Distinct Orbitofrontal Feedback Signals Shape Sensory Behavioral Strategies during Flexible Learning

Abstract

Animals adapt their behavior by integrating sensory evidence with prior experience and contextual information. Understanding how animals employ specific behavioral strategies to integrate these variables and how neural mechanisms support such strategies remains unclear. We trained mice on a tactile reversal learning task and used a trial-by-trial computational model to infer latent decision strategies. Early in learning, mice relied on action-based ("choice-driven") policies but progressively transitioned to stimulus-guided choices ("cue-driven") as they learned the task. Following a rule reversal, mice flexibly reinstated this policy to adapt their behavior. Chemogenetic silencing of the lateral orbitofrontal cortex (lOFC) delayed this transition and impaired reversal learning. To identify the neural basis of strategy learning, we developed a novel approach combining low-dimensional analysis of neural activity with decoding of behavioral strategies from longitudinal two-photon imaging. We revealed reward- and error-strategy representations in excitatory layer 2/3 neurons in the primary somatosensory cortex (S1) that were differentially modulated by OFC instructive signals. Within S1, distinct subpopulations of positive and negative-valence-coding neurons tracked evolving decision strategies through trial-history integration. Together, these findings revealed how mice flexibly deploy distinct exploratory strategies during adaptive behavior, highlighting OFCs role in supporting reward and error-guided learning through distinct corticocortical interactions.