Filling the Empty DNA

In the sake of fairness, it’s worth reporting that mitochondrial DNA (mtDNA) may not be as flawed as previously thought (see the original paper here). I spent a bit of time earlier this year barking about how mtDNA is a less than stellar marker for inferring demography because it’s subject to recurrent selective sweeps which wipe out all signatures of demographic history. Thanks to Dr. Rob, I revised my position to take into account the effects of population size on selection.

If mtDNA mutations are only weakly advantageous, then they should only be under selection in large populations (this is the essence of the nearly neutral theory). This was brought up by Adam Eyre-Walker in his review of the Bazin et al mtDNA/population size paper. Eyre-Walker pointed out that mtDNA-based estimates of the MRCA of humans agree with those based on nuclear markers. That’s because the human population size is too small for mtDNA mutations to be under adaptive evolution. No selective sweeps means mtDNA can be used to infer demographic history.

That brings us to this paper in which Mulligan et al examine the relationship between mtDNA polymorphism and allozyme diversity (an estimate of nuclear DNA polymorphism, which should be a good proxy for population size) in eutherian mammals. While mtDNA may be a poor estimate of population size in organisms with large populations (insects, for example), it appears to work quite well for small populations (I have reproduced their graph below the fold).

There must be some threshold of effective population size below which selective sweeps on mtDNA are near impossible. This result also implies that there are next to no highly advantageous mutations in mtDNA. But we must also take into account the fact that the effective population size of mtDNA will be 1/4 that of a nuclear gene (it’s a haploid genome and only transmitted maternally — I hope I didn’t screw this up, again). So, a mutation in an mtDNA gene will need confer a greater fitness benefit than the same mutation on an equivalent nuclear gene in order to be fixed by selection.

Also, I still maintain that one should use more than one marker to infer demographic history. Even though mtDNA correlates well with population size for small populations, one should sample a couple more loci before reaching any conclusions about demography.



  1. #1 John Hawks
    December 2, 2006

    Sure, mtDNA coalescence time in humans is consistent with nuclear coalescence times. How could it not be? With Ne=10000 for autosomes, mtDNA has Ne=2500. So to be “significantly different”, the coalescence time of mtDNA would have to be a lot less than 2500 generations, say around 1000 generations. For humans, that’s around 25,000 years ago. This is how recent the coalescence would have to be to trigger the interpretation of selection.

    Now you can see why that’s simple nonsense. This argument that mtDNA is neutral comes entirely from the low power of this test of selection. A coalescence date anywhere between 2 million and 25,000 years ago is quite consistent with the human effective size.

    In fact, there is no human gene that looks selected by this test. That’s how weak it is! By Eyre-Walker’s argument, human evolution didn’t involve any selection at all — every gene is neutral!

    So let’s consider some different tests of selection that don’t have such low power. We don’t have an LD test for mtDNA because it is entirely linked. But we do have the frequency spectrum, which shows that the current mtDNA increased vastly in numbers. That is compatible with rapid population growth and/or selection.

    We have functional variants of mtDNA in humans that are linked to disease. We also have geographic variants that look like they have been selected for greater heat production or ATP production efficiency, respectively.

    None of this necessarily proves that there was a recent selective sweep, but I think you’re entirely right in wanting to see multiple loci for demographic reconstruction. There was a time when mtDNA was cheap and easy and nothing else was, but that time is slipping into the past.

  2. #2 John Hawks
    December 2, 2006

    OK, I’m in danger of piling on….

    They are quite right in pointing out that there is an effective size that is too low for adaptive evolution to be likely. Let’s consider what that effective size is:

    The fixation probability of an advantageous mutation is 2s (Haldane 1927). The fixation probability of a neutral mutation is 1/2N. So selection becomes a more important factor than drift when 2s > 1/2N, or 4Ns > 1.

    In humans, the Ne for mtDNA is around 2500. What does s have to be for 4Ns > 1? The answer is s > 1/10000, or s > 0.0001.

    So for this argument about the impossibility of selection to be true, we have to imagine that no adaptive mutation with s > 0.0001 could possibly have arisen on the mtDNA.

  3. #3 RPM
    December 2, 2006

    Piling on is A-ok.

    There are also a couple of letters in this issue of Science — one by Oliver Berry and one by Wares et al — that are critical of the Bazin et al paper. Bazin et al reply to the five critisisms quite well.

    My favorite criticism comes from Berry, who claims that, because the comparisons are being made between distantly related taxa, that the selective sweeps that affect mtDNA diversity are ancient. This represents a total lack of understanding of how natural selection affects DNA polymorphism.

  4. #4 John Hawks
    December 3, 2006

    Interesting how many mistakes can be easily avoided without reading anything more recent than 1940…