The Cerebellum Contributes to Prediction Error Coding in Reinforcement Learning in Humans

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• The cerebellum is traditionally linked to sensory prediction error (PE) processing.
• Recent rodent studies suggest cerebellar involvement in reinforcement learning prediction errors (RL-PEs).
• The role of the cerebellum in human RL-PE processing remains unclear.

Purpose of the Study:

• To investigate if cerebellar output is essential for RL-PE processing in human forebrain regions.
• To determine the cerebellum's necessity for action-outcome learning and RL-PE coding.

Main Methods:

• Two complementary human experiments utilizing electroencephalography (EEG) and a probabilistic feedback learning task.
• Experiment 1: Compared RL-PE processing (via feedback-related negativity - FRN) in patients with cerebellar stroke versus healthy controls.
• Experiment 2: Employed single-pulse cerebellar transcranial magnetic stimulation (TMS) to create a temporary 'virtual lesion' in healthy participants.

Main Results:

• Behavioral learning of action-outcome associations remained largely intact, with minor effects on flexibility.
• No significant RL-PE processing (FRN) was detected in patients with cerebellar stroke.
• Cerebellar TMS also abolished significant RL-PE processing in healthy participants.
• Forebrain RL-PE coding was demonstrably blunted when cerebellar output was compromised.

Conclusions:

• Cerebellar output is necessary for RL-PE processing in human forebrain regions.
• These findings extend rodent data, supporting a direct role for the cerebellum in RL-PE processing.
• The cerebellum is essential for coding RL-PEs in the forebrain.