Adaptive decision-making enables animals to select appropriate goal-directed strategies based on the evaluation of recent stimulus/action-outcomes. The orbitofrontal cortex (OFC) play an important role in invoking rule-based strategies to enable flexible learning. However, the neural circuit mechanisms in the lateral OFC (lOFC) and its interactions with sensory cortical areas underlying such processes remain elusive. To investigate this, we trained head-fixed mice on a tactile-discrimination-based reversal-learning task and enforced them to relearn the task after a ‘Go/No-go’ rule-switch. Mice exhibited high performance during learning and re-learned the task upon rule-switch. To investigate how distinct neuronal subpopulations in lOFC respond during the task, we employed two-photon imaging through a rod-like GRIN lens. Longitudinal imaging of trial-by-trial Ca2+ responses from the same subsets of lOFC neurons in mice expressing GCaMP6 in layer 2/3 neurons revealed that OFC neurons are a key substrate encoding value prediction error. We also measured neuronal responses in the primary somatosensory cortex (S1). In S1, distinct neuronal subpopulations showed ‘stimulus-selective’ as well as ‘outcome-selective’ responses. The ‘outcome-selective’ neurons were differentially modulated by reward history and altered response selectivity upon reversal. Silencing lOFC following rule-switch impairs plastic changes in S1 neurons. Taken together, our experiments shed light on the long-range cortical circuit interactions underlying behavioural flexibility, indicating a crucial role of mouse OFC neurons in encoding predictive ‘teaching signals’ that drive adaptive changes in behaviour.