Confronting the missing epistasis problem: on the reproducibility of gene–gene interactions

Abstract

Epistasis (gene-gene interaction) is thought to play an integral role in the genetic basis of complex traits, and a significant amount of research has been invested into identifying this phenomenon in human disease. However, the overall success of empirical studies of epistasis in humans is unclear, as such studies are rarely systematically evaluated. Here, we have selected asthma as an example of a well-studied, complex human disease, and provide a critical analysis and replication attempt of nearly all prior reports of epistasis for this disease. Of 191 previously reported interactions, we find that 39.8% were not originally identified using an explicit test for interaction and thus may not have been true epistatic effects to begin with. Moreover, directions of effect were not described for 46.1% of the interactions, which prevents their rigorous replication. In the original studies, attempts at replication were made for 15.2% of the interactions, and 7.3% were actually replicated. In the current study, we were able to evaluate 85.9% of the interactions using a large asthma dataset from the GABRIEL Consortium. None of these interactions could be replicated based on strict criteria. However, we found nominally significant (p < 0.05) evidence in support of 23.8% of the evaluated interactions. Although many reports of epistasis are not robustly supported in the published literature, our results suggest that at least some of these reports may have been true-positive examples of epistasis. In general, improvements in empirical studies of epistasis are called for, in order to better understand the importance of this phenomenon in human disease.

Are Interactions between cis-Regulatory Variants Evidence for Biological Epistasis or Statistical Artifacts?

“…Most studies rely on the statistical association between genetic variants and phenotype to identify signs of epistasis, and the interactions identified are notoriously difficult to replicate. 6,7 This may be attributable to the inherent inability to tightly control a variety of factors when studying phenotypes in humans, or to the fact that most phenotypes studied are several steps removed from the underlying biological processes that influence them. These methodological limitations make it unclear whether the lack of observed epistasis in humans is a true feature of the genetic architecture, or whether epistasis is simply much more difficult to observe outside experimental systems.…”

Neuropeptide S (NPS) variants modify the signaling and risk effects of NPS Receptor 1 (NPSR1) variants in asthma

“…We must emphasize that the sample size analyzed in this study is underpowered to detect epistatic effects. However, the facts that the interaction signals in the two independent cohorts were narrowed to the same gene regions (Fig 4) and a recent meta-analysis on a larger dataset such as GABRIEL [49] confirmed the interaction signals that we originally identified in this population between NPSR1 and RORA [17], suggest that the interactions between NPS and NPSR1 may occur and are detectable at present resolution. Additional studies using larger sample sizes more suitable to test for epistasis are needed to confirm these findings.…”

Accurately annotate compound effects of genetic variants using a context-sensitive framework

“…It leverages the SNP–SNP interaction between different genes to explain the lack of heritability of numerous types of complex human disease (39,40). COPE is a variant effect annotator that considers the compound effects of multiple variants within the same gene, in contrast to epistasis, to annotate their effects accurately.…”