Search

Natural decision-making often involves extended decision sequences in response to variable stimuli with complex structure. As an example, many animals follow odor plumes to locate food sources or mates, but turbulence breaks up the advected odor signal into intermittent filaments and puffs. This scenario provides an opportunity to ask how animals use sparse, instantaneous, and stochastic signal encounters to generate goal-oriented behavioral sequences. Here we examined the trajectories of flying fruit flies (Drosophila melanogaster) and mosquitoes (Aedes aegypti) navigating in controlled plumes of attractive odorants. While it is known that mean odor-triggered flight responses are dominated by upwind turns, individual responses are highly variable. We asked whether deviations from mean responses depended on specific features of odor encounters, and found that odor-triggered turns were slightly but significantly modulated by two features of odor encounters. First, encounters with higher concentrations triggered stronger upwind turns. Second, encounters occurring later in a sequence triggered weaker upwind turns. To contextualize the latter history dependence theoretically, we examined trajectories simulated from three normative tracking strategies. We found that neither a purely reactive strategy nor a strategy in which the tracker learned the plume centerline over time captured the observed history dependence. In contrast, “infotaxis”, in which flight decisions maximized expected information gain about source location, exhibited a history dependence aligned in sign with the data, though much larger in magnitude. These findings suggest that while true plume tracking is dominated by a reactive odor response it might also involve a history-dependent modulation of responses consistent with the accumulation of information about a source over multi-encounter timescales. This suggests that short-term memory processes modulating decision sequences may play a role in natural plume tracking., Author summary Many important behaviors require animals to make extended decision sequences in response to complex stimuli. In this study we investigated the sequences of navigational decisions made by fruit flies and mosquitoes while tracking odor plumes, which are generally subject to turbulence. By examining video-taped 3D trajectories of these insects flying in a wind tunnel containing an attractive odor plume, we asked what features of encounters with the plume influenced odor-triggered flight decisions. Most notably, we found that although the average response was dominated by a reflexive upwind turn, its strength was modulated by the history of prior plume encounters throughout the trajectory. While no theoretical strategy we simulated captured all aspects of the data, the algebraic sign of the history dependence was only recapitulated in a model where a simulated tracking agent maximizes information about the position of the plume source. This suggests that real odor tracking may involve short-term memory processes over multi-encounter timescales that are consistent with the accumulation of information about source location.

Methods available for quickly and objectively quantifying the behavioral phenotypes of the fruit fly, Drosophila melanogaster, lag behind in sophistication the tools developed for manipulating their genotypes. We have developed a simple, easy-to-replicate, general-purpose experimental chamber for studying the ground-based behaviors of fruit flies. The major innovative feature of our design is that it restricts flies to a shallow volume of space, forcing all behavioral interactions to take place within a monolayer of individuals. The design lessens the frequency that flies occlude or obscure each other, limits the variability in their appearance, and promotes a greater number of flies to move throughout the center of the chamber, thereby increasing the frequency of their interactions. The new chamber design improves the quality of data collected by digital video and was conceived and designed to complement automated machine vision methodologies for studying behavior. Novel and improved methodologies for better quantifying the complex behavioral phenotypes of Drosophila will facilitate studies related to human disease and fundamental questions of behavioral neuroscience.