Adaptive reward representations integrate expected uncertainty signals in orbitofrontal cortex.

Abstract

Incorporating expected uncertainty into reward evaluation is critical for adaptive learning and behavior. The orbitofrontal cortex (OFC) represents reward attributes, such as delay and magnitude, alongside expected uncertainties about upcoming rewards, serving its established role in both value- and risk-based behaviors. However, the cellular basis of how reward and uncertainty interact remains unclear. To investigate, we trained rats on a task where odor cues predicted sucrose rewards with varying uncertainties in delay or magnitude. Single-unit recordings revealed that many OFC neurons encoded expected uncertainties related to reward delay or magnitude. Population-level analysis showed distinct neural codes for delay and magnitude uncertainties, separate from those for reward delay and magnitude. Signals for each reward attribute and associated uncertainty converged onto the same neurons, reducing the sensitivity of delay and magnitude encoding as uncertainty increased. These findings suggest a cellular-level computation in OFC neurons, whereby uncertainty signals are integrated into attribute-specific reward representations, potentially supporting flexible learning and risk control.