Visuomotor integration gives rise to three-dimensional receptive fields in the primary visual cortex

Distinguishing near and far visual cues is an essential computation that animals must carry out to guide behavior using vision. When animals move, self-motion creates motion parallax — an important but poorly understood source of depth information — whereby the speed of optic flow generated by self-motion depends on the depth of visual cues. This enables animals to estimate depth by comparing visual motion and self-motion speeds. As neurons in the mouse primary visual cortex (V1) are broadly modulated by locomotion, we hypothesized that they may integrate visual- and locomotion-related signals to estimate depth from motion parallax. To test this hypothesis, we designed a virtual reality (VR) environment for mice, where visual cues were presented at different virtual distances from the mouse and motion parallax was the only cue for depth, and recorded neuronal activity in V1 using two-photon calcium imaging. We found that the majority of excitatory neurons in layer 2/3 of V1 were selective for virtual depth. Neurons with different depth preferences were spatially intermingled, with nearby cells often tuned for disparate depths. Moreover, depth tuning could not be fully accounted for by either running speed or optic flow speed tuning in isolation, but arose from the integration of both signals. Specifically, depth selectivity of V1 neurons was explained by the ratio of preferred running and optic flow speeds. Finally, many neurons responded selectively to visual stimuli presented at a specific retinotopic location and virtual depth, demonstrating that during active locomotion V1 neuronal responses can be characterized by three-dimensional receptive fields. These results challenge the traditional view of V1 as a feed-forward filter bank, and suggest that the widespread modulation of V1 neurons by locomotion and other movements plays an essential role in estimation of depth from motion parallax.