Evolutionary biology

Relativity for molecular clocks

Article metrics

An analysis of genetic data sets from primates and birds provides firm evidence that molecular evolution is faster on shorter than on longer timescales. The estimated times of various evolutionary events require a rethink.

The relative constancy of the rate at which DNA sequences evolve has been a treasured icon of molecular evolution for nearly 40 years. The occurrence of such a stochastic ‘molecular clock’ was initially quite unexpected, and was explained by Motoo Kimura1 by assuming that most changes to amino-acid and nucleotide sequences were neutral — “neither beneficial nor injurious”, in Charles Darwin's prescient phrase.

However, there have been several inklings2,3,4 that the rate of molecular evolution accelerates when measured over evolutionarily short timescales. As they report in Molecular Biology and Evolution, Ho and colleagues5 have now put the evidence together. Their analyses of primate and bird data sets reveal that there is indeed a decided acceleration of molecular evolution on short timescales. This is an effect that demands explanation; moreover, estimates for the timing of recent events in population biology will need to be reconsidered.

First, some background. Traditionally6, the total mutation rate μ is subdivided into the proportion of mutations that are advantageous (α+), neutral (α0) or deleterious. Most rate calculations focus on neutral and deleterious mutations. Kimura's argument1 for the long-term rate is that for a fixed population size of N individuals, with a single copy of their genome (haploid), each neutral variant has the same probability of 1/N of ending up in 100% of the population (that is, becoming fixed). Kimura's basic argument is that the number of neutral mutations per unit time is Nμα0, and because the probability of fixation of a single mutation is 1/N, the rate of fixation is μα0. If the rate of fixation is equated with the rate of long-term molecular evolution, they will both equal the rate of neutral mutation, μα0.

Here I'll consider four questions. Is the short-term acceleration of molecular evolution real? If so, is it explainable? If so, why wasn't it detected earlier? Finally, what are the consequences?

First, is the acceleration real? Ho et al.5 observed the effect in three data sets, two based on protein-coding genes (avian and primate) and the third on the control regions in the mitochondria of higher primates. In each case, at times less than about 1–2 million years there is an increasing acceleration, with the highest rates at the shortest times. The rate is highest between generations (that is, in pedigree studies), and decreases continuously for local and then for widespread populations. Finally it reaches the low plateau that we know for long-term evolution. Overall, there is the J-shaped rate curve shown as the solid line in Figure 1a — in Texas-talk, the lazy J.

Figure 1: The J-shaped rate curve and the multiscale problem.
figure1

a, The shape of the rate curve of molecular evolution as estimated by Ho et al.5. The higher rate at shorter times is interpreted here as having a neutral component (below the dashed line) combined with an effect from deleterious mutations. b, In the classic long-term problem in evolutionary biology, each taxon is represented by a single sequence (t1, t2 ...) and a tip-labelled binary tree is appropriate; the internal points are 'missing data'. c, Within populations there will be many sequences (s1, s2 ...) per population. At these shorter times the tree is usually multifurcating (not binary), and sequences occur at internal nodes (representing ancestral nodes still present in the population). Comparison of b and c shows that there is a multiscale problem, where different aspects of the same underlying process are observed as the scale varies; in this case, the scale is time.

The authors5 eliminate both sequencing and calibration errors, as well as saturation of mutations at fast-evolving sites, as explaining the change in rates — although calibration errors at the shorter times (within the past million years) contribute strongly to the variance. Overall, however, the implication is that the timing of many recent events in human evolution has been overestimated by past studies. Examples include the divergence between humans and Neanderthals — new estimate 354,000 years ago (range 222,000 to 705,000 years ago, against a current range of 317,000 to 853,000 years ago); and the last split within those Neanderthals that have been sequenced — 108,000 years ago (range 70,000 to 156,000 years ago compared with a current range of 151,000 to 352,000 years ago).

Second, can the acceleration be explained? One approach is as follows. We can subdivide the deleterious mutations into the proportions that are very slightly deleterious, α, slightly deleterious, α=, and deleterious (but not lethal), α. Deleterious mutations are not expected to become fixed in large populations, but nevertheless can persist in the population for long periods of time. The average time before loss (see ref. 7, for example) correlates with deleteriousness, so persistence time increases from α to α0. Thus, as observation times diminish, we should observe a greater proportion of slightly deleterious mutations that have yet to be lost, with the most deleterious (α) observed only in the short-term pedigree studies. This gives the apparent acceleration in mutation rate as the separation times between sequences decrease. A qualitative explanation such as this is straightforward, but the effect requires quantitative treatment and thorough testing.

Third, why wasn't the short-term acceleration picked up earlier? There were hints. Kimura1 (page 45ff.) comments that his calculations assumed sufficient time for loss or fixation of a neutral mutant, and that at shorter intervals there would be higher variability. Some of the increase in rate comes from neutral mutations (Fig. 1a), and some from slightly deleterious mutations. In early molecular work, the genetic diversity within populations — heterozygosity — was measured by the differences in the behaviour of variant proteins in electrophoresis, and long-term evolution by differences in amino-acid sequences between species (with only one sequence studied per species). Thus, direct comparability was not possible in those early studies — although in 1977 Ohta did point out8 that “In the future, it is likely that genetic variability will be detected at the level of amino acid or nucleotide sequences”.

For some reason, the continuum between population heterozygosity and long-term evolution has not been adequately studied. Although it is a continuum, the techniques required may change as the timescale decreases. For example, some concepts from long-term evolution (binary evolutionary trees with sequences studied only at the tips) have been extended into populations where trees are no longer binary, and ancestral sequences (at internal nodes) are still present in the population. There are hints that a formal multiscale study9 is necessary, because even though the same underlying process is occurring, different features of trees are observed as the timescale changes (Fig. 1b, c).

Finally, what are the consequences of Ho and colleagues' conclusion5? As they point out, the obvious ones are practical, in that many time estimates require recalculation - including the times of events in recent human evolution (such as origin of the Polynesian genetic marker that distinguishes Polynesians from all other human populations10), and the origin of RNA viruses such as HIV11. In some cases the constraints are from recent events, and it is the long-term events that require re-analysis.

Much more remains to be done. There is the challenge of formulating a single theory that operates smoothly over disparate timescales, from current heterozygosity to the long-term rate of evolution. In addition, a single mutation rate (μ) does not really exist. Even for nucleotides there are many 'mutation rates', at least one between each pair of nucleotides, and these can be estimated separately using three-dimensional matrices12. The J-shaped curve cannot rest until a single theory holds for it: we live in interesting times.

References

  1. 1

    Kimura, M. The Neutral Theory of Molecular Evolution (Cambridge Univ. Press, 1983).

  2. 2

    Garcia-Moreno, J. J. Avian Biol. 35, 465–468 (2004).

  3. 3

    Howell, N. et al. Am. J. Hum. Genet. 72, 659–670 (2003).

  4. 4

    Lambert, D. M. et al. Science 295, 2270–2273 (2002).

  5. 5

    Ho, S., Phillips, M. J., Cooper, A. & Drummond, A. J. Mol. Biol. Evol. 22, 1561–1568 (2005).

  6. 6

    Dickerson, R. E. J. Mol. Evol. 1, 26–45 (1971).

  7. 7

    Garcia-Dorado, A., Caballero, A. & Crow, J. F. Evolution 57, 2644–2646 (2003).

  8. 8

    Ohta, T. Theor. Popul. Biol. 10, 254–275 (1976).

  9. 9

    Penny, D. & Holmes, S. Trends Ecol. Evol. 16, 275–276 (2001).

  10. 10

    Oppenheimer, S. J. & Richards, M. Nature 410, 166–167 (2001).

  11. 11

    Rambaut, Q., Posada, D., Crandall, K. A. & Holmes, E. C. Nature Rev. Genet. 5, 52–61 (2005).

  12. 12

    Ota, R. & Penny, D. J. Mol. Evol. 57, S233–S240 (2003).

Download references

Author information

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Penny, D. Relativity for molecular clocks. Nature 436, 183–184 (2005) doi:10.1038/436183a

Download citation

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.