William Provine and the Biological Meaning of Genetic Drift
I had the pleasure of attending a seminar by William Provine yesterday, in which he presented a history of population genetics theory focusing on genetic drift and neutral evolution. Provine argued that genetic drift is a biologically meaningless fallacy perpetuated to this day by people misinterpreting the early work of Sewall Wright (partly due to Wright’s sloppy terminology). For some of my more informed readers this will come as quite a shock – genetic drift is one of the fundamental concepts covered in any population genetics course (along with mutation, natural selection, and population structure). The antagonistic nature of the material, however, should come as no surprise to those familiar with Provine. I present for you a brief review of neutral evolution, both at the population level and at larger time scales, and a summary Provine’s case against genetic drift.
The blockquoted, red text below provides a brief review of population genetics (specifically, inbreeding, genetic drift, and neutral theory). Skip it if you have taken a decent evolutionary biology or population genetics course.
Before we can talk about genetic drift, we must first discuss inbreeding. Sewall Wright developed a statistic for understanding the effects of inbreeding on homozygosity (known as F). Inbreeding increases the homozygosity of a population because closely related individuals tend to have more alleles in common than more distantly related individuals. Inbreeding is higher in smaller populations because the probability of mating with a close relative increases as the number of possible mates decreases. Later, Wright extended his F statistics to deal with subdivided populations – note the connection between inbreeding and population structure.The first point Provine brought up is the similarity of genetic drift and inbreeding – both are measured by F statistics, both increase in small populations – but I’m more convinced by some of his later arguments, mostly dealing with genetic linkage. In the early treatments of drift by Wright and colleagues, the terms gamete, chromosome, and allele were used interchangeably. It’s true that chromosomes can be modeled like marbles being drawn from jar, and genetic drift seems plausible under this model. The same goes for individual loci, but when we begin to model multiple linked loci, recombination is not strong enough to allow us to assume independence. Also, in reality, gametes are randomly sampled from each individual to go on to the next generation, so modeling alleles or chromosomes is not biologically realistic. The problem with the model is compounded when we focus on nucleotide sites within a gene, and even more so when we look at individual codons. Tightly linked sites cannot drift independently even if both are under no selective constraint. Furthermore, neutral sites are far too close to non-neutral sites to evolve due to genetic drift while a neighboring site is selectively constrained. By modeling alleles in a population as if they were independent chromosomes, early theory on genetic drift could overcome the problems of genetic linkage, but Provine does not think these models are biologically meaningful. Individual chromosomes can undergo genetic drift, but it is implausible to think that loci within a chromosome are independent enough to undergo substantial genetic drift relative to each other.
Genetic drift refers to the change in allele frequencies due to random sampling in finite populations. For example, if we have 10 marbles, five of them blue and the other five red, we have a fifty percent chance of drawing a blue marble and a fifty percent chance of drawing a red one. Now, if we draw ten marbles (with replacement) we can start a new jar which represents the next generation. The most likely scenario is for us to draw five blue marbles and five red marbles, but we may also draw zero blue marbles, one blue marble, two blue marbles, etc. These changes in marble color frequency are like the changes in allele frequency due to genetic drift. Eventually, one allele will be lost from the population, and it’s said that the population has fixed for the other allele. Much like inbreeding, the effects of genetic drift are greater in smaller populations – smaller populations undergo more drastic changes in allele frequencies from generation to generation and are more prone to fixation of alleles due to genetic drift.
If we look at this so called neutral evolution (it’s called “neutral” because it does not invoke natural selection) over larger time scales, we can no longer study populations, but we must compare different species. When the first DNA and protein sequence data became available, it appeared that there were too many differences between the sequences to be accounted for by natural selection alone. This led Motoo Kimura (among a few others) to come up with the neutral theory of molecular evolution. He argued that the majority of DNA substitutions between species were due to random chance alone and not natural selection. This has been confirmed as more sequence data has become available and provides the foundation for the molecular clock (the idea that DNA change occurs in a clocklike manner due to neutral mutations, rather than in an irregular manner as expected if selection drove most sequence evolution). Tomoko Ohta later revised the theory to include slightly deleterious mutations and population size in proposing the nearly neutral theory of molecular evolution.
I probably got a rise out of some folks when I said some sights within in a gene could not evolve by genetic drift while other sites evolved by selection, for this seems to go against everything we know about the evolution of non-synonymous sites, silent sites, and introns. Note that Provine never rejects neutral evolution; he merely does not find a compelling case that genetic drift drives neutral divergence. The effects of genetic linkage are too strong to allow for genetic drift anywhere near a site under selection, due to the overwhelming effects of hitchhiking or background selection. These effects are compounded in small populations, where the population recombination factor (4Nr, or a product of population size and per nucleotide recombination rate) is even smaller. I guess you could consider this analogous to the proof that the rate of fixation of new neutral mutations in a population is independent of population size (the rate of fixation is equal to the neutral mutation rate). Therefore, even though small populations should, theoretically, be more prone to allele frequency changes due to genetic drift, genetic drift becomes implausible in these small populations due to a minuscule population level recombination rate. In fact, Provine thinks that inbreeding, not genetic drift, drives evolution in small populations.
One phrase that really bothers Provine (and is often part of an introductory treatment of population genetics) is the idea that, “Populations drift apart.” This is often accompanied by a description of FST and a discussion of allopatric speciation. Ernst Mayr pointed to allopatric populations diverging (often due to genetic drift) as a major cause of speciation. If drift is implausible, though, we must come up with some other explanation for how populations diverge (and, ultimately, speciate). Now, I’m probably gonna piss off a few people out there by bringing this up, but Coyne and Orr view speciation (especially in sexually reproducing organisms) as a byproduct of divergent selection between two populations. Their hypothesis is well supported by the current molecular and experimental data on speciation, and it jives well with Provine’s anti-drift stance. If drift is biologically meaningless, then we need something else to account for the genetic divergence observed between reproductively isolated populations, and Provine favors inbreeding in small populations.
I know a lot of people are familiar with the evolutionary models that include genetic drift and fit the observed data quite well. Please allow me to remind you that even though a model fits the data, if the parameters don’t make sense, it’s still a crappy model. For example, I could come up with a simple model whereby aliens cause mutations in DNA that drive evolution, but we have no evidence for aliens, so my parameters are bunk, and, hence, my model isn’t worth anything. According to Provine, the same could be said for genetic drift. Yes, binomial (or multinomial) sampling of alleles at independent loci does produce the same patterns of divergence between populations, but all the loci in a genome are not truly independent. As I mentioned before, because many drift models fail to account for the substantial genetic linkage observed along a chromosome, they do not contain the appropriate parameters.
Even the few experiments performed to simulate drift (the most often cited one by Kerr and Wright ), are not reproducing the same process described in the models. First, the experiments proceed as repeated founder events whereby a fixed number of individuals are used in each generation. One could argue that these events are merely reproducing a population of constant size, but if they are not “founder events” the only way the population would remain the same size would be if each individual mating only produced two individuals or if there was intense selection eliminating most of the progeny. Provine assumes that populations of constant size are under selection; therefore all the loci linked to sites under selection cannot evolve by genetic drift. Also, the experiments involve the random sampling of individuals rather than gametes, chromosomes, or alleles, but the models all involve sampling independent alleles. The divergence between these experimental populations cannot be attributed to genetic drift as it is typically modeled because the experiments do not reproduce that model.
Regardless of the limitations of genetic drift to change allele frequencies within populations, Provine still sees a role for neutral evolution on larger time scales. He acknowledges that the majority of nucleotide sites in a genome evolve according to neutral processes. How this process occurs, however, is where he differs in opinion from traditional population genetics theory. Neutral evolution and genetic drift must be viewed as different processes rather than the byproduct of a process (as the former is thought to relate to the later). If genetic drift is biologically meaningless, then some other force must account for what we view as neutral evolution. Provine sees inbreeding in small populations as a better explanation than genetic drift for neutral divergence.
Provine does not want to overhaul the entire field of population genetics. He merely thinks that genetic drift must be reconsidered and evaluated in regards to its biological meaning. The neutral theory is one of main foundations of molecular evolution, and when it’s disproved in a particular case (i.e., using a dN/dS or HKA test) we have strong evidence for selection at the molecular level. The genetic linkage between sites along a chromosome, however, challenges the biological tractability of drift. Uniting the neutral theory with Darwinian evolution, Provine made a nice comparison that I’d like to close with. He said that Kimura’s neutral theory, Ohta’s nearly neutral theory, and Darwin’s theory of natural selection all have their place within the framework of evolutionary biology. The neutral theory describes almost all DNA sequence evolution; the nearly neutral theory takes care of protein evolution; and natural selection provides an explanation for phenotypic evolution. While there may be exceptions (DNA sequences under selection and phenotypes evolving neutral), I see this as a good way to reconcile the different mechanisms of evolution at different levels.