Visual direction selectivity is most often characterized using coherent, unidirectional stimuli in laboratory settings. Natural vision, in contrast, exposes the visual system to complex and heterogeneous motion fields produced by self-motion and independently moving objects. The authors highlight this discrepancy and ask how early visual circuits extract a single, behaviorally relevant direction from such mixtures of motion vectors within a cell’s receptive field.
The study examined direction coding in mouse On-Off direction-selective ganglion cells (DSGCs). Rather than presenting unidirectional motion, the authors probed DSGC responses to stimuli composed of multiple motion components distributed across each receptive field. The central finding is that DSGCs report an integrated motion direction that reflects the combined pattern of component vectors falling within their receptive fields.
Using plaid stimuli with distinct velocity components, the authors show that DSGCs maintain a single preferred direction even when component vectors diverge by more than 90 degrees. In these compound stimuli, the DSGC tuning did not reflect competing peaks or multimodal direction responses; instead, cells exhibited a unidirectional tuning that tracked an aggregated direction computed from the component motions.
The authors attribute the observed angular averaging behavior to the directional tuning properties of the inhibitory inputs impinging on DSGCs. In other words, the inhibitory circuitry that shapes DSGC responses is tuned in a way that causes the cell’s output preference to reflect the angular mean of component directions rather than a simple arithmetic or vector-sum computation. The manuscript links the network-level directional properties of inhibition to the resultant population encoding of direction.
Although angular mean and vector sum are distinct mathematical operations that can yield different directions for the same set of component vectors, the authors analyzed the statistics of natural movie motion fields and report that these two measures tend to converge in real-world conditions. Natural scenes and motion patterns produce distributions of component vectors such that the angular mean and vector sum align, reducing the computational ambiguity for the retina: regardless of whether downstream circuits were to compute an angular average or a vector sum, they would receive largely consistent direction estimates under typical natural inputs.
The authors extend their analysis beyond synthetic plaid stimuli to naturalistic input by testing optic flow and object motion movies. DSGCs encoded the aligned direction derived from natural motion statistics for both classes of movies. Importantly, the precision of direction encoding depended on the homogeneity of motion vectors within the receptive field: more homogeneous local motion yielded more precise direction estimates, while heterogeneous motion reduced precision.
These findings support the view that the retina performs an early-stage integration of motion vectors using angular averaging. Because natural motion statistics make different possible summary computations (e.g., angular mean vs. vector sum) converge to a single aligned direction, angular averaging is an effective strategy to report the consensus motion direction in real environments. By providing a coherent direction signal at the level of retinal output, DSGCs may simplify downstream processing that underlies navigation, optic flow perception, and object motion discrimination.
This report is a preprint posted to bioRxiv and has not been certified by peer review. The authors declare no competing interests. Funder information listed in the source includes NIH Common Fund, NEI R01EY035268, and R90DA060338. For further methodological details, quantitative analyses, and supplemental data, consult the full preprint and supplementary materials available from the source.