Modifier effects between regulatory and protein-coding variation

Abstract

Genome-wide associations have shown a lot of promise in dissecting the genetics of complex traits in humans with single variants, yet a large fraction of the genetic effects is still unaccounted for. Analyzing genetic interactions between variants (epistasis) is one of the potential ways forward. We investigated the abundance and functional impact of a specific type of epistasis, namely the interaction between regulatory and protein-coding variants. Using genotype and gene expression data from the 210 unrelated individuals of the original four HapMap populations, we have explored the combined effects of regulatory and protein-coding single nucleotide polymorphisms (SNPs). We predict that about 18% (1,502 out of 8,233 nsSNPs) of protein-coding variants are differentially expressed among individuals and demonstrate that regulatory variants can modify the functional effect of a coding variant in cis. Furthermore, we show that such interactions in cis can affect the expression of downstream targets of the gene containing the protein-coding SNP. In this way, a cis interaction between regulatory and protein-coding variants has a trans impact on gene expression. Given the abundance of both types of variants in human populations, we propose that joint consideration of regulatory and protein-coding variants may reveal additional genetic effects underlying complex traits and disease and may shed light on causes of differential penetrance of known disease variants.

Figure 1. Illustration of a hypothetical epistatic interaction between a regulatory and a protein-coding variant.

Two double heterozygote individuals may be genotypically identical, but the phasing of alleles can be different and may result in very distinct phenotypes between individuals. In one individual (i) the A allele of the rSNP drives high expression levels of the protein arising from the C allele of the nsSNP. In another individual (ii) the G allele of the rSNP drives low expression levels of the protein arising from the C allele of the nsSNP. If the protein-coding variant is functionally important then this can give rise to different means in the distribution of a complex trait phenotype as shown on the right.

Figure 2. Strategies applied to discover differentially expressed (DE) nsSNPs and linkage disequilibrium properties between rSNP-nsSNP pairs.

(A) Two approaches were employed to discover DE nsSNPs: nsSNPs mapping in genes with a known rSNP (i) and nsSNPs that were associated with expression levels of the gene they map in (ii). In (ii) the presence of a cis-acting regulatory variant is implied. For some nsSNPs with a significant association, an identified cis rSNPs also exists (iii). In all other cases the nsSNPs interrogated were not inferred to be DE (iv). (B) (i) 909 nsSNPs map in a gene with an identified rSNP; (ii) 884 nsSNPs were found to be associated with levels of gene expression of the gene they reside in; (iii) the overlap between i and ii (nsSNPs with an identified rSNP that also showed a significant association) is 291 (iii). 6731 nsSNPs show no evidence for DE. (C) The distribution of r-squared (a measure of LD) was compared between rSNP-nsSNP pairs in which the nsSNP showed a significant association (at the 0.01 permutation threshold) and SNP pairs in which the nsSNP was not associated. As expected, r-squared values are much higher in the first case, in which the nsSNP is thought to act as a tag of the functional regulatory variant nearby.

Figure 3. Impact of rSNP-nsSNP genetic interaction on trans gene expression.

(A) QQ plot of observed –log₁₀pvalues of the interaction term in the ANOVA over expected (under the assumption of a Uniform distribution of p-values). (B) QQ plot of observed –log₁₀pvalues of the interaction term in the ANOVA over the –log₁₀pvalues of the interaction term in the permuted data. (C) Example 1: The interaction between rs13093220 (rSNP) and rs3009034 (nsSNP) on chromosome 3, is associated with changes in expression of NDN (probe ID GI_10800414-S) on chromosome 15 (interaction p = 4.5*10⁻¹¹). (D) Example 2: The interaction between rs6776417 (rSNP) rs17040196 (nsSNP) on chromosome 3 is associated with changes in expression of RLF (probe ID GI_6912631-S) on chromosome 1 (interaction p = 2.2*10⁻⁵).