 1986). It has been proposed that damage to the SC results in diplopia, due to impaired vergence eye movements, which causes a misalignment of the visual axes. However, precise binocular eye movements before and after SC lesions are needed to test this proposal.

In addition to lesion studies there is also mounting anatomical and electrophysiological evidence suggesting that the SC is involved in driving vergence eye movements. The SC

187

receives disparity-sensitive inputs from cortical and subcortical areas (Ferraina et al. 2000;

Genovesio and Ferraina 2004; Gnadt and Beyer 1998; Gnadt and Mays 1995; Mimeault et al.

2004); Toda et al. 1991) and the SC neurons project to premotor neurons, which encode vergence-related information (Ohtsuka and Nagasaka 1999; Scudder et al. 1996). Moreover, in the cat, vergence eye movements have been elicited by microstimuation of rostral SC (Suzuki et al. 2004). In particular, low current electrical stimulation (<30µA) was found to evoke convergence eye movements, which were often accompanied by conjugate eye movements in the contralateral direction. An evaluation of the binocular effects of microstimulation in the primate rostral SC, at sites where brief microstimulation resulted in the prevention or interruption of conjugate eye movements, also prevented or interrupted vergence movements (Chaturvedi and

Van Gisbergen 2000).

In the present experiment we provide specific evidence that the rostral SC plays a prominent role in generating and maintaining binocular eye movements. We microstimulated the intermediate/deep layers of the rostral SC where we had identified rostral SC neurons that responded during fixation, smooth pursuit and small saccades (Supplemental Fig. 2). Although not as frequent as in the cat, we identified sites where microstimulation resulted in small converging movements, which could be evoked regardless of whether the eyes were converged or diverged. At the majority of the microstimulation sites we found that when the monkey fixated a central, far fixation point, the resulting eye movements were very small conjugate saccades in the contralateral direction. However, when microstimulation was applied when the eyes were initially converged the resulting movements were disconjugate saccades, in the contralateral direction. Notably, the average latency to the initiation of vergence was comparable to that of the initiation of the saccade and was also similar to latencies that been reported in previous studies of 188

microstimulation in the rostral SC (e.g., ~20msec)(Gandhi and Keller 1999). Interestingly, at the same sites that evoked diverging and converging responses we indentified, using single unit recording techniques, a population of neurons that were linearly related to changes in vergence angle. These findings extend a single preliminary report in cat (Jiang et al. 1996), by demonstrating the existence of a specific group of neurons whose discharge encodes changes in vergence angle suggesting that the rostral SC contributes to the control of vergence eye movements.

Neurons related to slow vergence have been previously described near the oculomotor nucleus in monkey (Judge and Cumming 1986; Mays 1984; Zhang et al. 1991; Zhang et al.

1992). The majority of these vergence neurons were identified dorsal and dorsalateral to the oculomotor nucleus and, similar to the neurons described in the present study, were found to be unresponsive during conjugate saccades but either positively or negatively modulated during convergence. Judge and Cummings (1986) noted in their discussion that a few of the recorded neurons were noticeably more dorsal and indicated that they may have been in the pretectum or rostral SC. However Judge and Cummings did not thoroughly test the responses of nearby neurons and thus were unable to conclusively determine the location of these few neurons. In the present study we used microstimulation and single unit techniques to be certain that the vergence neurons that were recorded were indeed in the rostral SC.

Although accommodation normally occurs during convergence, making it is difficult to determine if neurons are responding to changes in vergence or accommodation, the results of previous studies suggest that neurons might preferentially encode vergence, accommodation or both vergence and accomodation (Judge and Cummings 1986; Zhang et al. 1992). The current study was not designed to specifically test for sensitivities to accommodation however we expect

189

that the neurons recorded in the present study would show a similar distribution of sensitivities to vergence and accommodation.

5.5.2 The neural control of fast versus slow vergence

One of the most striking results of the present study was the finding that vergence neurons, identified in the rostral SC, responded only to slow changes in vergence angle. This finding is particularly interesting given what is known about the behavior of 3-dimensional reorienting eye movements. In particular, these movements are characterized by a period of fast saccadic vergence, which quickly redirect the eyes by unequal amounts (i.e., disconjugate saccade; Fig. 5.8A; blue traces), in addition to periods of initial and late slow vergence, which are needed to binocularly position the two eyes to ensure stereopsis (Fig. 5.8A; red traces).

There is extensive data demonstrating that brainstem premotor saccadic neurons (e.g., saccadic burst and burst-tonic neurons) encode integrated conjugate and vergence commands during the fast saccadic component of a 3D eye movement (Sylvestre et al. 2003; Van Horn et al.

2008; Van Horn and Cullen 2008; Zhou and King 1996; 1998). However, studies have also shown that saccadic neurons are silent during the slow component of a 3D gaze shift (e.g., before and after the disconjugate saccade)(Van Horn et al. 2008). Taken together it has been suggested that while the saccadic burst neurons function to rapidly redirect the eyes, an additional command is required to align the fovea of each eye on a target and ensure accurate binocular perception (King and Zhou 2002; 2000; Van Horn et al. 2008; Van Horn and Cullen 2008).

Notably, additional evidence for separate fast and slow pathways for vergence eye movements

190

resides in human lesion studies, which have shown specific deficits in fast vergence eye movements in pontine lesions (Rambold et al. 2005).

At the level of the oculomotoneurons there is evidence that different motoneurons contribute more to certain oculomotor behaviours than others. For example, a neural simulation of the population drive generated by oculomotoneurons during conjugate versus disconjugate saccades has shown that the simulated activity is approximately 10% smaller during disconjugate saccades as compared to conjugate saccades (Sylvestre and Cullen 2002; Van Horn and Cullen

2009). Accordingly it has been proposed that an under sampling of motoneurons more specialized in driving vergence movements could account for this missing drive. Interestingly, retrograde studies have shown that large motoneurons, which lie within the motor nuclei, project to “fast” or singly innervated extraocular fibers (SIFs) and are more involved in generating fast eye movements, whereas slow multiply innervating fibers (MIFs) receive innervations from small “c-group” motoneurons, which tend to lie separately around the periphery of the nucleus

(Buttner-Ennever 2006; Ugolini et al. 2006; Buttner-Ennever et al. 2001; Spencer and Porter

1988). Since the motoneurons recorded in previous studies were most likely from larger motoneurons within the nucleus an under sampling of small MIF motoneurons may account for the missing drive seen during disconjugate saccades.

It has also been shown that that abducens motoneurons that innervate MIFs receive their innervations from premotor sources involved in executing slow eye movements (e.g., vestibular nucleus, prepositus hypoglossi and supraoculomotor nucleus) whereas SIF motoneurons receive premotor signals from SBNs in the saccadic burst generator (Ugolini et al. 2006; Wasicky et al.

2004). Given that the SC projects to the cMRF, we propose that projections from neurons in the rostral SC to neurons in the cMRF, which dynamically encode integrated conjugate and vergence

191

information (Waitzman et al. 2008), to small motoneurons in the abducens could serve as a pathway for mediating slow changes in vergence (Fig. 8B).

To date, the source of fast vergence to the brainstem saccadic burst neurons remains unknown. A likely candidate is saccade-related burst neurons in the deep layers of the caudal SC.

They have direct projections to saccadic burst neurons and receive inputs from disparity sensitive regions including the superficial layers of the SC (Mimeault et al. 2004). Surprisingly, only study has examined the relationship of caudal SC neurons during 3-dimensional viewing (Walton and

Mays 2003). However, although this study described a number of neurons whose discharge was clearly modulated by vergence eye movements, the investigators concluded that the SC was not involved in driving vergence eye movements because the movement fields did not systematically shift as predicted. Notably, there were a number of limitations to this study that justify a re- examination of the SC for evidence of a fast vergence command signal. For example, the study did not attempt to match the generated eye movement command to the optimal amplitude of each specific SC neuron. This is a problem since each neuron in the SC is tuned to a specific saccade vector thus if the task requires an eye movement that does not fall within the optimal saccade vector it will be impossible to properly characterize the neuron‟s discharge characteristics during disconjugate saccades. Given the limitations of this study, in addition to our present finding that the rostral encodes vergence, we suggest that movement related neurons in the caudal SC should be re-examined for evidence of vergence-related information that would be suitable for shaping the discharge dynamics of upstream premotor saccadic neurons.

Taken together, we propose that distinct groupings of neurons within the SC, that encode slow versus fast vergence, underlie the ability to accurately position each of the two eyes when fixating targets in 3-dimensional space. Specifically, inputs from neurons in the rostral SC could

192

serve as a pathway for facilitating slow changes in vergence (Fig. 8B) whereas inputs from caudal SC burst neurons to saccadic burst neurons in the brainstem to large motoneurons in the abducens could mediate fast vergence. Future work is needed to fully understand how coordinated inputs from distinct fast and slow vergence pathways work together to ensure accurate gaze positioning.

193

Fig. 5.8: Premotor control of vergence. (A) Top traces illustrate an example disconjugate eye movement which has a period of fast vergence to quickly redirect the eyes (blue dashed area) as well as initial and late periods of slow vergence to binocularly position the eyes and ensure accurate visual perception. Bottom traces illustrate the typical unit activity of saccadic burst neurons (SBNs; blue) and rostral SC neurons (rSC; red) associated with this movement. (B) Proposed neural circuitry encoding slow (left) versus fast (right) vergence. Slow vergence originating in the rSC projects to neurons in the cMRF and eventually to motoneurons in the abducens nuclei. The premotor saccadic circuitry has been shown to encode signals appropriate for fast vergence.

194

Supplemental Fig. 5.1: Example of a typical parafoveal visual neuron recorded dorsal to the example divergence illustrated in Fig. 2 during a visual blink paradigm. Raster plots are aligned on target onset (left) and blink onset (right).

195

Supplemental Fig. 5.2: Example of a rostral SC neuron typically recorded amongst the convergence and divergence neurons. (A) Raster plot aligned on small saccade onset. (B) Firing rate during conjugate smooth pursuit. (C) Firing rate as a function of motor error calculated during smooth pursuit.

196

Chapter 6

General Discussion

“The real voyage of discovery consists not in seeking new landscapes but in having new eyes.”

-Marcel Proust

How is it that you are able to quickly and accurately switch your gaze from something near to something far? 3-dimensional reorienting eye movements are typically characterized by a period of fast saccadic vergence, which quickly redirects the eyes by unequal amounts (i.e., disconjuagate saccade). In addition, these movements have periods of initial and late slow vergence, which are needed to binocularly position the two eyes to ensure stereopsis.

Theoretically, any disconjugate eye movement can be deconstructed into its conjugate (i.e., average movement of the two eyes) and vergence (i.e., difference in movement of the two eyes) components.

In this thesis, I have investigated how individual neurons contribute to redirecting gaze in

3-dimensional space. Overall, I provide neuronal evidence for two separate neural pathways that encode fast saccadic vergence versus slow vergence rather than conjugate and vergence. I first show that individual neurons within the brainstem saccadic network, which are commonly

197

assumed to drive the conjugate component of eye movements, dynamically encode the movement of an individual eye during disconjugate saccades. Although I found that premotor saccadic neurons carry sufficient vergence information to shape the activity of their targets motoneurons during disconjugate saccades I also show that these neurons are silent during slow vergence that precede and follow a saccade. Interestingly, I subsequently discovered a novel group of neurons in the rostral SC that are appropriate for signalling slow changes in vergence.

In particular, these neurons dynamically encoded changes in vergence angle during slow vergence tracking, fixation in 3-dimensional space and binocular realignment that occurs after disconjugate saccades, but were completely unresponsive during saccades and conjugate smooth pursuit. Taken together, the results from this thesis provide new insight to how the brain precisely coordinates the movement of our eyes during 3-dimensional viewing. In particular, I show that there are separate neural pathways controlling fast versus slow vergence. These results indicate a need to update textbooks and review articles to emphasize the physiological differences between the neural circuitry controlling fast and slow vergence.

6.1 Signals encoded by extraocular motoneurons

Previous studies have provided a complete description of the discharge dynamics of lateral rectus motoneurons (i.e., ABNs) during conjugate and disconjugate saccades (Sylvestre and Cullen 2002). However, in order to create a realistic model of eye movements, it is necessary to also consider the contribution of the medial rectus muscle. Accordingly, in Chapter 2, I provide the first detailed characterization of the dynamics of individual medial rectus neurons in the oculomotor nucleus (OMNs) during conjugate and disconjugate saccades. Like ABNs, I found that a simple, first-order model (i.e., containing eye position and velocity terms) provides 198

an adequate model of neural discharges during both ON and OFF-directed conjugate saccades. I also provide evidence that the activation of the antagonist, as well as agonist motoneuron pools must be considered to understand the neural control of horizontal eye movements across different oculomotor behaviors.

To understand how the medial rectus motoneurons drive disconjugate eye movements, I also evaluated whether the sensitivities estimated during conjugate saccades could be used to predict responses during disconjugate saccades. For the majority of neurons a conjugate-based model failed and instead neurons preferentially encoded the position and velocity of the ipsilateral eye. A neural simulation of the population drive of OMNs revealed that the response generated during disconjugate saccades was slightly smaller (~10%) than that generated during conjugate saccades. A similar result was shown for ABNs (Sylvestre and Cullen 2002). Overall, this general finding is not that surprising if we consider the distribution of ocular preferences of both OMNs and ABNs - a proportion of both are sensitive to the movement of the contralateral eye. Moreover, this finding is also consistent with what is known about the premotor inputs to

OMNs and ABNs – a significant percentage (~40%) of these neurons are also tuned to the movement of both eyes (i.e., conjugate) and OMNs receive converging premotor inputs from a number of different neurons which encode vergence angle (e.g., near response neurons) (Judge and Cumming 1986; Zhang et al. 1991; Zhang et al. 1992) and/or the movement of each eye

(e.g., abducens internuclear neurons, and cMRF) (Sylvestre and Cullen 2002; Waitzman et al.

2008).

Taken together, these results suggest that other mechanisms, such as selective weighting, or a sampling bias may be responsible for the apparent “missing” motoneuron drive during disconjugate saccades. There is evidence that suggests that different motoneurons may contribute

199

more to certain oculomotor behaviors than others. For example, extraocular muscle, which has two types of muscle fibers: twitch and non-twitch fibers (Buttner-Ennever et al. 2001; Spencer and Porter 1988), receives innervations from different types of motoneurons. Twitch fibers receive innervations from large motoneurons, which lie within the oculomotor nuclei, whereas non-twitch fibers received innervations from small motoneurons, which tend to lie separately around the periphery of the nucleus (Buttner-Ennever et al. 2001; Buttner-Ennever 2006; Ugolini et al. 2006). Moreover, small motoneurons, in the abducens have been found to receive innervations from premotor sources involved in executing slow eye movements (e.g., vestibular nucleus, prepositus hypoglossi and supraoculomotor nucleus), whereas large motoneurons receive premotor signals from the saccadic burst generator (Ugolini et al. 2006; Wasicky et al.

2004). I suggest that an under sampling of neurons that could be more specialized for vergence movements (e.g., small motoneurons) could account for the smaller drive observed during disconjugate saccades. Future studies aimed at characterizing and differentiating the responses of large and small motoneurons is needed to fully understand how each type of motoneuron is involved in generating specific oculomotor behaviours.

6.2 Premotor coding in 3D

For over a century, researchers have disagreed how the brain programs conjugate saccades versus vergence eye movements. A highly referenced view is that the brain is organized with two independent pathways - one pathway encoding the conjugate component, and the other pathway encoding the vergence component of any given movement - and that these signals are combined at the level of the motoneurons to move the eyes appropriately (Busettini and Mays

2005a; b; Gamlin 2002; Hering 1977; Mays 1998; Mays and Gamlin 1995). However, numerous 200

studies have provided evidence that argue against this view. First, during disconjugate saccades, vergence velocities are greater than what would be predicted by a linear summation of a conjugate saccade with a saccade-free vergence movement (Collewijn et al. 1997; Enright 1984;

Enright 1992; Kenyon et al. 1980; Kumar et al. 2006; Kumar et al. 2005b; Maxwell and King

1992; Ono et al. 1978; Oohira 1993; Zee et al. 1992). Notably, this phenomenon is commonly referred to as „vergence facilitation‟. Second, a number of traditional “conjugate” premotor neurons have been found to show monocular tuning (Sylvestre et al. 2003; Zhou and King 1998).

The neural mechanism behind vergence facilitation during disconjugate saccades has remained a topic of debate for decades (Cova and Galiana 1995; Gamlin 2002; 1999; King and

Zhou 2002; 2000; Kumar et al. 2005a; Kumar et al. 2006; Kumar et al. 2005b; Mays 1998; Mays and Gamlin 1995; Sylvestre et al. 2003; Zee et al. 1992). The goal of the second set of studies presented in this thesis was to determine if SBNs could account for the increased vergence velocities observed during disconjugate saccades. Previous studies have shown that premotor saccadic neurons show monocular tuning during disconjugate saccades (Sylvestre et al. 2003;

Zhou and King 1998), however this left open the question of whether the overall contribution of the saccadic circuitry is relatively unimportant compared to that of the vergence subsystem. For example, the most recent model explaining how vergence velocities are facilitated during disconjugate saccades indicates that SBNs exclusively encode conjugate saccadic dynamics, and that projections from these conjugate SBNs to the vergence premotor pathway underlie the vergence facilitation during disconjugate saccades (Busettini and Mays 2005b).

Here we provide conclusive evidence that the saccadic burst generator encodes sufficient vergence-related information to drive disconjugate saccades. I found that the onset of facilitated vergence velocities, observed during disconjugate saccades, was synchronized with the burst

201

onset of excitatory and inhibitory brainstem SBNs. Furthermore, the time-varying discharge properties of the majority of SBNs (>70%) preferentially encoded the dynamics of an individual eye during disconjugate saccades. Overall, when the experimental results were implemented into a computer-based simulation (Chapter 3), I found that the premotor saccadic circuitry carries all the vergence drive that is necessary to shape the activity of the abducens motoneurons to which it projects and rules out the alternative hypothesis that an additional input (i.e., from a separate vergence subsystem) is required to shape the activity of abducens motoneurons during disconjugate saccades.

In Chapter 4, I furthered this finding by recording the discharge dynamics of SBNs during vergence facilitated by a vertical saccade. Notably, the task used in this chapter was particularly unique because it required a vertical conjugate command, which would originate from burst neurons in the rostral interstitial nucleus, but did not require the simultaneous production of a horizontal conjugate command. I conclusively showed that the discharge of horizontal SBNs was tightly linked to the onset of facilitated vergence velocities associated with a vertical saccade and that the majority of SBNs consistently encode the velocity of the ipsilateral eye during this task. Taken together, the results of Chapters 3 and 4 strongly support the hypothesis that the premotor commands from the brainstem saccadic circuitry are sufficient to control rapid shifts of gaze in three dimensions (Cova and Galiana 1996; King and Zhou 2002;

2000; Sylvestre et al. 2003).

6.3 Superior colliculus ensures binocular realignment of gaze

While I clearly demonstrate that SBNs carry important vergence-related information to control saccades in three-dimensional space, I also show that SBNs are silent during slow 202

vergence movements. For example, neither IBNs nor EBNs fire any action potentials during periods of slow vergence that precede or follow disconjugate saccades. This finding suggests that while the SBNs function to rapidly redirect the eyes, an additional slow vergence command is required to binocularly position the two eyes to ensure stereopsis (King and Zhou 2000).

In Chapter 5 of this thesis, I identified a novel group of neurons in the rostral SC whose discharges are also appropriate for driving the slow vergence preceding and following a disconjugate saccade. Specifically, I identified two types of neurons in the rostral SC: the first type increased its neuronal discharge during convergence eye movements and the second type decreased its activity during convergence eye movements. Strikingly, while both neuron types were significantly modulated during vergence pursuit neither type responded during fast changes in vergence position (i.e., during disconjugate saccades). This is a very important finding as it complements the results found in Chapters 3 and 4 and suggests that neurons in the rostral SC are involved in the binocular realignment of the eyes. In particular, we propose that inputs from the rostral SC to neurons in the cMRF, which have been shown to encode integrated conjugate and vergence information (Waitzman et al. 2008), to small motoneurons in the abducens, which project to non-twitch fibers in the lateral rectus muscle and are involved in executing slow eye movements (Ugolini et al. 2006), could serve as pathway for mediating slow vergence.

6.4 Clinical applications

In order to have a clear view of the world, the brain must precisely rotate each eye such that they are both aimed at the same point. A failure in binocular coordination can result in ocular misalignment, which is associated with serious visual problems such as double vision and inaccurate depth perception. To appreciate how important it is to have both eyes aligned on a 203

single object, try gently pressing your finger against the side of your eye when you are trying to read this text. A perception of seeing double arises since you are essentially deflecting the image you are trying to focus on onto non-corresponding retinal locations.

It is estimated that 12% of the population suffers from some form of binocular vision disorder, the causes of which often remain unknown. One particularly common disorder is strabismus, or a condition more commonly referred to as „cross eyed‟. This can results from a misalignment of the eyes, which prevents each eye from being accurately aimed at the same point in space. To avoid double vision, the brain will often ignore the image coming from one eye. If the visual system corresponding to one eye develops abnormally, this can further result in a condition called amblyopia, more commonly known as „lazy eye‟.

To date there is no consensus as to how or why strabismus developments. On the one hand, strabismus could be the result of a dysfunctional eye muscle. If the muscle of one eye is weak it would be unable to move the right amount despite receiving appropriate motor commands from the brain. On the other hand, it has been proposed that strabismus is the result of the brain‟s inability to successfully send motor commands to the eye muscle. If strabismus is the result of a neurological problem, it logically follows that treatment should focus on enhancing the brain‟s ability to control eye alignment. For example, the Vision Therapy program uses vision exercises and other treatment devices (e.g., corrective lenses, prisms, eye patches etc.) to treat the visual-motor system in its entirety.

To gain a better understanding of the physiological mechanisms affected during development of strabismus, researchers have begun to perform studies using animal models. For example, eye misalignments can be induced using visual sensory deprivation or surgical methods

(e.g., cutting eye muscles or nerves) in the first few months of life. Importantly, animal models

204

of strabismus seem to be mechanistically similar to the human condition. It has been shown that eye movement behaviours (e.g., eye alignment) in naturally occurring strabismic monkeys are similar to those seen in humans, as well as those seen in the animal models (Das et al. 2005; Fu et al. 2007; Hasany et al. 2008; Tychsen et al. 2008).

Overall, the results of my research are vital for understanding how the brain controls binocular eye movements. Ultimately, by comparing how the brain is normally circuited, to how it develops in strabismic animals, will help establish what factors contribute to developing of strabismus. Consequently, approaches for the rehabilitation of strabismus can further be developed and refined.

7. 5 Conclusions and future directions

Taken together, this thesis has significantly contributed to our understanding of how the brain redirects and aligns gaze between near and far targets. I recorded from individual neurons, at multiple stages of processing within the premotor saccadic circuitry, and found that the majority of the neuronal responses reliably encode the movement of an individual eye rather than the conjugate component of a saccadic eye movement. These results contradict the traditional view that the brainstem saccadic circuitry is strictly for conjugate control. These findings are complimented by discovering of a novel group of neurons in the rostral SC that are appropriate for signalling slow changes in vergence angle, which occur before and after disconjugate saccades, and are essential for accurate binocular perception. Overall, the results of this thesis emphasize the physiological differences between fast and slow vergence and provide neuronal evidence for separate pathways controlling the two distinct phases of disconjugate eye

205

movements. Here I review a number of questions that remain open and require further investigation.

6.5.1 Are small motoneurons more specialized in encoding vergence movements?

In Chapter 2, I found that a simulated population drive of OMNs during disconjugate saccades was slightly less (~10%) than that estimated during conjugate saccades. I suggested that an under sampling of neurons that could be more specialized for vergence movements (e.g., small motoneurons) could account for the smaller drive observed during disconjugate saccades.

Future studies aimed at characterizing and differentiating the responses of large and small motoneurons are needed to fully understand how each type of motoneuron is involved in generating specific oculomotor behaviours.

6.5.2 What are the upstream sources of fast vergence?

In Chapters 3 and 4, I found that the SBNs encode integrated conjugate and vergence information. In particular, I proposed that SBNs are part of a circuitry involved in encoding fast changes in vergence during disconjugate saccades. The source of fast vergence information required to accurately shape the high frequency discharges of SBNs remains unknown. Burst neurons in the more caudal SC are a likely source of this vergence information. Their neural activity is tightly linked to saccadic eye movements, they have strong anatomical connections to

SBNs in the PPRF and receive inputs from disparity sensitive cortical (Ferraina et al. 2000;

Gamlin and Yoon 2000; Gnadt and Beyer 1998) and subcortical regions.

Surprisingly, only one study has explored the caudal SC for signs of vergence-related activity. While this study did describe neurons in the SC that were modulated during vergence 206

movements the authors found that the shifts in the movement fields were not appropriate and concluded that the SC was involved in primarily driving conjugate movements (Walton and

Mays 2003). There are a number of limitations to this study that justify a re-examination of the

SC for evidence of an individual eye command signal. Most notably, the study did not attempt to match the generated eye movement command to the optimal saccade amplitude of each specific

SC neuron. This is problematic since each neuron in the SC is tuned to a specific saccade vector thus if the task requires an eye movement that does not fall within the optimal saccade vector it will be impossible to properly characterize the neuron‟s discharge characteristics during disconjugate saccades. In Chapter 5 of this thesis I show that there exists a population of neurons located in the rostral SC that encode slow changes in vergence and suggest that the rostral SC is part of a circuitry encoding slow vergence. Given that caudal SC burst neurons receive vergence related activity from cortex and project directly to SBNs it seems appropriate that the activity of burst neurons in the caudal SC should be re-examined to determine if they are contribute to the generation of fast saccadic vergence.

Another possible source of vergence inputs to SBNs is the fastigial nucleus of the cerebellum. This nucleus projects directly to the SBN region (Noda et al. 1990), and contains neurons with saccade-related activity (Fuchs et al. 1993). While it is not known whether these neurons encode vergence modulations during disconjugate saccades, neurons in the adjacent interposed nucleus can carry vergence and disparity information (Zhang and Gamlin 1998).

207

208

Appendix A

Local neural processing and the generation of dynamic motor commands within the saccadic premotor network

In the main chapters of this thesis I recorded the spiking activity of individual neurons involved in generating eye movements. In this appendix I simultaneously recorded spikes and local field potentials (LFPs) in the network that commands saccadic eye movements. LFPs are considered to reflect the input to a given brain area while spiking activity represents the output that is sent to other parts of the brain. I compared the information dynamically encoded by LFPs and spikes recorded from individual premotor and motoneurons to evaluate the local computations that take place to produce eye movement commands. This chapter was adapted from Van Horn MR, Mitchell DE, Massot C and Cullen KE. Local neural processing and the generation of dynamic motor commands within the saccadic premotor network. J. Neurosci.

30(32):10905-17, 2010.

209

A.1 ABSTRACT

The ability to accurately control movement requires the computation of a precise motor command. However, the computations that take place within premotor pathways to determine the dynamics of movements are not understood. Here we studied the local processing that generates dynamic motor commands by simultaneously recording spikes and local field potentials (LFPs) in the network that commands saccades. We first compared the information encoded by LFPs and spikes recorded from individual premotor and motoneurons (saccadic burst neurons, omnipause neurons and motoneurons) in monkeys. LFP responses consistent with net depolarizations occurred in association with bursts of spiking activity when saccades were made in a neuron‟s preferred direction. In contrast, when saccades were made in a neuron‟s non- preferred direction, neurons ceased spiking and the associated LFP responses were consistent with net hyperpolarizations. Surprisingly, hyperpolarizing and depolarizing LFPs encoded movement dynamics with equal robustness and accuracy. Second, we compared spiking responses at one hierarchical level of processing to LFPs at the next stage. Latencies and spike- triggered averages of LFP responses were consistent with each neuron‟s place within this circuit.

LFPs reflected relatively local events (<500µm) and encoded important features not available from the spiking train (i.e., hyperpolarizing response). Notably, quantification of their time varying profiles, revealed a precise balance of depolarization and hyperpolarization underlies the production of precise saccadic eye movement commands at both motor and premotor levels.

Overall, simultaneous recordings of LFPs and spiking responses provides an effective means for evaluating the local computations that take place to produce accurate motor commands.

210

A.2 INTRODUCTION

Recordings of high frequency electrical events in the brain (i.e., spikes) have become a standard tool for examining how individual neurons are involved in guiding behavior. Recently, there has been increased interest in investigating the information that is carried by low frequency electrical activity (i.e., local field potentials; LFPs). While spiking activity represents the action potentials (i.e., output signal) produced by a neuron, LFPs are generally thought to reflect the summed activity of synaptic potentials, dendritic spikes and spike afterpotentials occurring around the tip of the recording electrode (Mitzdorf 1985; 1987). Accordingly, LFPs are considered to reflect the input to a given brain area whereas spiking activity represents the output that is sent to other parts of the brain.

A central goal in systems neuroscience is to understand how the brain generates precise motor commands to guide behavior. Concurrent recordings of LFPs and spikes could theoretically provide an effective means for evaluating the local computations that take place to produce accurate motor commands. Paradoxically, however, most previous studies have emphasized the strong correlations – rather than differences - that can be measured between spiking activity and LFPs. For example, studies investigating the control of movement execution have established that LFPs encode many movement-related variables, such as direction, with similar accuracy as the spiking activity (Mehring et al. 2003; Pesaran et al. 2002). While several recent studies have revealed significant differences between spiking activity and LFP activity, these have focused on the decoding of static state variables rather than movement dynamics. For example, in a reaching task, integrated LFP activity codes behavioral states (i.e., planning versus movement execution) more efficiently than do single spikes (Hwang and Andersen 2009;

Pesaran et al. 2002; Scherberger et al. 2005). 211

Thus, a key question that remains unanswered is: Can LFP responses provide insight into the local computations that take place within premotor pathways to determine the precise dynamics of a planned movement trajectory? To address this question, we simultaneously recorded LFPs and the spiking activity of neurons in the well characterized premotor network that commands saccadic eye movements (Fig. A.1). First, to determine how local neural processing precisely shapes the generation of dynamic commands, we characterized the relationship between spiking activity and LFPs recorded in association with individual neurons at both premotor and motor levels of processing. We found that both hyperpolarizing and depolarizing LFPs, like spiking activity, robustly encode saccade trajectories. Moreover, in contrast to previous comparisons in cortical neurons (Rasch et al. 2009), minimal filtering occurs between a neuron‟s input and output. Second, we compared the spiking outputs at one stage of processing to the LFPs recorded at the next hierarchical stage of motor processing to understand the computations that take place within the saccadic premotor network. Our results show that

LFPs encode valuable information that is not available when evaluating spike trains alone.

Notably, the analysis of LFPs revealed that a dynamic balance of depolarization and hyperpolarization underlies the production of precise eye movement commands.

A.3 MATERIALS AND METHODS

A.3.1 Surgical procedures

Two rhesus monkeys (Macaca mulatta) were prepared for chronic extracellular recordings using the aseptic surgical procedures described previously (Sylvestre and Cullen 1999). Briefly, a stainless steel post was attached to the animal's skull with stainless steel screws and dental acrylic, permitting complete immobilization of the animal's head. Two stainless steel recording 212

Fig. A.1: (A) Diagram of saccadic brainstem circuitry. Inhibition is shown in red and excitation is shown in green. Abducens motoneurons (MNs) initiate ipsilateral eye movements by sending signals related to eye velocity and position to lateral rectus eye muscles. When the lateral rectus acts as an agonist (i.e., ipsilateral direction) saccadic burst neurons (SBNs) provide motoneurons with an excitatory input. During saccades in the contralateral direction MNs are inhibited by contralateral inhibitory SBNs (red). Omnipause neurons (OPNs) tonically inhibit SBNs, except during saccades where their activity pauses. (B) Example extracellular recordings of spiking activity of brainstem saccadic neurons. In the ipsilateral direction the MN has position and velocity signals and the SBN has eye velocity information. While the input drive to MNs and SBNs can be inferred from the spikes during ipsilateral saccades (green box) the input drive to MNs and SBNs during contralateral saccades (OPNs during all saccades) cannot be predicted based on the spiking activity (red boxes). Overall, we have limited knowledge of the functional weighting/dynamics of these inputs onto their target neurons.

213

chambers, oriented stereotaxically toward the abducens nucleus on the right and left sides of the brain stem, were also secured to the implant. To record eye position an eye coil (three loops of

Teflon-coated stainless steel wire, 18- to 20-mm diameter) was implanted under the conjunctiva

(Judge et al. 1980). All procedures were approved by the McGill University Animal Care

Committee and complied with the guidelines of the Canadian Council on Animal Care.

A3.2 Experimental paradigms and data acquisition

During experiments monkeys were seated in primate chairs located within the center of a

1-m3 magnetic eye coil system (CNC Engineering). Monkeys were trained to make saccadic eye movements in response to the onset and offset of a red HeNe laser target that was projected onto a cylindrical screen located 55 cm away from the monkey's eyes. The timing and location of target illumination, data acquisition, and on-line data displays were controlled using REX (real- time experimentation), a UNIX-based real-time acquisition system (Hayes et al. 1982).

Horizontal and vertical eye position signals were measured using the magnetic search coil technique (Fuchs and Robinson 1966b; Judge et al. 1980). Position signals were low-pass filtered at 250 Hz (analog eight-pole Bessel filter) and sampled at 1 kHz. Since ocular saccades include very little power at >50 Hz (Cullen et al. 1996a; Van Opstal et al. 1985; Zuber et al. 1968) eye position signals were further digitally filtered (with a 51st-order finite-impulse-response filter with a Hamming window and a cutoff at 125 Hz), before being differentiated to obtain eye velocity signals (using zero-phase forward and reverse digital filtering to prevent phase distortion).

214

Simultaneous extracellular spiking activity and LFP activity were recorded using single enamel-insulated tungsten microelectrodes (FHC; for details, see Sylvestre and Cullen 1999).

Recordings were made from i) omnipause neurons (OPNs) in the nucleus raphe interpositus

(N=12); ii) saccadic burst neurons (SBNs) in the paramedian pontine reticular formation (N=9); and iii) motoneurons in the abducens motor nucleus (MNs) (N=22). Neurons were defined based on their characteristic discharge pattern during saccades. Notably, motoneurons were separated from internuclear neurons based on the relationship between eye position sensitivities and eye position threshold, as initially characterized by Fuchs et al. (Fuchs et al. 1988). The electrode signals were amplified with a high-input impedance head stage (>1G , 2 pF of parallel input capacitance) and filtered by a Multichannel Acquisition Processor (Plexon, Dallas, TX).

Field potential recordings were filtered (0.7-170Hz) and post-processed using the Plexon

FPAlign utility, which effectively compensates for frequency-dependent phase shifts in the LFP band caused by the filters in the Plexon LFP preamplifier boards (Nelson et al. 2008). Unit activity, LFPs, horizontal and vertical eye position, and target position were digitized and saved using a computer-based data acquisition system (Plexon, Dallas, TX). Unit activity and LFPs were digitized and sampled at 1kHz. Subsequent analysis was performed using custom algorithms (Matlab, The MathWorks).

A.3.3 Data analysis

The onset and offset of all saccades was determined using a 20°/s saccade velocity (i.e., horizontal or vertical) criterion. Horizontal saccades were defined as movements for which changes in vertical eye position were <10% of the change in horizontal position. A spike density function, in which a Gaussian function was convolved with the spike train (SD of 5 ms), was 215

used to estimate neuronal firing rate (Cullen et al. 1996a; Sylvestre and Cullen 1999). In order to evaluate and compare the timing and precision of the information carried by LFPs and spike trains, three different approaches were used:

A.3.3.1 Metric analysis

First, the LFPs and spiking activities for each neuron were characterized using classical metric-based analyses (Cullen and Guitton 1997). Specifically, we quantitatively analyzed the time course and magnitude of saccade-related LFPs/spiking activity and correlated each with saccade parameters (e.g., duration, radial eye velocity, and amplitude). The duration of the saccade-related LFP modulation was defined as the interval where the absolute potential was greater than 20% of the maximum value reached during the saccade consistent with previous analyses of intracellular recordings (Yoshida et al. 1999). Saccade duration was defined as the time interval during which velocity was greater than ±20o/s. For each neuron, standard linear regression techniques were used to describe the relationships between 1) LFP duration and saccade duration, 2) peak LFP and peak eye velocity, and 3) time integral of saccade induced

LFP and saccade amplitude.

A.3.3.2 Dynamic Analysis

Second, linear optimization techniques were used to quantify how well neuronal responses dynamically encoded eye movements as has been previously described in the analysis of spike trains (Cullen et al. 1996a; Sylvestre and Cullen 1999; Van Horn et al. 2008).

Specifically, we compared the ability of different dynamic models to describe the relationship between saccadic neuronal discharges/LFP responses and the concurrently recorded eye 216

movement behavior. Briefly, for each neuron we first estimated the sensitivity to ipsilaterally directed saccades (i.e., preferred -direction responses) using the following dynamic model:

 FRest (t  td )  b  r E(t) eq. A.1

Where FRest is the modeled firing rate, b is the firing rate at rest, r is a constant related to

 eye velocity, E is the instantaneous eye velocity and td refers to the dynamic lead time, which was estimated by shifting the FR activity in time until an optimal fit was obtained (Sylvestre and

Cullen 1999).

A similar approach was then taken to describe LFPs responses:

 LFP (t  t )  b  r E(t) eq. A.1b est d

As described in the RESULTS the inclusion of additional terms (e.g., eye position, acceleration) were also evaluated to establish whether we could further improve our ability to describe the relationship between neuronal discharges/LFP responses and behavior. Moreover, since motoneurons are also sensitive to eye position the relationship between firing rates/LFPs obtained from MNs was modeled using the following equation:

 FR (t  t )  b  kE(t)  r E(t) eq. A.2 est d where k and r are constants related to eye position and velocity, respectively. E and are instantaneous eye position and velocity, respectively. This formulation has been previously shown to provide an accurate description of motoneurons spike trains during saccades (Sylvestre and Cullen 2002; 1999; Van Horn and Cullen 2009).

217

To quantify the goodness of fit provided by a given model to neuronal data, we computed the variance-accounted-for VAF=1-[var(FRest-FR)/var(FR)], where FRest represents the modeled firing rate, FR represents the actual firing rate and var, the variance of the signal. Note, that when estimating linear models, the VAF is mathematically equivalent to the correlation of determination, R2. Accordingly, a VAF value of 1 indicated a perfect fit to the data, whereas a value of 0 indicated a fit equivalent to the mean value of the firing rate models.

In addition to the dynamic lead time (td) that was estimated for each model formulation above, we also calculated the timing of the response onset using a 2 standard deviation criterion.

Specifically, in this analysis LFP response onset was defined as the time at which the LFP reached a value of 2 standard deviations above (preferred direction for MNs and SBNs) or below

(MNs and SBNs non-preferred directions) baseline activity. Similarly, firing rate onset was defined as the time at which the firing rate reached a value of 2 standard deviations above

(preferred direction for MNs and SBNs) baseline activity. In the analysis of OPN responses, LFP and firing rate onset was defined as the time at which either signal reached a value of 2 standard deviations below baseline activity. The difference between the firing rate and saccade or LFP and saccade onset was defined as the response lead time.

A.3.3.3 Eye velocity reconstruction

To further compare differences in the information conveyed by spike trains versus LFPs, neuronal responses were characterized using the stimulus reconstruction technique (Rieke et al.

1996). This approach uses an optimal linear decoding algorithm, to assess the quality of the reconstruction of information carried by a given signal. Specifically, we quantified the ability of

218

spike trains versus LFPs to reconstruct eye velocity (i.e., the motor response rather than

„stimulus‟) by computing the coding fraction (CF).

First, it is assumed that the time-dependent eye movement can be described by a first-

order Voltera series (Gabbiani 1996; Rieke et al. 1996) for each neuron i:

 E est (t)  dK ( ) R (t  ). eq.A.3 i  i i

Where Ri is the neuronal activity of the neuron i (e.g., spiking activity or LFP). The kernel K( )

  2 2 that minimizes the mean-square error = <[ E (t) – E est(t)] > for each neuron i, can be

computed in the Fourier domain from the cross spectrum and power spectrum of the response as

follows (Dayan and Abbot 2001; Rieke et al. 1996):

~ P ( f ) K( f )  rs . eq.A 4 Pr ( f )

where (f) is the Fourier transform of K(t). We assessed the quality of linear stimulus estimation  by computing the coding fraction as follows (Gabbiani, 1996; Rieke et al., 1996):

 CF 1 eq.A. 5  min

 where  is the root mean squared error and  is the standard deviation of the „stimulus‟ E (t).  min The coding fraction ranges between 0 and 1 and represents the fraction of the eye velocity signal   correctly estimated, such that CF = 1 represents theoretical optimal reconstruction performance

whereas CF = 0 represents a complete lack of coding. However, even in model sensory systems,

perfect reconstructions based on single neurons are not observed. For example, values of ~0.55

219

are obtained when this same measure to vestibular regular afferents (Sadeghi et al. 2007) or the

P-type neurons of the electric fish (Wessel et al. 1996) was used.

In contrast to the dynamic analysis, the reconstruction technique does not require the

assumption of a particular relationship between the signals. The characteristics of the linear

relation are instead described by the shape of the estimated filter. For example, in the case that a

neuronal signal largely encoded eye velocity signals with a simple scaling factor, i) the filter

shape would be narrow and monophasic, and ii) the filter‟s gain and phase will be related to the

gain and lead time estimated by the dynamic analysis as follows, for each neuron i:

 E est (t)  b  d (r (  t )) Ri (t  ) eq. A.6 i i  i d ,i

where (t)is the Dirac delta function; r, b and td the parameters estimated by the dynamic

analysis in eq. 1 for each neuron i.   For each neuron, the CF was calculated for a data set that was comprised of ~40

saccades. Only neuronal responses associated with saccadic eye movements were used for this

analysis. Intersaccadic data were discarded and replaced with zeros (Harris 1998) to ensure

reconstruction performance (filter length = 1024 ms) over the relevant frequency range.

A.3.3.4 Spike triggered average and spike field coherence

The temporal relationship between the spiking activity and the LFP dynamics was further

characterized by computing spike triggered averages of LFP activity for MNs and SBNs. We

defined the „STA onset‟ as the time when the STA profile crossed a two standard deviation

threshold where the mean and standard deviation of the LFP signal was measured 120-40ms

220

preceding spike onset. To quantify this result, the spike-field coherence (SFC) was calculated by normalizing the STA power spectrum by the LFP power spectrum (Fries et al. 2001).

A.3.3.5 Spectogram analysis

To assess whether the frequency content of LFP activity changed over the course of the saccade, we computed spectrograms of the saccade-induced LFPs recorded at each site using the spectrogram function in Matlab with a window size of 100 (Mathworks). Average LFP spectra were then computed by aligning LFP traces with saccade onset (20deg/s) and averaging the resultant power spectra of LFPs corresponding to saccades of comparable amplitude and direction (e.g., ipsilateral, contralateral, up, down versus oblique).

A.3.3.6 LFP and spike tuning curves

To quantify and compare the tuning of LFPs and spiking responses, we analyzed data from saccades made across a full range of directions. Since LFP amplitudes/spiking activity increase linearly with saccade amplitude we limited this analysis to saccades ranging from 5-

10deg in amplitude. Tuning curves relating the amplitude of the LFP response/number of spikes during the saccade, as a function of saccade direction, were fit with Gaussian curves with the following equation:

- -µ)/2σ² TuningLFP ()=Ae eq. A.7 where deviation and the amplitude, respectively, of the Gaussian curve.

Unless otherwise stated values are means ±SDs.

221

A.4 RESULTS

Extracellular spiking activity and LFPs were recorded simultaneously from single electrodes, and characterized in the context of a well-defined neural network: the brainstem saccadic circuitry (Fig. A.1). Activity was recorded from saccade-related neurons in 3 distinct nuclei whose anatomical connections and spiking output properties are well known (see review

Sparks, 2002), but whose LFP responses have not yet been studied. Notably, while excitatory and inhibitory drives to motor and premotor neurons are assumed to reciprocally excite/inhibit action potential firing (and OPNs during all saccades), to date we have limited knowledge of the relative weighting and/or dynamics of these inputs. Accordingly in this study, simultaneous LFP and spiking responses were recorded from i) omnipause neurons (OPNs) in the nucleus raphe interpositus, which are tonically active during visual fixation and cease firing during saccades, ii) saccadic burst neurons (SBNs) in the paramedian pontine reticular formation, which have reciprocal inhibitory connections to OPNs (Nakao et al. 1980; Strassman et al. 1987) and send saccade-related command signals (e.g., saccade velocity, duration, direction) to motoneurons and iii) motoneurons in the abducens motor nucleus (MNs), which are excited by excitatory

SBNs during ipsilateral saccades and carry eye position and velocity commands to activate the lateral rectus muscle and are inhibited by contralateral inhibitory SBNs, which relax the lateral rectus muscle during contralateral saccades (Hikosaka et al. 1980; Hikosaka et al. 1977; Langer et al. 1986; Maciewicz et al. 1977; Strassman et al. 1986a; b).

A.4.1 LFPs: a reflection of intracellular activity

To test whether LFPs dynamically encode information that is consistent with a given neuron‟s input, we first analyzed the LFPs recorded from OPNs. Notably, previous analyses of

222

intracellular potentials recorded from OPNs had revealed tight relationships between the saccade metrics (e.g., duration, radial eye velocity, and amplitude) and OPN hyperpolarizations (Yoshida et al. 1999). Moreover, this same study showed that the dynamic time course of OPN hyperpolarizations closely resembles that of the velocity profile of the corresponding saccade.

To address whether these same trends could be measured in LFPs recorded with OPNs, we carried out a comparable series of analyses. Fig. A.2A illustrates a continuous recording of extracellular spiking activity and LFPs from a typical OPN during saccades. A cessation in spiking activity, as well as an LFP response consistent with a net hyperpolarization, occurred during saccades in all directions. (Note that in order to better compare LFP responses and eye velocity, we plot the inverted LFP trace in this and all subsequent figures). Fig. A.2B illustrates this finding for an ensemble of saccades made in the ipsilateral (left panel) and contralateral

(right panel) directions. Consistent with the intracellular results, we found a significant correlation between the duration of the LFP modulation and saccade duration in both the ipsilateral (Nsacc=28) and contralateral (Nsacc=26) directions (Fig. A.2C; R=0.8 and R=0.8, respectively), between the peak of the LFP response and peak radial eye velocity (data not shown; R=-0.41 and R=-0.36) as well as between the area of the saccade induced LFP (i.e., time integral) and saccade amplitude (Fig. A.2D; R=-0.8 and R=-0.7). The average correlation coefficients, for each of these three relationships, across the population of OPNs, are summarized in Supplemental Fig. A.S1. Strikingly, a comparison of the dynamic profile of the LFP modulations and corresponding saccadic eye velocity revealed a close resemblance (Fig. A.2E).

To facilitate comparison of the timing and amplitude of the peaks in the LFP trace and the behavior, the eye velocity traces have been superimposed on the LFP trace. As shown in Fig.

A.2E this finding was consistent for both typical (Fig A.2E: left) as well as atypical (i.e., display

223

Fig. A.2: Characteristics of LFPs recorded from typical OPN. (A) Example traces of horizontal eye velocity, LFP recording, and spiking activity are shown in A. Note that, regardless of direction, whenever a saccade is made there is a pause in the unit activity and a deflection in the LFP trace that is consistent with a net hyperpolarization. (B) A raster displaying spiking activity, as well as average LFP, eye velocity and eye position traces are shown for both ipsilateral (left) and contralateral (right) directed saccades. C-D: Plots of (C) LFP duration versus saccade duration and (D) LFP time integral versus saccade amplitude for both ipsilateral (left) and contralateral (right) directed saccades. (E) The LFP profile (not inverted) matched eye velocity (superimposed on the LFP trace for comparison) during a typical (left panel) and atypical (e.g., complex temporal dynamics; right panel) saccade. The timing and amplitude of the peaks in the velocity and LFP profiles are almost synchronous. Inset shows the lead time of the LFP and firing rate responses relative to eye velocity. * additional examples are illustrated in Supplemental A.S2.

224

complex temporal dynamics) saccades (Fig. A.2E: right). Additional examples of the time varying OPN LFP response for additional OPNs are shown in Supplemental Fig. A.2. OPN LFP hyperpolarizing responses, like intracellular membrane potentials, show a tight link with saccadic eye movements. Notably, this was the case in the complete absence of spiking activity

(since OPNs do not fire during saccades), indicating a surprisingly strong relationship between the motor tuning of LFPs and behavior.

SBNs are reciprocally connected to OPNs (Fig. A.1) and generate high frequency bursts of action potentials during ipsilateral saccades (i.e., preferred direction) for which the number of spikes generated, the duration and peak firing rate of the burst is related to ipsilateral saccade amplitude, duration and velocity, respectively [reviewed in (Sparks 2002)]. However, while the spiking output properties of SBNs are well known, to date no study had characterized the LFPs recorded simultaneously from these neurons. We predicted that the LFP responses of SBNs should be consistent with their known inputs [e.g., OPNs, superior colliculus, contralateral inhibitory SBNs (Strassman et al. 1986a; b)] and could potentially provide information that would not be available from the analysis of spike trains.

Fig. A.3A shows the LFPs and spiking activity of a typical SBN. A high frequency burst of action potentials was recorded during saccades made in the preferred direction and the neuron was silent during saccades made in the non-preferred (i.e., contralateral) direction. Similarly, simultaneously recorded LFP responses were consistent with a net depolarization in the neuron‟s preferred direction and a net hyperpolization during the non-preferred direction (Fig. A.3A; filled vs. unfilled boxes). In addition, consistent with our OPN results, the dynamic profile of the LFP response resembled eye velocity during both typical (Fig. A.3B; left) and atypical saccades (e.g., complex temporal dynamics; Fig A.3B, right; and additional examples from SBNs recorded in a

225

Fig.A.3: Characteristics of LFPs recorded from typical SBN. (A) Example traces of horizontal eye velocity, LFP recording, and single-unit activity are shown in A. Note that whenever a saccade is made in the cell‟s preferred direction a LFP response that is consistent with a net depolarization occurs, whereas when the saccade is in the anti-preferred direction, the response is consistent with a net hyperpolarization. (B) The LFP profile matched eye velocity during typical (left panel) and atypical (e.g., complex temporal dynamics; right panel) saccades. The timing and amplitude of the peaks in the velocity and LFP profiles are almost synchronous. Inset shows the lead time of the LFP and firing rate responses relative to eye velocity. * additional examples are illustrated in Supplemental 2. (C) A raster displaying unit activity, as well as average LFP, eye velocity and eye position traces are shown in226 the preferred (C1) and non-preferred directions (C2).

second monkey are shown in Supplemental Fig. A.S2). Moreover, the amplitude of the LFP response increased with saccade amplitude (Fig. A.3C1 and A.3C2, left versus right panels).

Specifically, during saccades in either the preferred or non-preferred direction, the magnitude of the saccade induced LFP modulation was well correlated with saccade amplitude (R=0.7 and

R=-0.7, respectively), the duration of LFP modulation was well correlated to the duration of the corresponding saccade (R=0.9 and R=0.8, respectively) and the peak of the LFP modulation was correlated with the peak eye velocity (R=0.7 and R=-0.7). An analysis of spiking and LFP data recorded from MNs revealed equivalent trends (data not shown). Supplemental Fig. A.S1 summarizes the average correlation coefficients for the three relationships, across each population of neurons recorded. Overall, the findings are consistent with our prediction that LFP responses provide insights into local processing that are not available in the spike train.

A.4.2 LFP response timing is consistent with sequential processing within saccadic network

We next determined whether LFPs provide a precise measure of timing relative to the generation of a motor command during premotor processing. Specifically, we compared the response latencies and information conveyed by spike trains versus LFPs in relation to the generated motor behavior (i.e., saccade) at two sequential stages of processing. Response latencies were first determined by calculating the difference between the onset of the saccade- induced LFP/spiking activity and the onset of the saccade using a 2 standard deviation criterion

(see METHODS). Overall, we found that the average latencies for each neuron type agreed with its place within the circuit‟s hierarchical processing order. In particular, LFP responses recorded

227

from MNs, which are the last neurons in the circuit to be activated to produce a saccade, had shorter latencies (4.4±0.85 msec) compared to SBNs (6.6±1.8 msec) and OPNs (5.5±2.0 msec).

Notably, when we compared the estimated latencies of LFPs to those of spiking activity for a given neuron there were no differences between the two response latencies, (insets of Fig A.2E,

3B). Thus, these results suggest that LFP response latencies are consistent with a neuron‟s place within the neural circuit.

To verify the robustness of this finding, we calculated neuronal response lead times using a second approach. LFP/spiking activity was shifted in time relative to the saccade by the lead time (the optimal dynamic delay (td)) that provided the best fit of the neuronal response when a simple but accurate model was used). Notably, in this analysis the entire portion of the temporally shifted LFP/spike train that was coextensive with the saccade duration was used to fit the model. We used the formulations that had been previously shown to well describe the spike train rate dynamics of SBNs and MNs [(Cullen and Guitton 1997; Sylvestre et al. 2003;

Sylvestre and Cullen 2002; 1999; Van Horn et al. 2008; Van Horn and Cullen 2009)]. The precise relationships between MN and SBN spiking activity and eye motion are described by eq.

1 and 2 in METHODS, respectively. Notably, SBN spike trains encode eye velocity during saccades, while MN spike trains encode both eye position and velocity. Again, the estimated response latencies for MNs (7.6±0.9msec) were shorter than the response latencies of its premotor input the SBNs (9.5±1.3 msec).

A.4.3 The temporal dynamics of LFPs predict motor behaviour

The results presented above introduced and employed a dynamic approach for the estimation of LFP and spike lead time. We next further used this approach to understand the

228

relationship between LFP dynamics and eye movements. Fig. A.4A illustrates the relationships between LFP/spike train dynamics and eye motion for representative recordings made from an example SBN, MN and OPN. As shown by the fits (red traces) superimposed on the LFP trace

(gray trace, third row; VAFSBN=0.66; VAFMN=0.67; VAFOPN=0.65, respectively) and spiking activity (gray filled profile, fourth row; VAFSBN=0.36 and VAFMN=0.62) the dynamics of both neural signals encoded eye velocity well when saccades were made in a given cell‟s preferred direction. Note that the VAF values were always calculated when fitting an entire data set, which included a wide range of saccade amplitudes (see Supplemental Fig. A.S3A for the distribution of saccade amplitudes; average data set included 41.8±5.8 saccades; average saccade amplitude was 12.5±8.5 deg). To ensure that the robustness of the LFP fit was not influenced by a higher sampling of larger or smaller saccades we also predicted the LFP response of MNs for a subgroup of small (4-10deg) and large saccades (>20deg) using the parameters that were estimated for the entire data set. Indeed, a comparison of the VAFs revealed no significant differences between the two groups of saccades (P>0.05; see Supplemental Fig A.S3B).

In contrast, in the absence of spiking activity (e.g., during contralateral saccades for

SBNs and MNs, and ipsilateral and contralateral saccades for OPNs), LFPs well encoded the dynamics of the corresponding saccade (Fig. A.4A; fits superimposed in red on gray LFP trace; third row; VAFSBN=0.52; VAFMN=0.64; VAFOPN=0.76). This discrepancy was due to the fact that LFP activity continued to show a robust time varying modulation in association with these saccades, while there was an absence of corresponding action potentials.

Overall, these main findings were confirmed for each population of neurons recorded.

The time varying profiles of spiking activity and LFPs equally well predicted the dynamics of saccades made in a neuron‟s preferred direction (P>0.05), but only LFPs responses dynamically

229

Fig. A.4: Dynamic analysis of LFP and single unit responses recorded from an example SBN, MN and OPN. (A1-A3) The fits (red traces) obtained using eq. 1 (SBN and OPN) or 2 (MN) (see METHODS for details) are superimposed on the LFP trace (gray trace, third row) and spiking activity (gray filled profile, fourth row) for the neuron‟s preferred (left panels) and non-preferred directions (right panels). Both the firing rate and LFPs encoded eye velocity in the preferred direction. In contrast, in the neuron‟s non-preferred direction, eye velocity was well encoded by LFPs, but not by spiking activity. Insets show the eye velocity coefficient obtained using eq. 1 (SBNs and OPNs) or 2 (MNs) for both ipsilateral and contralateral saccades. Note that the VAF values were always calculated when fitting the entire data set of saccades. (B1-B3): Histograms of VAFs estimated for LFPs during ipsilateral (i.e., preferred) and contralateral (i.e., non-preferred) saccades for each population of neurons recorded.

230

encoded eye velocity during saccades in the non-preferred direction. Notably, in the case of

OPNs (Fig. A.4B3), only the LFPs robustly encoded eye velocity in either direction since there was no concurrent spiking activity. The distributions of VAFs estimated for LFPs, across the three populations of neurons are illustrated in Fig. A.4B. A more complex dynamic model – incorporating an eye acceleration term increased our ability to predict eye movement using the

LFP signal by <10% (ipsilateral direction: VAFLFP-OPN=0.74±0.03; VAFLFP-SBN=0.57±0.09;

VAFLFPMN=0.51±0.05, and contralateral direction: VAFLFP-OPN=0.65±0.05; VAFLFP-

SBN=0.42±0.1; VAFLFPMN=0.44±0.03).

Interestingly, for SBNs and MNs, a comparison of the estimated velocity coefficients showed that the amplitude of modulation of the measured depolarizations and hyperpolarizations of the LFPs, was the same for saccades made in the preferred versus non-preferred directions

(P>0.05; Figs. A.4A1 and A2, insets). As addressed further in the DISCUSSION, this finding indicates a precise balance in the relative strength of depolarization versus hyperpolarization during preferred versus non-preferred saccades, respectively. Additionally, for OPNs, the modulation of measured hyperpolarizing LFPs was comparable in both ipsilateral and contralateral directions (P>0.05; Fig. A.4A3, inset).

To further assess the amount of information conveyed by LFPs and spike trains, we next characterized both signals using a reconstruction technique that is based on a linear decoding algorithm (Rieke et al. 1996) (see MATERIALS AND METHODS). This approach has proven useful for quantifying the amount of information transmitted by spikes (Rieke et al., 1996) and in contrast to the linear regression technique used above, it does not require prior assumptions regarding the relationship between two signals. Instead, the characteristics of the linear relation are revealed by the shape of the optimal filter, which is estimated.

231

Comparison between the original and reconstructed eye velocities computed from LFPs recorded from an example SBN revealed that LFPs encode variations in the eye velocity in both directions with great accuracy (Fig. A.5A and B; top row; blue traces; CFpref =0.56 and CFnon- pref=0.49). In contrast, eye velocity could only be reconstructed from spiking activity when saccades were made in the preferred-direction (Fig. A.5A and B; second row; red traces;

CFpref=0.55 versus CFnon-pref=0). These findings were confirmed for each population of neurons; variations in LFP and spiking activity were equally good at predicting saccade dynamics for saccades in the preferred direction (P>0.05), but reconstructions could only be computed from the LFP signal that occurred during non-preferred directed saccades (e.g., contralateral direction for SBNs and MNs and both ipsilateral and contralateral direction for OPNs). For comparison, we also reconstructed eye velocity using an estimate of neuronal firing rate, computed by convolving a Gaussian function (SD of 5 ms) with the spike train (Cullen et al. 1996). If LFPs represent a low pass filtered version of the spike train, then this estimate of firing rate should convey similar information as LFPs. Fig. A.6 shows that this prediction is contradicted by our quantification; reconstructions based on firing rate (black histograms) failed to predict eye velocity during saccades made in the non-preferred direction. Thus, LFPs convey more information than the spike trains as well as the associated firing rate and are not simply a filtered version of the spike train.

The estimated optimal filters for both LFP and spiking based reconstructions were monophasic and narrow (Fig. A.5, insets), indicating that both neuronal signals could be approximated, in large part, by appropriate scaling of the eye velocity signal consistent with the results of our regression analysis. Because MNs are one synapse closer to the motor output than are the premotor SBNs, we predicted that dynamics of the LFPs recorded with MNs should be

232

Fig.A.5: Stimulus reconstruction of the eye velocity using LFPs and spikes recorded from SBN neuron. Eye velocity (black trace; duplicate for clarity in first and second row), stimulus reconstruction of eye velocity using LFPs (first row; blue trace) or spikes (second row; red trace), LFPs and spiking activity are shown for saccades made in the (A) preferred and (B) non-preferred direction. Note that in the preferred direction, both the LFPs and spikes were able to accurately reconstruct the corresponding eye velocity signal (CFLFP=0.56; CFspike=0.49). In the non-preferred direction, however, only the LFPs precisely described the eye velocity (CFLFP=0.55; CFspike=0). Notably, the CFs were calculated for the entire signal length, which was made up of at least 40 saccades. Optimal filters were calculated for each individual neuron. The insets show the mean optimal filter estimated from LFPs for the reconstructions of the eye velocity profiles for the population of neurons.

233

the more tightly linked to eye movement. Indeed, we found that the CFs of MNs were significantly higher than those of SBNs (P<0.05 for saccades made in the preferred and non- preferred direction, respectively). Average CFs estimated for neuronal responses during preferred and non-preferred direction saccades are summarized in Fig. A.6, for each group of neurons.

A.4.4 Temporal and spatial relationships between spike trains and LFPs

Given the similarity in the filters estimated between spiking activity and eye movement versus LFPs and eye movement, we next asked whether the spike trains of individual neurons could explain the time course of the simultaneously measured LFPs. Our finding that LFPs encode saccades with comparable accuracy as the corresponding spike train, suggested that little filtering occurs between LFP responses and spike activity. To directly test this proposal, we again applied the linear reconstruction approach (Rieke et al. 1996) but this time we estimated the time course of the LFP response from the corresponding spiking activity. We found that the activity of a single neuron well predicted the dynamic profile of the corresponding LFP response

(Supplemental Fig. A.S4A). The shape of the optimized filter was monophasic and narrow and was highly stereotyped at each stage of processing (Supplemental Fig. A.S4B). Interestingly, this filter differs markedly from that estimated in a recent study of neurons in monkey visual cortex

(Rasch et al. 2009) in which the optimal filter shape was substantially wider. We consider the implications of this result in the DISCUSSION.

We further examined the degree of temporal correlation between spikes and LFPs by calculating spike-triggered averages (STAs) of the LFPs (Fries et al. 2001). The STA obtained for MNs and SBNs during saccades towards the preferred direction was consistent with a causal

234

Fig. A.6: Average population coding fractions for (A) MNs, (B) SBNs and (C) OPNs when the reconstruction of the eye velocity was applied with the LFP signal (gray), spike train (white) and firing rate (black). Note that LFPs recorded from MNs, SBNs and OPNs are able to reconstruct the eye velocity in both the neurons‟ preferred (i.e. ipsilateral) and non-preferred (i.e. contralateral) directions. Although the spike train and firing rate accurately reconstruct the eye velocity in the preferred direction of the MNs and SBNs, these signals fail to do so in the neurons‟ non-preferred direction. (C) Spiking activity and firing rate recorded from OPNs were unable to reconstruct the eye velocity in both the ipsilateral and contralateral directions.

235

relationship between LFP and spiking activity (Fig. A.7A). We found that on average onset in the STA preceded the onset of spiking of activity by 5.8ms ±3.3ms for MNS and 6.0ms ±5.3ms for SBNs. Notably, the STA response of the MNs was higher at the end of the saccade as compared to SBNs, which reflects the fact that MNs are also sensitive to eye position and continue to receive inputs related to eye position even once the saccade has finished. The later suppression seen in the burst neurons could be due to a transient increase in inhibition from the

OPNs that occurs at the onset of visual fixation (Bergeron and Guitton 2002). To quantify the

STA modulation, we next calculated the spike-field coherence (SFC) for each neuron (Fries et al.

2001; Fries et al. 1997). The SFC measures the temporal correlation between spikes and LFP oscillations as a function of frequency and we found that the average peak SFC for SBNs was

0.78±0.07 and 0.85±0.1 (mean±SE) for MNs, which occurred at 7.5Hz and 7.6Hz, respectively

(Fig. A.7, inset). Thus, taken together these results are consistent with the proposal that there is a causal temporal relationship between LFPs and spiking activity.

The results presented so far show that LFPs can encode the detailed time course of eye velocity during horizontal saccades (ipsilateral and contralateral) even when there are no spikes generated. To gain a more complete understanding of the spatial tuning of the LFP response, in relation to spiking activity, we next compared the two signals recorded from each neuron for saccades made in all directions. Fig. A.8 shows the spiking activity and average LFPs recorded simultaneously from a typical SBN during saccades, of equal amplitude (10 deg), made in 8 different directions (e.g., left, right, up, down, and oblique). The two signals are superimposed on a corresponding spectrogram. Notably, both spiking and LFP activity showed a marked decrease as the direction of the saccade approached vertical relative to the preferred direction (e.g., towards UP; 0 to 90 degrees and towards DOWN; 0 to 270 deg). However, as saccades became

236

Fig. A.7: Spike triggered average (STA) and spike field coherence (SFC) for MNs (left column) and SBNs (right column). A1 and B1 show the STAs of 2 example cells centered on the first spike of each trial. The onset of the first 2 standard deviation threshold crossing in the STA was computed 120-40ms preceding spike onset (see gray areas). The latency was obtained by measuring the duration between the spike onset and the STA onset (see vertical arrow). For both example neurons, the STA onset precedes the onset of spiking activity. A2 and B2 show population averages of STA superimposed on ±1 SEM (gray area). On average, the STA onset precedes the onset of spiking activity. The insets show the average SFC computed by taking the maximum of the normalized power spectrum for each neuron. Both classes of neurons display high SFCs indicating a strong positive correlation between the LFP signals and the spiking activity.

237

oblique with a contralaterally-directed horizontal components (relative to vertical) again only

LFP responses showed clear modulation, such that responses for purely contralateral saccades were comparable in magnitude (but opposite in polarity) to those observed for saccades in the preferred direction. Nevertheless, regardless of movement direction, saccade-related LFPs were limited to lower frequencies (e.g., <20Hz; compare most leftward and rightward spectrograms in

Fig. A.8A). Notably, no significant power was seen at higher frequencies even when the spectrogram was normalized to baseline activity and plotted using semi-log units (see

Supplemental Fig. A.S5A).

To develop a more complete quantification of LFP tuning, we analyzed data from saccades (N=400) made across a full range of directions (0-360 deg) and amplitudes (± 25 deg).

The results of the analysis are shown in Fig. A.8B. As saccade amplitude increased, LFP activity became increasingly positive (i.e., hotter in color) for ipsilateral saccades and increasingly negative (i.e., cooler in color) for contralateral saccades. Additionally, as expected from the data shown in Fig. A.8A, LFP responses were minimal for vertically directed saccades (baseline = light green). During ipsilaterally directed saccades, the directional tuning of LFP and spiking responses were both well described by a Gaussian function (Fig. A.8B; right inset; compare black and gray circles). In contrast, during contralateral saccades, only LFP responses showed clear Gaussian tuning (Fig A.8B; left inset). Similar results were obtained for the population of

SBNs; the LFPs were tuned for saccades of -1.2±4.4deg and 178±3.7deg (R2=0.60±0.04;

R2=0.55±0.0.4 respectively). Thus, while the spiking activity of SBNs shows tuning for ipsilateral saccades, LFPs are tuned (with opposite polarities) for both ipsilateral and contralateral saccades.

238

Fig. A.8: Spatial relationship of SBN LFP responses. (A) Spiking activity and average LFPs recorded simultaneously from a typical SBN during saccades, of equal amplitude (10deg), made in 8 different directions (e.g., left, right, up, down, and oblique) are superimposed on average spectrograms. 0 and 180 deg correspond to the cells preferred and anti-preferred direction, respectively; 90 and 270 deg correspond to vertical up and vertical down saccades, respectively. The x and y axes represent time and frequency, respectively, while LFP power is color coded. (B) Plot of saccade endpoint relative to the origin where the x and y axis represent horizontal and vertical components, respectively, and the time integral taken over the saccade interval of the corresponding LFP response is color-coded. Left and right insets show tuning curves for saccades (5 to 10 deg) made in contralateral and ipsilateral directions, respectively. 239

Fig. A.9 shows the same analysis as shown in Fig. A.8 but for a typical OPN. In contrast to the

SBN, the OPN ceased firing spikes for 10 deg saccades in all directions including vertical and oblique as well as horizontal movements (Fig. A.9A). Notably, the cessation in spiking activity was accompanied by a corresponding LFP response that was of equal magnitude for saccades made each direction. Additionally, as shown in the corresponding spectrograms, the power in the saccade-induced LFP was comparable and limited to lower frequencies (e.g., <20Hz) regardless of saccade direction (see also Supplemental Fig. A.5B in which the spectrogram is normalized and plotted using semi-log units). Further analysis of responses during saccades spanning a more complete range of directions and amplitudes (see Fig. A.8B), revealed that LFP responses became increasingly negative (e.g., cooler in color = darker blue) as saccade amplitude became larger irrespective of saccade direction (Fig. A.9B). The left and right insets illustrate LFP response amplitude as a function of saccade direction for ipsilaterally and contralaterally directed saccades, respectively. In contrast to SBNs, neither the LFP nor spiking responses of the example OPN showed directional tuning (Fig. A.9B insets; gray circles). This observation was confirmed for the population of OPNs (P=0.59±0.07 and 0.51±0.07 for ipsilateral and contralateral directed saccades, respectively). Therefore, while the LFPs of OPNs encode the speed, duration and amplitude of a saccade, they do not indicate its direction.

LFPs are routinely used as a measure of neuronal activity, however their spatial specificity remains a matter of debate and is likely to differ as a function of the geometry of the structure (e.g., relatively discrete nuclei versus layered structures such as the neocortex or cerebellum) (Katzner et al. 2009; Kreiman et al. 2006; Logothetis et al. 2001). To explicitly assess how electrode location influences the amplitude/reliability of LFP responses in our study, we were able to take advantage of the fact that neurons with similar properties are contained in

240

Fig. A.9: Spatial relationship of OPN LFP responses. (A) Spiking activity and average LFPs recorded simultaneously from a typical OPN during saccades, of equal amplitude, made in 8 different directions (e.g., left, right, up, down, and oblique) are superimposed on average spectrograms. 0 and 180 deg correspond to the cells preferred and anti-preferred direction, respectively. The x and y axes represent time and frequency, respectively, while LFP power is color coded. (B) Plot of saccade endpoint relative to the origin where the x and y axis represent horizontal and vertical components, respectively, and the time integral taken over the saccade interval of the corresponding LFP response is color-coded. Note that, irrespective of direction, as the amplitude of the saccades made become larger, the peak LFP becomes increasingly negative. Left and right insets show a constant LFP response as a function of saccade direction and a lack of unit activity during saccades (5 to 10 deg) made in the contralateral and ipsilateral direction, respectively. 241

distinct nuclei. Notably, the abducens nucleus is ideal for this analysis; it is characterized by a group of neurons that have predominately homogenous discharge properties and are organized in a spherical volume with a radius of ~1mm. During each recording session where a penetration was made in the abducens (N>20) we found that saccade induced LFP responses decreased as the electrode was moved away from the abducens. Indeed, the LFPs were not modulated by saccades once the electrode was more than 500µm from lateral edge of the abducens.

To quantify this finding LFPs were recorded from MNs when the electrode was approximately at the lateral edge of the right abducens nucleus. In two separate sessions the electrode was moved in 50µm increments and we recorded until the electrode was >500µm from the original site. At each site (N=20) we recorded the average amplitude of the LFP response for saccades of ~10 degrees (Nsacc=38.5±3.7). A baseline LFP activity was calculated at each site by averaging the peak LFP value during the 100 msec interval before each saccadic eye movement.

Fig. A.10 plots the average peak (gray diamonds) and baseline (black squares) LFP value as a function of distance from the center of the nucleus during one session. The peak LFP values decreased within the first 150µm and more substantially when the electrode was located 450µm from the center. At 600µm the peak LFP value decayed to a baseline level of activity, represented by the dotted line (average baseline values over all distances). The solid line shows

2 2 an exponential fit to the peak LFP data points (R 1=0.98 and R 2=0.89; and Tau1=0.36 and

Tau2=0.25). These findings indicate that the placement of the electrode relative to the recording site of interest has a significant effect on the measured response such that the LFPs recorded using standard single unit recording approaches in discrete nuclei reflect relatively local (i.e. within a ~ 500 µm radius) events.

242

Fig. A.10: Effect of electrode distance on LFP response. (A) LFPs traces when electrode was placed 0, 150, 300, and 600 µm away from the lateral edge of the abducens nucleus. Note that as the electrode is placed further away, the LFP trace becomes noisier. The average eye position trace is shown below. (B) Plot of the peak (black circles) and baseline (gray squares) LFP response as a function of distance from the nucleus. As the electrode is moved further away the LFP activity decays to a baseline level. Solid line represents an exponential fit to peak LFP values (Tau=0.36; R2=0.98). Dotted line represents average of baseline values for all distances plotted.

243

A.5 DISCUSSION

We studied the local neural processing required for the precise control of eye movement dynamics by simultaneously recording spiking activity and LFPs in the saccadic premotor network. We found that LFPs encode important information that cannot be extracted from single spikes. To understand how local neural processing precisely shapes the generation of dynamic eye movement commands, we first characterized the relationship between spiking activity and

LFPs recorded in association with individual premotor and motoneurons. During saccades made in a neuron‟s preferred direction, bursts of spikes were recorded with depolarizing LFP responses, consistent with a net excitatory synaptic input. LFPs reflected movement dynamics as robustly as spiking activity, and there was a consistent temporal relationship between LFP modulations and spike onset suggesting minimal filtering occurs between a neuron‟s input and output. Conversely, when saccades were made in the opposite direction, neurons ceased spiking and the associated LFP responses were consistent with a net hyperpolarization. Remarkably, hyperpolarizing LFPs dynamically encoded saccade trajectories with equal fidelity as depolarizing LFPs observed during preferred direction saccades. Next, to understand sequential processing within the saccadic premotor network, the spiking output recorded at one stage of processing was compared to the LFP response recorded at the next hierarchical level. Response latencies were in agreement with each neuron‟s place within the circuit. Quantification of the time varying profiles of the LFPs, showed a balance of depolarization and hyperpolarization underlies the production of precise saccadic eye movement commands at both motor and premotor levels such that neuronal activity is symmetrically modulated around a stable resting network state.

244

A.5.1 LFPs display similar properties to intracellular recordings

To date, surprisingly few studies have explicitly compared LFPs and intracellular responses measured in nearby neurons. Similarities have been described in onset latencies, frequency tuning (auditory system: (Kaur et al. 2004) as well as response phase of the two signals [(olfactory system: (Tanaka et al. 2009; Wehr and Laurent 1999); sleep: (Haider et al.

2006; Mukovski et al. 2007)]. Our analysis of OPN LFP responses during saccades provides an unambiguous test of the proposal that this signal encodes the input, rather than the filtered output of a given group of neurons. Here, we show that, in the complete absence of spiking output

LFPs, recorded simultaneously with OPNs, display similar characteristics to the intracellular responses of these neurons. Notably, OPNs exhibit an abrupt hyperpolarization that lasts throughout the entirety of the saccade and is correlated with the saccade‟s amplitude and peak velocity. Moreover, we show that inhibitory LFP responses are not only correlated with these saccadic parameters, but that this signal also robustly encodes saccade dynamics. Accordingly, our results strongly suggest that the pause in OPN spiking activity is caused by IPSPs (as opposed to the removal of an excitatory input) and that this inhibitory input can be measured using extracellular recordings of the associated LFP activity. Together these results support the proposal that LFPs provide a dynamic measure of local synaptic activity, using extracellular recording techniques.

A.5.2 LFPs provide precise timing information and predict motor responses for movements in the non-preferred as well as preferred directions

The finding that LFPs provide a dynamic measure of intracellular responses supports the proposal that simultaneous recordings of LFPs (i.e., input) and spikes (i.e., output) provide a 245

means for evaluating local neural processing that take place in a given brain area. Previous studies of movement execution control have established that LFPs can encode behavioral states

(i.e., planning versus movement execution) more efficiently than single spikes (Pesaran et al.,

2002; Scherberger et al., 2005; Hwang and Andersen, 2009). However, prior to our experiments, no study had addressed how the local computations that take place within premotor pathways determine the precise dynamics of a movement trajectory.

Our recordings show that LFPs dynamically encode important information that cannot be extracted from spiking responses. First, we found LFP response latencies were consistent with the associated neuron‟s place within the premotor circuit‟s hierarchical processing order for saccade generation; estimated latencies for motoneurons were shorter than those recorded for upstream neurons. Notably, the finding that OPN latencies were slightly shorter than SBN latencies is consistent with the proposal that the pause in OPN activity during saccades is the result of an inhibitory input from SBNs (Yoshida et al. 1999). Second, we also found, using both standard regression as well as optimal linear reconstruction approaches, that the relationship between LFP responses and eye movements were equally well described by the same formulations for saccades made in preferred and non-preferred directions. Thus, strikingly LFP activity robustly encoded eye movement dynamics even when there was no corresponding spiking activity. This latter observation has important implications, since it directly refutes the possibility that LFPs reflect a filtered version of the temporally clustered spike trains of ensembles of neurons (Fries et al. 2001; Mitzdorf 1985).

A.5.3 From local field potentials to spike trains: temporal and spatial relationships

246

In a reconstruction analysis, we estimated the optimal linear filters that describe the relationship between neuronal responses and behavior. We chose this approach since it i) requires no inherent assumptions regarding model structure (apart from the requisite assumption of linearity), and ii) has proven to be useful for quantifying the ability of spikes to encode a stimulus in sensory processing pathways (Rieke et al. 1996; Sadeghi et al. 2007). We found that optimal filters were remarkably similar for the LFPs and spike trains recorded from each group of neurons; the shape was narrow and monophasic indicating that both neuronal signals predominantly encoded eye velocity during saccades. On the basis of this result, we predicted that spike train of individual neurons (i.e., output) should have comparable dynamics as the associated LFPs (i.e., input). Indeed, there was minimal filtering between each neuron‟s input and output. While this finding differs markedly from that (Rasch et al. 2009) in neocortex, it is likely that differences in the intrinsic properties of neurons as well as the geometry of the neocortex versus other structures is an important factor in determining the precise relationship between individual spikes and measured LFP activity.

Spectral analysis further demonstrated that LFP responses were predominantly confined to a relatively low frequency range (< 20 Hz; i.e., β and lower frequency band responses). While the polarities of the LFPs associated with recordings of SBNs and MNs were oppositely directed for ipsilateral versus contralateral saccades, the spectrum was comparable for saccades made in either direction. Similarly, the power of the LFPs recorded in association with the pause in OPN firing rate during saccades was concentrated in this frequency range. Our finding that movement- related LFP responses are principally low frequency is consistent with the spectrum of LFPs recorded from neurons in primary motor cortex during arm movements (O'Leary and

Hatsopoulos 2006) as well as the finding that arm movement direction is significantly better

247

encoded in this relatively lower frequency band than by gamma band fluctuations (Moschovakis and Highstein 1994). Taken together, these results suggest that low frequency LFP modulations encode vital information required for the control of eye as well as arm movements.

A.5.4 Functional implications of LFPs as related to the analysis of neural circuits

The saccadic circuitry is well suited to analyze local circuit computations because its connectivity and spiking output properties have been extensively characterized. Previous intracellular recording studies have revealed that abducens MNs receive excitatory input from ipsilateral, excitatory SBNs during ipsilateral saccades, and inhibitory input from the contralateral inhibitory SBNs during contralateral saccades (Curthoys et al. 1984; Strassman et al. 1986b) (see Fig A.1). In addition, OPNs make direct inhibitory connections with SBNs

(Langer and Kaneko 1990; Yoshida et al. 1999) and there is evidence that OPNs receive reciprocal inhibition from SBNs (Strassman et al. 1986b).

While the inhibitory drives to MNs are assumed to inhibit neuronal firing during contralateral saccades (and OPNs during all saccades), to date little is known about the functional weighting/dynamics of these inputs onto their target neurons since previous studies have used traditional extracellular recording techniques that can only describe the excitatory command signals encoded by spiking activity. In this study, we showed that hyperpolarizing

LFPs responses encodes eye velocity signals consistent with the inhibitory drives to MNs and

OPNs.

Finally, our observation that hyperpolarizing LFP responses are also recorded in association with premotor SBNs during contralaterally directed saccades is novel. Prior studies have not quantitatively characterized the intracellular responses of SBNs during saccades. We

248

show that SBN LFPs dynamically encode eye velocity during contralateral saccades. This result has important implications, since it suggests that SBNs are actively inhibited by neurons that carry information about the dynamics of the on-going saccades. Likely candidates for this inhibitory command signal are contralateral SBNs and burst neurons in the superior colliculus, however these projections remain to be verified. What is even more striking is that an analysis of the LFP profiles reveals that a dynamic balance of depolarization and hyperpolarization underlies the production of precise eye movement commands at both the premotor and motor stages.

249

Supplemental Figures: Appendix A

250

Supplemental Fig. A.S1. Average correlation coefficients estimated between the duration of the LFP modulation and saccade duration (dark gray histograms), between the peak of the LFP response and the peak radial eye velocity (black histograms), as well as between the area of the saccade induced LFP (i.e., time integral) and saccade amplitude (light gray histograms) for the population of (A) MNs, (B) SBNs and (C) OPNs in both ipsilateral and contralateral directed saccades.

251

Supplemental Fig. A.S2: Additional examples of the time varying profiles of the LFPs for three OPNs (left) and three SBNs (right). Examples are shown for both monkeys to illustrate that the time varying LFP profile matched eye velocity (superimposed on the LFP trace for comparison) during typical and atypical saccades in both animals.

252

Supplemental Fig. A.S3: (A) Distribution of saccade amplitudes that were used to estimate the LFP and firing rate. (B) To confirm that the robustness of fit was not influenced by an oversampling of large or small saccades we broke the data recorded for MNs into large (>20deg) and smaller saccades (4-10deg). We then predicted the LFP response based on the parameters estimated for the entire data set. A comparison of the VAFs across the two groups of saccades revealed no significant differences (P>0.05).

253

Supplemental Fig. A.S4:Average coding fractions (A1) MNs and (A2) SBNs when the LFP signal was reconstructed from the corresponding spike train (open bars) or firing rate (filled bars). (B) The optimized filters were monophasic and narrow for (B1) MNs and (B2) SBNs.

254

Supplemental Fig.A.S5. Average spectrogram corresponding to an (A) OPN and (B) SBN during the ipsilateral direction, as shown in Fig. 8 and 9, normalized to baseline activity and plotted using a semi-log scale to illustrate that no additional power is seen in the higher frequencies. The x and y axes represent time and frequency, respectively, while LFP power is color coded and normalized to baseline activity.

255

Supplemental Fig. A.S6: Distributions of the amplitudes (A), means (B), and widths (C) of turning curves (respectively) for LFP activity recorded from saccadic burst neurons during ipsilaterally-directed (gray bars) and contralaterally-directed saccades (white bars). The tuning curves for ipsilateral and contralateral saccades were approximately equal in amplitude but in opposite directions (i.e. positive for ipsilateral and negative for contralateral). (B) The tuning curves for ipsilateral and contralateral saccades were centered around 0 and 180deg, respectively. (C) the distributions of the tuning curve widths for ipsilateral and contralateral saccades were identical.

256

References

Barlow HB, Blakemore C, and Pettigrew JD. The neural mechanism of binocular depth discrimination. J Physiol 193: 327-342, 1967.

Bergeron A, and Guitton D. Fixation neurons in the superior colliculus encode distance between current and desired gaze positions. Nat Neurosci 3: 932-939, 2000.

Bergeron A, and Guitton D. In multiple-step gaze shifts: omnipause (OPNs) and collicular fixation neurons encode gaze position error; OPNs gate saccades. J Neurophysiol 88: 1726-1742, 2002.

Busettini C, and Mays LE. Pontine omnipause activity during conjugate and disconjugate eye movements in macaques. J Neurophysiol 90: 3838-3853, 2003.

Busettini C, and Mays LE. Saccade-vergence interactions in macaques. I. Test of the omnipause Multiply Model. J Neurophysiol 94: 2295-2311, 2005a.

Busettini C, and Mays LE. Saccade-vergence interactions in macaques. II. Vergence enhancement as the product of a local feedback vergence motor error and a weighted saccadic burst. J Neurophysiol 94: 2312-2330, 2005b.

Buttner-Ennever J, Horn A, Scherberger H, and D' Ascanio P. Motoneurons of twitch and nontwitch extraocular muscle fibers in the abducens, trochlear, and oculomotor nuclei of monkeys. J Comp Neurol 438: 318-335, 2001.

Buttner-Ennever JA. The extraocular motor nuclei: organization and functional neuroanatomy. Progress in brain research 151: 95-125, 2006.

Buttner-Ennever JA, and Akert K. Medial rectus subgroups of the oculomotor nucleus ant their abducens internuclear input in the monkey. J Comp Neurol 197: 17-27, 1981.

Buttner-Ennever JA, Cohen B, Pause M, and Fries W. Raphe nucleus of the pons containing omnipause neurons of the oculomotor system in the monkey, and its homologue in man. J Comp Neurol 267: 307-321, 1988.

Buttner U, Buttner-Ennever J, and Henn V. Vertical eye movement related unit activity in the rostral mesencephalic reticular formation of the alert monkey. Brain Res 130: 239-252, 1977.

Cannon S, and Robinson D. Loss of the neural integrator of the oculomotor system from brain stem lesions in monkey. J Neurophysiol 57: 1383-1409, 1987.

257

Carpenter J, and Bithell J. Bootstrap confidence intervals: when, which, what? A practical guide for medical statisticians. Statist Med 19: 1141-1164, 2000.

Carpenter MB, Mc MR, and Hanna GR. Disturbances of conjugate horizontal eye movements in the monkey. I. Physiological effects and anatomical degeneration resulting from lesions of the abducens nucleus and nerve. Arch Neurol 8: 231-247, 1963.

Chaturvedi V, and Van Gisbergen J. Perturbation of combined saccade-vergence movements by microstimulation in monkey superior colliculus. J Neurophysiol 81: 2279-2296, 1999.

Chaturvedi V, and Van Gisbergen J. Stimulation in the rostral pole of monkey superior colliculus: effects on vergence eye movements. Exp Brain Res 132: 72-78, 2000.

Choi WY, and Guitton D. Responses of collicular fixation neurons to gaze shift perturbations in head-unrestrained monkey reveal gaze feedback control. Neuron 50: 491-505, 2006.

Clendaniel RA, and Mays LE. Characteristics of antidromically identified oculomotor internuclear neurons during vergence and versional eye movements. J Neurophysiol 71: 1111- 1127, 1994.

Collewijn H, Erkelens CJ, and Steinman RM. Trajectories of the human binocular fixation point during conjugate and non-conjugate gaze-shifts. Vision Res 37: 1049-1069, 1997.

Collewijn H, Erkelens CJ, and Steinman RM. Voluntary binocular gaze-shifts in the plane of regard: dynamics of version and vergence. Vision Res 35: 3335-3358, 1995.

Collins C. The Control of Eye Movements. New York: ed American Press, 1971, p. 283-325. Collins CC, O'Meara D, and Scott AB. Muscle tension during unrestrained human eye movements. J Physiol 245: 351-369, 1975.

Cova A, and Galiana H. A bilateral model integrating vergence and the vestibulo-ocular reflex. Exp Brain Res 107: 435-452, 1996.

Cova A, and Galiana H. Providing distinct vergence and version dynamics in a bilateral oculomotor network. Vision Res 35: 3359-3371, 1995.

Crandall WF, and Keller EL. Visual and oculomotor signals in nucleus reticularis tegmenti pontis in alert monkey. J Neurophysiol 54: 1326-1345, 1985.

Crawford JD, and Vilis T. Axes of eye rotation and Listing's law during rotations of the head. J Neurophysiol 65: 407-423, 1991.

Crawford JD, and Vilis T. Symmetry of oculomotor burst neuron coordinates about Listing's plane. J Neurophysiol 68: 432-448, 1992.

258

Cullen K, and Guitton D. Analysis of primate IBN spike trains using system identification techniques. II. relationship to gaze, eye, and head movement dynamics during head-free gaze shifts. J Neurophysiol 78: 3283-3306, 1997b.

Cullen K, Guitton D, Rey C, and Jiang W. Gaze-related activity of putative inhibitory burst neurons in the head-free cat. J Neurophysiol 70: 2678-2683, 1993b.

Cullen K, and McCrea R. Firing behavior of brain stem neurons during voluntary cancellation of the horizontal vestibuloocular reflex. I. Secondary vestibular neurons. J Neurophysiol 70: 828- 843, 1993.

Cullen K, Rey C, Guitton D, and Galiana H. The use of system identification techniques in the analysis of oculomotor burst neuron spike train dynamics. J Comput Neurosci 3: 347-368, 1996a.

Cullen KE, Chen-Huang C, and McCrea RA. Firing behavior of brain stem neurons during voluntary cancellation of the horizontal vestibuloocular reflex. II. Eye movement related neurons. J Neurophysiol 70: 844-856, 1993.

Cullen KE, and Guitton D. Analysis of primate IBN spike trains using system identification techniques. I. Relationship To eye movement dynamics during head-fixed saccades. J Neurophysiol 78: 3259-3282, 1997.

Cullen KE, and Guitton D. Inhibitory burst neuron activity encodes gaze, not eye, metrics and dynamics during passive head on body rotation. Evidence that vestibular signals supplement visual information in the control of gaze shifts. Ann N Y Acad Sci 781: 601-606, 1996.

Cullen KE, Rey CG, Guitton D, and Galiana HL. The use of system identification techniques in the analysis of oculomotor burst neuron spike train dynamics. Journal of computational neuroscience 3: 347-368, 1996b.

Curthoys IS, Markham C, and Furuya N. Direct projection of pause neurons to nystagmus- related excitatory burst neurons in the cat pontine reticular formation. . Exp Neurol 83: 414-442, 1984.

Das VE, Fu LN, Mustari MJ, and Tusa RJ. Incomitance in monkeys with strabismus. Strabismus 13: 33-41, 2005.

Dayan P, and Abbot L. Theoretical neuroscience:computational and methematical modeling of neural systems. Cambridge, MA:MIT, 2001.

Eckmiller R, Blair S, and Westheimer G. Fine structure of saccade bursts in macaque pontine neurons. Brain Res 181: 460-464, 1980.

Eckmiller R, and Mackeben M. Pre-motor single unit activity in the monkey brain stem correlated with eye velocity during pursuit. Brain Res 184: 210-214, 1980.

259

Enright J. Changes in vergence mediated by saccades. J Physiol 350: 9-31, 1984.

Enright JT. The remarkable saccades of asymmetrical vergence. Vision Res 32: 2261-2276, 1992.

Erkelens CJ, and Collewijn H. Control of vergence: gating among disparity inputs by voluntary target selection. Exp Brain Res 87: 671-678, 1991.

Erkelens CJ, Van der Steen J, Steinman RM, and Collewijn H. Ocular vergence under natural conditions. I. Continuous changes of target distance along the median plane. Proc R Soc Lond B Biol Sci 236: 417-440, 1989.

Ferraina S, Pare M, and Wurtz R. Disparity sensitivity of frontal eye field neurons. J Neurophysiol 83: 625-629, 2000.

Freedman E, Stanford T, and Sparks D. Combined eye-head gaze shifts produced by electrical stimulation of the superior colliculus in rhesus monkeys. J Neurophysiol 76: 927-952, 1996.

Fries P, Reynolds JH, Rorie AE, and Desimone R. Modulation of oscillatory neuronal synchronization by selective visual attention. Science 291: 1560-1563, 2001.

Fries P, Roelfsema PR, Engel AK, Konig P, and Singer W. Synchronization of oscillatory responses in visual cortex correlates with perception in interocular rivalry. Proceedings of the National Academy of Sciences of the United States of America 94: 12699-12704, 1997.

Fu L, Tusa RJ, Mustari MJ, and Das VE. Horizontal saccade disconjugacy in strabismic monkeys. Invest Ophthalmol Vis Sci 48: 3107-3114, 2007.

Fuchs AF, and Robinson DA. A method for measuring horizontal and vertical eye movement chronically in the monkey. J Appl Physiol 21: 1068-1070, 1966a.

Fuchs AF, and Robinson DA. A method for measuring horizontal and vertical eye movements in the monkey. J Physiol 191: 609-631, 1966b.

Fuchs AF, Robinson DA, and Straube A. Role of the caudal fastigial nucleus in saccade generation. I. Neuronal discharge pattern. J Neurophysiol 70: 1723-1740, 1993.

Fuchs AF, Scudder CA, and Kaneko C. Discharge patterns and recruitment order of identified motoneurons and internuclear neurons in the monkey abducens nucleus. J Neurophysiol 60: 1874-1895, 1988.

Fukushima K, Kaneko C, and Fuchs AF. The neuronal substrate of integration in the oculomotor system. Prog Neurobiol 39: 609-639, 1992.

260

Furuya N, and Markham CH. Direct inhibitory synaptic linkage of pause neurons with burst inhibitory neurons. Brain Res 245: 139-143, 1982.

Gabbiani. Coding of time varying signals in spike trains of linear and half-wave rectifying neurons. Network 7: 61-85, 1996.

Gamlin PD. Neural mechanisms for the control of vergence eye movements. Ann N Y Acad Sci 956: 264-272, 2002.

Gamlin PD. Subcortical neural circuits for ocular accommodation and vergence in primates. Ophthal Physiol Opt 19: 81-89, 1999.

Gamlin PD, Gnadt JW, and Mays LE. Abducens internuclear neurons carry an inappropriate signal for ocular convergence. J Neurophysiol 62: 70-81, 1989.

Gamlin PD, and Mays LE. Dynamic properties of medial rectus motoneurons during vergence eye movements. J Neurophysiol 67: 64-74, 1992.

Gamlin PD, and Yoon K. An area for vergence eye movement in primate frontal cortex. Nature 407: 1003-1007, 2000.

Gamlin PD, Zhang Y, Clendaniel RA, and Mays LE. Behavior of identified Edinger-Westphal neurons during ocular accommodation. J Neurophysiol 72: 2368-2382, 1994.

Gandhi NJ, and Keller EL. Comparison of saccades perturbed by stimulation of the rostral superior colliculus, the caudal superior colliculus, and the omnipause neuron region. J Neurophysiol 82: 3236-3253, 1999.

Genovesio A, and Ferraina S. Integration of retinal disparity and fixation-distance related signals toward an egocentric coding of distance in the posterior parietal cortex of primates. J Neurophysiol 91: 2670-2684, 2004.

Gnadt JW, and Beyer J. Eye movements in depth: What does the monkey's parietal cortex tell the superior colliculus? Neuroreport 9: 233-238, 1998.

Gnadt JW, Gamlin PD, Mays LE, and Zhang Y. Vergence-related cells near the abducens nuclei. . Soc Neurosci Abstr 14: 612, 1988.

Gnadt JW, and Mays LE. Neurons in monkey parietal area LIP are tuned for eye-movement parameters in three-dimensional space. J Neurophysiol 73: 280-297, 1995.

Hafed ZM, Goffart L, and Krauzlis RJ. A neural mechanism for microsaccade generation in the primate superior colliculus. Science 323: 940-943, 2009.

261

Hafed ZM, and Krauzlis RJ. Goal representations dominate superior colliculus activity during extrafoveal tracking. J Neurosci 28: 9426-9439, 2008.

Haider B, Duque A, Hasenstaub AR, and McCormick DA. Neocortical network activity in vivo is generated through a dynamic balance of excitation and inhibition. J Neurosci 26: 4535-4545, 2006.

Harris CM. The Fourier analysis of biological transients. Journal of neuroscience methods 83: 15-34, 1998.

Hasany A, Wong A, Foeller P, Bradley D, and Tychsen L. Duration of binocular decorrelation in infancy predicts the severity of nasotemporal pursuit asymmetries in strabismic macaque monkeys. Neuroscience 156: 403-411, 2008.

Hayes A, Richmond BJ, and Optican LM. A UNIX-based multiple process system for real-time data acquisition and control. In: WESON Conf Proc1982, p. 1-10.

Hazel T, Sklavos S, and Dean P. Estimation of premotor synaptic drives to simulated abducens motoneurons for control of eye position. . Exp Brain Res 146: 184-196, 2002.

Hepp K, and Henn V. Spatio-temporal recoding of rapid eye movement signals in the monkey paramedian pontine reticular formation (PPRF). Exp Brain Res 52: 105-120, 1983.

Hering E. Lehre vom binokularen Sehen. (The theory of binocular vision) (1868). New York: Plenum Press, 1977.

Highstein SM, and Baker R. Excitatory termination of abducens internuclear neurons on medial rectus motoneurons: relationship to syndrome of internuclear ophthalmoplegia. J Neurophysiol 41: 1647-1661, 1978.

Highstein SM, Karabelas A, Baker R, and McCrea RA. Comparison of the morphology of physiologically identified abducens motor and internuclear neurons in the cat: a light microscopic study employing the intracellular injection of horseradish peroxidase. J Comp Neurol 208: 369-381, 1982.

Hikosaka O, Igusa Y, and Imai H. Inhibitory connections of nystagmus-related reticular burst neurons with neurons in the abducens, prepositus hypoglossi and vestibular nuclei in the cat. Exp Brain Res 39: 301-311, 1980.

Hikosaka O, Igusa Y, Nakao S, and Shimazu H. Direct inhibitory synaptic linkage of pontomedullary reticular burst neurons with abducens motoneurons in the cat. Exp Brain Res 33: 337-352, 1978.

Hikosaka O, and Kawakami T. Inhibitory reticular neurons related to the quick phase of vestibular nystagmus--their location and projection. Exp Brain Res 27: 377-386, 1977.

262

Hikosaka O, Maeda M, Nakao S, Shimazu H, and Shinoda Y. Presynaptic impulses in the abducens nucleus and their relation to postsynaptic potentials in motoneurons during vestibular nystagmus. Exp Brain Res 27: 355-376, 1977.

Horn A, Buttner-Ennever J, Wahle P, and Reichenberger I. Neurotransmitter profile of saccadic omnipause neurons in nucleus raphe interpositus. J Neurosci 14: 2032-2046, 1994.

Hubel DH, and Wiesel TN. Receptive fields, binocular interaction and functional architecture in the cat's visual cortex. J Physiol 160: 106-154, 1962.

Hubel DH, and Wiesel TN. Stereoscopic vision in macaque monkey. Cells sensitive to binocular depth in area 18 of the macaque monkey cortex. Nature 225: 41-42, 1970.

Hwang EJ, and Andersen RA. Brain control of movement execution onset using local field potentials in posterior parietal cortex. J Neurosci 29: 14363-14370, 2009.

Jiang H, Guitton D, and Cullen KE. Near-response-related neural activity in the rostral superior colliculus of the cat. Soc Neurosci Abstr 22: 1996.

Judge SJ, and Cumming B. Neurons in monkey midbrain with activity related to vergence eye movement and accomodation. J Neurophysiol 55: 915-930, 1986.

Judge SJ, Richmond BJ, and Chu FC. Implantation of magnetic search coils for measurement of eye position: an improved method. Vision Res 20: 535-538, 1980.

Kaneko C, Evinger C, and Fuchs AF. Role of cat pontine burst neurons in generation of saccadic eye movements. J Neurophysiol 46: 387-408, 1981.

Kaneko CR. Saccade-related, long-lead burst neurons in the monkey rostral pons. J Neurophysiol 95: 979-994, 2006.

Katzner S, Nauhaus I, Benucci A, Bonin V, Ringach DL, and Carandini M. Local origin of field potentials in visual cortex. Neuron 61: 35-41, 2009.

Kaur S, Lazar R, and Metherate R. Intracortical pathways determine breadth of subthreshold frequency receptive fields in primary auditory cortex. J Neurophysiol 91: 2551-2567, 2004.

Keller EL. Accomodative vergence in the alert monkey. Motor unit analysis. Vision Res 13: 1565-1575, 1973.

Keller EL. Participation of medial pontine reticular formation in eye movement generation in monkey. . J Neurophysiol 37: 316-332, 1974.

263

Kenyon R, Ciuffreda K, and Stark L. Unequal saccades during vergence. Am Journ Opt & Physiol Opt 57: 586-594, 1980.

King WM, and Fuchs AF. Reticular control of vertical saccadic eye movements by mesencephalic burst neurons. J Neurophysiol 42: 861-876, 1979.

King WM, and Zhou W. Neural basis of disjunctive eye movements. Ann N Y Acad Sci 956: 273- 283, 2002.

King WM, and Zhou W. New ideas about binocular coordination of eye movements: is there a chameleon in the primate family tree? Anat Rec 261: 153-161, 2000.

King WM, Zhou W, Tomlinson RD, McConville KM, Page WK, Paige GD, and Maxwell JS. Eye position signals in the abducens and oculomotor nuclei of monkeys during ocular convergence. J Vestib Res 4: 401-408, 1994.

Krauzlis RJ, Basso MA, and Wurtz RH. Discharge properties of neurons in the rostral superior colliculus of the monkey during smooth-pursuit eye movements. J Neurophysiol 84: 876-891, 2000.

Krauzlis RJ, Basso MA, and Wurtz RH. Shared motor error for multiple eye movements. Science 276: 1693-1695, 1997.

Kreiman G, Hung CP, Kraskov A, Quiroga RQ, Poggio T, and DiCarlo JJ. Object selectivity of local field potentials and spikes in the macaque inferior temporal cortex. Neuron 49: 433-445, 2006.

Kumar AN, Han Y, Dell'osso LF, Durand DM, and Leigh RJ. Directional asymmetry during combined saccade-vergence movements. J Neurophysiol 93: 2797-2808, 2005a.

Kumar AN, Han YH, Kirsch RF, Dell'Osso LF, King WM, and Leigh RJ. Tests of models for saccade-vergence interaction using novel stimulus conditions. Biol Cybern 95: 143-157, 2006.

Kumar AN, Han YH, Liao K, and Leigh RJ. Tests of Hering- and Helmholtz-type models for saccade-vergence interactions by comparing visually guided and memory-guided movements. Ann N Y Acad Sci 1039: 466-469, 2005b.

Langer TP, Kaneko C, Scudder CA, and Fuchs AF. Afferents to the abducens nucleus in the monkey and cat. J Comp Neurol 245: 379-400, 1986.

Langer TP, and Kaneko CR. Brainstem afferents to the oculomotor omnipause neurons in monkey. J Comp Neurol 295: 413-427, 1990.

Lawler KA, and Cowey A. The effects of pretectal and superior collicular lesions on binocular vision. Exp Brain Res 63: 402-408, 1986.

264

Leigh RJ, and Zee DS. The neurology of eye movements. New York: Oxford University Press, 1999, p. 286-320.

Logothetis NK, Pauls J, Augath M, Trinath T, and Oeltermann A. Neurophysiological investigation of the basis of the fMRI signal. Nature 412: 150-157, 2001.

Luque MA, Perez-Perez MP, Herrero L, Waitzman DM, and Torres B. Eye movements evoked by electrical microstimulation of the mesencephalic reticular formation in goldfish. Neuroscience 137: 1051-1073, 2006.

Luschei E, and Fuchs AF. Activity of brain stem neurons during eye movements of alert monkeys. J Neurophysiol 35: 445-461, 1972.

Maciewicz R, Kaneko C, Highstein SM, and Baker R. Morphophysiological identification of interneurons in the oculomotor nucleus that project to the abducens nucleus in the cat. . Brain Res 96: 60-65, 1975.

Maciewicz RJ, Kaneko CR, Highstein SM, and Eagen K. Vestibular and medullary brain stem afferents to the abducens nucleus in the cat. Brain Res 123: 229-240, 1977.

Maxwell JS, and King WM. Dynamics and efficacy of saccade-facilitated vergence eye movements in monkeys. J Neurophysiol 68: 1248-1260, 1992.

Mays LE. Has Hering been hooked? Nature Med 4: 889-890, 1998.

Mays LE. Neural control of vergence eye movements: convergence and divergence neurons in midbrain. J Neurophysiol 51: 1091-1108, 1984.

Mays LE, and Gamlin PD. A neural mechanism subserving saccade-vergence interactions. In: Eye movement research: mechanisms, processes and applications, edited by Findlay J. WR, Kentridge R.W. Amsterdam: Elsevier, 1995b, p. 215-223.

Mays LE, and Gamlin PD. Neuronal circuitry controlling the near response. Curr Opin Neurobiol 5: 763-768, 1995.

Mays LE, and Porter JD. Neural control of vergence eye movements: activity of abducens and oculomotor neurons. J Neurophysiol 52: 743-761, 1984.

Mays LE, Porter JD, Gamlin PD, and Tello CA. Neural control of vergence eye movements: neurons encoding vergence velocity. J Neurophysiol 56: 1007-1021, 1986.

McConville KM, Tomlinson D, King WM, Paige GD, and Na EQ. Eye position signals in the vestibular nuclei: Consequences for models of integrator function. J Vest Res 4: 391-400, 1994.

265

McCrea RA. Neuroanatomy of the oculomotor system. The nucleus prepositus. Rev Oculomot Res 2: 203-223, 1988.

McFarland J, and Fuchs A. Discharge patterns in nucleus prepositus hypoglossi and adjacent medial vestibular nucleus during horizontal eye movement in behaving macaques. J Neurophysiol 68: 319-332, 1992.

Mehring C, Rickert J, Vaadia E, Cardosa de Oliveira S, Aertsen A, and Rotter S. Inference of hand movements from local field potentials in monkey motor cortex. Nat Neurosci 6: 1253-1254, 2003.

Melis BJ, and van Gisbergen JA. Short-term adaptation of electrically induced saccades in monkey superior colliculus. J Neurophysiol 76: 1744-1758, 1996.

Miller JM, and Robins D. Extraocular muscle forces in alert monkey. Vision Res 32: 1099-1113, 1992. Mimeault D, Paquet V, Molotchnikoff S, Lepore F, and Guillemot JP. Disparity sensitivity in the superior colliculus of the cat. Brain Res 1010: 87-94, 2004.

Missal M, de Brouwer S, Lefevre P, and Olivier E. Activity of mesencephalic vertical burst neurons during saccades and smooth pursuit. J Neurophysiol 83: 2080-2092, 2000.

Mitzdorf U. Current source-density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomena. Physiol Rev 65: 37-100, 1985.

Mitzdorf U. Physiological sources of evoked potentials. Electroencephalogr Clin Neurophysiol Suppl 42: 47-57, 1991.

Mitzdorf U. Properties of the evoked potential generators: current source-density analysis of visually evoked potentials in the cat cortex. Int J Neurosci 33: 33-59, 1987.

Moschovakis AK, and Highstein SM. The anatomy and physiology of primate neurons that control rapid eye movements. Annu Rev Neurosci 17: 465-488, 1994.

Moschovakis AK, Karabelas AB, and Highstein SM. Structure-function relationships in the primate superior colliculus. II. Morphological identity of presaccadic neurons. J Neurophysiol 60: 263-302, 1988.

Moschovakis AK, Scudder CA, and Highstein SM. Structure of the primate oculomotor burst generator. I. Medium-lead burst neurons with upward on-directions. J Neurophysiol 65: 203-217, 1991a.

Moschovakis AK, Scudder CA, Highstein SM, and Warren JD. Structure of the primate oculomotor burst generator. II. Medium-lead burst neurons with downward on-directions. J Neurophysiol 65: 218-229, 1991b.

266

Mukovski M, Chauvette S, Timofeev I, and Volgushev M. Detection of active and silent states in neocortical neurons from the field potential signal during slow-wave sleep. Cereb Cortex 17: 400-414, 2007.

Munoz DP, and Wurtz RH. Fixation cells in monkey superior colliculus. I. Characteristics of cell discharge. J Neurophysiol 70: 559-575, 1993.

Nakao S, Curthoys IS, and Markham CH. Direct inhibitory projection of pause neurons to nystagmus-related pontomedullary reticular burst neurons in the cat. Exp Brain Res 40: 283-293, 1980.

Nelson MJ, Pouget P, Nilsen EA, Patten CD, and Schall JD. Review of signal distortion through metal microelectrode recording circuits and filters. Journal of neuroscience methods 169: 141- 157, 2008.

Noda H, Sugita S, and Ikeda Y. Afferent and efferent connections of the oculomotor region of the fastigial nucleus in the macaque monkey. J Comp Neurol 302: 330-348, 1990.

O'Leary JG, and Hatsopoulos NG. Early visuomotor representations revealed from evoked local field potentials in motor and premotor cortical areas. J Neurophysiol 96: 1492-1506, 2006.

Ohgaki T, Markham CH, Schneider JS, and Curthoys IS. Anatomical evidence of the projection of pontine omnipause neurons to midbrain regions controlling vertical eye movements. J Comp Neurol 289: 610-625, 1989.

Ohtsuka K, Maeda S, and Oguri N. Accommodation and convergence palsy caused by lesions in the bilateral rostral superior colliculus. American journal of ophthalmology 133: 425-427, 2002.

Ohtsuka K, and Nagasaka Y. Divergent axon collaterals from the rostral superior colliculus to the pretectal accommodation-related areas and the omnipause neuron area in the cat. J Comp Neurol 413: 68-76, 1999.

Ono H, Nakamizo S, and Steinbach MJ. Nonadditivity of vergence and saccadic eye movement. Vision Res 18: 735-739, 1978.

Oohira A. Vergence eye movements facilitated by saccades. Jpn J Ophthalmol 37: 400-413, 1993.

Paul K, and Gnadt JW. Activity of omnipause neurons during "staircase saccades" elicited by persistent microstimulation of the superior colliculus. Vision Res 46: 3430-3442, 2006.

Pesaran B, Pezaris JS, Sahani M, Mitra PP, and Andersen RA. Temporal structure in neuronal activity during working memory in macaque parietal cortex. Nat Neurosci 5: 805-811, 2002.

267

Ramat S, Somers JT, Das VE, and Leigh RJ. Conjugate ocular oscillations during shifts of the direction and depth of visual fixation. Invest Ophthalmol Vis Sci 40: 1681-1686, 1999.

Rambold H, Neumann G, Sander T, and Helmchen C. Pontine lesions may cause selective deficits of "slow" vergence eye movements. Ann N Y Acad Sci 1039: 567-570, 2005.

Rasch M, Logothetis NK, and Kreiman G. From neurons to circuits: linear estimation of local field potentials. J Neurosci 29: 13785-13796, 2009.

Rashbass C, and Westheimer G. Disjunctive eye movements. J Physiol 159: 339-360, 1961a.

Rashbass C, and Westheimer G. Independence of conjugate and disjunctive eye movements. J Physiol 159: 361-364, 1961b.

Reyes-Puerta V, Philipp R, Lindner W, and Hoffmann KP. Role of the rostral superior colliculus in gaze anchoring during reach movements. J Neurophysiol 103: 3153-3166.

Rieke F, Warland D, Steveninck RR, and Bialek W. Spikes: exploring the neural code. Cambridge, MA:MIT, 1996.

Robinson DA. The Mechanics of Human Saccadic Eye Movement. J Physiol 174: 245-264, 1964.

Robinson DA. Models of the saccadic eye movement control system. Kybernetik 14: 71-83, 1973.

Robinson DA. Oculomotor unit behavior in the monkey. J Neurophysiol 33: 393-403, 1970.

Robinson DA, and Keller EL. The behavior of eye movement motoneurons in the alert monkey. Bibl Ophthalmol 82: 7-16, 1972.

Roy JE, and Cullen KE. Brain stem pursuit pathways: dissociating visual, vestibular, and proprioceptive inputs during combined eye-head gaze tracking. J Neurophysiol 90: 271-290, 2003.

Roy JE, and Cullen KE. Vestibuloocular reflex signal modulation during voluntary and passive head movements. J Neurophysiol 87: 2337-2357, 2002.

Sadeghi SG, Chacron MJ, Taylor MC, and Cullen KE. Neural variability, detection thresholds, and information transmission in the vestibular system. J Neurosci 27: 771-781, 2007.

Sasaki S, and Shimazu H. Reticulovestibular organization participating in generation of horizontal fast eye movement. Ann N Y Acad Sci 374: 130-143, 1981.

268

Scherberger H, Jarvis MR, and Andersen RA. Cortical local field potential encodes movement intentions in the posterior parietal cortex. Neuron 46: 347-354, 2005.

Schiller PH. The discharge characteristics of single units in the oculomotor and abducens nuclei of the unanesthetized monkey. Exp Brain Res 10: 347-362, 1970.

Schiller PH, and Stryker M. Single-unit recording and stimulation in superior colliculus of the alert rhesus monkey. J Neurophysiol 35: 915-924, 1972.

Schwartz G. Estimating the dimension of a Model. The Annals of Statistics 5: 461-464, 1978. Scudder CA. A new local feedback model of the saccadic burst generator. J Neurophysiol 59: 1455-1475, 1988.

Scudder CA, and Fuchs AF. Physiological and behavioural identification of vestibular nucleus neurons mediating the horizontal vestibuloocular reflex in trained rhesus monkeys. . J Neurophysiol 68: 244-264, 1992.

Scudder CA, Fuchs AF, and Langer TP. Characteristics and functional identification of saccadic inhibitory burst neurons in the alert monkey. J Neurophysiol 59: 1430-1454, 1988.

Scudder CA, Kaneko CS, and Fuchs AF. The brainstem burst generator for saccadic eye movements: a modern synthesis. Exp Brain Res 142: 439-462, 2002.

Scudder CA, Moschovakis AK, Karabelas AB, and Highstein SM. Anatomy and physiology of saccadic long-lead burst neurons recorded in the alert squirrel monkey. I. Descending projections from the mesencephalon. J Neurophysiol 76: 332-352, 1996.

Shinoda Y, Sugiuchi Y, Izawa Y, and Takahashi M. Neural circuits for triggering saccades in the brainstem. Progress in brain research 171: 79-85, 2008.

Sokal RR, and Rohlf FJ. Biometry: The Principles and Practice of Statistics in Biological Research. New York: Freeman, 1995.

Sparks DL. The brainstem control of saccadic eye movements. Nature reviews 3: 952-964, 2002. Spencer RF, and Porter JD. Structural organization of the extraocular muscles. Rev Oculomot Res 2: 33-79, 1988.

Stanford TR, Freedman EG, and Sparks DL. Site and parameters of microstimulation: evidence for independent effects on the properties of saccades evoked from the primate superior colliculus. J Neurophysiol 76: 3360-3381, 1996.

Steiger HJ, and Buttner-Ennever J. Relationship between motoneurons and internuclear neurons in the abducens nucleus: a double retrograde tracer study in the cat. Brain Res 148: 181-188, 1978.

269

Strassman A, Evinger C, McCrea RA, Baker RG, and Highstein SM. Anatomy and physiology of intracellularly labelled omnipause neurons in the cat and squirrel monkey. Exp Brain Res 67: 436-440, 1987.

Strassman A, Highstein SM, and McCrea RA. Anatomy and physiology of saccadic burst neurons in the alert squirrel monkey. I. Excitatory burst neurons. J Comp Neurol 249: 337-357, 1986a.

Strassman A, Highstein SM, and McCrea RA. Anatomy and physiology of saccadic burst neurons in the alert squirrel monkey. II. Inhibitory burst neurons. J Comp Neurol 249: 358-380, 1986b.

Suzuki S, Suzuki Y, and Ohtsuka K. Convergence eye movements evoked by microstimulation of the rostral superior colliculus in the cat. Neuroscience research 49: 39-45, 2004.

Sylvestre PA, Choi JT, and Cullen KE. Discharge dynamics of oculomotor neural integrator neurons during conjugate and disjunctive saccades and fixation. J Neurophysiol 90: 739-754, 2003.

Sylvestre PA, and Cullen KE. Dynamics of abducens nucleus neuron discharges during disjunctive saccades. J Neurophysiol 88: 3452-3468, 2002.

Sylvestre PA, and Cullen KE. Premotor correlates of integrated feedback control for eye-head gaze shifts. J Neurosci 26: 4922-4929, 2006.

Sylvestre PA, and Cullen KE. Quantitative analysis of abducens neuron discharge dynamics during saccadic and slow eye movements. J Neurophysiol 82: 2612-2632, 1999.

Sylvestre PA, Galiana HL, and Cullen KE. Conjugate and vergence oscillations during saccades and gaze shifts: implications for integrated control of binocular movement. J Neurophysiol 87: 257-272, 2002.

Takahashi M, Sugiuchi Y, Izawa Y, and Shinoda Y. Commissural excitation and inhibition by the superior colliculus in tectoreticular neurons projecting to omnipause neuron and inhibitory burst neuron regions. J Neurophysiol 94: 1707-1726, 2005.

Tanaka NK, Ito K, and Stopfer M. Odor-evoked neural oscillations in Drosophila are mediated by widely branching interneurons. J Neurosci 29: 8595-8603, 2009.

Toda H, Takagi M, Yoshizawa T, and Bando T. Disjunctive eye movement evoked by microstimulation in an extrastriate cortical area of the cat. Neuroscience research 12: 300-306, 1991.

270

Tychsen L, Richards M, Wong A, Foeller P, Burhkalter A, Narasimhan A, and Demer J. Spectrum of infantile esotropia in primates: Behavior, brains, and orbits. J Aapos 12: 375-380, 2008.

Ugolini G, Klam F, Doldan Dans M, Dubayle D, Brandi AM, Buttner-Ennever J, and Graf W. Horizontal eye movement networks in primates as revealed by retrograde transneuronal transfer of rabies virus: differences in monosynaptic input to "slow" and "fast" abducens motoneurons. J Comp Neurol 498: 762-785, 2006.

Van Gisbergen JA, Robinson DA, and Gielen S. A quantitative analysis of generation of saccadic eye movements by burst neurons. J Neurophysiol 45: 417-442, 1981.

Van Horn M, Sylvestre PA, and Cullen KE. The Brain Stem Saccadic Burst Generator Encodes Gaze in Three-Dimensial Space. J Neurophysiol 99: 2602-2616, 2008.

Van Horn MR, and Cullen KE. Dynamic characterization of agonist and antagonist oculomotoneurons during conjugate and disconjugate eye movements. J Neurophysiol 2009.

Van Horn MR, and Cullen KE. Dynamic coding of vertical facilitated vergence by premotor saccadic burst neurons. J Neurophysiol 100: 1967-1982, 2008. van Leeuwen AF, Collewijn H, and Erkelens CJ. Dynamics of horizontal vergence movements: interaction with horizontal and vertical saccades and relation with monocular preferences. Vision Res 38: 3943-3954, 1998.

Van Opstal A, Van Gisbergen J, and Eggermont J. Reconstruction of neural control signals for saccades based on an inverse method. Vision Res 25: 789-801, 1985. von Helmholtz H. Helmholtz's Treatise on Physiological Optics. New York: Dover Publications, Inc., 1962.

Wackerly DD, Mendenhall W, and Scheaffer RL. Mathematical statistics with applications. Belmont, Calif.: Duxbury Press, 1996, p. xvi, 798 p.

Waitzman DM, Van Horn MR, and Cullen KE. Soc Neurosci Abstr 33: 2007.

Waitzman DM, Van Horn MR, and Cullen KE. Neuronal evidence for individual eye control in the primate cMRF. Prog Brain Res 171: 143-150, 2008.

Walton MM, and Mays LE. Discharge of saccade-related superior colliculus neurons during saccades accompanied by vergence. J Neurophysiol 90: 1124-1139, 2003.

Wasicky R, Horn AK, and Buttner-Ennever JA. Twitch and nontwitch motoneuron subgroups in the oculomotor nucleus of monkeys receive different afferent projections. J Comp Neurol 479: 117-129, 2004.

271

Wehr M, and Laurent G. Relationship between afferent and central temporal patterns in the locust olfactory system. J Neurosci 19: 381-390, 1999.

Wessel R, Koch C, and Gabbiani F. Coding of time-varying electric field amplitude modulations in a wave-type electric fish. J Neurophysiol 75: 2280-2293, 1996.

Westheimer G, and Mitchell AM. Eye movement responses to convergence stimuli. AMA 55: 848-856, 1956.

Yoshida K, Iwamoto Y, Chimoto S, and Shimazu H. Saccade-related inhibitory input to pontine omnipause neurons: an intracellular study in alert cats. J Neurophysiol 82: 1198-1208, 1999.

Yoshida K, McCrea R, Berthoz A, and Vidal PP. Morphological and physiological characteristics of inhibitory burst neurons controlling horizontal rapid eye movements in the alert cat. J Neurophysiol 48: 761-784, 1982.

Zee DS, Fitzgibbon EJ, and Optican LM. Saccade-vergence interactions in humans. J Neurophysiol 68: 1624-1641, 1992.

Zhang H, and Gamlin PD. Neurons in the posterior interposed nucleus of the cerebellum related to vergence and accommodation. I. Steady-state characteristics. J Neurophysiol 79: 1255-1269, 1998.

Zhang Y, Gamlin PD, and Mays LE. Antidromic identification of midbrain near response cells projecting to the oculomotor nucleus. Exp Brain Res 84: 525-528, 1991.

Zhang Y, Mays LE, and Gamlin PD. Characteristics of near response cells projecting to the oculomotor nucleus. J Neurophysiol 67: 944-960, 1992.

Zhou W, and King WM. Ocular selectivity of units in oculomotor pathways. Ann N Y Acad Sci 781: 724-728, 1996.

Zhou W, and King WM. Premotor commands encode monocular eye movements. Nature 393: 692-695, 1998.

Zuber B, Semmlow J, and Stark L. Frequency characteristics of the saccadic eye movement. Biophysical J 8: 1288-1298, 1968.

272