In one mouse retina, starburst contact direction predicted preferred motion direction within a median 12° for 26 On-Off and 23° for 7 transient On DS ganglion cells; arbor direction did not
We asked which wiring measure predicts the preferred motion direction (PD) of each retinal ganglion cell (GC). We used the public e2198 data set: two-photon calcium responses to bars that moved in 8 directions and serial electron microscopy of the same mouse retina patch, with 47 GC types and human-verified matches for 326 GCs (Briggman et al. 2011; Bae et al. 2018). We compared two predictors with the calcium PD of 315 typed cells: the direction of starburst amacrine cell (SAC) dendrites that touch the GC, plus 180° (anti-parallel rule), and the soma-to-arbor direction, also from a disc that cannot reach the volume border. A registration of calcium regions to EM somata (median residual 0.91 µm) gave the image-to-EM map. We chose the remaining 90° step with the On-Off DS cells only, then kept the map fixed. For 26 tuned On-Off DS cells, the SAC rule gave a median absolute PD error of 12.3° (95% CI 8–15°; PD noise 2.5°). For 7 held-out transient On DS cells (7o), the error was 22.7° (12–38°), with a common +23° offset. The arbor direction did not predict PD in these types (median errors 60–144°). Sustained On DS cells had large SAC asymmetry but responded weakly to this fast bar. For the asymmetric Off types 2an and 2aw, both predictors pointed within 7–40° of the type PD, but neither predicted which cell had which PD (shuffle p ≥ 0.38). For 58 tuned cells of 25 other types, no predictor did better than random. Weak tuning of non-DS types shared one direction (234 cells, mean vector 0.115 toward 91°), and 2an and 2aw pointed 6–9° from it. The baseline before each bar correlated with the previous response (median r 0.31), and directions were shown in a fixed order. Thus this data set cannot show that arbor asymmetry causes 2an or 2aw tuning. Within types, SAC direction deviations did not consistently predict PD deviations; the type-mean SAC direction predicted as well as each cell's value. Across 37 types, SAC contact fraction followed DSI (Spearman 0.47). Limitations: one retina, one bar speed, contacts rather than synapses, and a shared tuning bias of unknown cause.
retina · direction selectivity · starburst amacrine cell · retinal ganglion cell · connectomics · electron microscopy · calcium imaging · structure-function · Eyewire · e2198 · dendritic asymmetry · J-RGC · mouse
References
6- usesdoi:10.1016/j.cell.2018.04.040
Source of the GC reconstructions, types, SAC contact counts, calcium-EM matches and code (e2198 Eyewire museum).
- supportsdoi:10.1038/nature09818
Original e2198 data set and the anti-parallel SAC wiring rule; we extend it to all contacts, a held-out On DS type and per-cell errors with a fixed angle map.
- doi:10.1038/nature06739
J-RGCs: asymmetric dendrites and ventral motion preference; motivates the arbor predictor.
- doi:10.1016/j.celrep.2016.04.069
F-mini and F-midi asymmetric GC types (2an, 2aw, 63 in e2198).
- doi:10.1038/nature16468
Functional GC types in mouse retina, including the transient On DS type that corresponds to 7o.
- doi:10.1038/nature22818
DS cell preferred directions vary with retinal position; a possible source of within-type PD variation.
Cited by
11 publicationsPublic catalog entries on Sciscape.
- usesA 128 ms window shift changed a depth model's error decrease from -6.4% to 20.1% in 178 cells from a second mouse retina
This study documented e2198 source formats, cell matches, and depth source locations.
- With published labels and axon depth control, adding ribbon volume raised local plateau test error by 4.3% in eight mouse ON bipolar types
Related structure–function analysis in a different retinal volume. No data or code are reused.
- usesDendrite elongation gave no clear gain in a fixed linear test of motion axis responses in 315 cells from one mouse retina
We used its verified skeleton scale, source format checks, and stimulus coordinate convention. We wrote independent code. Its primary target was motion direction.
- 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
Same e2198 cells; it found type-level but little cell-specific structural prediction of motion direction, which is compatible with our polarity and kinetics result.
- usesIn one mouse retina, image position reduced motion vector test error in a short trailing edge window but not a broad window
We use its fixed angular convention and file-format findings. Its shared direction bias motivates the position test.
- usesA linear inhibitory rule for starburst contacts gave no clear added axis tuning prediction in 269 ganglion cells from one mouse retina
We use its file format information and compare our first-moment controls with its direction results.
- In one mouse retina, dendrite depth predicted the polarity and part of the time course of 47 held-out ganglion cell types; bipolar-type overlap and bipolar kinetics added little
Same e2198 data; reports that directions came in a fixed order and that baselines carry the previous response, which motivated our response-frames-only check.
- usesIn 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)
We reuse its e2198 loaders, calcium preprocessing and ROI-to-EM registration.
- With independent repeat errors, starburst contact features captured 1.25% of residual motion-vector energy in 302 cells from one mouse retina
This prior study tested contact direction and motion preference in the same retina.
- usesA fixed rule for separate ON and OFF contact directions gave no clear gain over one direction per cell in 27 cells from one mouse retina
We use its file-format rules, local response windows, and fixed stimulus-to-anatomy angle map. We test a new difference between the ON and OFF layers within each cell.
- usesWith surface opportunity fixed, removal of 1.16–17.35% of contact counts moved predicted direction by 45 degrees in 27 mouse retinal cells
We reuse its unchanged scalar cell table and fixed physiological angle frame. We independently read the external contact arrays for all new sensitivity calculations.