I am currently in Oslo (Norway) at "Centre for Ecological Synthesis and Evolutinary Synthesis" (CEES) where I have been visiting theoretical evolutionary biologist Thomas F. Hansen. Also visiting here are evolutionary geneticists David Houle and Gunter P. Wagner. So it has been an interesting group of people to discuss, as you understand.
I talked to David Houle about the paradox that the molecular markers explain only a tiny fraction of total genetic variation in human height (see my previous bloggpost), whereas classical quantitative genetic studies (based on covariances between relatives) indicate that the real amount of genetic variation is substantially higher (between 80 and 90 %). David Houle interpreted this discrepancy very much as I did: it provides support for the so-called "infinitisemal model" in quantitative genetics, as formulated (among others) by Russel Lande.
In other words, many loci, perhaps the vast majority of all coding genes, influence human height, but each has a very small contribution in terms of percentages. This is not surprising as many independent genetic factors are likely to influence "condition" or growth, each giving a very small contribution. This makes the prospects for molecular genetic association studies quite bleach, since this approach can only (because of statistical power issues) only detect those few factors that have a relatively large effect. We are left with a situation where quantitative genetic approaches "capture" more of the existing variation than do molecular association studies that fail to detect all these small genetic factors.
This also means that we should perhaps be aware of the fact that many beatiful genetic association studies and studies of candidate genes such as the melanocortin receptor (MC1R) studied by Hopi Hoekstra's laboratory might be untypical and not representative for the vast majority of quantitative traits governed by many different genes (height being one of them). This is not surprising to me, however, it also shows that there is a lot of interesting work remaining to be done and a lot of theoretical and empirical challenges ahead.
Very interesting posts. I would like to offer a bit of a moderating perspective, just for fun. As you mention, Erik, the studies on human height suggest that many, many loci are involved, and although not always explicitly stated, many quantitative genetic models assume a very large number of loci of very small effect. This is an empirical question, of course, and it doesn't have to work that way. Indeed, many QTL studies have revealed a smaller number of loci of major effect (for instance, Schemske's studies of monkey flowers, but there are loads of other studies). I remember reading a few years ago (I don't remember the source) a paper where the question was being asked: Given such results, is the infinitesimal model so unrealistic as to be of little value? I remember at the time wondering how the breeders equation could work as well as it does if the assumptions regarding genetic architecture were generally seriously flawed. But it is an interesting argument, and worthy of approaching with an open mind. Perhaps QTL studies that fail to explain much variation are harder to publish, too? At any rate, I'll make a (actually very safe) prediction: the number loci controlling traits will be a continuum, and sometimes the standard QG models will be a perfect fit, sometimes an OK fit, and sometimes models that include a small number of loci of major effect will be needed. If I remember correctly, Lynch and Walsh, in the preface to their tome, mention that QG and molecular genetics are not at odds (there will be no winner or loser), and that the tools for synthesis are largely already in place.
ReplyDeleteShawn:
ReplyDeleteThe problem with QTL-studies, as I indicated above, is the issue of statistical power:
Only QTL:s of large effects have a real chance of being detected as contributing to explaining trait-variation. A locus which explains only a small fraction of variation will almost never be detected in a QTL-scan, i. e. there will be a VERY high Type-II error rate. This will, IN ITSELF, bias the results towards QTL:s of large effects, especially if they tend to published more than "negative" results.
Regarding the issue of genes of major effects, and their impact on quantitative genetics, I would say that they do certainly not make the quantiative genetic approach invalid as a tool to estimate parameters through e. g. parent-offspring comparisions. The potential problem arises when it comes to PREDICTING the evolutionary response to selection. This is because if there are some genes of large effects and the genetit architecture is not "infinitisemal" the expected mutational input on a per-generation basis will be lower.
This is easy to understand, as the mutational input per generation will be more or less directly related to the number of loci influencing the trait. With fewer loci of major effects, there will be fewer new mutations per generation ("all else being equal"). However, there is no problem at all, in principle, to estimate heritabilities if the trait is governed by a few QTL:s of large effects, the statistical principles of parent-offspring covariance are still valid (as they are for single-locus traits).
So, the presence of QTL:s of large effect is (maybe!) a problem when predicting the long-term evolutionary response to selection (as this will require new mutational input that replaces the variation removed by selection each generation), but it is NOT a problem (at least not a very big problem) when it comes to estimating quantitative genetic parameters per se.
This is a very interesting and enjoyable discussion. Erik, I feel that some of your feelings are biased by a need to say "HA, in your face" to all those who considered QG dead in light of new molecular techniques. They have, obviously, been proved wrong. Sometimes established techniques still have lots to add in the face of new and exciting times. It is also important not to snap back and push in the opposite direction. QTL studies still have a lot to bring to the table, they may not be the holy grail, but equally they are not a piece of toast with jesus's face on it.
ReplyDeleteJust had an interesting chat with Steve about this (I have sent him the link, and he may add a comment of far better construction than this), and he pointed out that QTL will struggle to explain standing genetic variation within a population due to any gene of major effect being pushed to high frequency. Frequency dependent selection (and dominance in the case of Hopi's system) will play major roles in the effectiveness of any QTL study. I hope this makes sense, I am struggling to understand it all myself as it is monday afternoon here.
ReplyDeleteif you guys are interested in these sorts of question and on the implications of processes of standing variation on G and other quantitative genetics parmeters, i would advise you to read this paper: Agrawal et al. 2001 Genetica "possible consequences of genes of major effect: transient changes in the G-matrix" this is a theoretical study and a bit tough to understand at points but very interesting i think....
ReplyDeletecheers
fabrice
Great to see this discussion taking off! The Agrawal-paper I read quite a long time ago, but it is certainly relevant to this discussion. And I certainly think that QTL-studies have their merit, although neither that approach nor any other approach (including QG) will solve all the issues by themself. As usual, methodological pluralism will be more successful than methodological dogmatism.
ReplyDelete