Central Complex representations of self-movement are sufficient to compute wind direction in flight

Abstract

External forces such as gravity and wind fundamentally shape how an animal navigates through the world, but force magnitude and direction cannot be directly measured by an animal in motion. How the brain integrates multiple sensory cues to determine the direction of such external forces remains unclear. In the fruit fly (Drosophila melanogaster) columnar inputs to the navigation center, known as PFNs, have been shown to represent both visual and mechano-sensory cues about self-motion in a similar vector format, but the dynamics and integration of these cues, particularly during complex natural movements, are not known. Here we used 2-photon imaging to investigate the dynamic representation and integration of multi-sensory self-motion cues - optic flow and airflow - across PFNs. We find that one type, called PFNd, linearly integrates optic flow and airflow direction signals with distinct dynamics. We next reveal airflow speed as well as direction encoding by PFNp_c neurons. Imaging from available genetic lines suggests that many PFN types likely encode functions of airflow speed and direction. Based on these data, we construct and validate computational models of how visual and mechano-sensory dynamics are integrated and represented across PFN populations. We then use these models to estimate neural responses during real and simulated flight maneuvers. Applying a nonlinear empirical observability analysis to these responses, we show that active turning and deceleration maneuvers combined with PFN representations can decode the direction of an ambient external cue (wind) during free flight, and identify the theoretical contribution of different PFNs to this ability. Finally, we show that an anatomically-inspired artificial neural network can successfully decode wind direction from PFN representations and identify components of the algorithm used by both the network and real flies to orient upwind during olfactory navigation in flight. Our work provides evidence that active sensation, combined with multi-sensory encoding, can allow a compact system like the fly navigation center to infer a property of the external world that cannot be directly measured by a single sensory system.