The (Illusory) Rise and Fall of the "Depression Gene"

Big psych news of the day is that a big JAMA study debunked the "depression gene" -- that is, this big new study (by Risch et alia, in JAMA, today) found that, contrary to a famous earlier big study (Caspi et alia, in Science, 2003), the short ("bad") form of a particular gene called 5-HTT does NOT make a person more vulnerable to depression. Or, to flip it:: Caspi 2003 had found that having a short version of 5-HTT, which affects processing of serotonin, put someone at more risk of depression if they experienced (as adults) repeated stressful life events. Risch 2009, crunching data from a bunch of studies (including Caspi 2003) to ask the same question -- Does the short version of 5-HTT make you more vulnerable to depression if you suffer stressful events as an adult? -- found the answer was No.

The headlines are predictable enough, "Sad News for Depression Gene" being perhaps the funnest.

But wait; not so fast. Has an empire crumbled here? A hypothesis evaporated?

You need only look at this briefly, I think, to see that the question addressed by both papers is fairly limited, and does not, crucially, cover variations in how early life experiences might amplify any risk conferred by the short 5-HTT allele. (Caspi & Moffitt clearly did not include such events in theirs excluded such early experiences from some of their analyses, and in fact took measures in some of their measures, such as removing from analysis anyone who suffered depression before age 21, that would be likely to exclude some people who suffered particularly rough early years.. And unless I missed something in reading the Risch paper, it too makes no effort to look at early experience in particular -- and, since it pulled Caspi's data from Caspia, would reflect the same possible filtering out of such early-onset depression cases.)

This failure to look at early experience is important, for amid the very large body of studies of gene-environment interactions and how genes and experience might combine to heighten risk of depression, the starkest effects have been found in studies that look at very early experience. Those studies, in both animals and humans, appear to find that very stressful early years -- tumultuous times, lousy parenting -- seem to heighten the sensitivity of genes such as 5-HTT that have been more broadly associated with risk of depression and other problems. The most significant activator of these "risk" genes, in other words, seems to be very early experience; and the Risch paper -- for perfectly good reasons, as it is seeking to answer a simpler question -- doesn't address those studies at all.

So what we have, really, is a big paper -- Risch 2009 -- that appears to refute another big paper (Caspi 2003), that is but one leg supporting a broad model of gene-environment interaction as a risk for depression. So what happens if you kick out that leg? Does the house fall? Not if it has a bunch of other legs under it. In fact, it may be that kicking out that leg merely will show you that the leg was in fact an early bit of scaffolding, and that the house will stand quite well without it.

The Risch paper will doubtless spur a lot of critical looks at the body of work on gene-environment dynamics in psychiatry and behavior. This can only be a good thing. But to declare dead both the "depression gene" and an entire model of risk seems rather premature.

Not that that's stopping the press. The press in general is rushing to declare, in sensational fashion, its earlier simplistic sensationalism erroneous, and I've little doubt that many lay readers will conclude, Hmm, that business about genes putting you at risk for depression (or other ailments) is hokum.

That said, among the most valuable takes so far are Ben Carey's at the Times (which acknowledges lightly some of the caveats I spell out); a short piece at the Wall Street Journal Health Blog; and Constance Holden's at Science.

The media, as you quite rightly say, have been quick to jump on this story with headlines proclaiming the original finding of the Caspi paper "flawed" (I'm looking at you, NYT). I agree this all has a touch of the sensationalism about it, and unfortunately many readers will not be able to take the time to draw their own conclusions.

I do agree on the whole with what you write, but I do have to take you to point on a couple of issues. Firstly, according to the Caspi paper, there is a report of the effect of early stress x 5-HTT on depression using a measure of childhood maltreatment for ages 3-11, and it shows that the results are the same as for later life stress. Secondly, the individuals that are removed because they were depressed before 21 were removed for one of the analyses only, that looking at first depressive episodes after age 21. For all the other analyses there is no indication they were taken out. Early stress certainly seems to be an important factor in depressive risk and is postulated to involve brain development, and the animal work in the area is showing that early stress and 5-HTT variation is an important factor in later behavioral differences.

I have some alternate reservations about the Risch study. Firstly, the authors' claim that they aimed to re-analyse only the data that matched the original study the closest. From their manuscript, they list 14 studies which constitute the meta-analysis dataset. These 14 studies are by no means equal. For a start, the demographics vary considerably. The youngest sample set is age 12-19, the oldest is older than 65- the original study sample was 26, and I'd like to guess that the symptoms, expression and underlying biological profiles of depression differ throughout the life-span. On top of that, at least one study was not Caucasian in origin (Kim et al., Korean elderly individuals). Different ethnic groups have different genotype frequencies for many markers (population stratification), which disrupts the underlying genetic linkage within gene regions. This means that often you cannot compare directly samples derived from different ethnic populations, especially if you do not know for sure that you have honed in on causal variant. Also, none of the studies were selected based on the similarity of their stress measure. The way 'stress' is measured varies widely from study to study and there doesn't seem to be a consensus on how it should be classified. Also the study designs differ so that some stress measures are in-depth and follow individuals over a long period of time, others are retrospective and require the individuals to remember bad things that have happened to them in the past using short questionnaires, opening them up to recall error (especially if subjects have suffered with mental problems- some people just can't remember all the bad times). The same goes for the depression measurements, some are clinical diagnoses, some are questionnaire-based self reports- are these really analogous? There was an excellent review on all these problems published last year by Monroe and Reid (Psychol Sci 2008 Oct; 19(10): 947-956.), which Risch and colleagues seem to have missed.

You and others have highlighted the neurobiological work with humans and animals that is basically supporting the original finding- SS individuals are more susceptible to depression after stress. I believe that 5-HTT and stress probably do have an interactive role in depression susceptibilty, but that it is by no means deterministic. It certainly doesn't explain all the variation in response- some people who are stressed and have the SS genotype still don't get depressed and vice versa. Instead of trying to undermine findings in the relentless search for tiny p-values, maybe scientists should be embracing the opportunity that these accumulated findings provide to find out what is the biology behind risk for depression.

Thanks for these excellent points, Susan, for your correction regarding the exclusion of early-age experience, and for the ref to the Monroe and Reid paper.

The ethnic group mixing seems particularly problemmatic, for the reasons you cite, as do the stress and depression measures and definitions. These don't utterly deflate the Risch study, it seems to me, but add to the reasons it should be taken provisionally and with a grain of salt.

And I'm glad you raised the issue of whether these alleles are deterministic -- for the general view among researchers seems to be (and the data certainly suggest) that they are not deterministic but probablistic. In addition, the thinking is increasingly that they depend on multigenic interactions -- and, doubtless, mult-environmental interactions, if you will: a complex play of genes and environment. Too complex to ferret out? The danger in the reaction to the Risch study is that people will conclude or argue that this line of research is a dead end, when that is far from apparent.