Visual system design from dynamic stimulus constraints

Marion Silies – Universität Mainz
Julijana Gjorgijeva – TU München

In natural environments, visual inputs are incredibly dynamic, influenced by external motion and self-motion of the animal. The ability to generate appropriate behaviors from dynamic visual environments is essential for animal survival. Thus, visual systems have evolved specific neural coding strategies. Here, we will combine theory and experiments to ask if and how the use of parallel temporal channels and specific connectivity motifs has tuned the visual system of Drosophila melanogaster to efficiently encode dynamic visual stimuli relevant for behavior. We extend this analysis to different fly species to determine the evolutionary design principles driven by dynamic stimulus constraints. We will analyze natural visual stimuli as seen by freely walking or flying flies and extract the statistics of dynamic stimuli features. We will characterize the statistical distribution of temporal features, inspired by power-law statistics described previously for static scenes. On the theory side, we will develop circuit models based on efficient coding, a powerful framework which assumes that sensory systems have evolved their properties to match the organisms’ environment. We will derive the functional and connectivity properties of core visual circuitry to maximize the encoding of dynamic stimuli. The project will examine the emergence of optimal visual system design at two levels. First, within a single hierarchy level, we investigate the hypothesis that neuronal populations have diversified into multiple parallel processing channels, each operating at its own timescale, and together covering the range of timescales in dynamic stimuli. Second, across levels of hierarchy we investigate the hypothesis that neural circuits represent dynamic visual stimuli by integrating temporal information from earlier stages. On the experimental side, we will compare theoretical predictions with measured physiological or behavioral responses to naturalistic dynamic stimuli. Specifically, we will investigate the role of multiple temporal channels in the lamina, the first layer of visual processing in the fly, for the efficient coding of the stimulus statistics experienced by a fly. Next, we will study how neuronal properties or connectivity of subsequent levels of the visual hierarchy contribute to the efficient coding of natural dynamic stimuli. We will complement measurement with perturbations of specific circuit elements to understand the functional role of specific circuit elements and computational strategies. Measuring behavior will allow us to evaluate how wildtype or perturbed visual systems accommodate the stimulus distribution present in natural environments. Last, we take a comparative approach to study visually-guided behaviors and optimal temporal filters of parallel visual pathways in different Drosophila species. Our work promises insight into common evolutionary design principles linking dynamic temporal stimulus features to circuit coding strategies.