Neural Basis of Visuomotor Learning in the Frontal Eye Field: An Integrated Computational and Neurophysiological Approach

Abstract

The frontal eye field (FEF) is the primary cortical output for generating saccadic eye movements. FEF neurons that respond to visual stimulation display three fundamental types of movement sensitivity. First, they are often tuned to movement, as well, specifically for saccades toward their receptive field. Second, for many neurons, their receptive fields shift to their post-saccadic location around the time of a saccade. Finally, they can exhibit an enhanced response to visual stimuli which are the target of a saccade. Our objective was to determine how FEF visual responses and their associated movement-related sensitivities are shaped by learning. First, we present a model of oculomotor visuomotor responses that accounts, at the single neuron level, for all three motor-related phenomena through a single mechanism: learning to anticipate their input activity. Second, we tested the predictions of our model in vivo. FEF neural activity is typically examined after extensive training of the animal to establish stable, predictable behaviors. In natural behaviors, the appropriate motor responses depend on contextual characteristics such as reward, timing, and prior behaviors, all of which can change frequently and unexpectedly. We recorded from single neurons in the FEF of monkeys and investigated how FEF cognitive signals changed during a variety of behavioral contexts, and ultimately concluded that the signals are adjusted through a simple reinforcement learning rule. The results suggest that FEF and connected structures may integrate cortical inputs with motor commands and information about reward to compute a signal for the guidance of learning. Taken together, these studies lay the groundwork for future research to determine fully the learning rules implemented by FEF.