A Peripheral Blood DNA Methylation Signature of Hepatic Fat Reveals a Potential Causal Pathway for Nonalcoholic Fatty Liver Disease

The estimated global prevalence of nonalcoholic fatty liver disease (NAFLD) in adults is 24% and has increased substantially along with the increasing rates of obesity (1). NAFLD correlates with increased risk of type 2 diabetes (T2D) (2), and it is the second leading contributor to hepatic failure necessitating transplantation (3). A prior study in three family-based cohorts estimated the heritability of hepatic steatosis to be 27%; however, common genetic variants from genome-wide association studies (GWAS) account for <5% of interindividual variance in hepatic fat (4). Epigenetics may explain part of the interindividual variance of steatosis. Several studies have demonstrated altered DNA methylation profiles in liver biopsy samples collected from individuals with NAFLD (5,6). One study showed that peripheral blood–derived DNA hypermethylation at one cytosine-guanine dinucleotide (CpG) (cg06690548) located in an intron of SLC7A11 may be associated with a lower risk of steatosis (7). In general, however, prior studies were limited by small sample sizes to discover DNA methylation sites associated with hepatic fat accumulation.

To bridge this knowledge gap, we examined the epigenome-wide association between DNA methylation at >400,000 CpGs and hepatic fat in European ancestry (EA), African ancestry (AA), and Hispanic ancestry (HA) participants from five population-based cohort studies with hepatic fat measurements derived from noninvasive imaging. For hepatic fat–associated CpGs, we further examined their relations to genetic variants, gene expression, and regulatory functions and explored potential relations of DNA methylation to NAFLD and T2D.

The current study included multiethnic participants from five population-based cohorts including the Coronary Artery Risk Development Study in Young Adults (CARDIA), the Framingham Heart Study (FHS), the Genetic Epidemiology Network of Arteriopathy (GENOA), the Multi-Ethnic Study of Atherosclerosis (MESA), and the Rotterdam Study (RS) (Supplementary Table 1). All cohorts excluded participants who consumed a high amount of alcohol, equivalent to ≥21 drinks/week in men or ≥14 drinks/week in women (8). Cohort-specific exclusion is detailed in Supplementary Data. Due to potential differences in DNA methylation patterns between different ethnicities (9), we analyzed the association between DNA methylation and hepatic fat separately in EA (n = 3,400), HA (n = 401), and AA (n = 724) participants. The protocol for each study was approved by the institutional review board in each cohort. All participants provided written informed consent.

The study design flowchart is presented in Fig. 1. We first conducted the epigenome-wide association studies of hepatic fat among EA participants, including both discovery and replication. We then examined differential DNA methylation in relation to hepatic fat in the HA and AA participants. We further examined the functional and regulatory annotations for the replicated CpGs and tested the potential associations of the identified CpGs with NAFLD and T2D.

A detailed description for hepatic fat assessment in each cohort is presented in Supplementary Data. The RS used ultrasound to estimate hepatic fat and diagnosed steatosis on a dichotomized scale. The other cohorts used computed tomography (CT) to quantify hepatic fat on a continuous scale by using either mean Hounsfield units of the liver image or the ratio of the Hounsfield units of the liver image to that of a control.

Methylation profiles were measured by Illumina BeadChip using DNA derived from all leukocytes or CD14+ monocytes in peripheral blood (Supplementary Table 1). Details for DNA preparation, bisulfite conversion, methylation profiling, quality control procedures, and raw data normalization in each cohort are provided in Supplementary Data. We analyzed either methylation signal M values (in CARDIA [calculated as logit transformation of β values]) or β values (in all other cohorts [calculated as methylated signals divided by the sum of methylated and unmethylated signals]). Nonautosomal probes were excluded from the current study.

We conducted the discovery epigenome-wide association study (EWAS) in FHS and interrogated the differentially methylated CpGs at false discovery rate (FDR) <0.05 in the replication cohorts (EA samples in CARDIA, MESA, and RS). Linear regression models or mixed models with consideration of family structures were conducted to examine directionality and calculate P values in each cohort. Because hepatic fat was measured using different scales, we meta-analyzed the P values in the replication cohorts using logit method based on the general fixed effect model. The statistical significance in the replication analysis was determined using the Bonferroni-corrected P value threshold, defined as 0.05 divided by the number of significant CpGs in the discovery phase. We used sex- and age-adjusted models (model 1) in the discovery and replication analyses. Estimated leukocyte composition (10) and technical variables were also adjusted for in a cohort-specific manner (Supplementary Data). Included in the sensitivity analyses, we conducted global meta-analyses in all EA participants to examine the impact of potential confounders. We performed the same sex- and age-adjusted model (model 1). We additionally adjusted for lifestyle factors including smoking status, physical activity levels, and alcohol intake in model 2. We further adjusted for BMI in model 3. We also performed a discovery and replication analysis using model 3 in EA participants. Covariates are assessed using cohort-specific methods (Supplementary Data). We conducted similar EWAS with adjustment for the same covariates to identify hepatic fat–related CpGs in MESA HA participants and AA participants in CARDIA, GENOA, and MESA. We calculated adjusted R² using multiple linear regression models with and without adjustment for the replicated CpGs and used the difference of the two adjusted R² values to represent the interindividual variation of hepatic fat from the replicated CpGs.

In FHS, we examined the associations between DNA methylation, hepatic fat, and gene expression. To prioritize genes in these analyses, we selected Illumina-annotated genes. For CpGs without annotated genes, we identified a set of genes by overlapping cis-meQTLs (methylation quantitative trait loci ±500 kilobases [kb] from CpG) with cis-eQTLs (expression quantitative trait loci residing within 500 kb of a nearby gene) at P value <5 × 10⁻⁷. The association between DNA methylation and gene expression was analyzed in 4,561 participants in the FHS as previously described (11). For genes significantly associated with CpGs (P value <5 × 10⁻⁷), we further examined their association with hepatic fat using similar statistical procedures in 2,317 FHS participants. For genes that associated with both CpGs and hepatic fat, we conducted mediation tests to estimate the proportion of mediation by gene expression on the association of CpGs and hepatic fat. We used the R package mediation with sex- and age-adjusted linear mixed models as described above and used the quasi-Bayesian Monte Carlo method with 1,000 simulations to calculate CIs (12).

Owing to the well-documented association between NAFLD and T2D (13), we conducted cross-sectional and prospective analyses between replicated CpGs and T2D. We defined T2D as fasting glucose ≥7.0 mmol/L, HbA_1c ≥6.5% (48 mmol/mol), or current use of insulin or an oral hypoglycemic drug. In the cross-sectional analysis, we analyzed 4,068 FHS participants who attended the eighth Offspring cohort examination (2005–2008) or the second examination of the Third Generation cohort (2008–2011). In the prospective analysis, we analyzed 1,764 FHS participants who attended both the eighth and ninth (2011–2014) Offspring cohort examinations. In the prospective analysis, prevalent T2D cases at baseline were excluded. The first model in the cross-sectional analysis adjusted for sex, age, alcohol intake, smoking status, physical activity level, DNA methylation measurement laboratory and technical variables, top three principal components for DNA methylation measurement, and estimated leukocyte composition. The first model in the prospectively analysis adjusted for the same covariates and baseline fasting glucose and HbA_1c. In the second model for both analyses, we additionally adjusted for BMI. We used generalized estimating equations to estimate the odds ratio (OR) after accounting for the family structure in this study sample.

We conducted Mendelian randomization (MR) analyses (Supplementary Fig. 1) to test for association of the replicated CpGs with NAFLD. We also examined the association between these CpGs and T2D. We performed MR analysis using a two-sample MR approach (14). We used independent cis-meQTLs (n = 4,170), with pairwise linkage disequilibrium r² <0.1 and F statistics >10, as instrumental variables (IVs). Using the TwoSampleMR R package, we performed the primary analysis using the inverse variance weighted (IVW) method and sensitivity analysis using the MR-Egger method when association was significant using IVW. The effect sizes and SEs for IVs-CpG were obtained from the FHS and the effect size and SEs were obtained for IVs-NAFLD and for IVs-T2D from published GWAS (4,15).

We conducted hypergeometric tests with Bonferroni correction to compare the genomic characteristics of the replicated CpGs with the whole set of CpGs that we analyzed and calculated the probability for these characteristics, e.g., whether CpGs were more likely to reside in a CpG island or gene region, using the Infinium HumanMethylation450 BeadChip annotation files. We queried cis-meQTLs in the platform of Functional Mapping and Annotation of Genome-Wide Association Studies (FUMA GWAS) (16). Using this platform, we examined the overlap between cis-meQTLs with signals in the NHGRI-EBI Catalog of published GWAS (17). We also studied genes using differentially expressed genes measured in human whole blood and liver samples in the Genotype-Tissue Expression (GTEx v6) database (18) and visualized the genomic region for cis-meQTLs. To assess the relevance of the identified peripheral blood–derived CpGs in liver, we compared DNA methylation levels in whole blood with that in liver using data deposited in the Gene Expression Omnibus (GEO Series accession number GSE48472) (19). Gene Ontology (GO) biological process enrichment analysis was performed using the GO Consortium website (http://www.geneontology.org/).

The data sets analyzed in the current study are available at https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000285.v3.p2 (CARDIA), https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000007.v29.p10 (FHS), https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs001238.v1.p1 (GENOA), and https://www.ncbi.nlm.nih.gov/projects/gap/cgi-bin/study.cgi?study_id=phs000209.v13.p3 (MESA). Data can be obtained upon request. Requests should be directed toward the management team of the RS (secretariat.epi@erasmusmc.nl), which has a protocol for approving data requests. Because of restrictions based on privacy regulations and informed consent of the participants, data cannot be made freely available in a public repository (RS).

Of the 400,129 CpGs analyzed in age- and sex-adjusted models, 58 CpGs were significantly associated with hepatic fat in the FHS discovery cohort (n = 1,496) at FDR <0.05 (corresponding P value = 6.9 × 10⁻⁶ [Manhattan plot in Fig. 2 and Supplementary Fig. 2, quantile-quantile plot in Supplementary Fig. 3 and Supplementary Table 2). The t statistics for the 58 CpGs were correlated between the FHS and each of the replication cohorts (Supplementary Fig. 4). The associations between the 58 CpGs and hepatic fat remained materially unchanged following additional adjustment for the time gap between assessments of DNA methylation and hepatic fat in FHS (Supplementary Fig. 5). In CARDIA, hepatic fat was measured at the year 25 examination. The t statistics calculated between hepatic fat and DNA methylation measured at the year 25 examination were highly correlated with those calculated using DNA methylation measured at the year 15 examination (Supplementary Fig. 5). Twenty-four (41%) CpGs replicated (Bonferroni-corrected P value <0.05/58 = 8.6 × 10⁻⁴) in the meta-analysis of the replication cohorts (n = 1,904) (Table 1). We removed two CpGs from the replication analysis because they were highly correlated (|r| ≥ 0.7) and close to other CpGs with lower P values in the same chromosome (Supplementary Table 3). Forest plots for all 22 CpGs using regression coefficients and SEs (standardized by the cohort-specific SD of the regression coefficients) are shown in Supplementary Fig. 6.

In the global meta-analysis of all EA participants, additional adjustment for lifestyle factors did not materially change the association between DNA methylation levels and hepatic fat (Fig. 3). Further adjustment for BMI, which is correlated with hepatic fat (Spearman r = 0.45 in FHS), reduced the strength of the associations (Fig. 3); however, all 22 CpGs remained nominally associated with hepatic fat (P value <0.05) (Table 1). Additionally, two CpGs, cg06690548 (SLC7A11) and cg19693031 (TXNIP), remained significant in FHS at FDR <0.05 (corresponding P value = 1.2 × 10⁻⁷) (Supplementary Table 4) and in the replication samples (Bonferroni-corrected P value <0.05). Both CpGs were among the replicated CpGs in the sex- and age-adjusted analysis. Leave-one-cohort-out analysis in EA participants showed that P values in the global meta-analysis with exclusion of one cohort were highly correlated with those in all samples, with r ranging from 0.83 to 0.87 (Supplementary Fig. 7). We further examined whether lipid traits, serum fasting triglycerides (TGs), and total cholesterol concentrations (TC) affect the associations of the 22 CpGs with hepatic fat. As shown in Supplementary Fig. 8, additional adjustment for TG but not TC attenuated the associations for most of CpGs in FHS. Additionally, after Bonferroni correction (corresponding P value <0.002), TGs significantly mediated the association with hepatic fat for 12 CpGs, with a mediation percent ranging from 13 to 39% (Supplementary Table 5). However, the association between all 22 CpGs and hepatic fat remained significant after additional adjustment for TG (all P values <0.002) (Supplementary Fig. 8). Additional adjustment for glycemic traits (T2D, fasting glucose, and HbA_1c) attenuated the associations between the 22 CpGs and hepatic fat; however, all associations remained significant (Supplementary Fig. 8). We found that adjustment for caloric intake and total dietary fat intake did not materially alter the association between the 22 CpGs and hepatic fat (Supplementary Fig. 9).

In CARDIA, the 22 replicated CpGs captured 10% of interindividual variation (i.e., the adjusted R²) in hepatic fat after we accounted for sex and age (P value = 1.4 × 10⁻⁷) and 4.3% of interindividual variation after we additionally accounted for lifestyle factors and BMI (P value = 0.001). The proportion of variation explained by the 22 replicated CpGs was similar to that observed in the discovery cohort (FHS): 14.9% and 5.8%, respectively. With additional adjustment for serum alanine transaminase, serum aspartate transaminase, and a genetic risk score for NAFLD, the 22 replicated CpGs explained 4.6% of interindividual variation in hepatic fat in FHS (P value = 1.4 × 10⁻¹²).

For the 22 CpGs that replicated in the EA participants, cg19693031 (TXNIP) remained significant in HA participants after Bonferroni correction (P value <2.3 × 10⁻³) (Supplementary Table 6). Additionally, four CpGs were nominally significant in HA participants (P value <0.05) (Supplementary Table 6) and three CpGs were nominally significant in the meta-analysis of AA participants (P < 0.05) (Supplementary Table 7). In epigenome-wide analysis, no CpG was detected at FDR <0.05 in HA participants. We discovered 26 CpGs at FDR <0.05 (corresponding P value = 2.7 × 10⁻⁶) (Supplementary Table 8) in the meta-analysis of epigenome-wide studies in AA participants, of which two CpGs were nominally significant (P value < 0.05) in the global meta-analysis of EA participants.

The mean whole blood DNA methylation levels of the 22 replicated CpGs in EA participants were moderately correlated with those measured in liver tissue (19) (r = 0.59) (Supplementary Fig. 10). These CpGs were more likely to reside in the south shore (0–2 kb downstream of CpG island; P value = 5.5 × 10⁻⁴) or south shelf (2–4 kb downstream of CpG islands; P value = 4.1 × 10⁻⁴), in DNase I hypersensitivity sites (P value = 1.7 × 10⁻³), in reprogramming-specific differentially methylated regions (P value = 3.8 × 10⁻⁴), and in gene body regions (P value = 2.4 × 10⁻⁴).

Of the 22 CpGs, 19 CpGs have been annotated to 18 unique genes (Table 1 and Supplementary Table 9). Based on liver-specific differentially expressed genes in GTEx (18), the 18 annotated genes were enriched with genes that are upregulated in the liver, including DHCR24, SLC43A1, CPT1A, SREBF1, SC4MOL, and SLC9A3R1 (Bonferroni-corrected P value = 0.005) (Supplementary Table 10). GO analysis showed enrichment for 18 biological processes (Supplementary Table 11). The most significant enriched pathway was positive regulation of the cholesterol biosynthesis (GO:0045542; ABCG1 and SREBF1; >100-fold enrichment; FDR adjusted P value = 0.02).

By overlapping cis-meQTL variants with GWAS results in the NHGRI-EBI Catalog of GWAS (17), we found that cis-meQTLs or their proxies (linkage disequilibrium R² >0.8) for nine CpGs were associated with 26 unique traits in GWAS (Supplementary Table 12). For example, rs637868 for cg14476101 (PHDGH) was associated in GWAS with alanine aminotransferase levels (20) and rs2834288 for cg08309687 (LINC00649) was associated in GWAS with abundance of gut microbiota (21).

In the FHS, 7 of the 22 replicated CpGs were associated with peripheral blood–derived expression of six annotated genes (e.g., cg17901584 with DHCR24) at a P value threshold of <5 × 10⁻⁷ (Supplementary Table 13). Additionally, cg17501210 was associated with expression of one nonannotated cis-gene, RNASET2 (transcription start site residing 301 kb downstream from cg17501210; P value = 9.4 × 10⁻¹¹). Among these seven genes, expression levels of ABCG1 and CPT1A were significantly associated with liver fat (P value = 1.2 × 10⁻³⁰ and 2.0 × 10⁻¹⁷, respectively). Expression of ABCG1 and CPT1A mediated the association between corresponding CpGs and hepatic fat by ∼20% and 10%, respectively (Fig. 4).

We identified eight cis-genes for cg08309687 (LINC00649) by overlapping their cis-meQTLs (P value <5 × 10⁻⁷) with cis-eQTLs (P value <5 × 10⁻⁷) (regional plot in Supplementary Fig. 11). None of the CpG was associated with gene expression at the predefined threshold (P value <5 × 10⁻⁷). Among the five nominally significant genes (P value <0.05) (Supplementary Table 13), expression of TMEM50B mediated 8% (95% CI 2, 16; P value <2.2 × 10⁻¹⁶) of the association of cg08309687 with hepatic fat (Fig. 4).

The IVW analysis (Supplementary Table 14) showed that hypomethylation at cg08309687 (LINC00649) was significantly associated with NAFLD (P value = 2.5 × 10⁻⁴) (Fig. 5). Hypomethylation at cg14476101 (PHDGH) was nominally associated with NAFLD (P value = 0.01) (Supplementary Fig. 12). Neither CpG was significant in the sensitivity analysis using the MR-Egger method (P value = 0.17 and 0.97, respectively). No horizontal pleiotropy effect was detected (P value = 0.66 and 0.15).

In the FHS, 18 of the 22 replicated CpGs in EA participants were cross-sectionally associated with T2D after Bonferroni correction (P value <0.002) (Supplementary Table 15). In addition, 6 of the 18 CpGs were prospectively associated with incident T2D (P value <0.05) (Table 2). In analysis with additional adjustment for BMI, associations remained significant (P value <0.05) for 13 CpGs in the cross-sectional analysis (Supplementary Table 15) and for 4 CpGs in the prospective analysis (Table 2). For all significant CpG-T2D pairs, the direction of the association was concordant with the expected direction; e.g., hypermethylation at cg08309687 (LINC00649) was expected to be associated with decreased risk of T2D because hypermethylation at this locus was associated with lower hepatic fat. In concordance with the prospective analysis using a BMI-adjusted model, MR analysis showed consistent results for cg21429551 (GARS); i.e., hypermethylation at this locus was nominally associated with lower risk of T2D (P value = 0.03) (Supplementary Table 16).

We found that differential methylation of peripheral blood–derived DNA at 22 CpG sites was associated with hepatic fat in EA participants. These CpGs reside at several loci regulating key biological processes relevant to the development of steatosis. The 22 CpGs explained ∼10% of interindividual variation in the sex- and age-adjusted model, which was larger than that explained by single nucleotide polymorphisms identified from GWAS (4). In addition, the current study associated several hepatic fat–associated CpGs with increased risk of new-onset T2D; e.g., hypomethylation at cg08309687 was associated with increased hepatic fat and higher risk of T2D.

Previous studies using array-based methods have examined epigenome-wide DNA methylation patterns in liver samples of individuals with biopsy-proven NAFLD (5,6). While these studies showed a strong contrast between DNA methylation profiles of individuals with nonalcoholic steatohepatitis compared with control subjects (i.e., individuals without steatosis or steatohepatitis), differences in DNA methylation patterns of individuals with steatosis (fatty liver alone) versus control subjects were less obvious (5,6). In contrast to prior studies using liver biopsies, we used noninvasive imaging to assess hepatic fat. We therefore had a much larger sample size and statistical power to detect epigenetic signals associated with elevated hepatic fat.

Two recent large EWAS identified CpGs associated with BMI (11,22). The majority (17 of the 22) of the replicated hepatic fat–associated CpGs in the current study were also observed in at least one of the two large-scale BMI-DNA methylation studies. Several CpGs associated with both liver fat and BMI are annotated to key genes involved in lipid metabolism pathways, namely, SREBF1, CPT1A, ABCG1, and DHCR24 (23,24). This is concordant with our observation that TGs mediated the associations for half of the replicated CpGs. These genes are in a top gene network associated with general adiposity identified in a previous analysis in MESA (25). Therefore, the overlap between hepatic fat–associated CpGs and those linked to BMI supports a key role of adiposity and adiposity-related pathways in the pathogenesis of steatosis.

Our MR analyses implicated cg08309687 as a putatively causal factor for NAFLD. This CpG is located in a long intergenic noncoding region (LINC00649) that may play a role in transcription regulation relevant to steatosis. We could not examine whether LINC00649 mediated the observed association between cg08309687 and hepatic fat because we lacked expression data for LINC00649 in our study cohorts. However, we observed that cis-meQTL variants for cg08309687 coincide with cis-eQTLs of several nearby genes. We also established three-way associations of cg08309687 methylation, hepatic fat, and gene expression levels for a nearby gene, TMEM50B (Fig. 3). Together with data from GTEx, our analysis indicates that DNA methylation at cg08309687 may affect LINC00649 and alter expression of nearby genes (e.g., TMEM50B) and impact hepatic fat accumulation. In addition, our data suggest that cg08309687 may be involved in the regulation of the gut microbiome, which has been postulated to play a critical role in the development of NAFLD (26). Future studies are needed to explore pathways underlying the observed association between DNA hypomethylation at cg08309687 and NAFLD.

Using models with additional adjustment for BMI, we showed that two CpGs, cg06690548 (SLC7A11) and cg19693031 (TXNIP), are independently associated with hepatic fat in the EA population. Although cg19693031 (TXNIP) has not been previously linked to obesity-related traits, several studies have shown that hypomethylation at cg19693031 (TXNIP) was associated with increased risk of T2D (27). As NAFLD is strongly associated with T2D (13), these observations are in accordance with our finding that DNA hypomethylation at this locus is associated with increased hepatic fat.

The present analyses included participants from multiple ancestries, and differences in DNA methylation patterns associated with hepatic fat were observed between ethnic groups. Such variability is consistent with other observations that DNA methylation levels differ by race and/or ethnicity (9). However, the sample size in the current study may not be sufficient for robust comparisons of EA with AA and HA participants. The MR analyses in the current study may be limited for a few reasons. First, the effect sizes and SEs for the IVs were estimated using relatively small-scale GWAS. Second, effect sizes and SEs for instrument-exposure and instrument-outcome associations were estimated in partially overlapping study samples, which may lead to instrument bias (28). Moreover, one participating cohort (RS) used ultrasound, which could not be used to quantify liver fat on a continuous scale as other participating cohorts did. This difference may partly explain the inconsistency in the direction of regression coefficients for some of the replicated CpGs in RS compared with the other cohorts. Further, because not all cohorts measured liver fat using the same scales, we meta-analyzed P values rather than regression coefficients and SEs and could not estimate the overall effect size for the association of DNA methylation and hepatic fat accumulation. However, we observed that DNA methylation signals were correlated among the cohorts of EA participants. As demonstrated by accounting for the time gap between assessments of DNA methylation and hepatic fat in FHS and CARDIA, DNA methylation signals were relatively stable over time (Supplementary Fig. 5). In addition, CpGs measured in monocytes were correlated well with those measured using whole blood samples (i.e., all leukocytes). This finding is consistent with observations from one prior BMI-DNA methylation study in which the observed associations between DNA methylation and BMI were shared across different cell subsets (22).

In conclusion, the current study demonstrates a unique DNA methylation pattern related to hepatic fat in EA participants. In addition, our study showed that several hepatic fat–associated CpGs may be useful biomarkers for prediction of new-onset T2D. Future studies with larger and more ethnically diverse sample sizes are needed to validate our findings and to explore the biological relevance of peripheral DNA methylation in the development and progression of NAFLD and T2D.

Acknowledgments. The authors thank Michael Verbiest, Mila Jhamai, Sarah Higgins, Marijn Verkerk, and Lisette Stolk (Erasmus Medical Center and Erasmus University, Rotterdam, the Netherlands) for their help in creating the methylation database. The authors are grateful to the study participants, the staff from the RS, and the participating general practitioners and pharmacists. The authors also thank the families that participated in GENOA.

Funding. Infrastructure for the CHARGE (Cohorts for Heart and Aging Research in Genomic Epidemiology) Consortium is supported in part by the National Heart, Lung, and Blood Institute (NHLBI), National Institutes of Health (NIH), grant R01-HL-105756. This work was supported in part by the Intramural Research Program of the NIH: NHLBI; National Institute on Aging (NIA); and National Institute of Environmental Health Sciences. The FHS is funded by NIH contract N01-HC-25195. The laboratory work for this investigation was funded by the Division of Intramural Research, NHLBI, NIH, Bethesda, MD. The analytical component of this project was funded by the Division of Intramural Research, NHLBI, and the Center for Information Technology, NIH. The generation and management of the Illumina HumanMethylation450K BeadChip array data (EWAS data) for the RS was executed by the Human Genotyping Facility of the Genetic Laboratory of the Department of Internal Medicine, Erasmus University Medical Center. The EWAS data were funded by the Genetic Laboratory of the Department of Internal Medicine, Erasmus University Medical Center, and by the Netherlands Organization for Scientific Research (NWO) (project number 184021007) and made available as a Rainbow Project (RP3 [Biobank-based Integrative Omics Study (BIOS)]) of the Biobanking and Biomolecular Research Infrastructure Netherlands (BBMRI-NL). The RS is funded by Erasmus University Medical Center and Erasmus University, Rotterdam; Netherlands Organization for Health Research and Development (ZonMw); the Research Institute for Diseases in the Elderly; the Ministry of Education, Culture and Science; the Ministry for Health, Welfare and Sports; the European Commission (Directorate-General XII); and the Municipality of Rotterdam. MESA and the MESA SHARe (SNP Health Association Resource) project are conducted and supported by the NHLBI in collaboration with MESA investigators. Support for MESA is provided by contracts NIH N01-HC-95159, N01-HC-95160, N01-HC-95161, N01-HC-95162, N01-HC-95163, N01-HC-95164, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168, N01-HC-95169, UL1-TR-001079, UL1-TR-000040, and DK063491. The MESA Epigenomics and Transcriptomics Study was funded by NIA grant 1R01HL101250-01 to Wake Forest University Health Sciences. CARDIA is supported by contracts HHSN268201300025C, HHSN268201300026C, HHSN268201300027C, HHSN268201300028C, HHSN268201300029C, and HHSN268200900041C and grant R01-HL098445 for the year 25 CT exam (J.J.C.) from the NHLBI. The laboratory work and analytical component were funded by the American Heart Association (17SFRN33700278 and 14SFRN20790000 [L.H.]). L.B.V. is supported by NHLBI grant K23-HL-136891. Support for GENOA was provided by the NHLBI of the NIH (HL-054464, HL-054457, HL-054481, HL-100185, HL-119443, HL-085571, and HL-133221).

The views expressed in this manuscript are those of the authors and do not necessarily represent the views of the NHLBI, the NIH, or the U.S. Department of Health and Human Services. All authors have completed the Unified Competing Interest form (available on request from the corresponding author) and declare no support from any organization for the submitted work, no financial relationships with any organizations that might have an interest in the submitted work in the previous 3 years, and no other relationships or activities that could appear to have influenced the submitted work.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. J.M. and D.L. designed the analysis. J.M., J.N., J.D., Y.Z., T.H., R.J., and L.B. analyzed data. C.L. provided statistical support. J.M., J.N., J.D., Y.Z., and J.A.S. wrote the manuscript. R.H., E.K.S., C.S., M.M.M., M.T.L., L.L., L.M.R., M.G., T.M., J.B.J.v.M., L.J.M.A., O.H.F., A.D., S.R., W.Z., S.L.R.K., P.A.P., H.N., L.B.V., D.M.L.-J., J.J.C., P.G., A.H.L., F.B.H., Y.L., L.H., S.D.M., and D.L. provided critical editorial comments. D.L. had primary responsibility for the final content. All authors read and approved the final manuscript. D.L. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.