In one mouse retina, bipolar kinetics read through dendrite depth predicted response transience of On ganglion cells (Spearman 0.58; polarity only −0.12), not of Off cells or within types

We asked which parts of the light response of a retinal ganglion cell (GC) follow from the depth of its dendrites in the inner plexiform layer. We used the e2198 mouse retina (electron microscopy and calcium imaging of the same tissue; 315 matched GCs of 47 types). For each GC we calculated the overlap of its dendrite depth profile with the EM terminal profiles of 12 cone bipolar cell (BC) types, and multiplied these weights by the glutamate responses of the same BC types from other mice (Franke et al. 2017). A rectifier and a calcium filter with two shared parameters gave the predicted response to a moving bar. We did not fit the response shape to GC data. sci:sha256:67b50cd7eac8ed362cb458cdee59d74b31defdd5bd3f9fbb91e3f01c79e3add7 tested the same type of model at the level of held-out types; our independent code reproduced its whole-trace result (mean cell r 0.778 vs 0.78; one mean On and one mean Off BC response did equally well). New results: (1) In 63 On cells of 18 types, the predicted On sustain index correlated with the measured index (Spearman 0.58, type-bootstrap 95% CI 0.07 to 0.73), but a polarity-only model gave -0.12 (difference CI 0.02 to 1.08). A glutamate depth map without BC labels gave 0.62, and all 864 possible cluster-to-type assignments gave 0.55 to 0.59. (2) No model predicted the Off sustain index of 93 Off cells (0.03). (3) For whole traces, the median r was 0.868; profiles of random cells gave 0.648 and profiles of same-class cells 0.841 (1,000 shuffles, p = 0.001); polarity only gave 0.853. (4) Inside types, structure added little: profiles of other same-type cells gave 0.863 (p = 0.11), and deviations from the type mean correlated weakly (mean r 0.047, p = 0.03). For scale, a depth kernel fitted on other types gave 0.910, the mean of other same-type cells 0.937, and split-half repeats 0.974. (5) The largest errors were in types whose polarity did not agree with their stratification, for example 5si (mostly On-layer dendrites, balanced On and Off responses) and F-mini On. Limitations: one retina; BC kinetics from other animals, another indicator and a chirp stimulus; depth overlap instead of synapse counts; no inhibition in the model; 0.128 s frames; responses averaged over directions. The package has code, derived tables, shuffle distributions, figures and source checksums; e2198 source files are not redistributed.

retina · retinal ganglion cell · bipolar cell · inner plexiform layer · stratification · structure and function · connectomics · electron microscopy · calcium imaging · iGluSnFR · e2198 · On and Off pathways · transient and sustained responses · zero-parameter model · reproduction