A recent paper on the relative number of genes that have undergone positive selection in chimps and humans recieved quite a bit of press (see Razib's comments here, here, and here). The title is quite provocative ("More genes underwent positive selection in chimpanzee evolution than in human evolution"), so I finally gave it a read. Frankly, if you haven't read it already, don't waste your time.

Let's grant the authors their starting position-- that there is a "common belief" that more genes have undergone positive selection in the human lineage than in the chimpanzee lineage (I would argue that this belief isn't all that widespead, though ultimately the reasons for the intiation of the study are irrelevant). In theory, addressing the veracity of this claim is easy-- make a list of the genes that have undergone positive selection along the human lineage, make a list of the genes that have undergone positive selection along the chimp lineage, and start counting. The devil, of course, is in the details.

Due to the fact that not every selected gene will leave a detectable signature, the major assumption of the authors' analysis, then, is that the fraction of detected selected genes along the human lineage is the same as the fraction of detected selected genes along the chimp lineage. That is, if the number of genes that have undergone positive selection in both lineages is the same, but 75% are detected in chimps and only 50% are detected in humans, one might erroneously conclude that more genes have undergone selection in chimps than in humans, while in truth the number of selected genes is the same. This, I will argue, is precisely the mistake made in this paper.

Let's take a look at how the authors identified genes that have undergone positive selection. The basis of the test is essentially the ratio of non-synonymous to synonymous changes in a given gene along a given lineage (non-synonymous changes alter the amino acid sequence of a protein and are presumed to be functional, while synonymous changes do not changes the sequence of a protein and provide a sort of background substitution rate). So if there is an excess of non-synonymous changes (a ratio > 1), one might conclude that the gene has been subject to positive selection. The power of this test to dectect selection is contingent on finding an excess of amino acid-changing substitutions in a lineage.

So what could alter said power? First, it's clear that a single selected site will alter the ratio only slightly, two selected sites will alter it a little more, three even more, etc. So the more selective fixations that occur in a gene, the more power the test will have to conclude for selection. On the other hand, take the number of synonymous substitutions-- if there are more of these, the levels of "noise" are elevated relative the levels of "signal", and there is lower power to conclude for selection.

There is a major difference between historical human and chimpanzee populations that alters the power of the test in the two lineages; indeed, the authors mention this difference without really grasping why it discounts their conclusions. That difference is population size. Humans have historically had a smaller effective population size than chimpanzees and, as the authors note, natural selection is more efficient in a larger population. Thus, advantageous alleles can be pushed to fixation with greater probability, while neutral or deleterious alleles are fixed at a lower rate. So smaller populations should have overall higher levels of substitution (assuming positively selected changes are a minority of all fixations). This is exactly what is seen in the data-- humans have 30,083 synonymous fixations and 19,000 non-synonymous fixations, while the numbers for chimp are 29,644 and 17,701, respectively.

These changes in the rates of allele fixations should lead to a weaker signal of selection in humans, and thus less power to detect it. It's no surprise, then, that the authors find less selected genes in humans than in chimps. Even if the number of selected genes were exactly the same, the relatively stronger signal of selection in chimps should produce exactly the same result. Perhaps the authors want to argue that fewer amino acid changes have been fixed by positive selection in humans than in chimps; this is what population genetics theory predicts, and may be true. However, to extrapolate from a number of amino acid changes to a number of genes is problematic; a single adaptive change in a gene could have major phenotypic consequences without being detected with the sorts of tests employed in this study.

Labels: Evolution, Genetics, Population genetics