Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior

Abstract

Learning motivated behaviors requires both anticipating rewards and reinforcing actions that yield them. Although cerebellar activity encodes natural rewards like water and food, it also coordinates physical movements like eating and drinking. To disentangle reward from consummatory movements, we trained mice to push for delayed dopamine rewards delivered directly into the brain. Here we show that cerebellar input streams use both predictive and instructive codes for dopamine reward. Two-photon imaging revealed that many cerebellar granule cells (GrCs) predictively encoded dopamine rewards with sustained activity that "stretched" to match 1- or 2-s delay intervals before terminating upon reward receipt. By contrast, most cerebellar climbing fibers (CFs) spiked just after dopamine delivery. In mice also trained with water rewards, encoding strength for dopamine matched or exceeded that for water. Both cell types contributed causally: chronic GrC inhibition disrupted self- stimulation learning, and CF self-stimulation "rewards" drove moderate operant learning in naive animals. Thus, cerebellar encoding of dopamine reward helps drive motivated behavior, suggesting deeper cerebellar integration in brain reward prediction networks.