Random drift and phenotypic evolution

This week we have a guest post from Markku Karhunen. Markku’s research at the University of Helsinki included the development and implementation of a number of very interesting and useful population genetics methods. In his guest post Markku discusses these novel methods with which researchers can distinguish neutral and non-neutral evolution of quantitative traits.
What does the word ‘random drift’ remind you of? A rowing boat, or a negligent walk through the market street, perhaps. However, there are many instances of this concept also in science. In financial theory, the stock prices are often modeled as randomly drifting, and animal movement could be seen as random drift, too.
In evolutionary theory, people speak of random genetic drift. This means merely the gradual change of allele frequencies from one generation to another. The existence of this phenomenon is trivial; consider for example a lab experiment of 100 generations of fruit flies, with a breeding population of 10 individuals on each generation. Surely you would expect the gene frequencies to change as a result of repeated random sampling.
Things start to get really interesting when we turn to polygenic, quantitative phenotypes. Would you expect the distribution of such phenotypes to change as a result of random genetic drift, or not really? Perhaps the polygenic basis of inheritance forms some sort of obstacle against random change? – Well, this is not a matter of verbal speculation. Recently Ovaskainen et al. (2011) ¹ derived the distribution of quantitative (potentially) multivariate phenotypes under random genetic drift. This was obtained by assuming additive genotypes and ignoring new mutations in course of the evolutionary process, but to my knowledge, it still represents the last word of science.
Practical usefulness
If you can derive the distribution of quantitative phenotypes under random drift, it means that you can say whether any given pattern of differentiation could have arisen by random. This is precisely what our software (RAFM and driftsel) attempts to do. It gives P values for different aspects of phenotypic differentiation under the null hypothesis of random genetic drift.
Of course, this method is not the first one to address the role of random genetic drift in phenotypic evolution. Previously, so-called F_ST – Q_ST comparisons have been used extensively (see e.g. Merilä & Crnokrak 2001 ²). These comparisons ask, whether an index of phenotypic differentiation (Q_ST) is compatible with an index of random genetic differentiation (F_ST), given all uncertainty.
However, our method is much more flexible. In principle, it allows you to focus on any aspect of phenotypic differentiation and test it against the baseline scenario of random change. So far, we have developed this into two statistical tests, the S test (Ovaskainen et al. (2011) ¹) and the H test (Karhunen et al. 2013 ³). The S test asks, whether the population means have drifted too far away from the ancestral source, taking into account their pattern of inter-relatedness, and on the other hand, the presumed (i.e. estimated) amount of ancestral genetic variation.
In my mind, the H test is even more interesting. It asks, whether the population mean phenotypes correlate too much with their environment. Normally, you would interpret such correlations as a case for local adaptation – but then again, correlation could have arisen by random, or because populations found in similar environments often have a shared ancestry. The H test controls for these possibilities by making clever use of the marker data.
Future directions
That being said, I need to state that driftsel and RAFM are not particularly well-suited for interspecific comparisons, because the model has been derived assuming a negligible mutation rate. For the very same reason, SNPs are to be preferred over microsatellites for use with this software. These are obviously issues of interest for the molecular ecologist, and further study is needed to assess the role and usage of different molecular markers in this field of science.
From the conceptual side, a vast landscape of intellectual opportunity awaits us. How can we fit in mutation and coalescent theory in this model? Is the additive genetic architecture the only option, or could dominance and epistasis be integrated in the model, and how does dominance variation change in course of random evolution? Also, the user interface of the software could be more proficient, and figures more beautiful.
Personally, I am not going to develop the model of driftsel and RAFM any further, because I have shifted my interests to epidemiology and biostatistics – right or wrong. However, I of course hope that someone would undertake the task of developing these methods. I am certain that my former supervisors Otso Ovaskainen and Juha Merilä are open to suggestions.
¹ Ovaskainen, O., Zheng, C., Karhunen, M., Cano Arias, J.M., Merilä, J. (2011). A new method to uncover signatures of divergent and stabilizing selection in quantitative traits. Genetics 189: 621-632.
² Merilä, J., Crnokrak, P. (2001). Comparison of genetic differentiation at marker loci and quantitative traits. J. Evol. Biol. 14: 892-903.
³ Karhunen, M., Ovaskainen, O., Herczeg, G., Merilä, J. (2013). Bringing habitat information into statistical tests of local adaptation in quantitative traits: a case study of nine-spined sticklebacks. Evolution, accepted manuscript online: 10 SEP 2013.