Advances in population genetics and genome sequencing have made it possible to identify anonymous fragments of DNA that have undergone selection. This yields some evolutionary answers, and a panoply of puzzles. See Letter p.360
When an affenpinscher named Banana Joe took the top honours at last month's Westminster Kennel Club dog show in New York, the judge referred to his “fantastic face [and] great body”. Some of Joey's success may be attributed to the awesome power of artificial selection, but we sometimes overlook the fact that dogs are also our companions in natural selection, and have adapted to similar changes in lifestyle and nutrient availability over the past 10,000 years or so.
The parallel evolution of humans and our animal companions is apparent in a study by Axelsson et al.1 on page 360 of this issue, in which the authors used a population-genomics approach to identify regions of the dog genome that have undergone selection during domestication. This method promises to revolutionize evolutionary biology, by challenging us to detect traits affected by evolution on the basis of genotype rather than an organism's characteristics, or phenotype. Two other recent papers, published in Cell by Grossman et al.2 and Kamberov et al.3, rise to this challenge and show how hypotheses about an adaptive human genotype can be tested in controlled experiments. Together, the three papers are a wonderful intersection between genomics, population science and experimental genetics — a synergy that has tremendous potential for teaching us more about how and why organisms evolve.
At first glance, the approach taken by Axelsson and colleagues is simple: to understand what distinguishes domesticated dogs from their wolf ancestors, the authors sequenced the genomes of 12 wolves and 60 dogs and identified DNA fragments that show little variation among dogs, but contain a high density of differences between dogs and wolves (a selective signature). They then determined which genes lie in those fragments.
But the devil is in the details of the dog DNA — the features that underlie a selective signature are exactly the same features that occur after a population bottleneck (the reduction in genetic diversity that occurs when the size of a population is rapidly and drastically reduced). To partition DNA fragments selected for during 10,000 years of domestication from those caused by population bottlenecks during the past several hundred years of breed formation, Axelsson et al. created a kind of 'virtual mutt' by choosing 60 dogs that represent 14 diverse breeds and analysing these data as a single large pool. What should emerge from this analysis are not genomic regions associated with herding, hunting or pointing, but regions that distinguish all dogs from their wild ancestors. The authors identified several functional categories of genes enriched in dog 'candidate domestication regions'. They chose to further investigate those involved in starch metabolism, reasoning that domestication was, in part, a consequence of the agricultural revolution and was therefore facilitated by adaptation to a starch-rich diet (Fig. 1).
Indeed, Axelsson et al. report evidence for gain-of-function alterations in dog genes encoding proteins that break down complex carbohydrates (AMY2B), hydrolyse oligosaccharides (MGAM) and help to transport glucose across the intestinal wall (SGLT1). The AMY2B result is particularly notable because it seems that the increased activity of the enzyme it encodes, amylase, has occurred as a result of an increase in the number of copies of AMY2B. Gene amplification is also known to underlie4 the increased activity of human amylase in populations in which ancestors consumed diets rich in starch. Thus, the same molecular mechanism has acted on similar genes in different species exposed to the same dietary pressure — a striking example of parallel evolution.
The authors of the Cell papers have taken a different tack. They focus on ways to analyse the flood of human sequence data that has resulted from the next-generation sequencing revolution. In cases in which it is possible to determine whether adjacent sequence variants lie on the same chromosome or homologous chromosomes, the range and complexity of potential analytical approaches expand5,6 and, together with data on the frequency of specific sequence variants, these approaches can sometimes be used to pinpoint causative variants that underlie recent selective sweeps7. (A selective sweep refers to the reduction in genetic variation that occurs in regions adjacent to a mutation that confers a strong selective advantage.) Grossman et al.2 applied this approach to human sequence data from the 1000 Genomes Project to generate a comprehensive catalogue of several hundred regions of around 30 kilobases long, each of which contains dozens of potential adaptive mutations.
The thousands of selected variants that emerge from such analyses highlight the challenge of 'reverse evolutionary genetics' — determining which phenotypic change has been brought about by a specific selected region. Even when causal relationships seem obvious, caution is warranted. For example, the KITLG gene, which encodes a crucial signalling molecule for the migration of melanocytes (skin cells that produce the pigment melanin), bears a strong signature of selection in European populations, but was not detected as a skin-colour gene in a recent association study of an African–European admixed population8.
Classical genetic studies are the optimal way to establish causal relationships, but in many cases these are impossible because the appropriate populations do not exist. Kamberov et al.3 suggest a way forward using a model organism. Their study focused on a well-recognized mutation in the human gene that encodes the ectodysplasin A receptor (EDAR), a signalling molecule that has a role in the development of hair, teeth and exocrine glands9. The mutation, which occurred around 30,000 years ago in East Asian populations, results in a valine-to-alanine substitution at amino-acid-residue 370 of the protein. It has been associated with increased hair thickness10, and several lines of evidence suggest that it may have other phenotypic effects10,11,12. Kamberov et al. genetically engineered this mutation in mice, and found that the animals had thicker hair, altered mammary-gland morphology and an increased density of eccrine sweat glands compared with normal mice. The authors then closed the phylogenetic circle by showing that the mutation is also associated with an increased number of eccrine sweat glands in a population of Han Chinese. Thus, the phenotypic effect of an adaptive human variant that arose around 30,000 years ago was explored, confirmed and extended to a species whose most recent common ancestor with humans was 3,000 times older than this.
Kamberov and colleagues' study is an exceptional example of experimental genetics, but does it provide, as the authors suggest, a general framework for assessing candidate adaptive mutations? Genetically altered mice are a powerful experimental tool, but the extent to which recent positive selection in humans acts on pathways and amino-acid residues that have been conserved across mammalian evolution is uncertain. More importantly, it is often not clear how to investigate positively selected genomic regions for which the target gene, let alone its action, is unknown. And so a major challenge for population genomics remains the construction of meaningful null hypotheses. As Charles Darwin, the best known evolutionary biologist, once said13, “It is always advisable to perceive clearly our ignorance”.