Poster B38, Sunday, March 26, 8:00 – 10:00 am, Pacific Concourse
Reward identity prediction error signaling in human orbitofrontal cortex
James D. Howard1, Thorsten Kahnt1; 1Northwestern University
Goal-directed behavior requires flexible neural representations of the complex associations between task structures, stimuli, and rewards. Recent studies have highlighted a critical role of the orbitofrontal cortex (OFC) in signaling these variables, including a conjunction between the value and identity of expected rewards. However, while the origin of predictive value signals is well-described by the firing of dopaminergic neurons in response to value-based prediction errors (PEs), whether and how violations in reward identity inform learning of expected outcome representations remains unknown. Here we implemented a transreinforcer reversal learning task in which hungry human participants (N=23) chose between two visual conditioned stimuli to receive either high-intensity (i.e. high-value) or low-intensity (i.e. low value) versions of sweet or savory food odors (2 value x 2 identity) while undergoing functional magnetic resonance imaging (fMRI). Critically, the stimulus-reward associations were intermittently changed throughout the task to independently induce either value or identity PEs. A regression model revealed that in addition to the value of rewards, changes in reward identity had a significant effect on choice behavior, prompting us to develop a modified Q-learning model that discounted value calculations by a weighted identity PE term. Group-level analysis of fMRI data regressed against model-derived identity PE estimates revealed robust activity in lateral posterior OFC, while orthogonal value PE evoked fMRI responses in the basal ganglia. These findings suggest that violations in reward identity may elicit an identity “teaching signal” in OFC that is used to construct representations of expected reward identity to guide goal-directed behavior.
Topic Area: EXECUTIVE PROCESSES: Goal maintenance & switching