Science Editor

MONTREAL – It may be the annual meeting of the American Society of Human Genetics. But scientists here devoted a lively discussion yesterday morning to what will be needed in addition to genetics to wrest greater clinical significance from genomewide association studies (GWAS).

GWAS have led to an explosion of known genes that are associated with complex traits. In 2007, a paltry seven such associations had been reported in the scientific literature. By the end of 2010, that number had risen to more than 1,200.

But that rise, to date, has not been accompanied in any sort of simple way by a corresponding increase in understanding about the biology of those genetic variants – let alone therapeutic approaches.

Partly, of course, that is because the genetic variants are so new. But at a session titled "Beyond Genome-Wide Association Study: Integrating Transcriptome, Proteome, and Pathway Data in the Genetic Dissection of Complex Traits," the University of Chicago's Nancy Cox started out by arguing that, even now, "there's a lot more function that we can bring to genetics and genomics studies."

At the plenary, several speakers highlighted that part of the reason such biology has been hard to extract is that the sheer number of statistical tests performed means that most associations will miss the statistical cutoff for being considered significant, meaning that such studies are susceptible to false negatives – which occurs when a study fails to detect a relationship between a gene variant and a complex disease because the relationship is not strong enough to rise above the statistical noise.

Oftentimes scientists are more concerned with the opposite type of error, when a study "finds" an association that turns out to be nothing more than a statistical fluke. But while the egg-in-the-face aspect of such Type I errors makes researchers more gun shy about them, Type II errors amount to leaving a treasure trove buried.

One way of mining that trove is by combining GWAS data with other types of data. In the talks at the symposium, Pennsylvania State University's Marylyn Ritchie described her lab's Analysis Tool for Heritable and Environmental Network Associations, or ATHENA, which combines data from GWAS with biological information on transcripts, proteins and protein interactions from public databases, and even variables such as structural imaging data and education.

In a related approach, Trey Ideker, of the University of California, San Diego, described combining SNPs with protein-protein interaction data gleaned from yeast. Adding in such data, in one sense, would seem only to make the haystack larger.

"DNA is certainly the easiest to measure," Penn State's Ritchie said, and once it has been measured, "for the most part, DNA variation doesn't change," while RNA and protein levels, and interactions between proteins, vary over time and between tissues. "It's time that we start embracing variation at all levels," she said.

Another approach, described by moderator Cox, is to look at genetic variation, not at the SNP level, but at the level of a "functional unit." Most of the time, she added, such a functional unit would be a gene – but it could be a pathway, as well.

In her talk, Cox described combining different SNPs to look at whole genes and how they influence the response to chemotherapy drugs. One advantage of doing so, she said, was that testing by the protein rather than for each individual SNP reduces the number of statistical tests from literally millions to a much more manageable 13,000 or so. That reduction of the number of tests, too, makes false associations less likely, and so allows weaker associations to be detected.

Testing by the gene, Cox and her team were able to identify four genes that influence the response to DNA-damaging chemotherapy agents cisplatin and carboplatin.

Some of the genes her team identified are involved in DNA repair, as one might expect from genes that affect the response to DNA-damaging agents. "Perhaps this is not a surprise," Cox said. "But it's not the sort of signal or biology you could learn looking at SNPs."

Indeed, Ideker said, the use of pathways and networks in one sense is "returning us to an era of the candidate gene approach," rather than the unbiased approach that was once thought to be one of the great strengths of whole-genome sequencing.

And one audience member had what might amount to a near-heretical question at a genetics meeting: "At what point do we start considering proteins and transcripts the functional unit, rather than genes?"