Reward Prediction Encoded by Single-Neuron Responses in the Human Nucleus AccumbensSameer A. Sheth, MD, PhD1, Shaun Patel, BS1, Matthew Mian, BSE1, Emad Eskandar, MD11Boston, MA United States Keywords: neuron, microelectrode recording, functional, obsessive-compulsive disorder, depression
A central feature of intelligent behavior is the ability to link values with actions and to evaluate expectations relative to outcomes. These functions form the basis of theories such as reinforcement learning (RL), and also apply to uniquely human behaviors, such as financial decision making.
Converging evidence suggests that the nucleus accumbens (NAc) is critical to this process, and ultimately is involved in governing the motivational aspect of goal-oriented behavior.
The patients were undergoing obessive-compulsive disease and major depression.
We performed single-neuronal recordings in the NAc using microelectrode recordings in patients undergoing deep brain stimulation for obsessive-compulsive disease and major depression. Subjects performed a gambling task that required financially contingent decisions.
We isolated 19 individual NAc neurons in eight subjects. In situations of unexpected wins, NAc activity was potentiated, and in situations of unexpected loss, activity was attenuated. These neurons therefore encode the difference between expectation and outcome, a reward prediction error (RPE) signal, consistent with RL theory. Furthermore, when the outcome was uncertain, the activity of individual NAc neurons predicted the subject’s bet 2 seconds before the subject expressed the bet.
This is a retrospective study.
We demonstrate for the first time direct evidence for a RPE signal in the human NAc. This signal encodes the difference between expected and actual outcomes, and is critical in biasing actions towards rewarding behaviors.
These results suggest a role for the NAc in encoding or making financial judgments in the absence of any predictive information, a process akin to the subject’s intuition. Project Roles:
S. Sheth (), S. Patel (), M. Mian (), E. Eskandar ()