Population-level geometric filtering generates motion-coherence selectivity in the retina.

Object-motion-sensitive (OMS) circuits must ignore self-motion while remaining sensitive to independently moving objects. We show that the mouse retina solves this by geometrically filtering motion coherence. Correlative light and electron microscopy reveal that TH2 amacrine cells (GABAergic Type-2 wide-field amacrine cells in the TH-Cre mouse strain) combine a dense proximal arbor with sparse radiating distal neurites, distributing synapses uniformly across both compartments. Two-photon imaging and compartmental modeling establish that sparse inhibition and voltage-gated sodium channels allow this bipartite arbor to multiplex signals: Proximal dendrites compute locally, whereas distal neurites integrate over hundreds of micrometers. A population model built from this architecture predicts a daisy-shaped inhibitory surround for OMS ganglion cells (W3/Ultra-High-Definition), with a circular near-surround and radially orientation-selective far-surround subunits. Patch-clamp recordings confirmed this organization and revealed that the surround suppresses coherent global motion while preserving object-motion detection, even amid high densities of independently moving objects. Thus, single-cell morphology and dendritic processing scale through plexus geometry to implement motion-coherence selectivity.