How do you really feel, Dr. Wakeley?


I'm currently working my way through John Wakeley's book on Coalescent Theory. (The website has a few pre-publication chapters if you want to take a peek.) In his introductory chapter, Wakeley introduces the concept of gene genealogies. He's careful to point out that, unlike the phylogenies we construct using inter-specific data, we don't actually use intra-specific gene genealogies to infer the relationships of the sequences we've sampled:

Readers used to "tree thinking," which is the subject of Section 1.1, will have little trouble seeing the close connection between gene genealogies and patterns of variation in a sample. It will be clear in what follows, however, that there is a danger in carrying over too much from the field of phylogenetics. Gene genealogies are by their nature unobservable, and they are treated as random variables in a statistical setting. This is necessary because genetic reproduction in a finite population is a random process. Sources of randomness include simple Mendelian inheritance as well as other factors, such as the effects of weather, predation, and competition on the birth and death of individuals.

Thus, while in phylogenetics the tree structure itself is significant, within species it is often the case that particular gene genealogies provide little information about population-level processes and events. These issues are manifest in recent controversies surrounding the use of nested clade analysis (Templeton et al., 1995; Templeton, 1998) or, more broadly, about differing sensibilities between the fields of intraspecific phylogeography (Avise et al, 1987; Avise, 1989, 2000) and coalescent theory. See Knowles and Maddison (2002) and Hey and Machado (2003) for two perspectives on these issues.

Basically, Wakeley is calling out two high-profile researchers for their role in blurring the line between gene genealogies and phylogenetics: Alan Templeton and John Avise. Avise gets credited for creating phylogeography, while Templeton is responsible for nested clade analysis. Both of these approaches have classically ignored the role stochastic processes play in shaping intra-specific genetic variation. In doing so, practitioners of phylogeography often invoke just-so-stories to explain the demography history of their species of interest.

Now, this isn't the first time a population geneticists has questioned the statistical rigor of phylogeography (Wakeley provides references for further discussion of the issue, but also see Beaumont and Mahesh's Panchal's critique). In fact, in Rasmus Nielsen's review of a different book on coalscent theory he also pointed out the necessity of applying coalscent theory in understanding demographic history:

One of the important insights we have gained from coalescent theory is that the same population history may generate very different gene trees if repeated and that very different historic scenarios may sometimes generate gene trees that are surprisingly similar.

Perhaps we're in need of a t-shirt to wear at next year's Evolution Meeting:


More like this

Scott Edwards made the same point at ESEB last year.

Tell you what, you wear that t-shirt in Idaho in June, and I'll wear the same one (or a different one with the same design) in Turin in August.

If you wear that shirt, it may behoove you spell "coalescent" correctly, lest the phylogeneticists mock you mercilessly.

As a phylogeneticist who sometimes poses as a population geneticist, I agree. It should be noted, however, that some phylogeneticists (e.g., Wayne Maddison) have long been aware of gene tree/species tree issues. What you're seeing among "good" phylogeneticists (like Scott Edwards) now is a lot of excitement over the ability to actually incorporate stochastic processes into the sort of evolutionary hypothesis testing that we used to do with NCA and other such methods. Coalescent theory is definitely seeping into phylogenetics, and the software is out there now to allow us to do some of this stuff a lot more easily. Remember, there are a lot of phylogeneticists who are put off if you ask them to use any software package that's less user-friendly than PAUP*.

I'm looking forward to snagging Wakeley's book myself. I had to chuckle at this, though:

Gene genealogies are by their nature unobservable...

The overall point is well-taken, and I don't mean to nitpick, but isn't this true of all genealogies not generated in the lab or a computer, including all phylogenies?