Dendritic spines implement specific connectivity and support neuronal ensembles in cortical circuits

Dendritic spines are the principal postsynaptic targets of excitatory synapses, yet the network logic governing their organization and contribution to cortical computation remains unclear. Here, we combine the MICrONS ultrastructural reconstruction of mouse visual cortex with matched in vivo two-photon calcium imaging to explore the network roles of dendritic spines. Structurally, we show that spine density predicts the number and diversity of both excitatory and inhibitory presynaptic partners, supporting the long-standing "connectivity and diversity" hypothesis. Beyond expanding input space, spines exhibit distinct network-level principles: excitatory axons preferentially target spines with increasing axonal distance from the soma; highly broadcasting ("hub") neurons preferentially innervate spines; and neurons sharing presynaptic partners preferentially route their own outputs onto spines, revealing an input-to-spine/output-to-spine wiring correlations. Functionally, we find that these same structural motifs are found in neuronal ensembles, defined as groups of neurons with correlated calcium activity. Ensemble members are interconnected at more than twice the rate of spatially matched controls; these synapses connecting them are routed almost exclusively onto dendritic spines; and their shared presynaptic partners are almost exclusively inhibitory neurons that themselves also preferentially target spines. We suggest that local recurrent spine-targeted excitation effectively binds ensemble members into a coactive subnetwork, while shared spine-directed inhibition gates, synchronizes, and stabilizes their collective activity, providing a circuit-level architecture that may support attractor dynamics, pattern completion, and other cortical computations.