A computer model of tooth evolution designed to assess the impact of developmental dynamics on natural selection reveals that complexity reduces the likelihood of maximum fitness being attained. See Letter p.361
The evolutionary biologist Leigh Van Valen famously defined evolution as the control of development by ecology1. By this, he meant that evolution is governed by the fitness of the phenotype of an organism (its physical form); that fitness depends on the ecological context in which the phenotype exists; and that an individual's phenotype is not a static structure built directly from genes, but a dynamic structure that unfolds sequentially from fertilization to adulthood. The developmental sequence determines and constrains which phenotypes can emerge, such that evolutionary change depends on modifications to development. On page 361 of this issue, Salazar-Ciudad and Marín-Riera2 present a groundbreakingly realistic computational model of this process. They use their model to demonstrate that developmental complexity often prevents natural selection from reaching optimal fitness when fitness is directly linked to attaining a particular phenotype, but that these 'adaptive peaks' can be reached when fitness is instead linked to functional properties of the phenotypeFootnote 1.
The authors' model revolves around a simulation of the evolution of mammalian tooth development, using well-understood details of molecular developmental biology and functional ecology3,4,5. Mammalian teeth develop from two main tissue types, the mesenchyme and the epithelium. The growth of these tissues helps to shape the tooth germ (the aggregation of cells that eventually form the tooth) as it develops through bud, cap and bell stages (Fig. 1a). A suite of activator and inhibitor molecules controls the rates of cell proliferation, differentiation and death in these tissues6. Three-dimensional changes in the developing tissues modulate the distances over which these signals interact, causing local changes in their concentration that result in the formation or loss of signalling centres called enamel knots. The shape into which the dynamically controlled tooth germ folds determines the topography of the crown of the mature tooth and thus its functional properties.
A huge variety of teeth can be produced by this developmental system, ranging from a dog's high-cusped molars to the elaborately ridged molars of some rodents, such as the coypu (Fig. 1b). Tall cusps in mammals are usually associated with the puncturing, shearing and tearing functions of carnivores and insectivores, whereas low cusps, crests and basined surfaces are associated with the grinding and chewing functions of omnivores and herbivores5.
Salazar-Ciudad and Marín-Riera simulated this developmental system as two sheets of cells, the epithelium and mesenchyme, with parameters that represented activator and inhibitor molecules produced by these tissues. Changes in the values of these parameters simulated up- or downregulation of the molecular signals, which, in turn, altered the pattern of growth and folding in the simulated tissue layers, just like in real tooth germs. To simulate evolution of tooth structure using this model, the authors created artificial populations of developing teeth that they subjected to mutation and selection. Each population started with a randomly chosen phenotype. Mutations were applied by stochastically altering the molecular signalling parameters, which resulted in new phenotypes. Fitness was assigned to each individual on the basis of either its phenotypic or its functional similarity to an arbitrarily chosen target phenotype. The latter can be thought of as the phenotype that conveys the greatest fitness in the ecological context of the simulation — a peak on the adaptive landscape, as described by the geneticist Sewall Wright7. The fittest individuals were then selected as the parents of the next generation and the simulation was repeated.
This process was expected to cause the population's mean phenotype to evolve towards the model's adaptive peak. And so it did, although only to a point. The authors found that when the criterion for fitness was overall similarity to the optimal phenotype, fewer than 40% of the simulations reached their adaptive peak. They suggest that failure to reach the adaptive peak is due to the complexity of developmental interactions; the sequence of parameter changes that is needed to generate the optimum phenotype includes changes that temporarily reduce fitness, but selection prevents these changes from occurring, and thus prevents the phenotype from evolving to the adaptive peak. It has long been known that natural selection might leave populations stranded on suboptimal peaks8. Indeed, Wright himself introduced the shifting balance theory to explain how an evolving species might jump from a suboptimal peak onto a higher one7. Salazar-Ciudad and Marín-Riera have shown that not only are suboptimal dead ends an evolutionary possibility, but they are also exceedingly likely to occur in real, developmentally complex structures when fitness is determined by the exact form of the phenotype.
However, the authors also found that when fitness was determined by functional properties instead of the phenotype itself, the adaptive peak was usually reached. This is because many different phenotypes can have the same functional properties9 — a herbivorous mammal, for example, simply needs grinding and chewing surfaces on its teeth, regardless of how the surfaces are constructed. Thus, there are many more paths to a functional adaptive peak than to a phenotypic one, especially for a phenotype that has a complex developmental system, such as a tooth.
By developing a model of Van Valen's hypothesis of evolution as the control of development by ecology, Salazar-Ciudad and Marín-Riera have demonstrated that one fundamental assumption that underpinned his concept was wrong. Van Valen argued that the effect of genes on evolution is negligible because developmental interactions — through mutation, recombination and epigenetic alteration — are “sufficiently flexible” for any phenotype to be selected1. By contrast, Salazar-Ciudad and Marín-Riera's results demonstrate that developmental interactions in fact prevent selection from finding the optimal phenotype, even with sufficient mutation in the genetic parameters.
Salazar-Ciudad and Marín-Riera's results reveal a lot about how evolution may work in practice. The originality of their work lies in how the model realistically maps the interactions between genetic parameters, developmental processes, adult phenotypes and functional properties. These interactions introduce dynamic properties that allow for morphological transitions and evolutionary novelties that were not captured by earlier computational models of the evolution of complex phenotypes10. The authors' results highlight interesting questions. If phenotypic evolution is likely to get stuck on suboptimal adaptive peaks, what happens to the evolving populations? Do they simply persist in a suboptimal state? Do their ecological relationships change to fit their phenotype, thereby creating new adaptive peaks? Or do they become extinct as they are out-competed by populations that can reach a more optimal phenotype? Salazar-Ciudad and Marín-Riera's focus on teeth — which can be studied empirically in living populations, among distantly related clades and in the fossil record — offers considerable potential for testing their evolutionary predictions and closing the knowledge gaps between genetics, development and macroevolution.
*This article and the paper under discussion2 were published online on 1 May 2013.
Van Valen, L. J. Herpetol. 8, 109–121 (1974).
Salazar-Ciudad, I. & Marín-Riera, M. Nature 497, 361–364 (2013).
Jernvall, J., Keränen, S. V. E. & Thesleff, I. Proc. Natl Acad. Sci. USA 97, 14444–14448 (2000).
Tucker, A. & Sharpe, P. Nature Rev. Genet. 5, 499–508 (2004).
Ungar, P. S. Mammal Teeth: Origin, Evolution, and Diversity (Johns Hopkins Univ. Press, 2000).
Salazar-Ciudad, I. & Jernvall, J. Evol. Dev. 6, 6–16 (2004).
Wright, S. Proc. 6th Annu. Congr. Genet. 1, 356–366 (1932).
Whitlock, M. C. & Phillips, P. C. Trends Ecol. Evol. 15, 347–348 (2000).
Wainwright, M. et al. Am. Nat. 165, E140–E154 (2005).
Polly, P. D. Palaeontol. Electron. 2, 7.2.7A (2004).
About this article
Deletion/loss of bone morphogenetic protein 7 changes tooth morphology and function in Mus musculus: implications for dental evolution in mammals
Royal Society Open Science (2018)
Scientific Reports (2017)