DNA methylation in primary myelofibrosis is partly associated with driver mutations and distinct from other myeloid malignancies

Abstract

Background: Primary myelofibrosis (PMF) is a clonal blood disorder characterized by mutually exclusive driver mutations in JAK2, CALR, or MPL genes. So far, it is largely unclear if the driver mutations have a specific impact on DNA methylation (DNAm) profiles and how epigenetic alterations in PMF are related to other myeloid malignancies.

Results: When we compared DNAm profiles from PMF patients we found very similar epigenetic modifications in JAK2 and CALR mutated cases, whereas MPL mutations displayed less pronounced and distinct patterns. Furthermore, induced pluripotent stem cell (iPSC) models with JAK2 mutations indicated only a moderate association with PMF-related epigenetic changes, suggesting that these alterations may not be directly driven by the mutations themselves. Additionally, PMF-associated epigenetic changes showed minimal correlation with allele burden and seemed to be largely influenced by shifts in the cellular composition. PMF DNAm profiles compared with those from other myeloid malignancies-such as acute myeloid leukemia, juvenile myelomonocytic leukemia, and myelodysplastic syndrome-showed numerous overlapping changes, making it difficult to distinguish PMF based on individual CpGs. However, a PMF score created by combining five CpGs was able to discern PMF from other diseases.

Conclusion: These findings demonstrate that PMF driver mutations do not directly evoke epigenetic changes. While PMF shares epigenetic alterations with other myeloid malignancies, DNA methylation patterns can distinguish between PMF and related diseases.

Keywords: AML; CpG; DNA methylation; Epigenetic; JMML; MDS; MPN; Myeloid malignancies; Primary myelofibrosis.

Fig. 1

Aberrant DNA methylation in primary myelofibrosis. a Multidimensional scaling plot of DNA methylation profiles in PMF patients with different driver mutations (JAK2, CALR, MPL) and healthy controls (812,274 CpGs). b Scatter plot of mean methylation beta values of PMF patients and healthy controls. 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. c, d) Enrichment analysis of significant hyper- and hypomethylated CpGs in PMF patients in c genomic regions and d) CpG islands (Hypergeometric distribution: * = p < 0.05, # = p < 10^–10, + = p < 10^–20, and $ = p < 10^–100. e Differential mean DNAm of CpGs adjacent to all significant hypermethylated and hypomethylated CpGs (1 kb window). f Comparison of DNA methylation and gene expression changes (GSE26049) between PMF patients and healthy controls, with genes showing significant differences in both categories highlighted

Fig. 2

Changes in DNA methylation according to driver mutations. a-c Scatter plots illustrating the mean DNAm beta values in PMF patients compared to healthy controls, stratified by driver mutation: a) JAK2 mutation, b) CALR mutation, and c) MPL mutation. d-f Additional scatter plots comparing the mean DNAm beta values of PMF patients based on their driver mutation: d) CALR versus JAK2; e) MPL versus JAK2; and f) MPL versus CALR. Significant hypo- and hypermethylated CpGs are indicated in blue and red, respectively (mean DNAm difference > 20%; adjusted p-values < 0.05). g Comparison of significantly differentially methylated CpGs between JAK2 and MPL versus CALR and MPL, with gene names for overlapping CpGs highlighted

Fig. 3

iPSCs with JAK2 V617F fail to recapitulate disease-associated changes. a, b Scatter plots showing mean DNAm beta values of a) wild type (WT) versus JAK2 V617F heterozygous (het) iPSCs, and b) WT versus JAK2 V617F homozygous (hom) iPSCs. The numbers of CpGs with > 20% DNAm difference are indicated, but none reached statistical significance. c, d To determine if DNAm changes in PMF patients with JAK2 V617F are reflected in iPSCs with or without JAK2 V617F, we focused on CpGs that were significantly differentially methylated in JAK2 V617F PMF versus healthy control (from Fig. 2a). Average DNAm changes were then analyzed in these CpGs in iPSCs with either c) WT versus heterozygous JAK2 V617F, or d) WT versus homozygous JAK2 V617F. e, f Following differentiation of iPSC lines into hematopoietic progenitor cells (iHPCs), scatter plots depict mean DNAm beta values for e) WT versus JAK2 V617F heterozygous iHPCs, and f) WT versus JAK2 V617F homozygous iHPCs (none of the CpGs reached statistical significance). Gray numbers indicate all CpGs exceeding the mean DNAm difference > 20%, irrespective of statistical significance. g, h The CpGs with significant differences in JAK2 V617F PMF versus healthy controls were reanalyzed in iHPCs: g) heterozygous and h) homozygous JAK2 V617F iHPCs exhibited an overall decrease in DNAm at CpGs that gained or lost methylation in JAK2 V617F PMF. Statistical significance was evaluated using one-way ANOVA

Fig. 4

Age-associated DNAm changes in primary myelofibrosis. a–d The correlation between epigenetic age predictions with chronological age and the difference between predicted and chronological age (delta-age) was determined with epigenetic clocks developed by a,b) Horvath [26] and c,d) Han et al. [25]. Statistical significance was determined using an unpaired t-test. e–h The DNAm at an age-associated region in PDE4C was analyzed by bisulfite amplicon sequencing (BA-seq). The heatmaps exemplify the presence of methylated (red) and non-methylated (blue) CpGs within the PDE4C amplicon, covering 26 neighboring CpGs. The age-associated CpG of the aging signature is indicated by arrow. Exemplary heatmaps are depicted for e) a healthy donor, and f) a PMF patient blood sample of the same age. The frequency of reads is clustered according to their DNAm patterns. The same analysis was performed in colony-forming units (CFUs) on day 14 that were either g) a wild type (WT), or h) harbored the JAK2 V617F mutation. Unlike the blood samples from PMF patients or controls, the CFUs exhibited a distinct DNAm pattern that appears to reflect the clonal characteristics of the colony-initiating cells

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

Fig. 6

Comparison of myeloid malignancies after exclusion of cell type-specific CpGs. a, b Multidimensional scaling plots of DNAm profiles (216,532 CpG sites) in patients with PMF, MDS, JMML, AML, and healthy controls (peripheral blood (PB) and bone marrow (BM)): a) first versus second dimension; b) third versus fourth dimension. c Scatter plots showing mean DNAm beta values of healthy control versus PMF, MDS, JMML, or AML, accounting for potential differences between peripheral blood and bone marrow by using appropriate control sets. 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. d, e Venn diagrams illustrating CpGs that are overlapping d) hyper- or e) hypomethylated in the above-mentioned comparisons of four myeloid malignancies after the exclusion of cell type-specific CpGs

Fig. 7

A five-CpG signature can discern PMF from other malignancies and controls. a, b The DNAm levels at five CpGs (cg02210934, cg02739280, cg21708058, cg07197092, and cg08069247) were combined into a simple PMF score (PMF score = 5—sum of the five DNAm values). The PMF score is provided for datasets of the a) training set, and b) independent validation set. The datasets of PMF, other myeloid malignancies and healthy samples used from both our and public datasets are indicated with GSE numbers. c The PMF score did not correlate with mutation allele burden. d Box plot demonstrating that the driver mutations did not have significant impact on the PMF score. However, triple negative (TN) samples of publicly available PMF datasets (GSE152519) revealed a lower PMF score. Statistical significance was determined using an unpaired t-test