Fig. 1

Identification of JMML-specific aberrant DNA methylation patterns. a Three-dimensional principal component analysis (PCA) of DNA methylation dynamics across 12 normal hematopoietic cell types. JMML samples were projected as additional data points. CMP, common myeloid progenitors; GMP, granulocyte-macrophage progenitors; HSC, hematopoietic stem cells; HSPCs, hematopoietic stem and progenitor cells; L-MPP, lymphoid-primed multipotent progenitors; MEP, megakaryocyte-erythroid progenitors; MPP, multipotent progenitor cells; NK cells, natural killer cells. b Relative proportions of hematopoietic cell types in each sample from the discovery cohort (n = 20)²². c Strategy used to identify JMML-specific differentially methylated probes (jmmlDMPs). Step 1: differentially methylated probes (DMPs) exhibiting dynamic changes during normal hematopoietic differentiation were identified between HSCs and each of six differentiated blood cell types (granulocytes, monocytes, NK cells, CD8⁺ T-cells, CD4⁺ T-cells, and B-cells). Probes were considered as DMPs if the adjusted p-value was < 0.05 and the methylation difference (Δmeth) was ≥ 0.2. Step 2: 59,230 unique hematopoiesis-specific DMPs (hemDMPs) were identified and removed from further analysis, resulting in 308,199 CpGs that are non-variable in hematopoiesis (nvCpGs). Step 3: consensus clustering identified stable JMML subgroups, for which JMML-specific DMPs were identified using adjusted p-value < 0.05 and Δmeth ≥ 0.2 as filtering criteria. Step 4: identification of jmmlDMPs and clustering of JMML samples into subgroups. d Consensus clustering of the 5,000 most variable nvCpGs identified 2 stable groups (k = 2) separating JMML samples. The consensus matrix shows pairwise cluster assignment frequencies derived from 500 iterations based on Manhattan distance metric and Ward’s linkage. Consensus values range from 0 (white) to 1 (dark blue). e Boxplots depicting the distribution of mean DNA methylation levels per JMML sample according to methylation group assignment across the 5,000 most variable CpG probes (top) and the 1,000 most variable CpG islands (CGI; bottom). Boxes represent the interquartile range and whiskers depict the minimum and maximum of the distribution. P-values are calculated using the two-sided unpaired Welch’s t-test. f Hierarchical clustering of the 1,000 most variable jmmlDMPs using Manhattan distance metric and Ward’s linkage. Samples (columns) are ordered according to consensus clustering results