Holm et al¹ first identified a signal comprising multiple low frequency SNPs (risk allele frequency [RAF] 0.01-0.26) on chr14 through GWAS and imputation of 1000 Genomes Project² variants into a study sample of 792 cases compared to 37,585 controls. To refine this association 7 cases and 80 controls were whole-genome sequenced at a mean depth of 10x. Eleven million discovered variants were then imputed into the full set of GWAS samples using long-range haplotype phasing approaches³, and strong association was established with a missense mutation, R721W, in MYH6 (RAF 0.0038, OR 12.53). No significant association remained at chr4 after accounting for R721W. The authors subsequently validated the association by directly-typing this variant in the discovery samples, and by replicating the association in an independent modestly-sized set of 469 Icelandic cases and 1,185 controls (RAF 0.0021, OR 12.59). R721W is not present in the HapMap or 1000 Genomes Project data and was not identified in any of the 1,776 European-descent controls or 135 US cases investigated¹. The lifetime risk of disease is 50% for carriers and 6% for non-carriers and the R721W sibling recurrence risk ratio (λ_s) for SSS is 1.52, substantially higher than most complex disease common-frequency risk loci. This is likely to be an Iceland-specific variant, estimated to have arisen approximately 30 generations ago.

The work of Holm and colleagues¹ applies a novel and powerful approach for complex trait association studies (Figure 1), representing state-of-the-art at least until WGS costs enable sequencing of the full study sample. For example, simulations⁴ suggest that sequencing 200 individuals from an outbred European-descent population at 6× depth can provide near-comprehensive coverage of variants down to MAF 0.01 and ~40% of variants with MAF between 0.001 and 0.01. Imputation approaches can then take advantage of variants discovered within the study sample subset, in conjunction with publicly available reference sets, to infer untyped variants in the full dataset using GWAS data as a scaffold. Indeed, this represents one of the key aims of large-scale investments to create repositories of whole genome sequence variation, such as the 1000 Genomes² and UK10K (www.uk10k.org) Projects.

Holm’s study also highlights the added value to be gained by focusing on well-characterised sample collections with deep phenotype data. When information on multiple quantitative or dichotomous traits is available, sequencing a representative, carefully-selected subset of the study sample (for example to maximise imputation efficiency by preferentially targeting distantly related individuals for the WGS set) can enable downstream association testing for a wide variety of phenotypes.

A potential drawback is uncertainty about the generalisability of findings in further populations. For example, it remains to be seen whether different variants within MYH6 are associated with SSS outside of Iceland. Nevertheless, the detected association, even if not directly transferrable beyond the study sample population, provides novel insights into the biology of disease. It is widely accepted that one of the main reasons behind decades of unsuccessful candidate gene studies has been poor understanding of the underlying disease aetiopathology, a point clearly-illustrated by a host of unexpected GWAS findings. Importantly, to establish the biological underpinnings of pathogenicity, functional studies tailored to the phenotype under study are required.