Genome-wide association studies are widely used to identify "disease genes" conferring resistance/susceptibility to infectious diseases. Using a combination of mathematical models and simulations we demonstrate that genetic interactions between hosts and parasites (GxG interactions) can drastically affect the results of these association scans and hamper our ability to detect genetic variation in susceptibility. When hosts and parasites coevolve, these GxG interactions often make Genome-wide association studies unrepeatable over time or across host populations. Reanalyzing previously published data on Daphnia magna susceptibility to infection by Pastueria ramosa, we identify genomic regions consistent with GxG interactions. We conclude by outlining possible avenues for designing more powerful and more repeatable association studies.
CoGWAS_Supplement
Mathematica notebook deriving analytical models, numerical simulations, and for producing figures 1-4 and S1.
CoGWAS_Supplement
CDF version of Mathematica notebook covering analytical model derivation, numerical simulations, and for producing figures 1-4 and S1.
CoGWAS_Supplement
PDF version of Mathematica notebook covering analytical model derivation, numerical simulations, and for producing figures 1-4 and S1.
Fig. S1 - Matching-alleles model
Row 1: Estimated allelic effects as a function of pathogen allele frequency (black: \beta_0, solid red: \beta_{Hi}, dashed red: \beta_{H1,H2}, blue \beta_{Pi}, blue dashed: \beta_{P1,P2}, purple dashed: \beta_{Hi,Pj}). Row 2: Variation explained by host additive effects only (solid red), host additive and epistatic effects (dashed red), host and pathogen additive and epistatic effects (dashed blue, overlaps with dashed red), and a full host pathogen model (dashed purple).
FigS1.gif
Fig. S2 - GWAS of *Daphnia magna* susceptibility
Genetic associations of each SNP with C1 (red circles), C19 (blue, squares), and "mixed'' susceptibility (green, triangles). Here, the mixed case was obtained by taking each host clone and randomly choosing the data upon exposure to C1 or C19 (not both, as incorporated in the overall analysis in figure 5). Hence each SNP is represented three times, once for each genome wide scan. Note that closely linked SNPs often overlap with one another and are not all individually visible. Significant SNPs are shown in colour while those below the log-likelihood of 2 threshold or that are not clustered within a 10cM region of three other significant SNPs are shown in grey. The 10 linkage groups are delineated by vertical dashed lines.
FigS2.gif
