Impaired reinforcement learning and coding of prediction errors in patients with cerebellar degeneration - a study...

This study investigated cerebellar involvement in reinforcement learning and prediction error (RL-PE) processing. Participants with pure cerebellar degeneration and demographically matched healthy controls performed a probabilistic feedback-based learning task while brain activity was recorded using electroencephalography (EEG). Structural magnetic resonance imaging was used to quantify cerebellar gray matter volume (GMV). Data from 21 cerebellar and 25 control participants were included in the analysis. We aimed to determine if feedback-based learning was impaired in patients relative to controls, and if single-trial RL-PEs were reflected in FRN, P3a, and P3b in the event-related potential (ERP) in patients and controls. Analysis of behavioral data revealed no differences in accuracy between patients and controls. Crucially, ERP analysis revealed that, while in controls, coding of RL-PEs was found in FRN and P3a for positive and in P3b for positive and negative feedback, these effects were absent in patients. Voxel-based morphometry revealed widely distributed cerebellar GMV reduction in patients, most pronounced in bilateral Crus I/ II and bilateral lobules I-IV. Multiple regressions in patients revealed a negative correlation between GMV in bilateral Crus I and II and FRN amplitudes. The present study extends previous evidence for cerebellar involvement in RL-PE processing in humans and advances our understanding of the cerebellum's role in performance monitoring and adaptive control of behavior.