Supplementary Fig. 13: Clustering of single-cell ChIP-seq H3K27me3 profiles and scRNA-seq profiles of human tumor cells from the HBCx-95 model.

(a) Left: Plot of copy number in 0.5 Mb non-overlapping regions in Capecitabine-resistant PDX (HBCx-95-CapaR) versus untreated PDX (HBCx-95), obtained from the input of the bulk ChIP-seq experiments. Regions with a deviation greater than n = 2 standard deviations (dashed black lines) compared to the mean (diagonal black line) were removed for subsequent scChIP-seq analysis. Middle and right: snapshots of loci affected by copy number variation for bulk DNA profiles of Capecitabine-resistant PDX and untreated PDX indicated in gray. (b) Left panel: hierarchical clustering and corresponding heatmap of cell-to-cell Pearson’s correlation scores. Sample of origin (green for HBCx-95 [n = 474 single-cells] and pink for HBCx-95-CapaR [n = 838 single-cells]) and unique read count are indicated above the heatmap. Middle panel: t-SNE plots of scRNA-seq tumor cells, dots are colored according to the sample of origin and consensus clustering segmentation. Right panel: Consensus clustering scores for hierarchical clustering of scRNA-seq tumor cells. Consensus score ranges from 0 (white: never clustered together) to 1 (dark blue: always clustered together). Cluster membership is color coded beneath the dendrogram. (c) Consensus clustering analysis for scChIP-seq dataset. Left panel: mean of all pairwise correlation score between cluster’s members is plotted for k clusters ranging from 2 to 10. At k = 2 clusters, the intra-cluster correlation is maximized. Right panel: hierarchical clustering and corresponding heatmap of cell-to-cell consensus clustering scores for scChIP-seq on tumor cells (HBCx-95 and HBCx-95-CapaR PDXs). Consensus scores ranges from 0 (white: never clustered together) to 1 (dark blue: always clustered together). Cluster membership is color coded above heatmap. (d) Barplot displaying the -log10 of adjusted P values from pathway analysis for regions with depletion of H3K27me3 in resistant cells. The gene sets are indicated on the barplot. (e) Left panels: Aggregated H3K27me3 chromatin profiles for Chrom_c1 and Chrom_c2 (n = 457 and 794 single-cells for Chrom_c1 and Chrom_c2 respectively) are shown for the loci identified in Fig. 3f, as significantly differentially enriched and expressed (two-sided Wilcoxon signed-rank test). For each window indicated in gray, the log2 fold-change, the adjusted P value, the number and the proportion of cells with H3K27me3 enrichment within each cluster are indicated. Right panels: t-SNE representation of scRNA-seq datasets. Dots are colored according to expression signal in each cell.