Predictive and instructive cerebellar encoding of dopamine reward drives motivated behavior

Predicting rewards is critical for learning. Although the cerebellum predicts rewards like water and food, it also famously coordinates physical actions like drinking and eating. To disentangle reward prediction from the movements to consume reward, we trained mice to push for delayed dopamine rewards delivered directly into the brain. Via two-photon imaging, we found that many cerebellar granule cells (GrCs) anticipated dopamine with sustained activity that terminated at reward delivery. GrC activity also "stretched" to match 1- or 2-s intervals before dopamine, thereby linking the action to the expected reward time. By contrast, most cerebellar climbing fibers (CFs) spiked just after dopamine delivery. GrC-CF activity also generalized between dopamine versus water rewards, both in individual mice and many individual neurons. Finally, naïve mice 'rewarded' only with CF stimulation also learned to execute a modest number of pushing movements, with similar predictive GrC activity. Thus, cerebellar circuits help animals learn to anticipate rewards even when no consummatory action is needed, suggesting deeper cerebellar integration in brain reward prediction networks.