Control of Neural Activity with Adaptive, Closed-Loop Patterned Stimulation

Abstract

Intelligent, complex goal-directed behaviors arise from the brain’s ability to exert cognitive control. Understanding how the brain exerts such control can provide insights into its dysfunctions, which are implicated in a myriad of neurological diseases. The ability to experimentally control neural activity is therefore an important tool for understanding and testing theories of cognitive control, for which causal manipulation of neural activity is a critical component. Importantly, our current understanding of the mechanisms underlying cognitive control is limited. Clinically, stimulation of neural activity has also emerged as a treatment for neurological diseases. However, current stimulation approaches are limited in their ability to control the brain: they typically control relatively few sets of neurons and treat those neurons uniformly. Complex behaviors require communication between many neural populations and cortical regions. Thus, nuanced control over behavior requires stimulation approaches that can independently control neural populations and, in turn, flexibly route information between neural populations. This thesis aims to develop technologies to control large scale neural activity and behavior. To this end, we describe a framework for adaptive closed-loop stimulation (ACLS) and its implementation. This system pairs a machine learning algorithm with bi-directional electrophysiology to learn a stimulation pattern that evokes a desired response. We show that the ACLS can both generate arbitrary patterns of neural activity and replicate responses from visual stimulation in mouse primary visual cortex. Next, we develop both a behavioral training system and a method to test the system’s ability to control large-scale neural dynamics. We demonstrate successful implementation of an efficient behaviorally trained task and capture large scale neural dynamics. Due to restrictions of the COVID-19 pandemic, we articulate two alternate hypotheses and propose analysis for this data in the discussion.