Brain network reconfiguration during reward prediction error processing

Learning from experience is theorized to be driven by reward prediction error (RPE) signals that reflect updates to our expectations of reward. Despite numerous studies on the neural correlates of RPEs, the question of how large-scale networks (LSN) in the brain reconfigure in response to an RPE learning signal remains open. Here, we examine how functional networks change in response to RPEs depending on the context. In our study, participants performed a probabilistic reversal learning task while we acquired fMRI data in two experimental settings: reward-seeking and punishment-avoiding. Participants' behavior was best explained by models with different learning rates for positive and negative RPEs. Furthermore, no evidence was found for context-dependent learning rates. Using behaviorally fitted RPE models, we performed a whole-brain network analysis. This analysis revealed classical reward structures, where striatal reward networks emerge as modules when the community structure is examined at a finer resolution, and a ventromedial prefrontal network emerges at a coarser resolution. Using the same behavioral model, we found that, compared with negative RPEs, positive RPEs increased within-network integration and decreased between-community integration. This indicates that there are distinctly different neural processes for positive and negative RPEs.