State-dependent network interactions differentially gate sensory input at the motor and command neuron level in Caenorhabditis elegans

Neural responses are influenced by both external stimuli and internal network states. While network states have been linked to behavioral and stimulus states, little is known about how sensory inputs are filtered by whole-brain activity to affect motor and command neurons. Here, we recorded whole-brain activity of Caenorhabditis elegans experiencing bacterial food stimuli, and modeled how sensory inputs affect motor and command neurons in a network state-dependent manner. First, we classified active neurons into six functional clusters: two sensory neuron clusters (ON, OFF), and four motor/command neuron clusters (AVA, RME, SMDD, SMDV). Using encoding models, we found that ON and OFF sensory neurons that respond to onset and removal of bacteria, respectively, employ different adaptation strategies. Next, we used decoding models to show that bacterial onset and removal differentially drive AVA and RME cluster activity. To explore state-dependent effects on AVA and RME clusters, we developed a model that identified network states and fitted submodels for each state to predict how each of the six functional clusters drive AVA and RME cluster activity. We also identified network states in which AVA and RME clusters were either largely unperturbed by or receptive to bacterial sensory input. Furthermore, this model allowed us to disentangle the state-dependent contributions of stimulus timescales and bacterial content to neural activity. Collectively, we present an interpretable approach for modeling network dynamics that goes beyond implication of neurons in particular states, and moves toward explicitly dissecting how neural populations work together to produce state dependence.