Evidence for a universal driver of evolution across all timescales could mean that the venerable paradox of stasis is dead. But even with such evidence, some biologists would be reluctant to accept its passing.
Disagreement has long swirled around the relative importance of various forces that might drive evolution on timescales ranging from dozens to millions of generations. Writing in The American Naturalist, Estes and Arnold1 offer a provocative contribution to this debate: they propose that evolutionary changes on all timescales might be explained by a single, simple model of adaptation.
Much of the challenge can be distilled down to what has been called the 'paradox of stasis'2. For me, the most obvious manifestation of this paradox is that neo-darwinian theory, with its emphasis on the power of selection, predicts the potential for rapid adaptation, whereas most lineages of organisms instead show long-term stasis: that is, very little cumulative change over long periods of time3,4. Several hypotheses have been advanced in the hope of resolving this seeming discontinuity between short- and long-term evolution2,4,5, but none has been convincing enough to resonate across the various camps.
Estes and Arnold1 point out that the best way to discriminate between the hypotheses is to confront the predictions of alternative evolutionary models with the reality of data. This sort of comparison has recently been made possible by compilations of data on phenotypic changes (such as in mean body size) within animal lineages at a variety of different timescales4,6,7. The pattern emerging from these data is that phenotypic changes over dozens of generations can range from small to large, and that this range remains roughly the same even over millions of generations (Fig. 1). This pattern thus affirms the original paradox — that phenotypic change can be dramatic on short timescales, but rarely accumulates into substantial evolutionary trends.
Estes and Arnold1 evaluate the degree to which six evolutionary models fit the observed data. All of these models are based on 'adaptive landscapes' — an analytical framework that relates mean phenotypes (mean body size, for instance) to the expected mean fitness (that is, number of offspring) of a population (Box 1). Evolution on such landscapes tends towards 'hill climbing', where the mean phenotype of the population moves towards that which maximizes population fitness (a local fitness peak)8,9.
Three of the models tested by Estes and Arnold represent different flavours of randomness (Box 1). One model ('neutrality') evokes a flat adaptive landscape, on which the mean phenotype of a finite population will drift at random. The other two random models have a single fitness peak that moves randomly according to either 'brownian motion' or 'white-noise motion'. In these two models, mean phenotypes forever chase the randomly moving fitness peaks, like a new task for Sisyphus. Estes and Arnold argue that these three models fail to fit the data well, suggesting that randomness, at least in these forms, may not be a primary driver of phenotypic change. This conclusion will be reassuring, or perhaps just obvious, to the innumerable evolutionary biologists who believe that adaptation plays a central role in evolution.
The other three models involve a directional shift in the position of a fitness peak. In one ('moving optimum'), the adaptive landscape has a single peak that moves step-by-step in a particular direction, with the phenotypic mean of the population following along. This model predicts too little evolution on short timescales and too much on long timescales relative to the observed data. In another directional model ('peak shift'), the adaptive landscape has two peaks and, under some conditions, the phenotypic mean can shift rapidly from one peak to the other. This model predicts too little evolution on short timescales, and it only fits the data well on long timescales when populations are unrealistically small. In the final model ('displaced optimum'), the adaptive landscape has only one peak, and the position of this peak shifts abruptly, but just once. In the authors' estimation, this last model fits the data quite well.
The displaced-optimum model can be visualized by reference to a hypothetical population that is well adapted to its local environment, in which case the mean phenotype of the population (let's say a body mass of 10 g) will be centred near a local fitness peak (also 10 g). Then imagine that the environment changes abruptly and displaces the optimum to a new location of 12 g, leaving the population mean behind at 10 g. Now the largest individuals in the population will be favoured by natural selection, and the mean phenotype will increase across generations until it is positioned at the new fitness peak (that is, 12 g), where it will then stay in the absence of further environmental change. This all makes sense, but the surprising part is that the displaced-optimum model assumes that this happens only once for a given lineage — regardless of timescale. The key general point, however, is that the peak can be displaced only a restricted amount (that is, within defined bounds), even if it takes several steps to get there.
Conveniently, Estes and Arnold1 provide an Excel file in which every parameter in every model can be varied and the resulting outcomes compared with actual data. After playing with these models myself, I tend towards general agreement with the authors, adding the caveat that achieving the observed changes on short timescales requires a very large displacement of the optimum, coupled with a very sharp fitness peak. These properties mean that a population will have a substantial fitness decline immediately after the optimum moves — a possible recipe for extinction. Perhaps the largest changes on short timescales are the result of phenotypic plasticity (when a genotype expresses a different phenotype in a new environment), rather than of genetic change, the latter being the focus of the models. Other factors that may inflate short-term changes are sampling errors, conflation of geographical variation with lineage evolution, and publication biases (perhaps published studies on short timescales tend towards those that find the largest changes).
Have Estes and Arnold1 slain the paradox of stasis with a simple displaced-optimum model? In my opinion, the paradox might have been slain only in a broad sense, because the pattern in the data also might be replicated by other models that generate a range of short-term changes that do not accumulate into long-term trends. Such models might include various types of fluctuating selection, where fitness peaks on adaptive landscapes move back and forth owing to environmental variation10. Perhaps the paradox of stasis will have its final death at the point of a rapier, whereas Estes and Arnold have wielded a scimitar.
Whatever the model, it will have to generate rapid changes on short timescales, and yet still be constrained by boundaries on long timescales4. It is also possible that the paradox is a phantom, against which swords are of no use. Indeed, it may have been dead on arrival: way back in 1944, George Gaylord Simpson8 suggested that evolutionary stasis might be explained by 'adaptive zones', where fitness peaks move back and forth within constrained bounds.
For some, any report of the death of this paradox will probably evoke the same reaction as the death of Elvis, with a large number of fans reluctant to accept its passing. But in the end, evolutionary biologists will probably converge on more pertinent questions, such as 'What generates and maintains adaptive zones in the first place?', and 'How do some lineages ultimately bridge the gap between different adaptive zones?'. This convergence would probably both please and frustrate Simpson, were he still alive, given that he posed much the same questions more than 60 years ago.
Estes, S. & Arnold, S. J. Am. Nat. 169, 227–244 (2007).
Hansen, T. F. & Houle, D. in Phenotypic Integration (eds Pigliucci, M. & Preston, K.) 130–150 (Oxford Univ. Press, 2004).
Gould, S. J. & Eldredge, N. Paleobiology 3, 115–151 (1977).
Gingerich, P. D. Genetica 112–113, 127–144 (2001).
Charlesworth, B. et al. Evolution 36, 474–498 (1982).
Hendry, A. P. & Kinnison, M. T. Evolution 53, 1637–1653 (1999).
Kinnison, M. T. & Hendry, A. P. Genetica 112–113, 145–164 (2001).
Simpson, G. G. The Tempo and Mode of Evolution (Columbia Univ. Press, 1944).
Arnold, S. J. et al. Genetica 112–113, 9–32 (2001).
Grant, P. R. & Grant, B. R. Science 296, 707–711 (2002).
About this article
Philosophy of Science (2009)
Artificial Life (2008)