PMC

Summary of the workflow for cis-QTL, co-localization, partial correlation, and mediation analyses

Examples of six co-localized eQTL-meQTL pairs. P-values for the co-localizing eQTL (blue) and meQTL (red) are plotted against physical position. a ILMN_1658464 (GTF3A) and cg22138327. b ILMN_1694711 (MAD2L1BP) and cg14302083. c ILMN_1721978 (CARD11) and cg19214707. d ILMN_1737918 (C1QA) and cg10916651. e ILMN_2193591 (UNC93B1) and cg20272935. f ILMN_2282366 (IQSEC3) and cg10356759

Overlap with genomic annotations for three candidate causal SNPs for an eQTL that co-localizes with seven meQTLs and shows strong evidence of mediation. a The SYNGR1 gene region has seven CpG sites (in red) affected by a common causal variant. Possible causal variants include rs9611155 (b) and rs909685 and 2069435 (c)

The posterior probability (PP) of sharing a common causal variant (CCV) depends on the extent of LD in the region. a Average PPs of CCV at different priors (p₁₂) stratified by quartiles of LD score of the region. b PP of CCV at different prior (p₁₂) for the lead eSNPs with the lowest LD score (50 SNPs). c PP of CCV at different prior (p₁₂) for the 50 lead eSNPs with the highest LD score (50 SNPs)

Histogram of the relative posterior support for a CCV stratified by strength of LD (r²) between the lead eSNP and lead meSNP. The relative posterior support for a CCV is defined as the posterior probability of a CCV divided by the sum of the posterior probabilities for a CCV and for distinct causal variants (DCV). These data are restricted to co-localization tests for which the sum of the posterior probabilities for DCVand CCV are >0.8 (100 out of 5397 tests excluded).  Results are shown for five values of the prior p₁₂ (panels a-e)

Direction of QTL effects and associations between expression and methylation for co-localized eQTL-meQTL pairs. Results for 2913 potentially co-localized eQTL-meQTL pairs identified using a p₁₂ value 4.4 × 10⁻⁴ are presented (results for additional p₁₂ values are presented in Supplementary Figures -). Results are stratified according to P-values from mediation analysis (Sobel P) and partial correlation analysis (Corr. P). a Histograms of the percentage of eQTL-meQTL pairs showing the same or different direction of association. b Histograms of the percentage of eQTL-meQTL pairs for which the direction of association between gene expression and DNA methylation is positive or negative

Examples of eQTLs that co-localize with a meQTL that has opposite effects on two nearby CpG sites. Results for the CpG selected for co-localization analysis (red) are shown on the top of each plot (ascending), overlaid with the eQTL results (blue). For these scenarios, the expression-increasing allele is associated with increased methylation at a "primary" CpG. Results for a “secondary CpG” are shown below (descending) in orange, for which the expression-increasing allele is associated with decreased methylation. The lead eSNP is shown as a triangle. a eSNP: 10:91186842, gene expression probe: ILMN_1696654, first CpG: cg27582166, second CpG: cg13172359. b eSNP: 6:37669641, gene expression probe: ILMN_1720595, first CpG: cg26720545, second CpG: cg26129310. c eSNP: 3:45047987, gene expression probe: ILMN_2055477, first CpG: cg25593573, second CpG: cg06117855

Partial correlation and mediation analyses provide evidence for shared regulatory mechanisms. Results for 2913 potentially co-localized eQTL-meQTL pairs identified using a p₁₂ value 4.4 × 10⁻⁴ are presented (results for additional p₁₂ values are presented in Supplementary Figures –). All models include adjustments for age, sex, and PCs from both the expression and methylation data (n = 316). a Results from partial correlation analysis. b Mediation analysis results for both the SME (blue) and SEM (red) models. Mediation proportion outliers out of −3 to 3 ranges were removed from the figure and Sobel P outliers <10⁻¹⁵ were set to the 10⁻¹⁵. c Relationship between Sobel P from mediation analyses and the post-adjustment correlation P-values from partial correlation analysis. Sobel P outliers <10⁻¹⁵ were set to the 10⁻¹⁵. d Venn diagram showing the concordance between mediation analysis and partial correlation analyses