Baby steps versus long jumps: The “size” of evolutionary change, and why it matters

Evolution can make leaps—but how frequently? Photo by Flavio Martins.

ResearchBlogging.orgDoes evolutionary change happen in big jumps, or a series of small steps? The question may seem a little esoteric to non-scientists—how many mutations can dance on the head of a pin?—but it has direct implications for how we identify the genetic basis of human diseases, or desirable traits in domestic plants and animals.

That’s because the evolutionary path by which a particular phenotype, or visible trait, first evolved in a population is closely related to the genetics that underlie the trait in the present. Phenotypes that arose in a single mutational jump will probably remain connected to one or a few genes with large effects; phenotypes that evolved more gradually do so because they are created by the collective action of many genes. So what kind of evolutionary change is most common will determine which kind of gene-to-phenotype relationships we should expect to find.

In an excellent recent review article for the journal Evolution, Matthew Rockman, a biologist with the Department of Biology and Center for Genomics and Systems Biology at New York University, makes the case that the era of genomics has, so far, been much too focused on finding genes of large effect. Fortunately, Rockman also sees the beginnings of a new movement towards acknowledging the importance of small-effect genes—one which may ultimately make genomic association studies more useful.

A tale of two lamps

To understand how phenotypes created by many genes of small effect differ from those created by one or a few genes with large effects, consider two different designs for a reading lamp.

First, there’s the design you’re likely most used to, with one standard-issue compact-fluorescent bulb, turned on by a single switch. Depending on how the position of that one switch, the lamp creates two conditions: enough-light-to-read or not-enough-light-to-read. That’s basically what you want in a reading lamp. This setup is so simple, you immediately know what position the switch is in by how much light there is.

But what if the lamp had not one big bulb, but twenty little ones; and not a single switch, but one for each bulb? To use this lamp, you would have to turn on individual bulbs until, collectively, they generate enough light to read by. With this lamp, what counts as “enough” light (and, by extension, “enough” turned-on bulbs) is now a complicated question. Depending on the ambient lighting around your favorite reading chair, you probably don’t need to turn on all twenty bulbs. And you don’t have to turn on the same set of switches to get a given amount of light; there are thousands of possible ways to turn on just ten out of twenty bulbs!

Shedding some light on the subject. Photo by neoktisma.

Some phenotypes are like that first lamp. We call these “Mendelian” traits, recalling the flower color and seed shape traits in pea plants, which Gregor Mendel used to untangle the basic rules of genetics. In these traits, different forms, or alleles, of one gene (or just a few) make big differences in what form a trait takes. It’s easy to work out how the gene that creates the visible trait operates, just from observing the trait in several generations of close relatives, as Mendel did.

Other phenotypes, like body size or tolerance for heat, are more like the second lamp. They defy easy categorization into off/on dichotomies, and while we can see that these “quantitative traits” have a genetic basis by comparing parents and their offspring, the pattern of that inheritance is fine-grained: a tall mother and short father do not (on average) have uniformly tall or short children. We can tell that such traits have a genetic basis because of this parent-offspring association—children’s traits are more like their parents’ than expected by chance or explained by shared environmental factors—but this association doesn’t tell us much about the specific genes involved.

Of course, Mendelian traits and quantitative traits aren’t really discrete, binary categories, but different ends of a continuum from phenotypes created by single genes with large effects to phenotypes created by the accumulated small effects of many genes. Biologists have recognized this ever since R.A. Fisher demonstrated [PDF] that quantitative traits are simply what you get when many Mendelian genes contribute to a single measurable trait. (Razib Khan explains this nicely, in detail.) However, we’ve been less certain about which end of the continuum is most common in the living world. As Rockman makes clear, whether we live in Mendel’s world or Fisher’s is an important question, in part because it determines how we interpret the results from one of the most widely-used analyses of the genomics era: the genome-wide association study.

The trouble with GWAS

Back to those two reading lamps: Imagine, now, that I’ve put one of them under a lampshade, so you can see how much light it sheds, but not how many bulbs create that light. On the wall next to the shaded lamp, I’ve mounted a hundred switches. Any number of the switches might connect to any number of bulbs under the lampshade. Your job is to figure out which switches need to be flipped on for the lamp to produce enough light to read by.

Sounds like a fun way to spend an afternoon, right? This is a little bit like the problem a genome-wide association study tries to solve. Only there are millions of possible “switches”—genes—and it’s almost always impossible, not to say impractical, to systematically switch them all on and off and see what effect each has on the light, or phenotype of interest.

Instead, you can take a sample of individuals, measure the interesting phenotype for each, then take genetic fingerprints using thousands, or millions, of markers scattered across the genome of every measured individual. (High-throughput sequencing methods make this more practical every day.) You can then perform a statistical test asking whether the version of a marker—the allele—at one spot in the genome is significantly associated with one form of the phenotype or the other. That is, do individuals with allele A tend to have a higher risk of cancer (or grow taller, or produce more fruit, or grow finer wool) than individuals with allele B? You then repeat that same statistical test for each of your thousands of genetic markers.

Readers of the geeky comic xkcd will recognize that this approach creates a substantial multiple testing problem.

As you perform more iterations of the same statistical test, you become more likely to have results that pass the threshold of statistical significance by chance, rather than because they actually are significant—a “false positive.” Statistical significance is, after all, just a way of assessing how likely you are to observe a given result by chance, and the more times you roll the dice, the more likely it is that one of those rolls will land on double sixes. To compensate for this effect, it’s necessary to increase the threshold by which you judge a result to be “significant.”

So here’s the catch-22 of GWAS: To maximize the chance that some of your markers are in regions of genetic code that help determine the value of the phenotype you’re studying, you need to use as many markers as possible. But as you add more markers to the analysis, you have to increase the threshold of statistical significance, so that a marker must have a greater and greater effect on the phenotype before you can be sure it isn’t a false positive.

In short, if your favorite phenotype is Mendelian, GWAS will probably find the one or two genes that determine that phenotype. But if you’re studying a quantitative trait, many of the genes that determine that trait’s value—maybe most of them—will have effects too small to show up as significant in GWAS. This effect is a major cause of “missing heritability” [$a], cases in which many GWAS studies of traits known to be heritable (based on, for instance, parent-offspring comparisons) have nevertheless turned up “significant” markers that explain only a small fraction of the total genetic basis.

As Rockman explains in his review, conventional GWAS methods—and most other modern methods of identifying the genes underlying interesting phenotypes—can’t tell us anything about the size of gene effects in general, because they just aren’t able to detect genes of small effect. And on the other hand, the genes of large effect they do pick up often fail to explain a lot of the genetics underlying interesting traits. So a potentially large fraction of the genetic code underlying many interesting traits remains hidden to us, even though we know it’s out there—the dark matter of inheritance.

Rallying the small effects

Fortunately, biologists are devising ways to find the missing heritability. Most of these move past the question of “significance” of individual markers’ associations, and instead look at phenotypes associated with many markers taken together.

For instance, a study of human height—the epitome of highly heritable quantitative traits—explained a large portion of the genetic variance underlying the trait by considering almost 300 thousand markers simultaneously [$a]. A study of schizophrenia [$a] went a step further: the authors identified all markers in their genomic dataset that had any detectable association with schizophrenia, used them to successfully predict the probability of schizophrenia in an independent sample of genotyped individuals, and then additionally showed that the same set of markers did not predict the probability of several unrelated, non-psychiatric diseases.

How tall you grow is heavily determined by your genetics; but many genes contribute. Photo by woodleywonderworks.

Or consider human longevity, another trait that looks very quantitative even before you start studying its genetics. Indeed, a recent study based on whole-genome data from two 114-year-olds didn’t find any genes of large effect that differ between these super-centenarians and folks with shorter lifespans. But a second, larger, study by the same group of collaborators found that a suite of 281 genetic markers were, collectively, associated with living past 100. And this makes sense. Aging is a phenomenon that affects every part of the body, all in different ways. It’s not surprising that long life isn’t a product of Mendelian “X-men genetics,” because there are so many different metabolic and physical traits that have to work together to keep a person alive for 114 years.

Another recent genomic study, this one of adaptation to local climatic conditions by the lab-friendly plant Arabidopsis thaliana [$a], demonstrates what can happen when your phenotype of interest turns out to be created by many genes of small effect. The authors identified hundreds of genetic markers that are more likely to occur in plants collected from, for instance, warmer or cooler parts of Europe. To validate their results, they grew plant lines collected from across Europe in a single experimental plot at one of the sampling locations. They then compared the number of seeds produced by each plant to the number of markers it had in common with Arabidopsis plants that are native to that spot. Plants sharing more markers in common with the natives did, indeed, make more seeds—that is, the locally associated markers were also adaptive. But it was possible for a plant to carry more than 100 (out of about 130) locally adaptive markers and still produce almost no seeds! Some of that is probably because of random effects—but some of it is probably because it takes a lot of genes acting together to make a plant well-adapted to local conditions.

How to stop worrying and learn to love quantitative traits

So it’s quite possible—I would say likely—we will never find “the gene for” running a four-minute-mile, or answering I.Q. test questions well, or developing depression, or being attracted to members of the same sex. Nor for whatever your favorite measurable trait is, especially if that trait is reasonably complex. Does that mean the study of genetics is futile? Not at all.

However, it does mean that in the ever-nearing future, when you go to your family physician for a genome scan, she may very well not give you the widely imagined list of “genes for” a suite of diseases and conditions, with check marks next to the variants you carry. She may instead give you a report based on statistical models taking into account millions of individual genetic markers. That report will probably not treat a lot of the conditions it examines as binary “sick”/”not sick” questions, either. The movement towards analyzing genes of small effect has prompted folks beyond evolutionary biology to consider that a lot of diseases may be better understood as extreme values of quantitative traits.

A world full of quantitative traits created by genes of small effect is also one in which it’s easier to understand how phenotypes that seem to be disadvantageous can persist in the face of natural selection. As pointed out by the authors of the schizophrenia study described above, when many genes each contribute a little bit to a disadvantageous trait, the strength of natural selection against the disadvantageous allele at each gene is greatly reduced—the selective cost is effectively shared by many genes. With selection weakened in that way, natural populations also maintain more genetic variation, which is useful if they experience changing conditions and new selective pressures.

It also means that we should understand variation as a key characteristic of living populations, especially our own. And isn’t it that very diversity—in the shapes of our bodies, and in the workings of our minds—that makes life interesting? ◼


Fisher, R.A. (1918). The correlation between relatives on the supposition of Mendelian inheritance. Transactions of the Royal Society of Edinburgh, 52 (2), 399-433

Manolio, T., et al. (2009). Finding the missing heritability of complex diseases. Nature, 461 (7265), 747-53 DOI: 10.1038/nature08494

Hancock, A., Brachi, B., Faure, N., Horton, M., Jarymowycz, L., Sperone, F., Toomajian, C., Roux, F., & Bergelson, J. (2011). Adaptation to climate across the Arabidopsis thaliana genome. Science, 334 (6052), 83-6 DOI: 10.1126/science.1209244

Linnen, C., Kingsley, E., Jensen, J., & Hoekstra, H. (2009). On the origin and spread of an adaptive allele in deer mice. Science, 325 (5944), 1095-1098 DOI: 10.1126/science.1175826

Plomin, R., Haworth, C., & Davis, O. (2011). Common disorders are quantitative traits. Nature Reviews Genetics, 10 (12), 872-9 DOI: 10.1038/nrg2670

Purcell, S., et al. (2009). Common polygenic variation contributes to risk of schizophrenia and bipolar disorder. Nature, 460, 748-52 DOI: 10.1038/nature08185

Rockman, M. (2012). The QTN program and the alleles that matter for evolution: All that’s gold does not glitter. Evolution, 66 (1), 1-17 DOI: 10.1111/j.1558-5646.2011.01486.x

Sebastiani, P., Riva, A., Montano, M., Pham, P., Torkamani, A., Scherba, E., Benson, G., Milton, J., Baldwin, C., Andersen, S., Schork, N., Steinberg, M., & Perls, T. (2012). Whole genome sequences of a male and female supercentenarian, ages greater than 114 Years. Frontiers in Genetics, 2 DOI: 10.3389/fgene.2011.00090

Sebastiani, P., Solovieff, N., DeWan, A., Walsh, K., Puca, A., Hartley, S., Melista, E., Andersen, S., Dworkis, D., Wilk, J., Myers, R., Steinberg, M., Montano, M., Baldwin, C., Hoh, J., & Perls, T. (2012). Genetic signatures of exceptional longevity in humans. PLoS ONE, 7 (1) DOI: 10.1371/journal.pone.0029848

Stern, D., & Orgogozo, V. (2008). The loci of evolution: how predictable is genetic evolution? Evolution, 62 (9), 2155-2177 DOI: 10.1111/j.1558-5646.2008.00450.x

Yang, J., Benyamin, B., McEvoy, B., Gordon, S., Henders, A., Nyholt, D., Madden, P., Heath, A., Martin, N., Montgomery, G., Goddard, M., & Visscher, P. (2010). Common SNPs explain a large proportion of the heritability for human height. Nature Genetics, 42 (7), 565-569 DOI: 10.1038/ng.608