Pool-seq (pooled sequencing) combines DNA from multiple individuals prior to sequencing, enabling population-level allele frequency estimation without individual genotyping. When employed in a case-control Genome Wide Association Study (GWAS) framework, pool-seq faces a fundamental power limitation: Errors on allele frequency estimates are inversely proportional to sequence coverage and are large at modest coverage levels. Although this power limitation is appreciated, modestly sized pool-seq GWAS lacking unambiguous hits are often interpreted as showing a polygenic genetic architecture. We illustrate that this inference is unwarranted using empirical data from a Drosophila zinc resistance mapping study. Despite achieving >700× sequencing coverage in case and control pools, a directly ascertained SNP-based GWAS failed to reveal clear evidence for major-effect loci. A unique feature of the dataset is that an advanced intercross multiparent population, with known founders, was employed as the GWAS population. We leverage this unique population structure, in a manner that would not be possible in an outbred panel, to carry out two additional GWASs using imputed haplotype- or SNP-frequency estimates, which in contrast uncover localized regions of major effect. The key difference between approaches lies in statistical power: directly ascertained SNP counts have errors inversely proportional to sequencing coverage, whereas known founder imputation-based approaches can be considerably more accurate. In outbred populations where imputation cannot be used to obtain more accurate allele frequency estimates, substantially higher coverage than currently envisioned may be required to reliably detect modest allele frequency shifts. This work highlights that insufficiently powered GWAS studies can mask simple genetic architectures and create the illusion of polygenicity through statistical noise alone.
Long et al. (Wed,) studied this question.