Answers are grounded in this paper and cite their sources.
THE rate of protein evolution shows considerable variation among genes. This variation is thought to reflect differences in the proportion of the sequence that is critical to fulfill given functions (Li 1997), but local recombination intensity (Williams and Hurst 2000) might also have an influence. Recent analysis of mammalian genes has revealed another significant pattern: broadly expressed proteins (i.e., those expressed in many tissues) tend to evolve slower than tissue-specific ones (Kuma et al. 1995; Hastings 1996; Duret and Mouchiroud 2000). A possible interpretation of these findings is that proteins expressed in many tissues interact with a greater number of molecules and may also have to function under widely changing cellular conditions. Hence, high pleiotropic constraints on these genes reduce sequence divergence (Kuma et al. 1995; Hastings 1996; Duret and Mouchiroud 2000). This model can, however, be shown not to be a general explanation for the pattern if it is found in an organism without tissues. It also begs the question as to whether the rate, as well as breadth, of expression is a predictor of the rate of evolution. Analysis of the molecular evolution of proteins in yeast allows us to address these issues, not least because of the availability of large-scale microarray expression data from the yeast Saccharomyces cerevisiae. Additionally there are extensive sequence data from the same species and close relatives. Furthermore, the presence of numerous duplicates in yeast allows us to compare rates of evolution for functionally related and sequence-related genes. These duplicates appear for the most part to be the product of a genome duplication ∼100 million years ago that was followed by a period of gene loss (Wolfe and Shields 1997). Approximately 16% of the genes in the S. cerevisiae genome are retained as duplicates (Wolfe and Shields 1997; Seoighe and Wolfe 1999a). However, a potential problem with the use of these duplicates is the interference of gene conversion giving artificially low rates of divergence. For example, a recent whole gene conversion event between two duplicates would result in 100% identity between the sequences, even if prior to this event they were rapidly evolving. We conduct a number of analyses to establish whether this is a major problem. These tests also provide further evidence for an ancient genome duplication event in yeast. The dataset used in this article consists of the remnants of the ancient genome duplication including 376 gene pairs organized into 55 gene clusters (Wolfe and Shields 1997). Because of their possibly unusual evolution (high expression level, very low sequence divergence), genes coding for ribosomal proteins were excluded from the analysis. S. cerevisiae amino acid sequences were obtained from Saccharomyces Genome Database (available at ftp://genome-ftp.stanford.edu/pub/yeast/yeast_ORFs). Sequence data for Candida albicans were obtained from the Stanford DNA Sequencing and Technology Center website at http://www-sequence.stanford.edu/group/candida. Protein sequence alignments were carried out with CLUSTAL-W version 1.4 (Thompson et al. 1994), using the default settings. Alignments were checked by eye. Some sequence alignments were ambiguous and, therefore, were omitted from the analysis. Nucleotides were aligned using the protein alignment as a template. The numbers of synonymous substitutions per site (KS) were estimated using Li’s method (Li 1993), with modifications (Pamilo and Bianchi 1993), and applying Kimura’s two-parameter model to correct for multiple hits. Protein distances (dA) were calculated using PHYLIP (Felsenstein 1989), employing PAM substitution matrices and default settings. To err on the side of caution, protein sequences with 100% identity were excluded, as these might represent recent gene conversion events. This process led to 315 gene pairs. Whole genome transcription data from Holstege et al. (1998) were used as estimates of gene expression levels (http://web.wi.mit.edu/young/expression/). Using microarray technology, these authors estimated the number of mRNA molecules per cell and mRNA half-life for a large collection of genes in S. cerevisiae under conditions of growth to mid-log phase in YPD media. Only gene pairs where both genes have annotated mRNA levels and dA < 1 were retained, leading to a dataset of 185 gene pairs. —Correlation between the mRNA level and protein distance in duplicated genes of yeast. r = -0.584, P < 10-6. We analyzed the rate of protein evolution in the duplicated Saccharomyces genes using, as a reference, sequences from another related organism, C. albicans. This organism is a close relative of S. cerevisiae and it is also known to have branched off from the Saccharomyces genus before the genome duplication in yeast (Seoighe and Wolfe 1999b). We carried out BLASTP 2.0 amino acid sequence similarity search of the duplicated yeast genes against all available C. albicans proteins, using BLOSUM62 substitution matrices and SEG filter for low complexity regions (Altschul et al. 1997). The criteria used to define homologs were as follows: (1) BLASTP expectation value, E < 10-20 for both gene pairs; (2) highest score for both duplicated S. cerevisiae genes; and (3) the duplicate-to-duplicate distance is smaller than the duplicate-to-Candida distances. The alignment method and the calculation of protein distances were the same as above. Only gene pairs with dA < 1.5 were retained. Do highly expressed genes evolve at a low rate? There is a statistically highly significant negative association between gene expression level and protein distance (Figure 1; Pearson correlation, r = -0.584, P < 10-6). It was shown that codon usage bias [assessed by the codon adaptation index (CAI)] strongly correlates with mRNA abundance in yeast (Coghlan and Wolfe 2000). In further support of our finding, CAI and the rate of protein evolution show a strong negative association (r = -0.617, P < 10-6). A possible explanation for the above results is that highly expressed genes are overrepresented by functionally more important gene families or that they have unusual sequence characteristics that lead to them evolving slowly. To evaluate whether the correlation between sequence divergence and gene expression level merely reflects differences of this nature, two complementary tests were done. In the first approach, we used a smaller dataset of the whole sample containing only gene pairs within the same broad functional category. Analyzing only genes coding for enzymes of carbohydrate, amino acid, or nucleotide metabolism, the significant correlation between gene expression level and the rate of protein evolution remains (Spearman-rank correlation, N = 29, r = -0.485, P = 0.0071). This controls for functional equivalence but not for sequence similarity. In a second, more rigorous approach, we examined the relative rate of protein evolution of each member of the duplicated pair when compared with the same Candida ortholog. We could then ask whether the slower evolving of the two duplicates also has the higher expression level. We found a significant negative correlation between the relative difference in gene expression levels and the relative difference in protein distances (P < 0.01, r = -0.215; see Figure 2). The more highly expressed duplicate evolves at a lower rate in 87 out of 150 gene pairs, but this is not significant (sign-test, P = 0.06). Given that most of the duplicated genes in yeast have highly overlapping functions (Seoighe and Wolfe 1999b) and similar sequences, it is unlikely that the correlation between gene expression level and protein divergence merely reflects functional differences or sequence-specific differences. Rather, these results suggest that the gene expression level per se affects protein evolution. The results above are, however, of less utility if the duplicates were not all derived at the same time as a result of a genome duplication event. Were they derived independently at numerous times, low divergence between two genes might indicate recent duplication rather than slow evolution. A further problem with the use of these duplicates is the possibility of gene conversion. Highly expressed genes may be more prone to frequent gene conversion events, leading to reduced sequence divergence between the pairs (e.g., Petes and Hill 1988). This problem is, in effect, the same as that of multiple different times of duplication. We performed several tests, all of which suggest that these explanations are unlikely. —Correlation between pairwise differences of mRNA level and the rate of protein evolution in duplicated genes. r = -0.215, P < 0.01. In the first, we compared the duplicate-to-duplicate distance with the average duplicate-to-outgroup distance. Under the assumption of frequent gene conversion events (or independent duplication events), the two distances should be uncoupled. The reason is the following. Let us assume that after the duplication event, one copy has accumulated some mutations. As long as the divergence between the duplicates is small, one copy may convert the other. This process would retard the rate of divergence between the duplicates, while divergence of the duplicates from orthologous sequences may remain relatively unaffected. If some sequences have undergone gene conversion but most have not, those that have been homogenized should appear as outliers (below the best-fit line) in the regression of duplicate-to-duplicate distance against duplicate-to-outgroup distance. A BLAST search was evaluated to find C. albicans genes whose sequences were most similar to the 315 duplicated genes in yeast (for details see methods). With the resulting 160 gene pairs with appropriate out-group sequences, a remarkably strong correlation was found between the duplicate-to-duplicate distance and the average of the duplicate-to-outgroup distances (Pearson correlation, r = 0.870, P < 10-6; see Figure 3). To search for possible outliers, the standardized residuals (SR) from the regression line were computed for all data points. Only 3 gene pairs were detected with suspiciously low duplicate-to-duplicate distance, defined as SR < -2, corresponding to a 0.05 significance level (see Table 1A). The slope of the regression equation (≅0.6) also provides an estimate of the timing of the genome duplication event. The two species probably diverged ∼140-330 million years ago (Berbee and Taylor 1993; Pesole et al. 1995). By multiplying the divergence time with the slope of the regression line, we can get a date of the genome duplication event. This method dates the genome duplication to ∼0.8-2 × 108 years ago. This is in accordance with the previous estimate, of 108 years (Wolfe and Shields 1997). Another way to detect gene conversion is to examine the number of synonymous substitutions per site (KS) between the duplicates. Under the assumptions that synonymous sites evolve neutrally and no gene conversion occurs between the duplicates, KS is expected to be roughly constant. Unfortunately, in the case of yeast, the first assumption is likely to be invalid. Codon usage shows a strong bias in highly expressed genes, most likely resulting from selection for translational efficiency (e.g., Coghlan and Wolfe 2000). Indeed, the estimated number of synonymous substitutions per site is negatively correlated with codon usage bias assessed by the codon adaptation index (N = 160, r = -0.706, P < 10-6). We may then presume that gene pairs with suspiciously low KS compared to their codon adaptation indices might recently have undergone gene conversion. These gene pairs were detected using the standard residuals inferred from the regression line of the KS - CAI analysis. We found only 4 gene pairs with SR < -2 (see Table 1B). YDR312W-YHR066W is the most significant outlier gene pair in both analyses. After elimination of the outlier gene pairs inferred from the two analyses, the strong negative correlation between gene expression level and duplicate-to-duplicate distance remains (data not shown). —Correlation between duplicate-to-duplicate distance and duplicate-to-outgroup distance. r = 0.870, P < 10-6. We found that (1) highly expressed genes evolve slowly in yeast, (2) this is not due simply to these genes being functionally or sequence related, and (3) it is not due to gene conversion. Additionally, we can conclude that the pattern is not likely to be the result of mutational patterns alone. In yeast, increased transcription rates are associated with increased mutation rates (Datta and Jinks-Robertson 1995; Morey et al. 2000). This is the opposite of what is required for mutation rate differences to explain the negative correlation between transcription rate and amino acid substitution rate. Our results show that tissue-specific gene expression differences cannot fully account for the slower evolution of highly expressed genes. One possible explanation for the results is that, as previously postulated (Duret and Mouchiroud 2000), commonly expressed genes need to act in several different conditions and this may limit the number of sites within the gene at which mutations will be effectively neutral. We can, however, imagine other explanations. There could be an effect of selection on the half-life of proteins or mRNA. Alternatively, the correlation between expression and evolutionary rate may simply reflect another factor that happens to covary with expression levels, for example, the recombination rate. We leave a test of this hypothesis to future work. Gene pairs with suspiciously low duplicate-to-duplicate distances, possibly indicating recent gene conversion events Results inferred from the comparison of duplicate-to-duplicate with the duplicate-to-Candida distances. Results inferred from the KS - CAI analysis. Gene pairs with suspiciously low duplicate-to-duplicate distances, possibly indicating recent gene conversion events Results inferred from the comparison of duplicate-to-duplicate with the duplicate-to-Candida distances. Results inferred from the KS - CAI analysis. As regards selection on the half-life of proteins or mRNA, the logic could run in many ways. Gene dosage requirements might impose strong selection pressure to reduce the rate of degradation of protein or mRNA. If a gene product is required at a high expression level, selection might also favor longer persistence of the protein product and might also make the mRNA less likely to decay before being translated. Conversely, the product might be expressed at a high rate because it has a high decay rate. This might be because its optimal enzymatic configuration is unstable. Selection might preserve the configuration that optimizes the enzymatic abilities ahead of ensuring stability of the structure. In reality a trade-off of these parameters is expected. Analysis of the data casts little extra light on this possibility. Unfortunately, current data on protein degradation rates in yeast are too scarce for a comparative analysis. In contrast, data on mRNA half-life are available for a large number of yeast genes (Holstege et al. 1998). We found a weak, but significant, positive association between the rate of protein evolution and mRNA half-life (N = 2856, r = 0.11, P < 10-6). Furthermore, genes with elevated mRNA levels show reduced mRNA half-life (r = -0.10, P < 10-6). Thus, it is unlikely that selection for conserved mRNA structure could explain why highly expressed genes evolve slowly. Nevertheless, it remains an open question whether broadly expressed proteins have to face extra difficulties in achieving their proper protein structure. Regardless of the uncertainty in interpretation, the results presented here indicate that in looking for the reasons that highly expressed genes evolve slowly, we need not concentrate exclusively on breadth of expression. The absolute rate of expression appears to be of some importance. That is not to say that in mammals tissue breadth has no effect, but it seems unlikely to be the only effect. We are grateful to Deborah Charlesworth, Ken Wolfe, Laurent Duret, and the two anonymous referees for their helpful suggestions. We also thank the European Science Foundation TBA program for providing a travel grant to C.P.
No takes yet. Share an insight, caveat, or question.
Pál et al. (2001) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: