| Summary: | Analyses of genome-wide association study data have demonstrated that there are potentially thousands of loci associated with schizophrenia (Sullivan et al. 2003). Although risk is partially explained by the additive effects of top-ranking polymorphisms, genetic interactions may help to explain additional heritability (Hemani et al. 2014; Zuk et al. 2012). However, attempts to identify disease-associated pair-wise interactions through exhaustive testing have so far been unsuccessful due to the large burden of multiple testing and the absence of easily discoverable interactions of large effect (Moskvina et al. 2011). Here we investigate whether evidence for a contribution to disease risk from SNP-SNP interactions can be found by searching for sets of genes enriched for nominally associated interactions. When performing interaction analyses covariates were introduced to account for population structure. Where the effect of covariates needs to be accounted for, the most widely used method modifies the basic logistic regression interaction analysis by simply adding covariate terms into the model. The performance of this method was compared to two alternative approaches: adding covariate-SNP interactions terms in addition to the individual covariate terms, as suggested by (Yzerbyt et al. 2004); and testing for interactions in each population separately, then using meta-analysis to combine interaction effects. Results and running time were similar whether SNP-covariate terms were included or not, while the meta-analytic approach was found to be the most efficient in terms of running time. To try and identify sets of genes enriched for nominally associated interactions, two approaches were investigated: one based on genetic information alone, and one based on functional information using protein-protein interactions (PPI). The first approach analyzed the distribution of interaction p-values after ranking them by the gene-wide main effects of the contributing genes, allowing a comparison to be made between genes with high/low gene-wide association. The second approach asked whether genes encoding directly interacting proteins were enriched for nominally associated interactions, drawing upon two PPI datasets: one from a large experimental (yeast two-hybrid) screen, the other consisting of PPI data curated from the literature. In both of the genetic datasets studied there was evidence for enrichment of nominally associated interactions amongst genes with highest gene-wide association for schizophrenia. There was no evidence for an excess of nominally associated interactions when investigating either PPI dataset.
|
|---|
