Overlapping DNA methylation changes in enhancers in clonal cytopenia of undetermined significance and myelodysplastic neoplasm patients with TET2, IDH2, or DNMT3A mutations

The development of myelodysplastic neoplasms (MDS) is multiphasic, and the disease shares several clinical and molecular features with the precursor entities clonal hematopoiesis of indeterminate potential (CHIP) and clonal cytopenia of undetermined significance (CCUS) and the more aggressive acute myeloid leukemia (AML). Common to these is the presence of mutations in genes regulating DNA methylation.^1-3 Notably, two of the most mutated genes in CCUS and MDS are DNMT3A, catalyzing de novo methylation, and TET2, catalyzing the first steps of active DNA demethylation. We have previously shown that TET2-mutated CHIP and CCUS are associated with a distinct pattern of DNA hypermethylation especially enriched in genes related to immune response and leukocyte function.⁴ Another study found that cases of DNMT3A- and TET2-mutated CHIP have distinct and directionally opposing DNA methylation patterns but both likely promoting self-renewal of hematopoietic stem cells.⁵ Furthermore, IDH1 and IDH2 mutations have been shown to inhibit TET2-mediated demethylation in AML and have been associated with a DNA hypermethylation profile similar to that of TET2-mutated AML.⁶ The epigenome-wide impact of mutations in TET2, IDH1, IDH2, and DNMT3A in CCUS and MDS is, however, not yet fully elucidated nor has it been investigated whether the methylation patterns are shared across the range of clonal myeloid disorders.

We investigated whether 58 CCUS and 59 MDS patients with mutations in TET2, IDH2, or DNMT3A have distinct DNA methylation patterns in bone marrow mononuclear cells compared to patients without any of these mutations. Patients were recruited from hematology departments at three Danish hospitals with approval from The Danish National Committee on Health Research Ethics (ID NVK-1705391). All participants provided written informed consent prior to participation and the study was conducted in accordance with the Danish ethical regulation for work with human participants. Patients donated bone marrow aspirates at the time of diagnosis, from which mononuclear cells were isolated using density gradient separation and T-cell-depleted using the RoboSep Human CD3 Positive Selection Kit II and the RoboSep Magnetic Cell Separator (StemCell Technologies, Vancouver, Canada). DNA sequencing was performed during diagnostic work-up or using a targeted sequencing panel and Twist Libraries (Twist Bioscience, San Francisco, CA, USA; more information available upon request). Genome-wide DNA methylation levels were measured using Illumina’s Infinium MethylationEPIC v1.0 BeadChips at the Genomics Core at the Van Andel Institute (Grand Rapids, MI, USA) according to the manufacturer’s protocol. Data processing, quality control, and statistical analyses were done using R v4.2.0 or higher. Data were quality controlled using minfi v.1.36.0⁷ and normalized using functional normalization.⁸ Probes with detection P values >0.01, bead count <3 in at least 5% of samples, non-CpG sites, targeting to sex chromosomes, single nucleotide polymorphisms <5 bp from the target CpG, and probes which previously showed binding to multiple target CpG sites⁹ were excluded using ChAMP v.2.20.1 ¹⁰ After quality control and filtering, methylation levels of 743,875 CpG sites were available for analysis. ComBat from R package sva 3.38.0 was used for sample plate correction. Differentially methylated positions (DMP) were identified by linear modeling using Limma v.3.46.0¹¹ with M-values as outcome and mutation group as explanatory variable adjusted for age, sex, and batch. The wild-type group was defined as patients without mutations in TET2, DNMT3A, IDH1, and IDH2. P values were adjusted for multiple testing using Benjamini and Hochberg’s method to control the false discovery rate (FDR). If none of the DMP was significant after correcting for multiple testing, we selected the threshold for statistical significance to unadjusted P values <0.001. Associations between dichotomous outcomes were tested using the χ² test.

The patients were 41-93 years of age (CCUS 54-90 years, MDS 41-93 years), predominantly male, and most MDS patients had lower-risk disease (Online Supplementary Table S1).

We first investigated the association between TET2 mutations and genome-wide methylation levels in CCUS and MDS separately. Patients with mutations in DNMT3A, IDH1, or IDH2 were removed from the analyses resulting in 23 CCUS and 20 MDS cases with, and 20 CCUS and 26 MDS cases without, TET2 mutations (data on 20 CCUS patients from each group have been published previously)⁴ (Online Supplementary Figure S1A, B). After correcting for multiple testing, we identified 24,515 DMP in TET2-mutated CCUS compared to wild-type cases (Figure 1A). Almost all DMP (24,365 DMP, 99%) were hypermethylated. Similarly, we detected 7,242 DMP in TET2-mutated MDS compared to wild-type cases with 98% (7,139) DMP being hypermethylated (Figure 1B).

Figure 1.Differential methylation associated with TET2, IDH2, and DNMT3A mutations. Volcano plots of results from analysis of differentially methylated positions (DMP). (A) Twenty-three cases of TET2-mutated clonal cytopenia of undetermined significance (CCUS) compared to 20 cases of TET2 wild-type CCUS. (B) Twenty cases of TET2-mutated myelodysplastic neoplasms (MDS) compared to 26 cases of TET2 wild-type MDS. (C) Four cases of IDH2-mutant compared to 46 cases of IDH2 wild-type CCUS and MDS. (D) Eight cases of DNMT3A-mutant compared to 20 cases of DNMT3A wildtype CCUS. (E) Five cases of DNMT3A-mutated MDS compared to 26 cases of DNMT3A wild-type MDS. Red and blue dots represent hyper- and hypomethylated CpG sites, respectively, below the significance threshold. Horizontal dashed lines represent a false discovery rate of 0.05 in (A-C) and P<0.001 in (D-E). (F, G) IDH2-mutated CCUS and MDS (same as in Figure 1C) with purple dots representing the 24,365 CpG sites showing TET2 mutation-associated (ass.) significant hypermethylation in CCUS (F) and the 7,139 CpG sites showing TET2 mutation-associated significant hypermethylation in MDS (G). All IDH2-mutated cases had co-occurring mutations in SRSF2. P values and effect size estimates in all panels were derived using a linear regression model. All P values are two-sided and not adjusted for multiple comparisons.

Next, we investigated the association between IDH2 mutations and methylation levels. After removal of patients with TET2, DNMT3A, or IDH1 mutations, four cases had IDH2 mutations (1 CCUS and 3 MDS) and 46 were IDH2 wild-type (Online Supplementary Figure S1C). Like TET2 mutations, IDH2 mutations were associated with hypermethylation of 94% (24,259) of the 25,888 DMP identified after FDR correction (Figure 1C). Since 2-hydroxyglutarate, the metabolic product of mutated IDH2 proteins, inhibits the catalytic activity of TET2 enzymes, we investigated if the methylation profiles detected in IDH2-mutant cases were similar to those observed for TET2 mutations. We found that 98% (23,807 sites) and 97% (6,917 sites) of the TET2 mutation-associated hypermethylated sites in CCUS and MDS, respectively, were associated with increased methylation in IDH2-mutated cases (P<2.2x10^-16) (Figure 1F, G). IDH1 mutations were not studied due to a low mutation count (N=1).

Eight CCUS and five MDS cases had mutations in DNMT3A and 20 CCUS and 26 MDS cases did not, after removal of cases with TET2, IDH1, or IDH2 mutations (Online Supplementary Figure S1D, E). We did not detect any DMP in DNMT3A-mutated CCUS after FDR correction and only 704 DMP were detected at an unadjusted P<0.001 (Figure 1D). This may be due to differences in cell populations, lower clonality, and/or number of mutated cases. Similarly, only 1,808 DMP were detected in DNMT3A-mutated MDS at an unadjusted P<0.001 with 93% (1,681) being hypomethylated (Figure 1E). The larger number of DMP associated with TET2 than DNMT3A mutations may be due to type and location of the mutations (DNMT3A mutations were mainly missense whereas TET2 mutations were primarily frameshift and nonsense mutations, which have been associated with increased clonal selection¹²), cell type specificity, and/or power since a higher number of patients had TET2 than DNMT3A mutations. In summary, mutations in TET2 and IDH2 are associated with a hypermethylation phenotype affecting thousands of CpG sites in both CCUS and MDS, whereas DNMT3A mutations are associated with a hypomethylation phenotype, which we only detected in MDS. As we had previously found 2,741 hypermethylated CpG sites in TET2-mutated CHIP,⁵ we next investigated whether the same sites were affected by TET2 mutations across CHIP, CCUS, and MDS. We found that 2,464 CpG sites were available for evaluation after quality control, and 98% (2,458 sites) and 99.6% (2,455 sites) of those were positively associated with TET2 mutations in CCUS and MDS, respectively (P<2.2x10^-16) (Figure 2A, B). Similarly, 98% (23,973 sites) of the 24,365 sites hypermethylated in TET2-mutated CCUS were hypermethylated in TET2-mutated MDS (P<2.2x10^-16) (Figure 2C). This indicates that methylation changes associated with TET2 mutations are affecting many of the same loci irrespective of disease state.

Figure 2.Comparison of results from epigenome-wide analyses of TET2 and DNMT3A mutations in clonal hematopoiesis of indeterminate potential, clonal cytopenia of undetermined significance, and myelodysplastic neoplasms. Volcano plots of results of the analysis of differentially methylated positions (DMP) in: (A) TET2-mutated clonal cytopenia of undetermined significance (CCUS) (same as in Figure 1A) and (B) TET2-mutated myelodysplastic neoplasms (MDS) (same as in Figure 1B) with red dots indicating the 2,741 sites previously reported as being hypermethylated in TET2-mutated clonal hematopoiesis of indeterminate potential (CHIP).⁴ (C) TET2-mutated MDS (same as in Figure 2B) with red dots representing the 24,365 CpG sites showing TET2 mutation-associated (ass.) significant hypermethylation in CCUS. (D) DNMT3A-mutated MDS (same as in Figure 1E) with blue dots indicating the 3,803 CpG sites previously reported to be hypomethylated in DNMT3A-mutated CHIP.⁵ Margins display distributions of colored sites versus the rest of CpG sites available after quality control (either rest of Infinium HumanMethylation450-sites in the case of CHIP or Infinium HumanMethylationEPIC-sites in the case of CCUS or MDS). P values and effect size estimates in all panels were derived using a linear regression model. All P values are two-sided and not adjusted for multiple comparisons.

Next, we investigated whether DNMT3A mutations affect DNA methylation similarly in MDS and CHIP. We found that 85% (2,924 sites) of the CpG sites previously reported to be hypomethylated in DNMT3A-mutated CHIP⁵ (3,433 of 3,803 sites available for analysis) were associated with lower methylation in DNMT3A-mutated MDS (P<2.2x10^-16) (Figure 2D). In summary, the large overlap of mutation-associated DMP across different hematopoietic contexts indicates that methylation phenotypes associated with TET2 and DNMT3A mutations are remarkably specific, which is in line with previous findings in CHIP and AML.^4-6,13 Due to the large overlap in CCUS and CHIP, it is unlikely that mutation-associated DMP in MDS are influenced by blast count or Revised International Prognostic Scoring System score. Although mutation-specific DMP were distributed across the entire genome, they were significantly enriched at enhancer regions and depleted at CpG islands for both hypermethylated sites associated with TET2 and IDH2 mutations and hypomethylated sites associated with DNMT3A mutations. Indeed, 11% of hypermethylated sites versus 4% of non-hypermethylated sites detected in TET2-mutated CCUS were in enhancer regions (P<2.2x10^-16) (Figure 3A), and only 3% of hypermethylated TET2-mutated sites versus 20% of non-hypermethylated sites were in CpG islands (P<2.2x10^-16). Strikingly similar results were observed for hypermethylated sites associated with TET2-mutated MDS and IDH2-mutated CCUS and MDS (P<2.2x10^-16) (Figure 3B). Likewise, 18% of all hypomethylated sites versus 4% of the non-hypomethylated sites associated with DNMT3A mutations in MDS were in enhancer regions (P<2.2x10^-16) (Figure 3C) and only 8% of hypomethylated sites were in CpG islands while this was the case for 19% of non-hypomethylated sites (P<2.2x10^-16). Interestingly, TET2 mutation-associated hypermethylation was especially enriched in regions annotated as enhancers in monocytes (Figure 3D), whereas IDH2 mutation-associated hypermethylation was equally present at monocyte and hematopoietic stem cell enhancers (Figure 3E). In contrast, DNMT3A mutation-associated hypomethylation was enriched in hematopoietic stem cell enhancers (Figure 3F). These findings demonstrate that the methylation changes are non-random and context-dependent, implying that the activity of TET2 and DNMT3A may be targeted to specific genomic regions in hematopoietic cells. TET2 has been suggested to have the strongest demethylation effect proximal to bound transcription factors¹⁴ and we have previously shown a strong enrichment of ETS and C/EBP transcription factor binding motifs near TET2 mutation-associated hypermethylated sites in CHIP and CCUS.⁴

Figure 3.Annotation of TET2, IDH2, and DNMT3A mutation-associated differentially methylated positions to chromatin states and cell-specific enhancers. (A-C) Annotation of interrogated CpG sites and methylation changes to chromatin states of peripheral blood mononuclear cells. (A) Proportions of the 24,365 TET2 mutation-associated hypermethylated sites in clonal cytopenia of undetermined significance (CCUS) (red bars) and the rest of the sites (gray bars) at different chromatin states. (B) Proportions of the 24,259 IDH2 mutation-associated hypermethylated sites in CCUS and myelodysplastic neoplasms (MDS) (red bars) and the rest of the sites (gray bars) at different chromatin states. (C) Proportions of the 1,681 DNMT3A mutation-associated hypomethylated sites in MDS (blue bars) and the rest of the sites (gray bars) at different chromatin states. Differentially methylated positions (DMP) were annotated to genomic regions using the annotation package for Illumina’s methylation arrays and to chromatin states using the mononuclear dataset from the Roadmap Epigenomics project.¹⁵ (D-F) Proportions of DMP and the rest of available sites interrogated by the EPIC v1.0 array located in regions annotated as enhancers in hematopoietic stem cells (HSC), mononuclear cells (MNC), or monocytes. DMP are the 7,139 CpG sites hypermethylated in TET2-mutated MDS (95% confidence intervals (CI): 0.32-0.35 and 0.23-0.25 for monocyte and HSC enhancers, respectively) (D), the 24,259 CpG sites hypermethylated in IDH2-mutated CCUS and MDS (95% CI: 0.33-0.34 and 0.31-0.32 for monocyte and HSC enhancers, respectively) (E), and the 1,681 CpG sites hypomethylated in DNMT3A-mutated MDS (95% CI: 0.23-0.27 and 0.40-0.45 for monocyte and HSC enhancers, respectively) (F). TSS: transcription start site.

Gene ontology term enrichment analysis of genes in the vicinity of the DMP revealed strikingly similar results for TET2 and IDH2 mutation-associated hypermethylated sites and DNMT3A mutation-associated hypomethylated sites with enrichment of genes related to immune system process and leukocyte activation (Online Supplementary Figure S2A-D). Similarly, genes near the 442 DMP that were common to all three mutations were involved in immune regulation and immune cell activation (Online Supplementary Figure S2E). These findings support a shared target pathway of TET2, IDH2, and DNMT3A mutation-induced methylation changes. Taken together, our results demonstrate the existence of distinct and non-random DNA methylation changes associated with mutations in TET2, IDH2 and DNMT3A in CCUS and MDS. We observed similar mutation-induced methylation changes across a spectrum of myeloid clonal disorders, suggesting that mutation-associated methylation changes may be an early event in leukemogenesis. Moreover, mutations in TET2, IDH2, and DNMT3A especially induce methylation changes at enhancer regions but with differential preferences for enhancers according to cell type, suggesting that these mutations affect regulatory regions of genes at various stages of hematopoietic cell differentiation. This is in line with a previous study showing that DNMT3A mutations led to DNA hypomethylation in regulatory regions of genes that are related to hematopoietic stem cell activity, while TET2 mutations led to DNA hypermethylation at regulatory regions of genes that downregulate the activity of more differentiated progenitor cells, thus both maintaining a stem cell-like state.⁵ Future studies of the genes regulated by these enhancers may shed light on the potential pathogenic impact of methylation changes during leukemogenesis.

Footnotes

• Received April 4, 2024
• Accepted December 17, 2024

Correspondence

^#LG and KG contributed equally as senior authors

K. GRØNBÆK - kirsten.groenbaek@regionh.dk

Disclosures

KG is on advisory boards for GSK and Nanexa and has received research funding for a different project from Janssen. PAJ is a paid consultant for Zymo Research Corporation whose products were used in the DNA methylation analyses. KK, MT, JWH, CS, KRJ, ADØ, BP, JW and LG have no conflicts of interest to disclose.

Funding

The study is part of the Danish Research Center for Precision Medicine in Blood Cancers funded by the Danish Cancer Society grant number R223-A13071 and was supported by a Center grant from the Novo Nordisk Foundation (Novo Nordisk Foundation Centre for Stem Cell Biology, Dan Stem; grant number NNF17CC0027852) and Greater Copenhagen Health Science Partners, and through a grant from Dansk Kræftforskningsfond. KK was supported by a 1-year grant from the University of Copenhagen.