The mid-lateral cerebellum is necessary for reinforcement learning

Abstract

The cerebellum has long been considered crucial for supervised motor learning and its optimization1-3. However, new evidence has also implicated the cerebellum in reward based learning4-8, executive function9-12, and frontal-like clinical deficits13. We recently showed that the simple spikes of Purkinje cells (P-cells) in the mid-lateral cerebellar hemisphere (Crus I and II) encode a reinforcement error signal when monkeys learn to associate arbitrary symbols with hand movements4. However, it is unclear if the cerebellum is necessary for any process beyond motor learning. To investigate if the mid-lateral cerebellum is actually necessary for learning visuomotor associations, we reversibly inactivated the mid-lateral cerebellum of two primates with muscimol while they learned to associate arbitrary symbols with hand movements. Here we show that cerebellar inactivation impaired the monkeys ability to learn new associations, although it had no effect on the monkeys performance on a task with overtrained symbols. A computational model corroborates our results. Cerebellar inactivation increased the reaction time, but there were no deficits in any motor kinematics such as the hand movement, licking or eye movement. There was no loss of function when we inactivated a more anterior region of the cerebellum that is implicated in motor control. We suggest that the mid-lateral cerebellum, which provides a reinforcement learning error signal4, is necessary for visuomotor association learning. Our results have implications for the involvement of cerebellum in cognitive control, and add critical constraints to brain models of non-motor learning14,15.