Abstract

In order to safely move through the environment, visually-guided animals use several types of visual cues for orientation. Optic flow provides faithful information about ego-motion and can thus be used to maintain a straight course. Additionally, local motion cues or landmarks indicate potentially interesting targets or signal danger, triggering approach or avoidance, respectively. The visual system must reliably and quickly evaluate these cues and integrate this information in order to orchestrate behavior. The underlying neuronal computations for this remain largely inaccessible in higher organisms, such as in humans, but can be studied experimentally in more simple model species. The fly Drosophila, for example, heavily relies on such visual cues during its impressive flight maneuvers. Additionally, it is genetically and physiologically accessible. Hence, it can be regarded as an ideal model organism for exploring neuronal computations during visual processing. In my PhD studies, I have designed and built several autonomous virtual reality setups to precisely measure visual behavior of walking flies. The setups run in open-loop and in closed-loop configuration. In an open-loop experiment, the visual stimulus is clearly defined and does not depend on the behavioral response. Hence, it allows mapping of how specific features of simple visual stimuli are translated into behavioral output, which can guide the creation of computational models of visual processing. In closedloop experiments, the behavioral response is fed back onto the visual stimulus, which permits characterization of the behavior under more realistic conditions and, thus, allows for testing of the predictive power of the computational models. In addition, Drosophila’s genetic toolbox provides various strategies for targeting and silencing specific neuron types, which helps identify which cells are needed for a specific behavior. We have focused on visual interneuron types T4 and T5 and assessed their role in visual orientation behavior. These neurons build up a retinotopic array and cover the whole visual field of the fly. They constitute major output elements from the medulla and have long been speculated to be involved in motion processing. This cumulative thesis consists of three published studies: In the first study, we silenced both T4 and T5 neurons together and found that such flies were completely blind to any kind of motion. In particular, these flies could not perform an optomotor response anymore, which means that they lost their normally innate following responses to motion of large-field moving patterns. This was an important finding as it ruled out the contribution of another system for motion vision-based behaviors. However, these flies were still able to fixate a black bar. We could show that this behavior is mediated by a T4/T5-independent flicker detection circuitry which exists in parallel to the motion system. In the second study, T4 and T5 neurons were characterized via twophoton imaging, revealing that these cells are directionally selective and have very similar temporal and orientation tuning properties to directionselective neurons in the lobula plate. T4 and T5 cells responded in a contrast polarity-specific manner: T4 neurons responded selectively to ON edge motion while T5 neurons responded only to OFF edge motion. When we blocked T4 neurons, behavioral responses to moving ON edges were more impaired than those to moving OFF edges and the opposite was true for the T5 block. Hence, these findings confirmed that the contrast polarityspecific visual motion pathways, which start at the level of L1 (ON) and L2 (OFF), are maintained within the medulla and that motion information is computed twice independently within each of these pathways. Finally, in the third study, we used the virtual reality setups to probe the performance of an artificial microcircuit. The system was equipped with a camera and spherical fisheye lens. Images were processed by an array of Reichardt detectors whose outputs were integrated in a similar way to what is found in the lobula plate of flies. We provided the system with several rotating natural environments and found that the fly-inspired artificial system could accurately predict the axes of rotation.