In one mouse retina, bipolar release kinetics by layer depth ranked the sustainedness of 20 On ganglion cell types (ρ 0.77) but not of 13 Off types (ρ 0.05, below depth-scrambled maps)

Question: do ganglion cells (GCs) inherit sustained or transient responses from the bipolar cells (BCs) at the depth of their dendrites, as Bae et al. (2018) proposed? We joined two data sets only by depth in the inner plexiform layer: EM stratification and calcium responses to a moving bar of e2198 GCs (one mouse retina), and glutamate release of 8,452 BC terminal regions at known depths (Franke et al. 2017, other mice). A forward model without BC type labels added the BC responses at each depth of the GC profile, spread them in time by the arbor geometry, and applied a calcium filter. Two global parameters were selected by trace fit, not by sustainedness. We used the sustainedness index, type groups and quality cutoff of Bae et al. (177 cells). For 20 On types, predicted and measured type sustainedness correlated (Spearman 0.77, 95% type-bootstrap CI 0.49 to 0.90). One response kernel per polarity gave -0.20, and all 500 within-polarity scrambles of BC field depths were lower (p = 0.002). For 13 Off types the correlation was 0.05 (CI -0.49 to 0.66), below 99.4% of the scrambles. One kernel per polarity gave 0.45. The model predicted that outer marginal Off types are more sustained than outer central Off types by +0.32 (CI 0.28 to 0.37); the measured difference was +0.01 (CI -0.08 to 0.09). Split-half reliability of the type rankings was 0.78 to 0.94, and predicted polarity matched measured polarity across 41 types (0.89). The pattern stayed with EM BC-type overlaps, full-field BC responses, calcium decay times of 0.05 to 0.8 s, and depth shifts of up to 6%. Subtraction of an On-pathway crossover signal in the Off layer improved Off trace fits (r 0.896 to 0.915) but kept the Off ranking at or below 0.17. Thus BC kinetics by depth explain On but not Off sustainedness in this retina. The result reproduces the On/Off split of sci:sha256:2d0e54d60bc3fe8c4b7a7b4e435a9826ae56921d784c2714a9e789687c4cee72 with other code and selection. Limitations: one retina, BC data from other mice and another stimulus, depth overlap instead of synapses, few cells for many types, and model variants explored on the same data. The package has the code, derived tables and figures; raw data come from their sources.

retina · retinal ganglion cell · bipolar cell · inner plexiform layer · sustained transient · structure-function · connectomics · e2198 · Eyewire · iGluSnFR · calcium imaging · Off pathway · crossover inhibition · forward model · mouse