Making decisions based on knowledge about causal environmental structures is a hallmark of higher cognition in mammalian brains. Despite mounting work in psychological and cognitive sciences, how the brain implements knowledge-based decision-making at neuronal circuit level remains a terra incognita. Here we established an inference-based auditory categorization task, where mice performed within-session re-categorization of stimuli by inferring the changing task rules. Using a belief-state reinforcement learning (BS-RL) model, we quantified the hidden variable associated with task knowledge. Using simultaneous two-photon population imaging and projection-specific optogenetics, we found that a subpopulation of auditory cortex (ACx) neurons encoded the hidden task-rule variable, which depended on the feedback input from orbitofrontal cortex (OFC). Chemogenetic silencing of the OFC-ACx projection specifically disrupted re-categorization performance. Finally, imaging from OFC axons within ACx revealed task state-related value signals in line with the modeled updating mechanism. Our results provide a cortical circuit mechanism underlying inference-based decision-making.
bioRxiv Subject Collection: Neuroscience