Abstract
Flexible decision-making is crucial for adaptive behaviour. Such behaviour in mammals largely relies on the frontal cortex, and specifically, the orbitofrontal cortex (OFC). How OFC neurons encode decision variables and instruct sensory areas to guide adaptive behaviour is a key open question. Here we developed a reversal learning task for head-fixed mice together with two-photon calcium imaging to monitor the activity of lateral OFC neuronal populations and investigated their dynamic interaction with primary somatosensory cortex (S1). Mice trained on this task learned to discriminate go/no-go tactile stimuli and adapt their behaviour upon changes in stimulus–reward contingencies (‘rule-switch’). Longitudinal imaging at cellular resolution across weeks during all behavioural phases revealed a distinct engagement of S1 and lateral OFC neurons: S1 neural activity reflected task learning-related responses, while neurons in the lateral OFC saliently and transiently responded to the rule-switch. A subset of OFC neurons conveyed a value prediction error signal via feedback projections to S1, as direct anatomical long-range projections were revealed by retrograde tracing combined with whole-brain light-sheet microscopy. Top-down signals implemented an update of sensory representations and functionally reconfigured a small subpopulation of S1 neurons that were differentially modulated by reward-history. Functional remapping of these neurons crucially depended on top-down inputs, as chemogenetic silencing of lateral OFC neurons disrupted reversal learning and impaired plastic changes in these outcome-sensitive S1 neurons. Our results reveal the presence of long-range cortical interactions between cellular ensembles in higher and lower-order brain areas specifically recruited during context-dependent learning and task-switching. Such interactions crucially implement history-dependent reward-value computations and error heuristics, which, in turn, help guide adaptive behaviour.
