DISSECTING THE CONTRIBUTION OF CORTICOCORTICAL PROJECTIONS TO SHARED CORTICAL VARIABILITY

Perception and cognition rely on coordinated interactions across cortical areas. These interactions produce low-dimensional shared variability across areas, called a communication subspace, thought to describe how neural activity propagates between areas. However, it remains unclear whether such shared variability reflects direct projections versus indirect pathways or global modulatory signals. We addressed this question using mesoscale two-photon calcium imaging of mouse primary visual cortex (V1) and higher visual areas posteromedial (PM) and anterolateral (AL), combined with viral retrograde labeling of feedforward (V1→PM) and feedback (PM→V1) projection neurons. This enabled a comparison of predictive subspaces derived from labeled projection neurons with those from neighboring unlabeled neurons. Projection neurons accounted for more activity in their specific target area than unlabeled neurons, in both feedforward and feedback directions. However, the majority of shared variability across all recorded areas reflected behavior-related signals. Removing this component reduced overall cross-area predictions yet revealed a clearer advantage for projection neurons. A subspace generalization analysis indicated that shared latents between V1 and PM are also present in unlabeled V1 and PM cells but preferentially expressed in labeled cells. Additionally, feedforward neurons were most strongly correlated with feedback neurons projecting back to their source, consistent with looped corticocortical connectivity. Together, these results show that communication subspaces in awake mice largely capture behavioral-state related activity. Regressing this out reveals the component carried by direct corticocortical projections that is target-area specific and organized into looped circuits. Low-dimensional variability shared across cortical areas thus reflects in part direct interareal communication.