Meeting Abstract
A cornerstone of ethology is that behavior comes in discrete packets, i.e., behavior can be temporally segmented into natural units. In some behaviors, these discrete packets are readily recognizable. But, in most behaviors, there is enough variability in these discrete packets to make them unrecognizable without the help of sophisticated analytical tools. Here we deploy a powerful analytical tool to extract these discrete packets in the context of a Drosophila’s locomotor response to odors. We reasoned that in the case of a fly’s locomotion, as in other complex behaviors, stimuli do not act directly on the observables (such as speed and turn rate) that define locomotion but on a higher-level strategy, which in turn affect the observables. This scheme can be mathematically described by a Hierarchical Hidden Markov Model (HHMM). We show that a two-layered HHMM in which there are 6-high level states or discrete packets or “locomotor primitives” is an intuitive model for a fly’s locomotion. We also demonstrate that odors modulate locomotion by altering the time a fly spends in the 6 high-level states. Finally, we extract the relationship between neural response and behavioral response elicited by food odors. We show that most food odors activate 7 olfactory receptor neuron (ORN) classes. Using genetic mutants and optogenetics, we were able to activate precise subsets of these 7 ORN classes. We found a modular organization in which each ORN class affects the time a fly spends in a subset of locomotor primitives. In sum, in this study, we introduce HHMM as a powerful method to extract the natural units of behavior and use this method to unravel the neural underpinnings of a behavior that is critical to a fly’s ecology.
