The Mode and Tempo of Genome Size Evolution in Eukaryotes. Using phylogenetic contrasts derived from published genome data and 18 S

Transcription

1 The Mode and Tempo of Genome Size Evolution in Eukaryotes Matthew J. Oliver 1, Dmitri Petrov 2, David Ackerly 3, Paul G. Falkowski 1,4, Oscar M. Schofield 1 1 Institute of Marine and Coastal Sciences, Rutgers University, New Brunswick, NJ 08901, USA. 2 Department of Biology, Stanford University, Stanford, CA 93405, USA. 3 Department of Integrative Biology, University of California Berkeley, Berkeley, CA 94720, USA. 4 Department of Geological Sciences, Rutgers University, Piscataway, NJ 08854, USA. Using phylogenetic contrasts derived from published genome data and 18 S rdna divergence calculations, we estimate the rate of genome size evolution in eukaryotes from 168 species spanning major taxonomic groups. Our analysis indicates that the rate of genome size evolution is proportional to genome size, with the fastest rates of evolution occurring in the largest genomes. This trend is evident across all major clades analyzed, indicating that proportional change is the dominant mode of genome size evolution in eukaryotes. Our analysis suggests group specific selection pressures on genome size should be evaluated within the context of proportional evolution. The rate of genome size evolution is the balance between the rates of DNA insertion and deletion. In eukaryotes, these include unequal chromosome crossover (1), DNA replication errors (2-4), polyploidization (5), and the proliferation and recombination of transposable elements (6, 7). These mechanisms of DNA mutation potentially have variable responses to selection pressures and, depending on the organism in which they occur, will have variable rates of fixation, reflecting a mosaic of genome size evolution (8). However, many modes of insertion and deletion operate proportionally to the initial genome size. For example, the increase in DNA resulting from polyploidy is 1

2 proportional to the initial genome size, as is the probability of total insertions and deletions due to random replication errors. In addition, the probability of transposition is a function of the initial transposon copy number, as well as the number of potential target insertion sites (7, 9). Therefore, we might expect the rate of genome size evolution to also reflect these underlying proportional mechanisms that alter genome size. In this study, we examined the rate of genome size evolution in 20 traditionally recognized eukaryotic taxonomic groups comprising 168 species (Table 1), and use the concept of Brownian evolution and phenotypic contrasts to test the hypothesis of proportional genome size evolution in eukaryotes. The absolute magnitude of evolutionary change for a phenotypic trait under simple Brownian evolution behaves as if they were drawn randomly from a ½ normal distribution at each time step (10). In other words, the spectrum of possible evolutionary rates is fixed, and is not correlated with the preceding phenotype. However, a trait under proportional evolution violates the Brownian model because the spectrum of possible evolutionary rates scales with, and is dependent on the preceding phenotype. Therefore, if a phenotypic trait such as genome size were evolving primarily in a proportional manner, we would expect two clear patterns of genome size evolution to emerge in eukaryotes. First, the absolute rate of genome size evolution should be positively related to genome size, while the spectrum of evolutionary rates should clearly deviate from the ½ normal distribution predicted under Brownian evolution. Second, if genome size data were proportionally transformed a-priori (Log 10 ), thus removing the dependency of the evolutionary rate on the preceding phenotype, the absolute rate of genome size evolution should show no correlation to genome size. Furthermore, the proportional transformation 2

3 should result in a ½ normal distribution of evolutionary rates, thus approximating the simple Brownian model. We estimated the rate of genome size evolution in eukaryotes using the phylogenetic contrast method. This method uses a maximum likelihood estimation of a phenotypic trait at each node in a tree based on the trait at its tips (11). Contrasts are the quantitative difference between traits of the subtending branches for each node, scaled to their evolutionary distance. The absolute values of these contrasts represent the minimum rate of genome size evolution that has occurred since divergence from a common ancestor (12). In short, the rate of genome size evolution can be inferred by mapping genome size onto a phylogenetic tree based on 18 S rdna (13). We examined the relationship between genome size and the absolute value of contrasts in two ways. First, the maximum likelihood estimation of genome size at each node was compared to the contrast calculated at each node for the whole tree (Fig. 1A). Second, the tree was divided into 20 traditionally recognized taxonomic sub-trees, from which the median transformed genome size and median contrast for each sub-tree was taken as the representative for the group (Fig. 1B). These analyses show a significant positive relationship between the rate of genome size evolution and genome size (14), while analyses of the distribution of the absolute value of the contrasts show the expected significant deviation from a ½ normal distribution (Supplementary Fig. 2-3). An alternative and more direct test of proportional genome size evolution is to a-priori transform the genome size data, thus removing any proportional dependency between the rate of genome size evolution and genome size. Comparisons of Log 10 transformed genome size and their calculated contrasts reveal no significant correlation (14). In addition, analysis of the distribution of the contrasts 3

4 calculated from Log 10 transformed genome size approximate a ½ normal distribution, thus fit the Brownian model of evolution quite well (Supplementary Fig. 4-7). These results strongly indicate that the dominant mode of genome size evolution is proportional, with the tempo increasing with genome size. Hence, it would appear that in eukaryotes, the larger the genome, the faster its size is evolving (15). The proportional scaling of genome size evolution also predicts that it is less likely for small genomes to become large, and more likely for large genomes to become small. Therefore, we might expect far more small genomes than large genomes in eukaryotes. While rigorous testing of this hypothesis would require broad and even sampling of eukaryotic genome size, the distribution of genome sizes used in this analysis (Supplementary Fig. 8), as well as the distribution of genome sizes in Angiosperms (16) and Metazoans (17), support this hypothesis. Our analysis shows proportional evolution to be the dominant mode of genome size change in eukaryotes. It does not appear that selective pressures on genome size in any of the groups examined here are strong enough to cause large deviations in this overall trend. Therefore, we conclude that genome size evolution in eukaryotes is not constrained by strong selection to maintain an optimal size, as has been previously suggested (18-20). However, this does not imply that selection does not influence genome size evolution. Unexplained variation in the overall proportional trend may reflect taxon specific selection pressures. Therefore, in order to identify eukaryotic genomes that may be under unique selective pressures, our analysis suggests that it is necessary to first take into account the underlying proportionality of genome size evolution. 4

5 Clearly, larger sample sizes are necessary to determine if the proportionally corrected rate of genome size evolution between taxonomic groups are significantly different. However, despite a small sample size in each of the taxonomic groups, there are some interesting trends in these rates that are verified by pair-wise Mann-Whitney comparisons of contrasts (Fig. 2B, Supplementary Figure 8). For example, bird genomes have been hypothesized to evolve at a slower rate compared to other eukaryotes (21). However, our analysis suggests their rate of genome size evolution is not especially slow, but near the expected rate if the underlying proportionality of genome size evolution is considered. Furthermore, our analysis suggests that Magnoliophyta and Bacillariophyta genomes evolve at significantly higher specific rates than some other eukaryotic groups, perhaps indicating that selection is important for genome size evolution in these groups. While the specific selective forces driving the rapid tempo of genome evolution in Magnoliophyta and Bacillariophyta have yet to be fully explored, these high specific rates could reflect the importance of rapid mechanisms of genome size change such as polyploidy or transposable elements in these groups (5, 22-24). Our results suggest the tempo of genome size evolution is positively correlated to genome size across broad eukaryotic diversity. This relationship is consistent with a proportional model of genome size change as the dominant mode of genome evolution. Of the taxa examined here, none appeared to violate proportional genome size evolution; therefore, we conclude that taxa specific selection pressures operate within the umbrella proportionality. As a result, the identification of genomes under specific selection pressures requires the overarching pattern of proportional change be accounted for before selective forces driving genome size evolution can be properly assessed. 5

6 References 1. G. P. Smith, Science 191, 528 (1976). 2. A. M. Albertini, M. Hofer, M. P. Calos, J. H. Miller, Cell 29, 319 (1982). 3. K. Bebenek, T. A. Kunkel, PNAS 87, 4946 (1990). 4. T. A. Kunkel, Biochemistry 29, 8003 (1990). 5. D. E. Soltis, P. S. Soltis, TREE 14, 348 (1999). 6. K. M. Devos, J. K. M. Brown, J. L. Bennetzen, Genome Res. 12, 1075 (2002). 7. H. H. Kazazian, Jr., Science 303, 1626 (2004). 8. D. A. Petrov, TRENDS in Genetics 17, 23 (2001). 9. Y. Zhu, J. Dai, P. G. Fuerst, D. F. Voytas, PNAS 100, 5891 (2003). 10. F. L. Bookstein, Paleobiology 13, 446 (1987). 11. J. Felsenstein, The American Naturalist 125, 1 (1985). 12. T. Garland Jr., The American Naturalist 140, 509 (1992). 13. The phylogeny is based on 18 S rdna sequences that simultaneously allowed for broad coverage across the eukaryotic tree, as well as incorporated variable mutation rates in these sequences associated with various reproductive strategies and life histories [J. F. Gillooly, A. P. Allen, G. B. West and J. H. Brown, PNAS, 102, 140 (2005)]. Therefore rates of evolution are in terms of 18 S rdna divergence. These sequences were first automatically aligned using clustalx [J. D. Thompson, T. J. Gibson, F. Plewniak, F. Jeanmougin and D. G. Higgins, Nucleic Acids Res., 25, 4876 (1997)] and then hand edited. A Maximum Likelihood tree was computed using PHYML (GTR model, 1000 bootstraps). See [S. Guindon and O. Gascuel, Syst. Biol. 52, 696 (2003)]. Genome size (1C values) estimates for these sequences come from six sources: [B. J. Shuter, J. E. Thomas, W. D. Taylor, A. M. Zimmerman, Am. Nat. 122, 26 (1983); M. J. Veldhuis, T. L. Cucci, M. E. Sieracki, J. Phyc. 33, 527 (1997)], DOE Joint Genome Institute Royal Botanic Gardens, Kew, [M. D. Bennett, A. V. Cox, I. J. Leitch, The Animal Genome Size Database, Gregory, T.R. (2005) [Supplementary Table 1] With the exception of Veldhuis et al 1997, these sources tabulate genome sizes from other research efforts, and have those references within. In order to balance the phylogenetic tree, not all estimates of genome size from plants or animals were used. Instead, genome size estimates taken from the Kew database and the Animal Genome Size Database were chosen at random. Standardized independent contrasts were calculated for the 20 taxonomic groups in Table 1 using the Analyses of Phylogenetics and Evolution Package in the statistical program R [E. Paradis, K. Strimmer, J. Claude, G. Jobb, R. Opgen-Rhein, J. Dutheil, Y. Noel and B. Bolker (2004); R Development Core Team (2004) For Figure 1A,B the data are shown on a Log 10 transformed axis, but the statistics were done on the linear data. For Figure 1A, a standard OLS regression of the two variables indicated a significant positive correlation (R 2 = 0.67, P << 0.001). However, the maximum likelihood estimations of genome size at each node are not independent of each other, because the estimation of the genome size at any 6

7 node depends on the nodes surrounding it, therefore making a standard P value unreliable. Therefore, to determine if the positive correlation was significant, we used the PDSIMUL module of the PDAP program to simulate proportional evolution of genome size [Garland, T., Jr., A. W. Dickerman, C. M. Janis, and J. A. Jones. Sys. Biol. 42, 265, (1993)]. Parameterization of the model was based on the distribution of the genome sizes and the tree topology based on the 18 S rdna divergence used in this analysis. Correlations computed from 1000 Monte Carlo simulations of proportional evolution of genome sizes were used to estimate the significance of the of the OLS correlation coefficient computed in Figure 1A. The correlation fell within the 95% confidence interval of the expected correlation between the nodal estimation of genome size and the absolute value of the standardized contrast (P = 0.226) indicating the trend in Fig 1A was not significantly different than what would be expected under proportional evolution of genome size. Considerations of non-independence of regression variables were also taken into consideration for Figure 1B because of the hierarchical nature of the sub groups considered. For example, Vertebrata are not independent of Metazoa, because Metazoa subsumes Vertebrata. Therefore, regression analysis was done only on the medians of the mutually exclusive sub groups (R 2 = 0.84, P << 0.001). The same statistical precautions were taken for Figure 2A,B which were based on a-priori Log 10 transforming genome size. For Figure 2A, a standard OLS regression showed no significant relationship (R 2 = 0.021, P = 0.057). Monte Carlo simulation of proportional evolution of genome size indicated that the OLS correlation fell within the 95% confidence interval of the expected correlation between the nodal estimation of Log10 genome size and the absolute value of the standardized contrast (P = 0.137) indicating the trend in Fig 2A was not significantly different than what would be expected under proportional evolution of genome size. For Figure 2B, the median values of the mutually exclusive subgroups showed no significant correlation (R 2 = 0.006, P = 0.787). While figures 1B and 2B affirm the overall proportional relationship between genome size and the rate of genome size evolution, we emphasize that correlation of medians should be interpreted with caution and therefore should be treated as visual heuristic companions to Figures 1A and 2A. 15. Standardization of genome size contrasts in this study is based on 18 S rdna branch lengths. Thus, strictly speaking our results indicate the rate of Log 10 transformed genome size evolution is constant, relative to the rate of 18S rdna evolution. There are two possible interpretations of this pattern. First, genome size evolution may be directly coupled to 18 S rdna evolution through shared underlying mechanisms such as mutation rates and generation time. Second, GS evolution may be constant in real time, while the variation in 18 S clock rates is decoupled with the branch lengths across our tree, such that it introduces background noise into the data, but does not obscure the underlying pattern. Distinguishing these two hypotheses requires a time-calibrated eukaryote phylogeny, which is beyond the scope of this study. 16. C. A. Knight, N. A. Molinari, D. A. Petrov, Ann Bot 95, 177 (2005). 17. T. R. Gregory, Ann Bot 95, 133 (2005). 18. T. Cavalier-Smith, Journal of Cell Science 34, 247 (1978). 7

8 19. T. R. Gregory, P. D. N. Hebert, Genome Res. 9, 317 (1999). 20. T. Cavalier-Smith, Ann Bot 95, 147 (2005). 21. T. R. Gregory, Evolution 56, 121 (2002). 22. C. M. Vicient, M. J. Jaaskelainen, R. Kalendar, A. H. Schulman, Plant Physiol. 125, 1283 (2001). 23. V. A. Chepurnov, D. G. Mann, W. Vyverman, K. Sabbe, D. Danielidis, Journal Of Phycology 38, 1004 (2002). 24. E. V. Armbrust et al., Science 306, 79 (2004). 25. Acknowledgements: Funding for this research was provided by the NSF Biocomplexity OCE We are indebted to Dr. Theodore Garland Jr. of the University of California, Riverside for helpful discussions and access to the PDAP software package. We also thank Dr. Mark Moline, Dr. Charley Knight, and Dr. Kay Bidle helpful comments and discussions. Table 1: Number of Species in each Group Analyzed Taxonomic Group N Streptophyta (Green Plants) 37 Bryophyta (Mosses) 9 Moniliformopses (Horse Tails) 6 Magnioliophyta (Angiosperms) 12 Gymnosperms 10 Coniferopsida 7 Chlorophyta (Green Algae) 23 Dinophyceae 12 Stramenopiles (Heterokonts) 23 Bacillariophyta (Diatoms) 12 Pelagophyceae 6 Haptophyceae 11 Metazoa 52 Vertebrata 33 Mammalia 9 Aves (Birds) 7 Teleostei (Bony Fish) 7 Arthropoda 14 Crustacea 8 Insecta 6 8

9 Figure 1: A) A tree-wise analysis of the nodal estimated genome size and the calculated contrast at each node indicate that as genome size increases, the rate of evolution of genome size increases (shown on Log 10 axes for plotting purposes). B) Distribution of the median absolute contrast and the median genome size of the 20 traditionally recognized taxonomic groups analyzed in this study. Bars represent 95% bootstrapped confidence intervals. Again, a clear positive relationship between genome size and the rate of genome size evolution is evident (shown on Log 10 axes for plotting purposes). Figure 2: A-priori Log 10 transformation of genome size removes the proportional effect of genome size on the rate of genome size evolution so that neither, A) a tree-wise analysis of the nodal estimated genome size and the calculated contrast at each node, nor B), the distribution of the median absolute contrast and the median genome size of 20 traditionally recognized taxonomic groups show a significant correlation. Bars represent 95% bootstrapped confidence intervals. 9

10 Supporting Materials (On-Line) Supplementary Figure 1: Maximum Likelihood tree based on 18s sequences built using PHYML. Taxonomic groups highlighted in bold were analyzed for genome size evolution. Accession numbers of the 18 S rdna sequences used in this analysis are given. 10

11 Supplementary Figure 2: Distribution of the absolute value of the standardized contrasts showing approximately a strong deviation from the ½ normal distribution expected from a phenotypic trait under Brownian evolution. A strong deviation would be expected for a trait under proportional evolution. Supplementary Figure 3: Quantile distribution of the absolute value of the standardized contrasts. These contrasts do not show a near linear relationship to the positive quantile standard deviates indicating a strong deviation from a ½ normal distribution, which is expected for a trait under proportional evolution. 11

12 Supplementary Figure 4: Distribution of the absolute value of the standardized contrasts calculated from Log 10 transformed genome size data showing approximately a ½ normal distribution expected from a phenotypic trait under Brownian evolution. Supplementary Figure 5: Quantile distribution of the absolute value of the standardized contrasts calculated from Log 10 transformed genome size data. These contrasts show a near linear relationship to the positive quantile standard deviates indicating the expected ½ normal distribution of the contrasts for a phenotypic trait under Brownian evolution. 12

13 Supplementary Figure 6: There is not a strong correlation between the absolute value of the standardized contrasts and their standard deviations (R 2 = , P = 0.056) that would indicate a violation of the Brownian model of evolution. Supplementary Figure 7: Distribution of genome sizes used in this analysis. Proportional evolution of genome size predicts that there should be far more small genomes than large genomes. 13

14 Supplementary Figure 8: Pair-wise Mann-Whitney tests of contrasts calculated from Log 10 transformed genome size. Black squares indicate significant differences (P < 0.05) at the in the specific rate of genome size evolution. White squares are non-comparable groups because members of one are in the other. For example, Metazoa subsumes Vertebrata, so comparisons between the same organisms are invalid. 14

15 Supplementary Table 1: Genome sizes used in this study. When necessary, the number of base pairs was estimated from the mass of DNA using: Base Pairs = DNA Mass (pg)*0.978 X 10 9 [J. Dolezel, J. Bartos, H. Voglmayr, J. Greilhubner, Cytometry 51A, 127 (2003)]. Accession Number Genome Size (Base Pairs) X E+08 Y E+08 U E+08 AF E+08 Y E+08 X E+08 AF E+08 X E+08 X E+08 AB E+10 U E+10 U E+09 D E+09 L E+09 L E+09 X E+08 U E+08 AF E+09 AF E+08 AF E+09 AF E+10 U E+09 AF E+08 D E+09 AF E+10 AF E+09 AY E+09 AY E+09 D E+10 D E+10 D E+09 AF E+09 AF E+10 AF E+10 D E+10 X E+10 U E+10 AF E+08 AY E+08 AJ E+08 AB E+08 X E+07 AJ E+08 Z E+08 AF E+09 AF E+09 Z E+09 Z E+08 AY E+08 U E+07 AB E+09 AY E+08 X E+08 AJ E+08 AF E+08 AB E+08 U E+08 AF E+08 AF E+07 AF E+07 U E+09 AB E+09 AJ E+09 AJ E+08 AB E+10 AF E+10 AF E+10 AJ E+10 AF E+10 AF E+10 AJ AJ AF AJ AJ E E E E E+11 U E+08 AY AY AY E E E+08 X E+10 AF AY AY AJ AY AY AY AY AF E E E E E E E E E+07 U E+07 U E+08 U E+08 U E+09 U E+07 U E+08 U E+08 U E+08 U E+08 AF E+08 U E+08 M E+08 AJ AJ AF AJ E E E E+08 X E+08 AJ AJ AJ AF E E E E+08 U E+08 M E+08 U E+08 AY AY AY E E E+07 X E+09 M E+09 X E+09 AJ E+09 M E+09 AJ AJ AJ AJ AF AJ AF AF AF AF AF AJ AF AY AY AY E E E E E E E E E E E E E E E E+09 M E+09 AF E+09 AY E+09 AY E+09 L E+09 AF E+09 M E+09 AF E+08 AF E+08 AF E+08 AF E+08 AF E+09 M E+09 M E+09 AB E+08 L E+08 AY E+08 AF E+08 AF E+09 AF E+09 Y E+09 AF E+09 AF E+09 AF E+09 L E+09 AF E+09 AF E+09 AY E+08 AJ E+08 AF E+08 AY E+08 15