Sleep reveals dynamics integrating and segregating movement and stimulus representations in V1.

The primary visual cortex (V1) encodes multidimensional representations of ongoing movements, raising the question of how these signals are organized and interact with sensory representations within the same population. Here we addressed this question by using extracellular recordings in deep layers of mouse V1 to explore the relationship of movement and stimulus representations to intrinsic local circuit dynamics observed during non-REM (NREM) sleep. NREM revealed low-dimensional dynamics corresponding to an "on-manifold" subspace associated with the intrinsic manifold and an "off-manifold" subspace comprising dimensions suppressed by intrinsic dynamics, with remaining dimensions forming an "unstructured" subspace not systematically related to these dynamics. Movement and stimulus representations are both concentrated on-manifold, where they interact additively. However, stimulus representations are also concentrated off-manifold, where interference from movement-evoked activity is minimized. Off-manifold coding comprises population-sparse stimulus-evoked activity in chorister neurons, revealing an unexpected link between dimensionality, chorister/soloist cells, and sparse coding. Our findings suggest that intrinsic dynamics constrain neuronal activity in V1 and provide a structured substrate balancing the integration and segregation of movement and stimulus representations.