Beyond perception: the sensory cortex as an associative engine during goal-directed learning

The sensory cortex is widely considered to be specialized for perception by interpreting complex sensory patterns while also exhibiting structured forms of representational plasticity of behaviorally-relevant stimuli. Recent evidence, however, suggests that sensory cortical neurons directly encode non-sensory variables such as movement and valence. Conjoint representations of sensory and non-sensory variables in the same network could further hone perception or, alternatively, subserve more integrative cognitive processes. Here, we ask the extent to which the sensory cortex is an associative engine, directly linking stimuli with actions that produce desirable outcomes during goal-directed learning. We trained mice on an auditory go/no-go task and exploited a novel behavioral task to isolate rapid task acquisition (measured in non-reinforced trials) from slower, behavioral expression (measured in reinforced trials). Optogenetically suppressing the auditory cortex (AC) throughout learning only during the stimulus (n=4), or only during reward consumption (n=8), weakened stimulus-action associations. Strikingly, AC suppression throughout the trial (stimulus-action-outcome, n=8) led to even greater delays, suggesting that AC is used for associative purposes during learning. Interestingly, the impact of AC silencing gradually waned during learning arguing for a transient, associative, and teaching role for the AC, rather than one focused on task execution. Longitudinal, two-photon calcium imaging throughout learning combined with unsupervised low-rank tensor decomposition uncovered low-dimensional learning-related dynamics at distinct timescales. Learning networks rapidly developed non-overlapping ensembles that encoded the different task contingencies. In parallel, neural populations exhibited no large-scale improvement in stimulus decoding but, instead, exhibited fine-scale changes in stimulus selectivity that matched learning trajectories. Latent knowledge of the task was thus manifested in cortical networks to reflect contingency acquisition and stimulus selectivity despite poor behavioral expression. Overall, our work argues that in addition to being specialized for perception, the sensory cortex is associative in nature and continuous with regions involved in higher-order cognitive processes.