Brain circuits underlying sensorimotor transformations

BRAIN CIRCUITS UNDERLYING SENSORIMOTOR TRANSFORMATIONS by

TED K. DOYKOS

BA, University of Colorado at Boulder, 2008

A thesis submitted to the Faculty of the Graduate School of the University of Colorado in partial fulfillment

of the requirements for the degree of Doctor of Philosophy

Neuroscience Program

2020

This thesis for the Doctor of Philosophy degree by Ted K. Doykos

has been approved for the Neuroscience Program

by

Nathan Schoppa, Chair Gidon Felsen, Advisor

Abigail Person Diego Restrepo

Daniel Tollin Tim Lei

Date: May 15, 2020

Doykos, Ted K. (PhD, Neuroscience Program)

Brain Circuits Underlying Sensorimotor Transformations Thesis directed by Associate Professor Gidon Felsen

ABSTRACT

Sensorimotor transformations are orchestrated across many brain regions to produce adaptive behavior. A chief goal of systems neuroscience is to understand the unique

computations and contributions made within this network. This thesis examines the sensorimotor transformations performed by two crucial nodes of the motor system: the cerebellum and

superior colliculus (SC).

The cerebellum is essential for adaptive learning and predictive motor control.

Theoretical work highlighting the structure’s pattern discrimination capacity motivated our analysis of how afferent convergence constructs sensorimotor representations in the cerebellum.

We found that optogenetic activation of an int racerebellar “nucleocortical” (NC) feedback pathway bidirectionally modulates cerebellar granule cell sensory-evoked response magnitude and timing. We additionally observed that activation of NC and sensory pathways produces a balance between mutual excitation and competing excitation and inhibition in granule cells, and NC feedback delays Purkinje cell sensory-evoked responses. These results are important for understanding how sensory and motor information become integrated by the cerebellar circuit to facilitate learning.

The SC is critical for spatial attention and orienting behavior. Its internal circuity is

composed of intermixed cell types highly interconnected with one another. Its intermediate and

deep layers (SC id ) are also densely innervated by a network of brain regions involved in sensory

and motor functions. While it is established that SC id contains a topographically organized map

of movement space, the underlying computations that give rise to orienting behaviors is less well understood. We addressed this by employing monosynaptic tracing from distinct neuronal subtypes within the SC id and identified thematic input patterns to the structure. We found that excitatory SC neurons (eSCNs) receive stronger inputs from a larger number of areas than do inhibitory SC neurons (iSCNs), a subset of brainstem-projecting eSCNs were targeted by many fewer brain areas than the general population of eSCNs, and populations of commissurally connected SC neurons were located in similar rostrocaudal positions. These findings support the view that active intrinsic processes within the SC id give rise to orienting behavior.

This thesis examined how the unique sensorimotor transformations performed by the cerebellum and superior colliculus arise from their distinct anatomy and interconnectivity with other brain structures.

The form and content of this abstract are approved. I recommend its publication.

Approved: Gidon Felsen

ACKNOWLEDGEMENTS

I am infinitely grateful for my experience in the Neuroscience Graduate Program, especially for the mentorship I received from Abby Person and Gidon Felsen. I found Abby and Gidon to be complementary in many ways and feel that my growth as a scientific thinker and as a person would no doubt be impoverished if not for the time I spent in their labs. I am thankful to Abby for her relentless passion, creativity, mental sharpness, and extreme eagerness to engage in scientific discussions at any moment. I am thankful to Gidon for his calm and imperturbable demeanor, cautious attention to detail, and his methodical and analytical approach to science. I am thankful to each of them for their reliable willingness to help me with any issue that might arise, making me a better writer, and their caring support. I am grateful for the daily interactions and support I’ve received from members of both labs. I am thankful for the numerous

discussions I had with every member of my thesis committee and the director of the

Neuroscience Graduate Program. I also want to thank the many program and departmental

administrators that have helped me along the way. I am thankful for the many friends I have

made in the Graduate School. Finally, I want to thank my Mom and Dad for their constant love

and support.

TABLE OF CONTENTS CHAPTER

I. INTRODUCTION………..……….1

Motivation………..………1

Overview of Thesis………1

Sensorimotor Transformations………...2

The neural basis of sensorimotor transformations………...2

Contextual modulation of sensorimotor transformations………3

Cerebellum……….3

Cerebellar anatomy………..4

Gross cerebellar anatomy……….5

The canonical cerebellar circuit………...5

Granule cell layer anatomy………..7

Associative learning in the cerebellum………8

Temporal representations in the granule cell layer………..9

Superior colliculus………..9

Superior colliculus anatomy………..11

Gross superior colliculus anatomy………11

Inputs to the superior colliculus……….11

Intrinsic processing in the superior colliculus………...12

II. SENSORIMOTOR INTEGRATION IN THE CEREEBELLAR GRANULE

CELL LAYER………13

Introduction………..13

Acknowledgements……….…….16

Materials and Methods……….16

Animals………..16

Surgery and opsin expression………16

In vivo electrophysiology, optogenetics, and sensory stimulation………....17

Spike-sorting and cell-type identification………..18

Histological analysis of opsin expression………..18

Data analysis………..20

Stimulus responsivity……….20

Linearity analysis………...21

Results………..22

In vivo electrophysiology and cell- type identification………..22

Granule cell responsivity………...22

Granule cell population responses……….25

Granule cell response timing shifts………28

Purkinje cell response timing shifts………...28

Discussion………30

Limitations……….30

Implications………31

III. MONOSYNAPTIC INPUTS TO SUPERIOR COLLICULUS ……….35

Introduction………..35

Acknowledgements………..37

Materials and Methods……….38

Animals………..38

Viral injections………...38

Tissue preparation and imaging……….40

Starter neuron quantification and brain area classification………41

Input neuron quantification and brain area classification………..42

Starter and input neuron analyses………..43

Statistical comparison of projection strength……….44

Results………..44

Rabies expression and identification of starter neurons………44

Extrinsic inputs to eSCNs and iSCNs………....45

Laterality of inputs to eSCNs and iSCNs………..51

Relationship between projection strength and rsotrocaudal position of starter neurons………...53

Extrinsic inputs to subset of tectofugal eSCNs………..53

Layer- specific targeting of contralateral superior colliculus……….55

Discussion………..58

IV. Discussion………...………64

Summary………..64

Cerebellum……….64

Superior colliculus……….67

Sensorimotor Transformations in Cerebellum Versus Superior Colliculus…………69

Interactions Between Cerebellum and Superior Colliculus……….71

The saccadic system………...71

Saccade characteristics………...71

Saccadic control……….72

REFERENCES………...………..75

LIST OF TABLES TABLE

3.1 Anatomical abbreviations used in Figures 3.4 and 3. 5………...49-50

LIST OF FIGURES FIGURE

1.1 Cerebellar circuit diagram ……….6

2.1 Nucleocortical fibers………15

2.2 Experimental methodology………..19

2.3 Superlinear GrC responses to whisker and NC stimulation……….23

2.4 Sublinear GrC responses to whisker and NC stimulation………24

2.5 GrC population responses and convergence patterns………..26

2.6 NC activation increases PC response latency………..29

3.1 Rabies expression and identification of starter neurons………..39

3.2 Control injections into the SC id ………46

3.3 eSCNs receive more extrinsic inputs than iSCNs………47

3.4 Pr ojection strength to eSCNs and iSCNs……….48

3.5 Laterality of SC inputs……….52

3.6 Extrinsic inputs favor the rostral SC………54

3.7 CTRNs receive fewer inputs than eSCNs………56

3.8 Layer- specific targeting of contralateral SC………....57

LIST OF ABBREVIATIONS ABBREVIATION

ChR2……….………...….channelrhodopsin

CS………...conditioned stimulus

CTRN………..…...crossed tecto-reticular neuron

DV………..dorsoventral

eSCN………..………….excitatory superior colliculus neuron

FEF………...frontal eye field

GFP………..………..green fluorescent protein

GrC………..granule cell

iSCN….………..………….inhibitory superior colliculus neuron

M2………...…….secondary motor cortex

mGluR2………...metabotropic glutamate receptor 2

ML………..………mediolateral

MPRF………..medial pontine reticular formation

NA………....numerical aperture

NC………...nucleocortical

oG……….optimized glycoprotein

OT………..optic tectum

PC………..Purkinje cell

PSTH………peri-stimulus time histogram

SC…...………..superior colliculus

SC id ………...………superior colliculus intermediate/deep layers

SNr………...…substantia nigra pars reticulata

SSN………semantic segmentation artificial neural network

TVA……….tumor virus A

US……….……unconditioned stimulus

V1………...….primary visual cortex

VPM……….………..ventral posteriomedial thalamic nucleus

CHAPTER I INTRODUCTION

Motivation

Animals have evolved to transform sensory inputs into ethologically relevant motor outputs. This process is best carried out by using the full range of information available at the time of movement. Voluntary movement requires multistage sensorimotor transformations across a network of brain regions. A chief goal of systems neuroscience is to understand how the

nervous system implements sensorimotor transformations to produce behavior.

I investigated the functional and structural components of sensorimotor transformations within the cerebellum and superior colliculus (SC). The cerebellum is a hindbrain structure whose excitatory projections to midbrain and thalamic regions fine-tune motor control. The SC is a midbrain structure influenced by cerebellar inputs involved in orienting to spatial targets via excitatory projections to premotor nuclei in the brainstem and spinal cord. I aimed to understand the unique nature of their neural computations since these regions are critical for motor behavior and receive widespread sensory input.

Overview of Thesis

This thesis will discuss 1) how afferent convergence constructs sensorimotor

representations in the cerebellum and 2) the anatomical convergence of sensorimotor inputs onto excitatory and inhibitory neurons in the SC. Chapter one discusses sensorimotor transformations, how they are studied, and presents an overview of the cerebellum and SC as sensorimotor hubs of neural activity. Chapters two and three present the novel data contributed by this thesis.

Chapter two presents the results of neural modulation of sensory processing in cerebellar granule

cells, while chapter three examines monosynaptic inputs onto excitatory and inhibitory cells in the SC along a functional rostrocaudal axis. Chapter four summarizes the results and presents a framework through which cerebellum and SC may guide precise orienting movements.

Sensorimotor Transformations The neural basis of sensorimotor transformations

Animal behavior is fundamentally the transformation of sensory input into a coherent motor output (Gomez-Marin and Ghazanfar, 2019). Individual neurons are thought to give rise to this process by integrating activity of inputs to produce an output. Thus, proximity to sense organs and muscles largely determines how closely their current activity relates to preceding sensory inputs and upcoming movements. The simplest form of sensorimotor transformations produces the simplest behaviors, for example, spinal reflexes that stabilize joints and produce rapid postural adjustments. These sensorimotor transformations are mediated by a disynaptic loop between primary sensory afferents, spinal interneurons, and motor neurons (Kandel et al., 2013). The simplicity of this anatomical arrangement ensures a rapid and reliable response.

While a relatively small number of neurons performing a single sensorimotor transformation can

explain such simple behaviors, all neurons in the nervous system are fundamentally involved in

more elaborate sensorimotor transformations that evolve over a range of exponentially longer

timescales. Thus, more complex sensorimotor transformations are orchestrated by neurons acting

within a network of interconnected brain areas. Such processes typically involve multiple sense

modalities, are affected by prior experiences, and engage multiple muscle groups. Examples

cover a range of disparate behaviors, such as swerving to avoid traffic, to deciding on which

college to attend.

Contextual modulation of sensorimotor transformations

It is important to consider the role of contextual factors such as prior experience and current goals when interrogating neural substrates of animal behavior because sensorimotor transformations occur within a larger feedback loop with the environment (Gomez-Marin and Ghazanfar, 2019). For example, behavioral preferences acquired prior to experimental testing may engage other untargeted brain regions, complicating interpretations of the neural activity that is necessary and sufficient for a given behavior (Gomez-Marin, 2017). Similar complications may arise from goal-dependent changes in network activity, as the goals of a behaving animal are related to its motivational and arousal state, which are themselves physiologically coupled to neuromodulatory activity in well-defined brain circuits (Kandel et al., 2013). Therefore,

accounting for the range relevant to contextual information is critical when studying sensorimotor transformations at a behavioral level.

Cerebellum

The cerebellum is a key brain structure involved in sensorimotor transformations. The

cerebellum receives widespread sensory and motor input and sends excitatory projections to

premotor areas (Eccles et al., 1967), acting to enhance movement precision and enable fine

motor control (Holmes 1917). Executing movements with a high degree of precision requires

producing well-timed activity patterns, particularly around movement endpoints. Thus, one

perspective is that the cerebellum mediates movement precision by controlling the timing of

movement cessation (Goffart et al., 1998b). The cerebellum regulates movement precision for a

variety of behaviors by learning associations between sensory and motor contingencies to

produce adaptive motor responses (Ito, 1972; Ohyama et al., 2003).

Disease or damage to the cerebellum highlights its role in normal motor control while also indicating its involvement in other processes. Movement in cerebellar patients is

accompanied by tremor-like oscillations characterized by a lack of endpoint precision (Holmes, 1917; Bonnefoi-Kyriacou et al., 1998). These symptoms suggest a critical role for the cerebellum in regulating the timing of movement cessation. In addition to these symptoms of ataxia,

cerebellar patients also exhibit severe deficiencies during visuomotor adaptations tasks (Martin et al., 1996) or when asked to generate a rhythmic tapping motion (Spencer et al., 2003). Finally, cerebellar patients have also been shown to exhibit abnormalities in perception (Parsons et al., 1997; Baumann et al., 2015) and cognition (Ito, 1993; Koziol et al., 2013). The diversity of symptoms associated with cerebellar damage likely reflects its role in processing neural information arising from many brain and spinal cord areas.

Cerebellar anatomy

Several features of cerebellar anatomy have led to ideas pertaining to its role in motor control. The convergence of inputs onto a massive population of neurons is consistent with a role in pattern separation (Marr, 1969; Albus, 1971), while the regularity in cerebellar

cytoarchitecture and connectivity patterns (Eccles et al., 1967; Palay and Chan-Palay, 1974) suggest an underlying “universal transform” common to all cerebellar-dependent behaviors (Schmahmann, 2004). Additionally, the unique innervation of cerebellar Purkinje cells (PCs) by a single climbing fiber and many parallel fibers (Eccles et al., 1967) promoted theories of plasticity-dependent pattern storage supervised by the cerebrum (Marr, 1969; Albus, 1971).

These ideas have undergone extensive theoretical (Mauk and Buonomano, 2004; Kalmbach et

al., 2011; Gilmer and Person, 2017) and experimental (Ito, 1982; Linden et al., 1991; Steuber et

al., 2007) examination, and while these theories and subsequent work have contributed

significantly to our understanding of how the cerebellum mediates motor control, there are still unanswered questions.

Gross cerebellar anatomy

The cerebellum is composed of two primary structures; a cortical input structure and three sets of underlying nuclei that constitute an output structure. The cerebellar cortex is trilaminar, consisting of an input granule cell (GrC) layer that receives widespread excitatory input from spinal cord, brainstem, and corticopontine sources, a neurite-dense molecular layer involved in synaptic processing, and an output layer of PCs that inhibit cerebellar nuclear neurons (Fig. 1.1). Finally, cerebellar nuclei convey outputs of cerebellar computations to other brain areas via excitatory projections to motor areas of the thalamus and midbrain.

The canonical cerebellar circuit

The GrC layer is occupied predominantly by excitatory GrCs and by a much smaller percentage of inhibitory Golgi cells, both of which sample a similar array of diverse inputs (Eccles et al., 1967). GrCs additionally receive feedback and feed-forward inhibition from Golgi cells and send parallel fibers to the molecular layer that synapse onto the dendrites of PC (Eccles et al., 1967). Each PC has an elaborate dendritic arborization on which it is contacted by

hundreds of thousands of GrC parallel fibers and by a single climbing fiber arising from the

inferior olive (Eccles et al., 1967). A PC’s olivary input is excitatory and consists of hundreds of

individual synapses (Eccles et al., 1967) that, when active, drive a sufficiently large calcium

signal in the PC to induce long-term depression at the parallel fiber-PC synapses active

Figure 1.1. Cerebellar circuit diagram (Kandel et al., 2000)

immediately prior to the calcium signal (Ito, 1982). This plasticity mechanism is thought to underlie most, but not all (Carey, 2011), forms of cerebellar-dependent learning (Marr, 1969;

Albus, 1971; Ito, 1982).

Granule cell layer anatomy

The cerebellum is the most highly folded structure in the brain. Its foliations greatly expand its surface area and facilitate the tight packing of over half the brain’s neurons into the cerebellar GrC layer (De Schutter and Bjaalie, 2001; Herculano-Houzel and Lent, 2005), which receives input from dozens of distinct regions. GrCs have on average only four short dendrites, each of which receives a single extracerebellar excitatory input (Eccles et al., 1967) and at least one inhibitory input from a neighboring Golgi cell (Eccles et al., 1967; Tabuchi et al., 2019).

Ascending inputs to the cerebellum arise from spinal cord (Voogd et al., 1969; Lakke et al., 1986; Yaginuma and Matsushita, 1987; Verburgh et al., 1989), brainstem (Russchen et al., 1976;

Saigal et al., 1982), and cranial ganglia (Saigal et al., 1980a; b), and carry proprioceptive and cutaneous information (Garwicz et al., 1998), while descending corticopontine inputs (Freedman et al., 1975; Glickstein et al., 1994; Serapide et al., 2001) convey a mixture of processed sensory and motor information (Proville et al., 2014). Cerebellar inputs are both divergent and

convergent; individual cells can ramify broadly across widespread regions of cerebellum (Eccles et al., 1967), while distinct brain areas converge in the same regions of cerebellum (Wu et al., 1999; Shinoda et al., 2000; Houck and Person, 2015). This organization allows identical

information to be widely represented and for large populations of GrCs to represent a multitude

of unique combinations of information. Theoretical work suggests this anatomical arrangement is

well suited to represent the unique features of highly overlapping input patterns, functioning as a pattern separator (Marr, 1969; Albus, 1971).

Associative learning in the cerebellum

The importance of the cerebellum in associative learning is well illustrated by Pavlovian classical conditioning studies where the cerebellum is lesioned. A typical paradigm used in such studies pairs a neutral sensory cue (the “conditioned stimulus” [CS]) with an aversive puff of air to the eye (the “unconditioned stimulus” [US]). After repeated pairings, the CS induces an eyelid closure in the absence of the US. Lesions to cerebellum prevent the acquisition of learned

responses without impairing the underlying eyeblink behavior (Yeo et al., 1985), indicating the cerebellum’s importance in learning sensorimotor associations. Furthermore, electrical

stimulation of mossy fiber input pathways functions as an alternative CS (Steinmetz et al., 1989).

This indicates that contextual sensory information used to modulate motor responses arises from

the GrC layer. Additional eyeblink studies in which the inferior olive was lesioned abolish both

the acquisition and the extinction of learned responses (Medina et al., 2002), suggesting that the

olivary climbing fiber input to the cerebellum provide instructional signals that facilitate adaptive

motor responses. The neural substrate of such classically conditioned responses is thought to be

the climbing fiber-mediated long-term depression induced at GrC parallel fiber-to-PC synapses

(Ito, 1982). While possibly a key mechanism in cerebellar-dependent learning, it does not

address representations of temporal information critical for generating well-timed movements,

including eyeblinks. This form of associative learning highlights a key feature of sensorimotor

transformations by the brain, specifically that they are highly plastic. Thus, the brain maintains

the capacity to arbitrarily assign reflexive motor output to sensory cues when behaviorally adaptive.

Temporal representations in the granule cell layer

Understanding how timing is represented in the cerebellum is central to understanding its role in motor control. An open question from conditioned eyeblink studies asks how well-timed responses are triggered by a time invariant stimulus like a pure tone lasting for hundreds of milliseconds. In other words, how do animals learn the ideal amount of time between the onset of a CS and the onset of their response? Modeling work suggests GrC layer inhibition likely serves this important function (Mauk and Donegan, 1997; Mauk and Buonomano, 2004; Kalmbach et al., 2011). While a subset of GrCs will discharge in response to a sensory cue (Chadderton et al., 2004; Jörntell and Ekerot, 2006; Arenz et al., 2008; Barmack and Yakhnitsa, 2008; Duguid et al., 2012; Ishikawa et al., 2015), they receive varying levels of tonic and phasic inhibition (Brickley et al., 1996), and therefore are expected to be active at different times during a sensory stimulus.

In this way it is thought that a population of active GrCs representing a particular time point with respect to a stimulus can be decoded by PCs and used to learn an appropriately timed response (Marr, 1969; Albus, 1971; Ito, 1972). Experimental verification of this hypothesis is still lacking;

however, experiments I discuss in chapter II address how stimulating convergent inputs pathways affect GrC response magnitude and timing.

Superior colliculus

The superior colliculus (SC) is a fundamental structure involved in sensorimotor

transformations. It receives widespread sensory and motor input (Sparks and Hartwich-Young,

1989) and sends excitatory projections to premotor areas (Sparks and Hartwich-Young, 1989) where it regulates orienting behaviors (Basso and May, 2017).

Many studies have examined the role of the SC in primates making saccades to visual targets (Goldberg and Wurtz, 1972a; b; Wurtz and Goldberg, 1972; Lee et al., 1988); however, other work across a broader range of species demonstrates a more general role of the SC in other orienting behaviors (Sparks, 1999). For example, the SC (or the optic tectum (OT) the

nonmammalian homologue of the SC) encodes orienting movements of the head (Meyer and Sperry, 1973; du Lac and Knudsen, 1991; Freedman et al., 1996; Guillaume and Pélisson, 2001;

Valentine et al., 2002), limbs (Werner et al., 1997; Courjon et al., 2004; Steinmetz et al., 2018), and trunk (Herrero et al., 1998; Felsen and Mainen, 2008), and is also involved in producing escape behavior away from aversive stimuli (Dean et al., 1986, 1989; Sahibzada et al., 1986;

Evans et al., 2018) . In addition to the SC’s role in producing orienting movement toward or away from a stimulus, the SC is also important for higher-order cognitive functions such as decision making (Thompson et al., 2016; Crapse et al., 2018), action selection (Wolf et al., 2015), and spatial attention (Krauzlis et al., 2013). The large range of functions mediated by the SC speaks to its role in processing neural information from diverse brain and spinal cord areas.

Furthermore, it highlights the increased elaboration of sensorimotor transformations in diverse

regions: whereas the spinal cord contains ‘hard wired’ sensorimotor circuits and the cerebellum

can form plastic associations, the SC uses diverse evidence (sensory) accumulation processes to

decide on an action plan.

Superior colliculus anatomy Gross superior colliculus anatomy

The SC is a laminar structure typically divided into two functionally distinct zones:

superficial layers, and intermediate/deep layers (SC id ). The superficial layers process visual information via direct projections from the retina and descending projections from the neocortex, while SC id receives widespread sensory and motor inputs and is a source of motor commands to premotor centers in the brainstem and spinal cord.

Inputs to the superior colliculus

Inputs to SC id arise from cerebral cortex (Garey et al., 1968; Edwards et al., 1979; Fries, 1984), thalamic areas (Edwards et al., 1974, 1979; Graybiel, 1974; Grofová et al., 1978),

cerebellar nuclei (Batton et al., 1977; Kawamura et al., 1982), and several mesencephalic regions (Hopkins and Niessen, 1976; Grofová et al., 1978; Edwards et al., 1979). Various roles have been proposed for individual SC id afferents including conveying behaviorally-relevant

information about visual input (frontal eye field (FEF) Segraves and Goldberg, 1987; Sommer and Wurtz, 2000, 2001; Paré and Wurtz, 2001; lateral interparietal cortex: Paré and Wurtz, 2001;

Wurtz et al., 2001; V1: Liang et al., 2015), recent experience (FEF: Sommer and Wurtz, 2001;

secondary motor cortex: Duan et al., 2019), and target value (substantia nigra pars reticulata:

Handel and Glimcher, 2000; Basso and Wurtz, 2002; Sato and Hikosaka, 2002; Bryden et al., 2011), as well as more direct movement-related functions such as saccade initiation (FEF:

Schiller et al., 1980; Hanes and Wurtz, 2001) and cessation (Goffart et al., 1998b).

Intrinsic processing in the superior colliculus

These studies suggest an integrative role for the SC in mediating behavior (Wolf et al., 2015); however, a variety of cell types exist within the SC id , thus fully understanding its functional circuitry will require a better understanding of its cell-type-specific inputs (Oliveira and Yonehara, 2018; Masullo et al., 2019). SC id contains ~70% glutamatergic cells and ~30%

GABAergic cells (Mize, 1992) each with projections within and between SC layers, to the contralateral SC, and out of the SC (Pettit et al., 1999; Isa and Hall, 2009; Sooksawate et al., 2011; Ghitani et al., 2014), indicating that intrinsic SC processes may play a pivotal role in the SC id computations underlying spatial orienting behaviors. Therefore, a necessary piece of elucidating SC id function rests in characterizing the unique projections to excitatory and inhibitory SC neurons. The introduction of new tools in mouse transgenics (Branda and

Dymecki, 2004) and transsynaptic tracers (Wickersham et al., 2007; Wall et al., 2010; Luo et al., 2018) have recently enabled neuroscientists to probe microcircuit organization with greater specificity. Thus, we employed Cre-lox recombination in conjunction with a transsynaptic retrograde rabies virus tracer strategy to label monosynaptic inputs to excitatory and inhibitory SC neurons, as well as to a subset of brainstem-projecting SC neurons thought to drive orienting movements (Sooksawate et al., 2005, 2008). We found that projection patterns differed to these populations, suggesting cell-type-specific input integration.

Overall, the studies described here explore the circuit organization and function within two structures that mediate flexible sensorimotor transforms – fundamental neuronal processes with implications for numerous neurological functions that when impaired manifest as

neurological disorders.

CHAPTER II

SENSORIMOTOR INTEGRATION IN THE CEREBELLAR GRANULE CELL LAYER Introduction

The cerebellum is critical for associative learning (Ito, 2006). A large body of theoretical work has attempted to understand the cerebellar computations that underlie the structure’s capacity for learning (Marr, 1969; Albus, 1971; Mauk and Donegan, 1997; Ohyama et al., 2003;

Mauk and Buonomano, 2004; Mauk and Ohyama, 2004; Steuber et al., 2007; Gilmer and Person, 2017). An important theorized, and now identified, locus of learning is between granule cell (GrC) parallel fibers and Purkinje cell (PC) dendrites. The long-term depression that occurs at this synapse has been the focus of considerable study (Ito, 1982, 1989, 2001, 2002a; b; Linden et al., 1991, p; Daniel et al., 1998; Mittmann and Häusser, 2007; Steuber et al., 2007; Schonewille et al., 2011; Kakegawa et al., 2018), however, another important component of cerebellar learning theory rests on how the preprocessing of information by the GrC layer facilitates learning (Marr, 1969; Albus, 1971; Mauk and Buonomano, 2004; Gilmer and Person, 2017).

Much less attention has been paid to this processing in the GrC layer and is the focus of the present study.

Notably, more than half of the neurons in the mammalian brain lie within the input layer

to cerebellar cortex, the GrC layer (De Schutter and Bjaalie, 2001; Herculano-Houzel and Lent,

2005). GrCs receive sensory and motor-related signals via excitatory mossy fibers arising from

dozens of distinct brain regions (Eccles et al., 1967; Apps and Hawkes, 2009), and they are

highly unique in that each GrC has on average only four dendrites, each of which are contacted

by a single mossy fiber. This super abundance of neurons each receiving few inputs raises the

question of what computational advantages this organization serves. Foundational theories of

cerebellar function propose that this anatomical arrangement is ideal for representing the unique features of highly overlapping patterns, and that Golgi cells, the inhibitory interneurons of the GrC layer, regulate the degree of pattern separation performed by GrCs (Marr, 1969; Albus, 1971). Later modeling work expanded upon these ideas by demonstrating that temporal sparsification of GrC activity patterns can facilitate learning (Mauk and Buonomano, 2004;

Kalmbach et al., 2011). However, little information exists regarding the interaction of diverse mossy fiber populations on GrC encoding (Livet et al., 2007; Huang et al., 2013). Thus, further defining the physiological convergence between unique mossy fiber populations will help us to better understand the GrC layer computations that underlie learning (Gilmer and Person, 2017).

Mossy fiber inputs to the cerebellum arise from several areas of the nervous system.

Ascending inputs from spinal cord (Voogd et al., 1969; Lakke et al., 1986; Yaginuma and

Matsushita, 1987; Verburgh et al., 1989), brainstem (Russchen et al., 1976; Saigal et al., 1982),

and cranial ganglia (Saigal et al., 1980a; b) carry proprioceptive and cutaneous information

(Garwicz et al., 1998), while descending corticopontine inputs (Freedman et al., 1975; Glickstein

et al., 1994; Serapide et al., 2001) convey a mixture of processed sensory and motor information

(Proville et al., 2014). One unique source of input arises from collaterals of cerebellar nuclear

output neurons (Houck and Person, 2015), constituting an intracerebellar feedback loop involved

in motor learning (Gao et al., 2016). While unique mossy fiber sources converge in similar areas

of cerebellar cortex (Wu et al., 1999; Shinoda et al., 2000; Houck and Person, 2015), little is

known about how these unique information streams are combined in individual GrCs (Huang et

al., 2013; Ishikawa et al., 2015). Thus, I examined the functional interactions between mossy

fibers carrying somatosensory information and mossy fibers carrying cerebellar nucleocortical

(NC) feedback (Fig. 2.1) by performing in vivo extracellular recordings from GrCs and PCs

Figure 2.1. Nucleocortical fibers

Arrows indicate axons of neurons in CbN

(cerebellar nuclei). BDA (biotinylated dextran

amine); GCL (granule cell layer).

while independently driving each pathway using whisker air puffs and optogenetic stimulation.

These results have important implications for how these two information streams are integrated by the input layer of cerebellum.

Acknowledgements

Brenda Houck performed the histology and acquired the image displayed in Figure 2.1.

Materials and Methods Animals

All procedures followed the National Institutes of Health Guidelines are were approved by the institutional Animal Care and Use Committee at the University of Colorado Anschutz Medical Campus. Animals were housed in an environmentally controlled room, kept on a 12- hour light-dark cycle and had ad libitum access to food and water. Adult mice of both sexes were used in these experiments (n = 5 males; n = 21 females). No sex differences were observed between GrC sensory- and optogenetically-evoked responses. All mice were adult C57BL/6 (including Mutant Mouse Regional Resource Center Tg(Ntsr1-Cre)GN220Gsat/Mmucd mice;

Houck and Person, 2015) bred in house.

Surgery and opsin expression

The excitatory opsins, channelrhodopsin (ChR2) or Chronos, were injected bilaterally into the interposed nuclei of 65 mice (injection coordinates: AP: -2.0 mm from bregma; ML:

±1.1 mm; DV: -2.5 mm). Wild-type mice were injected with either rAAV2-hSyn-hChR2-

H134R-mCherry (14 mice) or rAAV8-Syn-Chronos-GFP (32 mice), and neurotensin receptor

type 1-Cre (Ntsr1-Cre) mice, a transgenic line in which Cre recombinase expression is restricted to glutamatergic projection neurons (Houck and Person, 2015), were injected with either AAV- FLEX-mCherry-ChR2 (5 mice), AAV1-CAGGS-FLEX-ChR2-tdTomato-WPRE-SV40 (7 mice), or AAV8-hSyn-FLEX-Chronos-GFP (7 mice). Opsin expression proceeded for 3-20 weeks prior to recordings.

In vivo electrophysiology, optogenetics, and sensory stimulation

I performed acute recordings in 65 adult, ketamine (50 mg/kg)/xylazine (5 mg/kg)

anesthetized, head-fixed mice. Extracellular, single-unit recordings were targeted to GrCs using a hydraulic micromanipulator (Siskiyou, model MX610) to advance high-impedance glass

microelectrodes (1- 40 MΩ; 0.5-2 M NaCl; ~1 µm diameter tip opening; Jörntell and Ekerot,

2006; Barmack and Yakhnitsa, 2008; Ruigrok et al., 2011; Hensbroek et al., 2014) through

cerebellar lobules Crus I and Crus II. Recordings were conducted with a microelectrode

amplifier (A-M Systems model 1800) sampled at 30 kHz with a Power3 1401 (Cambridge

Electronic Designs) while being monitored visually in Spike2 (Cambridge Electronic Designs),

and acoustically with an audio monitor (A-M Systems; Model 3300). Recordings were made

while using the Power3 1401 to send TTL pulses to deliver whisker and/or optical stimulation in

randomly interleaved trials (inter-trial interval: 3-5 s). Whisker stimulation consisted of an air

puff directed at the vibrissal pad (200 ms; 20 psi) ipsilateral to the recording electrode and was

delivered with a pico-liter injector (Warner Instruments; Model PLI-10). Chronos- or ChR2-

expressing terminals were stimulated by trains of 470 nm light (50 Hz; 5 or 10 5-ms width

pulses) delivered through an optical fiber (200, 300, or 400 µm core diameter; 0.39 numerical

aperture) coupled to an LED (Doric LEDRV_2CH; 36-85 mW/mm ² ). While ChR2 was used in

earlier experiments, I opted to use Chronos in later experiments due to its superior light sensitivity and faster channel gating kinetics (Klapoetke et al., 2014, Fig. 2.2d). Light was delivered proximal to the recording site by cementing an optical fiber to a glass microelectrode (Dondzillo et al., 2013; Fig. 2.2c) such that the tip of the microelectrode extended beyond the optical fiber by ~50-100 µm.

Spike sorting and cell-type identification

Spike sorting was performed using Spike2 software (Version 7.1, Cambridge Electronic Designs). Spikes were detected using a manually set threshold, and then sorted using a template matching algorithm in combination with principle component analysis, visual inspection of the overlaid waveforms, and in some cases, inspection of individual spikes. Forty-one GrCs were identified on the basis of accepted electrophysiological criteria (Chadderton et al., 2004; Jörntell and Ekerot, 2006; Barmack and Yakhnitsa, 2008; Ruigrok et al., 2011; Duguid et al., 2012;

Hensbroek et al., 2014): putative GrC recordings were required to meet one of the following two criteria to be included in the GrC population: 1) a spontaneous firing rate less than 1 Hz; or 2) the presence of at least 10 inter-spike intervals which were at least 50 times shorter than the median inter-spike interval. Additionally, six PCs were identified as described previously on the basis of their high firing rate and tendency to exhibit brief pauses (Hensbroek et al., 2014).

Histological analysis of opsin expression

At the end of a recording session mice were immediately overdosed with an

intraperitoneal injection of a sodium pentobarbital solution, Pentobarbital (Sigma-Aldrich Inc.),

and perfused transcardially with 0.9% saline followed by 4% paraformaldehyde. Brains were

Figure 2.2. Experimental methodology

(a) Circuit schematic of cerebellar cortex. Mossy fibers carrying whisker information (red) and motor-related information (blue) terminate on GrCs and Golgi cells. (b) Experimental paradigm. Recordings were performed in anesthetized mice receiving whisker air puff (red) and NC optogenetic (blue) stimulation (see Materials and Methods). (c) Image of a single-use electrode-fiber assembly used for recordings. (d) table illustrating channel gating kinetics and light sensitivity differences between ChR2 and Chronos (values taken from Klapoetke et al., 2014). (e) Upper: example GrC waveform overlay (left) and sample recording trace (right).

Middle panels: inter-spike interval (ISI) histograms displayed on different scales. Lower: All

ISIs plotted against the ISI before and ISI after. (f) Same as in e except for an example PC.

removed and postfixed for 4-24 hours then cryoprotected in 30% sucrose. Tissue was sliced in 40 µm serial coronal sections using a freezing microtome and stored in 0.1 M phosphate buffered saline. Every third section was mounted onto slides and opsin expression was examined using an epifluorescent microscope. Only animals in which opsin expression was readily apparent and largely restricted to cerebellar nuclei were included in analyses.

Data analysis

Spike2 software was used to export spike-clustered data files which were subsequently analyzed in MATLAB using custom-written software. Raster plots were made by binning spikes at 1 ms resolution. Peri-stimulus time histograms (PSTHs) were smoothed by convolving raw trial averaged firing rate vectors with a 10 ms width Gaussian filter created by the MATLAB function ‘fspecial’. The arithmetic sum PSTHs displayed in Figures 2.3 and 2.4 were generated by summing the two unistimulus raw firing rate vectors after first subtracting from each a baseline firing rate computed from the inter-trial intervals over the entire recording session. The summed PSTH was then smoothed as described above. The response time plotted in Figures 2.5 and 2.6 was defined as the center of mass of the response occurring within 500 ms across all trials (equivalent to the central-most spike of the raster plot).

Stimulus responsivity

Cells were determined to be responsive to a stimulus condition by iteratively comparing

trial by trial spike counts within a sliding window stimulus epoch to an equal duration of baseline

immediately prior to stimulus onset using a Wilcoxon signed rank test (two-tailed). The sliding

window incremented by 1 ms per iteration, had a width equal to the duration of stimulus

presentation, and ended once the leading edge reached 500 ms post-stimulus onset. The multiple comparisons performed during this process were corrected for by using a false discovery rate analysis (Curran-Everett, 2000) to compute a new significance threshold based on pooling all p- values across all conditions and all cells. A cell was then only determined to be responsive to a stimulus condition (Fig. 2.5a) if at least 5% of the p-values obtained during the sliding window procedure were below the new significance threshold.

Linearity analysis

The linearity of cell responses to dual stimulus presentation was computed by comparing trial by trial evoked spike counts (stimulus epoch spike counts minus baseline epoch spike counts) within the dual stimulus condition to the arithmetic sum of all possible unistimulus trial combinations using a Wilcoxon rank sum test (two-tailed). The stimulus epoch was defined as the duration of stimulus presentation and was compared to an equal duration of baseline

immediately prior to stimulus onset. Comparisons made across conditions were corrected for by

using a false discovery rate analysis (Curran-Everett, 2000) to compute a new significance

threshold. Cells with significant responses were classified as superlinear or sublinear responders

based on whether the mean of the dual stimulus-evoked spike counts was higher or lower than

the mean of the arithmetic sum evoked spike counts, respectively. Cells that met criteria for

superlinearity in one dual stimulus condition and sublinearity in another were classified as linear

responders, as were cells that did not reach significance.

Results In vivo electrophysiology and cell-type identification

To test for multi-modal integration in the GrC, I performed in vivo recordings from mouse GrC single units (Fig. 2.2a). Recordings were made while puffing air onto the whiskers of anesthetized mice while concurrently optogenetically exciting NC terminals in the GrC layer of cerebellar lobules Crus I and Crus II (Fig. 2.2b). While GrCs were specifically targeted using a high-impedance glass microelectrode, PCs were also encountered, and I report data from both cell types here. Once a single-unit was isolated, randomly interleaved trials were presented which consisted of either an air puff alone, optogenetic stimulation alone, or the combination of both stimuli presented at one of 4-6 temporal offsets from one another. Forty-one putative GrCs were identified based on their firing properties, specifically a low firing rate and high variability of inter-spike intervals (Fig. 2.2e; also see Materials and Methods). Additionally, six PCs were identified based on their overall high firing rate and brief periodic pauses (Hensbroek et al., 2014; Fig. 2.2f). Using this approach, I examined the sensorimotor integrative properties of GrCs and PCs in vivo.

Granule cell responsivity

GrC responses varied in response to the presentation of somatosensory and/or NC

optogenetic stimulation. Some GrCs responded with an increase in firing rate to both the whisker

and NC stimulus when presented in isolation (Fig. 2.3a, b), and exhibited a superlinear response

when the stimuli were presented together (Fig. 2.3c-f). However, other cells that were responsive

to only the whisker stimulus (Fig. 2.4a, b) exhibited a sublinear response in some dual stimulus

conditions (Fig. 2.4d, e) and a linear response in others (Fig. 2.4c, f). These observations indicate

Figure 2.3. Superlinear GrC responses to whisker and NC stimulation

(a) Raster plot (top) and peri-stimulus time histogram (PSTH; bottom) depicting single GrC

response to whisker stimulation (light red shaded area). (b) Same as in a except GrC response

is to 5 pulses of NC optogenetic stimulation (each light blue shaded area is one pulse). (c-f)

Raster plots (top) and PSTHs (middle) as in a-b except GrC response is to combination of

whisker and NC optogenetic stimulation. Label above raster plot indicates relative timing

between the two stimuli. PSTHs (bottom) depict GrC response to whisker stimulation alone

(red), NC optogenetic stimulation alone (blue), whisker and NC optogenetic stimulation

combined (purple), and the arithmetic sum of whisker alone and NC optogenetic stimulation

Figure 2.4. Sublinear GrC responses to whisker and NC stimulation

that NC activity is capable of both enhancement and suppression of whisker air puff-evoked GrC responses.

Granule cell population responses

GrCs were first grouped based on their response to three conditions: whisker air puff presented alone, optogenetic stimulation presented alone, or the two stimuli presented together.

The response pattern to these three conditions defines the “base response category”. For example, GrCs responsive to all three stimulus conditions are grouped into base response

category I, GrCs responsive only when both stimuli are presented together are grouped into base response category II, while GrCs responsive to both the two stimuli presented together and to the whisker air puff presented alone are grouped into base response category III. GrCs were then analyzed to determine whether their responses to the two stimuli presented together were

consistent with a linear, superlinear, or sublinear merging of individual stimuli. I then combined a GrC’s “base response” with the linearity of its response to infer the type of convergent

information it received from the two input pathways being driven experimentally. GrCs were classified as receiving convergent excitatory information, convergent excitatory and inhibitory information, or as receiving no convergence of information based on the rationale described below.

Only GrCs in which opsin expression was restricted to the cerebellar nuclei were

included in this analysis (26/41 GrCs). 17/26 GrCs (65%) responded to at least one stimulus

condition (Fig. 2.5a). 11/17 GrCs (65%) responded to whisker air puffs, 2/17 GrCs (12%)

responded to NC optogenetic stimulation, and all 17 GrCs (100%) responded when the two

stimuli were presented together. Three unique response patterns were observed in these

Figure 2.5. GrC population responses and convergence patterns

(a) Base responsivity of all 26 GrCs with opsin expression limited to cerebellar nuclei. Red

circles indicate significant responses (false discovery rate analysis used to correct for multiple

comparisons; see Materials and Methods) to whisker, NC, or dual stimulus conditions. Black

circles indicate there was no significant response. (b) Responsivity patterns of 17 GrCs

responsive to at least one stimulus condition (red circles in a). Check marks indicate cell was

responsive to particular condition. Upward and downward arrows indicate superlinear and

sublinear responses, respectively, and dash indicates no change from linearity. (c) Center of

mass response latency difference between whisker air puff alone and dual stimulus condition

for GrCs receiving putative excitatory-excitatory (EE) convergence (left) and excitatory-

inhibitory (EI) convergence (right). All dual stimulus conditions were included. Filled circle

and error bars indicate mean ± SEM (n = 6 EE GrCs; n = 6 EI GrCs). *: distributions are

different from zero; p < 0.05.

responsive GrCs (Fig. 2.5b): 1) two GrCs responded to both stimuli when presented together as well as to each of the individual stimuli presented alone (base response category I; Fig. 2.5bi), 2) six GrCs were only responsive when both whisker air puffs and NC stimuli were presented together (base response category II; Fig. 2.5bii), and 3) nine GrCs were responsive to the presentation of both stimuli together and when whisker air puffs were presented alone (base response category III; Fig. 2.5biii). These unique response patterns provide some insight into the nature of the convergence of excitatory and inhibitory information that arises from coactivation of somatosensory and NC pathways. To characterize these responses, I computed the linearity of the integrative response by comparing the magnitude of the dual stimulus-evoked response with that of the arithmetic sum of the individual stimulus-evoked responses (Fig. 2.5b, arrows; also see Materials and Methods). GrCs in the first base response category (Fig. 2.5bi; cells 1, 2) were excited by all stimulus conditions, indicating a convergence of excitatory somatosensory

information with excitatory NC information. Similarly, cells in the second base response category (Fig. 2.5bii; cells responsive to only the dual stimulus condition) exhibiting linear integrative responses (Fig. 2.5bii, cells 3-6) are cases where converging excitation was required to drive these cells above baseline. However, the remaining GrCs in the second base response category exhibiting sublinear integrative responses (Fig. 2.5bii, cells 7, 8) indicate that NC activation can inhibit whisker air puff-evoked excitation. Similarly, GrCs in the third base response category (Fig. 2.5biii; cells responsive to the dual stimulus condition as well as the whisker air puff alone stimulus condition) with a sublinear integrative response (Fig. 2.5biii, cells 9-12) illustrate instances of somatosensory excitation converging with NC inhibition.

Finally, GrCs in the third base response category that had a linear integrative response (Fig.

2.5biii, cells 13, 17) demonstrated no evidence of convergence as their response to the dual

stimulus is explained as simply being a product of their response to the whisker air puff. Taken together, these observations suggest that somatosensory and NC information converging on individual GrCs produces a balance between mutual excitation and competing excitation and inhibition.

Granule cell response timing shifts

GrCs were grouped according to their putative somatosensory/NC convergence patterns, as described above (Fig. 2.5b). Each group was then examined to determine whether NC

optogenetic activation changed their whisker air puff-evoked onset latency. GrCs receiving putative mutual excitation from somatosensory and NC pathways responded to whisker

stimulation sooner in trials with concurrent NC stimulation than in trials without NC stimulation (change in response time: 16 ms [mean]; 12 ms [median]; Wilcoxon signed rank test [two- tailed]; n = 30 dual stimulus conditions from 6 GrCs; p < 0.05; Fig. 2.5c). Additionally, GrCs receiving putative convergence of excitation and inhibition from somatosensory and NC

activation responded to whisker stimulation later in trials when NC optogenetic stimulation was present (change in response time: 25 ms [mean]; 1 ms [median]; Wilcoxon signed rank test [two- tailed]; n = 29 dual stimulus conditions from 6 GrCs; p < 0.05; Fig. 2.5c). These results indicate that NC drive is capable of recruiting excitation or inhibition to bidirectionally diversify the timing of GrC sensory-evoked responses.

Purkinje cell response timing shifts

In addition to examining how NC activation altered sensory processing in GrCs, I also

observed the effects of driving these convergent pathways on downstream PCs. I found that PC

Figure 2.6. NC activation

increases PC response latency

Center of mass response latency

difference between whisker air

puff alone and dual stimulus

condition. The dual stimulus

condition with the earliest

whisker air puff onset was used

for each cell. Filled circle and

error bars indicate mean ± SEM

(n = 6 PCs). *: distribution is

different from zero; p < 0.05.

sensory-evoked responses were delayed when NC activity was elevated (mean delay: 9 ms; one- sample t-test [two-tailed]; n = 6 PCs; p < 0.05; Fig. 2.6). These observations suggest a net inhibitory effect of NC activity on the GrC layer.

Discussion Limitations

While these studies provide insight into the nature of GrC layer processing of convergent

information, certain methodological limitations constrain the conclusions that can be drawn. A

major challenge of these experiments was obtaining a sufficiently large number of recordings

from stimulus responsive GrCs. This was true mainly due to challenges in recording as well as

post-hoc identification. Recording challenges consisted of a low probability of encountering

active GrCs due to their very low spontaneous firing rates (D’Angelo et al., 1995; Chadderton et

al., 2004; Barmack and Yakhnitsa, 2008; Hensbroek et al., 2014), difficulty isolating a single-

unit from a group of small, tightly-packed cells, and lack of a straightforward way to target the

GrC layer due to its atypical shape and variable thickness. Post-hoc identification challenges

stemmed from our reliance on spiking activity to determine cell identity, as extracellular

recordings precluded our using membrane properties to more definitively identify GrCs

(D’Angelo et al., 1995; Chadderton et al., 2004). While previous groups have developed and

validated GrC identification criteria using juxtacellular recordings techniques (Barmack and

Yakhnitsa, 2008; Hensbroek et al., 2014), our population of putative GrCs was somewhat

sensitive to which group’s criteria were used. This inconsistency might be due to differences in

species or experimental design, or the possibility that there is a diversity of GrC spiking

characteristics that reflects the diverse activity of their inputs (Jörntell and Ekerot, 2006; Arenz et al., 2008).

Another caveat is our use of non-cell-type-specific opsins in most experiments (see Materials and Methods). In addition to providing excitatory feedback to the cerebellar cortex, the cerebellar nuclei also send inhibitory projections to the cortex that target Golgi cells (Ankri et al., 2015). Therefore, it is possible that we were also driving the inhibitory NC pathway in some of our experiments. Nevertheless, our results still provide insight into the effect of NC feedback on the cerebellar cortex, albeit, in a manner lacking cell-type specificity.

A final limitation relates to characterizing the linearity of the integrative response. We did not assess the position of GrCs within their dynamic range. This raises the possibility that observed super- and sublinear responses reflected GrCs that were activated below or above their dynamic range, respectively. If this did bias our results, it likely would have been in the direction of seeing more superlinearity; since GrCs can burst at very high rates, it is unlikely that observed responses were saturated. Seeing that most of the nonlinear integrative responses we observed were sublinear, we concluded that this had a minimal impact on our results.

Implications

Notwithstanding the above-mentioned limitations, this study provides useful insight into

the dynamics of convergent streams of input into the cerebellar cortex. The cerebellum is a key

structure for forming associations needed for adaptive learning (Ito, 2006). Its vast number of

neurons and crystalline circuity make it ideally suited for pattern separation and storage (Marr,

1969; Albus, 1971; Ito, 1982; Mauk and Buonomano, 2004; Gilmer and Person, 2017). These

functions are critically informed by the interactions among incoming streams of information.

While little is empirically known about these interactions, my experiments investigated this gap by employing in vivo GrC and PC recordings and optogenetics to examine how somatosensory and NC information converge in cerebellar cortex. My results suggest that NC information can actively suppress or enhance GrC sensory-evoked responses, that activation of these pathways produces a balance between mutual excitation and competing excitation and inhibition, and that NC feedback temporally diversifies GrC sensory-evoked responses and delays PC sensory- evoked responses.

Prominent theories of cerebellar function predict that the GrC layer combines unique streams of information in order to facilitate pattern separation (Marr, 1969; Albus, 1971; Ito, 1989), however, little is known about the dynamics of GrC multimodal integration or the role played by Golgi cell inhibition in this process. Nor do we understand how intracerebellar feedback facilitates learning or augments sensory processing. Our finding that NC activation produces equal levels of excitation and inhibition in GrCs suggests that it engages the GrC layer differently than somatosensory inputs, which we only observed driving excitation. The increased level of inhibition seen following NC activation may be a consequence of NC terminals

preferentially targeting superficial regions of the GrC layer (Gao et al., 2016), which contain a higher density of inhibitory Golgi cells (Simat et al., 2007). Another possibility is that we promoted Golgi cell activity via a disinhibitory mechanism; inhibitory NC terminals exclusively target Golgi cells (Ankri et al., 2015), which inhibit each other (Hull and Regehr, 2012).

Regardless of the circuit mechanism, the unique capacity of NC feedback to bidirectionally modulate GrC sensory encoding raises the question of what computational role this serves.

Our findings that NC activation temporally diversifies GrC sensory-evoked responses

suggests a neural substrate for a cerebellar-based representation of time. Theoretical models have

implicated Golgi cells as a likely candidate for generating temporal representations (Mauk and Buonomano, 2004; Kalmbach et al., 2011), and, consistent with this, mGluR2-expressing Golgi cell ablations in mice recapitulates classic symptoms of cerebellar ataxia (Watanabe et al., 1998), indicating that these cells play an indispensable role in cerebellar computations. Work from delay eyeblink conditioning studies posits that the GrC layer represents the sensory context that over time becomes associated with adaptive motor responses (Mauk and Donegan, 1997;

Ohyama et al., 2003). Additionally, NC feedback is composed of collaterals of neurons projecting to pre-motor areas. Thus, NC feedback may enrich the contextual representation by complementing it with the current motor plan through a Golgi cell mediated mechanism of GrC temporal diversification. While these ideas have yet to be rigorously tested, optogenetic

experiments have demonstrated a bidirectional affect of NC activity on learning in conditioned eyeblink paradigms (Gao et al., 2016), suggesting that it may enhance the predictive capacity of the cerebellum.

Apart from the influence of NC activity on GrCs, an important question is how this information is filtered through the GrC layer and read out by PCs, the output cells of the cerebellar cortex. Our findings that NC activation delayed PC sensory-evoked responses is consistent with previous experiments in which NC optogenetic activation inhibited PCs (Gao et al., 2016), suggesting that the net effect of NC activity on GrC layer processing is inhibitory. In contrast to the NC-induced bidirectional modulation of GrC sensory-evoked activity that we observed, the unidirectional effect on PC sensory-evoked activity may function to temporally align their responses, thereby enhancing synchronicity at the population level, which is known to be important for driving their target cells in the cerebellar nuclei (Person and Raman, 2012a).

Overall, we have examined GrC layer dynamics while driving converging somatosensory and

NC pathways. Following up on these results will have important implications for how sensory

and motor information are integrated within the cerebellar circuit to facilitate learning.

CHAPTER III

MONOSYNAPTIC INPUTS TO SUPERIOR COLLICULUS Introduction

The superior colliculus (SC) is a highly conserved midbrain structure critical for orienting behavior (Basso and May, 2017), as well as other associated functions such as spatial attention (Krauzlis et al., 2013) and multisensory integration (Stein and Stanford, 2008). The SC is

organized into a superficial visual layer, which receives projections from the retina (Apter, 1945) and descending inputs from the neocortex (Kawamura et al., 1974), and intermediate and deep layers (SC id ) that receive widespread input from several cortical and subcortical regions (Sparks and Hartwich-Young, 1989). The SC id is organized into a topographic map of movement space, whereby small amplitude orienting movements are encoded rostrally and larger amplitude movements are represented caudally (Robinson, 1972; Wang et al., 2015). While much of our understanding of the role of the SC id during behavior originated with work in primates making saccades to visual targets (Goldberg and Wurtz, 1972b; c; Wurtz and Goldberg, 1972; Lee et al., 1988), other work across a wider range of species points to a broader involvement of the SC id (or the optic tectum (OT) the nonmammalian homologue of the SC) in other orienting behaviors (Sparks, 1999). For example, SC id /OT activity encodes orienting movements of the head in cats (Guillaume and Pélisson, 2001), monkeys (Freedman et al., 1996; Corneil et al., 2002; Walton et al., 2007), owls (du Lac and Knudsen, 1991), frogs (Meyer and Sperry, 1973), and bats

(Valentine et al., 2002). SC id /OT neural activity also controls limb movements in cats (Courjon

et al., 2004, 2015), monkeys (Werner et al., 1997; Philipp and Hoffmann, 2014), and mice

(Steinmetz et al., 2018) as well as full body orienting movements in goldfish (Herrero et al.,

1998) and rodents (Felsen and Mainen, 2008; Stubblefield et al., 2013). In addition to its role in

orienting to targets across a wide range of evolutionarily diverse species, the SC is also critical for producing escape behavior away from aversive stimuli (Dean et al., 1986, 1989; Sahibzada et al., 1986; Evans et al., 2018).

Alongside our understanding of the SC id ’s roles in behavior, a great deal is also known about which brain centers project to the SC id (Edwards et al., 1979; Sparks and Hartwich-Young, 1989; Wolf et al., 2015). Several studies have employed anterograde and/or retrograde tracers demonstrating SC id afferents originating from cerebral cortex (Garey et al., 1968; Edwards et al., 1979; Fries, 1984), thalamic areas (Edwards et al., 1974, 1979; Graybiel, 1974; Grofová et al., 1978), cerebellar nuclei (Batton et al., 1977; Kawamura et al., 1982), and several mesencephalic regions (Hopkins and Niessen, 1976; Grofová et al., 1978; Edwards et al., 1979). Potential roles for individual SC id afferents range from transmitting behaviorally-relevant information about visual input (frontal eye field (FEF): Segraves and Goldberg, 1987; Sommer and Wurtz, 2000, 2001; Wurtz et al., 2001; lateral interparietal cortex: Paré and Wurtz, 2001; Wurtz et al., 2001;

V1: Liang et al., 2015), recent experience (FEF: Sommer and Wurtz, 2001; secondary motor cortex (M2): Duan et al., 2019), and target value (substantia nigra pars reticulata (SNr): Handel and Glimcher, 2000; Basso and Wurtz, 2002; Sato and Hikosaka, 2002; Bryden et al., 2011), to more active roles such as saccade initiation (FEF: Schiller et al., 1980; Hanes and Wurtz, 2001) and cessation (cerebellum: Goffart et al., 1998).

While these and other studies point to an integrative role for the SC in mediating

behavior (Wolf et al., 2015), the SC id itself contains a variety of cell types, and in order to fully

elucidate its functional circuitry we need to better understand its cell-type-specific inputs

(Oliveira and Yonehara, 2018; Masullo et al., 2019). As a first step, we focused on inputs to

excitatory and inhibitory SC id neurons (“eSCNs” and “iSCNs,” respectively). The SC id is