- Quantification and statistical analysis of DNA methylation
- Enrichment analysis of genomic regulatory features
- Expression quantitative trait methylation analysis
- Causal analyses (Mendelian Randomisation and Colocalisation)
- Weighted Gene Correlation Network Analysis (WGCNA)
- Construction of Methylation Risk Score (MRS)
We ran EWAS of DCM in two independent cohorts to identify conditionally-independent sentinel CpGs (P<5.96E-08) with confirmed consistent directionality of association. Within the multi-ancestry cohort (MAGNet), ancestry-specific EWAS was performed followed by inverse variance-weighted meta-analysis using METAL. A 2-stage correction of inflation in effect sizes and standard errors was applied: once to ancestry-specific results prior to meta-analysis, and once after meta-analysis. Candidate sentinel CpGs were identified (Bonferroni P<0.05), following which sentinel CpGs were identified by assignment into distinct genomic loci as well as conditional analysis.
The main EWAS model is
DCM ~beta(quantile-normalised) + Age + Gender + principal components of control probes capturing ~95% of control probe variation
The scripts in Quantification and statistical analysis of DNA methylation facilitate various tasks in methylation data preprocessing and quality control, calculation of methylation betas, epigenome-wide association analysis, meta-analysis, adjustment for test-statistic inflation and finally the identification of sentinel CpGs.
We analysed sentinel CpGs for enrichment in various genomic regulatory features using a permutation testing approach. Specifically, sentinel CpG overlap of genomic regulatory features was compared to a background set comprising permutations of EPIC array CpGs matched by methylation levels and variability. This addresses bias inherent in methylation arrays, which preferentially assay pre-determined genomic sites and well-annotated genes.
The scripts in Enrichment analysis of genomic regulatory features were utilized to create background CpG sets, perform enrichment analysis, and determine significance at a permutation p-value threshold of p<0.001.
We identified significant associations between sentinel CpG methylation and proximal gene expression (+/- 1Mb) of the sentinel CpG. Replicated eQTMs were defined as significant eQTMs in the discovery cohort (MAGNet; disc FDR P<0.05) for which replication testing (BMCB) confirmed consistent directionality of association. Association testing was performed using the R package MatrixEQTL using the following model:
gene expression ~ methylation + Age + Gender + ancestry + RNA Integrity Number(RIN) + 5 PEER factors
PEER factors correspond to hidden sources of variation, learned by Bayesian probabilistic estimation of residual factors to accurately capture and account for latent influences in the data, thereby enhancing the robustness of statistical models. The choice of number of PEER factors to learn was informed by the GTex eQTL mapping study: for N < 150, use 15 PEERs; 150 <= N < 250, use 30 PEERs; N >= 250, use 35 PEERs.
The scripts in Expression quantitative trait methylation analysis were used for learning PEER factors from gene expression data and implementing linear models to identify eQTMs (MatrixEQTL).
We elucidated the putative causal contribution of sentinel CpG methylation to DCM and proximal gene expression using SMR. Significant associations were validated using one-sample MR and underwent colocalisation analysis to assess the posterior probability of a shared causal variant underlying the association between the two assessed traits
The primary SMR analysis utilizes SNPs linked to sentinel CpG methylation (known as 'methylation quantitative trait loci') derived from MAGNet left ventricular samples in the current investigation, as well as publicly-available genetic associations from the following sources:
Individual-level genotype, methylation and expression/disease status data from the MAGNet cohort were used for one-sample MR analyses.
The scripts in Causal analyses detail the primary SMR analysis, one-sample MR validation of SMR-significant hits, and colocalisation analysis of genetic associations for the assessed traits.
We employed WGCNA to detect co-methylation modules amongst DCM-associated CpGs (discovery EWAS FDR<0.05, directionality of association confirmed in replication cohort, SD>0.02).
The scripts in Co-methylation analyses identify common co-methylation patterns by comparing ancestry-specific co-methylation networks.
Following fine-mapping of 28 DCM sentinel CpGs and testing individual CpGs within fine-mapped regions for CVD trait associations, we constructed CVD-trait specific MRS as a weighted sum of methylation betas from CpGs within fine-mapped regions and tested MRS for association with CVD.
The scripts in MRS detail the construction of region and trait-specific MRS and testing their associations with the respective CVD traits.