Efficient navigation is achieved through state-dependent strategies in C. elegans

Animals employ different behavioral strategies to achieve the goal of sensory navigation. However, the factors that drive switches between different strategies---and the neural mechanisms underlying them---remain largely unknown. The nematode worm C. elegans is known to exhibit distinct “steering” and “turning” strategies during navigation, where steering involves continuous, gradual changes in orientation, and turning involves infrequent, major reorientation events. It is commonly assumed that the worm’s responses to sensory input, whether steering or turning, are independent behavioral actions without persistence. Here we tested this assumption explicitly by developing a novel statistical model for worm navigation in salt and odor environments, and showed that this existing account is incomplete. Instead, we show that the worm’s navigation is well described by a Hidden Markov Model (HMM) with two distinct states, each persisting over time and producing different mixtures of strategies. One of the identified states exhibits greater prevalence of steering, while the other a greater prevalence of turning. This hierarchical description challenges the assumption that strategies are static over time and driven solely by immediate sensory input. To verify the proposed state-switching model, we optogenetically perturbed an olfactory neuron in worms and showed that sensory input causally drives transitions to a turn-enriched state. We also demonstrated that worms transition from a turn-enriched state to a steering enriched state when the local concentration increases. This sensory-driven state transition accounts for a previously unexplained phenomenon of “directed turns”, where the animals are able turn and orient toward the goal direction. Finally, we showed that a data-constrained reinforcement learning (RL) model suggests state-switching emerges through optimization and enables more efficient gradient climbing. By combining experimental measurements with computational models we revealed that worms alternate between persistent states to achieve efficient navigation.