State-dependent behavioral strategies in C. elegans olfactory navigation*

Animals can dynamically adjust their behavioral response depending on the odor environment, their internal states, and learned experiences. To understand these behavioral dynamics, we study olfactory learning and navigation strategies in the roundworm C. elegans. We train worms to associate butanone odor with food (appetitive training) or starvation (aversive training) and record navigation in a controlled odor environment. We developed a mixture of Generalized Linear Models (MoGLM) that constitutes two known navigation strategies in worms: biased random walk that modulates turning probability and weathervaning that steers the direction of motion in response to change in odor concentration. By fitting MoGLM to navigation trajectories from different training conditions, we find that the worms can differentially alter strategies depending on the appetitive or aversive training. Given finite data, MoGLM can decode the learned experience with >90% accuracy and outperforms the classic chemotaxis metric. The MoGLM also correctly predicts behavioral responses to optogenetic perturbation in an olfactory neuron and captures behavioral variability across the population. Lastly, we extend MoGLM with a hidden Markov model and show that it better explains experimental observations. The recovered latent states last for seconds and consist of different navigation strategies. We discuss progress towards identifying the neural mechanisms underlying these state-dependent behavioral strategies.