Monday, June 4, 2012

Whole genome sequencing, yes or no?

We all should have our genomes sequenced
Does it strike anyone else as odd that Science has published an article recommending that everyone have their whole genome sequenced for medical purposes, written by the CEO of a DNA sequencing company? Odd as in conflict of interest? Of course the guy thinks everyone should have it done! And surely preferably by his company? How on Earth are we supposed to evaluate his arguments without taking into account his vested interest?

But, ok, let's soldier on anyway. He argues that rather than the ad hoc sequencing now being done for known causal variants or to look for novel ones, "[a] more effective approach might be to routinely sequence individuals' entire genome once, preferably early in life, and to continue to use this information to make health-care decisions throughout their lifetime."

Well, this might in fact be more useful than targeted genotyping or sequencing when the clinician is looking for a novel mutation in a known gene for a Mendelian disease, that is, one known to be caused by single genes. But a decade of experience with GWAS has shown that Mendelian diseases are much less common than polygenic diseases, for which the full suite of associated genes are proving to be difficult to identify. Never mind the full suite, any genes with more than small effect.

But, you say, eventually we'll figure those out, and it will prove useful to have our whole genome sequence.

Well, not necessarily. Not when everyone's genome is unique, and there are multiple genetic pathways it's probably safe to say to every disease or disorder, plus environmental factors, and the environment is impossible to predict. So, if we're going to be able to predict anything with whole genomes, it's most likely going to prove to be just about what we can now predict with much cheaper, much simpler, targeted tests for the maybe 3000 Mendelian diseases for which causal mutations are currently and clearly known.

What about people with apparently genetic disorders whose causal gene is not known? Can't those be identified with whole genome sequencing? Well, there are a number of "single-gene" disorders that are so-called even when a mutation in a known gene is not found in everyone with the same condition. In such cases, often we define the trait in terms of its causal gene (reminiscent of how the Obama administration defines a terrorist as being whoever they kill with a drone strike), leaving others with the same problem but no known gene without a diagnosis, or ignored by research. Or we may make the mistake of thinking the gene can only be causal if its protein-sequence is altered, whereas it's possible the cause is the gene's regulation in some people.

Such traits are often rather rare, with a substantial fraction not gene-specific as far as currently known, and hence many are not well understood, and don't attract much research money. Now, everyone has 3 - 4 million sequence variants relative to the reference human genome, which means that the likelihood of identifying the cause of such a disorder, particularly when it's not familial, even if we had whole genome sequence, is slim.

That is, even if the cause is simple, and genetic, the number of possible variants among different people, in various places in the genome, can be such that it will usually not be possible to identify a causal mutation from sequence data. Exceptions will be found, of course, and they'll be touted in Nature or Science or the New York Times as justifying the overall costly enterprise, when the funds could be much more usefully spent on things already known but without known treatment.

So, if a whole genome sequence can't be used to explain a disorder like one of these, nor to predict common diseases like heart disease, obesity, asthma, etc., and testing for Mendelian disorders is already possible and readily done, then why again should we all be paying people like the author of the Science paper to give us our whole genome sequence at birth?

Not so fast!
All that said, a second paper in the June 1 Science does urge constraint. The authors rightly say that "the vast majority of genomic data is, at this time, not medically actionable." And, the authors further point out, and obviously we agree, that how multiple risk variants combine in additive, multiplicative, or compensatory ways to inform clinical risk is largely unknown. In addition, data from studies of disease concordance in monozygotic twins suggest that for many common diseases, such as cancer, a negative test result from WGS data would not appreciably reduce an individual's risk relative to the baseline population risk and would therefore not enable meaningful medical intervention. The challenge is to identify and validate genotypes via WGS that are robustly associated with disease phenotype and with effect sizes, sensitivity, and specificity that enable counseling about risk beyond what is predicted by traditional clinical factors.

Oh, and look what else they say: "A particular concern is that some of those making claims about the application of WGS may not be considered objective or dispassionate because of their commercial or even academic interests."  Apparently we're not the only people who noticed.

The authors also say that there aren't nearly enough trained interpreters of the data, which will constrain making sense of it once it starts rolling. This, though, assumes that it can be made sense of! (See above.) But, they further point out that clinical reasoning is Bayesian, meaning it depends on prior knowledge of the patient's symptoms. Knowing symptoms can determine what the clinician gleans from a patient's sequence; the same sequence in a healthy patient might be interpreted quite differently.

And finally, they point out that with few exceptions, genetic testing rarely leads to improved health outcomes. Apologists for untrammeled big science regularly argue that, yes, it may be a while, but we'll eventually figure complex causation out using such data and reap its rewards long-term. That is possible but is also possibly a rationale--or excuse--for keeping the taps open. After all, there are other more likely successfully researchable problems that the same funds could make more with.

Note also that the genomic techniques will have most population impact on common diseases that develop during life. It's clear that these mainly require environmental effects, and risk estimates are perforce made after the fact. The same genotype's risk, if any, may be very different in the future because our lifestyles change so rapidly as is very, very well known. Current risk estimates are clearly off, perhaps way off, but we have literally no way to know since we can't predict future lifestyle exposures. So there is a serious built-in problem that nobody seems interested in acknowledging.

There are always nay-sayers and resistance to bold, arrogantly asserted futuristic claims, mostly associated with demand for lots of public resources. Some technologies unambiguosly aid science, and nobody can doubt that DNA sequencing is one of many like that. But that's not the same as saying it should be applied to everything someone has a whim for. Of course, there is huge momentum of all sorts, sincere but also heavily vested, to keep the foot on the accelerator--but the gas being used is yours.

A lot of people were unhappy when Ken suggested in an April post that the push for whole genome sequencing may be fading. Well, if it's not fading, we reiterate: it should at least be tempered, and we're gratified to see that the authors of one of these papers would agree.


Holly Dunsworth said...

Non specialists don't know the difference between sequencing and genotyping. This guy didn't need to have his genome sequenced, for example, but the article makes sequencing out to be the hero: A Geneticist's Research Turns Personal. Genotyping would have gotten that same information to him faster and cheaper. But notice how this is in the BUSINESS SECTION!

Ken Weiss said...

There is sooo much wrong with this article besides your points (which are also correctly noted!) that it's hard to comment.

The lifetime risk of diabetes in the US today is about 33%. Even for lower risk groups (people of the right color and who are handsomely svelt) it's something like 10%. Thus, a story of a 'success' after the fact is absolutely of zero relevance. What we need to know is the relative risk associated with this variant, the conditional reporting of the success (that is, negative predictions not being reported), and we need to take into account what 'diabetes' means in this case.

Diabetes may be diagnosed based on some cutoff blood glucose level, which is a subjective CDC kind of judgment criterion. The measure may change over time. The definitions change over time. And if the guy's treatment worked, by the simple lifestyle changes we all know are relevant to diabetes!

No genetic test was needed or called for, nor did it do any good in this case. With or without the Bad genotype, all he needed to have done was what he's now doing.

But even more seriously than his obvious vested interest and the conditional reporting and definition issue, is that one true success doesn't justify in itself the cost or stress on the (highly profitable to his company) approach.

The real legitimate question is how useful or cost-effective (to use the US's only accepted criterion these days) is it to do this rather than other things? Maybe if everybody biked it would be better for everyone: more jobs for bike makers and fixers, less health care costs, more funds for truly ill people with truly genetic diseases.

But the alternative-approach balance question isn't asked, especially by those who are committed to a particular approach. This is true in general, I think, not just of genetics.

Holly Dunsworth said...

I knew you'd like that one.