Neural dynamics for working memory and evidence integration during olfactory navigation in Drosophila

Working memory and evidence integration are fundamental components of cognition thought to arise from distributed circuits throughout the brain^1-2. Theoretical^3,4 and behavioral studies^5,6 argue that both processes are required for plume navigation, an innate task in which animals use stochastic sensory cues to navigate towards the unknown location of an odor source^7-10. Here we identify a small population of local neurons in the navigation center of Drosophila ^11-13 that exhibits both evidence integration and working memory dynamics during goal-directed olfactory navigation. Developing a closed-loop virtual plume navigation paradigm, we show that a bump of activity in this population ramps up with successive odor encounters, and can persist for variable intervals after odor loss. While bump activity persists, the fly maintains the goal heading it adopted during odor. Silencing these neurons impairs the persistence of upwind heading after odor loss. Simulations show that the time constant of persistence observed in these neurons optimizes navigation in a turbulent boundary layer plume. Our work localizes working memory and evidence integration to a specific group of genetically-identified neurons, which will facilitate the mechanistic dissection of these building blocks of cognition.