A Dissection of Volatility in Yeast
http://www.100md.com
分子生物学进展 2005年第10期
* Section of Evolutionary Biology, Department Biology II, Ludwig-Maximilians-University Munich, Planegg-Martinsried, Germany; and Centre for the Study of Evolution, University of Sussex, Brighton, United Kingdom
E-mail: stoletzki@zi.biologie.uni-muenchen.de; a.c.eyre-walker@sussex.ac.uk.
Abstract
It has been suggested that volatility, the proportion of mutations which change an amino acid, can be used to infer the level of natural selection acting upon a gene. This conjecture is supported by a correlation between volatility and the rate of nonsynonymous substitution (dN), or the ratio of nonsynonymous and synonymous substitution rates, in a variety of organisms. These organisms include yeast, in which the correlations are quite strong. Here we show that these correlations are a by-product of a correlation between synonymous codon bias toward translationally optimal codons and dN. Although this analysis suggests that volatility is not a good measure of the selection, we suggest that it might be possible to infer something about the level of natural selection, from a single genome sequence, using translational codon bias.
Key Words: volatility ? codon bias ? selection ? nonsynonymous substitution rate ? yeast
Introduction
Understanding the nature of natural selection on DNA sequences is one of the central goals of molecular evolution. Plotkin, Dushoff, and Fraser (2004) and Plotkin et al. (2004) have recently suggested that it is possible to infer the level of natural selection, both positive and negative, acting upon a gene from a single genome sequence. They suggest that this can be achieved by measuring "volatility"—volatility is the proportion of point mutations in a gene, which do not yield a stop codon, which change an amino acid. They base their method on the prediction that genes which have recently undergone amino acid substitutions should be populated by codons with high volatility (Plotkin et al. 2004). In support of their thesis they show that in both Mycobacterium and Saccharomyces species, there is a correlation between volatility and the rate of nonsynonymous substitution (dN) or the ratio of nonsynonynous and synonymous substitution rates (dN/dS). This correlation is quite strong in yeast, which suggests that volatility might be a useful measure of selection.
The idea that volatility can measure the level of selection, either positive or negative, on a gene has been criticized on a number of grounds (Dagan and Graur 2004; Friedman and Hughes 2004; Sharp 2004; Chen, Emerson, and Martin 2005; Hahn et al. 2005; Nielsen and Hubisz 2005; Zhang 2005). Much of the debate has centered around the reasons why volatility is not expected to correlate to dN and dN/dS. For example, it has been suggested that volatility is unlikely to measure selection because (1) it only depends on four or five amino acids (Dagan and Graur 2004; Sharp 2004; Chen, Emerson, and Martin 2005), (2) it has low variance (Dagan and Graur 2004), and (3) simple models of evolution fail to yield a correlation between dN/dS and volatility (Dagan and Graur 2004; Nielsen and Hubisz 2005; Zhang 2005). However, volatility is correlated to dN/dS (and dN); so much of this discussion, while interesting is slightly tangential. The crucial question is why there is a correlation.
Almost all of these critiques point out that volatility is a measure of codon usage bias. As such, the apparent correlation between volatility and dN/dS in yeast may, in fact, as Hahn et al. (2005) suggest, be due to a correlation between translational codon bias and dN/dS. Although Hahn et al. suggest that the correlation between volatility and dN/dS may be due to a correlation between translational codon bias and selective constraint they do not resolve whether this is the case. They show that a measure of translational codon bias, codon adaptation index (CAI), explains more of the variance in volatility than dN/dS in yeast, but they do not pursue the matter further. Plotkin, Dushoff, and Fraser (2005) investigate the partial correlation between dN/dS and volatility controlling for CAI and show it is significant, but they fail to give the magnitude of the effect.
It, therefore, remains very unclear if the principle correlation is between dN/dS and translational codon bias, with the correlation between dN/dS and volatility a by-product of this, or whether the principle correlation is between dN/dS and volatility. Also, it might be that both translational codon bias and volatility separately correlate to dN/dS.
To investigate the matter further, we take advantage of the fact that in yeast there is a strong correlation between dN/dS (or dN) and both translational codon bias (Pal, Papp, and Hurst 2001) and volatility (Plotkin, Dushoff, and Fraser 2004) and that in yeast some of the translational optimal codons have relatively high volatility while others have relatively low volatility (table 1). It is well established that codon bias and gene expression are correlated in yeast (see e.g., Coghlan and Wolfe 2000). So volatility per amino acid is expected to increase (Ile, Leu, and Ser) or decrease (Arg and Gly) with translational codon bias or expression level (table 1). For example, the most optimal codon in yeast for argenine is AGA, which has relatively high volatility. If the principle correlation is between dN/dS (or dN) and translational codon bias, then we expect AGA usage to be negatively correlated to dN/dS (or dN), but if the principle correlation is between dN/dS (or dN) and volatility, then we expect AGA usage to be positively correlated to dN/dS (or dN).
Table 1 Synonymous Codon Use of Volatility-Affecting Amino Acids
Our results are unequivocal; in yeast dN/dS (and dN) is negatively correlated to the use of translational optimal codons for all amino acids whose synonymous codons differ in their volatility, even in those whose optimal codons have high volatility. We further show that the correlation between dN/dS (or dN) and translational optimal codon use is universal across all amino acids, including those synonymous codons which do not differ in their volatility. The observed correlation between dN/dS (or dN) and volatility is a by-product of the correlation between dN (or dN/dS) and translational codon bias.
Materials and Methods
We downloaded the gene alignments from the four yeast species sequenced by Kellis et al. (2003). From these we excluded all genes which were not present in all the four yeast species (Saccharomyces cerevisiae, Saccharomyces paradoxus, Saccharomyces mikatae, and Saccharomyces bayanus), which did not have start and stop codons in all species, which had premature stop codons, and which had frameshifting indels. This left 1,077 genes. This is smaller than the data set analyzed by Plotkin, Dushoff, and Fraser (2005) but has less chance of containing pseudogenes.
Plotkin, Dushoff, and Fraser (2004) suggest using a statistic, volatility P value, to measure volatility. This is the probability of a gene having the observed volatility given the average synonymous codon use of the genes in the genome. The volatility P measure of Plotkin, Dushoff, and Fraser 2004 is unlikely to be a very good statistic because it will depend to some extent on gene length (Sharp 2004) and amino acid composition (Dagan and Graur 2004; Zhang 2005)—any statistic based on probability values depends on sample size, and the variance between synonymous codons for volatility differs between amino acids. To account for these shortcomings, we calculated an alternative measure, the average volatility:
where
and Xi is the number of times codon i is used for the amino acid aa, Vi is the volatility of that codon, and n is the number of amino acids whose synonymous codons differ in their volatility. When considering amino acids separately we used Vaa, the average volatility per amino acid. Note that the volatility is only affected by five amino acids whose synonymous codons differ in their volatility—Arg, Gly, Ile, Leu, and Ser (the codons of Ile only differ when the transition:transversion ratio is different from unity). We assume that the transition:transversion ratio = 4.1, to calculate the volatilities of individual codons, as suggested by Plotkin et al. (http://volatility.cgr.harvard.edu/cgi-bin/volatility.pl). We compute Plotkin's volatility P values using their Web site (http://volatility.cgr.harvard.edu/cgi-bin/volatility.pl).
We measured translational codon bias per gene and per amino acid separately. To measure translational codon bias, we computed the CAI according to Sharp and Li (1987) with the corrections suggested by Bulmer (1988). We also calculated the frequency of optimal codons (FOP) according to the list of optimal codons for S. cerevisiae given by Kliman, Naheelah, and Santiago (2003). Volatility values and codon bias statistics were calculated for the S. cerevisiae sequence because this is the best studied of the yeasts.
We used PAML (Yang 1997) to compute dN, dS, and dN/dS for each gene using the F3 x 4 model in which codon frequencies are estimated from the nucleotide frequencies at the three codon positions. Because a physical definition of a site is more appropriate for the measurement of the synonymous substitution rate (dS), we express dS per codon (Bierne and Eyre-Walker 2003). We performed all our analyses on both dN and dN/dS. Although dN/dS is often regarded as a better measure of the selection acting upon nonsynonymous sites, it may not be in organisms, like yeast, in which there is selection on synonymous codon use. Indeed we note that there is a strong correlation between dS per codon and codon usage bias in our data (tables 2–4).
Table 2 Spearman's Rank Correlation Coefficients Between dN, dN/dS, dS Per Codon, or dS and Volatility or Translational Codon Usage Bias for Each Gene
Table 4 Spearman's Rank Correlation Coefficients Between dN, dN/dS, and dS Per Codon and Translational Codon Usage Bias for the Individual Amino Acids Not Effecting Volatility
Table 3 Spearman's Rank Correlation Coefficients Between dN, dN/dS, and dS Per Codon and Volatility or Translational Codon Usage Bias for the Five Amino Acids Effecting Volatility
Results
Confirming the analysis of Plotkin, Dushoff, and Fraser (2004), we found a highly significant correlation between the volatility P value of Plotkin, Dushoff, and Fraser 2004, or average volatility, and dN/dS (or dN) per gene (table 2). We also confirm the result of Pal, Papp, and Hurst (2001) that there is a strong correlation between measures of translational codon bias (FOP and CAI) and dN/dS (or dN) per gene (table 2).
So, is the observed correlation between volatility and dN/dS (or dN) due to the correlation between translational codon bias and dN/dS (or dN) or vice versa? To answer this, we look at the five volatility-affecting amino acids individually (table 3). We only observe a positive correlation between volatility and dN/dS (or dN) for three of the amino acids (Ile, Leu, and Ser). The two amino acids which show a negative correlation between volatility and dN/dS (or dN), opposing the expectation of Plotkin, Dushoff, and Fraser 2004, are those (Arg, Gly) for which high translational codon usage (in high expression genes) leads to low volatility (see table 1). There is also no indication that volatility affects the correlation; the correlation between translational codon bias and dN/dS (or dN) is as strong for Arg and Gly, as for Ile, Leu, and Ser (table 3).
The correlation between translational codon bias and dN/dS (or dN) is very consistent across amino acids—for almost every amino acid the correlation is negative and often significant, and if it is positive, the correlation is small and nonsignificant (table 4).
Discussion
We have shown that the observed correlation between dN/dS (or dN) and volatility is an incidental correlation caused by a correlation between dN/dS (or dN) and translational codon bias—dN/dS (or dN) correlates negatively with translational codon bias and volatility, for those amino acids in which the translationally optimal codons are high in volatility. This suggests that dN/dS (or dN) is not directly correlated to volatility and that volatility is therefore not the best, or even a good, predictor of dN/dS (or dN). This is not unexpected given recent theoretical work, which suggests that volatility will only be a measure of selection under rather specific conditions (Plotkin et al. 2004).
Our results may seem surprising given that Plotkin, Dushoff, and Fraser (2005) report a significant partial correlation between volatility P value and dN/dS in yeast using CAI to control for translational codon bias, a result we can confirm on our smaller data set (table 5). However, volatility P value is not normally distributed, so the probability of the partial correlation is not necessarily accurate, and the significance of the partial correlation depends critically on the volatility statistic used. If we use our average volatility instead of the volatility of Plotkin et al., which will depend to some extent on gene length and amino acid composition (see Materials and Methods), then the partial correlation between dN/dS (or dN) and average volatility, controlling for translational codon bias, becomes very small and nonsignificant, while the partial correlation between dN/dS (or dN) and translational codon bias remains (table 5). The strongest correlations, either simple or partial, that we observe are between translational codon bias and dN/dS (or dN), which suggests that these are the primary correlations (tables 2 and 3).
Table 5 Partial Correlations of Measures of Translational Codon Bias and Measures of Volatility Measures for Each Gene with dN and dN/dS
It is also interesting to note that the correlation between codon bias and dN is consistently stronger than the correlation between codon bias and dN/dS. This is probably due to the fact that dS is correlated to codon bias and that this correlation is due to selection on codon usage bias and not variation in the mutation rate.
Although volatility does not appear to be a good measure of selection, Plotkin, Dushoff, and Fraser (2004) may have been correct in asserting that it may be possible to infer something about dN in a gene from a single genome sequence. A negative correlation between translational codon bias and dN has now been described in three different organisms: enteric bacteria (Sharp 1991; Rocha and Danchin 2004), Drosophila (Akashi 1994; Betancourt and Presgraves 2002; Marais et al. 2004), and yeast (Pal, Papp, and Hurst 2001), and we have shown that the correlation is consistent for all amino acids in yeast. Furthermore, although the basis of this correlation is unknown and subject to much debate (Betancourt and Presgraves 2002; Marais et al. 2004), at least one of the explanations is likely to lead to the correlation being widespread. It has been suggested that the correlation between codon bias and dN arises through a correlation in the strength of selection acting upon synonymous and nonsynonymous mutations, probably as a consequence of selection for translational accuracy—important amino acid sites in a protein will be subject to strong selection to be conserved during evolution and to be accurately translated (Akashi 1994). Thus any genome, in which selection for translational accuracy is effective, should show the correlation, and it may therefore be possible to use codon bias, maybe in combination with other information, such as amino acid composition or structural data (Tourasse and Li 2000), to predict which genes are likely to be fast-evolving genes. So, although volatility has come in for much criticism, Plotkin and colleagues may have drawn our attention to an approach to an important problem of some utility.
Acknowledgements
We thank Daniel Jeffares for some initial work, Stephan Hutter and Pieter van Beek for help with Perl, and an anonymous referee for helpful comments.
References
Akashi, H. 1994. Synonymous codon usage in Drosophila melanogaster: natural selection and translational accuracy. Genetics 136:927–935.
Betancourt, A., and D. Presgraves. 2002. Linkage limits the power of natural selection in Drosophila. Proc. Natl. Acad. Sci. USA 99(21):13616–13620.
Bierne, N., and A. Eyre-Walker. 2003. The problem of counting sites in the estimation of the synonymous and nonsynonymous substitution rates: implications for the correlation between synonymous substitution rate and codon usage bias. Genetics 165:1587–1597.
Bulmer, M. 1988. Are codon usage patterns in unicellular organisms determined by selection mutation balance? J. Evol. Biol. 1:15–26.
Chen, W., J. J. Emerson, and T. M. Martin. 2005. Not detecting selection using a single genome. Nature 433:E6–E7.
Coghlan, A., and K. H. Wolfe, 2000. Relationship of codon bias to MRNA concentration and protein length in Saccharomyces cerevisiae. YEAST 16:1131–1145.
Dagan, T., and D. Graur. 2004. The comparative method rules! Codon volatility cannot detect positive Darwinian selection using a single genome sequence. Mol. Biol. Evol. 22:1260–1272.
Friedman, R., and A. L. Hughes. 2004. Codon volatility as an indicator of positive selection: data from eukaryotic genome comparisons. Mol. Biol. Evol. 22:542–546.
Hahn, M., J. G. Mezey, D. J. Begun, J. H. Gillespie, A. D. Kern, C. H. Langley, and L. Moyle. 2005. Codon bias and selection on single genomes. Nature 433:E5.
Kellis, M., N. Patterson, M. Endrizzi, and E. S. Lander. 2003. Sequencing and comparison of yeast species to identify genes and regulatory elements. Nature 423:241–254.
Kliman, R. M., I. Naheelah, and M. Santiago. 2003. Selection conflicts, gene expression, and codon usage trends in yeast. J. Mol. Evol. 57:98–109.
Marais, G., T. Domazet-Loso, D. Tautz, and B. Charlesworth. 2004. Correlated evolution of synonymous and nonsynonymous sites in Drosophila. J. Mol. Evol. 59:771–779.
Nielsen, R., and M. J. Hubisz. 2005. Detecting selection needs comparative data. Nature 433:E6.
Pal, C., B. Papp, and L. D. Hurst. 2001. Highly expressed genes in yeast evolve slowly. Genetics 158:927–931.
Plotkin, J. B., J. Dushoff, M. M. Desai, and H. B. Fraser. 2004. Synonymous codon usage and selection on proteins. (http://www.arxiv.org/abs/q-bio/0410013).
Plotkin, J. B., J. Dushoff, and H. B. Fraser. 2004. Detecting selection using a single genome sequence of M. tuberculosis and P. falciparum. Nature 428:942–945.
———. 2005. Reply. Nature 433:E7–E8.
Rocha, E. P. C., and A. Danchin. 2004. An analysis of determinants of amino acid substitution rates in bacterial proteins. Mol. Biol. Evol. 21(1):108–116.
Sharp, P. M., 1991. Determinants of DNA sequence divergence between Escherichia coli and Salmonella typhimurium: codon usage, map position, and concerted evolution. J. Mol. Evol. 33:23–33.
Sharp, P. M. 2004. Gene "volatility" is most unlikely to reveal adaptation. Mol. Biol. Evol. 22:807–809.
Sharp, P. M., and E. Cowe. 1991. Synonymous codon usage in Saccharomyces cerevisiae. Yeast 7:657–678.
Sharp, P. M., and W.-H. Li. 1987. The codon adaptation index—a measure of directional synonymous codon usage bias, and its potential applications. Nucleic Acids Res. 15:1281–1295.
Tourasse, N., and W.-H. Li. 2000. Selective constraints, amino acid composition and the rate of protein evolution. Mol. Biol. Evol. 17(4):656–664.
Yang, Z. 1997. PAML: a program package for phylogenetic analysis by maximum likelihood. Comput. Appl. Biosci. 13:555–556.
Zhang, J. 2005. On the evolution of codon volatility. Genetics 169:495–501.(Nina Stoletzki*, John Wel)
E-mail: stoletzki@zi.biologie.uni-muenchen.de; a.c.eyre-walker@sussex.ac.uk.
Abstract
It has been suggested that volatility, the proportion of mutations which change an amino acid, can be used to infer the level of natural selection acting upon a gene. This conjecture is supported by a correlation between volatility and the rate of nonsynonymous substitution (dN), or the ratio of nonsynonymous and synonymous substitution rates, in a variety of organisms. These organisms include yeast, in which the correlations are quite strong. Here we show that these correlations are a by-product of a correlation between synonymous codon bias toward translationally optimal codons and dN. Although this analysis suggests that volatility is not a good measure of the selection, we suggest that it might be possible to infer something about the level of natural selection, from a single genome sequence, using translational codon bias.
Key Words: volatility ? codon bias ? selection ? nonsynonymous substitution rate ? yeast
Introduction
Understanding the nature of natural selection on DNA sequences is one of the central goals of molecular evolution. Plotkin, Dushoff, and Fraser (2004) and Plotkin et al. (2004) have recently suggested that it is possible to infer the level of natural selection, both positive and negative, acting upon a gene from a single genome sequence. They suggest that this can be achieved by measuring "volatility"—volatility is the proportion of point mutations in a gene, which do not yield a stop codon, which change an amino acid. They base their method on the prediction that genes which have recently undergone amino acid substitutions should be populated by codons with high volatility (Plotkin et al. 2004). In support of their thesis they show that in both Mycobacterium and Saccharomyces species, there is a correlation between volatility and the rate of nonsynonymous substitution (dN) or the ratio of nonsynonynous and synonymous substitution rates (dN/dS). This correlation is quite strong in yeast, which suggests that volatility might be a useful measure of selection.
The idea that volatility can measure the level of selection, either positive or negative, on a gene has been criticized on a number of grounds (Dagan and Graur 2004; Friedman and Hughes 2004; Sharp 2004; Chen, Emerson, and Martin 2005; Hahn et al. 2005; Nielsen and Hubisz 2005; Zhang 2005). Much of the debate has centered around the reasons why volatility is not expected to correlate to dN and dN/dS. For example, it has been suggested that volatility is unlikely to measure selection because (1) it only depends on four or five amino acids (Dagan and Graur 2004; Sharp 2004; Chen, Emerson, and Martin 2005), (2) it has low variance (Dagan and Graur 2004), and (3) simple models of evolution fail to yield a correlation between dN/dS and volatility (Dagan and Graur 2004; Nielsen and Hubisz 2005; Zhang 2005). However, volatility is correlated to dN/dS (and dN); so much of this discussion, while interesting is slightly tangential. The crucial question is why there is a correlation.
Almost all of these critiques point out that volatility is a measure of codon usage bias. As such, the apparent correlation between volatility and dN/dS in yeast may, in fact, as Hahn et al. (2005) suggest, be due to a correlation between translational codon bias and dN/dS. Although Hahn et al. suggest that the correlation between volatility and dN/dS may be due to a correlation between translational codon bias and selective constraint they do not resolve whether this is the case. They show that a measure of translational codon bias, codon adaptation index (CAI), explains more of the variance in volatility than dN/dS in yeast, but they do not pursue the matter further. Plotkin, Dushoff, and Fraser (2005) investigate the partial correlation between dN/dS and volatility controlling for CAI and show it is significant, but they fail to give the magnitude of the effect.
It, therefore, remains very unclear if the principle correlation is between dN/dS and translational codon bias, with the correlation between dN/dS and volatility a by-product of this, or whether the principle correlation is between dN/dS and volatility. Also, it might be that both translational codon bias and volatility separately correlate to dN/dS.
To investigate the matter further, we take advantage of the fact that in yeast there is a strong correlation between dN/dS (or dN) and both translational codon bias (Pal, Papp, and Hurst 2001) and volatility (Plotkin, Dushoff, and Fraser 2004) and that in yeast some of the translational optimal codons have relatively high volatility while others have relatively low volatility (table 1). It is well established that codon bias and gene expression are correlated in yeast (see e.g., Coghlan and Wolfe 2000). So volatility per amino acid is expected to increase (Ile, Leu, and Ser) or decrease (Arg and Gly) with translational codon bias or expression level (table 1). For example, the most optimal codon in yeast for argenine is AGA, which has relatively high volatility. If the principle correlation is between dN/dS (or dN) and translational codon bias, then we expect AGA usage to be negatively correlated to dN/dS (or dN), but if the principle correlation is between dN/dS (or dN) and volatility, then we expect AGA usage to be positively correlated to dN/dS (or dN).
Table 1 Synonymous Codon Use of Volatility-Affecting Amino Acids
Our results are unequivocal; in yeast dN/dS (and dN) is negatively correlated to the use of translational optimal codons for all amino acids whose synonymous codons differ in their volatility, even in those whose optimal codons have high volatility. We further show that the correlation between dN/dS (or dN) and translational optimal codon use is universal across all amino acids, including those synonymous codons which do not differ in their volatility. The observed correlation between dN/dS (or dN) and volatility is a by-product of the correlation between dN (or dN/dS) and translational codon bias.
Materials and Methods
We downloaded the gene alignments from the four yeast species sequenced by Kellis et al. (2003). From these we excluded all genes which were not present in all the four yeast species (Saccharomyces cerevisiae, Saccharomyces paradoxus, Saccharomyces mikatae, and Saccharomyces bayanus), which did not have start and stop codons in all species, which had premature stop codons, and which had frameshifting indels. This left 1,077 genes. This is smaller than the data set analyzed by Plotkin, Dushoff, and Fraser (2005) but has less chance of containing pseudogenes.
Plotkin, Dushoff, and Fraser (2004) suggest using a statistic, volatility P value, to measure volatility. This is the probability of a gene having the observed volatility given the average synonymous codon use of the genes in the genome. The volatility P measure of Plotkin, Dushoff, and Fraser 2004 is unlikely to be a very good statistic because it will depend to some extent on gene length (Sharp 2004) and amino acid composition (Dagan and Graur 2004; Zhang 2005)—any statistic based on probability values depends on sample size, and the variance between synonymous codons for volatility differs between amino acids. To account for these shortcomings, we calculated an alternative measure, the average volatility:
where
and Xi is the number of times codon i is used for the amino acid aa, Vi is the volatility of that codon, and n is the number of amino acids whose synonymous codons differ in their volatility. When considering amino acids separately we used Vaa, the average volatility per amino acid. Note that the volatility is only affected by five amino acids whose synonymous codons differ in their volatility—Arg, Gly, Ile, Leu, and Ser (the codons of Ile only differ when the transition:transversion ratio is different from unity). We assume that the transition:transversion ratio = 4.1, to calculate the volatilities of individual codons, as suggested by Plotkin et al. (http://volatility.cgr.harvard.edu/cgi-bin/volatility.pl). We compute Plotkin's volatility P values using their Web site (http://volatility.cgr.harvard.edu/cgi-bin/volatility.pl).
We measured translational codon bias per gene and per amino acid separately. To measure translational codon bias, we computed the CAI according to Sharp and Li (1987) with the corrections suggested by Bulmer (1988). We also calculated the frequency of optimal codons (FOP) according to the list of optimal codons for S. cerevisiae given by Kliman, Naheelah, and Santiago (2003). Volatility values and codon bias statistics were calculated for the S. cerevisiae sequence because this is the best studied of the yeasts.
We used PAML (Yang 1997) to compute dN, dS, and dN/dS for each gene using the F3 x 4 model in which codon frequencies are estimated from the nucleotide frequencies at the three codon positions. Because a physical definition of a site is more appropriate for the measurement of the synonymous substitution rate (dS), we express dS per codon (Bierne and Eyre-Walker 2003). We performed all our analyses on both dN and dN/dS. Although dN/dS is often regarded as a better measure of the selection acting upon nonsynonymous sites, it may not be in organisms, like yeast, in which there is selection on synonymous codon use. Indeed we note that there is a strong correlation between dS per codon and codon usage bias in our data (tables 2–4).
Table 2 Spearman's Rank Correlation Coefficients Between dN, dN/dS, dS Per Codon, or dS and Volatility or Translational Codon Usage Bias for Each Gene
Table 4 Spearman's Rank Correlation Coefficients Between dN, dN/dS, and dS Per Codon and Translational Codon Usage Bias for the Individual Amino Acids Not Effecting Volatility
Table 3 Spearman's Rank Correlation Coefficients Between dN, dN/dS, and dS Per Codon and Volatility or Translational Codon Usage Bias for the Five Amino Acids Effecting Volatility
Results
Confirming the analysis of Plotkin, Dushoff, and Fraser (2004), we found a highly significant correlation between the volatility P value of Plotkin, Dushoff, and Fraser 2004, or average volatility, and dN/dS (or dN) per gene (table 2). We also confirm the result of Pal, Papp, and Hurst (2001) that there is a strong correlation between measures of translational codon bias (FOP and CAI) and dN/dS (or dN) per gene (table 2).
So, is the observed correlation between volatility and dN/dS (or dN) due to the correlation between translational codon bias and dN/dS (or dN) or vice versa? To answer this, we look at the five volatility-affecting amino acids individually (table 3). We only observe a positive correlation between volatility and dN/dS (or dN) for three of the amino acids (Ile, Leu, and Ser). The two amino acids which show a negative correlation between volatility and dN/dS (or dN), opposing the expectation of Plotkin, Dushoff, and Fraser 2004, are those (Arg, Gly) for which high translational codon usage (in high expression genes) leads to low volatility (see table 1). There is also no indication that volatility affects the correlation; the correlation between translational codon bias and dN/dS (or dN) is as strong for Arg and Gly, as for Ile, Leu, and Ser (table 3).
The correlation between translational codon bias and dN/dS (or dN) is very consistent across amino acids—for almost every amino acid the correlation is negative and often significant, and if it is positive, the correlation is small and nonsignificant (table 4).
Discussion
We have shown that the observed correlation between dN/dS (or dN) and volatility is an incidental correlation caused by a correlation between dN/dS (or dN) and translational codon bias—dN/dS (or dN) correlates negatively with translational codon bias and volatility, for those amino acids in which the translationally optimal codons are high in volatility. This suggests that dN/dS (or dN) is not directly correlated to volatility and that volatility is therefore not the best, or even a good, predictor of dN/dS (or dN). This is not unexpected given recent theoretical work, which suggests that volatility will only be a measure of selection under rather specific conditions (Plotkin et al. 2004).
Our results may seem surprising given that Plotkin, Dushoff, and Fraser (2005) report a significant partial correlation between volatility P value and dN/dS in yeast using CAI to control for translational codon bias, a result we can confirm on our smaller data set (table 5). However, volatility P value is not normally distributed, so the probability of the partial correlation is not necessarily accurate, and the significance of the partial correlation depends critically on the volatility statistic used. If we use our average volatility instead of the volatility of Plotkin et al., which will depend to some extent on gene length and amino acid composition (see Materials and Methods), then the partial correlation between dN/dS (or dN) and average volatility, controlling for translational codon bias, becomes very small and nonsignificant, while the partial correlation between dN/dS (or dN) and translational codon bias remains (table 5). The strongest correlations, either simple or partial, that we observe are between translational codon bias and dN/dS (or dN), which suggests that these are the primary correlations (tables 2 and 3).
Table 5 Partial Correlations of Measures of Translational Codon Bias and Measures of Volatility Measures for Each Gene with dN and dN/dS
It is also interesting to note that the correlation between codon bias and dN is consistently stronger than the correlation between codon bias and dN/dS. This is probably due to the fact that dS is correlated to codon bias and that this correlation is due to selection on codon usage bias and not variation in the mutation rate.
Although volatility does not appear to be a good measure of selection, Plotkin, Dushoff, and Fraser (2004) may have been correct in asserting that it may be possible to infer something about dN in a gene from a single genome sequence. A negative correlation between translational codon bias and dN has now been described in three different organisms: enteric bacteria (Sharp 1991; Rocha and Danchin 2004), Drosophila (Akashi 1994; Betancourt and Presgraves 2002; Marais et al. 2004), and yeast (Pal, Papp, and Hurst 2001), and we have shown that the correlation is consistent for all amino acids in yeast. Furthermore, although the basis of this correlation is unknown and subject to much debate (Betancourt and Presgraves 2002; Marais et al. 2004), at least one of the explanations is likely to lead to the correlation being widespread. It has been suggested that the correlation between codon bias and dN arises through a correlation in the strength of selection acting upon synonymous and nonsynonymous mutations, probably as a consequence of selection for translational accuracy—important amino acid sites in a protein will be subject to strong selection to be conserved during evolution and to be accurately translated (Akashi 1994). Thus any genome, in which selection for translational accuracy is effective, should show the correlation, and it may therefore be possible to use codon bias, maybe in combination with other information, such as amino acid composition or structural data (Tourasse and Li 2000), to predict which genes are likely to be fast-evolving genes. So, although volatility has come in for much criticism, Plotkin and colleagues may have drawn our attention to an approach to an important problem of some utility.
Acknowledgements
We thank Daniel Jeffares for some initial work, Stephan Hutter and Pieter van Beek for help with Perl, and an anonymous referee for helpful comments.
References
Akashi, H. 1994. Synonymous codon usage in Drosophila melanogaster: natural selection and translational accuracy. Genetics 136:927–935.
Betancourt, A., and D. Presgraves. 2002. Linkage limits the power of natural selection in Drosophila. Proc. Natl. Acad. Sci. USA 99(21):13616–13620.
Bierne, N., and A. Eyre-Walker. 2003. The problem of counting sites in the estimation of the synonymous and nonsynonymous substitution rates: implications for the correlation between synonymous substitution rate and codon usage bias. Genetics 165:1587–1597.
Bulmer, M. 1988. Are codon usage patterns in unicellular organisms determined by selection mutation balance? J. Evol. Biol. 1:15–26.
Chen, W., J. J. Emerson, and T. M. Martin. 2005. Not detecting selection using a single genome. Nature 433:E6–E7.
Coghlan, A., and K. H. Wolfe, 2000. Relationship of codon bias to MRNA concentration and protein length in Saccharomyces cerevisiae. YEAST 16:1131–1145.
Dagan, T., and D. Graur. 2004. The comparative method rules! Codon volatility cannot detect positive Darwinian selection using a single genome sequence. Mol. Biol. Evol. 22:1260–1272.
Friedman, R., and A. L. Hughes. 2004. Codon volatility as an indicator of positive selection: data from eukaryotic genome comparisons. Mol. Biol. Evol. 22:542–546.
Hahn, M., J. G. Mezey, D. J. Begun, J. H. Gillespie, A. D. Kern, C. H. Langley, and L. Moyle. 2005. Codon bias and selection on single genomes. Nature 433:E5.
Kellis, M., N. Patterson, M. Endrizzi, and E. S. Lander. 2003. Sequencing and comparison of yeast species to identify genes and regulatory elements. Nature 423:241–254.
Kliman, R. M., I. Naheelah, and M. Santiago. 2003. Selection conflicts, gene expression, and codon usage trends in yeast. J. Mol. Evol. 57:98–109.
Marais, G., T. Domazet-Loso, D. Tautz, and B. Charlesworth. 2004. Correlated evolution of synonymous and nonsynonymous sites in Drosophila. J. Mol. Evol. 59:771–779.
Nielsen, R., and M. J. Hubisz. 2005. Detecting selection needs comparative data. Nature 433:E6.
Pal, C., B. Papp, and L. D. Hurst. 2001. Highly expressed genes in yeast evolve slowly. Genetics 158:927–931.
Plotkin, J. B., J. Dushoff, M. M. Desai, and H. B. Fraser. 2004. Synonymous codon usage and selection on proteins. (http://www.arxiv.org/abs/q-bio/0410013).
Plotkin, J. B., J. Dushoff, and H. B. Fraser. 2004. Detecting selection using a single genome sequence of M. tuberculosis and P. falciparum. Nature 428:942–945.
———. 2005. Reply. Nature 433:E7–E8.
Rocha, E. P. C., and A. Danchin. 2004. An analysis of determinants of amino acid substitution rates in bacterial proteins. Mol. Biol. Evol. 21(1):108–116.
Sharp, P. M., 1991. Determinants of DNA sequence divergence between Escherichia coli and Salmonella typhimurium: codon usage, map position, and concerted evolution. J. Mol. Evol. 33:23–33.
Sharp, P. M. 2004. Gene "volatility" is most unlikely to reveal adaptation. Mol. Biol. Evol. 22:807–809.
Sharp, P. M., and E. Cowe. 1991. Synonymous codon usage in Saccharomyces cerevisiae. Yeast 7:657–678.
Sharp, P. M., and W.-H. Li. 1987. The codon adaptation index—a measure of directional synonymous codon usage bias, and its potential applications. Nucleic Acids Res. 15:1281–1295.
Tourasse, N., and W.-H. Li. 2000. Selective constraints, amino acid composition and the rate of protein evolution. Mol. Biol. Evol. 17(4):656–664.
Yang, Z. 1997. PAML: a program package for phylogenetic analysis by maximum likelihood. Comput. Appl. Biosci. 13:555–556.
Zhang, J. 2005. On the evolution of codon volatility. Genetics 169:495–501.(Nina Stoletzki*, John Wel)