Genome-wide association analysis using multiple Atlantic salmon populations

BackgroundIn a previous study, we found low persistence of linkage disequilibrium (LD) phase across breeding populations of Atlantic salmon. Accordingly, we observed no increase in accuracy from combining these populations for genomic prediction. In this study, we aimed to examine if the same were true for detection power in genome-wide association studies (GWAS), in terms of reduction in p-values, and if the precision of mapping quantitative trait loci (QTL) would improve from such analysis. Since individual records may not always be available, e.g. due to proprietorship or confidentiality, we also compared mega-analysis and meta-analysis. Mega-analysis needs access to all individual records, whereas meta-analysis utilizes parameters, such as p-values or allele substitution effects, from multiple studies or populations. Furthermore, different methods for determining the presence or absence of independent or secondary signals, such as conditional association analysis, approximate conditional and joint analysis (COJO), and the clumping approach, were assessed.ResultsMega-analysis resulted in increased detection power, in terms of reduction in p-values, and increased precision, compared to the within-population GWAS. Only one QTL was detected using conditional association analysis, both within populations and in mega-analysis, while the number of QTL detected with COJO and the clumping approach ranged from 1 to 19. The allele substitution effect and -log10p-values obtained from mega-analysis were highly correlated with the corresponding values from various meta-analysis methods. Compared to mega-analysis, a higher detection power and reduced precision were obtained with the meta-analysis methods.ConclusionsOur results show that combining multiple datasets or populations in a mega-analysis can increase detection power and mapping precision. With meta-analysis, a higher detection power was obtained compared to mega-analysis. However, care must be taken in the interpretation of the meta-analysis results from multiple populations because their test statistics might be inflated due to population structure or cryptic relatedness.