So, the latest (hottest, and certainly this time just must be true) report is that obesity (that is, Body Mass Index, or weight-for-height) isn't so clearly damaging to health and disease as billions of dollars and millions of pages of punditry and scientific hyperbole have suggested over a mere fifty years. Whoopie! We can eat again! What a relief!
Or is it?
The study we posted on yesterday seemed to say that. But even forgetting our usual (of course always cogent and well-placed) reservations about science news bulletins, perhaps there is something else to note, that might cause at least a few milliseconds of thoughtful contemplation.
If this is a causal, material world, then as the argument goes, everything must be understandable and predictable strictly in terms of molecules and energy--because that's all there is! And since, the argument continues, evolution has molded life around DNA as the primary causal molecule, we simply must be the product of, and hence predictable from, our genes.
The BMI study was not a genetic report, and only concerned the predictive power of the net measure, BMI. It was about the long-assumed health risks, or not, of obesity. But there is a bit of slippage here: BMI is easy to measure (your weight related to how tall you are), and so many different studies can collect comparable data, etc. It is thus a convenient measure of choice for obesity.
Taking the current dogma of our time that everything simply must, obviously, necessarily be 'genetic', many studies you've paid for with your taxes have naturally done their best to find the genes 'for' this important health-risk trait.
No, not at all
Thus a major and very large GWAS on the genetics of BMI was published a couple of years ago (Nature Genetics, Nov. 2010). This study of a mere 250,000 individuals found a small number of modest (statistically 'significant') locations in the genome, including one confirmatory gene (called FTO) that a blind person could find without using his hands. Other 'known' obesity risk-factor genes weren't in this list, and of course there is the plethora of excuses--er, that is, alternative explanations--for why these genes didn't show up in the hit-list.
Now that is mysterious enough (unless you've been thinking critically about genetic causation, its evolutionary history, and the nature of such studies), but at least it's a large study that should illuminate at least the nub of the causal truth.
However, also in Nature Genetics in Nov. 2010 was another obesity GWAS. This time the measure used was not BMI, but the waist-to-hip ratio (WHR). This is another convenient, non-invasive, and cheaply measured index of obesity. The study was the pooling of 61 studies of a total of a mere 114,000 participants. Now this study essentially found no overlap in genome region 'hits' with the BMI study! It also failed to show several genes well known to relate to obesity and related dieases, from many actually focused studies including mouse experimental work.
One can rationalize all one wants about this 'discrepancy' (to use a kind word for it). But if 'obesity' is a meaningful trait with any sort of unitary causal nature, then measuring it in two ways should generate essentially the same result, after accounting for statistical vagaries. Just as using a metric (Celsius) thermometer won't tell you anything more about water, ice, and steam than using a Fahrenheit scale.
 |
| 237 traits linked to genomic loci by 1449 GWAS studies (Source: www.genome.gov/GWAStudies); 2012 |
This issues seems not at all to have been noticed (openly, at least) by anybody. Instead, it should be seen as a clear and devastating indictment of the GWAS and related 'omics' grand-sample, meta-analysis, quick-and-dirty enterprise that we are investing so heavily and mechanically in. It should be the miner's canary, telling us clearly that we are not going about this in a right way.
We can't blithely accept the current BMI and health finding as related to obesity in an interpretable way and are rather forced to recognize, as we said in our prior post, that BMI is a stand-in for some confounding factor(s) that may or may not have been measured. That's because if different genes predict one measure of 'obesity' compared to another measure, there must be some seriously complex or heterogeneous causal variation in our data that we are not measuring, may not know about, but are not highly correlated with each other or, at least, are not consistently correlated with different ways we choose to define something as a trait, or risk factor.
'Obesity' is in some ways an obvious trait in its extremes (from skinny to very over-weight), and body weight is clearly related to health measures of various kinds. But the Omics Way that is being taken is falling short, and this also means that the Epidemiological Way, of parsing a large plateful of variables into this or that correlation coefficient with various statistical significance levels, is also badly wanting.
We don't have the answers. Indeed, the problem is not just that nobody has the answers, it's that the only reaction is to claim we need more and more, larger and larger, studies of essentially the same sort to get the answers! But, bigger isn't always better!
3 comments:
"Now this study essentially found no overlap in genome region 'hits' with the BMI study! " ... "But if 'obesity' is a meaningful trait, then measuring it in two ways should generate essentially the same result,"
Not at all. First BMI and WHR are what's being correlated and obesity inferred. The fact that there's little overlap between the two methods indicates that obesity is, like you say later, a multifactorial, complex, polygenic disorder not unlike, say, dyslexia, for instance. There are multiple ways to test for dyslexia, such as homophone spelling and tactile localization, none of which have overlapping genetics. Which isn't at all surprising. GWAS isn't really the problem. Instead more skeptical energy needs to be aimed at how these disorders are measured, the parameters of the population.
"Indeed, the problem is not just that nobody has the answers, it's that the only reaction is to claim we need more and more, larger and larger, studies of essentially the same sort to get the answers!"
Too often in reactionary posts from the science blogosphere there's a great confusion of audience. For instance, is your above broadside directed at the media or people who do GWAS and are all too aware of the caveats and limitations of their research?
I obviously largely do not agree with you, I'm sorry to say. The only reason measures like WHR and BMI (and others) are studied is because they purportedly measure 'obesity' and that is justified because 'obesity' is considered a health risk factor.
It is true that these are different ways of triangulating on an elusive and poorly defined problem, but when the results hardly overlap (among other issues not covered in our post) then it is time to accept this reality and move on to something more productive, rather than perpetuating ever more, ever bigger, ever more costly studies of the same types.
The problem is inclusive: the media who hype this stuff, the investigators who let them get away with it (and do their own hyping, blatantly lobbying for funding), the funders who don't want to lose their grant portfolios, and more.
I would largely agree with you that a central problem at present is to define things better, rather than the largely thought-free, or at least rushed kinds of studies being done. Better definitions could, perhaps, lead to more focused studies that are more likely to reveal something useful--and even that is not guaranteed given the kind of causal complexity we face.
But the proper scientific approach to something that is important and known not yet to be well-defined is: Don't GWAS 'it' until you first figure out what 'it' is!
In this light, my view is that GWAS really _is_ the problem: it is too just easy to set up a GWAS rather than to stop and take a breath and think what better to do instead, or to do cogent focused studies of genes we already know have a clear-cut causal affect, or to do studies with a sharper focus than mindless chasing after countless ephemeral, miniscule effects as so much GWASing and genomewide sequencing do.
For this reason, in my view, people who know what the story is, as you imply they do (which early in the GWAS era may--debatably--have been at least slightly less clear), and yet continue to do the same thing, deserve criticism.
By the way, your comment about 'reactionary posts from the science blogosphere' seems to suggest that critics don't know what they're talking about. I believe it's relevant to say that we are involved in genomic research to try to grapple to understand complex causation in our own lab, and have been for years, so we are not just taking pot-shots from a comfortable scientific armchair.
What is going on in this area is generally par for the course in human societies, because going along with current views is natural, normal, safer, and easier than creative or innovative thinking. We celebrate the latter because of its rarity. But there needs to be criticism to point out the issues, to try to nudge the system to try to take more profitable directions.
One could argue that, with what we know now, those who are actually 'reactionary' are those comfortably in the system who react vehemently to any suggestion that they might have to change.
"In this light, my view is that GWAS really _is_ the problem: it is too just easy to set up a GWAS rather than to stop and take a breath and think what better to do instead, or to do cogent focused studies of genes we already know have a clear-cut causal affect, or to do studies with a sharper focus than mindless chasing after countless ephemeral, miniscule effects as so much GWASing and genomewide sequencing do."
But we have fancy words for "countless, ephemeral, minuscule effects" and its called quantitative trait loci and polygenic effect.
There are plenty of people interested in following up with functionally characterizing these genes so I don't really see your point, but further saying that those who are simply satisfied with detecting associations are part of some problem is facile. Especially when no one is saying GWA studies are the final word on the genetic component of a disorder. Besides, the low-hanging fruit of already discovered genes tend to be over-picked fields in cancer or autism.
What GWAS and other genome wide studies begin to show is that many common variants of small effect do explain the vast majority of genetic effects to phenotype associations * and that we need something more fine-grained such as whole genome sequencing (obviously the next step in GWAS). I say more and better genomics not shifting focus away from it, especially when the clarion call is couched in self-styled fighting the system Matrix-style rhetoric.
*Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature, 2009.
Post a Comment