Evolutionary biology is, fundamentally, the study of how populations of living things change over time. Different creatures live different lives, and at any given point in time they seem to do so relatively well, which poses a question: how do you get from one reasonably well-adapted form to another? At first glance, some of the transformations we see in the fossil record and reconstruct from genetic relationships are incredible. Even over millions of years, can there really be an adaptive path from something that looked like a hyaena to a modern blue whale? From a Tyrannosaurus to a mockingbird? From a lily to a Joshua tree?
Consider some trait related to a critter’s fitness, like the thickness of a bird’s beak. If a small beak is good for collecting small seeds, and a bigger beak is good for cracking big seeds, and there aren’t many seeds of intermediate size in the local environment, maybe the relationship between beak size (the trait) and fitness (the ability to efficiently turn seeds into more birds) looks something like this:
It appears that there’s no way to evolve from point A (a small beak) to point B (a large beak) without going downhill, or becoming less fit. This puzzle was the context in which the geneticist Sewall Wright introduced the metaphor of adaptive landscapes, in a 1932 paper for the Sixth International Congress of Genetics [PDF]. Wright considered fitness associated with genetic variants, rather than the observable trait values they create, and pointed out that 1,000 genetic loci, each carrying 10 possible variants, offered 101000 possible combinations of genetic variants. Even if a small fraction of those were well adapted, this would offer an enormous number of possible adaptive “peaks.” But, Wright reasoned:
In a rugged field of this character, selection will easily carry the species to the nearest peak, but there may be innumerable other peaks which are higher but which are separated by “valleys.” The problem of evolution as I see it is that of a mechanism by which the species may continually find its way from lower peaks to higher peaks in such a field. In order that this may occur, there must be some trial and error mechanism on a grand scale by which the species may explore the region surrounding the small portion of the field which it occupies.
Wright thought that the random shifting of genetic variant frequencies in small populations — genetic drift — would be necessary to get across adaptive valleys. We now know that drift is a common, and likely important, source of evolutionary change. But in fact, an additional answer lay in Wright’s original formulation of the adaptive landscape as multidimensional. To build from my earlier sketch, consider that there’s more to a bird than its beak. Maybe birds that are sufficiently efficient fliers can seek out seeds to fit any beak size. If we add that new dimension to my original crude sketch, the valley between small beaks and big beaks turns out to be not an unbridgeable chasm, but more of a cirque, with a path from A to B that never loses altitude, provided flight efficiency (“another trait”) can adapt at the same time.
If you consider that the dimensions of the adaptive landscape are limited only by the dimensions in which living things vary, this logic suggests that there might actually be many paths between phenotypes that look like well-separated peaks when viewed in only two or three dimensions. You might also start to feel a sense of the overwhelming space of possibilities that natural selection sorts through. This is no longer a landscape that Charles Darwin can stroll across on a thoughtful afternoon; it needs more science fictional imagery, like the starship Enterprise warping space to cross light-years in a day, or the Doctor’s TARDIS sliding between geological epochs the way I might walk down the street for a cup of coffee.
Yet it is possible to map the multidimensional geography of adaptive landscapes. In Arrival of the Fittest: Solving Evolution’s Greatest Puzzle (Current, $27.95 in hardback), computational biologist Andreas Wagner provides a brisk guided tour through his own research to understand the full scope of Wright’s “rugged fields.” Wagner does not merely chart adaptation’s course from one point to another — he aims to show how the structure of the adaptive landscape makes it possible for mutation and natural selection to create entirely new types of living things. This, the origin of new adaptive forms, is the “greatest puzzle” of his subtitle.
The logic of Arrival hinges on viewing the adaptive landscape not as a topographical map, but a network. A good example is presented in a 2009 PLOS Computational Biology paper, in which Wagner and coauthor João Mathias Rodrigues modeled the biochemistry of bacteria. Rodrigues and Wagner simulated the core metabolism of Escherichia coli as a set of discrete biochemical reactions. For their purposes, a “genotype” was the list of reactions included (or not) in a hypothetical bacterium. By simulating the combined input and output of those reactions, Rodrigues and Wagner could estimate whether a bacterium could use them to survive on a particular food source — any carbon-containing compound from ethanol to glucose. Then, the ability to survive (or not) on a particular carbon compound (all other environmental variables being held equal) is the “phenotype” of a bacterium using a given reaction-genotype.
In this framework, any one bacterial genotype is a single “mutation” away from genotypes that differ by either adding or subtracting one of 5,870 different possible metabolic reactions considered by Rodrigues and Wagner. That is, the adaptive landscape is a network of possible genotypes, in which each genotype is a node linked to almost six thousand other nodes! In Arrival, Wagner builds an extended metaphor of such a network as a hyperspatial “universal library,” with each book “shelved” next to not two other books, but thousands. (He apparently wrote this before Matthew McConaughey browsed a multidimensional bookshelf on cinema screens around the world, which makes this choice fortuitous as well as poetic.)
Even with modern computing power, it is effectively impossible to completely map such a network. Instead, in the 2009 study, Rodrigues and Wagner selected individual genotypes, and stepped away from them one mutation at a time to understand the network’s structure. For instance, the typical E. coli metabolism includes 726 of the reactions they consider, and only 210 of them are essential for survival using glucose as the sole carbon source; thus, 210/5,870, or only 3.6% of genotypes that differ from E. coli by one reaction are unable to survive on glucose. Evolving from the E. coli reaction set to any of the other 5,660 neighboring genotypes should (Rodrigues and Wagner reasoned) make no difference in viability on glucose.
Rodriguez and Wagner then navigated along the network, moving away from the core E. coli genotype via one randomly-chosen allowable mutation at a time. They found 1,000 different genotypes containing a similar number of reactions, each 10,000 mutations steps away from E. coli. Each of these had a “robustness” similar to that of the original E. coli genotype — able to tolerate the loss of about 60 to 70% of their reactions without losing the ability to survive on glucose. Any two of the 1,000 randomly-found genotypes that could survive on glucose shared, on average, about a third of their essential reactions. And out of 1,420 reactions that were essential to at least one of the 1,000 randomly-found genotypes, only 103 were essential in all of the 1,000 genotypes.
That is, most of the time, a given metabolic reaction might be vital to the survival of a particular genotype, but somewhere out in the network there exist other genotypes that get along just fine without it. This is presumably because they include different sets of reactions that together achieve similar results. Wagner’s conclusion from these findings is that the tremendous range of possibility offered by a complex metabolism — or any other set of interacting organismal traits — gives evolution room to explore, to find new combinations of traits, and to innovate.
In sharp, conversational prose, Arrival of the Fittest presents this and other, related results, though not before carefully building broader biological context. The opening chapters of the book cover the Central Dogma (DNA, transcribed to RNA, codes for the assembly of amino acids into proteins, the molecular components of all living things), the theory of evolution by natural selection, and the biochemical origins of life. I found Wagner’s walk through this familiar and basic territory brisk and pleasant, and it should make the book accessible to anyone with an interest in biology, even as its ideas are sure to start some rousing conversations among professional biologists.
For, while Wagner’s work is a fantastic application of modern understanding of biological systems and computing power, it has some very real limits. Alert readers will already have made my biggest objection: in the real world, natural selection doesn’t care only about whether an organism can survive in a given environment. Competed against each other, some of the genotypes simulated by Rodrigues and Wagner would surely be more efficient than others, and these differences would almost certainly narrow the scope of evolutionary paths through Wagner’s universal library. With so many paths available, maybe that constraint is not much of a constraint — but how much is a question we need to answer, to know how well Wagner’s results reflect reality.
One such experiment was posted to the bioRxiv preprint server just a month ago. Celia Payen and coauthors at the University of Washington painstakingly tested the fitness effects of mutations to thousands of genes in the genome of yeast, in competition in a common environment. They then sequenced the genomes of yeast populations that had evolved in that environment for more than a hundred generations, and compared the mutations that actually succeeded over the long term to the ones that looked promising in their initial assay.
This should sound familiar — it’s remarkably close to Rodrigues and Wagner’s simulations, performed with actual populations of living, mutating, competing cells. But its results are somewhat at odds with what Wagner’s research has found. Payen et al. found hundreds of mutations that improved fitness in their initial screening, but never turned up in the evolved populations. This is exactly what we’d expect to see when the definition of “adaptation” is continuously variable, not just binary survival or failure. Further work like Payen’s, tracking mutations during experimental and observed evolutionary change, will provide a “top-down” perspective to complement what Wagner has built almost from first principles.
Wagner A. 2014. Arrival of the Fittest: Solving Evolution’s Greatest Puzzle. New York: Current, 291 pages. Publisher website.
Payen C, AB Sunshine, GT Ong, JL Pogachar, W Zhao, and MJ Dunham. 2015. Empirical determinants of adaptive mutations in yeast experimental evolution. bioRxiv doi: 10.1101/014068.
Rodrigues JFM and A Wagner. 2009. Evolutionary plasticity and innovations in complex metabolic reaction networks. PLoS Computational Biology, 5(12) e1000613. doi: 10.1371/journal.pcbi.1000613.
Samal A, J Jost, OC Martin and A Wagner. 2010. Genotype networks in metabolic reaction spaces, BMC Systems Biology, 4(1) 30. doi: 10.1186/1752-0509-4-30.
Wright S. 1932. The roles of mutation, inbreeding, crossbreeding, and selection in evolution. Proc. Sixth Int’l Congress on Genetics. pp. 355–366. PDF.