The cerebellum contributes to prediction error coding in reinforcement learning - complementary evidence from stroke patients and from cerebellar transcranial magnetic stimulation

To survive and thrive in our ever-changing environment, we need to be able to predict the consequences of our actions. We update these predictions by learning through trial and error, and associated prediction errors (PEs). Recent rodent data suggest that the cerebellum – a region typically associated with processing sensory PEs in supervised error-based learning – also processes PEs in reinforcement learning (RL-PEs; i.e., learning from action outcomes). A proxy of action outcome processing in regions traditionally associated with RL-PE coding, such as striatum and anterior cingulate cortex, can be measured in a component of the feedback-locked event-related potential (ERP), i.e., the feedback-related negativity (FRN). We tested the hypothesis that cerebellar output is necessary for this RL-PE coding in the FRN in a probabilistic feedback learning task. In that case, altered cerebellar output should result in changes in the FRN. Two complementary experiments were performed. First, patients with chronic cerebellar stroke were tested. Second, single-pulse cerebellar transcranial magnetic stimulation (TMS) was applied in healthy participants, thus implementing a virtual lesion approach. Different from controls and control (vertex) TMS, no significant RL-PE processing was observed in the FRN in patients with cerebellar stroke, and in participants receiving cerebellar TMS. Only minor deficits in behavioural flexibility were found, with learning success preserved, possibly due to compensation by other brain areas within the reinforcement learning network. Findings in both experiments show that frontal RL-PE processing depends on cerebellar output. Our results provide evidence for involvement of the cerebellum in processing of RL-PEs in humans, complementing and extending previous findings in rodents.