August 09, 2012
The idea behind a genome wide association study (GWAS) makes perfect sense. Compare the DNA of one group of people with a disease to another group that doesn’t have the disease, identify the DNA region specific to the disease group, and then find the specific gene and mutations that lead to the disease.
In theory, this sort of study should have become routine once we had the human genome sequenced. In practice, it has turned out to be less useful than everyone hoped.
Now, this doesn’t appear to be any fault with the technique itself. Instead, it has more to do with the fact that many human diseases are simply too complex for GWAS to handle.
Most common human diseases appear to result from multiple genetic pathways and/or multiple genes. Throw in environmental effects and GWAS quickly becomes overwhelmed. At least for now, too many patients and controls would be needed for this powerful technique to have a real chance at deciphering most common human diseases.
But that doesn’t mean the technique isn’t useful. It is very good at finding single genes involved in strongly expressed traits. And this might be ideal for certain model organisms.
In a study just out in the latest issue of GENETICS, Connelly and Akey set out to investigate how well GWAS would work in the yeast, Saccharomyces cerevisiae. In many respects, this yeast appears to be made for GWAS.
It has a small, easily sequenced genome, there is on average a polymorphism every 168 base pairs or so, and its linkage disequilibrium is low. There are genome sequences from 36 wild and laboratory strains publicly available, all as diverse as can be.
But this yeast isn’t perfect. The chromosomal structure between strains tends to be much more varied than between two humans. This is predicted to introduce a high error rate. And this is just what Connelly and Akey saw when they ran some simulations.
They found that the error rate was too high in the simulations to draw any meaningful conclusions. But they also found that by using a more sophisticated analytical technique called EMMA, they were able to partly correct for some of these errors.
Simulations are one thing, but how about real life? Connelly and Akey next tested the method by applying it to a practical problem: identifying the genetic reasons for differences in mitochondrial DNA (mtDNA) copy number in yeast. What they found mimicked the simulation data.
Using more traditional analytical approaches on the data obtained from GWAS, they found 73 potential causative SNPs. But when they switched to analyzing the data with EMMA, they found a single SNP that was significant. It took a bit of hand waving, but the gene associated with this SNP could possibly be implicated in mtDNA copy number. And then again, it might not.
This “significant” SNP was found amidst lots of errors and in a background of high p values. In other words, this finding may not be a real one after all. This experiment does not give confidence that GWAS can be used when all known strains of yeast are compared.
But if the strains to be included are selected more carefully, it may still prove to be a useful tool. When Connelly and Akey focused on strains that were structurally similar, they found that the error rate was much lower. Low enough that in the near term, scientists may be using GWAS to figure out how things work in model organisms.
Hopefully the findings from GWAS applied to model organisms will illuminate disease mechanisms in humans. Then maybe GWAS can realize its full potential, although not in the way it was originally envisioned.
by D. Barry Starr, Ph.D., Director of Outreach Activities, Stanford Genetics
Categories: Research Spotlight