Most disease associated genetic risk factors are non-coding, making it challenging to design experiments to understand their functional consequences. Identification of expression quantitative trait loci (eQTLs) has been a powerful approach to infer downstream effects of disease variants but the large majority remains unexplained.. The analysis of DNA methylation, a key component of the epigenome, offers highly complementary data on the regulatory potential of genomic regions. However, a large-scale, combined analysis of methylome and transcriptome data to infer downstream effects of disease variants is lacking. Here, we show that disease variants have wide-spread effects on DNA methylation in trans that likely reflect the downstream effects on binding sites of cis-regulated transcription factors. Using data on 3,841 Dutch samples, we detected 272,037 independent cis-meQTLs (FDR < 0.05) and identified 1,907 trait-associated SNPs that affect methylation levels of 10,141 different CpG sites in trans (FDR < 0.05), an eight-fold increase in the number of downstream effects that was known from trans-eQTL studies. Trans-meQTL CpG sites are enriched for active regulatory regions, being correlated with gene expression and overlap with Hi-C determined interchromosomal contacts. We detected many trans-meQTL SNPs that affect expression levels of nearby transcription factors (including NFKB1, CTCF and NKX2-3), while the corresponding trans-meQTL CpG sites frequently coincide with its respective binding site. Trans-meQTL mapping therefore provides a strategy for identifying and better understanding downstream functional effects of many disease-associated variants.

We show that epigenome- and transcriptome-wide association studies (EWAS and TWAS) are prone to significant inflation and bias of test statistics, an unrecognized phenomenon introducing spurious findings if left unaddressed. Neither GWAS-based methodology nor state-of-the-art confounder adjustment methods completely remove bias and inflation. We propose a Bayesian method to control bias and inflation in EWAS and TWAS based on estimation of the empirical null distribution. Using simulations and real data, we demonstrate that our method maximizes power while properly controlling the false positive rate. We illustrate the utility of our method in large-scale EWAS and TWAS meta-analyses of age and smoking.Electronic supplementary materialThe online version of this article (doi:10.1186/s13059-016-1131-9) contains supplementary material, which is available to authorized users.

Genome-wide association studies (GWAS) have identified thousands of variants associated with complex traits, but their biological interpretation often remains unclear. Most of these variants overlap with expression QTLs, indicating their potential involvement in regulation of gene expression. Here, we propose a transcriptome-wide summary statistics-based Mendelian Randomization approach (TWMR) that uses multiple SNPs as instruments and multiple gene expression traits as exposures, simultaneously. Applied to 43 human phenotypes, it uncovers 3,913 putatively causal gene–trait associations, 36% of which have no genome-wide significant SNP nearby in previous GWAS. Using independent association summary statistics, we find that the majority of these loci were missed by GWAS due to power issues. Noteworthy among these links is educational attainment-associated BSCL2 , known to carry mutations leading to a Mendelian form of encephalopathy. We also find pleiotropic causal effects suggestive of mechanistic connections. TWMR better accounts for pleiotropy and has the potential to identify biological mechanisms underlying complex traits.

We tested whether DNA-methylation profiles account for inter-individual variation in body mass index (BMI) and height and whether they predict these phenotypes over and above genetic factors. Genetic predictors were derived from published summary results from the largest genome-wide association studies on BMI (n ∼ 350,000) and height (n ∼ 250,000) to date. We derived methylation predictors by estimating probe-trait effects in discovery samples and tested them in external samples. Methylation profiles associated with BMI in older individuals from the Lothian Birth Cohorts (LBCs, n = 1,366) explained 4.9% of the variation in BMI in Dutch adults from the LifeLines DEEP study (n = 750) but did not account for any BMI variation in adolescents from the Brisbane Systems Genetic Study (BSGS, n = 403). Methylation profiles based on the Dutch sample explained 4.9% and 3.6% of the variation in BMI in the LBCs and BSGS, respectively. Methylation profiles predicted BMI independently of genetic profiles in an additive manner: 7%, 8%, and 14% of variance of BMI in the LBCs were explained by the methylation predictor, the genetic predictor, and a model containing both, respectively. The corresponding percentages for LifeLines DEEP were 5%, 9%, and 13%, respectively, suggesting that the methylation profiles represent environmental effects. The differential effects of the BMI methylation profiles by age support previous observations of age modulation of genetic contributions. In contrast, methylation profiles accounted for almost no variation in height, consistent with a mainly genetic contribution to inter-individual variation. The BMI results suggest that combining genetic and epigenetic information might have greater utility for complex-trait prediction.

Genome-wide association and fine-mapping studies in 14 autoimmune diseases (AID) have implicated more than 250 loci in one or more of these diseases. As more than 90% of AID-associated SNPs are intergenic or intronic, pinpointing the causal genes is challenging. We performed a systematic analysis to link 460 SNPs that are associated with 14 AID to causal genes using transcriptomic data from 629 blood samples. We were able to link 71 (39%) of the AID-SNPs to two or more nearby genes, providing evidence that for part of the AID loci multiple causal genes exist. While 54 of the AID loci are shared by one or more AID, 17% of them do not share candidate causal genes. In addition to finding novel genes such as ULK3, we also implicate novel disease mechanisms and pathways like autophagy in celiac disease pathogenesis. Furthermore, 42 of the AID SNPs specifically affected the expression of 53 non-coding RNA genes. To further understand how the non-coding genome contributes to AID, the SNPs were linked to functional regulatory elements, which suggest a model where AID genes are regulated by network of chromatin looping/non-coding RNAs interactions. The looping model also explains how a causal candidate gene is not necessarily the gene closest to the AID SNP, which was the case in nearly 50% of cases.