RULE REINFORCEMENT IS CAUSALLY ENCODED IN STRIATAL ENSEMBLES

Adapting goal-directed actions to changes in the environment is greatly facilitated by inferring the latent structure underlying such changes. However, the specific brain mechanisms that modulate decision making during changing task rules are incompletely understood. Here, we developed an auditory discrimination task for rats where the latent rule, i.e. the probability to repeat the previous stimulus category, varied in trial blocks of repetitions or alternations. We found that a reinforcement model where rats update the evidence favoring one rule or another with every new trial fitted choices the best. Permanent lesions or brief optogenetic inhibition of the dorsal striatum between trials decreased the modulation of choice by the rule evidence. Specifically, results suggested that striatal activity mediates the updating of rule evidence at each rewarded trial and translates it into a flexible response bias. While the two hemispheres cooperated to encode the rule evidence before action selection, they competed to guide animals during the response. Inhibition of the prefrontal or parietal cortex had little impact on choice modulation. Single unit recordings identified populations of striatal neurons encoding the response bias or previous choice sequences that, when combined, faithfully represented the dynamics of rule evidence across trials. Together, our results point to a prominent role for the dorsal striatum in inferring and implementing task rules.