Posture dynamics of crawling behavior in D. melanogaster larvae

We explore the posture dynamics of the locomotory behavior of Drosophila melanogaster larvae. Using tracking microscopy, we imaged individual larva at high spatiotemporal resolution while the animal crawled freely on an agarose substrate. In each image, we captured larval posture by tracking body segment boundaries as well as the tips of the head and the tail using the machine vision algorithm SLEAP. We characterized each posture as a set of 11 normalized segment lengths and 10 between-segment angles. Interestingly, principal component analysis of either length or angle data did not reveal a linear low-dimensional space. We visualized locomotory behavior by constructing short sequences of segment lengths and angles for which UMAP embedding revealed stereotyped cycles corresponding to forward locomotion and turning. We quantified these behavior motifs through the eigenvectors of a maximally-predictive Markov model inferred from the sequences. Our results are a first step towards building a neuromechanical model which can quantitatively capture larval posture dynamics and behavior.