Identifying cis-mediators for trans-eQTLs across many human tissues using genomic mediation analysis

Abstract

The impact of inherited genetic variation on gene expression in humans is well-established. The majority of known expression quantitative trait loci (eQTLs) impact expression of local genes (cis-eQTLs). More research is needed to identify effects of genetic variation on distant genes (trans-eQTLs) and understand their biological mechanisms. One common trans-eQTLs mechanism is "mediation" by a local (cis) transcript. Thus, mediation analysis can be applied to genome-wide SNP and expression data in order to identify transcripts that are "cis-mediators" of trans-eQTLs, including those "cis-hubs" involved in regulation of many trans-genes. Identifying such mediators helps us understand regulatory networks and suggests biological mechanisms underlying trans-eQTLs, both of which are relevant for understanding susceptibility to complex diseases. The multitissue expression data from the Genotype-Tissue Expression (GTEx) program provides a unique opportunity to study cis-mediation across human tissue types. However, the presence of complex hidden confounding effects in biological systems can make mediation analyses challenging and prone to confounding bias, particularly when conducted among diverse samples. To address this problem, we propose a new method: Genomic Mediation analysis with Adaptive Confounding adjustment (GMAC). It enables the search of a very large pool of variables, and adaptively selects potential confounding variables for each mediation test. Analyses of simulated data and GTEx data demonstrate that the adaptive selection of confounders by GMAC improves the power and precision of mediation analysis. Application of GMAC to GTEx data provides new insights into the observed patterns of cis-hubs and trans-eQTL regulation across tissue types.

Figure 1.

Causal diagram demonstrating mediation and “mediator-outcome confounding.” Here, the variable set “U” represents a set of unmeasured or unknown variables that may show confounding effects in the mediation analysis. Mediation analysis can detect mediation of the effect of the eSNP on the trans-gene by the cis-gene, assuming mediator-outcome confounding is absent or adjusted for in the analysis. Mediation will not be detected if the effect of the eSNP on the trans-gene is through some alternative pathway that does not involve the cis-gene.

Figure 2.

Graphical illustrations of GMAC and its main ideas. (A) A summary of the GMAC algorithm; (B) a mediation relationship among an eQTL, L_i, its cis-gene transcript, C_i, and a trans-gene transcript, T_j, with confounders, X_ij, allowing L_i to affect T_j via a pathway independent of C_i; (C) a mediation trio where C_i and T_j have common child variable(s), Z_ij; (D) a mediation trio where C_i affects T_j through intermediate variable(s), W_ij. (E) The adaptive confounder selection procedure: Based on the P-value matrix for the association of each potential confounder variable to at least one of the cis- or the trans-gene transcripts, we apply a stratified FDR approach by considering the P-values for each potential confounder (each column) as a stratum, with the significant ones indicated by a check mark (√). When conducting the mediation test for each trio, we only adjust for the significant confounding variables (the ones with √ in each row). (F) A mediation trio L_i → C_i → T_j (left) and a trio under the null with both cis-linkage and trans-linkage but no mediation (right). Within-genotype permutation of the cis-gene expression levels maintains the cis- and trans-linkage (different mean levels) while breaking the potential correlation between the cis- and trans-expression levels within each genotype group. Note that X_ij, Z_ij, W_ij may vary by trios and are all subsets of H. We assume that either X_ij or a combination of variables in X_ij would capture the variation of the unmeasured confounder U in Figure 1.

Figure 3.

Plots of negative log base 10 of mediation P-values versus the percentages of reduction in trans-effects after accounting for cis-mediation. The P-values are calculated based on (A) mediation tests without adjusting for hidden confounders, and (B) mediation tests by GMAC considering all PCs as potential confounders. P-values are truncated at 10⁻¹⁶. The plots are based on the results from the adipose subcutaneous tissue. The percentage of reduction in trans-effects is calculated by (β₂^m − β₂)/β₂^m × 100%, where β₂^m is the marginal trans-effect of the eQTL on the trans-gene expression levels, and β₂ is the trans-effect after adjusting for a potential cis-mediator and other covariates. For trios with true cis-mediations, the marginal trans-effects are nonzero, and after adjusting for the true cis-mediators, we expect the adjusted trans-effects β₂ to be substantially reduced; that is, we expect the trios with very significant mediation P-values to have positive percent reduction in trans-effects. For results based on no adjustment of hidden confounders (A), we observed many trios with significant mediation P-values but the percentages of reduction in trans-effects are negative. At the 0.05 P-value threshold, 712 (21.4%) and 188 (5.6%) out of 3332 trios have P-values below the threshold and percent reduction in trans-effects being negative in A and B, respectively.

Figure 4.

A biologically interpretable example of a cis-eGene (IFI44L) that appears to mediate the effects of trans-eSNPs in multiple tissues. The gene IFI44L (A) resides <5 kb away from IFI44 (B), and expression of these genes is associated with a common cis-eQTL that also impacts the expression of multiple genes in trans in multiple tissues. Both IFI44 and IFI44L show statistical evidence of mediation for a similar set of interferon-related genes. Thus, based on this evidence, we infer that at least one of these genes is a cis-mediator, although we cannot know which is (or if both are) the true mediator.