If you want a succinct description of how scientists are finding human genes that have experienced natural selection over the past few thousand years, you could do worse than read Ann Gibbons’s three-page “news focus” in this week’s Science, “Tracing evolution’s recent fingerprints” (a bit of a mixed metaphor, that). The article is a useful summary for both biologists and laypeople, especially if you’re curious about whether we’re still evolving. (The answer, of course, is “yes.”)
Starting with recent suggestions that the EPAS1 gene could have been selected for oxygen transport in high-altitude populations, Gibbons sumarizes new statistical approaches (many of them taken here at the U of C) to detecting selection in H. sapiens. As you may know, these analysis have suggested that a surprisingly large fraction of our genome has been under fairly recent selection, with that selection based on adaptation to things like oxygen, diet, disease, and the like.
One of the problems of these studies, as Gibbons notes, is that statistics is not sufficient to show selection: “Finally, few teams have been able to prove that a particular allele actually affects the function of a trait under selection.” I think it’s unwise to say that your case for selection is conclusive without showing that the genetic variants you’re studying make a physiological difference to their carriers. And, of course, the ultimate “proof” of selection is to connect those physiological differences to reproductive output: i.e., that there really was selection.
Gibbons talks a bit about physiological studies (there aren’t many of these), but showing that genetic variants really do affect reproductive fitness is even harder. For one thing, that selection might have occurred in our ancestors, and not be going on so much today. Or, the selection could be very weak, and, though sufficient to cause significant evolution over centuries, might be undetectable in just one or two generations of an experiment. Long-term cohort studies, like those of Steve Stearns and his colleagues, might be useful here, but are still limited if selection is weak. I worry whether the new era of bioinformatics will gull us into accepting conclusions that are based solely on statical analysis of gene-frequency data.