Dept of Bioengineering, Imperial College London, London, UK
The nervous system selectively integrates an enormous amount of sensory inputs to generate behaviour that aids the survival of an animal. One example is of a fly controlling its flight using the visual information of its environment. It does with a relatively simple neural system while mammalian system would need to sample a large population of neurons to perform the same task reliably. Previous studies have looked at how the direction of motion of a pattern is used to control yaw. We ask the question: how does the fly realize the direction of motion of a 2D pattern?
Flies are known to have elementary motion detectors (EMDs) that detect visual motion between pairs of eye facets in the compound eye. Due to the geometric arrangement of the eye facets the information regarding motion is constrained along certain directions. Little is known about how information along different directions is then integrated to control flight. We recorded the response of the H1 neuron, the activity of which is involved in yaw control, and determined its direction tuning curves to different plaids (overlapping sine wave gratings moving in different directions). Observing how neurons deal with the information from each of the plaid components can tell us about its directional integrating properties. The experimental results show that the tuning width across the different plaids is similar.
We built a model based on the known structure of the fly visual system. It consists of EMDs tuned to different directions (as constrained by the geometry of the compound eye) with a sigmoid input non-linear function, a weighting of the EMD responses, summation at H1 and an output non-linear function. We fitted the model to the recorded response to different plaids. We find that the sigmoid input non-linear function is a crucial component involved in predicting the response of the neuron. As an independent check, we compared the contrast tuning response predicted by the model to the actual response and found a similar trend, which other models fail to predict. Using the model fitting exercise, we were able to determine the tuning width of EMDs and the non-linear function operating on their outputs. We conclude that by having a sigmoid input non-linearity at the EMDs, the fly visual system can effectively compute the direction of motion of a 2D pattern.