Supplementary Figure 4: Artifacts in surface intensity sampling due to cell geometry and depth-dependent fluorescence variation.
(A) A simulated cell image in which intensity varies only with the distance from the cell surface but is uniform along the surface shows correlation between curvature and intensity (B, blue). Inset, red: Illustration of sampling sphere. (C) The correlation is due to variation in effective sample depth with curvature. Inset: cartoon to illustrate that the effective mean sample depth varies with curvature. Cell is shown in red, sampling sphere in blue. Different cell curvatures give different sample geometries (green, compare top and bottom) n = 10 simulations. (D) Similar depth variation is seen in experimental data. (E) Depth normalization (see Supplementary Information ‘Curvature and Image Intensity Correlation’) removes systematic intensity variation with depth in experimental images, but when applied to simulated image some artifact remains (B, red) due to variation in sample size with curvature. n = 10 simulations. (F). (G) An image with only noise shows the same correlation between intensity and curvature (H), and a correlation between intensity and sample size (I). A random signal with spatial autocorrelation of σAC = 1 (J) shows diminished artifactual correlation (K), which decreases further with σAC = 2 (L). Green vertical lines in (H,K,L) indicate the percentile levels of image intensities below/above which the surface curvature and intensity display a significant correlation, for example, the curvature values for a small range of particular intensity values deviate significantly from the mean surface curvature value over the entire cell. (M) Applying depth normalization to the experimental data only alters the observed correlations at very low intensities, that is, below the 10th percentile. In B,H,I,K,L and M dashed lines indicate boot-strapped 95% confidence intervals and solid lines indicate means from n = 10 simulations each.
