DUAL ENCODING OF SLOW AND FAST FLUCTUATIONS IN TASK-RELATED VIGOR IN RAT DORSO-MEDIAL STRIATUM

Interacting with the environment requires not only planning sequences of actions but also setting the duration of each step in the sequence, based on time, energy, and cognitive constraints. Here we studied in a perceptual decision-making task how rats set the time to initiate a choice response (“reaction time”, RT), perform the response movement (“movement time”, MT) and initiate a novel trial (“inter-trial interval”, ITI). All three behavioral epochs co-varied strongly within each session and across sessions, reflecting fluctuations in motivation and fatigue. These co-fluctuations were captured by a single latent dimension that we term “slowly varying vigor”, which was inferred using Gaussian Process Factor Analysis (GPFA). Strikingly, this motivational variable explained a much larger part of the variability of all three behavioral epochs (RT, MT, ITI) than cognitive variables such as sensory evidence, history-based expectations, or the previous outcome. Moreover, we found that the firing rate of a large proportion of individual neurons in the dorso-medial striatum (DMS) tracked the slow fluctuations of vigor across trials and sessions. Time-resolved regression analyses revealed a double dissociation in DMS activity: while the slowly varying component of vigor was encoded by tonic activity in the population, the variability of each behavior epoch above this slow component was encoded by phasic responses in DMS around the corresponding epoch. Our computational framework to dissociate slow and fast variations of vigor in decision-making tasks sheds a new light on the function of striatum in motivating behavior.