Interactive auditory task reveals complex sensory-action integration in mouse primary auditory cortex

Predictive coding theory postulates that the brain achieves perception by actively making predictions about the incoming sensory information and correcting them if errors signals arise. These signals are likely the most relevant when the individual is actively interacting with the environment and where the sensory outcome determines the ongoing action. In addition, the cerebral cortex is thought to play a key role in generating these signals. Thus, to study the representation of error signals in the primary sensory cortex, we trained mice to perform an interactive auditory task that coupled their actions to the generated sound and perturbed this coupling to evoke putative error responses. We imaged Layer 2/3 (L2/3) and Layer 4 (L4) neurons in the mouse primary auditory cortex, and we identified not only neurons that mainly encoded action related information but also neurons encoding the mismatch between the action and the sound. These results show that a subset of A1 neurons encode the nonlinear interactions between the sound and the action. Furthermore, more L2/3 neurons encoded action related information than L4, indicating that action-sound integration emerges hierarchically in A1 circuits. Together, our results show that complex interactions between action and sound happen in A1 and that some A1 neurons responses reflect the violation of the learnt relationship between the action and sound feedback. Thus, primary sensory cortices not only encode sensory driven activity but also represent the complex interplay between sensory inputs, expectations, and errors.