Fig. 4: Methylation levels in T-ALL define different clusters of CGIs.

a, PCA based on the mean methylation of the variable commonly covered CGIs (n = 8,863 CGIs) of precursor T cells and patients with T-ALL. T-ALL samples show continuous CGI methylation levels instead of forming groups of high and low CGI methylation (n = 10 precursor T cell and 48 T-ALL samples). b, PCA based on the methylation status of the variable commonly covered CGIs (methylated defined as >0.2) of precursor T cells and patients with T-ALL (n samples and CGIs same as in a). c, Hierarchical clustering representing four clusters of CGIs identified by consensus clustering of variable CGIs across patients with T-ALL and healthy precursor T cells. Cluster 1 shows already a range from low to high hypermethylation across patients with T-ALL although targets are still rather sample-specific. Cluster 2 and 3 show dynamic, increasing hypermethylation across patients. CGIs in cluster 1 to 3 are unmethylated in precursor T cells, whereas cluster 4 shows higher methylation levels that become fully methylated in almost all patients with T-ALL. d, Fraction of CGIs per cluster overlapping promoters, active promoters (defined for genes with an average transcript per million (TPM) ≥1 across different precursor T-cell stages), DMVs, PMDs, gene bodies and intergenic regions. The fraction of CGIs in promoters decreases with increasing cluster-wise methylation level, whereas the proportion overlapping gene bodies, PMDs and intergenic regions rises (defined as 20% of a CGI or 20% of a feature overlapping). e, Fraction of CGIs per cluster overlapping chromatin states in HSCs and the T-ALL cell line DND41 (as a proxy for T-ALL). A CGI was assigned to the chromatin state with the largest overlap. f, Median methylation per CGI cluster for ALL subtypes, other hematopoietic malignancies and solid tumors. Although defined based on T-ALL, the clusters of CGIs show a similar tendency to gain methylation from cluster 1 to 4 as well as for the low/high group in other cancer types. Samples per tumor type (n same as in Fig. 1d) were averaged to generate a subtype-specific methylation signature (Methods).