Fig. 5

DNA methylation changes are largely attributed to the cellular composition. a The multidimensional scaling plot demonstrates that PMF sample did not cluster by mutation allele frequency. b, c Correlation between DNAm and allele burden (across all driver mutations, given that they hardly affected DNAm) for the top two candidate CpGs: b) cg14658896_BC21 and c) cg16965444_BC21. d An epigenetic deconvolution algorithm [51] was applied to estimated fractions of granulocytes, monocytes, B cells, NK cells, CD4 and CD8 T cells. e, f To determine how many significant DNAm changes in PMF versus controls is attributed to CpGs that have high variation between leukocyte subsets we compared scatter plots e) before and f) after exclusion of CpGs with more than 10% DNAm between any of the leukocyte subsets (only CpGs measured across all datasets are shown). Significant hypo- and hypermethylated CpGs are indicated in blue and red (mean DNAm difference > 20%; adjusted p-values < 0.05). Gray numbers indicate all CpGs exceeding the mean DNAm difference > 20%, irrespective of statistical significance