BEHAVIORAL STATE SELECTIVELY ENGAGES SUPRAGRANULAR INHIBITORY NEURONS TO CONTROL CORTICAL DYNAMICS

Behavioral state strongly shapes cortical activity, yet how specific cortical interneuron types link behavior to circuit dynamics remains poorly understood. Here, we first quantify how locomotion and pupil-linked arousal correlate with neuronal activity across inhibitory neuron classes in mouse visual cortex using in-vivo two photon calcium imaging data. We find that behavioral variables are differentially associated with interneuron activity, with neurogliaform cells (NGFCs; LAMP5-expressing interneurons) and VIP neurons in superficial layers showing strong and consistent coupling to locomotion and pupil dynamics. To model these relationships, we developed a nonlinear decoding framework that integrates a multivariate autoregressive (MVAR) model with a multilayer perceptron (MLP), enabling prediction of neuronal activity from behavioral inputs alone. Notably, NGFC and VIP neuronal activity was reliably predicted from behavior, whereas other inhibitory classes showed weaker or less predictive associations. Having identified strong behavioral correlations in NGFCs, we next asked how these cells functionally shape cortical dynamics by implementinga biologically grounded leaky integrate-and-fire (LIF) network model that enables causal manipulation of NGFC output. Using this framework, we found that NGFCs dynamically stabilize excitatory neuron activity during locomotion in a contrast- and layer-dependent manner. Together, our results link behavioral state to interneuron-specific activity patterns and identify NGFCs as key modulators of state-dependent cortical dynamics.