Understanding how the canonical circuit implements computations requires that we characterize the components of the circuit and determine how they interact. Cortical components are often equated to the various classes of excitatory and inhibitory neuronal cell-types that occupy the six cortical layers. Excitatory cortical neurons are known to exhibit stereotyped, cell-type specific patterns of connectivity; these excitatory synaptic pathways are a central feature of cortical organization. It has been less clear whether inhibitory neurons are organized according to a similar logic. In this thesis I describe my efforts to map the inhibitory pathways of the neocortex, using the mouse primary somatosensory barrel cortex as a model.

In Chapter 1, I will give an introduction to theories of cortical function and general overview of the organization of cortical circuits. I will also provide a focused review of the inhibitory circuitry in layer 5 of the cortex. This chapter is presented, in part, in the form of reproductions of material from two reviews (one published and one in preparation) that I co-authored with my advisor. In Chapter 2, I describe the discovery of a novel disynaptic inhibitory pathway between layer 4 and layer 5 of the somatosensory cortex, and present evidence that this pathway contributes to selective sensory representations in layer 5. This work is presented as a published manuscript, which I co-authored with Dr. Scott Pluta. In Chapter 3, I describe a series of experiments probing the organization of two subnetworks of dendrite-targeting interneurons. These subnetworks exhibit complementary patterns of connectivity which enable them to differentially modulate the dynamics of cortical activity in a layer-specific manner. This work is presented as a first author manuscript (currently in review). Finally, in Chapter 4 I offer closing thoughts and directions for future work.

Circuit-mapping experiments combining whole-cell electrophysiology with two-photonoptical stimulation of potentially presynaptic neurons (“2p-mapping”) have produced rich data on monosynaptic connectivity of neural circuits. However, mapping densely- packed presynaptic populations at cellular resolution has proven challenging, making the precise localization and identity of connected neurons difficult. To interpret data resulting from these experiments, it is therefore critical to develop statistical methods which can infer the properties of neural circuits despite the limited spatial resolution of state-of-the-art 2p-mapping technologies. In this work, we present several novel statistical methods for mapping neural circuits at cellular resolution. In Chapter 1, we present background on mapping neural circuits and the methods used. In Chap- ter 2, we present a novel statistical method for inferring monosynaptic connections when the resolution of two-photon stimulation is not sufficient given the density of the opsin-expressing cells. In Chapter 3, we present a high-resolution approach for infer- ring common input to two simultaneously patched neurons during mapping. And in Chapter 4, we present a high accuracy method for inferring the timing of post-synaptic currents - the key response observation when mapping neural circuits.

Abstract

The speed and reliability of computation in neural circuits depends on fast chemical transmission between neurons in the brain. Excitatory transmission in the central nervous relies on the activity-dependent release of glutamate onto ionotropic receptors clustered at the postsynaptic membrane. AMPA receptors mediate the majority of synaptic communication in the mature central nervous system, but NMDA receptors play a critical role under conditions of strong or correlated activity. The precise mechanisms governing the number and subtype of AMPA receptor at synapses are essential for the proper regulation of synaptic strength. Recent work has identified many neuronal proteins involved in the trafficking and scaffolding of AMPA receptors at synaptic sites and has begun to clarify how these proteins interact to support excitatory synaptic transmission, though several fundamental questions remain.

To assess directly the trafficking of AMPA receptors in real time, we used a novel, photo-activatable, irreversible AMPA receptor antagonist. Inactivation of surface receptors reveals that AMPA receptors cycle rapidly between an intracellular domain and the extrasynaptic somatic membrane. The total surface pool at synaptic sites is exchanged with intracellular receptors every 18 hours. Following insertion to the extrasynaptic domain, AMPA receptors can also move laterally within the plasma membrane where they can ultimately stabilize at postsynaptic densities.

Trafficking of AMPA receptors to the neural surface requires isoforms of the protein stargazin. Stargazin controls AMPA receptor trafficking by an interaction involving its intracellular c-terminus. Surprisingly, stargazin also modulates AMPA receptor gating through its first extracellular domain, which acts to stabilize the open state of the AMPA receptor channel. Overexpression of a stargazin with a mutated first extracellular domain reduces charge movement during transmission by accelerating the decay of postsynaptic quantal events. This suggests that stargazin acts as an AMPA receptor auxiliary subunit and is necessary for the normal function of AMPA receptors during ongoing synaptic transmission.

Stargazin recruits AMPA receptors to synapses by its interaction with the postsynaptic scaffolding protein, PSD-95. Through an additional interaction, PSD-95 binds to the transmembrane protein ADAM22, which forms a receptor/ligand complex with the secreted factor Lgi1. Binding of Lgi1 to ADAM22 results in an increase in AMPA-receptor mediated synaptic transmission. This newly identified signaling complex may represent an alternate pathway of AMPA receptor recruitment to synapses.

Long-term potentiation (LTP) at the CA3 to CA1 synapse is thought to rely on the rapid insertion of AMPA receptors into the synaptic membrane following strong NMDA receptor activation. However, LTP does not involve the insertion of calcium permeable AMPA receptors that lack the GluR2 subunit, nor does LTP require ongoing activity to be maintained.

LTP-like processes are believed to play a major role in wiring neural circuitry during development by recruiting AMPA receptors to nascent synapses allowing them to sense glutamate. Although LTP is critically dependent on activation of the NMDA receptor, deletion of the essential NMDA subunit, NR1, in pyramidal neurons results in a net increase in AMPA-receptor mediated transmission. When the deletion occurs embryonically, synapses form in normal number but exhibit increased quantal strength. When NR1 is deleted postnatally in oganotypic culture, synapses exhibit normal quantal strength, but increase dramatically in number. Reintroduction of NR1 to the knockout results in a net loss of synapse number. Thus NMDA receptor activity limits rather than promotes synaptic maturation during brain ontogeny.

The supralinearity of neural responses shapes neurons’ activity. When depolarized close to threshold, they sensitively transduce small changes in their input; however, when hyperpolarized, they fail to spike despite large changes in their input. This means that a neuron which combines inputs from across sensory cortical space will nonlinearly encode stimuli, representing a conjunction of multiple stimuli differently from the sum of their parts. Further, via interaction with synaptic connections, this supralinearity has profound consequences for how signals propagate between nearby neurons. Signals might strongly propagate through highly active neurons in the network, while not propagating through inactive neurons. This shapes the functional interaction of neurons across sensory space: two neurons which cooperate under one stimulus condition, each mutually driving activity in the other, might compete in another condition. Further, the net effect of neural connections might be to supralinearly amplify some part of their inputs over another, shaping how their signals are added across space. In this work, we explore several questions related to the consequences of supralinearity for neurons’ encoding of stimuli of varying spatial extent and complexity.

First, we found a sparse but comprehensive population code for higher order tactile and visual features that depended on a heterogeneous and neuron-specific logic of spatial summation beyond the receptive field. Different pyramidal cells (PCs) in primary somatosensory (S1) and primary visual (V1) cortex, summed most combinations of sensory inputs sublinearly, but integrated specific inputs supralinearly, leading to selective responses to higher order features. This may explain how the brain exploits the thalamocortical expansion of dimensionality to encode arbitrary complex features of sensory stimuli.

Next, in V1, PCs integrate widely across space when signals are weak, but narrowly when signals are strong, a phenomenon known as contrast-dependent surround suppression. Theoretical work has proposed that local interneurons could mediate a shift from cooperation to competition of PCs across cortical space, underlying this computation. Our calcium imaging and optogenetic data, along with computational modeling, showed that recurrent amplification drives a transition from a positive PC→VIP⊣SST⊣PC feedback loop at small size and low contrast to a negative PC→SST⊣PC feedback loop at large size and high contrast to contribute to this flexible computation. This may represent a widespread mechanism for gating competition across cortical space to optimally meet task demands.

Finally, this mechanism has been proposed to support integration of weak signals across space to support sensitive detection. For example, observers are better able to detect large low contrast stimuli than small low contrast stimuli, obeying a power law relationship between contrast sensitivity and stimulus size, and several psychophysical models have related the exponent in the power law relationship to the intrinsic nonlinearity of contrast coding in the visual system. We measured contrast sensitivity vs. stimulus size for the first time in mice and combined this with calcium imaging and optogenetics to constrain the role of V1 in this computation. V1 was necessary both for low contrast detection thresholds and, based on a psychophysical model, nonlinear encoding of contrast. This mechanism could provide a general framework for understanding sensory cortices’ role in flexibly assigning cognitive significance to arbitrary small environmental changes. Together, these studies of sensory encoding of space, ranging from the scale of synaptic connections to behavior, shed light on the role of nonlinearities in primary sensory cortex in sensory processing.

The cerebral cortex is responsible for neural functions ranging from basic sensory processing to complex decision making. However, the underlying neural processes and the precise function of some cortical areas are not well understood. For humans the cortex is invaluable, providing us with our most powerful cognitive traits. For some species, the cortex can be removed leaving only minor deficits. Stark contrasts like these blur the overall function and importance of cortex. Recent advancements in recording and analysis technology present us with a unique opportunity to search for neural processes previously impossible to find. Recent work has found cortical regions in mice that do more than their functional name implies, in contrast to the specialized cortical regions in higher order species. In this study we explore the hidden functions of mouse motor cortex and elaborate on its role in the whisker system. The mouse whisker system is traditionally divided into sensory regions and motor regions, with sensory cortex (vS1) and motor cortex (vM1) sitting on the border. The functional divisions between sensory and motor cortex have recently blurred, both regions able to drive movements and encode sensory information. Whisker motor cortex, specifically, exhibits disparate functions, making it an ideal place to study multirole cortex. Here we take advantage of advanced neural recording and analysis techniques and uncover a motor cortex that acts as a sensory and possibly a higher order cortical region as well.In chapter 1, I provide an overview of how cortical regions were defined and then summarize how modern research is blurring the lines between functionally defined cortical areas. I then introduce sensory processing in the mouse whisker system with a focus on motor cortex, a cortical region where its function is increasingly blurring, and then describe how my research explored non-motor functionality in a motor region. In chapter 2, I present my first first-author publication where we determined if somatosensory cortex integrates sensory information over short or long timescales in order to estimate “mean” variables. In this work I first use neural decoding to quantify how well each neuron represents pieces of sensory information and find that some neurons correlate with choice. Chapter 3 is work that I collaborated on, we determined how multiple sensors contribute to the receptive field of individual neurons and the broader population. We discovered a map of sensory space distributed across somatosensory cortex and determined the map was dependent on neurons integrating information from multiple sensors in parallel. Chapter 4, we explore whether somatosensory cortex is necessary for a whisker dependent discrimination task and further determine what arrangement of sensors are required for task completion. Finally, in Chapter 5 I present my main project investigating vibrissae motor cortex. Here I study the sensory responses present in motor cortex, quantifying vM1 sensory tuning for the first time, and ultimately determining that vS1 does not drive the activity as previously thought. Incredibly, vM1 is not required for whisker movements in general but is required during demanding whisker dependent contexts, such as a whisker discrimination task, affecting both choice, whisker movements, and onset of lick response. Finally, in Chapter 6, I summarize what this tells us about sensory processing, propose some ideas for future research, and discuss how modern tools can enable us to find hidden functionality.

Organisms scan their sensors around their environment to build an internal representation of that environment in a process known as active sensation. The integration of information across time and space is critical to providing context as to what is the organism is perceiving. However, the neural circuits that encode and underlie the integration of incoming sensory information have predominantly been studied in the context of passive sensation. Studying these circuits in the context of active sensation is imperative to generating a better understanding of how the brain naturally encodes sensation. This would have profound impacts on understanding the mechanisms of a number of neural disorders, including autism and attention-deficit/hyperactivity disorder, as well as how to improve the acuity of artificial sensation implanted into disabled individuals.

To better understand how the mammalian brain encodes and integrates information during active sensation, my collaborators and I developed several novel paradigms to study surround integration in the mouse barrel cortex during active whisking. In Chapter 1 I establish why this is an important problem, and briefly summarize what is already known about sensory coding in the mouse whisker system. In Chapter 2 my collaborators and I probe how mice represent the location of an object within its whisking field, and how the integration of information across surround whiskers affects this representation. In doing so we discover a novel thalamocortical transformation where surround integration in the cortex suppresses activity in layer 4 of the cortex, ultimately generating a smooth map of scanned space in cortical layer 2/3. In Chapter 3 I utilize a novel tactile display to better understand the logic of multi-whisker integration in two cortical layers. In this unpublished work, I show that contrary to the previous literature in anesthetized mice, cortical neurons in awake, whisking mice powerfully summate specific whisker combinations supralinearly, generating a sparse code representing the entire combinatoric space of whisker touch. In Chapter 4, I conclude with some closing thoughts and propose some future lines of inquiry to further this research.

Understanding what aspects of neural activity causally drive sensory perception is essential to understanding the nature of the neural code. Direct perturbation of neural activity is needed to conclusively distinguish between competing theories, and in many cases precise perturbations that alter neural activity patterns in specific ways are needed to distinguish coding mechanisms. Optogenetics has ushered in a new era of causal manipulation, but conventional methods have several shortcomings. We addressed several of these and used our new methods to study detection of contrast in the mouse visual system, focusing on the primary visual cortex (V1). We first established that visual cortex provides necessary information for contrast detection. To do this we developed an all-optical “reversible lesion” paradigm to enable interpretation of the necessity of a region for a task in the face of oft-contradicting optogenetic and mechanical lesion results. Next, we leveraged new 2-photon targeted optogenetics to test a long-held theory that the more stimulus sensitive neurons, i.e. the neurons that carry the most information about the stimulus, are responsible for detection and perception of that stimulus. We stimulated small ensembles of V1 neurons during visual task performance and found that small groups could exert a large influence on behavioral output. Surprisingly, the visual response properties of the stimulated neurons had no bearing on their behavioral influence. Instead, the network influence of the ensemble strongly predicted the behavioral effect of stimulation. This result supports the idea that perception requires large-scale network activity and suggests that the neural code for stimulus detection may be a simple summation of activity. Finally, we developed a new method to enable testing theories of coding that go beyond neural identity, such as the high-dimensional geometry of population codes or spike synchrony. These ideas are not testable with conventional targeted optogenetics because of the inability to write specific spike trains. To overcome this obstacle, we first developed a transgenic mouse line that co-expresses a soma-targeted opsin fused to a strong calcium indicator for all-optical experiments. We leveraged this line to create a paradigm for all-optical read-out and write-in of precise population vectors of neural activity using cell-by-cell calibration. This will enable new testing of how the topology of the neural code influences behavior. All together, these studies establish the role of visual cortex in contrast detection and argue that the code for detection is a simple summation of all excitatory activity within visual cortex. The new methods developed in Chapter 4 should enable further investigation and confirmation of this theory, and the methods presented in Chapters 2-4 can be applied to other sensory tasks to determine how these conclusions generalize across regions and tasks.