Vertebrate motor behaviors vary widely both in form and complexity, and so do the brains that generate them. Despite this variability, there is a high degree of conservation across vertebrate taxa in the organization of the neural circuits which control the patterning and expression of motor behavior, which are usually distributed across multiple regions within the nervous system. Attempts to understand the principles of how nervous systems generate motor outputs are aided by taking a broad perspective, comparing how neural circuits at different levels of the brain interact and cooperate to produce behaviors with differing levels of complexity. We investigated the question of how variable motor patterns of a single class of behaviors are generated and expressed by motor control circuits in a relatively simple vertebrate model system, the vocalizations of the African clawed frog, Xenopus laevis. Female and male Xenopus make temporally stereotyped sex-specific calls using a single pair of muscles. Calls vary in complexity, and female calls are considerably simpler than those of males in terms of temporal structure, but both sexes switch between different components of a sex- specific vocal repertoire in response to external stimuli and internal states. Sex differences in vocal behavior are regulated by gonadal hormones, and both the patterning and the expression of sex-specific call types can be modified by manipulating hormone levels in adulthood. We took advantage of this flexible control of otherwise stereotyped motor behavior to analyze how motor circuits pattern and express different vocal behaviors in Xenopus using an ex vivo isolated brain preparation. Fictive calling episodes closely matching the temporal structure of in vivo calls are readily induced ex vivo in both male and female Xenopus by bath application of serotonin (5- HT). We used 5-HT-elicited fictive calling episodes to probe how the activity of vocal circuits varies between male and female calling patterns, and investigated the mechanisms that generate these differences using female brains whose vocal circuits had been masculinized by treatment with exogenous androgen. We show that vocal patterning circuits can be masculinized even where there is no expression of vocal behavior in vivo, that sex differences in vocal patterns are expressed at multiple levels of the vocal pattern generating circuit, and that individual characteristics that vary as a function of sex differ in their sensitivity to masculinization by exogenous androgen. Masculinization of ex vivo vocal patterns without masculinization of vocal behavior in vivo suggests that the circuits governing patterning are distinct from those governing action selection in this system. Using a combination of tracing and microstimulation techniques in the isolated brain, we outline a putative top down control circuit for vocal control in Xenopus. This circuit is centered on the anuran central amygdala nucleus (CeA), located in the ventral subpallium of the telencephalon. We show that this forebrain nucleus receives auditory input from a thalamic sensory nucleus, and projects directly and indirectly to vocal pattern generating circuits in the hindbrain. Electrically stimulating CeA in the ex vivo preparation induces fictive calling episodes in the absence of exogenous 5-HT. Electrical stimulation is equally effective in a neighboring subpallial nucleus, the bed nucleus of the stria terminalis (BNST). BNST and CeA share several common targets within the diencephalon and isthmo-mesencephalic tegmentum, however BNST does not project directly to hindbrain vocal pattern generating nuclei. The fictive calling generated by these two subpallial nuclei is indistinguishable, indicating that the ability of microstimulation to drive activity in hindbrain vocal circuits is mediated through indirect connections. In female (but not male) brains, the temporal characteristics of fictive calling induced by microstimulation differ from those induced by 5-HT application, occurring at faster repetition rates that resemble the call made by receptive female Xenopus in response to auditory stimulation. We propose that fictive calling induced by stimulation of the telencephalon represents an ex vivo correlate of this behavioral pattern. The reproductive state of the female frog at the time of brain isolation does not alter the response of the ex vivo preparation to either 5-HT or microstimulation. We hypothesize that the effects of behavioral state on neural circuits are mediated by sensory nuclei upstream of forebrain motor control circuits. Considered as a whole, the work presented in this thesis shows that variations in motor behaviors are expressed through multiple levels of motor control circuits throughout the central nervous system. These results emphasize the benefits of studying motor behaviors with a view to diversity and variation, both when considering the behaviors themselves and when analyzing the circuits that pattern and govern them.

The ability to store and recall memories is an essential function of nervous systems, and at the core of subjective human experience. As such, neuropsychiatric conditions that impair our memory capacity are devastating. Learning and memory in mammals have long been known to depend on the hippocampus, which has motivated widespread research efforts that converge on two broad themes: determining how different cell types in the hippocampus interact to generate neural activity patterns (structure), and determining how neural activity patterns implement learning and memory (function). Central to both these pursuits are pyramidal cells (PCs) in CA1, the primary hippocampal output, which transform excitatory synaptic inputs into the action potential output patterns that encode information about locations or events relevant for memory. CA1 PCs are embedded in a network of diverse inhibitory (GABA-releasing) interneurons, which may play unique roles in sculpting the activity patterns of PCs that implement memory functions. As a consequence, investigating the functional impact of defined GABAergic interneurons can provide an experimental entry point for linking neural circuit structure to defined computations and behavioral functions in the hippocampal memory system. In this thesis I have applied a panel of novel methodologies to the mouse hippocampus in vitro and in vivo to link structure to function and behavior, and determine 1) how hippocampal inhibitory cell types shape distinct patterns of PC activity, and 2) how these inhibitory cell types contribute to the encoding of contextual fear memories. To first establish the means by which interneuron subtypes contribute to PC activity patterns, I used optogenetic techniques to activate spatiotemporally distributed synaptic excitation to CA1 in vitro, and recorded from PCs to quantify the frequency of output spikes relative to input levels. I subsequently used a dual viral and transgenic approach to combine this technique with selective pharmacogenetic inactivation of identified interneurons during synaptic excitation. I found that inactivating somatostatin-expressing (Som+) dendrite-targeting interneurons increased the gain of PC input-output transformations by causing more output spikes, while inactivating parvalbumin-expressing (Pvalb+) soma-targeting interneurons did not. Inactivating Som+ inhibitory interneurons allowed the dendrites of PCs to generate local NMDA receptor-mediated electrogenesis in response to synaptic input, resulting in high frequency bursts of output spikes. This discovery suggests neuronal coding via hippocampal burst spiking output can be regulated by Som+ dendrite-targeting interneurons in CA1. Specific types of neural codes are believed to have different functional roles. Neural coding with burst spikes is known to support hippocampal contributions to classical contextual fear conditioning (CFC). In CFC the hippocampus encodes the multisensory context as a conditioned stimulus (CS), whose burst spiking output is paired with the aversive unconditioned stimulus (US) in the amygdala, allowing for fear memory recall upon future exposure to the CS. To investigate the contribution of Som+ interneurons to this behavior, I designed a CFC task for head-fixed mice, allowing for optical recording and manipulation of activity in defined CA1 cell types during learning. Pharmacogenetic inactivation of CA1 Som+ interneurons, but not Pvalb+ interneurons, prevented the encoding of CFC. 2-photon Ca2+ imaging revealed that during CFC the US activated CA1 Som+ interneurons via cholinergic input from the medial septum, driving inhibition to the PC distal dendrites that receive coincident excitatory input from the entorhinal cortex. Inactivating Som+ interneurons increases PC population activity, and suppressing dendritic inhibition during the US alone is sufficient to prevent fear learning. These results suggest sensory features of the US reach CA1 PCs through entorhinal inputs, and thus require active inhibitory filtering by Som+ interneurons to ensure hippocampal output exclusively encodes the CS during CFC. In conclusion, I found that Som+ interneurons in CA1 are an effective regulator of PC burst spiking because they inhibit dendritic electrogenesis. This function is used by the hippocampus to prevent the US from influencing the burst spike output of PCs that encode the CS, ensuring successful CFC. This work bridges the gap between cells, circuits, and behavior, and provides mechanistic insight into one of our most essential cognitive functions - the ability to learn and remember.

To efficiently process natural environments, many species have sensory systems that selectively encode behaviorally relevant information. Vocal communicators such as humans and songbirds rely on their auditory systems to recognize vocalizations and to extract vocalizations from complex auditory scenes. Yet many of the neural correlates of these perceptual abilities remain poorly understood. In this dissertation, I describe neural mechanisms by which the songbird auditory system produces sparse, informative and background-invariant neural representations of vocalizations. First, I show that auditory midbrain neurons encode vocalizations differently than other complex sounds, and that subthreshold excitation and inhibition may facilitate stimulus-dependent encoding of vocalizations. Second, I show that the responses of individual midbrain neurons can be unreliable, and that pooling the responses of correlated and similarly tuned neurons facilitates the neural discrimination of vocalizations. Third, I show that sparse coding neurons in the songbird forebrain extract individual vocalizations from auditory scenes at signal-to-noise ratios that match behavior. Lastly, I show that a simple neural circuit of delayed inhibition transforms a dense and background-sensitive neural representation into a sparse and background-invariant representation, in as little as one synapse. Together, these findings illuminate previously unknown mechanisms for selective vocalization coding, suggest a behaviorally relevant role for the ubiquitous phenomenon of sparse neural coding, and provide a neural correlate for the perceptual extraction of vocalizations from complex auditory scenes.

Mood disorders are chronic and debilitating psychiatric diseases that affect 450 million people worldwide. Despite their overwhelming prevalence, the etiology and pathophysiology of these disorders are poorly understood. As a result, mood disorders are diagnosed by symptom presentation, not disease processes. Furthermore, our incomplete understanding of the biological underpinnings of these disorders is a major impediment to the development of effective treatments. Animal models offer a tractable means of examining the molecular and cellular processes that contribute to the pathogenesis of psychiatric disorders. Chronic social defeat (SD) stress is a novel ethologically-relevant mouse model of affective psychopathology. Like all animal models, face, construct, and predictive validity must first be established for SD before findings in this model can be extrapolated to the clinic. Though the depressive-like and anxious phenotypes induced by SD are well-established, cognitive symptoms have yet to be validated. As cognitive impairment is a significant but understudied core symptom of affective disorders, we sought to determine if SD would recapitulate this dimension of psychopathology. First we confirmed that SD induced depressive-like and anxious behavior (ethological validity), as well as decreased adult neurogenesis in the dentate gyrus of the hippocampus--an established correlate of depressive behavior (biomarker validity), in our experimental mouse strain. We then tested mice in two learning paradigms: 1-shock contextual fear conditioning (CFC) and novel object recognition (NOR). SD mice were significantly impaired in CFC fear memory recall, as well as in NOR. Having identified a robust cognitive impairment in 1-shock CFC, we sought to locate a neural correlate of this deficit. As both 1-shock CFC and NOR are hippocampus-dependent tasks, and knowing that SD alters hippocampal architecture by decreasing adult hippocampal neurogenesis, we chose to examine cellular activity patterns in the dentate gyrus and its downstream target, CA3. We found that impaired fear expression during context re-exposure correlated with decreased reactivation in CA3. Having confirmed SD as a viable model for the study of affective disorders, we then used this model to explore the antidepressant potential of ketamine. Classical antidepressants have a delayed onset of therapeutic efficacy of approximately four to six weeks. Ketamine, an NMDA receptor antagonist, has recently been identified as a rapid-acting antidepressant in humans. In order to explore ketamine's antidepressant mechanism of action, mouse models of ketamine administration need to be established and optimized. Though several groups have begun to investigate the antidepressant effect of ketamine in mice, dose, strain, and behavioral paradigms have yet to be systematically titrated. We found only a modest antidepressant effect of ketamine following SD. In conjunction with other murine ketamine studies, this modest effect argues for a more rigorous optimization of ketamine administration paradigms in mice. We next sought to determine if ketamine could protect against the induction of psychopathology. Psychiatric disorders are not traditionally approached from a preventive perspective. This is in part because the etiology of these disorders remains largely unknown. It is known, however, that stress can precipitate affective disorders such as major depressive disorder and post-traumatic stress disorder, as well as trigger symptomatic episodes in patients with prior psychiatric diagnoses. However, stress does not ubiquitously induce psychopathology in all exposed individuals. Stress resilience, the capability to withstand stress without developing an affective disorder, varies across individuals. Using the SD model of chronic stress, we sought to determine if ketamine could enhance stress resilience, thereby protecting mice from the depressive-like sequelae of chronic stress exposure. We found that a single subanesthetic dose of ketamine was protective against stress-induced depressive-like behavior for at least three weeks following administration. The drug was given a week prior to SD and, as ketamine has a half-life of only a few hours, was fully washed out by the commencement of the stress paradigm. This finding demonstrates that ketamine induces an actively self-maintaining form of resilience. Though we observed a protective effect of ketamine out to three weeks, it is possible that this effect persists even longer in duration. If this prophylactic effect translates to humans, ketamine could potentially be used in at-risk populations, such as active-duty soldiers, to inoculate against stress-induced psychopathology. In summary, this thesis establishes chronic social defeat stress as a valid model of the cognitive behavioral symptoms of affective disorders, as well as identifies decreased reactivation in CA3 as a cellular correlate of stress-induced cognitive impairment. Furthermore, we find that ketamine has a modest antidepressant effect when administered following two weeks of defeat stress. Most importantly, however, we show that a single subanesthetic dose of ketamine can induce robust, long-lasting, self-maintaining stress resilience. The work in this thesis establishes prophylactic ketamine as a novel model of stress resilience and identifies ketamine as the first clinic-ready pharmaceutical with the potential to prevent psychiatric disorders.

Lineage commitment of pluripotent cells is a critical step in the development of multicellular organisms and a prerequisite for efficient differentiation of stem cells into terminal cell types. During successful neuroectodermal lineage commitment, extracellular signals terminate the pluripotency program, activate neural transcriptional program, and suppress alternative mesendodermal fate. Retinoic acid (RA) has been identified as a potent inducer of neural differentiation in embryonic stem cells (ESCs), yet the transcriptional program initiated by RA is poorly understood. Expression profiling of differentiating ESCs revealed delayed response of the pluripotency marker Oct4 and neural marker Sox1 following RA treatment, suggesting that RA regulates the pluripotency program and neural transcriptional program indirectly through induction of additional transcription factors. In this study, I identified a zinc finger factor Zfp703 as a downstream effector of RA-mediated neuroectodermal lineage commitment. Zfp703 expression in ESCs resulted in Oct4 repression, Sox1 induction, and neural differentiation. Moreover, Zfp703 strongly suppresses mesendodermal fate by repressing genes such as Brachyury, Eomes, and Mixl1 even under conditions favoring mesendoderm specification. Zfp703 binds to and represses Lef1 promoter, raising the possibility that it might modulate Wnt signaling via regulating Lef1. Finally, Zfp703 is not required for RA-mediated Oct4 repression and Sox1 induction. However, it is necessary for efficient Brachyury repression by RA. Based on these data, I propose that Zfp703 is involved in the transcription regulation during neural progenitor specification. Through downregulating of both mesendodernal fate and pluripotency, Zfp703 de-represses neural transcriptional program and indirectly promotes the default neuroectodermal lineage commitment.

Any animal must be able to predict and cancel the sensory consequences of its own movements to avoid ambiguity in the origin of sensory input. Theoretical and human behavioral studies suggest that nervous systems contain internal models that use copies of outgoing motor signals along with incoming sensory feedback to predict the consequences of movements. Many studies propose the cerebellum as one possible site of such internal models. Yet whether such an internal model exists and how such an internal model might be implemented in neural circuits is largely speculative. Early work in cerebellum-like structures of mormyrid fish identified neural mechanisms of sensory predictions at the levels of synapses, cells, and circuits, and successfully linked those mechanisms to the systems-level function--the cancellation of electrosensory input due to the fish's own behavior. However, those early studies were restricted to predicting and cancelling the electrosensory consequences of relatively simple and rather specialized electromotor behavior. The research described here takes an in vivo electrophysiological approach to generalize the previous work in mormyrid fish to the more ubiquitous problem of predicting and cancelling the sensory consequences of movements. First, I demonstrate that neurons in the electrosensory lobe of weakly electric mormyrid fish generate predictions at the cellular level, termed negative images, about the sensory consequences of the fish's own movements based on ascending spinal corollary discharge signals. Second, I examine the interactions between corollary discharge and proprioceptive feedback under conditions that simulate real movements. Using experiments and modeling, I show that plasticity acting on random, nonlinear mixtures of corollary discharge and proprioceptive signals can account for key properties of negative images observed in vivo. Mossy fibers originating in the spinal cord carry randomly mixed, though linear, corollary discharge and proprioceptive signals, while properties of granule cells observed in vivo are consistent with a nonlinear re-coding of these signals. The conclusion of these studies is that both corollary discharge and proprioception, in combination with an associative neural network endowed with synaptic plasticity, provide a powerful and flexible basis for solving the ubiquitous problems of predicting the sensory consequences of movements.

Layer 1 (L1) of the cerebral cortex is a largely acellular layer that consists mainly of long-range projection axons and apical dendrites of deeper pyramidal neurons. In the rodent barrel cortex, L1 contains axons from both higher motor and sensory areas of the brain. Despite the abundance of synapses in L1 their actual contribution to sensory processing remains unknown. We investigated the impact of activating long-range axons on barrel cortex L2/3 pyramidal neurons in vivo using a combination of optogenetics and eletrophysiological techniques. The reason we target our investigation on L2/3 is because of its well-known sparse sensory responses. We hypothesize that long-range top-down inputs via L1 can provide the additional inputs necessary to unleash L2/3 and strongly influence sensory processing in S1. We focused on three main sources of BC-projecting synapses: the posterior medial nucleus of the thalamus (POm, the secondary somatosensory nucleus), the primary motor cortex (M1), and the secondary somatosensory cortex (S2). Here we report that while activation of POm axons elicits strong EPSPs in most recorded L2/3 cells, activation of M1 or S2 axons elicited small or no detectable responses. Only POm activation boosted sensory responses in L2/3 pyramidal neurons. We also found that during wakefulness and under sedation, POM activation not only elicited a strong fast-onset EPSP in L2/3 neurons, but also a delayed persistent response. Pharmacological inactivation of POM abolished this persistent response but not the initial synaptic volley to L2/3. We conclude that the persistent response requires intrathalamic or thalamocortical circuits and cannot be mediated by specialized synaptic terminals or intracortical circuitry. Overall, our study suggests that the higher order thalamic nucleus provides more powerful network effect on L2/3 sensory processing than higher order cortical feedback inputs. POm activation not only directly boosts L2/3 sensory responses, but is also capable of influencing S1 signal processing for prolonged periods of time after stimulus onset and can potentially be important for other cognitive aspects of sensory computation.

Animals use their nervous system to translate signals from their sensory environment into appropriate behavioral responses. In some cases, these responses are hard-wired through genetic sculpting of neural circuits, such that certain stimuli drive innate behavioral responses in the absence of prior experience [Ewert, Burghagen, and Schurg Pfeiffer 1983; Yilmaz and Meister 2013; Wu et al 2014]. But most often, responses to stimuli are modified over the course of an organism's lifetime via associative learning, in which past experience is used to adaptively modify the neural circuits controlling behavior. The remarkable regularity of cerebellar circuitry made it an early target of experiments seeking a link between neural circuit structure and computational function (Eccles, Ito, and Szentgothai, 1967). These efforts led to a first generation of models describing cerebellar cortex as a device for associative learning, remarkable for their focus on linking each cell type of cerebellar cortex to a computational aspect of associative memory formation and adaptive control ([Marr 1969; Albus 1971; Ito 1972). In subsequent decades, specialized neural architecture resembling that of the cerebellum has been identified in several other brain regions, including the dorsal cochlear nucleus of most mammals (Oertel and Young, 2004), the mushroom body of the insect olfactory system (Farris, 2011), and a region evolutionarily and developmentally related to the cerebellum in the brains of weakly electric fish, the electrosensory lobe (Bell, Han, and Sawtell, 2008). This has raised the hope that a similar computational mechanism is at work in these structures. It is not easy to find behavioral paradigms that isolate learning in the cerebellum, and a complete mechanistic account of learning during commonly studied behaviors has remained elusive. In this thesis, I analyze two cerebellum like structures, the electrosensory lobe of the mormyrid fish and the mushroom body of the fly olfactory system, in which mapping out associative learning is more tractable, due to the availability of well controlled learning paradigms and the development of powerful biochemical and genetic techniques. With the help of my experimental collaborators, I constructed computational models of the electrosensory lobe and mushroom body from electrophysiological and anatomical data, and studied the process of associative learning in these models. In both systems, an initial sensory representation is first projected up into a high dimensional space, and then read out via convergent input onto individual neurons. Learning adjusts the input to readout neurons, causing changes in their responses to future stimuli that alters their drive to downstream nuclei. Two details shape how each circuit handles associative learning: the way in which sensory inputs are represented, and the mechanism of learning. Together, these two pieces determine what transformations each circuit is able to learn and how it generalizes after learning. In the four chapters of this thesis I present four related projects dealing with sensory representation and learning in cerebellum-like structures. The first chapter has previously been published as a paper and describes a model for cancellation of self generated sensory input in the passive electrosensory system of the mormyrid fish. In the second chapter, I adapt this model to a more high dimensional cancellation problem in the fish's active electrosensory system, which deals with the effects of the fish's body on the electric fields it generates. In the next two chapters, I construct a network model of odor representation in fly olfactory system, terminating at the mushroom body. Finally, I use this model in conjunction with recent experimental findings on the output of the mushroom body, to build a model of associative odor learning in the fly.

The present thesis is divided in two parts. In the first part I discuss two modeling efforts to analyze extracellularly recorded spiking activity of auditory neurons. First, in Chapter 2 I introduce a receptive field estimation method based on a generalized linear model with a sparse prior (L1-GLM). I apply this method to the estimation of spectro-temporal receptive fields (STRFs) of songbird auditory midbrain neurons from natural and synthetic stimuli, and show that the L1-GLM outperforms a traditionally used STRF estimation method by reducing estimation biases and increasing predictive power. Second, in Chapter 3 I describe a computationally efficient approach to the spike sorting problem that can automatically track non-stationarities in electrophysiological recordings. In the second part of this thesis I describe a series of electrophysiological experiments and computational tools for characterizing several information coding properties of single cells and ensembles of cells in the songbird primary auditory cortex (A1). In Chapter 4 I demonstrate that, despite the absence of a laminar structure, the avian A1 displays the same information coding principles that define the canonical cortical microcircuit in mammals, at the single neuron, cell types and pairwise interaction levels. Lastly, in Chapter 5, I study the emergence of song selectivity in the songbird A1 and demonstrate that vocalization selectivity is a network-level effect rather than a single cell property. I show that increased firing of single neurons to songs occurs jointly with a decrease in trial to trial variability in song responses that is shared across neurons in the population. Using a probabilistic model of population responses I characterize the spatial and temporal structure of shared response variability, providing insight into the potential mechanisms underlying vocalization selectivity in the songbird primary auditory cortex. The results presented in this chapter are the product of a collaborative effort between myself and Lars Buesing (Columbia University).

Spinal circuits can generate locomotor output in the absence of sensory or descending input, but the principles of locomotor circuit organization remain unclear. We sought insight into these principles by considering the elaboration of locomotor circuits across evolution. The identity of limb-innervating motor neurons was reverted to a state resembling that of motor neurons that direct undulatory swimming in primitive aquatic vertebrates, permitting assessment of the role of motor neuron identity in determining locomotor pattern. Two-photon imaging was coupled with spike inference to measure locomotor firing in hundreds of motor neurons in isolated mouse spinal cords. In wild type preparations we observed sequential recruitment of motor neurons innervating flexor muscles controlling progressively more distal joints. Strikingly, after reversion of motor neuron identity virtually all firing patterns became distinctly flexor-like. Our interneuron imaging experiments demonstrate a new approach for functionally mapping the types of inputs that motor neurons might receive during locomotor firing. These data revealed that En1-derived inhibitory spinal interneuron activity appears to be dominated by a flexor-like pattern across the ventrolateral extent of the lumbar spinal cord–even in the regions surrounding flexor and extensor motor pools. Together, these findings show that motor neuron identity directs locomotor circuit wiring, and indicate the evolutionary primacy of flexor pattern generation.