Digesting GWAS

nce upon a time, medical genetics was a relaOtively straightforward business: one gene—one protein—one disease was sufficient to explain most inherited disease and was neatly in keeping with Gregor Mendel’s discovery of alleles with his famous peas. The concept remains a good starting point but the biology of life is somewhat more complex, and increasingly we have become impressed and perplexed, in equal measure, at the striking variability observed in many inherited conditions, particularly those hitherto believed to be determined by a heterozygous single-gene mutation or DNA variant, and therefore following autosomal-dominant inheritance. A good example is Alagille syndrome, and in this issue of Cellular and Molecular Gastroenterology and Hepatology, Tsai et al proposed a modifier genetic locus to help explain the great variation in severity of liver disease in this condition. The authors reached this conclusion by using a technique that is barely more than 10 years old, namely a genome-wide association study (GWAS), which is part of the arsenal of techniques that has heralded the era of genomic medicine. A GWAS, rather similar to DNA fingerprinting, exploits the immense variation in DNA sequence between individuals, these differences consisting of single-nucleotide polymorphisms (SNPs), deletions, insertions, and copy number variations, the majority of which are regarded as essentially benign. Although a genetic linkage study will set out to identify the rare single-gene mutations or variants that clearly cause a particular disease or phenotype, an association study is potentially useful at finding common genetic variants with a relatively weak effect in more complex disorders. A typical study will adopt a case–control model, comparing 2 cohorts of subjects, 1 with the phenotype or disorder of interest and the other functioning as normal controls. It goes without saying that the criteria for selecting the cohort with the phenotype should be clinically rigorous whereas the control group should be as pure as possible (ie, free of contamination from subjects who might harbor the phenotype). Generally speaking, the larger the number of subjects in each group, the better the chances of achieving significant results. The genomes of all subjects then are analyzed for the presence of at least a million common SNPs, after which the investigation moves to the statistical phase: are there certain SNPs or alleles that are seen more often in the disease group compared with controls, and, if so, by how much (ie, what is the odds ratio?). This is all well and good, but there are pitfalls and shortcomings aplenty. Apart from the crucial aspect of carefully selecting the study cohort by clinical phenotype, the population background should be taken into account lest there are common origins with a range of unique but misleading polymorphisms that give rise to false associations. The normal variation data accumulating through many large-scale, biobank-style initiatives and the Human