Figure 1.
Comparison of cell-typing using responses to the full field flash, the spot, and the white noise stimuli.
Schematics of the full field flash (A), spot (B), and white noise (C) stimuli used to characterize RGC RF centers. (D) Averaged single frames taken at the maximum (ON-dominating) or minimum (OFF-dominating) of the STA time course (E) illustrating their RF center locations and sizes. A 1-standard deviation (s.d.) Gaussian contour estimate of the RF center is overlaid in red. (E) STA time course for an ON-dominating cell (black) and an OFF-dominating cell (gray). (F) Scatter plot of the Spot Response Bias versus the Full Field Flash Response Dominance Index (RDI). The measures were correlated (R = 0.73), but there were many cells with discrepant classifications by the two measures, particularly for ON-OFF cells. (G) Boxplot of the distribution of the Spot Response Bias for ON and OFF STAs shows that many ON-OFF cells revealed through the spot stimulus were classified as ON or OFF cells by the STA. Horizontal box lines represent the lower quartile, median, and upper quartile values of the distribution. Notch represents the 95% confidence interval around the median. Whiskers (dashed lines) show the extent of the remaining data and the plus sign represents an extreme outlier.
Figure 2.
The linear filter discovered by non-centered spike triggered covariance (STC-NC) analysis aligns with the STA for ON or OFF cells and the high variance STC for ON-OFF cells.
(A–B) Scatter plots of the STE projected onto the STA and the high variance STC (STC1) for an ON cell (A) and an ON-OFF cell (B) shows that the STC-NC aligned extremely well with the STA for the ON cell, but not for the ON-OFF cell, in which the STC-NC clocks toward the STC1. The STA was projected out of the STE prior to computing the STC only in A–B to ensure that the STA was orthogonal to the discovered STC vectors. (C) Scatter plot of the STE against the high and low variance STC vectors for the same cell in (A). The STC-NC aligned better with the STA in panel A than the low variance STC (STC500). (D) Scatter plot of the STE against the two STC vectors of greatest variance for the same cell in (B). The STC-NC aligned extremely well with the STC of highest variance. The length of the STC-NC vector (red) corresponded to the degree to which it projects onto the 2D plane. (E–F) Spatial frame of maximal contrast, 1D projection of the STE, and recovered static nonlinearity using the STA (E) and STC-NC (F, STC-NC bias = −0.11) for the ON-OFF cell from (B) and (D). The STA misidentified this cell as an ON cell and provided a poorly defined RF. (G–H) Some ON-OFF cells possessed spatiotemporally overlapping and inverted ON and OFF RFs (G), and some possessed slower OFF RF filters (H). The STC-NC accurately captured the character of both the ON and OFF filters for the cell in panel G, but not in panel H. The insets show the spatial RF centers (Scale bar: 100 µm).
Figure 3.
STC-NC Bias correlates well with the Spot Response Bias.
(A) ON-center RGC response to spot stimulation. Time-course of stimulus above the histogram. The dark horizontal line indicates a dark spot stimulus. The cell spikes at the offset of the dark spot. (B) ON-center RGC 1D STE projection onto the STC-NC direction and resulting static nonlinearity. (C) ON-OFF RGC response to spot stimulation. The cell spikes following both the onset and offset of the dark spot. This cell also gave a delayed ON response to the offset of the dark spot. (D) ON-OFF RGC 1D STE projection onto the STC-NC direction and the resulting static nonlinearity. (E) Scatter plot of the Spot Response Bias against the STC-NC Bias showed that they were well correlated (R = 0.84).
Figure 4.
Spiking in response to both positive and negative contrasts (ON-OFF behavior) produces bimodal distributions and symmetric nonlinearities, while ON or OFF cells produce unimodal distributions and asymmetric nonlinearities.
(A) Scatter plot of the STC-NC bias against the p value for the null hypothesis of a unimodal distribution. Bimodality was measured using the Hartigan and Hartigan Dip Test. Red lines at −0.6 and 0.6 show the cutoffs for ON and OFF classification. (B–C) The majority of ON and OFF cells were unimodal (e.g., B#1 and B#3) with highly asymmetric static nonlinearities, however, a few demonstrated highly unbalanced bimodality (C#4). Most ON-OFF cells were bimodal with a symmetric static nonlinearity (B#2), though some possessed less symmetric nonlinearities and were not bimodal (C#5).
Figure 5.
NT-3 OE mice exhibit stronger STC-NC signal strength at P15 and at P18.
(A) Average cell count for NT-3 OE and WT controls at the three ages analyzed by the STC-NC or the STA method. (B) Plot of the STC-NC or the STA signal strength in s.d. against mouse age for WT and NT-3 OE mice. (C–D) Immunostainings with Brn-3a and SMI-32 antibodies demonstrated that WT and NT-3 OE retinas did not differ in total RGC number or cell body size at P16. Error bars represent S.E.M. throughout. ***: p<0.001; **: p<0.01; *: p<0.05 in Student's t-test.
Figure 6.
The number of ON-OFF cells decreases with age.
(A) Cumulative distribution of the absolute value of the STC-NC bias for WT P18 and P25. A lower percentage of cells with ON-OFF character are observed in P25 retina (p = 0.003 in K-S Test). (B) The percentages of ON, OFF, and ON-OFF RGCs classified based on their STC-NC Bias in WT retinas at P18 and P25. The distributions differed significantly (two sample χ2 p = 0.015), with P25 retina possessing fewer ON-OFF cells and a corresponding increase in ON cells. (C) No difference was observed in the percentage of cell types between WT and NT-3 OE retinas at P18 (two sample χ2 p = 0.366).
Figure 7.
STC-NC analysis measures a reduction in the RF center size for both ON and ON-OFF cells during WT development.
(A) Both ON and ON-OFF RGCs show a decrease in RF center size at P25 compared to P18, while the size of OFF RGC RF centers is unchanged. (B) Cumulative distributions of RF center sizes for ON, OFF, and ON-OFF cells in WT. *: p<0.05; ***: p<0.001 in Wilcoxon rank sum test (same in Fig. 8). (C) The correlation between the STC-NC signal strength and RF size is negative for WT P18 (black) and P25 (gray) ON cells. The ANCOVA technique removes the effect of the confounding STC-NC signal strength prior to calculating the significance of differences in RF size due to grouping. The parallel lines represent best fits. Following standard ANCOVA analysis, the parallel condition was enforced after demonstrating that the slopes of the lines of best fit through the two data sets were not significantly different and that P25 had smaller RF sizes than P18 for WT ON cells ( p = 4×10−5 in ANCOVA test).
Figure 8.
STC-NC analysis reveals smaller RF center size of ON-OFF cells in NT-3 OE mice.
(A) ON and ON-OFF RGCs in NT-3 OE retinas had smaller RF centers at P18 than WT mice. (D) At P25, only ON-OFF RGCs in NT-3 OE had smaller RF centers compared to WT. (B–C, E–F) Cumulative distributions of RF sizes for ON (B, E) and ON-OFF cells (C, F) in WT and NT-3 OE retinas at P18 (B–C) and P25 (E–F).