Epigenome-Wide Association Study in Asian Cohort Identifies Novel DNA Methylation Markers for Carotid Intima-Media Thickness

Figures

• 

  Figure 1:. Study design— The analytical workflow integrates Asian-focused discovery with trans-ancestry validation. The Asian-focused screening stage conducted meta-analysis across ~850,000 CpG sites across three major Asian ethnic groups—Chinese, Malay, and Indian participants—from the HELIOS cohort (n=1,357). Signals reaching P<0.05 were advanced to trans-ancestry meta-analysis with European cohorts for validation. A Bonferroni-corrected significance threshold (P<9.35E-07, 0.05/53,470 CpGs tested in trans-ancestry stage) was applied to select three robust associations (‘Sentinel CpGs’) for targeted interrogation through multi-omics integration (meQTL, eQTL, eQTM analyses) and causal inference methods to examine evidence for methylation-mediated effects on cIMT, CVD and gene expression. An MRS was constructed from the three validated methylation markers and evaluated for its association with cIMT elevation above the 75th percentile based on Asian normative ranges. Abbreviations: CHN, Chinese; cIMT, carotid intima-media thickness; DNAm, DNA methylation; eQTL, expression quantitative trait loci analysis; eQTM, expression quantitative trait methylation analysis; IND, Indian; meQTL,…

• 

  Figure 2:. Associations between sentinel CpG sites and cIMT— (a) Manhattan plot of cIMT associations. Associations for the 53,470 CpGs included in the targeted trans-ancestry meta-analysis are shown. Sentinel CpGs (Bonferroni-corrected P<9.35E-07) are highlighted and annotated with proximal: genes: nearest gene (n) and top cis-eQTM-associated gene (e; PeQTM <0.05). (b-d) Forest plots of sentinel CpG associations in (b) trans-ancestry, (c) Asian-specific, and (d) European-specific meta-analyses, respectively. For the European-specific meta-analysis, LBC1936 cohort (measured on 450K array) was not included since all sentinel CpGs were unique to the EPIC array. Data point sizes correspond to marker-specific effective sample sizes that are normalised to sum to one across all analysed groups. Test statistics from Stouffer’s sample-size weighted meta-analysis approach (Zscores, P) are indicated. Heterogeneity between meta-analysed groups was assessed using Cochran’s Q test and quantified by I2 (HetISq), with p-values indicating significance. Abbreviations: cIMT, carotid intima-media thickness; eQTM, expression quantitative trait methylation.

• 

  Figure 3:. Association between cg05575921 and cIMT— The cg05575921-cIMT association was previously reported at epigenome-wide significance in a meta-analysis of European cohorts (n=6,400, Portilla-Fernández et al., 2021). (a-c) Forest plots of the cg05575921-cIMT association in the (a) trans-ancestry meta-analysis, (b) Asian-specific analysis and (c) European-specific analysis from the current investigation. The association reached P<0.05 only in the European-specific analysis, but not in the Asian-specific analysis. Data point sizes correspond to normalised effective sample size. Abbreviations: cIMT, carotid intima-media thickness

• 

  Figure 4 :. Overrepresented gene sets among genes mapped to cIMT-linked CpGs— Genes mapped to cIMT-linked CpGs were analysed for pathway enrichment against a background of genes mapped to 200 permutations of EPIC array CpGs of matched methylation level and variability to the cIMT-linked CpGs. CpG-gene mapping was done via cis-eQTM analysis (PeQTM<0.05). Enriched gene sets (P<0.005, log2 fold change >1) are displayed.

• 

  Figure 5:. Evidence for cg08227773 methylation-mediated effects on CAD risk— (a) Schematic overview of the SMR experiment. cis-meQTL data (exposure) were derived from Asian cohorts in SG10K_Health (n=5,273). CAD GWAS summary statistics (outcome) were obtained from Biobank Japan (n=212,453 including 29,139 cases and 183,134 controls). The instrumental SNP is chr3:46971962:A:T (PmeQTL=2.79E-57, F-statistic=306.25). (b) Regional association plots displaying genetic associations for cg08227773 methylation and CAD. Plots display cg08227773 meQTL (top) and CAD GWAS (bottom) signals within a ±1Mb window centered on cg08227773. The ±1Mb window size reflects the cis-region used for meQTL mapping in the SG10K_Health cohort. Colocalisation analysis supported shared causal variant (PP.H4=0.91). SNPs are colored according to their linkage disequilibrium (r2) with the instrumental SNP, calculated using the full SG10K_Health cohort with available genetic data (N=9,766). Gene tracks show locations of protein-coding genes based on Ensembl Release 86 (hg38). Abbreviations: BBJ, Biobank Japan; CAD, coronary artery disease; CHN, Chinese; IND, Indian; JPN, Japanese; MLY, Malay; meQTL, methylation quantitative…

• 

  Figure 6:. Evidence for cg08227773 methylation-mediated effects on NBEAL2 expression— (A) Schematic overview of the SMR experiment. cis-meQTL data (exposure) were derived from Asian cohorts in SG10K_Health (n=5,273). Blood-based cis-eQTL data (outcome) were obtained from the Asian HELIOS cohort (n=1,168) for primary analyses. The instrumental SNP for SMR is chr3:46971962:A:T (PmeQTL=2.79E-57, F-statistic=306.25). (B) Regional association plots displaying genetic associations for cg08227773 methylation and NBEAL2 expression. Plots display cg08227773 meQTL (top) and NBEAL2 eQTL (bottom) signals within a ±1Mb window centered on cg0822773. The ±1Mb window size reflects the cis-region used for meQTL mapping in the SG10K_Health cohort. Colocalisation analysis supported a shared causal variant (PP.H4=0.69). SNPs are colored according to their linkage disequilibrium (r2) with the instrumental SNP, calculated using the full SG10K_Health cohort with available genetic data (N=9,766). Gene tracks show locations of protein-coding genes based on Ensembl Release 86 (hg38). Abbreviations: CHN, Chinese; IND, Indian; MLY, Malay.

• 

  Figure 7. MRS of cIMT predicts clinically meaningful cIMT elevation. The MRS was built from sentinel CpGs, weighting each CpG by its Stouffer-combined z-score estimated in the training set (model: log-transformed cIMTmean ~sentinel CpG methylation + age + sex). Models were fit separately by ethnicity, and the perethnicity association z-scores were combined using Stouffer’s method to obtain the final CpG weight. (a) Distribution of MRS by cIMT status. MRS values were normalised within each group for visualisation purposes. P-values are from two-sample t-tests using the original (unnormalised) scores. (b) Discriminatory ability of MRS. ROC analysis compares models with and without the MRS; AUCs are plotted (c) Stratification by MRS quartiles. Participants were grouped by MRS quartile; relative odds of clinically concerning cIMT elevation are reported relative to Q1 (lowest MRS). Odds ratios and 95% CI were obtained from logistic regression (cIMTelevated ~ MRS quartile + Age + Sex + Ethnicity) by exponentiating the MRS quartile coefficients. Abbreviations: AUC, Area Under Curve; CHN, Chinese; IND, Indian; MLY, Malay; MRS, methylation risk score; ROC, Receiver Operating Characteristic