QUANTIFYING, CODIFYING, AND CONTROLLING CORTICAL NEURAL POPULATION DYNAMICS

Abstract

Cognition arises from the processing of information by populations of neurons across the brain. Neural populations encode and transform information about the surrounding world and our internal state to enable behaviors that match our current demands. Disrupting the patterns of activity across neural populations disrupts cognition and is thought to underlie many neurological and neuropsychiatric conditions. Despite their importance, the ways in which neural populations represent, transform, and communicate information remain poorly understood. This dissertation presents a set of experiments that aim to understand the form, function, and control of neural population dynamics. First, we use a novel analytical approach to identify spatiotemporal ‘motifs’ that capture the moment-to-moment flow in neural activity across the cerebral cortex of mice. We establish that a relatively small set of motifs capture the majority of neural population dynamics. These motifs are shared across mice and extend to a diversity of behavioral contexts, suggesting that a low-dimensional (few in number) set of neural population dynamics may facilitate efficient control of communication between brain areas. Second, we probe the relationship between these motifs and individuals’ behavioral phenotypes. We identify unique sets of motifs related to either motor or sensory-memory processes and find evidence that the sampling of interactions between neural populations supports behavioral individuality. Third, we test a flexible framework for controlling neural population dynamics. We demonstrate that this approach, which uses ‘model free’ learning algorithms to identify electrical stimulation patterns that elicit specific neural responses, can learn to recapitulate the natural neural response to visual stimuli in awake mice. Collectively, the work presented in this dissertation advances both our scientific understanding of neural population dynamics and our methodological approaches for quantifying and manipulating these dynamics. Our results reveal that neural population dynamics are organized such that a parsimonious set of interactions can support a diversity of behaviors and establish that ‘model-free’ approaches for brain stimulation allow flexible control of activity across neural populations. In this way, this dissertation sets the stage for research into the mechanisms governing the flow of information across the brain, and the clinical manipulation of neural population dynamics to treat disease.