DNA methylation is associated with von Willebrand factor and coagulation factor VIII plasma levels: the atherosclerosis risk in communities study

Abstract

Background

Limited information is available on how DNA methylation of cytosine-phosphate-guanine (CpG) sites across the genome regulate circulating von Willebrand Factor (VWF) and factor VIII (FVIII) levels.

Methods

We performed an epigenome-wide association study to examine the association of leukocyte DNA methylation levels at 483,735 CpG sites with VWF antigen plasma levels and FVIII activity in 2,597 Black and 1,011 White participants from the Atherosclerosis Risk in Communities (ARIC) study. VWF antigen levels and FVIII activity were measured at the baseline exam, while DNA methylation was measured with the Infinium HumanMethylation450 BeadChip array at visit 2 or 3. Discovery analysis was performed in the Black population, with replication in the White population.

Results

We identified 55 and 46 significant CpG sites associated with VWF and FVIII (P < 1.03 × 10^–7) in the discovery analysis, respectively. Among these, for VWF, 13 were replicated, mapping to one novel (NBEAL2) and one known (ABO) locus. For FVIII, 12 associations were replicated which mapped to 1 novel locus (B3GALT4) and 2 known but previously unreplicated loci (ABO and CORO1A). When we adjusted for the variants previously identified as independently associated with VWF and FVIII at the ABO locus, all attenuated at least 10% towards the null after adjustment for genetic variants.

Conclusion

We identified novel epigenetic associations with VWF and FVIII levels at the NBEAL2 and B3GALT4 loci. NBEAL2 is critical for the biosynthesis of platelet alpha granules, which store proteins that enable platelet adhesion initiation, including VWF while B3GALT4 is involved in glycosylation of glycoproteins like VWF and FVIII.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13148-025-02028-2.

Introduction

Factor VIII (FVIII) is a glycoprotein that is synthesized primarily in liver sinusoidal endothelial cells, and it circulates in the bloodstream inactively, bound to von Willebrand Factor (VWF), until an injury to the blood vessels is detected [1, 2]. FVIII interacts with activated factor IX to activate factor X which initiates the common pathway of the coagulation cascade. Activated factor X then converts prothrombin to thrombin, which ultimately converts fibrinogen to fibrin, and a stable fibrin clot or a mesh is formed.

VWF, the carrier protein for FVIII, is a glycoprotein that is almost exclusively produced by endothelial cells, and its plasma levels fluctuate according to the different states of endothelial damage [3]. It is also an important factor that mediates adherence of platelets to a damaged vascular wall and thus in facilitating platelet aggregation [4, 5]. Since VWF is a carrier protein for FVIII, without its presence, FVIII is rapidly catabolized. Conversely, when FVIII is bound to VWF, its half-life is extended six-fold [6]. Thus, VWF levels are the primary determinant of circulating FVIII levels.

Given their well-established roles in coagulation, it has been hypothesized that elevated levels of FVIII and VWF are risk factors for thrombotic diseases. Epidemiological studies show that elevated plasma levels of FVIII and VWF are associated with increased risk of venous thromboembolism [7–12], ischemic stroke [12–15], myocardial infarction [16] and peripheral artery disease [16–18].

Epigenetic modifications are potentially heritable biochemical alterations to the genome that do not change the underlying genetic makeup, but they can alter the phenotypic expression [19, 20]. DNA methylation is an epigenetic mechanism whereby a methyl group is attached to the DNA sequence, specifically to a cytosine nucleotide followed by a guanine nucleotide, also referred to as a CpG site. DNA methylation can be modified by both genetic and environmental factors, and, as such, investigating the variation in DNA methylation may further explain the phenotypic variation in FVIII and VWF levels.

Epigenome-wide association studies (EWAS) have successfully explored relationships between DNA methylation and a range of clinical phenotypes such as blood pressure, body mass index, lipid levels and C-reactive protein (CRP) [19, 21–25]. However, to the best of our knowledge, to date there has not been an EWAS study performed for VWF, and only an EWAS of small sample size was conducted for FVIII, which identified 30 CpG sites in the Jackson Heart Study (JHS), but the findings did not replicate [13]. Thus, the aim of this study was to explore and further expand knowledge on epigenetic determinants of FVIII and VWF by performing an epigenome-wide association study to identify epigenetic markers that are associated with FVIII and VWF.

Methods

Study population and design

This study was performed in the Atherosclerosis Risk in Communities (ARIC) study to examine the association between leukocyte DNA methylation levels at CpG sites and circulating VWF antigen and FVIII activity levels. The ARIC study is a population-based prospective cohort study of cardiovascular disease sponsored by the National Heart, Lung, and Blood Institute (NHLBI) [26, 27]. The baseline visit took place from 1987 to 1989, followed by a second visit in 1990–1992 and a third visit in 1993–1995. Additional details about the ARIC study are provided in the Supplementary Methods.

For both VWF and FVIII, the discovery study population included methylation data from the Black population group (n = 2597), and the replication study population included methylation data from the White population group (n = 1011). All individuals had measures of both DNA methylation and VWF or FVIII levels. The population group was determined based on self-reporting.

VWF and FVIII measurements

VWF antigen, measured at the baseline exam, was determined by ELISA kits from American Bioproducts Co (Parsippany, NJ). Factor VIII activity was also measured at the baseline exam by determining the ability of the tested sample to correct the clotting time of human factor VIII–deficient plasma obtained from George King Biomedical Inc (Overland Park, KA). Circulating VWF antigen and FVIII activity levels were measured in IU/dL among participants [16, 28]. VWF and FVIII levels were natural log-transformed to approximate the normal distribution, and outliers with trait values greater than 3 standard deviations (SD) from the log-transformed mean were removed [29].

DNA methylation quantification

Genome-wide DNA methylation levels in whole blood from visit 2 or 3 were measured using the Infinium HumanMethylation450 BeadChip array (Illumina Inc, San Diego, CA, USA). DNA methylation β value normalization was conducted using Noob background and dye bias correction [30] and the Beta Mixture Quantile dilation (BMIQ) method [31]. The degree of DNA methylation was expressed using β values, which ranged from 0 to 1. The β value of a specific CpG site can be interpreted as the proportion of copies of that CpG site that are methylated. Quality control procedures of DNA methylation quantification are described in the Supplementary Methods.

Discovery and replication analyses

For each phenotype, an EWAS of the Black population group was performed. A linear mixed model regression was used to evaluate the association of inter-individual variation in DNA methylation and VWF and FVIII levels. This model included VWF or FVIII levels as an independent variable, and DNA methylation values as the outcome variable. This model was adjusted for age, sex, the time difference between the measurements of DNA methylation and phenotypes, white blood cell proportions, and smoking (never vs. current). Measured or estimated white blood cell proportions were included to account for various methylation patterns within many blood cell subpopulations [32, 33]. Estimation was performed using the minfi package in R, implementing the Houseman method [34, 35]. Technical covariates including plate number, Chip ID and Chip position, to control for batch effects were also included in the model as random effects variables. A Bonferroni corrected p-value was used to determine the significance threshold (0.05/number of CpG sites included).

Any significant associations from the discovery analysis were replicated in the White population group, using the same statistical model as used in the discovery analysis. A Bonferroni correction was applied to adjust the significance threshold for the number of tested associations.

Sensitivity analyses

Genetic variants at the ABO locus are strongly associated with VWF and FVIII levels [29]. A sensitivity analysis was performed to further control for this association, by additionally adjusting associations with CpGs at the ABO locus for genetic variants at the ABO locus. The genetic variants that were adjusted for were identified as independently associated with VWF and FVIII and population-specific in a conditional analysis using Genome-wide Complex Trait Analysis (GCTA) from a previous genome-wide association study (GWAS) [29, 36]. We adjusted for rs56392308 and rs8176719 for VWF in the Black population group and for rs74503122, rs10901252, rs56392308, and rs975381715 in the White population group. For FVIII, we adjusted for rs5945282, rs8176746, and rs8176719 in the Black population group, and for rs56392308, rs8176721, and rs975381715 in the White population group. The population-specific adjusted results were then subsequently meta-analyzed using fixed-effect inverse-variance weighting as implemented in METAL [37].

A number of ARIC participants at the Jackson, Mississippi site was later enrolled in JHS. Thus, to ensure independent replication of the findings previously reported in the JHS [13], we reported the analysis in ARIC while excluding any JHS participants. For this replication, a Bonferroni correction was used to adjust the p-value threshold for the number of tested CpG sites.

To assess potential cis-regulatory methylation effects at the genes encoding VWF and FVIII, we conducted a targeted analysis in Black and White population groups restricted to CpG sites located within the VWF (chr12:6,058,040 − 6,233,936) and F8 (chrX:154,064,063–154,255,215) genes according to their genomic coordinates. Bonferroni correction was applied according to the number of CpGs tested within each locus to account for multiple comparisons.

Annotation of associated CpG sites

For the epigenome-wide significant and replicated CpG sites, the nearest genes were identified and annotated using the genomic coordinates provided by Illumina (GRCh37/hg19). They were also checked to see whether any of the probes are cross-reactive, if any CpG sites were single-nucleotide polymorphisms (SNPs) and whether a SNP existed in the probe binding sites [38]. Additionally, the results were cross-referenced with the EWAS Catalog to gain insights on existing findings on those associations present in the literature [39]. Finally, any associations of gene expression levels with methylation at epigenome-wide significant and replicated CpG sites, also known as expression quantitative trait methylation sites (eQTMs), were identified using published results from the BIOS consortium [40].

To provide insight into the inter-dependence of the significant and replicated CpG sites, we evaluated their correlation structure by calculating the pairwise Pearson correlation coefficients. Correlation coefficients were computed separately in the White and Black population groups to allow for differences in the correlation structure between population groups.

Bidirectional Mendelian randomization

We performed bi-directional two-sample Mendelian randomization (MR) analyses to provide evidence in support of causal effects underlying the associations between VWF or FVIII and DNA methylation at significant and replicated CpG sites. By using a bidirectional approach, we explored whether genetically determined DNA methylation levels alter VWF or FVIII levels or whether genetically determined VWF or FVIII levels influence DNA methylation levels. For both directions of MR, the primary analyses combined evidence from multiple genetic instruments using fixed-effect inverse variance-weighted meta-analysis if there was more than one instrument. For the case with only one instrument, the Wald test was utilized, as implemented in the R package TwoSampleMR [41].

We examined the association between genetically determined DNA methylation levels at each CpG site with VWF or FVIII levels. For significantly associated and replicated CpG sites from each phenotype, genetic instruments were identified using a publicly available database of methylation quantitative trait loci (meQTLs) from the Genetics of DNA Methylation Consortium (GoDMC) [42]. Genetic variants associated with each CpG site at p < 0.05 were distance-pruned to select the most significant variant at each associated locus to use as genetic instruments (± 1 Mb). We obtained effect estimates for these instruments on VWF or FVIII levels from summary statistics of the largest published GWAS of VWF and FVIII [29]. A Bonferroni corrected significance threshold (0.05/number of CpG sites that had one or more instruments) was applied to determine significance. Linkage disequilibrium was also checked with rs8176719, the deletion that determines the ABO blood group, using LDlink [43].

We also examined the association between genetically determined VWF or FVIII levels with DNA methylation levels at each CpG site. We utilized the largest published GWAS of VWF and FVIII to identify genetic variants associated with VWF or FVIII using a significance threshold of p-value < 2.5 × 10^{− 8} [29]. Genetic variants associated with FVIII or VWF were then distance-pruned to select the most significant variant at each associated locus to use as genetic instruments (± 1 Mb). We obtained effect estimates and standard errors for the association of these instruments with DNA methylation levels obtained from association analyses within the ARIC study that are further described in the Supplementary Methods. In addition, we also performed sensitivity analyses, excluding the ABO locus, to better define the causal association between genetically determined VWF or FVIII levels with DNA methylation levels.

Methylation-derived scores for VWF and FVIII

Methylation-based scores were developed in the Black population group and then tested in the White population group, to avoid bias from sample overlap. For each phenotype we first identified epigenome-wide significant CpGs in Black population group and then applied correlation-based pruning (r² < 0.1) to obtain a set of independent CpGs. For the resulting independent CpGs, we estimated effect sizes (weights) using models that were identical to those used in the EWAS, but with VWF or FVIII as the dependent variable and each CpG as the dependent variable. These weights were then applied to the White population groups to generate methylation-derived scores by multiplying standardized CpG beta values by their corresponding weights and summing across CpGs. Scores were standardized to mean = 0 and standard deviation = 1.

As part of score validation, we quantified how much variance in VWF or FVIII VWF levels was explained by each methylation-derived score using mixed-effects models. The percent variance explained was calculated as the difference in marginal R-squared values between models with and without the score term, multiplied by 100 to express it as a percentage. Marginal R-squared values were obtained using the MuMIn R package.

After excluding participants with prevalent ischemic stroke or coronary heart disease, we then tested associations of the scores with incident ischemic stroke, and coronary heart disease (CHD) from the ARIC study using Cox proportional hazards models, adjusted for age, sex, current smoking status, BMI, systolic blood pressure, type 2 diabetes, high-density lipoprotein cholesterol, total cholesterol, lipid lowering medications, and antihypertensive medications. Ischemic stroke and CHD were defined as described previously [44]. All covariates were taken from the same visit as methylation measurement and follow-up for incident events started at that visit. Additionally, as a sensitivity analysis all outcome models were further adjusted for blood group O.

Results

Study characteristics

Descriptive characteristics of the ARIC study (n = 3,638) are shown in Table 1. The mean VWF level for the discovery group was 133.37 IU/dL (SD = 55.69) and 111.51 IU/dL (SD = 40.42) for the replication group. For FVIII, the mean was 145.90 IU/dL (SD = 44.86) and 123.07 IU/dL (SD = 29.17) for the discovery and replication group, respectively. Among the discovery group, 63.42% were females, and the mean age was 53.33 years (SD = 5.75) while the replication group had 58.50% females and a mean age of 56.14 years (SD = 5.09).

Table 1.

Characteristics of the ARIC study

	Population
	Black	White
Visit 2
Sample Size	2272	759
Sex (% Female)	63.07	61.00
Smoking (% Current Smoker)	29.01	22.27
Age (SD)	53.27 (5.73)	55.66 (5.14)
BMI (kg/m2) (SD)	29.71 (6.05)	25.08 (4.27)
Visit 3
Sample Size	325	282
Sex (% Female)	65.85	51.77
Smoking (% Current Smoker)	29.85	21.63
Age (SD)	53.76 (5.87)	57.43 (4.74)
BMI (kg/m2) (SD)	29.47 (5.87)	25.85 (4.20)
VWF* (IU/dL) (SD)	133.37 (55.69)	111.51 (40.42)
FVIII* (IU/dL) (SD)	145.90 (44.86)	123.07 (29.17)

*VWF and FVIII measurements are from the baseline visit

Abbreviations: SD, standard deviation; BMI, body mass index; VWF, von Willebrand factor; FVIII, coagulation factor FVIII

Epigenome-wide association analysis (discovery)

A total of 483,735 CpG sites were included in the analysis and 55 CpG sites were significantly associated (Bonferroni corrected p-value < 1.03 × 10^− 7) with VWF levels in the Black population group, and these mapped to 16 independent loci (Table S1, Fig. 1). For FVIII, we identified 46 significantly associated CpG sites that mapped to 15 independent loci (Table S2, Fig. 2).

Fig. 1.

Manhattan plot of VWF levels in the Black population group. On the x-axis is the chromosomal position, and the y-axis represents the significance threshold on a logarithmic scale. The dotted line represents a significance threshold. When a CpG site is located higher along the y-axis, it refers to higher significance. There are many epigenome-wide signals that meet the significance threshold (P < 1.03 × 10^− 7). *Chr 23 and Chr 24 represent chromosomes X and Y, respectively

Fig. 2.

Manhattan plot of FVIII levels in the Black population group. On the x-axis is the chromosomal position, and the y-axis represents the significance threshold on a logarithmic scale. The dotted line represents a significance threshold. When a CpG site is located higher along the y-axis, it refers to higher significance. There are many epigenome-wide signals that meet the significance threshold (P < 1.03 × 10^− 7). *Chr 23 and Chr 24 represent chromosomes X and Y, respectively.

Replication

Among the 55 significant CpG sites for VWF, 13 were replicated in the White population group (Bonferroni corrected p-value < 9.09 × 10^− 4), mapping to one novel (NBEAL2) and one known (ABO) locus (Table 2; Table S1; Table S3; Fig. 3). For FVIII, 12 significant associations were replicated in the White population group (Bonferroni corrected p-value < 1.09 × 10^− 3) which mapped to 1 novel locus (B3GALT4) and 2 known but previously unreplicated loci (ABO and CORO1A) (Table 3; Table S2; Table S4; Fig. 4). Those that were significant but did not replicate for VWF and FVIII mapped to a total of 13 independent loci (Tables S1-S2; Figures S1-S2).

Table 2.

Significant and replicated DNA methylation sites associated with Circulating VWF levels

CpG site	Chr	Position	Effect size (Black)	P-value (Black)	Effect size (White)	P-value (White)	Gene
cg26432350	3	47,040,357	0.0123	1.22E-08	0.0150	2.56E-04	NBEAL2
cg24605046	6	33,245,895	0.0133	8.59E-08	0.0152	4.24E-04	B3GALT4
cg12020464	9	136,131,183	−0.0310	2.62E-37	−0.0220	2.66E-06	ABO
cg11879188	9	136,149,908	0.0321	1.69E-22	0.0656	4.34E-15	ABO
cg21160290	9	136,149,941	0.0844	3.76E-39	0.1494	5.55E-31	ABO
cg22535403	9	136,150,032	0.0811	2.02E-38	0.1393	1.42E-29	ABO
cg13506600	9	136,150,361	0.0252	1.46E-22	0.0312	8.74E-09	ABO
cg07241568	9	136,150,823	0.0105	2.29E-12	0.0106	3.04E-08	ABO
cg24267699	9	136,151,359	0.0329	2.42E-30	0.0390	2.61E-22	ABO
cg06818865	9	136,151,958	0.0098	3.87E-10	0.0209	3.31E-06	ABO
cg13683939	9	136,152,547	−0.1284	2.39E-17	−0.0665	3.07E-04	ABO
cg14271713	9	136,153,846	−0.0049	2.10E-08	−0.0067	2.15E-04	ABO
cg13660174	9	136,238,392	0.0209	4.09E-09	0.0219	1.14E-04	SURF4

Bolded CpG sites represent the most significant variant at a given locus

Fig. 3.

Significant and Replicated CpGs in VWF

Fig. 4.

Significant and Replicated CpGs in FVIII

Table 3.

Significant and replicated DNA methylation sites associated with Circulating FVIII levels

CpG site	Chr	Position	Effect size (Black)	P-value (Black)	Effect size (White)	P-value (White)	Gene
cg00052772	6	33,245,804	0.0156	6.50E-09	0.0144	7.92E-04	B3GALT4
cg24605046	6	33,245,895	0.0197	2.14E-08	0.0208	1.56E-04	B3GALT4
cg12020464	9	136,131,183	−0.0427	3.25E-36	−0.0319	1.49E-07	ABO
cg11879188	9	136,149,908	0.0447	6.61E-22	0.0960	3.45E-18	ABO
cg21160290	9	136,149,941	0.1156	2.35E-37	0.2007	5.64E-33	ABO
cg22535403	9	136,150,032	0.1110	1.47E-36	0.1852	1.91E-31	ABO
cg13506600	9	136,150,361	0.0330	1.56E-19	0.0468	2.07E-11	ABO
cg07241568	9	136,150,823	0.0158	1.23E-13	0.0110	1.20E-05	ABO
cg24267699	9	136,151,359	0.0487	2.06E-33	0.0549	9.90E-27	ABO
cg06818865	9	136,151,958	0.0120	5.37E-08	0.0327	9.81E-09	ABO
cg13683939	9	136,152,547	−0.1660	5.36E-15	−0.1255	3.76E-07	ABO
cg06038367	16	30,198,370	0.0188	1.97E-08	0.0209	1.99E-06	CORO1A

Bolded CpG sites represent the most significant variant at a given locus

Sensitivity analyses

The original models for VWF and FVIII were additionally adjusted for previously reported genetic variants at the ABO locus [29]. For both VWF and FVIII, the effect sizes for the associations with CpG sites at the ABO locus were all attenuated at least 10% towards the null after adjustment for genetic variants. All p-values decreased more than 1 order of magnitude after adjustment, although some remained significant (Table S5-S6).

We also performed a sensitivity analysis for FVIII, excluding individuals also in JHS. Out of 30 significant CpGs reported by Raffield et al. [13] using the more comprehensive Illumina EPIC array, 15 were included in our meta-analysis that excluded JHS. Of the 15 CpGs available in our sensitivity analysis excluding JHS samples, 13 were significantly associated (Bonferroni corrected p-value < 3.33 × 10^{− 3}) (Table S7).

In gene-restricted analyses focusing on 72 CpG sites within the VWF gene and 42 CpG sites within the F8 gene, several CpGs within VWF demonstrated significant associations (Bonferroni corrected p-value < 0.001) with circulating VWF levels, including three in the Black population group and two in the White population group (Table S8). No CpGs within F8 gene reached Bonferroni significance (Bonferroni corrected p-value < 0.002) for association with FVIII levels (Table S9).

Bidirectional Mendelian randomization

MR analysis evaluating the association between genetically determined DNA methylation levels and VWF levels yielded 9 CpG sites that had one instrument, and 2 CpG sites with two instruments (Table S10). Each independent locus was represented in the MR analysis by at least one CpG site. With a Bonferroni corrected significance threshold of 4.55 × 10^− 3, we identified two CpG sites at ABO and the nearby SURF4 gene where genetically determined DNA methylation was associated with VWF levels (Tables S10-S11): cg13660174 (beta_WaldRatio = 36.7, p-value = 5.18 × 10^− 182) and cg12020464 (beta_WaldRatio = −13.2, p-value = 1.07 × 10^− 252). Similarly, for FVIII, genetically determined DNA methylation of one of the same two CpG sites (cg12020464; beta_WaldRatio = −11.8, p-value = 2.20 × 10^− 242) was associated with FVIII levels (Tables S12-S13). Of note, the meQTLs for cg12020464 and cg13660174 were in linkage disequilibrium with the deletion that determines the ABO blood group O (rs8176719) (Tables S11, S13).

MR analyses evaluating the association between genetically determined VWF or FVIII levels on DNA methylation identified 1 (cg12020464) and 3 (cg06818865, cg11879188, cg13506600) CpG sites at the ABO locus, using 19 and 20 genetic instruments, respectively, at their respective Bonferroni corrected significance threshold (Tables S14-S17). However, when we excluded genetic instruments located at ABO from the analysis, there were no significant associations between genetically determined VWF or FVIII levels and DNA methylation at CpG sites (Tables S14-S17).

Annotation

None of the significant and replicated CpG sites for both phenotypes had cross-reactive probes, were a SNP, or harbored a SNP existing in the probe binding sites (Table S3-S4). According to the EWAS Catalog, significant and replicated CpG sites for VWF and FVIII identified 37 and 24 distinct associations, respectively, in previous EWAS of various phenotypes including smoking, TNF-α concentration and ulcerative colitis, among many (Table S18-S19). None of the significant and replicated CpG sites were associated with altered gene expression levels in the eQTM database from the BIOS Consortium [40]. There was a widespread correlation amongst the significant and replicated CpG sites for both VWF and FVIII, as shown in Figures S3 and S4. Within each locus, the correlations were generally higher between CpG sites, and they seemed to lack independence.

Methylation-derived scores for FVIII and VWF

18 participants with prevalent ischemic stroke and 68 participants with prevalent CHD were excluded from analyses with incident events. Among 922 participants from the White population groups that were free of ischemic stroke at baseline, 61 incident ischemic stroke events occurred during follow-up. Among 872 participants from the White population groups that were free of CHD at baseline, 166 incident CHD events occurred during follow-up. The mean follow-up time among controls was 26.6 years for ischemic stroke and 25.5 years for CHD.

For FVIII, 46 independent CpGs were selected for inclusion into the methylation-based score, which explained 14.24% of the variance in FVIII levels (Table S20). For VWF, 55 independent CpGs were selected for inclusion into the methylation-based score, which explained 14.51% of the variance in VWF levels (Table S21). Though all associations were in the expected direction, associations with ischemic stroke were not significant (Table S22). After adjustment for blood group O, each 1 standard deviation increase in the VWF score was associated with 1.33 times the hazard of incident CHD (95% confidence interval = 1.11–1.61, p-value = 0.0027), and each standard deviation increase in the FVIII score was associated with 1.30 times the hazard of incident CHD (95% confidence interval = 1.06–1.58, p-value = 0.011).

Discussion

In this EWAS of VWF and FVIII, we identified 55 CpG sites that were associated with VWF and 46 that were associated with FVIII. Among these, for VWF, 13 associations were replicated, mapping to one novel (NBEAL2) and one known (ABO) locus. For FVIII, 12 associations were replicated which mapped to 1 novel locus (B3GALT4) and 2 known but previously unreplicated loci (ABO and CORO1A). All the associations at the ABO locus were attenuated when additionally adjusted for the conditional independent variants of GCTA analyses from a previous GWAS of VWF and FVIII.

DNA methylation at cg26432350 near NBEAL2 was positively associated with VWF. NBEAL2, neurobeachin-like 2, is critical for the biosynthesis of platelet alpha granules, which store proteins that enable platelet adhesion initiation, including VWF [45]. Therefore, mutations in NBEAL2 can disrupt the normal function of platelets, impairing their ability to form clots effectively. This disruption in hemostasis can result in excessive bleeding or easy bruising in affected individuals. In 2011, autosomal recessive mutations in NBEAL2 were identified as causative for gray platelet syndrome (GPS) [46–48]. Gray platelet syndrome is a rare autosomal recessive platelet disorder, marked by faulty formation of alpha-granules in megakaryocytes and platelets [49], leading to significantly impaired release of VWF during platelet activation [50].

Another novel finding in association with FVIII levels was cg00052772, located near the B3GALT4 gene. This gene family encodes glycoproteins like VWF and FVIII, with diverse enzymatic functions using different donor substrates (UDP-galactose and UDP-N-acetylglucosamine) and different acceptor sugars (N-acetylglucosamine, galactose, N-acetylgalactosamine) to add galactose residues during glycosylation [51]. Notably, B3GALT4 is involved in the synthesis of glycan structures, including those containing galactose residues that serve as ligands for Macrophage Galactose-Type Lectin (MGL) [51–53]. MGL is one of the known plasma clearance receptors of FVIII, and is expressed on macrophages and binds to glycoproteins expressing terminal N-acetylgalactosamine or galactose sugars [52–54]. In a study utilizing an MGL-deficient mouse model, the FVIII level was approximately 1.3-fold elevated, which highlights the important contribution of MGL to the regulation of FVIII clearance [53].

A CpG site, cg24044988, near CORO1A was previously identified to be associated with FVIII levels, but this finding lacked replication [13]. In our study, cg06038367 located near CORO1A was again significantly associated with FVIII levels and was also replicated in the White population group. The results remained the same when we excluded the overlapping samples between ARIC and JHS. CORO1A encodes a protein called coronin-1 A which is a family of actin-binding proteins that has been implicated in various cellular processes. It was shown to regulate platelet actin dynamics which modifies platelet size and cytoskeletal arrangement following activation, thereby attributing to arterial thrombosis [55–59]. In a mouse model of ferric chloride-induced arterial thrombosis, the absence of CORO1A was associated with reduced thrombus stability and consolidation [55]. It also was found to have a potential role in coronary artery disease as higher expression of CORO1A in the plasma of patients was found in myocardial infarction cases compared to controls [60].

Our MR found that genetically determined DNA methylation at cg12020464 was associated with both VWF and FVIII levels, while methylation at cg13660174 was associated with VWF levels. Notably, cg12020464 is located at the ABO gene and cg13660174 is located nearby at the SURF4 gene. Given that the meQTLs for cg06818865 and cg12020464 were both in linkage disequilibrium with the deletion that partially determines the ABO blood group, the MR fundamental assumption of ‘no horizontal pleiotropy,’ was likely violated. ABO variants have strong direct effects on hemostatic traits and thus, signals at CpGs nearing ABO possibly reflect underlying genetic effects rather than causal methylation effects. Consistent with this, sensitivity analyses adjusting for ABO variants for the analysis of genetically determined VWF or FVIII levels on DNA methylation showed associations only when ABO instruments were included, and all associations became null when ABO instruments were excluded. In contrast, an MR of DNA methylation level on the phenotypes, excluding ABO instruments, was not feasible, as the CpGs associated with VWF and FVIII are almost exclusively instrumented by ABO-linked variants, leaving no independent instruments. Taken together, our ABO-locus findings should be interpreted with caution.

Beyond locus-specific analyses, we also constructed methylation-derived scores for FVIII and VWF and evaluated their associations with incident cardiovascular outcomes. We found some evidence of associations between methylation-derived scores for FVIII and VWF with incident cardiovascular outcomes, in particular with CHD. Scores were developed in the Black population groups and tested in the White population group, and thus the association analyses were limited by the small sample size and low number of cardiovascular events in our testing population. Nevertheless, the results show potential for that larger EWAS to derive more comprehensive and generalizable methylation scores with potential translational utility in cohorts lacking direct FVIII or VWF measurements.

Our study had the strength of having independent discovery and replication populations, which ensured that the CpG sites that replicated were more generalizable across different population groups. This study also had limitations. A key limitation of our study is that DNA methylation was measured in leukocytes, which may not fully capture tissue-specific epigenetic regulation of VWF and FVIII, as these proteins are primarily synthesized in hepatocytes and endothelial cells. Although a few studies have examined methylation patterns in liver biopsies [61, 62], such approaches remain infeasible in large population-based settings due to the invasive nature of tissue collection. In addition, although we adjusted for estimated leukocyte cell type proportions, residual confounding due to unmeasured cell subtypes may persist [44, 45]. Future studies using induced pluripotent stem cells differentiated into hepatocyte- or endothelial-like cells, or employing cell type-specific assays such as cell-type deconvolution methods [63] and single-cell methylation profiling [64], will help clarify tissue-relevant mechanisms and be valuable for disentangling cell type-specific effects from true epigenetic associations.

While our analyses identified CpGs at B3GALT4 and CORO1A as associated with FVIII levels, the functional mechanisms underlying these associations remain speculative. There is currently no direct evidence that B3GALT4 contributes to the glycosylation of VWF or FVIII, or that CORO1A plays a mechanistic role in their regulation. Prior studies have implicated B3GALT4 more generally in glycoprotein synthesis and clearance, suggesting a possible but as yet untested pathway through which FVIII biology may be influenced. These findings should therefore be interpreted as hypothesis-generating. Future functional validation studies, such as targeted epigenetic perturbation using CRISPR editing or gene silencing of B3GALT4 and CORO1A in hepatocyte and endothelial models, will be important to determine whether these loci directly modulate FVIII levels or act through alternative pathways.

There were also limitations innate to MR analyses. We had limited statistical power to assess the association between genetically determined VWF or FVIII levels and DNA methylation in our MR analyses. The genetic instruments for VWF and FVIII explain approximately 29% and 25% of the variance in VWF and FVIII levels, respectively [29], and our sample size to assess the association of these instruments with DNA methylation was only relatively modest. As a result, null findings from the MR analyses should be interpreted with caution. Larger EWAS and MR studies will be needed to more definitively evaluate these relationships.

In conclusion, we performed an EWAS to examine the association of DNA methylation levels with VWF antigen levels and FVIII activity in the ARIC study. We identified 13 and 12 significant and replicated CpG sites where circulating leukocyte DNA methylation is robustly associated with VWF and FVIII levels, respectively. Our novel findings reflected glycosylation and platelet biology. Understanding the functions of NBEAL2, B3GALT4, and CORO1A and their associated pathways will further provide insights into the molecular mechanisms underlying genetic disorders associated with VWF and FVIII levels, and contribute to the development of potential therapeutic strategies.

Author contributions

E.B., and A.C.M. contributed to phenotype measurement or harmonization (including design/funding of contributing studies). J.B., and M.F. contributed to methylation data measurement and its quality control process. J.H., and P.d.V. contributed to statistical analyses. J.H., N.L.S., A.C.M., and P.d.V. contributed to the interpretation of results. J.H., J.B., P.d.V., and A.C.M. contributed to the drafting of the manuscript. All authors contributed to the revision of the manuscript.

Funding

This study was supported by National Heart, Lung, and Blood Institute (NHLBI) grant R01HL139553. The Atherosclerosis Risk in Communities study has been funded in whole or in part with Federal funds from the NHLBI, National Institutes of Health, Department of Health and Human Services, under Contract nos. (75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). The authors thank the staff and participants of the ARIC study for their important contributions. Funding was also supported by R01HL087641 and R01HL086694; National Human Genome Research Institute contract U01HG004402; and National Institutes of Health contract HHSN268200625226C for ARIC GWAS. Infrastructure was partly supported by Grant Number UL1RR025005, a component of the National Institutes of Health and NIH Roadmap for Medical Research. For DNA methylation, funding was supported by 5RC2HL102419 and R01NS087541.

Data availability

Summary statistics from the EWAS of VWF and FVIII conducted in this study will be made publicly available through the EWAS Catalog (https://ewascatalog.org) upon publication.

Declarations

Ethics approval and consent to participate

The ARIC study has been approved by a single Institutional Review Board (IRB) at Johns Hopkins School of Medicine and IRBs at all participating institutions: University of North Carolina at Chapel Hill IRB, Johns Hopkins University School of Public Health IRB, University of Minnesota IRB, Wake Forest University Health Sciences IRB, and University of Mississippi Medical Center IRB. Study participants provided written informed consent at all study visits.

Competing interests

The authors declare no competing interests.