Epigenome-wide DNA methylation association study of CHIP provides insight into perturbed gene regulation

A hallmark of aging is the accumulation of somatic mutations in dividing cells. The vast majority of these mutations do not affect cell fitness. In rare circumstances, however, a mutation can arise in a progenitor cell that confers a selective fitness advantage, culminating in its expansion relative to other cells. In the hematopoietic system, this process is termed clonal hematopoiesis (CH). Individuals with CH are at increased risk for the development of hematologic malignancies¹. A subset of CH is driven by pathogenic mutations in myeloid malignancy-associated genes, which is termed CH of indeterminate potential (CHIP) and has been shown to be associated with hematologic cancers, cardiovascular disease (CVD), chronic obstructive pulmonary disease, and mortality, among other conditions^2,3,4.

The prevalence of CHIP increases with advancing age^2,5,6,7. In a whole genome sequencing (WGS) study from the NHLBI Trans-Omics for Precision Medicine (TOPMed) program that included ~100,000 individuals across 51 separate studies, large CHIP clones were found to be uncommon (<1%) in individuals younger than 40 years of age and increased to 12% in those aged 70–89 and 20% in those aged 90 years and older⁵. This age-dependent pattern was consistent across CHIP driver genes⁵ and has been observed in other studies^2,6,7.

DNA methylation (DNAm), the addition of a methyl group to a cytosine followed by a guanosine (CpG) in DNA, is an epigenetic modification that reflects age and environmental exposures. The gene products of the three most frequently mutated CHIP driver genes, DNMT3A, TET2, and ASXL1, are epigenetic regulators⁵. DNMT3A (DNA-methyltransferase 3A) is a methyltransferase that catalyzes the transfer of methyl groups to CpG sites and catalyzes de novo DNA methylation⁸. Conversely, TET2 (ten-eleven translocation-2) is a DNA demethylase that catalyzes the conversion of 5-methylcytosine to 5-hydroxymethylcytosine, one of the steps leading to eventual demethylation of CpG sites⁹. ASXL1 (ASXL transcriptional regulator 1) is involved in histone modification¹⁰. Its function in CHIP remains relatively unknown¹¹.

CHIP has been shown to be associated with global DNAm changes, particularly for the DNMT3A and TET2 CHIP driver gene mutations¹². A previous epigenome-wide association study (EWAS) of CHIP in 582 Cardiovascular Health Study (CHS) participants, with replication in 2655 Atherosclerosis Risk in Communities (ARIC) participants, revealed several thousand CpG sites associated with CHIP and its two major CHIP driver genes, DNMT3A and TET2¹². DNMT3A and TET2 CHIP were also found to have directionally opposing DNAm signatures: DNMT3A CHIP mutations were associated with hypomethylation of CpGs, whereas TET2 CHIP was associated with hypermethylation of CpGs, consistent with the canonical regulatory functions of DNMT3A and TET2 elucidated in murine and human model systems^13,14,15.

Despite the wealth of information from the previous EWAS of CHIP¹², several limitations and knowledge gaps remain. These include the need to use larger sample sizes to enable analyses of less prevalent CHIP driver gene mutations such as ASXL1, explore downstream functions and pathways influenced by mRNA expression for any CHIP and CHIP subtypes, and identify underlying molecular mechanisms linking CHIP to CVD.

To address these knowledge gaps, we conduct a multiracial meta-analysis of separate EWAS of CHIP in four independent cohort studies (N = 8196; 462 with any CHIP, 261 DNMT3A, 84 TET2, and 21 with ASXL1 CHIP) along with analysis of the associations of CHIP-related CpGs with downstream gene expression. We expand upon the previous EWAS of CHIP study¹² by adding two cohorts—the Framingham Heart Study (FHS) and the African-American Jackson Heart Study (JHS) – in addition to the ARIC and CHS cohorts. The EWAS findings are functionally validated using human hematopoietic stem cell (HSC) models of CHIP. Expression quantitative trait methylation (eQTM) analysis identifies gene expression changes associated with CHIP-associated CpGs. Causal inference analysis using two-sample Mendelian randomization (MR) is performed to gain insight into the molecular mechanisms linking CHIP to CVD. A flowchart of the study design is shown in Fig. 1.

This flowchart outlines the sequential steps of the study, from data collection to downstream analyses. CHIP Clonal Hematopoiesis of Indeterminate Potential, WGS Whole Genome Sequencing, WES Whole Exome Sequencing, TOPMed Trans-Omics for Precision Medicine program, eQTM Expression Quantitative Trait Methylation.

Full size image

Clinical characteristics of study participants

The baseline characteristics of FHS, JHS, CHS, and ARIC participants included in this investigation are presented in Table 1. The mean age at the time of blood draw for whole-genome sequencing (WGS) was 57, 56, and 58 for FHS, JHS, and ARIC, respectively. Participants from CHS were considerably older, with a mean age of 74 years. All four cohorts had more women than men (54–63% women). Overall, CHIP mutations with a variant allele frequency (VAF) ≥ 2% were present in 5% (166/3295) of participants in FHS, 4% (68/1664) in JHS, 5% (142/2655) in ARIC, and 15% (86/582) in CHS. Consistent with previous reports⁵, the three most frequently mutated CHIP driver genes across all cohorts were DNMT3A, TET2, and ASXL1. Eighty percent of individuals with CHIP demonstrated expanded CHIP clones with VAF > 10%.

Epigenome-wide association analysis

Race was classified based on self-report. In the race-stratified analysis, we identified 2843 CpGs associated with any CHIP, 758 with DNMT3A, 4735 with TET2 CHIP in White participants and 5498 with any CHIP, 5065 with DNMT3A, and 290 with TET2 CHIP in Black participants at Bonferroni-corrected P < 1(times {10}^{-7}) (Supplementary Data 1–6). 1290, 675, and 254 CHIP-associated CpG sites were shared between White and Black participants at the Bonferroni-corrected threshold, with concordant directions of effect for any CHIP, DNMT3A, and TET2 CHIP, respectively.

In a multiracial, meta-EWAS of CHIP, 9615 CpGs were associated with any CHIP, and 5990, 5633, and 6078 CpGs were associated with DNMT3A CHIP, TET2 CHIP, and ASXL1 CHIP, respectively (at Bonferroni-corrected P < 1(times {10}^{-7})). The top ten CpGs for any CHIP and for each of the three CHIP driver genes are shown in Table 2. A full list of CpG signatures and their directions of effect are reported in Supplementary Data 7–10. There was minimal to moderate overlap of CpGs associated with DNMT3A, TET2, and ASXL1; 429, 904, and 1088 CpG sites were shared between DNMT3A and TET2, DNMT3A and ASXL1, and TET2 and ASXL1, respectively.

We identified 5987 CpGs (~100%) associated with DNMT3A CHIP and 4607 CpGs (~76%) associated with ASXL1 CHIP that showed decreased methylation (β < 0) (Fig. 2b, d). In contrast, 5079 CpGs (~90%) associated with TET2 CHIP showed increased methylation (β > 0) (Fig. 2c). Out of the 554 TET2-associated CpGs that showed decreased methylation, 171 (~31%) CpGs were found to overlap with DNMT3A CpG sites. The vast majority of CpGs associated with CHIP were remote (>1 Mb) from the driver gene including 5969/5990 (99.6%) for DNMT3A, 5632/5633 (~100%) for TET2, and 6070/6078 (99.9%) for ASXL1.

Volcano plots showing the effect size (β) and -log₁₀(P-value) from the multiracial meta-analysis of epigenome-wide association studies (EWAS) for (a) any CHIP (Clonal Hematopoiesis of Indeterminate Potential), (b) DNMT3A CHIP, (c) TET2 CHIP, and the EWAS in the Framingham Heart Study (FHS) for (d) ASXL1 CHIP. Genes annotated to the CpG sites are shown. For panels (a–c), the effect size (β) and P-values were derived from fixed-effect meta-analysis of multiple cohorts. For panel (d), because ASXL1 CHIP was only available in the FHS cohort, the effect size (β) and P-values were derived from linear regression models. The color green indicates a significant negative association between CHIP and DNA methylation while purple indicates a positive association between the two variables. Yellow signifies non-significant associations between CHIP and DNA methylation. Two-sided tests were used for all analyses. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg false discovery rate (FDR) method. Significant associations were defined as FDR < 0.05. Exact P-values, standard errors for the β, and 95% confidence intervals for significant results are provided in Supplementary Data 7–10. Source data are provided as a Source Data file.

Full size image

Although age was included as a covariate, there remains a possibility that common CpG sites across all three CHIP driver genes may be related to age rather than CHIP mutation. To assess this, we compared the 19 CpGs that are common among all three CHIP driver gene mutations from our meta-EWAS of CHIP with the CpGs from a recent EWAS of chronological age in the Generation Scotland cohort¹⁶ (N = 18,413). No CpGs overlapped, suggesting that common CpGs across all CHIP driver genes are related to CHIP mutation rather than age.

A sensitivity analysis was performed by excluding CHIP cases with VAF < 10%. The results are similar to the multiracial meta-EWAS of any CHIP and are provided in Supplementary Fig. 4 and Supplementary Data 11–13. Approximately 78% of CpGs (7460/9615) in the meta-EWAS of any CHIP were re-identified in the sensitivity analysis, while 312 CpGs were newly identified.

Human hematopoietic stem cell models of CHIP validate EWAS

We sought to experimentally validate our multiracial meta-EWAS methylation findings with an in vitro model of CHIP. CHIP was modeled by introducing loss-of-function mutations in DNMT3A, TET2, and ASXL1 in mobilized peripheral blood CD34+ hematopoietic cells, using CRISPR-Cas9¹⁷. After seven days in culture, these cells were flow sorted to isolate a purified population of CD34⁺CD38^–Lin^– cells. Following fluorescence-activated cell sorting, genomic DNA (gDNA) was extracted, and methylation was assayed using biomodal duet evoC (see Methods)¹⁸.

The analysis focused on the subset of CpG sites that were significantly associated with CHIP in the EWAS data and nominally differentially methylated (P < 0.05) in the in vitro model of CHIP. When comparing CpG site subsets with their respective engineered cells, DNMT3A-associated CpG sites showed significant enrichment in DNMT3A-engineered cells (P < 2.88 × 10^-239) (Fig. 3A), while TET2-associated CpG sites were significantly enriched in TET2-engineered cells (P < 8.39 × 10^-56) (Fig. 3B), and ASXL1-engineered cells (P < 1.65 × 10^-14) (Supplementary Fig. 5). ASXL1-associated CpG sites showed no significant hits in ASXL1-engineered cells (Fig. 3C), but a slight trend in TET2-engineered cells (Supplementary Fig. 5). The any CHIP-associated CpG sites were significantly enriched in DNMT3A-engineered primary cells only (P < 9.29 × 10^-121), unlike TET2 and ASXL1 (Supplementary Fig. 5).

Dot plots of methylation change from −1.0 (no methylation) to 1.0 (complete methylation) seen in engineered primary cell cultures compared to correlation of EWAS results ranging from −0.1 to 0.1. Following an initial CpG filtering using an uncorrected Student’s t-test (p < 0.05), significance was determined with a two-sided binomial test. a DNMT3A-associated CpG sites (n = 855 CpG sites) compared to DNMT3A engineered human stem cells (n = 4). b TET2-associated CpG sites (n = 312) compared to TET2-engineered human stem cells (n = 4). c ASXL1-associated CpG Sites (n = 139) compared to ASXL1-engineered human stem cells (n = 3). Source data are provided as a Source Data file. KO Knockout, mC methylcytosine.

Full size image

CpG association with gene expression and pathway analyses

To investigate the functional consequences of CHIP-associated CpGs, we performed gene ontology (GO) and pathway enrichment analysis for genes harboring CHIP-associated CpGs. For any CHIP, DNMT3A CHIP, and ASXL1 CHIP, the enriched GO terms related to broad cellular developmental and organismal processes, while for TET2 CHIP the top GO terms related to cellular regulation and cell signaling (Supplementary Data 14–17). For example, for any CHIP, DNMT3A CHIP, and ASXL1 CHIP, the top ten most significant ontologies included multicellular organism development, anatomical structure development, system development, and developmental process. For TET2 CHIP, the most significant ontology terms related to a cellular response to stimulus, regulation of cellular processes, and cell signaling. Notably, for the genes annotated to the 554 TET2-associated CpGs that were found to be demethylated, the top GO terms were enriched for cellular developmental and organismal processes such as multicellular organism development and system development—similar to the enriched GO terms for genes annotated to DNMT3A CHIP-associated CpGs (Supplementary Data 29).

To understand how differentially methylated CpGs associated with CHIP might alter cellular function, we identified gene expression changes associated with CHIP-linked CpGs. We analyzed the associations of CHIP-associated CpGs with changes in cis gene expression (expressed gene [eGene] within 100 kB of CpG) in 2115 FHS participants whose DNA methylation data and whole-blood RNA-seq data were available. At P < 1(times {10}^{-7}), we identified 467 significant cis CpG-transcript pairs for any CHIP, 258 for DNMT3A CHIP, 293 for TET2 CHIP, and 234 for ASXL1 CHIP (Supplementary Data 18–21 provide the full expression quantitative trait methylation (eQTM) results)¹⁹. The vast majority of the associations between methylation and gene expression changes were negative, where decreased methylation changes were associated with increased gene expression changes or increased methylation changes were associated with decreased gene expression changes. For any CHIP, DNMT3A, TET2, and ASXL1 CHIP, ~68% (317/467), ~71% (184/258), ~77% (224/293), and ~72% (168/234) of CpGs had a negative association between methylation and gene expression changes, respectively. For any CHIP, the top enriched GO terms related to lipid metabolism. eGenes associated with DNMT3A CHIP were enriched in cell motility and adhesion processes. For TET2 CHIP, the top enriched terms related to immune processes, such as leukocyte differentiation. ASXL1 CHIP eGenes were enriched in cellular and immune processes, including cell importation and antigen processing and presentation (Supplementary Data 22–25).

Association of methylation with variants and MR analysis

Cis-methylation quantitative trait loci (cis-mQTL)—genetic loci that are significantly associated with CpG methylation levels and located within 1 Mb of their associated CpG—linked 8642 CpGs associated with any CHIP and CHIP subtypes to GWAS Catalog traits/diseases^19,20. Of the cis-mQTL variants, a subset were associated with clonal hematopoiesis traits, particularly myeloid clonal hematopoiesis and the number of clonal hematopoiesis mutations (Supplementary Data 26).

Additionally, enrichment tests of CHIP-associated CpG sites with EWAS catalog traits²¹ were performed across 4023 traits using a significance threshold of (1.24times {10}^{-5}) (0.05/4023) (Supplementary Data 27). For any CHIP, DNMT3A CHIP, TET2 CHIP, and ASXL1 CHIP, the top outcomes reflected CpG sites related to age/aging, alcohol consumption, smoking, and multiple CVD-related traits including body mass index (BMI), type II diabetes, and fasting insulin. In support of previous studies reporting ASXL1 CHIP enrichment among smokers^22,23, 24% (1462/6078) of ASXL1 CHIP-associated CpGs overlapped with smoking-associated CpGs.

Two-sample MR analysis of CHIP-associated CpGs (as exposures) with cis-mQTLs as the instrumental variables in relation to CVD-related traits and mortality (as outcomes) was performed to infer whether differential methylation at CHIP-associated CpGs may causally influence the outcomes. The significantly associated CpGs for any CHIP and for the three CHIP driver genes were tested for causal associations with 22 traits, including all-cause mortality, BMI, LDL cholesterol, hypertension, diabetes, CVD, and smoking. The top 20 CpGs and annotated genes for each trait are reported in Table 3 (Supplementary Data 28 displays the full MR results). 261 CHIP-associated, differentially methylated CpG sites were identified that were putatively causally associated with CVD-related traits and/or all-cause mortality, including eight CpGs for myocardial infarction (MI) (e.g., cg11879188 (ABO), β_MR = −0.99, P_MR = (4.8times {10}^{-18})), 108 CpGs for blood pressure (e.g., cg20305489 (SEPT9), β_MR = 10, P_MR = (1.7times {10}^{-31})), 86 CpGs for lipid traits (e.g., cg11250194 (FADS2), β_MR = −0.89, P_MR = (2.0times {10}^{-33})), and two CpGs for mortality (e.g., cg08756033 (C13orf33), β_MR = 0.016, P_MR = (1.3times {10}^{-4})). 53 CpGs were associated with more than one trait. For example, cg11879188 is annotated to the ABO gene and was associated with seven traits, including diastolic blood pressure (β_MR = 2.7, P_MR = (1.9times {10}^{-23})), MI (β_MR = −0.99, P_MR = (4.8times {10}^{-18})), and triglycerides (β_MR = 0.20, P_MR = (2.0times {10}^{-9})).

We report the results of a multiracial meta-EWAS of CHIP and identified thousands of CpG sites across the genome that are significantly associated with any CHIP and with DNMT3A, TET2, and ASXL1 CHIP. Of note, the vast majority of the CpGs were trans- relative to the CHIP driver gene. This appears to be consistent with the functions of DNMT3A, TET2, and ASXL1 in globally altering DNA methylation levels of CpG sites genome wide, as seen in the EWAS of each of the three CHIP driver genes, where the significantly associated CpGs were numerous and located diffusely across the genome. The methylomic signatures of CHIP and CHIP driver genes were experimentally validated with human-engineered CHIP cells. Downstream analyses were conducted to assess whether these alterations in DNA methylation levels may be causally associated with CVD-related outcomes and all-cause mortality. Causal inference analyses using two-sample MR revealed evidence of a possible causal role of CHIP-associated CpGs in various CVD-related traits and all-cause mortality.

For the experimental validation of our meta-EWAS results, any CHIP-associated CpG sites were significantly enriched in DNMT3A-engineered cells, which was expected given the overwhelming predominance of DNMT3A CHIP among total CHIP cases reported in our study and several others^2,5,12. Interestingly, TET2-associated CpG sites were enriched in ASXL1-engineered cells. This finding is consistent with the substantial CpG overlap ( ~ 1000 shared CpGs) between TET2 and ASXL1 CHIP from the meta-EWAS and suggests that the epigenetic regulators TET2 and ASXL1 impact several of the same genome regions and may lead to similar downstream consequences. Notably, ASXL1-associated CpGs showed no significant enrichment in the ASXL1-engineered cells. The lack of enrichment of ASXL1-associated CpGs in the ASXL1-engineered cells may limit the validity of the study’s downstream analyses with ASXL1 CHIP. This observation may be due to the limited number of ASXL1 CHIP cases in the EWAS as well as several biological factors. ASXL1 mutations primarily affect histone modifications, particularly H2AK119 ubiquitination, which indirectly influences chromatin accessibility²⁴. Recent studies have shown that ASXL1 loss-of-function mutations increase chromatin accessibility, potentially resulting in individualistic methylation changes influenced by other genetic and environmental factors²⁵. Furthermore, the effects of ASXL1 mutations on methylation might be temporally dynamic or cell-type specific, aspects not fully captured in our current experimental design. Future studies with larger sample sizes, particularly for ASXL1 CHIP, and longer observation periods may help elucidate the relationship between ASXL1 mutations and DNA methylation patterns in CHIP and determine with greater certainty the epigenetic signatures of this CHIP driver mutation.

By incorporating an engineered reductionist system, we provide an orthogonal approach to confirm that the patterns observed in CHIP donors result directly from the somatic mutation. By engineering DNMT3A, TET2, and ASXL1 mutations into healthy CD34+ cells and performing DNA methylation profiling, we can recapitulate the DNA methylation patterns seen in CHIP donors. Our study also establishes this reductionist system as a robust representation of the methylation phenotype. Future studies leveraging this system will enable more precise dissection of the causal relations between these mutations and changes in DNA methylation than would be possible from population-scale epidemiology data alone.

In the in vitro validation analysis, we focused on the subset of CpG sites that were significantly associated with CHIP in the EWAS and nominally differentially methylated in the in vitro CHIP model to improve the validity of our findings and reduce the likelihood of false positives. By concentrating on overlapping CpG sites, we prioritized CpGs with a stronger potential biological relevance, as they were consistent across population-level and experimental settings. Importantly, despite the strong directional concordance in the subset of overlapping CpG sites, a small proportion of the CpGs identified from the meta-EWAS were captured in the in vitro model: ~14% (855/5990) for DNMT3A, ~6% (312/5633) for TET2, and ~2% (139/6078) for ASXL1 CHIP. This may be because the in vitro system does not fully capture the complex in vivo environment in which CHIP is influenced by various cell types, environmental factors, and systemic interactions, such as immune system interactions.

Two-sample MR analysis identified 261 differentially methylated CpG sites that were putatively causally related to one or more CVD traits and/or all-cause mortality. For example, cg11250194 was putatively causally associated with four CVD-related cardiometabolic traits: LDL cholesterol, HDL cholesterol, triglycerides, and fasting glucose. Cg11250194 resides in the FADS2 gene. It is hypomethylated, associated with DNMT3A CHIP (β = −0.022, P = 1.6E-13), and replicated in the DNMT3A CHIP-engineered cells. The FADS2 gene encodes the enzyme fatty acid desaturase 2 – the first rate-limiting enzyme for the biosynthesis of polyunsaturated fatty acids²⁶. A recent study found that cg11250194 (FADS2) was associated with Alternative Healthy Eating Index and that hypermethylation of this CpG was associated with lower triglyceride levels²⁷. Moreover, cg11250194 was previously identified in an EWAS of lipid-related metabolic measures²⁸. Based on our findings, hypomethylation of this diet-associated CpG may be linked to higher triglyceride levels, putatively increasing the risk for CVD. FADS2 overexpression has also been found to promote clonal formation²⁶. Thus, FADS2 may be an important gene connecting CHIP with diet. Of note, of the 30 CpGs associated with either Mediterranean-style Diet Score or Alternative Healthy Eating Index or both in a 2020 study by Ma et al.²⁷, 17 were CHIP-associated CpGs (~57%) identified from our multiracial meta-EWAS of CHIP. The substantial overlap between diet- and CHIP-associated CpGs is consistent with the hypothesis that an unhealthy diet may be associated with CHIP through epigenetic mechanisms. Despite including smoking status as a covariate in the models for all participating cohorts in both studies²⁸, we recognize that our ability to fully separate the effects of diet from smoking on CHIP risk may be limited due to our cross-sectional study design. As a result, some of the observed association between unhealthy diet and CHIP through epigenetic mechanisms may reflect residual smoking effects.

Compared to a previously published EWAS of CHIP¹² (N = 3273, 228 CHIP cases), the present study has a substantially larger sample size (N = 8196, 462 CHIP cases), including all the samples from the previous study. With the larger sample size of the present study, we identified 6687, 3524, and 4678 novel CpGs significantly associated with any CHIP and with the top two CHIP driver genes DNMT3A and TET2. Of the CpG sites identified from the previous EWAS study at P < 1(times {10}^{-7}), a large proportion overlapped and have concordant directions of effect with CpGs from the multiracial meta-EWAS of CHIP at P < 1(times {10}^{-7}): 91% (2928/3217) for any CHIP, 89% (2466/2769) for DNMT3A CHIP, and 90% (955/1059) for TET2 CHIP. This is expected, as almost half of the CHIP cases in our meta-EWAS of CHIP are from the previous EWAS of CHIP¹². Additionally, we report thousands of ASXL1 CHIP-associated CpGs from an EWAS of ASXL1 CHIP. Through eQTM analysis that identified CpG-transcript pairs, the top eGenes in ASXL1 CHIP relate to various immune processes, suggesting that dysregulated immune function may contribute to ASXL1 CHIP-related disease outcomes. This putative role of ASXL1 CHIP in perturbing immune function, specifically T cell function, has been recently reported using an ASXL1 CHIP conditional knock-in mouse model²⁹. Notably, several of the ASXL1 CHIP-associated CpGs displayed putatively causal relations to CVD-related traits in MR analysis, including cg11879188 (in ABO).

While there are several strengths of our study, some limitations should be noted. Although smoking status was included as a covariate in the statistical models for all study cohorts, there could be residual confounding as smoking behavior may not be fully adjusted for in the analysis. Thus, smoking could still be driving the association between CHIP and CVD, as was reported in a recent study of CH³⁰. Of note, there are several studies across diverse populations and in different settings that controlled for smoking as a covariate and also found an association of CHIP (a specific form of CH) with CVD. For instance, in a recent study, Diez-Diez et al.³¹ clarified the directionality of the CHIP-CVD relationship with adjustment for smoking and concluded that CH confers an increased risk of developing atherosclerosis.

Additionally, the way that the DeCODE³⁰ investigators ascertained “CH” may have contributed to their finding that CH was not associated with CVD, as it is distinct from the definition of “CHIP.” CH includes clonal events with known leukemic driver gene mutations, such as CHIP and mosaic chromosomal alterations (mCAs), and clonal events without clear driver genes. Given recent findings that each of these distinct classes of CH has unique phenotypic consequences^2,32, the lack of association between CH and CHIP reported by Stacey et al. may be due to the grouping of heterogenous CH subtypes.

Moreover, it has been previously demonstrated that small changes in the stringency with which CHIP is ascertained can have an outsize effect on downstream analyses. For example, Vlasschaert et al.³³ reported the importance of CHIP detection stringency in relation to CHIP-associated CVD risk. Specifically, more stringent criteria (≥5 supporting reads) were associated with CVD risk, while less stringent criteria (≥3 supporting reads) attenuated the association of CHIP with CVD. Their study provides an up-to-date and nuanced explanation of the CHIP-CVD relationship.

Driverless CH is the occurrence of clonal expansions in blood without a known CHIP driver mutation and is estimated to drive the majority of clonal expansions in the elderly³⁴. Bernstein et al.³⁴ identified regions within exome sequences that are under positive selection to identify additional driver mutations in whole blood (large clones >0.1) together with validation of positive selection in single cell-derived hematopoietic myeloid and lymphoid colonies. The inclusion of mutations in these fitness-inferred CH genes increases prevalence of CH by 18% in the UK Biobank cohort. In our study, CHIP was defined when an individual harbored at least one deleterious insertion/deletion or single nucleotide variant in any of the 74 genes that have been previously linked to myeloid malignancy at a variant allele frequency of at least 2%. Given the study’s scope, we did not include driverless CH in our CHIP definition. Thus, CHIP prevalence in our study may be underestimated relative to studies that account for driverless clonal expansion.

A larger sample size is needed to examine less frequently mutated CHIP driver genes, such as TP53, JAK2, and PPM1D. Moreover, the reported putatively causal associations of CpGs with CVD outcomes and mortality were based on two-sample MR analysis. Despite our attempt to minimize horizontal pleiotropy by excluding cis-mQTLs that also serve as trans-mQTLs and excluding CpGs with three or more independent instrumental variables through MR-Egger using a threshold of P-value < 0.05, the two-sample MR approach has known limitations³⁵. Methods for detecting and addressing pleiotropy may be ineffective³⁶ and, thus, longitudinal and functional studies are needed to reinforce causal findings.

To account for the possibility that the VAF of CHIP driver gene mutation may influence DNA methylation, we used a threshold of FDR < 0.05 to detect associations between CpGs and VAF. There were no significant associations between CpG sites and VAF. Based on these results, we do not believe that VAF significantly influenced our DNA methylation findings. Notably, the sample size with available VAF information is limited. For example, in the FHS cohort, we have only 166 CHIP cases and cannot completely rule out the possibility that VAF may still have a modest impact on DNA methylation. Experimental studies or studies with larger sample sizes may be necessary to address the effect of VAF of mutation on DNA methylation.

While this study benefits from a large sample size, which allows for robust statistical comparisons, we acknowledge the distinction between statistical significance and biological significance. The statistical significance observed in this study does not necessarily equate to meaningful biological effects. The impact of cell type correction on downstream analyses, particularly in heterogeneous patient populations, has not been fully validated. Further research is needed to determine how this adjustment translates into biological outcomes. Additionally, our study evaluated individual CpG sites, however, differentially methylated regions consisting of several consecutive methylated CpGs have been shown to have important implications for disease pathogenesis³⁷. Thus, studies exploring these broader methylation patterns are warranted to better capture the functional relevance of epigenetic signatures of CHIP.

Last, although cell-type proportions were included as covariates for all cohorts, we cannot exclude the possibility that subtle uncorrected effects in cell-type proportions due to clonal selection in immune cells may contribute to the enrichment of immune function observed for TET2 and ASXL1 CHIP eGenes. While cell-type adjustments reduce confounding effects, residual contributions from altered immune cell proportions remain possible. Future studies investigating cell-type specific DNA methylation and gene expression may provide additional clarity on the impact of CHIP on immune gene expression.

Overall, our study sheds light on the epigenetic changes linked to CHIP and CHIP subtypes and their associations with CVD-related outcomes. The differentially expressed genes and pathways linked to the epigenetic features of CHIP may serve as therapeutic targets for CHIP-related diseases. For example, Fc receptor-like protein 3 (FCRL3) (cg17134153, Fx = -5.5, P = 1E-113) is the top differentially expressed gene for TET2 CHIP. FCRL3 encodes a type I transmembrane glycoprotein that is expressed by lymphocytes and plays a role in modulating immune responses³⁸. Polymorphisms in this gene have been implicated in the pathogenesis of autoimmune diseases^38,39. A recent study demonstrated that FCRL3 stimulation of regulatory T cells induced production of pro-inflammatory cytokines, including IL-17 and IL-26³⁸. This finding suggests that FCRL3 may play a critical role in mediating the transition of regulatory T cells to a pro-inflammatory phenotype and could potentially contribute to the increased inflammation observed among TET2 CHIP carriers^40,41. Additionally, Clark et al. identified FCRL3 as a gene for which DNA methylation at the CpG site cg17134153 in CD4⁺ T cells likely mediates the genetic risk for rheumatoid arthritis⁴². Given that CHIP, including TET2 CHIP, has been associated with rheumatoid arthritis (RA)⁴³, the regulation of FCRL3 expression by methylation changes at cg17134153 may, in part, serve as the functional basis of the observed association between CHIP and RA. Further experimental studies are warranted to better understand how differential expression of FCRL3 may impact TET2 CHIP development and the pathogenesis of RA. Taken together, our results provide insight into the molecular mechanisms underlying age-related diseases, namely cardiovascular disease.

Ethics

All participants provided written, informed consent. The study protocol was approved by the following institutional review boards at each collaborating institution: Institutional Review Board at Boston Medical Center (FHS); University of Washington Institutional Review Board (CHS); University of Mississippi Medical Center Institutional Review Board (ARIC: Jackson Field Center); Wake Forest University Health Sciences Institutional Review Board (ARIC: Forsyth County Field Center); University of Minnesota Institutional Review Board (ARIC: Minnesota Field Center); Johns Hopkins University School of Public Health Institutional Review Board (ARIC: Washington County Field Center); University of Mississippi Medical Center (JHS); Jackson State University (JHS); and Tougaloo College (JHS). All research was performed in accordance with relevant ethical guidelines and regulations. The study design and conduct adhered to all relevant regulations regarding the use of human study participants and was conducted in accordance to the criteria set by the Declaration of Helsinki.

Study cohorts

The Framingham Heart Study (FHS) is a prospective, observational community-based cohort investigating risk factors for CVD. For our discovery sample, DNAm was measured from FHS participants (N = 3295) in the Offspring cohort (N = 1860; Exam 8; years 2005-2008)³⁷ and in the Third Generation cohort (N = 1435; Exam 2; years 2008-2011)⁴⁴. CHIP calls were based on whole-genome sequencing of whole blood DNA samples, the majority of which were from FHS Offspring participants at Exam 8 and Gen 3 participants at Exam 2 and temporally concordant with the time of DNAm profiling. All FHS participants self-identified as White at the time of recruitment.

The Jackson Heart Study (JHS) is an observational community-based cohort studying the environmental and genetic factors associated with CVD in African Americans. For our discovery sample, data were collected from 1664 JHS participants¹². DNAm was measured from the majority of JHS participants at visit 1, with a small subset at visit 2. CHIP calls were concurrent with DNAm profiling and based on whole-genome sequencing of whole blood DNA samples, where the majority were from visit 1 (years 2000–2004) and a subset from visit 2 (years 2005-2008)¹². All JHS participants self-identified as Black or African American at the time of recruitment. No ancestry outliers were excluded, as inferred based on genetic similarity to reference panels. Similarity to the 1000 G AFR reference panel varied by individual (study q1, median, q3 77.9% 84.3% 89.0%) in the methylation and WGS overlap dataset, using estimates from RFMix.

The Cardiovascular Health Study (CHS) is a population-based cohort study of risk factors for CVD in adults aged 65 or older⁴⁵. DNAm was measured from blood samples from participants in years 5 and 9, year 5, or year 9 only. CHIP calls were based on whole-genome sequencing of blood samples, where the majority were taken 3 years before or concurrently with the first DNAm measurement¹². CHS participants self-reported their race at the time of recruitment.

The Atherosclerosis Risk in Communities (ARIC) is a prospective, multiracial cohort study of risk factor and clinical outcomes of atherosclerosis³⁸. DNAm was measured from 2655 ARIC participants at visit 2 (1990-1992) or visit 3 (1993-1995). CHIP calls were based on whole exome sequencing of blood samples from visit 2 and visit 3^12,39. ARIC participants self-identified their race at the time of recruitment. There is a subset of participants included in both ARIC and JHS. These overlapping participants were not excluded.

DNA methylation profiling

All the DNA samples were from whole blood. The four cohorts including FHS, JHS, CHS and ARIC, conducted independent laboratory DNAm measurements, quality control (including sample-wise and probe-wide filtering and probe intensity background correction; see Supplementary Information File). DNA methylation was measured in FHS, CHS, and ARIC participants using Illumina Infinium Human Methylation-450 Beadchip (450 K array) and in JHS participants using the Ilumina EPIC array^40,41.

CHIP calling

For the purposes of this investigation, CHIP was defined as a candidate driver gene mutation in genes that have been reported to be associated with hematologic malignancy, is present at a variant allele frequency (VAF) of at least 2% in peripheral blood, and is present in the absence of hematologic malignancy⁴². CHIP was detected in FHS, JHS, and CHS from WGS blood DNA in the NHLBI Trans-Omics for Precision Medicine (TOPMed) consortium using the Mutect2 software⁵. In ARIC, CHIP calls were based on whole exome sequencing of blood DNA using the same procedure⁵. CHIP is defined as when an individual harbors at least one pre-specified deleterious insertion/deletion or single nucleotide variant in any of the 74 genes linked to myeloid malignancy at a variant allele frequency (VAF) ≥ 2%⁵. TOPMed WGS samples were sequenced to a median depth of 40x, with the sequencing depth ranging from 30x-50x for a specific region. At this sequencing depth, CHIP can be reliably ascertained with a VAF > 10% but CHIP variants with a VAF ≤ 10% are unable to be robustly captured⁵. For a sensitivity analysis, race-stratified and multiracial meta-EWAS of any CHIP was performed using a more restrictive CHIP clone size of VAF > 10% (See Supplementary Fig. 4 and Supplementary Data 11–13).

Cohort-specific EWAS

The correction of methylation data for technical covariates was cohort specific. Each cohort performed an independent investigation to select an optimized set of technical covariates (e.g., batch, plate, chip, row, and column), using measured or imputed blood cell type fractions, surrogate variables, and/or principal components. Most cohorts had previous publications using the same dataset for EWAS of different traits, such as EWAS of alcohol drinking and smoking. In this study, those cohorts used the same strategies as they did previously for correcting for technical variables, including batch effects. Linear mixed models were used to test the associations between CHIP status as the predictor variable and DNAm β values as the outcome variable. Information about cohort-specific models is available in the Supplementary Information File.

Meta-analysis

All analyses were contingent on self-reported Black or White race. Previous ancestry inference in these cohort studies⁴³ suggests high genetic similarity of nearly all self-identified White participants to EUR reference panels (including 1000 Genomes). Self-identified Black participants have high but variable (average ~80% but may vary based on study and by study participant) genetic similarity to AFR reference panels and have some similarity to EUR reference panels as well. In some cases, extreme ancestry outliers may have been removed during study-specific QC. However, this has not been thoroughly documented in the data we received from participating studies. Importantly, we do not mean to imply that socially constructed racial identities reported by study participants are synonymous with genetic ancestry. Stratification by race may, however, capture differential social and environmental exposures within the US, which may impact the epigenome.

The meta-analysis was performed for any CHIP, DNMT3A, and TET2 in White participants from FHS, CHS, and ARIC (n = 4355) and Black participants from JHS, CHS, and ARIC (n = 3841) participants, respectively, using inverse variance-weighted fixed-effects models implemented in metagen() function in R packages (https://rdrr.io/cran/meta/man/metagen.html). The summary statistics were used from the previous EWAS of CHIP for the ARIC and CHS cohorts¹². Then, multiracial meta-analysis was performed for White and Black participants (n = 8,196). The meta-analysis was constrained to methylation probes passing filtering criteria in all cohorts.

Supplementary Fig. 1 presents QQ plots with genomic control (GC) inflation factor (λ) to illustrate the EWAS results in each cohort and in the meta-analysis. Our observations reveal a prevalence of high inflation factors (λ > 1.1) across nearly all studies. Such elevated inflation factors typically signal potential bias in the analysis process. However, it’s important to note that in cases where a significant portion of CpG sites exhibit differential methylation associated with the outcome (e.g., age and CHIP), this can contribute to the observed high λ values. Moreover, adjusting for additional PCs moderately associated with the outcome may alleviate lambda values, albeit at the expense of reduced power to detect CpGs related to the outcome. To address this, we adopted strategies consistent with those employed by the respective cohorts in previous analyses, focusing on correcting for technical variables and latent factors identified in prior studies across multiple outcomes^46,47,48. Furthermore, prior to meta-analysis, we implemented additional corrections for individual study results exhibiting λ > 1.5, ensuring the integrity of our findings. The statistical significance threshold was P < 0.05/400,000 ≈ 1 × 10⁻⁷. A less stringent threshold, the Benjamini-corrected FDR adjusted p-value < 0.05, was also used.

Expression quantitative trait methylation analysis

Association tests of DNAm and gene expression were previously performed in 2115 FHS participants in the Offspring (n = 686) and Third Generation (n = 1429) cohorts with available whole blood DNA methylation and RNA-seq gene expression data to identify CpG sites at which differential methylation is associated with gene expression⁴⁹. Approximately 70,000 significant cis CpG-transcript pairs were identified at P < 1 × 10^-7. Cis is defined as CpGs located within 100 kB of the transcription start site of a mRNA. When calculating the association between CpG sites and gene-level transcripts, linear regression models were used. Residualized gene expression served as the outcome and residualized DNA methylation β value as the primary explanatory variable, with adjustment for age, sex, white blood cell count, blood cell fraction, platelet count, five gene expression PCs, and ten DNA methylation PCs. Through integration of CpGs and gene-level transcripts (mRNAs) from RNA-seq, mRNAs were identified that were significantly associated with each of the CpGs in cis for any CHIP and the CHIP subtypes^49,50.

Pathway enrichment analysis

Enrichment analysis for CHIP EWAS signatures with a significance threshold of P < 1 × 10^-7 was conducted on gene sets comprising genes annotated to CpGs associated with CHIP and major CHIP subtypes using missMethyl R package. This package adjusts for known DNAm array bias⁵¹. For the enrichment analysis for eQTM gene sets, the DAVID Bioinformatics online tool was used (https://david.ncifcrf.gov/home.jsp). To improve the focus of this study, only the results of Gene Ontology (GO) terms related to biological process and KEGG pathways were used. Over-representation enrichment tests, specifically one-sided Fisher’s exact tests, were used to assess whether a GO/KEGG term is significantly enriched compared to the background. The significant threshold of FDR adjusted p-value < 0.05 was used, corrected by multiple tested terms⁵.

Cell culture of mPB CD34+ cells

Patients were given G-CSF ≤ 10 mcg/kg/day for up to 5 days. Peripheral blood mononuclear cells were collected and CD34+ cells were isolated using a MACs sorter. Samples were then counted and frozen down for future use. This research is funded from NIDDK. Mobilized peripheral blood (mPB) CD34+ cells were bought from StemCell technologies or the Cooperative Center of Excellence in Hematology (CCEH) at the Fred Hutch Cancer Research Center, Seattle, USA. The name and source of all cell lines used are the following: mPB-001: Sex: Female, Supplier Fred Hutchinson; mPB-002: Sex: Male, Supplier Fred Hutchinson; mPB-003: Sex: Female, Supplier StemCell Technologies; mPB-004: Sex: Male, Supplier: StemCell Technologies; mPB-005: Sex: Male, Supplier: StemCell Technologies. CD34+ cells were thawed and cultured in CD34+ expansion medium (StemSpan II (StemCell Technologies) + 10% CD34+ expansion supplement (Stemcell Technologies) + 20 U/mL penicillin-streptomycin (Gibco) + 500 nM UM729 (StemCell Technologies) + 750 nM Stemreginin-1 (StemCell Technologies)) for 48 h prior to editing with CRISPR-Cas9. After 48 h, samples were electroporated with RNP complexes and seeded at 400k cells per mL. Cells were maintained between 200k – 1M cells per mL.

CRISPR-Cas9 of mPB CD34+ cells

Ribonucleoprotein (RNP) complexes targeting scramble, AAVS, TET2, ASXL-1, and DNMT3A were made by incubating Cas9 (IDT Alt-R HiFi sp Cas9 Nuclease V3) and sgRNA (IDT Alt-R Cas9 sgRNAs) at a 1:3.26 ratio. Guides for each gene are present in Supplementary Table 1. On day 2 post thaw, mPB CD34 cells were counted and resuspended in Buffer R or GE Buffer. RNP complexes and cells were mixed and electroporated using Neon Pipette (Thermo Scientific Inc.) with the following settings: 1650 V 10 ms pulses 3 times. Samples were seeded in expansion media at 400k/mL.

Assessment of indel formation

Genomic DNA (gDNA) was isolated and amplified with the following conditions: 95 °C for 2 min followed by 35 cycles of 95 °C for 45 s, 61-62 °C for 1 min, 72 °C for 2 min with a final extension at 72 °C for 5 min using primers towards TET2, ASXL-1, and DNMT3A (Supplementary Table 2). PCR products were sent to GeneWiz (Azenta Life Sciences) where PCR cleanup and Sanger sequencing was performed. Indel formation was assessed using TIDE (Supplementary Table 3)⁵².

FACS sorting of mPB CD34+ cells

Edited CD34+ cells were sorted at day 7 post CRISPR-Cas9 using a FACSymphony™ S6 Cell Sorter or a BD FACS Aria II to remove differentiated cells. Briefly, CD34+ cells were washed in cell staining buffer (Biolegend) once and stained with antibodies targeting CD34 (Biolegend: 343614; dilution: 1:50), CD38 (Biolegend: 303532; dilution: 1:100), and Lineage Markers (Biolegend: 348805; dilution: 1:10) for 30 min at 4 °C in the dark (Supplementary Fig. 6). The antibodies are present in Supplementary Table 4.

Duet evoC library generation and primary methylation analysis

DNA was extracted using Micro kits (Qiagen) from flow sorted cells from 3–5 donors. EvoC libraries were created following manufacturer instructions (Biomodal). Briefly, DNA was sheared using a Covaris LE220 and assessment of input DNA was performed using Bioanalayzer instrument (Agilent) and Qbit (ThermoFisher). Library generation was performed according to the duet evoC library generation protocol (biomodal).

Sequencing of duet evoC libraries

Capture of CpG sites was performed using Twist Human Methylome Panel (Twist Biosciences) and next generation sequencing was completed by using the NovaSeq 6000 (150 bp paired-end reads) targeting 160 M reads per sample. Biomodal pipeline version 1.1.1 was used to analyze the raw FASTQs with default settings. Briefly adapter trimming was performed with cutadapt, resolution of R1 and R2 to generate single-end reads with epigenetic information, mapping onto the human genome (GRCh38), and quantification of the modification state of each CpG site.

Comparisons between EWAS and biomodal data

For each sample and for each CpG, read counts from the forward and reverse strand were summed and the mC fraction calculated as the number of reads supporting mC divided by the total number of reads with modified or unmodified C (excluding reads with A, T or G). The dataset was reduced to the CpGs with significant levels of association from each EWAS analysis. For each of these CpGs, methylation difference was calculated as the difference between the average mC fraction of multiple replicates of different KO primary cells (“DNMT3A”, “TET2”, “ASXL-1”) and the average mC fraction of multiple replicates of control cells (“Scramble” or “AAVS”). Only CpGs with uncorrected p-values < 0.05 (t-test) were carried forward. For each EWAS analysis (“any-CHIP”, “DNMT3A_chip”, “TET2_chip”, “ASXL1_chip”) and for each gene-KO (“DNMT3A”, “TET2”, “ASXL-1”), the mC fraction of these CpGs was plotted against the EWAS TE, and a binomial test was used to check for enrichment in the top-right and bottom-left quadrant indicating a sign correlation between the mC fraction change induced by the KO and the EWAS TE.

Cis-mQTLs

Methylation quantitative trait loci (mQTLs) – SNPs associated with DNA methylation – were identified from 4,170 FHS participants as previously reported⁴⁰, including 4.7 million cis-mQTLs at P < 2 × 10⁻¹¹. Genotypes were imputed using the 1000 Genomes Project panel phase 3 using MACH / Minimac software. SNPs with MAF > 0.01 and imputation quality ratio >0.3 were retained. Cis-mQTLs were defined as SNPs residing within 1 Mb upstream or downstream of a CpG site.

Association of methylation with complex diseases and traits

To annotate CHIP-associated CpGs and cis-mQTLs, we utilized both the EWAS Catalog (https://www.ewascatalog.org/)²² and the GWAS Catalog (https://www.ebi.ac.uk/gwas/)²⁰. The EWAS Catalog collected published CpG signatures for about 4000 traits and/or diseases. GWAS Catalog collected significant SNPs associated with thousands of traits and/or diseases. For the identified CHIP-associated CpGs, we matched these CpGs with reported trait-associated CpGs in the EWAS Catalog. To evaluate the enrichment of CHIP-associated CpGs for traits listed in the EWAS Catalog, we performed one-sided Fisher’s exact tests. We applied a Bonferroni-corrected significance threshold of P = 1.24E-05 (0.05/4023, accounting for 4023 traits in the EWAS Catalog). Additionally, to assess whether any cis-mQTLs of CHIP-associated CpGs demonstrated strong associations with human complex traits, we matched the cis-mQTLs against SNPs in the GWAS Catalog that were reported with P < 5E-8.

Mendelian randomization analysis

In order to investigate whether differentiation methylation at CHIP-associated CpGs causally influences risk of CVD and mortality, two-sample Mendelian randomization (MR) was performed between exposures (CHIP-associated CpGs) and a list of CVD- and mortality- related traits as outcomes. We utilized our in-house developed analytical pipeline called MR-Seek (https://github.com/OpenOmics/mr-seek.git) to perform the analysis. The full summary statistics of different GWAS datasets were downloaded from NHGRI-EBI. The list of CVD- and mortality- related traits and list of references of those GWAS results are included in Supplementary Data 28. Previously identified cis-methylation quantitative trait loci (cis-mQTL) were utilized as instrumental variables (IVs) in the MR analysis²⁰. The cis-mQTLs that were also identified as trans-mQTLs were excluded to avoid potential pleiotropic effects. For each CpG site, the IVs comprised independent cis-mQTLs pruned for linkage disequilibrium (LD) with an r² < 0.01. Only one cis-mQTL variant with the lowest SNP-CpG p-value was retained in each LD block. For CpGs with more than one IV, Inverse-variance weighted (IVW) MR tests were conducted. Heterogeneity and MR-EGGER pleiotropy tests were employed to assess the validity of IVs. Results with a significance level of P < 0.05 were excluded. If a CpG has only one instrumental variable (IV), the Wald MR method was applied. Significance levels of MR results were determined based on the Benjamini-Hochberg corrected FDR adjusted p-value with a threshold of <0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

The whole genome sequencing (WGS), DNA methylation, RNA sequencing data, and phenotypic data from the Framingham Heart Study (FHS), the Jackson Heart Study (JHS), the Cardiovascular Health Study (CHS), and the Atherosclerosis Risk in Communities (ARIC) study have been deposited in the dbGaP database under accession codes, phs000007 (FHS) [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000007.v34.p15], phs000964 (JHS) [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000964.v5.p1], phs001368 (CHS) [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs001368.v4.p2], phs000668 (ARIC) [https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000668.v6.p2]. The WGS, DNA methylation, and phenotypic data from all cohorts used in this study are available under restricted access to protect participant privacy and ensure confidentiality. Access can be obtained by submitting an ancillary study proposal and obtaining IRB approval. Timelines for the approval process range from 4–9 weeks for CHS and 3–6 weeks for ARIC ancillary studies, with specific criteria and proposal forms for the respective studies available at https://chs-nhlbi.org/node/6222 and https://sites.cscc.unc.edu/aric/ancillary-studies-pfg. Full summary statistics for several figures are available in two Zenodo repositories: Fig. 2, Supplementary Figs. 1–3, and Fig. 1 from Responses to Reviewers’ Comments File [https://doi.org/10.5281/zenodo.14712757]⁵³ and Supplementary Fig. 4 [https://doi.org/10.5281/zenodo.14712522]⁵⁴. Source data are provided with this paper.

1. Warren, J. T. & Link, D. C. Clonal hematopoiesis and risk for hematologic malignancy. Blood 136, 1599–1605 (2020).

   PubMed  PubMed Central  Google Scholar 

2. Jaiswal, S. et al. Age-related clonal hematopoiesis associated with adverse outcomes. N. Engl. J. Med. 371, 2488–2498 (2014).

   Article  PubMed  PubMed Central  Google Scholar 

3. Jaiswal, S. & Ebert, B. L. Clonal hematopoiesis in human aging and disease. Science 366, eaan4673 (2019).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

4. Miller, P. G. et al. Association of clonal hematopoiesis with chronic obstructive pulmonary disease. Blood 139, 357–368 (2022).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Bick, A. G. et al. Inherited causes of clonal haematopoiesis in 97,691 whole genomes. Nature 586, 763–768 (2020).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

6. Genovese, G. et al. Clonal hematopoiesis and blood-cancer risk inferred from blood DNA sequence. N. Engl. J. Med. 371, 2477–2487 (2014).

   Article  PubMed  PubMed Central  Google Scholar 

7. Xie, M. et al. Age-related mutations associated with clonal hematopoietic expansion and malignancies. Nat. Med 20, 1472–1478 (2014).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

8. Challen, G. A. et al. Dnmt3a is essential for hematopoietic stem cell differentiation. Nat. Genet 44, 23–31 (2011).

   Article  PubMed  PubMed Central  Google Scholar 

9. Fuster, J. J. et al. Clonal hematopoiesis associated with TET2 deficiency accelerates atherosclerosis development in mice. Science 355, 842–847 (2017).

   Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

10. Inoue, D. et al. A novel ASXL1-OGT axis plays roles in H3K4 methylation and tumor suppression in myeloid malignancies. Leukemia 32, 1327–1337 (2018).

    Article  CAS  PubMed  Google Scholar 

11. Fujino, T. & Kitamura, T. ASXL1 mutation in clonal hematopoiesis. Exp. Hematol. 83, 74–84 (2020).

    Article  CAS  PubMed  Google Scholar 

12. Uddin, M. D. M. et al. Clonal hematopoiesis of indeterminate potential, DNA methylation, and risk for coronary artery disease. Nat. Commun. 13, 5350 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

13. Zhang, X. et al. DNMT3A and TET2 compete and cooperate to repress lineage-specific transcription factors in hematopoietic stem cells. Nat. Genet 48, 1014–1023 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

14. Spencer, D. H. et al. CpG island hypermethylation mediated by DNMT3A is a consequence of AML progression. Cell 168, 801–816 e813 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

15. Li, Y. et al. Rapid and accurate remethylation of DNA in Dnmt3a-deficient hematopoietic cells with restoration of DNMT3A activity. Sci. Adv. 10, eadk8598 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

16. Bernabeu, E. et al. Refining epigenetic prediction of chronological and biological age. Genome Med. 15, 12 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

17. Hendel, A. et al. Chemically modified guide RNAs enhance CRISPR-Cas genome editing in human primary cells. Nat. Biotechnol. 33, 985–989 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

18. Fullgrabe, J. et al. Simultaneous sequencing of genetic and epigenetic bases in DNA. Nat. Biotechnol. 41, 1457–1464 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

19. Buniello, A. et al. The NHGRI-EBI GWAS catalog of published genome-wide association studies, targeted arrays and summary statistics 2019. Nucleic Acids Res. 47, D1005–D1012 (2019).

    Article  CAS  PubMed  Google Scholar 

20. Ma, J. et al. Elucidating the genetic architecture of DNA methylation to identify promising molecular mechanisms of disease. Sci. Rep. 12, 19564 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

21. Battram, T. et al. The EWAS catalog: a database of epigenome-wide association studies. Wellcome Open Res. 7, 41 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

22. Dawoud, A. A. Z., Tapper, W. J. & Cross, N. C. P. Clonal myelopoiesis in the UK biobank cohort: ASXL1 mutations are strongly associated with smoking. Leukemia 34, 2660–2672 (2020).

    Article  CAS  PubMed  Google Scholar 

23. Bolton, K. L. et al. Cancer therapy shapes the fitness landscape of clonal hematopoiesis. Nat. Genet 52, 1219–1226 (2020).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

24. Nagase, R. et al. Expression of mutant Asxl1 perturbs hematopoiesis and promotes susceptibility to leukemic transformation. J. Exp. Med. 215, 1729–1747 (2018).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

25. Kohnke, T. et al. Human ASXL1-Mutant hematopoiesis is driven by a truncated protein associated with aberrant deubiquitination of H2AK119. Blood Cancer Discov. 5, 202–223 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

26. Li, Y. L., Tian, H., Jiang, J., Zhang, Y. & Qi, X. W. Multifaceted regulation and functions of fatty acid desaturase 2 in human cancers. Am. J. Cancer Res. 10, 4098–4111 (2020).

    CAS  PubMed  PubMed Central  Google Scholar 

27. Ma, J. et al. Whole blood DNA methylation signatures of diet are associated with cardiovascular disease risk factors and all-cause mortality. Circ. Genom. Precis Med. 13, e002766 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

28. Gomez-Alonso, M. D. C. et al. DNA methylation and lipid metabolism: an EWAS of 226 metabolic measures. Clin. Epigenetics 13, 7 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

29. Liu, X. et al. CHIP-associated mutant ASXL1 in blood cells promotes solid tumor progression. Cancer Sci. 113, 1182–1194 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

30. Stacey, S. N. et al. Genetics and epidemiology of mutational barcode-defined clonal hematopoiesis. Nat. Genet 55, 2149–2159 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

31. Diez-Diez, M. et al. Unidirectional association of clonal hematopoiesis with atherosclerosis development. Nat. Med. 30, 2857–2866 (2024).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

32. Jakubek, Y. A. et al. Mosaic chromosomal alterations in blood across ancestries using whole-genome sequencing. Nat. Genet 55, 1912–1919 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

33. Vlasschaert, C., Heimlich, J. B., Rauh, M. J., Natarajan, P. & Bick, A. G. Interleukin-6 receptor polymorphism attenuates clonal hematopoiesis-mediated coronary artery disease risk among 451 180 individuals in the UK biobank. Circulation 147, 358–360 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

34. Bernstein, N. et al. Analysis of somatic mutations in whole blood from 200,618 individuals identifies pervasive positive selection and novel drivers of clonal hematopoiesis. Nature Genetics 56, 1147–1155 (2024).

35. Richmond, R. C. & Davey Smith, G. Mendelian randomization: concepts and scope. Cold Spring Harb. Perspect. Med 12, a040501 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

36. Reed, Z. E. et al. Exploring pleiotropy in Mendelian randomisation analyses: what are genetic variants associated with ‘cigarette smoking initiation’ really capturing? Genet Epidemiol. 49, e22583 (2024).

37. Kannel, W. B., Feinleib, M., McNamara, P. M., Garrison, R. J. & Castelli, W. P. An investigation of coronary heart disease in families. The Framingham offspring study. Am. J. Epidemiol. 110, 281–290 (1979).

    Article  CAS  PubMed  Google Scholar 

38. Wright, J. D. et al. The ARIC (Atherosclerosis risk in communities) study: JACC focus seminar 3/8. J. Am. Coll. Cardiol. 77, 2939–2959 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

39. Van Alsten, S. C. et al. Association between ABO and Duffy blood types and circulating chemokines and cytokines. Genes Immun. 22, 161–171 (2021).

    Article  PubMed  PubMed Central  Google Scholar 

40. Huan, T. et al. Genome-wide identification of DNA methylation QTLs in whole blood highlights pathways for cardiovascular disease. Nat. Commun. 10, 4267 (2019).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

41. Joehanes, R. et al. Epigenetic signatures of cigarette smoking. Circ. Cardiovasc. Genet 9, 436–447 (2016).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

42. Libb, P. et al. Clonal hematopoiesis: crossroads of aging, cardiovascular disease, and cancer: JACC review topic of the week. J. Am. Coll. Cardiol. 74, 567–577 (2019).

43. Taliun, D. et al. Sequencing of 53,831 diverse genomes from the NHLBI TOPMed program. Nature 590, 290–299 (2021).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

44. Tsao, C. W. & Vasan, R. S. Cohort profile: the framingham heart study (FHS): overview of milestones in cardiovascular epidemiology. Int J. Epidemiol. 44, 1800–1813 (2015).

    Article  PubMed  PubMed Central  Google Scholar 

45. Fried, L. P. et al. The cardiovascular health study: design and rationale. Ann. Epidemiol. 1, 263–276 (1991).

    Article  CAS  PubMed  Google Scholar 

46. Liu, C. et al. A DNA methylation biomarker of alcohol consumption. Mol. Psychiatry 23, 422–433 (2018).

    Article  CAS  PubMed  Google Scholar 

47. Huan, T. et al. Integrative analysis of clinical and epigenetic biomarkers of mortality. Aging Cell 21, e13608 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

48. Demerath, E. W. et al. Epigenome-wide association study (EWAS) of BMI, BMI change and waist circumference in African American adults identifies multiple replicated loci. Hum. Mol. Genet 24, 4464–4479 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

49. Keshawarz, A. et al. Expression quantitative trait methylation analysis elucidates gene regulatory effects of DNA methylation: the Framingham heart study. Sci. Rep. 13, 12952 (2023).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

50. Liu, C. et al. Whole genome DNA and RNA sequencing of whole blood elucidates the genetic architecture of gene expression underlying a wide range of diseases. Sci. Rep. 12, 20167 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

51. Phipson, B., Maksimovic, J. & Oshlack, A. missMethyl: an R package for analyzing data from illumina’s HumanMethylation450 platform. Bioinformatics 32, 286–288 (2016).

    Article  CAS  PubMed  Google Scholar 

52. Brinkman, E. K., Chen, T., Amendola, M. & van Steensel, B. Easy quantitative assessment of genome editing by sequence trace decomposition. Nucleic Acids Res. 42, e168 (2014).

    Article  PubMed  PubMed Central  Google Scholar 

53. Huan, T. Epigenome-wide DNA methylation association study of CHIP provides insight into perturbed gene regulation. Zenodo https://doi.org/10.5281/zenodo.14712757 (2025).

54. Huan, T. Epigenome-wide DNA methylation association study of CHIP provides insight into perturbed gene regulation. Zenodo https://doi.org/10.5281/zenodo.14712522 (2025).

Download references

Author notes

1. These authors contributed equally: Sara Kirmani, Tianxiao Huan.

2. These authors jointly supervised this work: Alexander G. Bick, Daniel Levy.

Authors and Affiliations

1. Framingham Heart Study, Framingham, MA, 01702, USA

   Sara Kirmani, Tianxiao Huan, Roby Joehanes, Nancy L. Heard-Costa, Joanne Murabito & Daniel Levy

2. Population Sciences Branch, Division of Intramural Research, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD, 20892, USA

   Sara Kirmani, Tianxiao Huan, Roby Joehanes & Daniel Levy

3. Division of Genetic Medicine, Department of Medicine, Vanderbilt University Medical Center, Nashville, TN, 37232, USA

   Joseph C. Van Amburg, David A. Ong & Alexander G. Bick

4. Medical and Population Genetics and Cardiovascular Disease Initiative, Broad Institute of Harvard and MIT, Cambridge, MA, 02142, USA

   Md Mesbah Uddin & Pradeep Natarajan

5. Cardiovascular Research Center, Massachusetts General Hospital, Boston, MA, 02114, USA

   Md Mesbah Uddin

6. Department of Epidemiology, Human Genetics, and Environmental Sciences, School of Public Health, The University of Texas Health Science Center at Houston, Houston, TX, 77030, USA

   Ngoc Quynh H. Nguyen, Bing Yu & Jan Bressler

7. Cardiovascular Health Research Unit, Department of Medicine, University of Washington, Seattle, WA, 98101, USA

   Jennifer A. Brody, Nona Sotoodehnia, James S. Floyd & Bruce M. Psaty

8. Brown Foundation Institute of Molecular Medicine, McGovern Medical School, University of Texas Health Science Center at Houston, Houston, TX, 77030, USA

   Myriam Fornage

9. Human Genetics Center, School of Public Health, University of Texas Health Science Center at Houston, Houston, TX, 77030, USA

   Myriam Fornage & Jan Bressler

10. Biomodal, The Trinity Building, Chesterford Research Park, Cambridge, CB10 1XL, UK

    Fabio Puddu

11. Department of Epidemiology, University of Washington, Seattle, WA, 98101, USA

    James S. Floyd & Bruce M. Psaty

12. Department of Medicine, Baylor College of Medicine, Houston, TX, 77030, USA

    Christie M. Ballantyne

13. Department of Health Systems and Population Health, University of Washington, Seattle, WA, 98101, USA

    Bruce M. Psaty

14. Department of Genetics, University of North Carolina, Chapel Hill, NC, 27599, USA

    Laura M. Raffield

15. Cardiovascular Research Center and Center for Genomic Medicine, Massachusetts General Hospital, Boston, MA, 02114, USA

    Pradeep Natarajan

16. Department of Medicine, Harvard Medical School, Boston, MA, 02115, USA

    Pradeep Natarajan

17. Department of Human Genetics, Emory University School of Medicine, Atlanta, GA, 30322, USA

    Karen N. Conneely & Joshua S. Weinstock

18. Department of Medicine, University of Mississippi Medical Center, Jackson, MS, 39216, USA

    April P. Carson

19. Department of Medicine, University of Colorado at Denver, Aurora, CO, 80045, USA

    Leslie A. Lange & Kendra Ferrier

20. Department of Neurology, Boston University School of Medicine, Boston, MA, 02118, USA

    Nancy L. Heard-Costa

21. Department of Medicine, Section of General Internal Medicine, Boston University School of Medicine and Boston Medical Center, Boston, MA, 02118, USA

    Joanne Murabito

Contributions

S.K., T.H., and D.L. conceived of and designed the study, with input from K.C. and P.N. T.H., S.K., R.J., K.F., L.L., and L.R conducted the analyses in the FHS and JHS cohorts. J.A. and D.O. performed the CHIP model validation experiments. F.P. performed the bioinformatic analyses for the CHIP model validation experiments. J.A.B., J.B, N.S., M.F., J.F., B.P., and C.B. collected and pre-processed previously published data in the CHS or ARIC cohorts that was analyzed in this study. P.N., A.B., and J.W. performed the calling of CHIP in all samples. B.Y. and N.N. conducted analyses in the ARIC cohort. M.U. led a previous study of which the published datasets were used in the current study. S.K and T.H. drafted the manuscript. D.L., L.R., K.F., and A.B. substantively revised the work. A.C. supervised analyses in the JHS cohort. N.C. and J.M. contributed to revisions of the manuscript. These authors contributed equally: S.K., T.H. These authors jointly supervised this work: A.B., D.L. All authors have read and approved the final version of the manuscript.

Corresponding authors

Correspondence to Tianxiao Huan, Alexander G. Bick or Daniel Levy.

Competing interests

F.P. is an employee of Biomodal. L.M.R serves as a consultant for the NHLBI TOPMed Administrative Coordinating Center (through Westat). P.N. reports research grants from Allelica, Amgen, Apple, Boston Scientific, Genentech / Roche, and Novartis, personal fees from Allelica, Apple, AstraZeneca, Blackstone Life Sciences, Creative Education Concepts, CRISPR Therapeutics, Eli Lilly & Co, Foresite Labs, Genentech / Roche, GV, HeartFlow, Magnet Biomedicine, Merck, and Novartis, scientific advisory board membership of Esperion Therapeutics, Preciseli, TenSixteen Bio, and Tourmaline Bio, scientific co-founder of TenSixteen Bio, equity in MyOme, Preciseli, and TenSixteen Bio, and spousal employment at Vertex Pharmaceuticals, all unrelated to the present work. B.P. serves on the Steering Committee of the Yale Open Data Access Project funded by Johnson & Johnson. The remaining authors – S.K., T.H., J.A., R.J., M.U., N.N., B.Y., J.A.B., M.F., J.B., N.S., D.O., J.F., C.B., L.R., K.C., J.W., A.C., L.L., K.F., N.H., J.M., A.B., and D.L. – declare no competing interests.

Peer review information

Nature Communications thanks Tamir Chandra, who coreviewed with Samuel Crofts, Christopher Bell and the other, anonymous, reviewer for their contribution to the peer review of this work. A peer review file is available.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions