Genome-wide interaction-based association (GWIBA) analysis has the potential to identify novel susceptibility loci. These interaction effects could be missed with the prevailing approaches in genome-wide association studies (GWAS). However, no convincing loci have been discovered exclusively from GWIBA methods, and the intensive computation involved is a major barrier for application. Here, we developed a fast, multi-thread/parallel program named “pair-wise interaction-based association mapping” (PIAM) for exhaustive two-locus searches. With this program, we performed a complete GWIBA analysis on seven diseases with stringent control for false positives, and we validated the results for three of these diseases. We identified one pair-wise interaction between a previously identified locus, C1orf106, and one new locus, TEC, that was specific for Crohn's disease, with a Bonferroni corrected P<0.05 (P = 0.039). This interaction was replicated with a pair of proxy linked loci (P = 0.013) on an independent dataset. Five other interactions had corrected P<0.5. We identified the allelic effect of a locus close to SLC7A13 for coronary artery disease. This was replicated with a linked locus on an independent dataset (P = 1.09×10−7). Through a local validation analysis that evaluated association signals, rather than locus-based associations, we found that several other regions showed association/interaction signals with nominal P<0.05. In conclusion, this study demonstrated that the GWIBA approach was successful for identifying novel loci, and the results provide new insights into the genetic architecture of common diseases. In addition, our PIAM program was capable of handling very large GWAS datasets that are likely to be produced in the future.

BackgroundCopy number variations (CNV) are important causal genetic variations for human disease; however, the lack of a statistical model has impeded the systematic testing of CNVs associated with disease in large-scale cohort.Methodology/Principal FindingsHere, we developed a novel integrated strategy to test CNV-association in genome-wide case-control studies. We converted the single-nucleotide polymorphism (SNP) signal to copy number states using a well-trained hidden Markov model. We mapped the susceptible CNV-loci through SNP site-specific testing to cope with the physiological complexity of CNVs. We also ensured the credibility of the associated CNVs through further window-based CNV-pattern clustering. Genome-wide data with seven diseases were used to test our strategy and, in total, we identified 36 new susceptible loci that are associated with CNVs for the seven diseases: 5 with bipolar disorder, 4 with coronary artery disease, 1 with Crohn's disease, 7 with hypertension, 9 with rheumatoid arthritis, 7 with type 1 diabetes and 3 with type 2 diabetes. Fifteen of these identified loci were validated through genotype-association and physiological function from previous studies, which provide further confidence for our results. Notably, the genes associated with bipolar disorder converged in the phosphoinositide/calcium signaling, a well-known affected pathway in bipolar disorder, which further supports that CNVs have impact on bipolar disorder.Conclusions/SignificanceOur results demonstrated the effectiveness and robustness of our CNV-association analysis and provided an alternative avenue for discovering new associated loci of human diseases.

Abstract