"Waste not, want not." The old proverb is about frugality, but a study published today gives it a whole new dimension. Lead author Sílvia Bonàs, directed by Josep Mercader and David Torrents and collaborating with many colleagues at the Barcelona Supercomputing Center, the Broad Institute, and other institutions (Bonàs-Guarch et al. (2018), Nature Communications 9), decided to investigate variants associated with type 2 diabetes (T2D) by re-analyzing existing GWAS data rather than initiating a new study.
This was a frugal strategy, conserving both time and resources. But the benefits of this approach went way beyond frugality. By aggregating multiple datasets and using unified, current methods for quality control, imputation, and association analysis, the researchers discovered nuggets of significant information that were not apparent in the original analyses of the individual sets. And all of these nuggets are freely available for browsing and searching in the T2D Knowledge Portal (T2DKP).
To amass these data, the researchers combined all of the individual-level T2D case-control GWAS data that were available from the European Genome-Phenome Archive (EGA) and the database of Genotypes and Phenotypes (dbGaP). After harmonization and quality control, data from 70,127 subjects (12,931 cases and 57,196 controls) remained, inspiring them to name the project "70KforT2D".
In the time since the original studies had been performed, better and more comprehensive reference panels for imputation had been generated by the 1000 Genomes and UK10K projects. By using both of these panels for imputation, the researchers were able to substantially increase the number of variants that could be imputed. They ended up with more than 15 million variants, including more than 5 million rare variants and over 1.3 million indels, which have previously been difficult to impute.
In performing association analysis, the authors took advantage of existing large datasets of T2D association summary statistics for meta-analysis, being careful to only combine non-overlapping samples. They also took advantage of the T2D Knowledge Portal to verify some associations for low-frequency variants that were located in coding regions and had suggestive, but not unambiguously significant, p-values. The significance of the T2D associations of these variants was confirmed by meta-analysis along with the associations seen in two large studies in the T2DKP (GoT2D exome chip analysis, with nearly 80,000 samples, and the 17K exome sequence analysis dataset with 17,000 samples).
The association analysis identified 57 loci associated with T2D risk at the genome-wide significance level or better (p-value ≤ 5x10e-8), seven of which had not previously been associated with T2D. The high quality of the data made it possible to fine-map the variants at each of these loci and construct credible sets. Many of the putative causal variants—including those in previously identified loci—were indels rather than single-nucleotide polymorphisms, underscoring the importance of an imputation procedure that discovers indels.
The T2D-associated loci discovered in this study give some tantalizing hints about genes potentially involved in T2D, and suggest new avenues for detailed wet-lab investigation. We can’t review all of them in this space, but one association is particularly interesting for the generalizable lessons it teaches us about case-control GWAS for T2D.
This association, which the authors validated and replicated using additional datasets, involves the X chromosome variant rs146662075. The risk allele confers a 2-fold elevated risk of developing T2D, in males. The variant appears to affect an enhancer that could regulate expression of AGTR2, a gene known to be involved in modulating insulin sensitivity—making it a very interesting subject for investigation with regard to T2D. More work is needed to figure out whether this is really a male-specific effect, or whether it was only detectable in males because imputation for the X chromosome is more accurate in males, who have only one copy of the chromosome.
The first lesson learned from this association is that the X chromosome harbors important loci, and deserves attention in association studies. While this seems obvious, since the X chromosome comprises 5% of the genome, it has been neglected in most studies to date.
The second lesson is that for an adult-onset disease like T2D, it’s very important to pay attention to the details of case-control classification. If there are young people in the control group, they may actually be future T2D cases, destined to develop the disease later in life. When the authors tried to replicate the initial discovery for this variant in different datasets, the associations were not as significant as expected. But after digging deeper into the experimental cohorts, they found that most of the replication datasets had many subjects younger than 55, which was the average age for T2D onset for these cohorts. Re-running the analysis after excluding controls younger than 55 and also excluding those who appeared to be pre-diabetic, based on an oral glucose tolerance test, brought the replication results into concordance with the discovery results and confirmed the significance of the rs146662075 association.
In keeping with the spirit of open access, the authors provided the summary statistics from this work to the T2DKP even before publication. These results are incorporated into the T2DKP and are visible on Gene and Variant pages as well as searchable via the Variant Finder. The authors have also made the full summary statistics available for public download.
The novel and important findings from this study strongly reaffirm the value of data sharing. Not only are data sharing and re-analysis the right things to do for reasons of fairness, equity, and frugality; they can also spark new insights and move science forward in unexpected ways.