Behavioral strategies and neural signatures underlying stay/switch decision-making in Drosophila

In natural environments, animals must decide when to commit to one option, such as searching for food, or switching to another option, such as escaping a predator. How the nervous system mediates this “stay/switch” decision-making process during naturalistic behavior remains largely unknown. Here, we study stay/switch decisions in fruit flies by presenting optogenetically-driven heat pulses during courtship of a female. We developed a closed-loop stimulation paradigm in which the male receives optogenetic stimulation conditioned on his real-time song production. In response to the stimulus, flies made one of two decisions: a “switch” decision characterized by a U-turn response away from the female, or a “stay” decision characterized by little to no change in the male’s rotational speed or male-female distance and a continuation of courtship. Using SLEAP-based automated pose tracking, we quantified multiple dimensions of courtship, and constructed a generalized linear model to predict the male’s upcoming decision. We found that among multiple social features, female forward velocity (fFV) has the highest predictive ability (mean cross-validated accuracy = 62%, n=51 flies). The associated fFV filter shape suggests that the male fly integrates fFV information over hundreds of milliseconds, and that high forward velocity leads to “switch” decisions. We next considered how the brain implements stay/switch decisions by recording brain-wide neural activity with 2-photon microscopy in a male fly while he received optogenetic heat pulses during courtship with a virtual female. Our imaging data revealed that multiple brain regions contain decision-predictive activity, which is consistent with a distributed computation. Furthermore, we found that the alignment between neural populations that encode the optogenetic stimulus and the fly’s rotational speed strongly captures the animal’s turning behavior on a trial-to-trial basis. Together, these behavioral and neural results demonstrate both the strategies and neural computations underlying decision-making during naturalistic behaviors.