Contributions of Spatial Orienting Networks to Abstract Cognition

A half-century of neurophysiological recordings from single electrodes established a `localized' viewpoint on function in the brain–that complex behavior occurs through computations and representations carried out across distinct brain areas, each of which plays a specialized role. New techniques for specific, high-throughput measurement of neuronal activity and behavior in rodents have prompted an alternative viewpoint, which posits that neural encoding of behavioral variables is distributed across a wide range of areas: `everything, everywhere, all at once'. In Chapter 1, we begin by briefly introducing these paradigms and evaluating which of them better describes cognition—the manipulation of internal variables that enables flexible behavior. Measurements of neuronal activity in rodents suggest that cognitive variables are reflected broadly across the brain, but generally during behaviors that do not fully dissociate cognition from response preparation. Neuronal recordings in non-human primates performing tasks that achieve this dissociation similarly implicate a distributed network—including, to a surprising degree, regions engaged in controlling movement. We close the chapter by discussing why cognitive signals may appear in such areas, as well as the factors that affect the breadth of the brain-wide network that is recruited for cognition. These last questions serve as the foundation for Chapters 2 and 3, my two primary thesis projects. Both chapters focused on better understanding the contributions that brain regions with roles in spatial processing and movement make to cognition. In Chapter 2, I describe a set of observations implicating the primate oculomotor system surprisingly centrally in abstract categorization. These observations emanated from a first-of-its-kind dataset of neuronal recordings and behavioral measurements from monkeys--spanning more than 10 unique subjects, 1000 recording sessions, and a quarter million correctly-completed trials of behavior. Because all datasets were collected in the context of the same cognitive task, this large dataset afforded us the unique opportunity to discover a robust, uninstructed feature of monkeys' behavior: tiny eye movements correlated with the categories of stimuli the subjects needed to hold in mind. We found that these eye movements also belied a broader trend borne out in the neural data: of the variety of brain regions implicated in visual and cognitive functions from which our lab had obtained recordings, oculomotor areas appeared to be especially strongly engaged in representing the abstract, experience-dependent categories the subjects used to guide their decisions during this task. This effect was further context-specific, reflecting the extent to which subjects' decisions required abstraction. These findings broaden our view of the cortical/subcortical network engaged in complex cognitive behavior, and underscore the potential for using large high-quality datasets to mine fundamental insights from small features of behavior. Motivated by these findings, I carried out an electrophysiology experiment investigating the engagement of the frontal eye fields (FEF) in category-guided movement selection (Chapter 3). Our previous work demonstrating engagement of oculomotor/spatial orienting areas by abstract categorization had a largely representational flavor, and (by design) left open questions of how these representations come to influence subjects' behavior. We addressed this through a carefully-designed abstract-memory-guided saccade task, during which subjects made saccades to a common set of targets across a variety of contexts. We found that FEF maintains a persistent, stable encoding of abstract memory content during this task, even as subjects plan and execute unrelated eye movements. These data demonstrate that engagement of the oculomotor and spatial orienting system in cognition noted in Chapter 2 extends to the frontal eye fields. Analysis of activity dynamics as subjects use abstract memory content to choose where to look also illuminates the initial contours of a general class of mechanism whereby internal states can affect behavior.