The review by Waxman & Gavrilets (henceforth W&G) gives a highly readable, nontechnical introduction to adaptive dynamics that is well suited to those who encounter the field for the first time. Unfortunately, however, the review also contains erroneous statements and recommendations. Since we do not have space here for corrections, we refer to Kisdi & Gyllenberg (2004) and to the commentaries by Geritz & Gyllenberg and Dieckmann & Doebeli (this issue). We also urge the reader to consult the original and fairly nontechnical paper of Geritz et al. (1998), which W&G follow closely in the first part of their review.

Adaptive dynamics (sensuMetz et al., 1996 and Geritz et al., 1998) is a framework to investigate evolution in a large variety of ecological models. Existing applications include structured populations, spatially heterogeneous and temporally fluctuating environments, metapopulations, altruism, sexual selection, mating systems, sex determination, microbial metabolism and prebiotic replicators, as well as resource competition, interference competition, predation, host–parasite systems, cannibalism and mutualism (references can be found at We are aware of over 50 applications to date, and this number is increasing fast. The review mentions few of these, which is a pity, because these diverse applications are the ‘fruits’ of adaptive dynamics.

The simplicity of the review hides many interesting evolutionary phenomena studied by adaptive dynamics, such as extensive adaptive radiations, evolution to extinction and ‘mass extinctions’, evolutionary cycles (also cycles involving changes in the level of polymorphism), multiple evolutionary attractors, evolutionary hysteresis effects, etc. The picture that emerges from the review is scanty compared with the rich behaviour of many models of adaptive dynamics.

Adaptive dynamics is an analytical rather than a simulation method. By ‘analytical’, we also refer to numerical analysis. All theorems and procedures of adaptive dynamics apply also when the ecological model at hand cannot be solved explicitly, as it is often the case with complex ecological models. Adaptive dynamics provides far better understanding than merely watching computer simulations; the latter are mainly aids for visualization. Thus while one can only agree with W&G that simulation (and other) techniques must be properly documented, the significance attached to simulation procedures by W&G is unwarranted.

W&G review adaptive dynamics from the stance of population genetics. In fact, the main roots and connections lay elsewhere: adaptive dynamics is primarily a phenotypic (clonal) approach, like evolutionary game theory or life history optimization. Matrix games and optimization models can be included as special cases in adaptive dynamics (Meszéna et al., 2001). Important forerunners were studies that found evolution to fitness minima in specific ecological models and suggested that polymorphism will evolve in such cases (see for references); some results were reached independently by Eshel et al. (1997). We do not consider multilocus simulation models of sympatric speciation, also reviewed by W&G, as part of adaptive dynamics, for they are not bound by the same basic assumptions.

Prediction of diversity.  Since adaptive dynamics is a method and not a model, testable predictions should come, and do come, from the specific applications. Obviously, one needs to specify the ecological model in some detail before specific predictions can be derived. The very few general assumptions of adaptive dynamics induce only a few universal predictions: Evolutionary branching occurs only at specific trait values (called branching points); one branch splits generically into two (not more) branches at a time; phenotypic differentiation at the onset of branching is slow compared with the speed of directional evolution; and changes in the environment generally prevent branching if their speed is comparable with the speed of the initial differentiation between the new branches (Metz et al., 1996). One tantalizing conjecture is that the fossil record might show a series of evolutionarily stable states, which change due to changes in the external environment on a paleontological time-scale. If so, the fossil record should correspond to a bifurcation diagram of adaptive dynamics (Rand & Wilson, 1993; Metz et al., 1996).

Many specific applications predict that the population evolves to an evolutionary branching point, i.e. to a fitness minimum, where disruptive selection acts to split the population into two diverging lineages. This sample of models is no doubt biased, because adaptive dynamics is particularly suited to study divergence without allopatry. Nonetheless, a general message emerges from the diverse applications: evolution to disruptive selection is expected to be common. There existed examples for this before the ascent of adaptive dynamics (Christiansen & Loeschcke, 1980; Abrams et al., 1993 and others), but now there are many more, and there is a theory that analyses the common structure underlying the specific models. Evolution to disruptive selection is a major conceptual change compared with the classic view, where evolution avoids fitness minima, disruptive selection is inherently unstable, and evolution leads to a single optimal phenotype. This brings new awareness to adaptive diversification in general, including (but not limited to) the emergence of new species.

In its quest for understanding adaptive diversification, adaptive dynamics does not stop at evolutionary branching in monomorphic populations. Coevolution of distinctly different phenotypes (either the products of evolutionary branching or present from the onset) may lead to further branching and also to extinctions (a detailed case study is given by Geritz et al., 1999). Looking beyond the first branching event is one important aspect in which adaptive dynamics goes beyond previous analyses. It must be emphasized that evolutionary branching is neither necessary nor sufficient for an evolutionarily stable polymorphism to exist [see e.g. Geritz et al. (1999), for counterexamples].

Evolutionary branching vs. speciation.  Evolutionary branching amounts to diversification, but it cannot by any means be equated to speciation. One of the first applications of adaptive dynamics (Kisdi & Geritz, 1999) was concerned with the emergence of a genetic polymorphism by evolutionary branching of alleles within a randomly mating population; speciation was thus ruled out by assumption. Van Dooren (1999) studied the joint adaptive dynamics of allelic values and dominance, again under the assumption of random mating. Within-species diversification may also take the form of sexual dimorphism or mixed strategies (Bolnick & Doebeli, 2003; Van Dooren et al., 2004). Adaptive dynamic analyses of these alternatives find that diversification itself is promoted by the same ecological conditions, but the subsequent evolution of already diverse systems depends on the form of diversity (e.g. Geritz & Kisdi, 2000).

Speciation is of course an exciting possibility, and the evolution of reproductive isolation between the emerging branches may be seen as a process competing with the evolution of dominance or of sexual dimorphism. In some cases, reproductive isolation may already be in place. Sexual selection (e.g. Seehausen, 2000) or sexual conflict (Gavrilets & Waxman, 2002) may split a population into reproductively isolated groups without ecological differentiation. In plants, polyploidization easily gives rise to instant reproductive isolation. These reproductively isolated new species, however, can coexist on an evolutionary time scale only if they become ecologically differentiated. Differentiation happens precisely if the population is near an evolutionary branching point at the time of speciation (Galis & Metz, 1998). In other cases, reproductive isolation may be very difficult to evolve. While the population is trapped at a fitness minimum, evolution has a long time to experiment with possible ways out (e.g. chromosomal rearrangements that reduce recombination and thus help speciation, sex-dependent phenotypic expression that allows for the evolution of sexual dimorphism, etc.). Any ‘solution’ that facilitates branching is favoured by disruptive selection.

Recent empirical findings direct attention to ‘byproduct’ speciation, where mate choice depends on the same trait that is under disruptive natural selection (e.g. Feder, 1998; Macnair & Gardner, 1998; Jiggins et al., 2001; Podos, 2001; Cruz et al., 2004; McKinnon et al., 2004). Adaptation to multiple habitats, accompanied by reduced migration and habitat-specific mating, can be seen as a form of byproduct speciation (where reduced migration substitutes for assortative mating; e.g. Rice & Salt, 1990; Hawthorne & Via, 2001). Theory predicts that in this case, mating assortativeness should increase as long as the trait is under disruptive selection, such that trait divergence is accompanied by reproductive isolation. If so, evolutionary branching would predict speciation.

There are, however, caveats to note (Kirkpatrick & Nuismer, 2004). Firstly, sexual selection may disfavour rare types. As assortativeness increases, sexual selection becomes stronger and may turn overall selection into stabilizing; if this happens, assortativeness stops increasing (Matessi et al., 2001). Secondly, environmental variance (reduced heritability) of the trait ‘blurs’ the information upon which mate choice is based, and therefore constrains the maximum attainable level of assortativeness on the genetic level. These caveats will hopefully motivate further research. For example, evolution is not necessarily limited to one specific form of mate choice. Since sexual selection is stabilizing under some mating rules but not under others, one should also consider the evolution of different mating rules. Or, given that several genetically correlated traits give more reliable information about the additive genetic value in the presence of environmental variance, mate choice based on a suite of traits should evolve in a noisy world. Only if all such possibilities fail will evolutionary branching be blocked.

Reproductive isolation is most difficult to evolve if mate choice depends on a trait different from the one under natural selection (Felsenstein, 1981). It is thus not surprising if the competing process of ‘byproduct’ speciation is more successful. The case of independent ecological and mating traits has been thoroughly investigated in population genetics (not reviewed here), but new ideas may still turn up. For example, the role of stochastic correlations between the mating trait and the ecological trait (Dieckmann & Doebeli, 1999) should be investigated in more detail. In the classic model of Udovic (1980), Geritz & Kisdi (2000) unexpectedly found a range of mating assortativity (penetrance) where initial allele frequency in the mating locus determines whether reproductive isolation is possible, such that a rare dominant allele in the mating locus is more conducive to speciation than a rare recessive allele. W&G unfortunately missed these points.

Going into more details, W&G picked only a side remark from the speciation model of Geritz & Kisdi (2000), ‘the evolution of assortative mating does not require very strong selection’. This was true in the model – when assortative mating evolved at all! Figure 5 of Geritz & Kisdi (2000) shows that mating assortativity must exceed a threshold for speciation to occur. With the ecological parameters used in this figure, selection against heterozygotes could not be stronger than s = 0.358 (compared with Hardy–Weinberg equilibrium, less than 22% of the heterozygotes could be ‘missing’ after selection).

Adaptive dynamics vs. population genetics: need for alliance.  Nonallopatric speciation needs stable polymorphism under disruptive selection as well as assortative mating. Traditionally both were considered unlikely (Maynard Smith, 1966; Felsenstein, 1981), but population genetic models of speciation focused mainly on the latter. Adaptive dynamics addresses the ecological side of speciation, and largely solves the first problem. To study the entire process of speciation, however, adaptive dynamics must join forces with population genetics.

In the population genetics literature, most speciation models use a very simple ecological background (either multiple habitats or the symmetric Lotka–Volterra competition model) in order to focus on the genetic details. Quantitative results are tied to these few ecological scenarios, although disruptive selection can arise in many other ways. It would be useful to explore whether population genetic models could instead be based on generic local approximations of fitness near an evolutionary branching point. For example, Matessi et al. (2001) developed an almost-ecology-independent approximation assuming only that fitness derives from pairwise interactions in polymorphic populations. They use this approximation to show that high levels of assortative mating will evolve only if the strength of disruptive selection exceeds a certain threshold. Adaptive dynamic analyses should establish whether disruptive selection is stronger than this threshold in various ecological models. Geritz & Kisdi (2000) made some steps in this direction based on the population genetic model of Udovic (1980). Approximations with different genetic assumptions could be combined with diverse ecological models without repeating the population genetic analysis in each and every case. In addition, these approximations would be easier to compare with one another. Decoupling the complexities of ecology and of genetics could benefit both parties, if they can form an alliance.


  1. Top of page
  2. Acknowledgments
  3. References

Our research is financially supported by the Academy of Finland.


  1. Top of page
  2. Acknowledgments
  3. References