Epigenome-wide association study for dilated cardiomyopathy in left ventricular heart tissue identifies putative gene sets associated with cardiac pathology and early indicators of cardiac risk

Abstract

Background: Methylation changes linked to dilated cardiomyopathy (DCM) affect cardiac gene expression. We investigate DCM mechanisms regulated by CpG methylation using multi-omics and causal analyses in the largest cohort of left ventricular tissues available.

Methods: We mapped DNA methylation at ~ 850,000 CpG sites, performed array-based genotyping and conducted RNA sequencing on left ventricular tissue samples from failing and non-failing hearts across two independent DCM cohorts (discovery n = 329, replication n = 85). Summary-data-based Mendelian Randomisation (SMR) was applied to explore the causal contribution of sentinel CpGs to DCM. Fine-mapping of regions surrounding sentinel CpGs revealed additional signals for cardiovascular disease risk factors. Coordinated changes across multiple CpG sites were examined using weighted gene co-expression network analysis (WGCNA).

Results: We identified 194 epigenome-wide significant CpGs associated with DCM (discovery P < 5.96E-08), enriched in active chromatin states in heart tissue. Amongst these, 32 sentinel CpGs significantly influenced the expression of 30 unique proximal genes (± 1 Mb). SMR suggested the causal contribution of two sentinel CpGs to DCM and two other sentinel CpGs to the expression of two unique proximal genes (P < 0.05). For one sentinel CpG, colocalisation analyses provided suggestive evidence for a single causal variant underlying the methylation-gene expression relationship. Fine-mapping revealed additional signals linked to cardiovascular disease-relevant traits, including creatinine levels and the Framingham Risk Score. Co-methylation modules were enriched in gene sets and transcriptional regulators related to cardiac physiological and pathological processes, as well as in transcriptional regulators whose cardiac relevance has yet to be determined.

Conclusions: Using the largest series of left ventricular tissue to date, this study investigates the causal role of cardiac methylation changes in DCM and suggests targets for experimental studies to probe DCM pathogenesis.

Fig. 1

Study design. Key abbreviations used: (Analyses) EWAS = epigenome-wide association study; eQTM = expression quantitative trait methylation analysis; WGCNA = weighted gene co-expression network analysis; (Ancestries) AA = African American; (Cohorts) MAGNet = Myocardial Applied Genomics Network; BMCB = Bruce McManus Cardiovascular Biobank; UKBB = UK Biobank; GTEx = Genotype-Tissue Expression project

Fig. 2

DCM Sentinel CpGs. a Manhattan plot for discovery-stage EWAS of DCM. The horizontal significance line (red) corresponds to the epigenome-wide significance threshold (P < 5.96E−08, 0.05/838,624 tests). The 194 DCM sentinel CpGs are highlighted. At genomic loci with > 1 epigenome-wide significant signal, secondary signals (green) were identified by conditioning on the lead signal (lowest P in region; red). b Distribution of DCM Sentinel CpGs across various gene features. The 194 Sentinel CpGs were annotated to gene features based on the Illumina manifest file (version b2)

Fig. 3

Unsupervised clustering using methylation levels of 194 sentinel CpGs. DCM, dilated cardiomyopathy

Fig. 4

Association of sentinel CpG methylation with gene expression. A cis-eQTMs by genomic region. Sentinel CpGs are annotated by their genic locations: 5' upstream (the region from upstream to + 100 bp downstream of the gene TSS), gene body, and 3' downstream (the region following the gene body). Replicated cis-eQTMs were defined using two criteria: discovery FDR P < 0.05 (MAGNet) and confirmed directionality in replication testing (BMCB). B Directionality of replicated cis-eQTMs by genic location of sentinel CpGs. In each region-specific plot, replicated cis-eQTMs are ordered by chromosomal location of sentinel CpGs along the x-axis. The y-axis represents beta values of individual cis-eQTM relationships based on discovery-stage analysis. The table below each plot summarises the inverse and positive cis-eQTMs by genomic region, including their ratio (inverse/positive). Gene coordinates and TSS are based on the GENCODE version 19 annotation. eQTM = expression quantitative trait methylation loci

Fig. 5

Regional plots for a sentinel CpG gene pair with suggestive evidence for colocalisation. A Associations of CpGs surrounding cg11793257 with DCM from discovery-stage EWAS (MAGNet). The displayed region (2 Mb) is centred on cg11793257. B Regional genetic associations with cg11793257 and ABHD12 Genetic associations were obtained for a region (± 500 kb) centred on the genetic variant (rs2424699) that was used as an instrumental variable in SMR to assess causal relationships between cg11793257 methylation and ABHD12 gene expression. The posterior probability (coloc PP.H4) of colocalisation between cg11793257 meQTLs and ABHD12 eQTLs was 0.47. Gene tracks for all regional plots consist of protein-coding genes obtained from Ensembl Release 75 (last release to be based on hg19 assembly). C Left ventricular chromatin interactions. At 5 kb resolution, multiple interactions were identified between the chromatin region containing cg11793257 and various regions along ABHD12. meQTL = methylation expression quantitative trait loci; eQTL = expression quantitative trait loci

Fig. 6

Co-methylation modules identified using DCM-associated CpGs. A Hierarchical cluster tree (dendogram) of co-methylation modules. WGCNA identified seven modules that exhibit conservation between co-methylation networks constructed independently within the White (n = 209) and African American (n = 118) ancestries. The colour band underneath the tree indicates distinct modules (grey = CpG sites that are not clustered into any module). B–D Bar plots of over-represented gene sets (FDR P < 0.05) in co-methylation modules. Genes were assigned to CpGs based on replicated cis-eQTM relationships. Gene set enrichment within a module was assessed against a background set consisting of all genes mapped to CpG sites that were used to identify co-methylation patterns (855 unique genes). Enriched gene ontology terms are displayed by decreasing order of significance (−log10p value)

Fig. 7

Regional associations of a DCM sentinel CpG with CVD-related traits. This plot shows the regional associations of the cg11793257 region with creatinine and FramCHD score. 12 CpGs were sequenced in this region. Three types of plots are included in each panel: (top) pairwise correlation in methylation levels between CpGs within the region; (middle) CpG-CVD trait associations; (bottom) protein-coding genes in the region. A Regional associations with creatinine. The lead signal is chr20_25218304 (P = 1.26E−03). B Regional associations with FramCHD score. The lead signal is chr20_25218276 (P = 4.08E−03). CVD = cardiovascular disease