Probing circuit mechanisms of feature selectivity in mouse visual cortex through synaptic-resolution connectomics

Feature selectivity, the ability of neurons to respond preferentially to specific stimuli, is a defining property of cortical computation. Competing theories attribute selectivity to structured, tuning-dependent "like-to-like" feedforward or recurrent connections, whereas others propose that it can emerge without specific structure in randomly connected, inhibition-dominated networks. The relative contribution of these mechanisms remains unclear. Here, we investigate the circuit basis of feature selectivity in mouse visual cortex, focusing on how orientation selectivity arises in layer 2/3 neurons driven by input from layer 4. We developed a data-driven modeling framework integrating network modeling with functional imaging and synaptic-resolution connectomics from the MICrONS dataset. Our analyses show that randomness in connectivity is the dominant source of selectivity, while structured "like-to-like" feedforward and recurrent connections play comparable secondary roles in amplifying it. These findings refine classical theories of cortical selectivity and demonstrate how connectome-constrained modeling can reveal the circuit principles underlying cortical computation.