Tuesday, June 16, 2009

Another candidate gene association bites the dust   posted by p-ter @ 6/16/2009 09:29:00 PM

In 2003, Avshalom Caspi and colleagues published an influential article (Google Scholar lists it as having almost 2000 citations in 6 years) claiming that genetic variation in the seratonin transposter gene influences how people respond to traumatic events--the particular, in terms of risk of depression. For years, this has been the poster-child example of gene-environment interactions (for whatever reason, finding significant interaction terms like this is the Holy Grail of human genetics for some). Like the more recent dubious breastfeeding-IQ-genetics story (led by the same group, it should be noted), the authors identified a phenotype they wished to study (depression), an environmental factor that plays a role in the phenotype (traumatic events), genotyped a couple markers in a gene they thought might reasonably be expected to play a role in that phenotype (seratonin), and found a "statistically significant" interaction. Voila.

The 2003 article, as I noted, received quite a bit of attention. This led to attempts to replicate it, and this week, a comprehensive meta-analysis was published of those studies. The result: nothing. There is no evidence for an interaction between genotype at the seratonin receptor and trauma on risk of depression. And in retrospect, why should there be? The probability of happening on the proper combination of genotype and environmental exposure when sampling one environmental exposure (out of an infinite number) and a few gene markers (out of millions) is miniscule--the statistical burden of proof should be much higher than a simple p-value cutoff of 0.05.

These sorts of candidate gene association studies were/are used in all fields, but in my mind are lent the most credence in psychiatric genetics (the place where they should probably be given the least credence, IMO). This is just an additional cautionary tale: the vast majority of associations found through small candidate gene studies, even ones with functional work, plausibility, and the status of publication in a high-profile journal--MAOA and social problems, FADS and IQ (actually, any study published to date on IQ), NPY and stress--are likely wrong.