Epigenome-wide association data implicate DNA methylation as an intermediary of genetic risk in rheumatoid arthritis

Abstract

Epigenetic mechanisms integrate genetic and environmental causes of disease, but comprehensive genome-wide analyses of epigenetic modifications have not yet demonstrated robust association with common diseases. Using Illumina HumanMethylation450 arrays on 354 anti-citrullinated protein antibody-associated rheumatoid arthritis cases and 337 controls, we identified two clusters within the major histocompatibility complex (MHC) region whose differential methylation potentially mediates genetic risk for rheumatoid arthritis. To reduce confounding factors that have hampered previous epigenome-wide studies, we corrected for cellular heterogeneity by estimating and adjusting for cell-type proportions in our blood-derived DNA samples and used mediation analysis to filter out associations likely to be a consequence of disease. Four CpGs also showed an association between genotype and variance of methylation. The associations for both clusters replicated at least one CpG (P < 0.01), with the rest showing suggestive association, in monocyte cell fractions in an independent cohort of 12 cases and 12 controls. Thus, DNA methylation is a potential mediator of genetic risk.

Figure 1

Differential cell counts in identifying RA-associated differentially methylated positions (DMPs). Volcano plot of −log₁₀ (p-value) against beta, representing the methylation difference between ACPA positive RA cases and controls, without a) or with b) adjusting for differential cell counts, in addition to control for age, sex and smoking status. The dashed red lines represent the threshold used for statistical cutoff (Bonferroni-adjusted p-value = 0.05).

Figure 2

Identification of epigenetically mediated genetic risk factors for RA disease. (a) Possible relationships between a causal factor (G), a possible mediator (M) and an outcome (Y). The diagram on the left represents the methylation-mediated relationship, in which genotype (G) acts on phenotype (Y) through methylation (M). The diagram in the middle represents the consequential methylation model, in which DNA methylation (M) changes are the consequence of phenotype (Y). The diagram on the right represents the independent model, in which the genotype (G) acts on DNA methylation (M) and phenotype (Y) independently. (b) Summary workflow for identifying epigenetically mediated genetic risk factors for RA. The diagrams on the right represent the relationships between genotype (G), DNA methylation (M) and RA phenotype (Y). The dashed lines represent the association relationship, whereas the arrowed lines represent the causal relationship.

Figure 3

Summary workflow and results for identifying epigenetically mediated genetic risk factors for RA disease.

Figure 4

Genotype-dependent candidate DMPs that mediate genetic risk within the MHC region. (a) Top panels: association between DNA methylation level at a DMP that mediates genetic risk in RA and phenotype (top left) or genotype (top right). Red lines mark average DNA methylation levels. Bottom left panel: association between genotype and RA phenotype. Red lines mark percentage of cases for each genotype. Bottom right panel: Coefficient (β) represents the dependence of RA phenotype (Y) on genotype (G), with or without adjusting for DNA methylation (M). The error bars represent the 95% confidence interval for the coefficient, β. In the case of the methylation-mediated model, the absolute value of the observed G:Y relationship strength reduces towards zero when adjusting for methylation (M). (b) Association between candidate genetic risk-mediating DMPs and genotype within a 1.5Mb section of the MHC region. Each dashed line represents a potential mediation relationship between a SNP and a DMP as determined by the Causal Inference Test (CIT). The color of the line indicates significance of the p-value for the statistical mediation test. Genotype data is imputed based on a large reference panel. Three previously identified RA-associated genetic variants are illustrated in red. Bottom panel: zoomed-in images of top panel with gene annotations. (c) Examples of RA-associated DMP in which genotype is associated with the change in both mean and variance of DNA methylation. The methylation density plot is color coded by genotype. The number of individuals in each genotype group is shown on the top left corner.

Figure 5

Replication data for ten candidate DMPs that mediate genetic risk in RA from sorted monocytes. DNA methylation data on CD14+ monocytes from a replication set of 12 ACPA positive RA cases and 12 controls was measured using the Illumina 450K methylation array. Results from PBLs are illustrated in blue while the results from monocytes are in red. For PBLs, all 10 DMPs have p-value less than 10⁻⁷. For sorted monocytes: one asterisk, p-value < 0.1; two asterisks, p-value < 0.01; three asterisks, p-value < 0.001.