EMERGENCE OF LOCOMOTION-INVARIANT REPRESENTATIONS OF DEPTH FROM MOTION PARALLAX

To estimate the distance of visual cues, the visual system must infer the missing depth information from two-dimensional retinal images. Motion parallax, the inverse relationship between optic flow speed and depth of visual cues during movement, is a fundamental but poorly understood cue for depth perception. We recently showed that excitatory neurons in layer 2/3 of the mouse primary visual cortex are selective for depth driven by motion parallax. These depth-selective responses result from integration of optic flow and locomotion-related signals. However, objects at the same depth generate different speeds of optic flow as the animals' movement speed changes and how depth representations remain robust to these changes is unclear. To address this question, we recorded neuronal activity across cortical layers using two-photon calcium imaging and Neuropixels probes while mice navigated a virtual reality environment where motion parallax was the only cue for depth. We found that many neurons changed their optic flow preference depending on the running speed of the animal, consistently responding to the same visuomotor gain corresponding to a specific depth. In layers 2/3 and 5, many neurons were well described by a model incorporating depth-selective responses gain-modulated by locomotion. However, the responses of neurons in layer 5 persisted over a broader range of locomotion speeds. We propose that these locomotion-invariant depth representations in layer 5 result from the summation of layer 2/3 inputs tuned to specific combinations of running and optic flow speeds.