Extended Data Fig. 3: Sequential staining of the biofilm and the subsequent spatial segmentation according to the final fluorescent image.

Sequential staining of the biofilm and the subsequent spatial segmentation according to the final fluorescent image. (a) Fluorescence images of a biofilm after labeling at different stages. Shown are composite images of the biofilm (phase contrast; mCherry, red; GFP, green; CFP, blue) right after the sequential labeling by TADA, NADA and HADA; Scale bar is 200 µm. Note that TADA molecules were absorbed by the microfluidic chamber (PDMS) during labeling, giving rise to the background signal; Further cultivation using medium without TADA washed out these absorbed dyes and reduced the background signal to the normal level. Actually, that is the main reason we used TADA as the first dye in the sequential labeling. (b) Extracting the fluorescent signals for each pixel of the biofilm in microscopic image; the pixel-fluorescence matrix is presented in the three-dimensional fluorescent space. Next, k-means clustering is executed to assign each pixel to one particular group according to its fluorescent pattern; shown are clustering of all pixels into (c) 4, (d) 8, (e) 12, (f) 20 and (g) 30 groups. According to the clustering result for all pixels, we segmented the biofilm into (h) 4, (i) 8, (j) 12, (k) 20 and (l) 30 groups (scale bar: 500 μm). Note that the number of groups to be clustered (N) is a manual parameter in k-means clustering; when it is set below a threshold (N ≤ 20), the biofilm can be perfectly segmented into multiple concentric rings from the periphery to interior of biofilm—the major direction of spatial heterogeneity (as shown in (h), (i), (j) and (k). However, when N is set too big, the rainbow structure in biofilm segmentation will be disrupted, as shown in (l) N = 30. This gives rise to an upper limit (N = 20) of computational segment of stained biofilm, which further limits the spatial resolution of current version of RAINBOW-seq.

Source data