Pleistocene speciation and the mitochondrial DNA clock

Science

Volumn 282, Issue 5396, 1998, Pages 1955-

  Arbogast, Brian S ^a   Slowinski, Joseph B ^b  

^a WAKE FOREST UNIVERSITY   (United States)

^b CALIFORNIA ACADEMY OF SCIENCES   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 80054744521     PISSN: 00368075     EISSN: 10959203     Source Type: Journal    
DOI: 10.1126/science.282.5396.1955a     Document Type: Note

References (20)

1. J. Klicka and R. M. Zink, Science 277, 1666 (1997).
2. J. B. Slowinski and B. S. Arbogast, Syst. Biol., in press.
3. * And Other Methods), Version 4.0 (Sinauer, Sunderland, MA, 1998)].
4. This program estimates pairwise sequence divergence between taxa and infers phylogenetic trees under parsimony, maximum-likelihood, and distance criteria. Use of LRTs to determine bestfit models of nucleotide substitution and to test for a molecular clock followed the procedures outlined by J. Huelsenbeck and B. Rannala [Science 276, 227 (1997)].
5. For each data set, we evaluated the fit of ten models, five models each evaluated with and without among-site rate variation incorporated via a discrete gamma distribution with four rate categories as described by Z. Yang [Mol. Biol. Evol. 10, 1396 (1993)].
6. These models were "JC" [T. H. Jukes and C. R. Cantor, in Mammalian Protein Metabolism, H. N. Munro, Ed. (Academic Press, New York, 1969), pp. 21-132]
7. "F81" [J. Felsenstein, Evolution 35, 1229 (1981)]
8. "K2" [M. Kimura, J. Mol. Evol. 16, 111 (1980)]
9. "HKY85" [M. Hasegawa, H. Kishino and T. Yano, J. Mol. Evol. 22, 32 (1985)]
10. and "GTR" [C. Lanave, G. Preparata, G. Serio, J. Mol. Evol. 20, 86 (1984);
11. Z. Yang, J. Mol. Evol. 39, 105 (1994)]. Because it is unrealistic to assume that any nucleotide site is wholly invariant
12. [Z. Yang, Mol. Biol. Evol. 10, 1396 (1993)], all sites were assumed to be potentially variable. The majority of these models require direct sequence data, which prevented us from re-analyzing RFLP data presented in table 1 of (1). In order to reduce sampling error and to avoid among-gene rate heterogeneity, we used only those 21 species of North American passerines represented by ≥1000 bp of cytochrome b sequence data to determine a best fit model and to test for a molecular clock (14). The gamma-HKY85 model (-Ln likelihood = 6779.1872; α = 0.20826) provided the best fit to these data with the fewest parameters (11).
13. W. M. Brown, M. George Jr., and A. C. Wilson, Proc. Natl. Acad. Sci. USA 76, 1967 (1979);
14. W. M. Brown et al. J. Mol. Evol. 18, 225 (1982)]. We use the same calibration point for the chimpanzee-human divergence as these authors (about 5 My B. P.).
15. We used the cytochrome b data presented by E. Randi [Mol. Phylogenet. Evol. 6, 214 (1996)]
16. and the estimated date of divergence for chicken-Alectoris of about 17 My B. P. presented in K. M. Helm-Bychowski and A. C. Wilson [Proc. Nat. Acad. Sci. USA, 83, 688 (1986)].
17. D. L. Swofford et al. [in Molecular Systematics, D. M. Hillis, C. Moritz, B. K. Mable, Eds. (Sinauer, Sunderland, MA, 1996), pp. 407-514] review the assumptions and performance of various models of molecular evolution. Comparisons of models that assume there is no variation in the rate of substitution among nucleotide sites versus those that assume that substitution rates follow a gamma distribution suggest that the former may often underestimate the real number of substitutions that differentiate two DNA sequences. The gamma-HKY85 model provided the best fit to the North American passerine, great ape, and chicken-partridge data sets that we examined in this study (in each case, the more parameter-rich gamma-GTR model provided a slightly better fit to the data, but the improvement was not significant based on LRTs). The low value of a estimated under the best fit gamma-HKY85 for each data set (6, 9) indicates that there is a highly skewed distribution of rates of substitution in the cytochrome b sequences we examined. In other words, a few nucleotide sites within each rate category are evolving quite rapidly, but most sites are evolving relatively slowly. This means that a few nucleotide sites are likely to receive superimposed substitutions very soon after divergence. Because the equal-rates models (which assume α = infinity) used to estimate the various 2% per My mtDNA clocks cannot address this type of among-site rate variation, they will often underestimate the true number of substitutions that have occurred since two sequences diverged, which in turn leads to an underestimate of the true rate of evolution.
18. S. V. Edwards, in Avian Molecular Systematics and Evolution, D. P. Mindell, Ed. (Academic Press, New York, 1997), pp. 251-278.
19. C. B. Cox and P. D. Moore, Biogeography: An Ecological and Evolutionary Approach (Blackwell, Oxford, ed. 5, 1993).
20. D. M. Hillis et al., in Molecular Systematics, D. M. Hillis, C. Moritz, B. K. Mable, Eds. (Sinauer, Sunderland, MA, 1996), pp. 515-543. When using a molecular clock to estimate dates of divergence there are two important sources of error that must be addressed: (i) expected error in the measurement of molecular divergence, and (ii) error associated with predicting dates of divergence from a regressionbased rate calibration (that is, the residual of regression). The former, although often large, is trivial compared to the residual error of the regression. However, Klicka and Zink only reported mean estimated dates of divergence and did not address either of these sources of error. Since a regression of avian cytochrome b divergence on time is not available (the calibrations cited in note 10 of the report cannot be used to perform such a regression because they are based on various gene regions of the mtDNA and are not comparable), we used the confidence limits depicted by Hillis et al. for the original 2% mtDNA clock to emphasize that even if the other problems we address are ignored, the mean dates of divergence presented by Klicka and Zink would be highly imprecise.