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)28 and in the Third Generation cohort (N=1435; Exam 2; years 2008-2011).29 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 participants.13 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).13 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 1000G 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 older.2 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 measurement.13 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 atherosclerosis.30 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 3.13, 31 ARIC participants self-identified their race at the time of recruitment.
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 additional file 1). DNA methylation was measured in FHS, CHS, and ARIC participants using Illumina Infinium Human Methylation-450 Beadchip (450K array) and in JHS participants using the Ilumina EPIC array as previously described32, 33.
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 malignancy.34 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 software as previously described.6 In ARIC, CHIP calls were based on whole exome sequencing of blood DNA using the same procedure.6 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%.6 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 captured.6 For a sensitivity analysis, ancestry-stratified and pooled ancestry meta-EWAS of any CHIP was performed using a more restrictive CHIP clone size of VAF >10% (See Supplementary Figure 4 and Supplementary Tables 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 Supplementary File 2.
Meta-analysis
All analyses were contingent on self-reported Black or White race. Previous ancestry inference in these cohort studies35 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 cohorts.13 Then, cross-ancestry 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 Figure 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 outcomes36-38. 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 (eQTM) 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 expression 19. Approximately 70,000 significant cis CpG-transcript pairs were identified at P <1 × 10-7. Cis is defined as CpGs located within 1 Mb 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 ꞵ 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.19, 39
Pathway Enrichment Analysis
Enrichment analysis was conducted on gene sets comprising genes annotated to CpGs associated with CHIP and major CHIP subtypes, with a significance threshold of P <1 × 10-7, along with their corresponding eQTM gene sets. The DAVID Bioinformatics online tool was used for the enrichment analysis (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. The significant threshold of FDR adjusted p-value <0.05 was used, corrected by multiple tested terms.6
Cell Culture of mPB CD34+ Cells
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. CD34+ cells were thawed and cultured in CD34+ expansion medium (StemSpan II (StemCell Technologies) + 10% CD34+ expansion supplement (Stemcell Technologies) + 20U/mL penicillin-streptomycin (Gibco) + 500nM UM729 (StemCell Technologies) + 750nM Stemreginin-1 (StemCell Technologies)) for 48 hours priorto editing withCRISPR-Cas9. After 48 hours, 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 Figure 7. 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: 1650V 10ms 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 minutes followed by 35 cycles of 95˚C for 45s, 61-62˚C for 1min, 72˚C for 2min with a final extension at 72˚C for 5 minutes using primers towards TET2, ASXL-1, and DNMT3A (Supplementary Figure 8). PCR products were sent to GeneWiz (Azenta Life Sciences) where PCR cleanup and Sanger sequencing was performed. Indel formation was assessed using TIDE (Supplementary Figure 9).40
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), CD38 (Biolegend), and Lineage Markers (Biolegend) for 30 minutes at 4˚C in the dark (Supplementary Figure 6). Antibodies from Biolegend are present in Supplementary Figure 10.
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 EvoC protocol (Biomodal).
Sequencing of 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 (150bp paired-end reads) targeting 160M reads per sample. Biomodal pipeline version 1.1.1 was used to analyze the raw FASTQs with default settings. Briefly adaptor 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 reported32, 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.
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 Table 28. Previously identified cis-methylation quantitative trait loci (cis-mQTL) were utilized as instrumental variables (IVs) in the MR analysis.21 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.