Standing and waiting. Photo by Image Zen.Ever since Charles Darwin and Alfred Russell Wallace first described the workings of natural selection, one popular way to summarize about selective change has gone something like this: A population of critters is well-adapted to its environment until that environment changes—maybe the critters move to a new climate, maybe the climate changes on them, maybe some new competitors or predators move in. Life gets harder for our critters, until one of them is born ... different. That lucky mutant has a never-before-seen trait that lets it cope in the new conditions, and in a few generations, every critter in the population is a descendent of that original mutant.
That narrative isn't wrong. But it does miss one of the key insights that led to the discovery of natural selection—natural populations are variable.
That population of critters encountering new conditions of life may very well not need to wait around for the lucky mutant before it can begin adapting to new conditions. Mutations happen at random, and continuously—and, particularly if they don't leave the mutant much less fit, can hang around in a population for generations. And this "standing" variation is raw material waiting for natural selection to act.
High-octane fuel for adaptation
There's good reason to think that natural selection is more efficient when it has standing variation to work with. Joachim Hermisson and Pleuni Pennings demonstrated this principle rather neatly in a 2005 theory paper, in which they modeled the fate of new genetic mutations that had a weak negative effect when they first appeared in a population, but then became beneficial after the population's environment changed.
Normally, when a new mutation appears in a population, it's almost immediately lost to the random effects of genetic drift, even if it confers a benefit. This means that a new mutation needs to be quite strongly favored by selection to have a high probability of "fixing," or spreading through an entire population.
However, under Hermisson and Pennings's model, the mutations considered are only those that survive the initial effects of drift. The flip-side of the randomness that can make a weakly beneficial mutation disappear can also help a weakly deleterious mutation spread, achieving an equilibrium between drift, selection, and new mutation events that create new copies of the same variant to replace the ones lost to selection or drift. So, when conditions changed, and the mutation became even weakly beneficial, it was ready to start spreading.
Natural selection is more effective when it works with standing variants. Figure 1 from Hermisson and Pennings (2005).This graph, the key figure from Hermisson and Pennings's paper, shows the probability that a mutation will "fix," or spread to dominate the population over the course of several generations, given the power of natural selection (alpha, the term on the horizontal axis). The dotted line tracks the probability of fixation for a brand-new mutation; the solid line tracks probability of fixation for a mutation that existed before selection began to act, and had achieved mutation-selection-drift equilibrium. No matter how strong selection is, the pre-existing mutation is more likely to "fix" than the new mutation—and that difference is most pronounced when selection favoring the mutation is weakest.
In other words, if mutations provide the variation that fuels evolution by natural selection, standing variation is fuel with a substantially higher octane rating.
Harder to spot
But the same features that make adaptation from standing variation so much more efficient also act as a sort of population genetic stealthing. This is because adaptation from standing variation has very different effects on the genetics of an adapting population than the spread of a single new mutation.
The key to this difference is that gene variants, or alleles, aren't transmitted from one generation to another one at a time. Instead, they come as part of chromosome regions, physically linked to genetic code that may have nothing to do with the function of the focal gene. And population geneticists use that fact to zero in on genetic regions that might have been recently affected by selection.
It's a little bit like buying LEGO bricks—or, at least, how it used to be when I was still buying a lot of LEGOs, back before you could custom-build your own sets online. Say you want a hundred copies of a particularly special type of LEGO brick, one that's only available in a single kit. To get those hundred bricks, you need to buy a hundred copies of that one kit. So you end up with a selection of bricks—the ones you wanted, and the ones that came with the ones you wanted—that probably doesn't have a very wide diversity of brick types.
But suppose you want a hundred copies of a more common LEGO brick, one that's included in dozens of different kits—kits for pirate ships and castles, race cars and railroads. You might still need to buy a hundred kits, but you can buy many different kinds of kits, and so in addition to the hundred copies of the brick you want, you also have bricks to build anything from a starship to a dragon.
LEGOs, evolving. Photo by Kaptain Kobold.Selection on a single beneficial mutation is like that first LEGO shopping case, where there's only one kit containing the brick you want. The one lucky mutation exists with only one "genetic background" of other, associated genetic code, and so when the mutation spreads through the population, a chunk of that background code spreads with it. (At least, until recombination can separate the favored mutation from its background; that takes time, sometimes a lot of time.)
Just as purchasing a hundred copies of the same LEGO kit would leave an obvious mark on the makeup of your brick collection, a selective sweep that starts with a single mutation—what's called a "hard sweep"—results in a region of genetic code with noticeably lower variation across the population, because everyone is carrying the original lucky mutation plus its associated background.
Figure 4 from Linnen et al. (2009), demonstrates the reduced diversity in a gene region associated with fur color in deer mice. Image from Linnen et al. (2009).In practice, biologists use this principle in two major ways. First, if a biologist has a particular gene in mind that might have recently experienced selection, she can collect DNA sequence data in the vicinity of that gene for many individuals in a popualtion, and see whether it's less diverse than it ought to be. This is how Catherine Linnen and her collaborators demonstrated that a population of deer mice living on light-colored soils in the Sand Hills of Nebraska had experienced natural selection for lighter color. In a study [PDF] I've discussed previously, the team identified a genetic region that was associated with coat color in the mice, then collected sequence data from that region in mice collected from the light-soil population. Compared to the same genetic region in mice from nearby sites with dark soil, the light-soil mice had markedly less variation in the coat-color region.
Alternatively, biologists who don't know which genes might have been targeted by natural selection can collect sequence data from a whole lot of gene regions—or even "scan" the whole genome—and compare the diversity at each region. Any region that has lower diversity than most of the other sampled regions may have experienced selection recently, and is probably a good candidate for follow-up study.
But selection froms standing variation doesn't leave such a clear mark on the genome. It's more like that second LEGO shopping spree, for a brick found in many different kits. If a useful variant is located on many different genetic backgrounds, than selection can make the variant more common in the population without necessarily reducing the diversity of gene regions near the focal variant. This is called a "soft sweep." Soft sweeps present a problem for those of us who want to find genes that have recently been affected by natural selection—without the loss of diversity, genetic regions that have undergone soft sweeps may not stand out in the genome as a whole.
Searching for soft sweeps
As we collect and analyze more genome-scale population genetic datasets, biologists are coming around to the idea that easy-to-detect hard sweeps may be the exception [$a], rather than the rule, for evolution in natural populations—in no small part because the evidence of hard sweeps just isn't there [PDF].
But the absence of hard sweeps doesn't mean that soft sweeps are going on all over the place instead. For instance, in an (ongoing) analysis I presented [PDF] at the recent Evolution meetings in Ottawa, I examined patterns of diversity in genetic regions close to genetic markers that are very strongly associated with differing climate conditions in the small but awesome wildflower Medicago truncatula—and I found little evidence of recent hard sweeps. Does that mean all those strongly associated gene variants are strongly associated as a result of adaptation from standing variation? Maybe; but some portion of the associations could also be due to population genetic processes like drift and isolation-by-distance—I'm still thinking about ways to kill the soft sweep hypothesis.
Pennings and Hermisson followed up their original theory paper with a study comparing the power of several different statistical tests to detect soft sweeps, and they found some promising results with an approach based on linkage between genetic variants in the vicinity of a favored variant. More recently, Pennings has approached the question of adaptation from standing variation from a somewhat different angle, by studying selective sweeps in human immunodeficiency virus, HIV. The evolution of HIV after it infects a patient, and as it adapts to antiviral drugs, is quite well understood—to the point that virologists know to expect particular mutations to sweep the viral population within a patient who starts taking a particular drug.
In an analysis recently published in PLoS Computational Biology, Pennings found that the virus's evolution of drug resistance could be based on standing variation in about 6% of patients on a standard anti-viral drug cocktail—which is to say, about 6% of all patients carry viral populations that are primed to evolve drug resistance the moment therapy begins. (Pennings's lab website has a good explanation of the clinical implications of this result, with video, even.)
Then, at the Ottawa Evolution meetings, Pennings presented [PDF] an examination of HIV genetic samples taken from multiple patients undergoing antiviral treatment. She identified cases when the virus's adaptation to the drugs was fueled by standing variation or based on a mutation that occurred after the drug treatment started; one resistance mutation evolved to fixation via a soft sweep in eight out of 23 patients. [Correction, 6 Aug 2012: See Pennings's comment below for a correction on this point; it's not known whether this particular soft sweep started from standing variation, or whether it's simply the case that two different mutations with the same effect managed to sweep the population together.]
If evolutionary biologists want to understand how natural selection helped make the living world we see around us today, it looks like we're going to have to learn to love soft sweeps. We're still learning how to differentiate the aftermath of soft sweeps from the results of other, non-selective processes. But fortunately, we live in an era when the genome-scale data that may let us untangle this question are increasingly easy to collect.◼
I started working on this post quite a while before the Ottawa Evolution meetings, when I was pleased to meet Pleuni Pennings for the first time. If there are mistakes in what I've written above, they're my own; but I hope she'll let me know if I've made any!
Flintoft, L. (2011). Human evolution: Sweep model is swept away. Nature Reviews Genetics, 12, 228-9 DOI: 10.1038/nrg2978
Hermisson, J., & Pennings, P.S. (2005). Soft sweeps: Molecular population genetics of adaptation from standing genetic variation. Genetics, 169 (4), 2335-52 DOI: 10.1534/genetics.104.036947
Hernandez, R. D., J. L. Kelley, E. Elyashiv, S. Melton, A. Auton, G. McVean, G. Sella, & M. Przeworski (2011). Classic selective sweeps were rare in recent human evolution. Science, 331, 920-4 DOI: 10.1126/science.1198878
Linnen, C. R., E. P. Kingsley, J. D. Jensen, & H. E. Hoekstra (2009). On the origin and spread of an adaptive allele in deer mice. Science, 325, 1095-8 DOI: 10.1126/science.1175826
Oleksyk, T. K., M. W. Smith, & S. J. O'Brien (2010). Genome-wide scans for footprints of natural selection. Phil. Trans. Royal Soc. B, 365, 185-205 DOI: 10.1098/rstb.2009.0219
Pennings, P.S. (2012). Standing genetic variation and the evolution of drug resistance in HIV. PLoS Computational Biology, 8 : 10.1371/journal.pcbi.1002527
Pennings, P.S., & J. Hermisson (2006). Soft sweeps III: The signature of positive selection from recurrent mutation. PLoS Genetics, 2 DOI: 10.1371/journal.pgen.0020186.eor
Pritchard, J. K., & A. Di Rienzo (2010). Adaptation—not by sweeps alone Nature Reviews Genetics, 11, 665-7 DOI: 10.1038/nrg2880