BME PhD Dissertation Defense: Tian Wang

Title: "Neural Mechanisms Underlying Flexible Decision Making in Prefrontal and Premotor Cortex”

Advisory Committee: Chandramouli Chandrasekaran, PhD – Anatomy & Neurobiology, Psychological and Brain Sciences, BME (Research Advisor) Kamal Sen, PhD – BME (Chair) Michael Economo, PhD –BME Brian DePasquale, PhD – BME Jonathan Kao, PhD – ECE, UCLA

Abstract: The neural mechanisms underlying flexible perceptual decisions, where the mapping between sensory input and motor action changes depending on context, remain unclear. Here, we trained monkeys to discriminate the dominant color of a red–green checkerboard and report their decision by choosing one of two targets. By randomizing the target configuration on each trial, the mapping between color (red vs. green) and action (left vs. right) became flexible, requiring a nonlinear exclusive-or (XOR) computation. We found that neural dynamics in dorsolateral prefrontal cortex (DLPFC) led to high-dimensional population representations of the task. Neurons in DLPFC were modulated by target configuration, color, and action choice, reflecting the XOR computation. In contrast, neural dynamics in dorsal premotor cortex (PMd) were lower-dimensional and primarily reflected action choice. Next, these higher-dimensional task representations were strongest in ventral DLPFC, whereas color signals were more dominant in anterior DLPFC, revealing a functional gradient across the frontal lobe. Finally, when monkeys performed two tasks with identical stimulus–motor mappings, DLPFC representations were initially orthogonal across tasks but converged during the motor stage. At the single-unit level, this geometric structure arose from neurons exhibiting task-specific modulations earlier in the trial, with the same neurons eventually encoding action choice in both tasks. Together, these findings reveal a prefrontal implementation of XOR computation, confirm functional specialization across the frontal lobe, and unveil the neural mechanisms underlying multi-task representations.