Determining tissue-specific methQTLs

To determine whether the effects observed in the four tissues independently were shared across the different samples, we employed colocalization analysis. More specifically, we used Summary-data-based Mendelian Randomization (SMR) and Heterogeneity in Dependent Instruments (HEIDI) analysis [39] implemented in the GWAS-MAP software tool. Briefly, SMR is a statistical test that indicates whether two traits (here CpG methylation states in two tissues) are significantly associated with the same genetic locus. The test is an extension of Mendelian Randomization (MR), which is used to test for a causal relationship between two traits using an instrumental variable. While classical MR requires that the two traits are measured on the same samples, these can be investigated in distinct samples or studies using SMR. The input to the SMR test are methQTL statistics (i.e., P-values, slopes of the regression line) obtained in two scenarios, and it returns a test statistic that indicates whether the effect observed in the two scenarios is significantly associated with the same SNP. Thus, SMR analysis determines whether the same genetic effect leads to the methQTL results that we obtained in the two tissues, but cannot discern pleiotropy (the same SNP influences two traits) from linkage (two highly correlated SNPs influence the traits independently). Thus, for the SNPs that pass the SMR test, we employed the HEIDI test in a second step to test whether the observed effects are likely driven by pleiotropy. Briefly, the HEIDI test utilizes linkage (correlation) information of SNPs from a reference panel to determine whether the observed heterogeneity in the methQTL statistics is more likely caused by linkage than by pleiotropy. By using colocalization analysis through SMR and HEIDI, we were able to determine whether the methQTLs identified in the four tissues/cell types independently were shared or tissue-specific. We employed colocalization analysis for all pairs of tissues/cell types to determine shared methQTLs (Fig. 3B).

We selected those CpGs for colocalization analysis, which were selected as tag-CpGs in at least two tissues and that had a significant association with a lead-SNP (P-value below 8.65 × 10^–11) at least in one tissue. Then, anchoring the analysis on the tissue showing the significant association, we performed the SMR test to detect if the same lead-SNP is associated with the same CpG in any of the other tissues. In case the same lead-SNP was identified in more than one tissue, the tissue/cell type with the lowest P-value was used as the starting point of the SMR analysis. In total, we performed 4253 SMR tests. The SMR P-values were adjusted for multiple testing using the Benjamini–Hochberg [64] method and we used a P-value cutoff of 0.05. In case the methQTLs measured in two tissues are significant according to the SMR test, this is an indication that the CpG methylation states are significantly correlated with the same SNP in the two tissues. Thus, we use the P-value of the SMR test as an indication of the shared effect of methQTLs in the two tissues.

For CpGs that passed the SMR test, we applied the HEIDI test to discern pleiotropy from linkage. We defined all those pairs of methQTLs with a P-value higher than 0.05 as pleiotropic interactions. The results for a different P-value cutoff (0.001) are shown in Additional file 4: Table S3. The methQTLs that had an SMR test P-value below the cutoff and had a HEIDI test P-value higher than the threshold were defined as shared across the two tissues. The methQTLs shared across all pairwise comparisons according to the colocalization analysis were termed shared methQTLs. Additionally, those shared methQTLs with a methQTL P-value below 8.65 × 10^–11 in all tissues were termed common methQTLs.

The methQTLs that either fail the SMR test or that pass the SMR test, but also pass the HEIDI test were defined as tissue-specific methQTLs (Additional file 4: Table S3). Tissue-specificity was defined for each tissue individually. Finally, three classes of methQTLs were defined: tissue-specific, shared, and common methQTLs. SMR and HEIDI analysis was performed using GWAS-MAP (https://www.polyknomics.com/solutions/gwas-map-biomarker-and-intervention-target-discovery-platform).