A debate has erupted over the reliability of a high-profile paper published last week in Science. The paper reported a genome-wide scan for variations in DNA sequence that are associated with unusually long lifespan, and found that 150 such variants could be used to predict, with 77% accuracy, whether or not a person is genetically predisposed to living a hundred years or longer.
But some researchers in the field are skeptical. Many are withholding judgment until the work is reproduced in a larger population, such as the European GEHA (Genetics of Healthy Aging) study of over 2,500 nonagenarian sibling pairs.
Furthermore, some, such as Jeffrey Barrett of the Wellcome Trust Sanger Institute near Cambridge, were also alarmed by the large effects these genetic variations seem to have on exceptional longevity. Most researchers in the field are uncovering variants that boost the likelihood of a given condition by something on the order of, say, 1.5 fold, he told the Guardian. But last week’s longevity study claimed to find variants that boosted a person’s chance of seeing their hundredth birthday by a whopping ten fold.
That, Barret said, could result if subtle biases in the work made the test “seem more accurate than it really is.”
Paola Sebastiani, a biostatistician at Boston University and an author on the Science paper, counters that researchers normally use genome-wide association studies to study common diseases, such as diabetes. Becoming a centenarian, however, is a relatively rare event that only 1 in every 6000 people in the United States will achieve. Sebastiani says her team discovered variants with larger effects than researchers are accustomed to seeing because she and her colleagues were studying a more extreme condition than normally tackled using this technique.
Meanwhile, Newsweek quoted genome-wide association guru David Goldstein’s concerns that the study’s control samples were analyzed in different labs and using different technology than the centenarian samples. Sebastiani rebuffs these claims, saying that both were analyzed using chips produced by Illumina. Furthermore, she says, her team fished out the same variants associated with exceptional longevity when yet a third lab reproduced the analysis of the control samples.
Goldstein, kind enough to elaborate in an email late on a holiday-weekend Friday, says yes, both sets of studies were performed using products – called SNP (pronounced “snip”) chips – from the same company (San Diego-based Illumina), but they were different versions of the SNP chips.
A subtle difference, perhaps, but Goldstein says that in his experience, using different types of chips or even the same chips run in different labs can bias the data. Such experiments sometimes turned up variants that seem to be associated with a given trait but in fact result from slight differences in experimental conditions. (This reminded me a bit of findings from Stephan Schuster’s lab at Pennsylvania State University that the different platforms used to sequence genomes were yielding slightly different sequences.)
Because the experiments are so delicate, Goldstein says, his lab always confirms their results using a different method – a step that was missing in the Science paper.
We may hear more on the subject next week: Sebastiani says that given all of the confusion surrounding the study, she and her colleagues have considered holding an online chat next Wednesday to answer questions.
Update (7 July): Sebastiani and co-author Tom Perls (both of Boston University) have just wrapped up their online chat. In light of the concerns about the different platforms used for their study, Sebastiani says her team is picking through their data again. Preliminary analysis suggests that the problem is limited, she said, “but one has to be thorough and so we are examining the effect of these issues on the analysis.”