Multi-ancestry epigenome-wide analyses identify methylated sites associated with aortic augmentation index in TOPMed MESA

Despite the prognostic value of arterial stiffness (AS) and pulsatile hemodynamics (PH) for cardiovascular morbidity and mortality, epigenetic modifications that contribute to AS/PH remain unknown. To gain a better understanding of the link between epigenetics (DNA methylation) and AS/PH, we examined the relationship of eight measures of AS/PH with CpG sites and co-methylated regions using multi-ancestry participants from Trans-Omics for Precision Medicine (TOPMed) Multi-Ethnic Study of Atherosclerosis (MESA) with sample sizes ranging from 438 to 874. Epigenome-wide association analysis identified one genome-wide significant CpG (cg20711926-CYP1B1) associated with aortic augmentation index (AIx). Follow-up analyses, including gene set enrichment analysis, expression quantitative trait methylation analysis, and functional enrichment analysis on differentially methylated positions and regions, further prioritized three CpGs and their annotated genes (cg23800023-ETS1, cg08426368-TGFB3, and cg17350632-HLA-DPB1) for AIx. Among these, ETS1 and TGFB3 have been previously prioritized as candidate genes. Furthermore, both ETS1 and HLA-DPB1 have significant tissue correlations between Whole Blood and Aorta in GTEx, which suggests ETS1 and HLA-DPB1 could be potential biomarkers in understanding pathophysiology of AS/PH. Overall, our findings support the possible role of epigenetic regulation via DNA methylation of specific genes associated with AIx as well as identifying potential targets for regulation of AS/PH.


Introduction
Arterial stiffness (AS) serves as an independent predictor of cardiovascular diseases (CVD) morbidity and mortality 1,2 . Structurally, AS is characterized by degeneration of the media layer of arterial walls and involves decreased intact elastin and increased collagen bers 3 . Functionally, AS impairs the cushioning capacity of arteries that transform pulsatile ow at the ascending aorta into steady ow through the arterioles 3 . Pulsatile hemodynamics (PH) resulting from AS involves the augmentation of systolic blood pressure (BP) in late systole 4 which partially results from wave re ection arriving at the aorta during ejection, increasing left ventricular load 5 .
Both genome-wide association studies (GWAS) and candidate gene approaches have identi ed genetic variants that are associated with AS/PH, including those in the renin-angiotensin-aldosterone (RAA) system, elastic ber structural components, metalloproteinases, the nitric oxide (NO) pathway, and βadrenergic receptors 6-8 . However, a large group of identi ed AS/PH-associated genes are not related to known pathophysiological mechanisms of AS/PH, re ecting the likely multifactorial etiology of AS/PH 6 .
Since variation in AS/PH is mediated through gene expression, exploration of factors that are involved in transcriptional pathways may offer new insights to variation of AS/PH. DNA methylation (DNAm), one of the most studied epigenetic modi cations, plays a critical role in the transcriptional regulation 9,10 . DNAm occurs when a methyl group is added to a cytosine followed by guanine (CpG) in the genome 10 . Despite the emphasis on the role of AS/PH in the development of CVD, few human studies have examined DNAm in relation to AS/PH. In contrast, studies in aged mice have demonstrated that compound H, a potential activator of DNA demethylases, attenuates aging-related AS and hypertension 11 . In hyperhomocysteinemia mice, 5-aza-2'-deoxycytidine (Aza), a DNA methyltransferase (DNMT1) inhibitor, reduced high BP and vascular stiffening via reduced expression of matrix metalloproteinases 9 (MMP9), and tissue inhibitors of metalloproteinases (TIMPs) 12 . Using umbilical cord blood DNA from 470 participants, an epigenome-wide association study (EWAS) identi ed differentially methylated CpGs associated with increased BP and AS at an 8-9-year follow-up examination 13 . Although single CpG positions have been associated with many phenotypes [14][15][16] , various studies have reported that methylation levels are strongly correlated across the genome, and the reported functionally-relevant ndings are generally associated with genomic region rather than single CpG position [17][18][19] . Hence, understanding DNAm associations with AS/PH at both single CpG position and genomic region levels are important to gain a broader understanding of DNAm changes in AS/PH.
Our study aimed to identify differentially methylated CpG positions and regions that are associated with AS/PH traits. First, we examined differently methylated positions (DMPs) and differentially methylated regions (DMRs) that are associated with eight AS/PH traits, using both epigenome-wide and literaturebased candidate gene approaches 7 . Second, we conducted three follow-up analyses including gene set enrichment analysis, expression quantitative trait methylation analysis, and functional enrichment analysis for signi cantly associated DMPs and DMRs for each AS/PH trait. Third, we prioritized CpGs and their annotated genes for each AS/PH trait using the overlap of signi cant ndings across three follow-up analyses.

Participant characteristics
The demographic and clinical characteristics of study samples for AIx are summarized in Table 1 The overview of study design is shown in Fig. 1. After applying standard methylation quality control and ltering out positions with low methylation variation (Methods), there were 491,174 CpGs identi ed for epigenetic association analyses in MESA. In the multi-ancestry epigenome-wide association studies (EWAS) of AS/PH traits, we rst reported the Quantile-Quantile (QQ) plots with corresponding λ, a measure to quantify the in ation in the test statistics ( Supplementary Fig. 1). The genomic in ation λ ranged from 0.97 (PTC1) to 1.05 (AIx). We observed two FDR-signi cant (FDR<0.05) differentially methylated positions (DMPs), cg20711926 (CYP1B1) and cg25309493 (NGEF), associated with AIx, with cg20711926 (p-value=1.87x10 -9 ) also passing Bonferroni signi cance (0.05/491,174=1.02x10 -7 ) (Fig. 2).
We further examined the association of these two FDR-signi cant DMPs with AIx in race/ethnic strati ed analyses. The two DMPs were nominally signi cant in all ancestry groups except cg20711926 in AA group (Fig. 2 Signi cant associations between gene expression and suggestive DNAm differences To identify genes whose expression is associated with suggestive DNAm differences, we conducted expression quantitative trait methylation (eQTM) analysis for suggestive DMPs and DMRs with their annotated genes (Methods). We identi ed 60 FDR-signi cant eQTMs at DMPs among 343 expression-DMP association tests for AIx (Supplementary Table 33 Table 35).
Signi cant enrichment in Heart enhancers for AIx-associated DNAm differences We conducted functional enrichment analysis to better understand the complex interpretation of suggestive DNAm differences (Methods). Speci cally, we checked the enrichment results of two AS/PHrelevant tissues (Blood and Heart) for suggestive DMPs and DMRs. The analysis for AIx highlighted enrichment in both Blood and Heart enhancers from 15 chromatin states for CpGs included in AIx-associated DMPs and DMRs ( Supplementary Fig. 2). Furthermore, the signi cant enrichment in Heart enhancers was observed in two AS/PH-relevant cell-types: E104 Right Atrium (Q-value=8.6x10 -3 ) and E65 Aorta (Q-value=0.04) (Supplementary Table 36). For the other seven AS/PH traits, we did not observe any signi cant enrichment in the 15 chromatin states for either Blood or Heart tissues (Supplementary Fig. 3 -9).
Three genes prioritized for AIX After conducting follow-up analyses for CpGs included in suggestive DMPs and DMRs, we checked CpGs included in the overlap of signi cant ndings across three follow-up analyses (gene set enrichment analysis, expression quantitative trait methylation analysis, and functional enrichment analysis). The work ow of prioritization is shown in Fig. 3a. We rst summarized results from the three follow-up analyses for the two FDR-signi cant DMPs associated with AIx in Supplementary Table 37. No signi cant results were observed in any of the three follow-up analyses for these two DMPs. However, there were three other CpGs (cg23800023-ETS1, cg17350632-HLA-DPB1, and cg08426368-TGFB3) prioritized by follow-up analyses for AIx.
Among the prioritized genes, both ETS1 and TGFB3 are on our candidate gene list, although the candidate gene list itself was not used in our overall approach for prioritization of genes. All three genes were involved in KEGG pathway hsa05166: Human T-cell leukemia virus 1 infection pathway and both ETS1 and TGFB3 were in KEGG pathway hsa05200: Pathways in cancer (Fig. 3b). Additionally, all three CpGs were FDR-signi cant eQTMs and both CpGs cg23800023 and cg17350632 were negatively associated with their annotated genes, ETS1 and HLA-DPB1, respectively (Fig. 3c). Furthermore, the CpG cg23800023 was enhancer-enriched in both AS/PH-relevant cell-types, Right Atrium (enrichment Q-value=8.57x10 -3 ) and Aorta (enrichment Q-value=0.0434); the other two CpGs were enhancer-enriched in Right Atrium Fig. 2 and Fig. 3c). Finally, we checked the correlation of gene expression levels between GTEx Whole-Blood and Aorta tissues respectively for these three genes (Methods). Both ETS1 and HLA-DPB1 have signi cant tissue correlation (ETS1, Pearson's correlation R=0.33 with p-value=1.50x10 -10 ; HLA-DPB1, Pearson's correlation R=0.24 with p-value=3.20x10 -6 ; Fig. 3d). However, TGFB3 was not signi cantly correlated in those two tissues in GTEx. We also checked the overlap of signi cant ndings across the three follow-up analyses for the other seven AS/PH traits, but no overlap was observed.

Discussion
Arterial stiffness (AS) is a subclinical condition that has a signi cant prognostic value for future development of CVD events and end organ damage 20 . AS results in excess pulsatile hemodynamics (PH), which can be measured by aortic augmentation index 21 (AIx). While CVD is often identi ed late requiring interventions to manage disease progress, targeting aortic AS and excessive PH present an early opportunity to identify individuals at greater risk as well as means to monitor effects of preventative interventions. Although the pathophysiology and prognostic signi cance of AS/PH have been well described, epigenetic in uences that impact transcript level expression of AS/PH remain poorly known.
The current study contributes to the eld by investigating individual CpG positions and co-methylated regions that are associated with multiple AS/PH traits using both epigenome-wide and candidate-gene approaches. Among a diverse cohort of individuals free from clinical CVD, we identi ed two signi cant differentially methylated positions (DMPs), cg20711926 (Bonferroni-signi cant) and cg25309493 (FDRsigni cant) that were associated with AIx independent of potential confounders. These two DMPs were positively associated with AIx across all groups in race/ethnic strati ed analysis.
AIx is a commonly used measure of late systolic pressure augmentation obtained from pulse wave analysis (PWA) 22 . PWA is a noninvasive method to generate the ascending aorta pressure wave from the pressure waveform measured in the radial artery 21 . The pressure waveform is a composite of the forward pressure wave (incident wave) and a re ected wave. In elastic vessels, because pulse wave velocity (PWV) is low, the re ected wave tends to arrive back at the aortic root during diastole. In stiff vessels, PWV is high and the re ected wave arrives back at the central arteries earlier, augmenting the systolic pressure. This augmented pressure (AP) is calibrated by PWA, and AIx is de ned as AP which is expressed as a percentage of pulse pressure 23 . While PWV is the gold standard of AS 24 , AIx can be affected by multiple factors (e.g., left ventricular ejection, PWV, timing of re ection, arterial tone, structure at peripheral re ecting sites, BP, and heart rate) 25  CYP1B1 (cytochrome P450 family 1 subfamily B member 1) is the gene annotated to cg20711926, the Bonferroni-signi cant DMP associated with AIx. CYP1B1 is a member of the CYP1 subfamily and encodes CYP1B1 enzyme which is involved in drug metabolism and synthesis of cholesterol, steroids, and other lipids 26 . While many studies reported the role of CYP1B1 in glaucoma and cancer 27,28 , recent studies have indicated that CYP1B1 is involved in cardiac pathophysiological changes. For example, CYP1B1 has been found to mediate angiotensin II-induced aortic smooth muscle cell migration, proliferation, and protein synthesis in rats 29 , as well as to contribute to cardiac hypertrophy induced by uremic toxin in mice 30 . Our nding further suggests that methylation of CYP1B1 may play a crucial role in determining AIx, although additional research is required to determine whether the association we observed is causal. Despite the need for further investigation, CYP1B1 may have noteworthy therapeutic implications for cardiac hypertrophy and AS/PH that are the important underlying mechanisms of CVD 31 .
In addition to CYP1B1, our study also prioritized three CpGs and their annotated genes (cg23800023-ETS1, cg08426368-TGFB3, and cg17350632-HLA-DPB1) for AIx, based on the overlap of signi cant results from three follow-up analyses of DMPs, including gene set enrichment analysis, expression quantitative trait methylation (eQTM) analysis, and functional enrichment analysis. Both ETS1 and TGFB3 have been previously prioritized as candidate genes 7 , although the candidate gene information was not used in our overall approach for the prioritization. Thus, our approach brings new independent evidence to support the roles of these two genes in processes regulating AS/PH.
First, the cg23800023 annotated gene ETS1 encodes the founding member of the family of ETS transcription factors. ETS family proteins have a conserved ETS DNA-binding domain that recognizes the GGAA/T in target genes and function as transcriptional activators or repressors of many genes 32 . In endothelial cells, vascular smooth muscle cells, and epithelial cancer cells, ETS is involved in regulating expressions of matrix metalloproteinase (MMP)1, 2, 3, and 9 and vascular endothelial growth factor [33][34][35][36] that are the well-known regulators of AS/PH 37 . Our study adds to the existing evidence supporting the link between ETS1 and AIx via the regulation of DNAm. However, further research is needed to establish the causality of this association.
Second, the cg17350632 annotated gene HLA-DPB1 (histocompatibility complex, class II, DP beta 1) belongs to a family of human leukocyte antigen (HLA) complex genes that help the immune system to distinguish the body's own proteins from proteins from foreign bodies. Among three subregions (DP, DQ, and DR) of HLA-D, HLA-DP molecule has not been extensively studied 38 , but available studies have demonstrated that HLA-DPB1 is associated with autoimmune disorder such as rheumatoid arthritis 39 and Behcet's disease 40 . The role of HLA-DPB1 in the vascular system has not been revealed yet. Our study presents novel information on the connection between HLA-DPB1 and AIx via the regulation of DNAm. Furthermore, both ETS1 and HLA-DPB1 have signi cant tissue correlations between Whole Blood and Aorta in GTEx, which suggests ETS1 and HLA-DPB1 could be potential biomarkers in understanding pathophysiology of AS/PH.
Lastly, the cg08426368 annotated gene TGFB3 (transforming growth factor beta 3) encodes a TGFB3 protein which is part of a large family of cytokines called TGFB superfamily 41 . TGFB3 regulates molecules involved in cell proliferation, cell differentiation, and apoptosis, and plays a critical role in the formation of blood vessels and wound healing 41,42 . Previous studies have indicated that TGFB pathways regulate the expressions of elastin, collagen, and MMP2 &9 43 , that are the major determinants of mechanical properties of large arteries 44 . Although TGFB3 did not show a signi cant tissue correlation between Whole Blood and Aorta in GTEx, the present study further con rms the potential relationship between TGFB3 and AIx mediated by DNAm.
It is noteworthy that all these three prioritized genes were involved in KEGG pathway hsa05166, Human Tcell leukemia virus 1 (HTLV-1) infection pathway. HTLV-1 infection can cause adult T-cell leukemia/lymphoma and HTLV-1-associated myelopathy 45,46 , as well as an in ammatory disease such as arthritis 47 . HTLV-1 is known to disturb the regulation of cytokines including interferon gamma (IFN-γ), tumor necrosis factor alpha (TNF-α), transforming growth factor beta (TGF-β), and IL-10 48 . Given the previous evidence that HTLV-1 triggers chronic in ammatory cascade, which is an important risk factor of atherosclerosis and CVD 49 , potential roles of HTLV infection in AS/PH are considered as an important gap that needs to be lled by future studies.
In summary, our study provides valuable information on epigenetic modi cations associated with AS/PH. The major strengths of our study include being one of the very few studies that examined the epigenomewide association with AS/PH and the use of population-based multi-ancestry data that include surrogate measures of AS and related hemodynamic parameters. In addition, we conducted a series of follow-up analyses to prioritize genes that are potentially relevant to AS/PH. A few limitations of our study should be noted. First, the study didn't use AS gold standard measure, carotid-femoral PWV 50 , due to data unavailability. Secondly, our study has limited statistical power of identifying signi cant DNAm differences associated with AS/PH traits due to the sample size. Thirdly, our study lacks a replication study due to the discrepancies in AS/PH measures across public data. Lastly, the study sample included relatively old adults with mean age of 60 (± 10) years; thus, the study ndings may be confounded with other CVD risks associated with aging. Future studies should aim to replicate the study ndings in younger and healthy populations and to investigate the biological processes in which the identi ed genes contribute to AS/PH. Such studies will help better understand pathophysiology of AS/PH and identify potential therapeutic targets of AS/PH.

Overview of approach
The overview of study design is shown in Fig. 1. We conducted analyses using Illumina's Methylation

Study participants
The study population consisted of participants free from clinical CVD enrolled in MESA, a longitudinal study of subclinical CVD and risk factors that predict progression to clinically overt cardiovascular disease or progression of the subclinical disease 51  to DNAm data prior to analysis, including color bias correction, median background adjustment, standard quantile adjustment, batch effect correction (using ComBat function in sva 58 R package), sex or race mismatch check, and outlier detection (using Gaphunter function in mini 59 R package). We further ltered out low methylation variation positions that the standard deviation of their methylation Betavalues is less than 0.02. Hence, 491,174 CpGs were left for epigenetic analyses. To preserve better statistical properties (i.e., homoscedasticity), the M-values (i.e, M = log(Beta/(1-Beta))) were used in the analyses.

Arterial stiffness and pulsatile hemodynamics traits
The study examined eight AS/PH traits, including: aortic augmentation index (AIx, %) measured by PWA; aortic arch pulse-wave velocity measured by cardiac magnetic resonance imaging (CMR-PWV, m/sec), ascending aortic distensibility (AAD, mmHg), descending aortic distensibility (DAD, mmHg); Young's Elastic Modulus (YM, mmHg) and distensibility coe cient (DC, mmHg), measured by carotid ultrasound; PTC1 (milliseconds) and PTC2 (milliseconds) obtained from radial artery pressure waveforms. The de nition and measurement of all eight AS/PH traits in MESA are described in previous studies 21,54,56 . Due to the skewness of raw phenotype data of eight AS/PH traits, the log-transformation was applied to AS/PH traits except for DC where the square-root transformation was applied.

EWAS at individual CpG
The linear regression adjusted for covariates was performed to test association between each AS/PH trait and M-value of individual CpG at epigenome-wide scale excluding sex chromosomes for 491,174 CpGs. The covariates included age, sex, race/ethnicity, BMI, smoking status (never, former, or current), smoking pack-years, mean arterial pressure, anti-hypertensive medication usage (yes or no), anti-diabetic medication usage (yes or no), lipid-lowering medication usage (yes or no), fasting glucose, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, triglycerides, estimated cell type proportions (Lymph, Mono, and Neu), and the rst four genomic principal components (PCs) of ancestry.
We carried out pooled analyses across self-reported race/ethnic groups for EWAS. The Illumina 60 was used for CpG annotation. We then generated a list of 59 candidate genes from literature-based approach 7 . The EWAS results for CpGs annotated to 59 candidate genes were referred to as candidategene approach results. All association tests were adjusted for multiple comparisons using false discovery rate (FDR) correction (Benjamini-Hochberg) at 5%. Due to the limited statistical power of this study, we considered CpGs with EWAS p-values passed suggestive signi cance as suggestive DMPs. The suggestive signi cance cut-offs for epigenome-wide approach and candidate-gene approach were 1x10 − 4 and 0.05, respectively. All suggestive DMPs were then used in the follow-up analyses.

EWAS at co-methylated genomic region
We applied coMethDMR 61 on 491,174 CpGs to identify DMRs. First, coMethDMR identi ed co-methylated sub-regions with closely located and co-methylated CpGs. We rst extracted clusters of CpGs located closely within genomic regions, i.e., the CpG cluster has at least three CpGs and the maximum separation between any two consecutive CpGs within the cluster is 200 base pairs. This step helps to ensure the subregions with similar CpG densities. Then we used the correlation between methylation levels among CpGs (i.e., rdrop statistics > 0.4 in coMethDMR) in a sub-region to identify co-methylated CpGs. Next, the median of M-values of CpGs within a co-methylated region was used to test association with AS trait in a random coe cient mixed model in coMethDMR that allowed us to model both variations between CpGs within the region and differential methylation simultaneously. The mixed model was also adjusted for the same covariates as for EWAS at individual CpG. We used the AnnotateResults function in coMethDMR to annotate co-methylated regions. The association results for co-methylated regions annotated to 59 candidate genes were referred to as candidate-gene approach results. Similar to the identi cation of suggestive DMP, the co-methylated region with association p-values passed suggestive signi cance was de ned as suggestive DMR. The suggestive signi cance cut-offs for DMR were 5x10 − 4 and 0.05 for epigenome-wide approach and candidate-gene approach, respectively. All suggestive DMRs were then used in the follow-up analyses as well.
Weighted gene co-expression network analysis (WGCNA) WGCNA 62 is a systems biology method that can be used to nd modules (clusters) with highly correlated methylation levels and to relate modules to clinical traits. We applied the WGCNA R package 63 on 491,174 CpGs to identify modules signi cantly associated with AS/PH traits. First, an unsigned comethylation network was constructed by using blockwiseModules function (soft thresholding power = 6, merge cut height = 0.25, and minimum module size = 30). 41 modules were identi ed from the WGCNA network. DNAm levels of CpGs within a module were summarized by the module eigengene (ME) value which represents the overall methylation level of CpGs clustering in a module. Next, the linear regression model adjusted for covariates was performed between ME value and AS/PH traits for each module to identify signi cantly associated modules with AS/PH traits. The covariates were the same as EWAS analysis. We considered the module with association p-value < 0.05 as a signi cant module. Finally, we used CpGs in the signi cant modules to carry out gene set enrichment analysis.

Gene set enrichment analysis (GSEA)
We conducted both Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis for each AS/PH trait on two sets of CpGs respectively. The rst set of CpGs, the signi cant EWAS set, was composed of CpGs from the union of both suggestive DMPs and DMRs identi ed based on suggestive signi cance of EWAS for each trait of interest. The second set of CpGs, the signi cant WGCNA-module set, contained CpGs in trait-associated signi cant modules (P < 0.05) from the WGCNA network. We used the gsameth function in the missMethyl 64 R package that was developed for genome-wide DNAm data to conduct GSEA. The FDR at 5% was applied to GSEA results to correct for multiple testing comparisons.

Expression quantitative trait methylation (eQTM) analysis
To identify genes whose expression is associated with signi cant DNAm differences, we conducted eQTM analysis for both suggestive DMPs and DMRs with their annotated genes (Illumina reference table was used to annotate genes for CpGs and DMR annotation was done by coMethyDMR). We used 587 multi-ancestry samples with both MESA Exam 1 RNA-seq normalized gene expression and DNA methylation for association analysis. First, we removed confounding effects in DNA methylation by tting the linear regression model M value ~ age + gender + race + rst 4 genomic PCs of ancestry + estimated cell type proportions (Lymph, Mono, and Neu) and extracting DNA methylation residuals from the model.
We used the median M value for the DMR. Similarly, we removed potential confounding effects in RNAseq by tting model normalized gene expression ~ age + gender + race + rst 4 genomic PCs of ancestry + PEER factors 1-10 and extracting gene expression residuals from the model. Next, for each gene expression and signi cant DNAm difference pair, we tested association between gene expression residuals (outcome) and DNA methylation residuals via a simple linear regression to quantify eQTM. The FDR at 5% was applied to eQTM results for suggestive DMPs and DMRs respectively to correct for multiple testing comparisons.

Functional enrichment analysis
To understand the complex interpretation of signi cant DNAm differences better, we applied eFORGE 65,66 to test whether our AS/PH trait-associated DNAm differences were enriched in regulatory elements from the Roadmap Epigenomics Consortium 67 across 20 tissues and cell types. The CpGs included in both suggestive DMPs and DMRs for each AS/PH trait were used for eFORGE 15 chromatin states enrichment analysis. eFORGE selects a background of 1,000 random CpGs with matching properties based on genecentric categories ( rst Exon, 3′ untranslated region or UTR, 5′UTR, Body, intergenic region or IGR, TSS1500 and TSS200) and CpG island-centric categories (CpG island, CpG island shore/shelf, N/A or "open sea"). Then the eFORGE uses a 1 kb proximity to lter out highly correlated input CpGs. Finally, eFORGE compares the number of CpGs overlapping regulatory elements from the reference panel with those obtained randomly to calculate enrichment scores for each of the selected cell types. eFORGE performs the Benjamini-Yekutieli approach to account for multiple testing corrections for a cell-type level signi cance.

Tissue correlation look-up in GTEx
We used GTEx v8 gene expression data 68 of both Whole-Blood and Aorta tissue (AS/PH-relevant tissue) to check their tissue correlation for the prioritized genes after follow-up analyses. The GTEx v8 gene expression data was downloaded from the GTEx portal. The inverse normalization was rst applied to 360 tissue-overlapped GTEx samples. Then both Pearson's correlation and its p-value were reported to measure tissue-correlation level.

Declarations Data availability
The arterial stiffness and pulsatile hemodynamics data for the Multi-Ethnic Study of Atherosclerosis Other authors declare no con ict of interests. Suggestive differentially methylated regions associated with aortic augmentation index.
AIx, aortic augmentation index; mean sd; median (25 th percentile, 75 th percentile). Table 2 Suggestive differentially methylated regions associated with aortic augmentation index. Study design. The suggestive DMP was de ned as CpG with association p-value less than 1x10 -4 and 0.05 respectively for epigenome-wide and candidate-gene approaches; the suggestive DMR was de ned as co-methylated region with association p-value less than 5x10 -4 and 0.05 respectively for epigenomewide and candidate-gene approaches; eQTM, expression quantitative trait methylation; AIx, aortic augmentation index; CMR-PWV, aortic arch pulse-wave velocity measured by cardia magnetic resonance imaging; AAD, ascending aortic distensibility; DAD, descending aortic distensibility; YM, Young's Elastic Modulus; DC, distensibility coe cient; PTC1 and PTC2, radial artery pressure waveform index 1 and 2.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.