Last year, at the American Society of Human Genetics meeting in Hawaii, Nick Eriksson gave a talk on 23andWe, the scientific arm of 23andMe. He reported a series of genetic association studies that used their customer base as a pool of active participants, by asking them to fill out questionnaires, and correlating their answers with their genotypes. They reported a few novel genes that were associated with semi-amusing/semi-serious traits, including a variant associated with curly hair, and the olfactory receptor variant that predicts whether or not you can smell asparagus in your own urine; these associations have since been published in PLoS Genetics.
This is all very interesting, but that’s not really what I want to talk about here. What I do want to talk about is the effect that knowledge of these results can have on the users of personal genomics, and how this could feed back into genetics research. Stay with me, I’m going somewhere with this.
In Hawaii, Nick Eriksson quoted an interesting result involving the gene ACTN3, and specifically the variant rs1815739, associated with endurance during exercise; they tested whether this variant correlated with whether customers considered themselves sprinters or endurance runners. Now, 23andMe also has a section on their customer website that talks about ACTN3, and informs individual customers of whether they have the endurance or the sprinter variant. Being relatively smart people, the researchers recorded which individuals in the study group had and had not viewed this page. While those that had not viewed the page showed no strong association between genotype and reported runner type, those that had viewed the page and knew they had the sprinter variant were significantly more likely to report that they were sprinters (and vice versa).
So, the self-reported runner type was influenced by what people believed their DNA told them about their runner type. Knowledge of your genotype can change your self-reported phenotype? Interesting, but perhaps not that surprising, and not really a problem for research, as we can just exclude these individuals, or, if they are too common, use objective measures that don’t rely on self reporting (heart rate, metabolite concentration, etc). But, I can’t help picturing a pair of customers completing the 23andWe survey, getting up and thinking “I’m an endurance runner… maybe I’ll run to Crystal Palace and back this afternoon!”, and “I’m a sprinter: I’m definitely trying out for Striker next season”. Think about how different these two people could be, physically, in 12 months? 23andMe may have created a genetic feedback loop.
Where could this lead to, especially when you take into account the influence of the pseudo-scientific reporting of genetics in parts of the media? Could the Daily Mail, by reporting a dodgy “Gene For” story, actually create a genotypic association between, say, OR51E2 and unemployment, by declaring those with a defective form of the olfactory receptor to be “more likely to be failures” or somesuch? What happens when it spreads out into families, creating strange genotype-phenotype associations by proxy? “We always went swimming as kids; Dad liked it, as he thought it would help prevent the muscle degeneration he thought he’d get because of his coding deletion in LMNA“.
We are talking here about creating an entirely new channel of genotype-phenotype causality, one that goes from Genotype, directly to Behavior, and then finally to Physical Traits; this will get even more complicated with psychiatric traits. Perhaps, ultimately, genetic researchers should be treasuring the current generation, who were born early enough to grow up while genetic testing is still rare enough to have no real influence on phenotypic development.
So this has the quality of a horoscope, is that what I’m to take away from this? Heh.
This post reminds me a bit of something I always told my family: I’m so glad that I was a child before the video camera became so ubiquitous. I’m sure the memory of my dance recitals is far superior to their actual quality. More data would not be better in that case.
But I think there’s something more insidious possible too. Parents are going to pressure their kids to sprint or to distance running. Or moms will get chastised on the mommy boards for not supporting their child’s genomes appropriately.
This is fascinating Luke. I wonder what implications this might have for future studies like those presented in 23andMe’s PLoS paper, or other studies using self-report for phenotyping, especially as access to personal genetic information soon becomes a cheap commodity rather than a luxury only available to a few. This is especially relevant in studies of social and psychological disorders. What otherwise mentally healthy person wouldn’t feel at least a little on edge after finding they “have the gene for” social anxiety, bipolar, or major depressive disorder?
@Stephen: It’s unlikely there will ever be a “gene” for the disorders you mention. And I think most people are perfectly capable of understanding that their genetic code doesn’t read in black and white. The sooner we get started down the path of educating the lay public about how our DNA plays a varying role in our health, the sooner we’ll get past these days of hand-wringing over how to disseminate the information. Transparency, as always, is critical.
On the subject of self-report, this is an issue on the front end data collection side of genetic discovery, not on the reporting side where people are learning of new, validated associations (whether made using self-reported data or not, and whether or not they were participants in the study itself). There are undoubtedly some phenotypic questions that individuals can’t reliably answer (sprinting capability being a good example since most of us have no appropriate frame of reference). For those traits that CAN be reliably self-reported, using web-based surveys has the chance to greatly expand the size and variability of study cohorts, as well as having the potential to add supplemental data around other comorbid traits that were unknown, since responders have the ability to opt into surveys of their choosing, rather than the researchers controlling which questions are posed. Bottom line, no research model is perfect; using a variety of approaches seems most prudent.
@Linda – agreed on both points. Psychosocial traits like those are especially likely to be influenced by lots of genes interacting with environmental effects that are difficult to measure, and the “gene for x” stories often don’t do this complexity justice.
Take home message – self reporting is not an accurate measure of behavioural phenotype or self-perceived phenotype? I think this exhemplifies the need for correlating quantitative measures with qualitative genotypes in order to maintain objectivity when using these resources.
Good point Luke, this could be really a problem in the future and we should think about alternative ways to assess phenotypes in order to do accurate associations. Maybe for some traits self-reporting is not the best option. For example, when you want to find associations with a psychological trait you should not ask directly if the person has the trait or does not have it: it would be better to propose a test in which people with the trait would respond differently from people without the trait. It’s only an example, but it shows how you could assess phenotypes in a less direct way.
@Mike: Didn’t mean to suggest that behavioral phenotypes can’t be self-reported, whether done in front of a computer or as a response to questions from a physician or other reaearcher. There are some studies indicating that people may report sensitive information more accurately to an electronic survey than to a human conducting it. Computers don’t judge and don’t react either positively or negatively to responses.
This would only be applicable to borderline traits that had a pretty weak genotype phenotype correlation. No one is going to imagine themselves out of phenylketonuria or blue eyes.
So this is like a genomic hypochondria?
@Mike
Its important to understand under what situations poor self reporting will and will not be a problem. In the general 23andWe case, self reporting creates noise, and thus decreases power; however, this can be got around with larger sample size, and is thus not a major problem (it’ll just delay findings until more samples can be collected). However, the problem I was talking about in my post could be a bigger problem, as it could cause spurious correlations that will only get worse with sample size.
@ABC
Interestingly, I have borderline blue/green eyes, as do a lot of people; no doubt knowledge of our own genetics could bias on which side we fall. And all quantitative traits will be susceptible to this, especially if knowledge of genotype can cause a behavioural change that then feeds back into the QT (think of weight, for instance – knowing you have an FTO variant causes you to eat more because you believe you feel less full than other people).
Luke, sure, but that is my point: you are borderline in this trait. In other fields the phenomenon is variously known as confirmation bias or availability bias or anchoring. There is a pretty obvious fix to this in our field, for at least a very large subset of traits, but the proof is too small to write in this margin.
Will self-reporting research be replaced with passive participation in the study of soft traits?
Here’s where I see us headed…
Some college students could build recreational applications that incorporate their genetic data as avatars in their social networks soon. A few years from now entities like Zynga, IMVU and Facebook could have a lot of data with powerful metrics to back-up potential associations for psychological traits. In fact, a persons entire web usage could be used in research for genetic associations. Add to that their location data (for environment) which is also being increasingly collected via the web and the research potential (as well as potential for abuse) is huge. The potential sample size for these sorts of studies and that the information will be available real time is mind boggling.
What will the “lab” be a few years from now?
Along these lines…
http://www.cell.com/AJHG/abstract/S0002-9297(10)00362-9
I feel compelled to note that Luke’s post has been sitting in draft format for several months, and it is pure coincidence that it was published exactly one day before that ASHG paper on the same topic!