Fig. 7

Translational application of epigenetic ITH to understand ccRCC metastasis. A Unsupervised hierarchical clustering of TCGA’s KIRC cohort using the 4333/5000 most variable CpGs that overlap with the EPIC array and do not have any missing data across samples. B Comparison of mean 5mC of the 4333 CpGs of the M0 tumors in cluster 1 to those in cluster 2, and the M1 tumors in cluster 1 to those in cluster 2. C Boxplot showing the over-representation of CNVs in cluster 2 irrespective of the metastasis status. Dots represent TCGA-KIRC samples, and the y-axis indicates the number of CNVs from a published set of CNVs linked to clinical features [1, 44]. D Comparison of mean 5mC of the 4333 CpGs of the M0 to the M1 tumors in cluster 1, and then the M0 to the M1 tumors in cluster 2. E Euclidian distance comparison between M0-like samples in our cohort and the normal kidneys, and the distance between the M1-like samples and the normal kidneys. F Difference in the number of observed CNVs in our M0-like and M1-like samples. G Phyloepigenetic tree for tumor w3 drawn using the 4333. M0-like and M1-like branches are colored differently. A metastatic sample obtained from the pancreas is included in this tree (red arrow) and branches off with two M1-like regions most distant from the normal kidney. H Phyloepigenetic tree for tumor w3 drawn using 236 enhancer-associated CpGs derived from Fig. 5. M0-like and M1-like branches are colored differently. The heatmap represents the average 5mC of the enhancer CpGs mapped to genes indicated on the tree in Additional file 2: Fig. S13: USP44 (n = 3), KRBA1 (n = 4), AKNS4B (n = 5), KDM8 (n = 4), and SATB2 (n = 2)