The X chromsome: WTF?
The X chromosome in humans is something of an exception with regards to the rest of the genome–as it’s diploid only in females, the population genetic forces on it are slightly different. In particular, the effective population size of loci on the X, in a standard neutral model, is 3/4 that of the autosomes. In different demographic models, this fraction can change, so comparing the X to the autosomes is potentially an important tool for understanding human demography.
In a paper published earlier this year, Hammer et al. analysed a data set they had collected of sequences at 40 loci (20 autosomal and 20 on the X) in a number of populations. They saw a striking pattern (the relevant figure from their paper is on the right): in every population they looked at, their estimate of the ratio of effective population sizes on the X and autosomes was greater than 0.75. After additional analyses, they interpreted this as the signature of polygamy in human history.
At the same time, another group (Keinan et al.) was independently looking at this issue in other datasets. Their analysis, published today is markedly different. In particular, they see the exact opposite of the pattern in Hammer et al.–a decrease in the X/autosome ratio in effective population size compared to 0.75 (a figure from their paper is on the right. Note that the y-axis is the same in both this and the Hammer et al. figure–the x/autosome ratio in Ne. In both, the solid horizontal line is at 0.75). . And this is not due to extremely different methodologies–one of the analyses presented by Keinan et al. is very similar to that in Hammer et al., only using different data.
So this is all a bit odd, to say the least.
Labels: Population genetics
Unless I’m crazy, polygamy ought to make the Ne inferred from X chromosomes smaller, not greater. Polygamy increases variance in male reproductive success but shouldn’t effect the variance in female RS. High variance in male RS means many X chromosomes in males will be dead ends. This should decrease the Ne of the male fraction of the X (1/3 of the 3/4), thus making the Ne of the X less than 3/4 that of the autosomes. What am I missing?
males account for 1/2 of the chromosomes on the autosomes.
yes, of course. duh. thanks for setting me straight, p-ter.
Over at Gene Expression Fils Razid noted that the two different conclusions were drawn from different data. I suspect that an analysis using all available data would produce a third conclusion. And that if one included all relevant data a different conclusion would be reached.
We have a habit of selecting that which supports our thinking, and shying away from observations and evidence that contradicts it. This being as true of those working in genetics as in any other field of inquiry.
There are only two directions of skew possible, so I don’t see what third and fourth conclusions you have in mind. And while it’s easy to (metaphorically) stroke your beard and pontificate about confirmation bias, the methodology used is described in pretty good detail. If there is a shortcoming in either team’s work it should be possible to outline it. Unfortunately I can’t see the full paper of Keinan et. al. to make the point by point comparison.
Just thinking out loud, but is it possible that estimators of Ne can exhibit nontrivial sampling variance?
My stats is certainly stronger than my pop gen, but recall the Genghis Y chromosomal incident. Could it be that Hammer et al. oversampled the descendants of polygamists and vice versa for Keinan?
Either that or someone may have a sign error a la poor Geoffrey Chang.
Could it be that Hammer et al. oversampled the descendants of polygamists and vice versa for Keinan?
don’t know, i kind of doubt it. I think it’s more likely to be something like what hawks suggests–different ways of calibrating mutation rates.
keinan et al have three analyses–one of differentiation, one of the allele frequency spectrum, and one of diversity. hammer et al. present data based on diversity only. looking that those other two aspects of the data in the hammer et al. dataset would probably shed some light on where the discrepancy is coming from.