Fig. 6: Model RGC responses to probe flashes following luminance steps.

a, b Spiking response of model ON (columns 1–2) and OFF (columns 3–4) RGCs to luminance steps alone (a, blue) and to luminance steps followed by probe flashes (b, black) at 50, 250, and 2000 ms (different rows; analogous to real RGCs in Fig. 1b, c). Luminance steps are depicted by intensity bars in (a). First column in each cell type: responses following a positive-contrast luminance step; second column: responses following a negative-contrast luminance step. Vertical blue lines: timing of luminance step; orange bars: timing of probe flashes. Note the ON RGC and OFF RGC did not spike in response to negative-contrast (column 2) and positive-contrast (column 3) luminance steps respectively. c Flash-induced responses, after subtracting (a) from (b), overlaid to show the modulation of probe flash responses at different times (analogous to real RGCs in Fig. 1d). Lines connecting the response peaks highlight the time courses of suppression relative to baseline flash-induced responses (2000 ms). d Modulation indices for probe flashes in ON (light gray) and OFF model RGCs (dark gray), following positive-contrast (top panel) and negative-contrast (bottom panel) luminance steps. Modulation indices were calculated based on model responses to probe flashes presented at 10-ms intervals after luminance steps, and baseline as response to a flash at 2000 ms. Circle markers indicate modulation indices based on probe flashes at 50 and 250 ms shown in (b, c). Cyan and red arrows highlight the suppression of opposite-contrast flashes at 50 ms in ON and OFF RGCs, respectively. e Simplified schematic of the model (left), and stimulus and response traces at the different processing steps (right). Signal resulting from light stimulus (in this case positive-contrast luminance step followed by dark flashes) is passed through the cone model and the resulting output is filtered and thresholded. Extended schematic of the model showing different filter transiency and nonlinearity threshold is shown in Supplementary Fig. 13. f Modulation indices as a function of RGC transiency for real OFF RGCs (dark gray circles; N = 92; red line: linear regression fit) and ON RGCs (light-gray circles; N = 228; cyan line: linear regression fit). Individual panels correspond to different flash times after positive-contrast (top row) and negative-contrast (bottom row) luminance steps. These RGCs are a subset of the population data shown in Fig. 4b for which we could compute a transiency index (“Methods”). Suppression after negative-contrast steps was weaker in less transient ON RGCs (bottom row, blue regression line has negative slope) and, after positive-contrast steps, suppression was weaker in less transient OFF RGCs (top row, red regression line has negative slope). Numbers in each panel indicate the slope of the fits and asterisk symbol indicates statistically significant slope (slope ≠ 0, P<0.01, two-tailed t test). g Modulation index of model OFF (red) and ON (cyan) RGCs as a function of the pathway’s transiency, where transiency was varied by changing the transiency of the filter shown in (e). h Modulation index of model ON and OFF RGCs as in (g) but as a function of model’s nonlinearity threshold. Arrows in (g) highlight the same data as in (d). In (a–d), the filter transiency was set to 1 and the nonlinearity threshold to 0.1. In (g), nonlinearity threshold was set to 0.1. In (h), the filter transiency was set to 0. Supplementary Fig. 12 shows model RGC responses based on real cone data of Fig. 5 instead of model cone responses.