Give GWAS a Chance

From 2007 to 2015, genome-wide association studies (GWAS) resulted in the discovery of over 260 genetic loci associated with obesity and type 2 diabetes (T2D) (1,2). Although GWAS have validated “old culprits” (e.g., PPARG, KCNJ11), illuminated novel disease-causing biological pathways (e.g., the zinc transporter SLC30A8), and have led to quick translational medicine applications (e.g., hs-CRP as a diagnostic tool for HNF1A maturity-onset diabetes of the young) (3–5), some have questioned their utility (6,7). They argue GWAS are expensive and that GWAS-derived single nucleotide polymorphisms (SNPs) explain only a fraction of the heritability for complex traits (6,7). They propose to focus instead on next-generation sequencing (NGS) studies or post-GWAS experiments (functional studies, animal models, clinical studies) (6,7). This debate has led to skepticism about the benefits of GWAS and hesitancy to fund additional GWAS. With the ever-decreasing costs in NGS technologies, we should seriously address the relevance of funding more GWAS. In this issue of Diabetes, Scott et al. (8) provide some elements of response to this important question.

The authors conducted a meta-analysis of 18 GWAS totalling 26,676 T2D case and 132,532 control subjects. They imputed 12.1 million SNPs using the multiethnic 1000 Genomes Project (1000G) reference panel. Twenty-nine SNPs showing promising associations (P < 10⁻⁵) with T2D in the stage 1 GWAS were present in the Metabochip array (9) and were followed up in 14,545 T2D case and 38,994 control subjects from 16 independent studies. Thirteen novel T2D loci in/near the ACSL1, HLA-DQA1, SLC35D3, MNX1, ABO, PLEKHA1, HSD17B12, MAP3K11, NRXN3, CMIP, ZZEF1, GLP2R, and GIP genes reached a genome-wide significant association (P < 5 × 10⁻⁸) with T2D in the combined stage 1 and stage 2 analysis. All study participants were of European ancestry, but the authors followed up those with the 13 novel T2D SNPs in the largest trans-ethnic GWAS meta-analysis published to date (10). No significant heterogeneity of allelic effects between major ancestral groups was observed at any of these loci. Conditional analyses within established and novel T2D GWAS loci revealed two novel independent association signals at the ANKRD55 and CCND2 loci. The authors fine-mapped 95 distinct T2D association signals at 82 loci and identified 99% credible sets of less than 10 SNPs at 16 loci. They used an impressive panel of approaches to bring new pathophysiological insights from the 13 novel T2D associations. These analyses highlighted the importance of cell types such as pancreatic islet cells, adipocytes, hepatocytes, and blood cells in T2D etiology. They also highlighted abdominal fat deposition (NRXN3, CMIP), lipid homeostasis (CMIP, ABO), insulin secretion and processing (MNX1, GIP, GLP2R), immunity (HLA-DQA1), and hematopoiesis (ABO) as key mechanisms in the pathophysiology of T2D.

The main strengths of this study include a sample size that exceeds that of the largest previous T2D GWAS meta-analysis in individuals of European ancestry (11). The authors imputed 12.1 million SNPs using the multiethnic 1000G reference panel, a considerable improvement compared with the 2.5 million SNPs classically imputed from the HapMap reference panel. They elegantly complemented the discovery of novel SNP associations with T2D with a thorough investigation of the biological mechanisms involved.

The study also presents several limitations. Only a subset of T2D case subjects and control subjects with normal glucose tolerance were diagnosed using an oral glucose tolerance test, increasing the risk of case control misclassification. Similarly, not requiring an islet cell autoantibody test for T2D case subjects increases the possibility of misclassifying autoimmune forms of diabetes as T2D. Considering that T2D case subjects usually display higher BMI and waist-to-hip ratios than control subjects with normal glucose tolerance, the lack of adjustment for these covariates in the GWAS meta-analysis is questionable. It may explain why the FTO, MC4R, and NRXN3 loci, three important contributors to obesity traits (12), have been associated with T2D at the genome-wide level of significance in this study, likely via their obesogenic effects (13). Considering the large blocks of linkage disequilibrium observed in Europeans, the fine-mapping of T2D loci in this population is unlikely to isolate the causal variant(s). Trans-ethnic fine-mapping has proven more effective in that perspective (14). Despite the improved quality of imputation compared with previous GWAS, the authors were unable to identify low-frequency/rare (minor allele frequency <5%) variants associated with T2D. However, their conclusion that common variants play an overwhelmingly predominant role in T2D genetic susceptibility may be premature. Several factors explain why no novel low-frequency/rare T2D variants were identified in this study. First, the sample size was insufficient to identify low-frequency/rare variants, especially in the context of a hypothesis-free GWAS (15,16). Second, only a fraction of the rare variants carried by T2D case and control subjects were imputed given the limited density (<1 million) of SNPs genotyped in most GWAS arrays used in this study, the limited sample size of the 1000G reference panel (N = 1,092), and the absence of individual whole-genome sequencing data in T2D case and control subjects (17). Third, single variant tests were used instead of cumulative burden association tests, which decreased statistical power (18). Low-frequency/rare variants associated with T2D have been reported in literature (17,19,20), and there is no doubt that future GWAS in larger sample sizes will uncover at least some of them.

Let’s go back to our essential question. In the light of this cutting-edge study, is it relevant to fund more GWAS? The answer, in my opinion, is four times yes with a push on the Golden Buzzer! GWAS are complementary to NGS approaches and represent a reliable, cost-effective, and user-friendly solution that is exceptionally successful in identifying novel genes and underlying biological mechanisms for complex traits in human populations (1). Future GWAS based on millions of samples, with array data imputed to a whole-genome sequencing data reference panel of hundreds of thousands of individuals, will help to crack the DNA code of T2D susceptibility in diverse ethnic groups (1). GWAS in increasingly large sample sizes will continue to be effective, as the magnitude of the genetic effect is not necessarily correlated with the importance of the pathways underlying the genetic signal in T2D etiology. All I am saying is give GWAS a chance.