Over my time as a dilettante observer of the science blogging community, I have noticed a certain frisson of controversy over the idea of random genetic drift. Sewall Wright, who with Ronald Fisher and J. B. S. Haldane (Bill Bryson’s observations on Haldane’s research into diving and decompression are entertaining) established the science of population genetics, is credited with coining the phrase in 1929. Thanks to Professor Joe Felsenstein for pointing out his seminal paper.
Will Provine has written a much admired biography of Sewall Wright, Sewall Wright and Evolutionary Biology, and yet has expressed doubt about the significance of genetic drift, culminating in his recent work, The “Random Genetic Drift” Fallacy, available as an E book here. This has caused a storm of comment on the internet. At the ID-friendly blog currently run by lawyer Barry Arrington, there was a portentous post by “News” which generated immense interest and four comments that I failed to notice and it was only when Larry Moran criticized Provine’s book recently at The Sandwalk that I became aware of its existence.
In ignorance of Provine’s doubts, I’ve recently expressed my own inability to grasp the significance of the effect for evolution. Allan Miller has been very gentle and patient with me and OM’s computer simulation was most helpful. In the commments, Joe Felsenstein links to his 1971 paper which should give the mathematically inclined food for thought. It just made my head spin!
So I’m hoping that someone can gently take me by the hand and lead me to the promised land where the effects of random genetic drift are clear to all. I have purchased Provine’s book. At less than three dollars and only 180 pages, why would I not?
It also provides a way for enabling mutations to get a foothold.
I’m in some danger of repeating myself, but …
I think it’s worth clearing out of the way a dichotomous view of selection and drift. Evolution is a stochastic process. Nothing is certain – except evolution itself. That certainty derives from the fact that evolution is a sampling process, and finite iterated sampling processes have a directional tendency away from any broad distribution towards a more narrow one. The biased element – selection – takes precedence for historic reasons, but the fundamental process is actually drift.
OM’s model shows the inevitability of fixation with drift alone, but that same process generates an inevitability of change, because it is coupled with new mutation. A refinement to the model would be to have N different starting colours represent the N members of the population, and then to add the capacity for novel colours to be added, singly, to the mix, representing mutation. This model would settle upon a balanced, but shifting equilibrium after a certain period. The initial population of N colours would broaden out to give fewer different alleles with more members each, each set descended from one of the original population members or a more recently mutated version of it. This is standing variation. The presence of this standing variation permits a potentially instant response to a change in the selective wind. The population does not always have to wait for novelty to arise; it already has it. Furthermore, each band of standing variation is probing a different neighbourhood. Further mutation from one or other neutral variant may uncover a selectable double-mutation in a more wide-ranging manner than a monotonous block.
Now, to this picture, we can add selection. Picture the model running on a supercomputer which is continually running origin-fixation series, and tallying and graphing the score. The winning allele in the neutral model can be any of the starting ones with equal probability. Now, you have a knob that you can turn to favour or disfavour an allele. It starts in the neutral position. One click in the positive direction, and fixations for the affected allele increase. Click the same amount in the negative direction, and fixations diminish – but these are not the same degree in either direction. It is much harder for a slightly deleterious allele to become fixed than a slightly beneficial one. Nonetheless, slightly deleterious alleles can reach an appreciable frequency before disappearing, and in doing so they too probe their neighbourhood, and may uncover benefit. This widens the dimensionality of evolutionary ‘search’ substantially, particularly given that every locus is engaged in this process.
When you turn the knob a tiny bit from the neutral position, you aren’t turning selection on and drift off. You are turning one up and the other down. Additionally, the effect of selection is dependent on population size. You have to turn the knob more in a small population to get the same effect.
As far as the ‘selectionist controversy’ goes, it is really an argument about the distribution of s values (selection coefficients) among real mutations. Drift is real and inescapable. And it doesn’t just affect small populations, because real populations don’t mix with the maximal efficiency assumed by the idealised models. Drift can drive divergence. Populations are perennially in the process of breaking apart due to the ‘viscosity’ of gene flow, the process opposing this fragmentary tendency.
Once reproductive isolation has reached a certain level (ie gene flow has slowed to a certain level), the further divergence of the populations could be regarded as entirely due to drift, even if a locus held in common is under selection in both species. An allele which is under selection within a single reproducing population can be treated as such because there is reciprocal recombination which unchains the loci and allows an allele to ‘colonise’ at locus level. But when there is no mating between subpopulations, there is no such subdivision at that level – loci do not recombine. Then, as far as the two subpopulations are concerned, the ‘allele’ is the entire genome, and if the subpopulations don’t even compete, selection cannot be considered to be even in operation at that level, even if it is in operation within the subpopulations.
As to whether drift is ‘really’ inbreeding … they are related. Because drift increases the representation of descendants of one allele in a population, it increases the chance that two copies will end up in the same genome, both descended from the same original. Inbreeding simply speeds this up. A maximally outcrossing fictional population will take longer to achieve this, but the end result is similar – you end up with, effectively, a clone, barring new variation.
Essentially, we are all ‘inbred’ to some extent. We have to be to create viable offspring. We just don’t think of it as such – a decorous remove allows us to find it less icky than close relations. But if our daughter is homozygous for a particular genetic stretch, it must be the case (ignoring the complication of internal recombination) that an ancestor not all that far removed from my wife and I had 2 children, one of which led to the copy in my wife, and one to me. Those children could readily have been of the same gender …
All that said, it is not just drift that increases the apparent inbreeding of a population, so I’m not sure why Provine singles it out.
In thinking about genetic drift and inbreeding, here are a few considerations:
1. There is the issue of whether an individual is inbred relative to its local population. If someone marries their cousin, their offspring will be more inbred than a random individual in their population.
2. There is the issue of whether two randomly-chosen genes from a population are more inbred than two genes chosen at random from the whole species. Thus if genetic drift increases the frequency of (say) allele A as opposed to allele a, two gene copies from that population are more likely to be AA than two genes sampled randomly from the whole species.
3. (Technically we are discussing not “more inbred” but “more identical by descent”, but let this pass for the moment).
4. Genetic drift causes the second kind of inbreeding. It does not necessarily make individuals within a population more inbred than that. It does not, for example, make the proportions of genotypes in a population violate Hardy-Weinberg proportions.
5. I’m back in Seattle, but not yet in my office, and I still need to read what Provine says.
Here‘s an interesting and reasonably accessible paper by James Crow on the relationship (or otherwise) between inbreeding and drift, and the differing views of Fisher and Wright.
I’m struggling with RL currently but did manage to send Professor Provine a heads-up to this thread. No response so far.
Alan Fox,
I believe he may have undergone an operation recently, so there may be a delay in his response.
Thanks for the link. I missed that one.
That was published when Jim had just turned 94 years of age (but of course when it was written he was a mere 93 years old). How many of us will be able to write that well at that age?
Joe Felsenstein,
Yes, thanks, Allan. Accessible, indeed!
What does that even mean, “very inviting”?
Rumraket,
I’d guess [a model of] a process that produces a loss of fitness, even if temporary, is not very [attractive or persuasive].
It really depends upon the depth and length of residency in the ‘fitness trough’. Borrowing from quantum physics (or Douglas Adams), a small amount of temporary relative improbability can be permitted, provided it is eventually paid back. You can’t sail against the selective wind forever (to mix metaphors), but you can for a while.
The ability of a population to sustain a proportion of currently deleterious alleles affects the dimensionality of its evolutionary exploration. If something ‘useful’ lies beyond a trough, it is not necessary for the entire population to go down there in order to find out.
Am I mistaken in thinking that fixation is not necessary in any sense, and that the path a species takes might often involve a serendipitous mutation in a minority allele?
petrushka,
No, you’re not mistaken; that was the point of the paragraph you quoted. If there is benefit on the other side of a particular detriment, then the more members have that detrimental allele, the more likely it is that the beneficial mutation will be hit. It’s not necessary (and is pretty unlikely) that 100% have it.
Although it has to be said that the routine traversal of detrimental regions has more reasonable surmise than empirical support in its favour.
Detrimental is not the same thing as fatal, and it depends on context.
In bacteria, an allele could be detrimental in the presence of a therapeutic drug, but neutral or beneficial in the absence.
petrushka,
Sure, one can construct plausible scenarios, and even find empirical support for specific examples – eg the K76T mutation in Plasmodium appears to be detrimental alone (though not lethal) and yet is an essential component of multi-locus chloroquine resistance pathways. It is likely that drift played a part in allowing the second mutation to meet up with the first despite its detriment.
But the issue is whether this is routine in evolution, or exceptional. There are theoretical objections to its being a widespread mechanism in adaptation. For example …
I’m not trying to make a case that something difficult is routine.
The fact that cherry pickers like Behe have trouble finding examples of edge hopping evolution suggests to me that most evolution does not jump over impediments.
petrushka,
No, of course. But actually, it’s not that difficult either. It is pretty easy to get a deleterious mutation to an appreciable frequency in a diploid population – particularly given the fact that most new mutations are recessive.
I remain neutral on the generality of the mechanism.
Kind of makes a hash of genetic entropy, however.
Allan Miller,
… which is a good example of a scenario where drift and inbreeding are NOT the same. A deleterious recessive can drift by being masked by its dominant allele. Inbreeding cannot do this, because it exposes the detriment.
Death of the fittest.
I imagine each of those dead but fit variants to be an unexplored dimension in Wagner World.
Rumraket,
Like! (As they say in Facebook World)
There is an interesting question about which force is in operation when an allele is increasing. If it is beneficial, we may say that ‘of course’ its increases are due to selection, and its decreases are due to drift. But that’s not really the case. The drunkard’s walk can move towards fixation for stochastic reasons even for a beneficial allele. It may have had no direct effect on individual lives during that period of increase. And it’s often a death somewhere else that pushes frequency up, rather than a directly enhanced survival here.
I once expended an embarrassing amount of energy trying to maintain a separation between selection and drift. I have long since abandoned the attempt, and fortunately the discussion has been erased.
Which would be:
1. Valhalla ?
2. The New York City Mayor’s office?
3. Hollywood?
OK, just kidding, I know it’s where Günter lives (or is it Andreas)? Actually I was just last month at the SMBE meeting in Vienna and was talking to the one of them when the other one came by.
Yeah it really is amazing how far one can take the lion vs antilope example.
Suppose your antilope, which is the fastest antilope there is, (lucky mutations have made it super fast) is out grazing, but it’s raining and the antilope gets mud in it’s eyes by accident, or maybe there’s the crack of lightning and rumbling of thunder in the distance drowning out nearby noises and it just so happens to coincide with the nearby slowest lion (unlucky mutations has made it a sloth) in the pack sneaking up and getting a nice meal of fast antilope. From the perspective of their genes (slow lion vs fast antilope), that antilope should have escaped that lion 49 out of 50 times perhaps under more ideal circumstances. Was it drift or selection?
Wouldn’t that depend on how much reproducing they had managed before becoming dead?
Rumraket,
Ditto for the antelope?
Just read the original Wright paper on fitness landscapes. It’s interesting that he sees Drift as having alleles wander about a mean without fixing. The fact that neutral alleles always fix (or at least, their descendants do) is remarkable and counter-intuitive.
Allan Miller,
fix or go extinct, I should say
Allan Miller,
In Wright’s fitness landscape examples, two forces are at work: natural selection and genetic drift. Fixation can be long-delayed. It will ultimately occur, though this could take a very long time.
If you add in mutation, which reintroduces variability, then fixation is not permanent, and having reached it, the population can escape it later.
Joe Felsenstein,
It certainly will occur, but I don’t read Wright as recognising this. I refer to this: “In a small […] group all gene frequencies can drift irregularly back and forth about their mean values […] without reaching fixation.”.
Larry Moran wrote here (in 2006):
The controversy is over how much of evolution is due to drift and how much is due to natural selection. Excellent arguments have been advanced to prove that most of evolution is due to random genetic drift and that’s the position I take. Thus, in a discussion about the role of chance and accident in evolution I would say that most of evolution is accidental because of the frequency of drift vs. selection. Note that this says nothing about the perceived importance of these mechanisms. That’s a value judgement. Some evolutionists think that adaptation, or evolution by natural selection, is the only interesting part of evolution. These evolutionists don’t deny that random genetic drift occurs; instead, they simply relegate it to the category of uninteresting phenomena. Others, like me, think that random genetic drift is far more interesting than natural selection because drift is responsible for junk DNA, molecular phylogenies, molecular clocks, and DNA fingerprinting.
If that is a reasonable analysis, then there doesn’t seem to be much of a controversy, after all. So; Professor Moran clarifies that drift is important in sections of DNA that have no essential function in the development and survival of the organism and that selection is only important as an aspect of evolution, in which drift plays no positive part.
What are Moran and Provine in disagreement about?
Well, I guess this statement:
Provine suggests in his book, if I read him right, that (whilst there are arguments and models concerning drift) there is no evidence and no experimental work to support them.
Were you there argument. Lensky seems to have experimented with drift. His work was not predicated on finding an unlikely adaptation.
His LTEE is certainly working with bottle-necks. I could be convinced that drift is a factor. I’d need the null hypothesis to be really convinced that drift makes a difference to the LTEE results. I’m beginning to wonder if there could be an analogy with drift and gravity. Difficult to eliminate the effect from your experiment!
Alan Fox,
There is plenty of experimental evidence for drift. As I read Provine (2nd hand, admittedly), he is saying that inbreeding has been mistaken for drift, not that drift never happens. That would be a bizarre position to take.
Drift must happen when s values are small or zero. And it does happen eg – comparative substitution rates at 3rd codon positions (which typically do not change the amino acid) show a substantial discrepancy with the other 2. That’s drift, that is.
Alan Fox,
Why wouldn’t drift itself be the null hypothesis?
I finished reading the book yesterday. Yes, I think you are reading Provine correctly.
After reading Provine’s last chapter, I’ve come to the conclusion that his book is really a critique of population genetics.
Absolutely, if we are speaking of the effect of random sampling for competing alleles for a locus and the tendency over time in small populations for any one of them to win out over time.
Agreed.
Well sure. I have not the slightest doubt about the effect as defined above and demonstrated by OM’s simulation. I’m just (sorry to repeat) not quite seeing what drift achieves. If we could design an experiment that somehow managed to eliminate drift, what would the results look like?
PS to Allan
Forgot to say
And of course there is a distinction between whether the section of DNA is under selective pressure, is functional or not. No drift – no molecular clocks.
I have only got through a little of the relevant sections of Provine. So far I see nothing that affects my understanding of random genetic drift, selection etc.
Folks, genetic drift happens just as random thermal noise happens. It happens all the time except in model populations that are infinite — and they are not in the real world.
That of course does not say whether random genetic drift is important. When we’re dropping cannonballs off the Leaning Tower of Pisa, we can model that extremely well by just talking about height, mass and gravity and ignoring thermal noise. That does not mean that thermal noise does not occur in such situations.
I find all the arguing about whether “most evolutionary change” is genetic drift to be a waste of time. Of course variation in most parts of the genome is neutral, and can be adequately modeled without natural selection. But when you’re asking about change in, say, the malate dehydrogenase gene, that is completely irrelevant. You have to figure out how much change in that particular gene is neutral, how much is selected, and whether they interact (which they may well do).
It is quite likely that most movement of atoms on Earth is due to thermal noise. That does not at all make the case that the fall of that cannonball from the Leaning Tower of Pisa is best modeled as due to thermal noise.
Alan Fox,
It’s not just small populations. New mutations are being added all the time. The bigger the population, the more are being added. The effectively neutral ones are being fixed at the mutation rate irrespective of population size, because population size cancels out of the equation.
It allows a population to have a reserve of variation, which greatly enhances its resilience. If you have a population exposed to a novel antibiotic, and there is variation in the response to it within the population despite it never having been exposed to it, how did that first-level resistance get into the population in the first place?
It increases the dimensionality of the evolutionary search. If evolution were condemned to be ever marching up a slope of increased benefit, all increase being by selection alone, the landscape of possible genetic paths would be severely constricted.
It drives divergence at least as much as selection, between separated gene pools.
Selection is a much stronger force than drift. But drift (and recombination) are important, both within evolution itself, and in the tools used to probe it.
Unless I misunderstood him, Provine is excluding new mutations in what he considers “drift”. He is discussing drift as change that occurs in the absence of selection and in the absence of new mutations. And if drift leads to fixation, that amounts to the loss of an allele, so a reduction in the amount of variation within the population.
Allan Miller,
I’m thinking you need to design an experiment where you could show how drift affected the result by eliminating its effect somehow. But as Joe says, eliminating random thermal noise might be difficult (as Penzias and Wilson found on another issue).
I don’t think anyone, including Provine, rejects the idea that in a theoretical situation in the absence of any selection, one allele will eventually fix and the other competing alleles will be lost.
I like the analogy! But it suggests random thermal noise is present but statistically non-directional. It doesn’t affect outcomes at the cannon-ball level.
I’m not sure who is arguing. I’m certainly not. I haven’t grasped what the differences are yet between advocates and non-advocates of drift as a positive element in evolution. I think I grasp what drift is but I’m still not convinced what it does.
As I said, I like this analogy.
Now, this is a bold claim! How does a process that in the absence of selection will, as Neil says, lead to loss of the alleles that don’t fix, increase the chances of new variation entering the gene pool? This might be the nub of my doubt over drift.
Unless I’ve totally misunderstood what has been said, I’d say that drift has the following attributes or characteristics:
1. Drift will be most pronounced in non-functional DNA, (Maybe this is a big DUH, but Moran also argues that 90 percent of DNA in humans is non-functional, so most evolutionary change will be drift in non-functional code.)
2. Drift doesn’t necessarily do anything, but it is useful as a molecular clock and in tracing lineages. This should be of great interest for evolutionary biologists, because it might provide a more fine-grained picture of lineages.
3. Drift in non-functional code may occasionally result in a new function.
4. Drift may also occur in functional code when alleles are equivalent or nearly equivalent.
5. The net result of all this is that mutations resulting in positive selection will be a minority of the total molecular change.
Neil Rickert,
True, drift is not mutation. But it is wrong to see drift as only significant in small populations, because the number of drifting alleles scales precisely with population size, and because populations do not mix with maximal efficiency.
Drift and selection both lead to reduction in the amount of variation in the population, absent frequency-dependent or heterozygote effects. Selection just does the same thing faster (on average). They are pretty much the same thing – respectively, the unbiased and the biased component of an iterated population sampling process.
I think it’s pointless to try and compartmentalise them. There is simply a range of possible s values, which includes zero. At every s value, fixation will tend to occur, and hence variation eliminated. Apart from s=0 (more generally, where s is less than 1/2Ne where Ne is effective population size), both selection and drift will be involved. A fractional move away from the effectively neutral zone, you get a tiny bit of bias but still a lot of drift. Drift is not merely what occurs in the absence of selection – evolution does not flip to being deterministic as soon as you favour an allele. All increasing s does is increase the probability that a given allele will fix.
Alan Fox,
You can’t turn it off, which is precisely why it is important. It makes evolution inevitable. To get a feel for the impossibility of eliminating randomness, imagine trying to write a working GA with no random component. It would have to always take the same path. What use is that? Or try conducting an exit poll without sample error.