DNA methylation signature of chronic low-grade inflammation and its role in cardio-respiratory diseases

Abstract

We performed a multi-ethnic Epigenome Wide Association study on 22,774 individuals to describe the DNA methylation signature of chronic low-grade inflammation as measured by C-Reactive protein (CRP). We find 1,511 independent differentially methylated loci associated with CRP. These CpG sites show correlation structures across chromosomes, and are primarily situated in euchromatin, depleted in CpG islands. These genomic loci are predominantly situated in transcription factor binding sites and genomic enhancer regions. Mendelian randomization analysis suggests altered CpG methylation is a consequence of increased blood CRP levels. Mediation analysis reveals obesity and smoking as important underlying driving factors for changed CpG methylation. Finally, we find that an activated CpG signature significantly increases the risk for cardiometabolic diseases and COPD.

Fig. 1. Results of multi-ethnic meta-analysis.

Panel A is a circos plot representation of the multi-ethnic meta-analysis. Outermost track is chromosome number followed by ideogram. The second track in blue is the Manhattan plot of the CpG CRP association results. Next track (in orange and light green) are effect sizes of CpG CRP associations, where orange represents positive associations and light green negative. Track represented in brown track color gives the overlap between the 1765 CRP associated CpG markers with CpG island in the genome. The innermost track (in dark green) gives the overlap with enhancer regions as defined by Roadmap project. Panel B is a qqplot of the genomic control corrected P-values from the multi-ethnic meta-analysis. Panel C shows replication rates of 1765 across ancestries. Each bar gives the number of replicated CpGs across ancestries indicated as dots below the barplot. Horizontal bars reflect the total number of replicated CpGs per ancestry group.

Fig. 2. Correlation structure of CRP-associated CpG methylation.

Panel A gives meta-analyzed correlation values (Pearson Rho) across 4 cohorts. Displayed are all CRP-associated CpGs from a genomic region from chromosome 5 alongside with CRP-associated CpGs on chromosome 6. CpG ids are color-coded according to correlation clusters. Panel B CpGs correlation depending on distance. Correlation values were binned according to their distance to each other. X-axis gives distance between CpGs. Y-axis gives Pearson Rho observed each distance bin. In blue font, we plotted mean and standard errors of Person Rho values. Panel C is a UMAP representation of correlation values of the 1511 independent loci. Dots are color-coded according to their correlation cluster membership.

Fig. 3. Driving forces of CpG signature.

Panel A is a comparison of Z-scores from the sensitivity analysis. Each dot represents a coefficient from the 1511 CRP-associated loci. The X-axis gives Z-scores derived from base model as applied in multi-ethnic meta-analysis. The Y-axis gives results from the same analysis adjusted for BMI. Panel B gives an overview of applied mediation analysis models. Panels C and D are representations of the Mendelian Randomization triangulation analysis. Each dot represents a CpG. The Y-axis is the observed effect, which is the association between the genetic instrument and outcome. The observed effects for Panel C originate from CpG instruments (SNPs) vs serum CRP levels. The predicted effect is the combined effect from the SNP CpG association and the CpG serum CRP association. Observed effects for D are the associations between a polygenic risk score for CRP (instruments) and CpG methylation. The predicted effects for panel D are the combined effects from the polygenic risk score for CRP (instruments) serum CRP association and serum CRP CpG methylation association (CRP_{Genetic risk score for CRP association} × CRP _{CpG association}). The observed effect is the association of the polygenic risk score for CRP (instruments) to CpG methylation (CRP_{Genetic risk score for CpG association}).

Fig. 4. Overrepresentation analysis.

Panel A gives the percentage of each CRP-associated gene set that overlaps with selected genomic feature. Orange bars represent overlapping features by chance; green bars give the percentage that actually overlap with the CRP-associated CpGs. Transcription start site and enhancer genomic region were used as defined by the Roadmap project. HiC regions were as reported in GSE63525, where component A was connected to highly transcribed genomic regions and component B to heterochromatin. Panel B shows enrichment analysis between CRP-associated CpG that were significantly associated with mRNA expression. Empirical P-values for the overlap derived from a permutation test (described in more detail in method section “Overrepresentation analysis”) are given as negative log10. Percent overlap indicates the percentage of CpGs present in each GO term set. Panel C gives overlaps between CpGs observed in this study and published gene lists from large scale EWAS.

Fig. 5. Associations of CRP DNA methylation signature to clinically relevant phenotypes.

Forest plots give estimate from logistic regression (logODDs) and confidence intervals (error bars) of CpG risk score regression against relevant phenotypes. N is the number of samples included in analysis. To produce adjusted relative risk estimates we transformed odds ratios as follows: RR = odds ratio/1 − (lifetime risk) + (life time risk × odds ratio). Those estimates indicate the theoretical maximum impact of the discovered CpG signature (100% DNA methylation change) on the tested traits. The risk conveyed by one percent change in the DNA methylation risk score on the tested traits was 1.007% for COPD, 1.7% for T2D, 2.9% for myocardial infarction 4.3% coronary artery disease, and 0.2% for hypertension. For continuous traits such as FEV1, FVC, systolic BP, and blood glucose estimates from linear regression including confidence intervals are given.