Stimulus-reward contingencies drive long-lasting alterations in neocortical somatostatin inhibition during learning

Learning involves the association of discrete events in the world to infer causality, likely through a cascade of changes at input- and target-specific synapses. Transient or sustained disinhibition may initiate cortical circuit plasticity important for association learning, but the cellular networks involved have not been well-defined. Here we show that sensory association learning drives a durable, target-specific reduction in inhibition from somatostatin (SST)-expressing GABAergic neurons onto pyramidal (Pyr) neurons in superficial but not deep layers of mouse somatosensory cortex. Critically, SST-output was not altered when stimulus and rewards were unpaired, indicating that these neurons are not modified by sensory input alone. Depression of SST output onto Pyr neurons could be phenocopied by chemogenetic suppression of SST activity outside of the training context. Thus, neocortical SST neuron output is persistently modified by convergent sensory and reinforcement signals to selectively disinhibit superficial layers of sensory neocortex during learning.