Contrast normalization affects response time-course of visual interneurons.

In natural environments, light intensities and visual contrasts vary widely, yet neurons have a limited response range for encoding them. Neurons accomplish that by flexibly adjusting their dynamic range to the statistics of the environment via contrast normalization. The effect of contrast normalization is usually measured as a reduction of neural signal amplitudes, but whether it influences response dynamics is unknown. Here, we show that contrast normalization in visual interneurons of Drosophila melanogaster not only suppresses the amplitude but also alters the dynamics of responses when a dynamic surround is present. We present a simple model that qualitatively reproduces the simultaneous effect of the visual surround on the response amplitude and temporal dynamics by altering the cells' input resistance and, thus, their membrane time constant. In conclusion, single-cell filtering properties as derived from artificial stimulus protocols like white-noise stimulation cannot be transferred one-to-one to predict responses under natural conditions.

Dynamic Signal Compression for Robust Motion Vision in Flies.

Sensory systems need to reliably extract information from highly variable natural signals. Flies, for instance, use optic flow to guide their course and are remarkably adept at estimating image velocity regardless of image statistics. Current circuit models, however, cannot account for this robustness. Here, we demonstrate that the Drosophila visual system reduces input variability by rapidly adjusting its sensitivity to local contrast conditions. We exhaustively map functional properties of neurons in the motion detection circuit and find that local responses are compressed by surround contrast. The compressive signal is fast, integrates spatially, and derives from neural feedback. Training convolutional neural networks on estimating the velocity of natural stimuli shows that this dynamic signal compression can close the performance gap between model and organism. Overall, our work represents a comprehensive mechanistic account of how neural systems attain the robustness to carry out survival-critical tasks in challenging real-world environments.