Human thalamic low-frequency oscillations correlate with expected value and outcomes during reinforcement learning

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Reinforcement learning relies on fronto-striatal circuits.
• The thalamus is a critical, yet understudied, component of these circuits in humans.
• Direct electrophysiological evidence for thalamic involvement in human reinforcement learning is limited.

Purpose of the Study:

• To investigate the role of the human thalamus in reinforcement-based adaptive decision-making.
• To analyze intra-thalamic electrophysiological recordings during a reinforcement learning task.
• To identify neural signals within the thalamus related to expected value and prediction errors.

Main Methods:

• Electrophysiological recordings were obtained from the anterior thalamus (ATN) and dorsomedial thalamus (DMTN) in eight human participants.
• Participants performed a reinforcement learning task involving rewards and punishments.
• Computational modeling was used to estimate expected value and reward prediction errors.

Main Results:

• Low frequency oscillations (LFO, 4-12 Hz) in the ATN and DMTN positively correlated with estimated expected value during both reward and punishment learning.
• Thalamic LFO were negatively correlated with outcomes, indicating signaling of reward prediction errors.
• Distinct prediction signals were observed between rewarding and punishing conditions, suggesting a role in action inhibition.

Conclusions:

• The human thalamus, particularly the ATN and DMTN, is directly involved in reinforcement-based decision-making.
• Thalamic LFO encode expected value and prediction errors, crucial for adaptive learning.
• The findings elucidate the neural mechanisms of decision-making and action inhibition in the human thalamus.