Brain network reconfiguration during prediction error processing

Abstract Learning from experience is driven by reward prediction errors—signals that reflect updates to our expectations of reward. Despite numerous studies on neural correlates of reward prediction errors, the question of how large-scale brain networks reconfigure in response to reward prediction error signalling remains open. Here we ask how functional networks change in response to reward prediction errors depending on the context. In our study participants performed the probabilistic reversal learning task in functional magnetic resonance imaging (fMRI) scanner in two experimental contexts: a reward-seeking setting and a punishment-avoiding. We found that the participants’ learning speed depended on the sign of the prediction error but not on the experimental context. Whole-brain network analysis revealed a multi-scale community structure with a separate striatal reward network emerging at a finer topological scale and a ventromedial prefrontal network emerging at a coarser scale. We also found that the integration between large-scale networks increased when switching from positive to negative prediction error events. This pattern of large-scale network reconfiguration aligns with the broad range of research showing increased network integration with increased cognitive demands. Our findings offer a first sketch of how processing reward prediction error affects the functional connectivity of brain-wide networks.