Column-Like Subnetwork Reconstruction in Motor Cortex from Graph-Based 3D High-Density Two-Photon Calcium Imaging

How precise 3D interactions among cortical neurons underlie layer-specific computations remains elusive. We developed a graph-framework to infer functional connectivity from fast volumetric two-photon Ca ²⁺ imaging in the awake mouse primary motor cortex. By converting deconvolved traces into binary spike trains, removing population bursts, and applying an adaptive, layer-specific statistical threshold, we reconstructed a directed, weighted network of ∼1,000 neurons. Decomposition into strongly connected components revealed ∼10-cell subnetworks, predominantly in layer II/III but often bridging to layer Va. Across six 20-min recordings, we found that (1) layer II/III dominates connectivity, (2) feedback (Va→II/III) links are more numerous and stronger than the feed-forward (II/III→Va) ones, and (3) information flows in ≤ 6 synapses. We uncovered seven geometrical and dynamical motifs—ranging from compact columns to elongated diagonals—each with characteristic event sizes and durations. These findings reveal diverse, column-like microcircuits in M1 with a net ascending flow, suggesting that such subnetworks form elemental processing modules for motor control.