C. elegans neuron classes with shorter connectome walks to body-wall muscle encode locomotion more strongly in two whole-brain data sets, beyond rewired-wiring nulls

Question: does the position of a neuron class in the connectome predict how strongly its activity encodes locomotion in freely moving C. elegans? Approach: for each class we calculated the expected random-walk steps to body-wall muscle in the Cook 2019 connectome (muscle distance). We fitted blocked, cross-validated ridge models of locomotion for 3,548 NeuroPAL-identified neurons in 40 off-food animals (Atanas 2023) and 814 neurons in 8 on-food animals (Dag 2023). Results: classes with a shorter muscle distance encoded locomotion more strongly (Spearman rho = -0.49, 58 classes, p = 0.0001; rho = -0.40, 54 classes, p = 0.003). Strength-preserving rewired connectomes gave mean rho near 0. Rewiring that kept every direct muscle synapse gave mean rho of -0.25 and -0.27, and the real value was stronger than 100% and 96.2% of 1,000 such nulls. After control for degree and direct muscle synapses, partial rho was -0.36 and -0.30. After control for degree, cell type, number of animals and label confidence, partial rho was -0.25 in each data set (p = 0.06 and 0.07). In interneurons only, rho was -0.40 and -0.41 (p = 0.03). Cross-animal transfer of encoding models also decreased with muscle distance, but within-animal encoding strength fully explained this (partial rho = +0.05 and +0.03). Limitations: across-class correlation, not a causal test; cell type explains part of the relation; nerve-ring-only Witvliet 2021 connectomes gave weaker relations of the same sign; on-food animals moved little; linear models with fixed time filters.

C. elegans · connectome · whole-brain imaging · behavior encoding · locomotion · random walk · hitting time · rewired null · NeuroPAL · structure-function · cross-animal reliability