As predicted by the nearly neutral model of evolution, numerous studies have shown that reduced Ne accelerates the accumulation of slightly deleterious changes under genetic drift. While such studies have mostly focused on eukaryotes, bacteria also offer excellent models to explore the effects of Ne. Most notably, the genomes of host-dependent bacteria with small Ne show signatures of genetic drift, including elevated Ka/Ks. Here, I explore the utility of an alternative measure of selective constraint: the per-site rate of radical and conservative amino acid substitutions (Dr/Dc). I test the hypothesis that purifying selection against radical amino acid changes is less effective in two insect endosymbiont groups (Blochmannia of ants and Buchnera of aphids), compared to related gamma-Proteobacteria. Genome comparisons demonstrate a significant elevation in Dr/Dc in endosymbionts that affects the majority (66–79%) of shared orthologs examined. The elevation of Dr/Dc in endosymbionts affects all functional categories examined. Simulations indicate that Dr/Dc estimates are sensitive to codon frequencies and mutational parameters; however, estimation biases occur in the opposite direction as the patterns observed in genome comparisons, thereby making the inference of elevated Dr/Dc more conservative. Increased Dr/Dc and other signatures of genome degradation in endosymbionts are consistent with strong effects of genetic drift in their small populations, as well as linkage to selected sites in these asexual bacteria. While relaxed selection against radical substitutions may contribute, genome-wide processes such as genetic drift and linkage best explain the pervasive elevation in Dr/Dc across diverse functional categories that include basic cellular processes. Although the current study focuses on a few bacterial lineages, it suggests Dr/Dc is a useful gauge of selective constraint and may provide a valuable alternative to Ka/Ks when high sequence divergences preclude estimates of Ks. Broader application of Dr/Dc will benefit from approaches less prone to estimation biases.
Citation: Wernegreen JJ (2011) Reduced Selective Constraint in Endosymbionts: Elevation in Radical Amino Acid Replacements Occurs Genome-Wide. PLoS ONE 6(12): e28905. https://doi.org/10.1371/journal.pone.0028905
Editor: Shin-Han Shiu, Michigan State University, United States of America
Received: September 24, 2010; Accepted: November 16, 2011; Published: December 14, 2011
Copyright: © 2011 Jennifer J. Wernegreen. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: Funding was provided by grants from the National Science Foundation (MCB-1103113) and National Institutes of Health (R01GM062626). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The author has declared that no competing interests exist.
Evolutionary significance of Ne
In considering the evolutionary fate of a new mutation, a critical parameter is the product of the selection coefficient (s) for or against that mutation and effective population size (Ne), which dictates the efficacy of selection. Selection determines the fate of mutations that are strongly deleterious or advantageous (|Nes|≫10), and mutations with negligible fitness effects (|Nes|≪1) should behave neutrally. However, for a narrow range between these values, the fate of mutations will depend on a balance between selection and the stochastic effect of genetic drift. Ohta ,  emphasized that many mutations fall into this ‘nearly neutral’ category, with selection coefficients near the reciprocal of Ne. In general, a random mutation is more likely to be deleterious than beneficial , . Ohta proposed that the fate of mildly deleterious mutations depends on Ne. Namely, when Ne is reduced, genetic drift plays a greater role and purifying selection against such mutations is less effective. This theory generated a key prediction: reduced Ne will lead to the greater accumulation of deleterious changes.
Given the biological significance of slightly deleterious mutations, it is not surprising that much effort has centered on quantifying their abundance and understanding the factors that govern their fate in natural populations. Deleterious mutations likely influence a wide range of biological phenomena, including inbreeding depression, mate choice, Y chromosome and mitochondrial genome degradation, and the maintenance of variation in fitness (reviewed in ). Major evolutionary transitions, such as the emergence of sex and recombination, may be mechanisms to avoid the accumulation of such changes –. Their persistence may also influence the demographic fate of small populations .
As predicted by the nearly neutral theory, empirical studies have documented a greater accumulation of deleterious changes under genetic drift in small populations. This work spans diverse taxonomic groups, including vertebrates (e.g., rodents, primates and birds), invertebrates, plants, fungi, and bacteria. Evidence supporting the importance of Ne includes faster rates of sequence evolution and apparent gene degradation in species with relatively small Ne – or short generation times, which may correspond to small populations , . Consistent with this trend, island species often show greater signatures of deleterious changes than do mainland relatives –. Moreover, observations that mitochondrial datasets often show deviations from neutrality may reflect the lower Ne of haploid, maternally inherited genomes –.
In addition to drift-based dynamics in small populations, genetic linkage can also influence the fate of deleterious changes. Species with little or no recombination, such as asexual species, may lack the ability to purge deleterious mutations, a phenomenon known as Muller's ratchet . Similarly, in regions of low recombination, linkage to selected sites may contribute to deleterious substitutions. For instance, selection against strongly deleterious mutations reduces variation at linked sites and can speed the fixation of slightly deleterious mutations , . Hitchhiking with selective sweeps of beneficial mutations may also drive the fixation of neutral and even deleterious changes at linked sites . In fact, this contribution of selective sweeps to deleterious evolution, termed genetic draft, may be more significant than genetic drift when recombination is low –. Such linkage during selective sweeps has likely contributed to the degradation of Y chromosomes  and perhaps mitochondria , as well as genomes of organisms under strong artificial selection during the process of domestication, including certain crops .
While most studies of deleterious evolution have focused on eukaryotic groups, bacteria also offer promising models to explore the significance of deleterious mutations and processes that influence their dynamics. In particular, the transition from a free-living existence to an obligate host association can profoundly influence Ne , –. By virtue of their constrained lifestyle and the potential for severe bottlenecks upon transmission to new hosts, host-dependent bacteria are thought to have much smaller effective population sizes than related bacterial lineages that can replicate independently –. Host-dependent species include several well-known bacterial pathogens, as well as long-term intracellular mutualists that are critical to the reproduction and success of many insect species . As extreme examples of host specialization, many such mutualists have coevolved with a particular insect group for tens to hundreds of millions of years. The near absence of recombination in these mutualists may exacerbate the effects of genetic drift due to Muller's ratchet  and linkage to selected sites (see above). Abundant evidence points to increased accumulation of deleterious changes in these long-term endosymbionts (reviewed in ) and other host-dependent bacterial species , . This evidence includes a genome-wide acceleration of evolutionary rates that is concentrated at nonsynonymous sites, thus elevating the ratio of nonsynonymous to synonymous substitutions (Ka/Ks and similar measures). In addition, population genetic analyses have revealed patterns that point to reduced Ne, such as exceptionally low levels of intraspecific polymorphism, an excess of nonsynonymous polymorphisms, and excess of rare alleles , . These findings indicate that evolutionary models developed in eukaryotic taxa have predictive power when applied to bacteria.
Measuring selective constraint
Empirical tests of the nearly neutral model often compare the effects of purifying selection in species or gene regions with relatively large versus small Ne, predicting lower selective constraint in the latter. Selective constraint is typically estimated by Ka/Ks, which offers a gauge for the effects of purifying selection on protein divergence. This use of Ka/Ks assumes that nonsynonymous mutations have greater fitness consequences than do synonymous changes, and that protein-altering mutations are more likely to be deleterious than beneficial. On this basis, Ka/Ks will be suppressed to the extent that purifying selection effectively removes nonsynonymous mutations. Conversely, elevated Ka/Ks suggests that purifying selection is relaxed due to reduction of selection coefficients, or less effective due to a greater influence of genetic drift. While relaxed selection is expected to affect particular loci or functional categories, reduced efficacy of selection under genetic drift is expected to elevate Ka/Ks across the genome –. Positive, diversifying selection at a particular gene can also elevate Ka/Ks, but this force is expected to be relatively rare. Ka/Ks has many attributes as a metric of purifying selection, but it is limited to close sequence comparisons for which Ks can be estimated reliably.
For more distant sequence comparisons, the relative abundances of radical and conservative amino acid changes may offer an alternative measure of selective constraint. Several classifications exist to categorize amino acid substitutions based on changes in physiochemical properties such as charge, polarity, and/or volume. While positive selection may favor radical amino acid changes at particular genes (but see , ), most radical changes are expected to be deleterious and therefore subjected to purifying rather than positive selection. The prediction that radical changes experience stronger purifying selection than conservative changes is supported by the far greater abundance of conservative changes in most proteins.
The ratio of per-site rates of radical and conservative amino acid substitutions (Dr/Dc) has been used as a measure of selective constraint, analogous to Ka/Ks, and is potentially valuable for measuring the strength or efficacy of purifying selection , , , , –. An advantage of Dr/Dc is its utility for species pairs in which high synonymous divergence confounds estimates of Ka/Ks. With some exceptions (e.g., ), several studies show elevated Dr/Dc in small populations , , , , a pattern consistent with reduced efficacy of purifying selection against radical amino acid changes.
In the present study, I explore the utility of Dr/Dc to assess selective constraint in bacteria. Although several studies of eukaryotic taxa have employed this ratio, to my knowledge this is the first study exploring its utility for studying bacterial evolution. Specifically, I test the hypothesis that purifying selection against radical amino acid changes is less effective in bacteria with reduced Ne. To this end, I track patterns of protein divergence within genome pairs that span ten bacterial lineages. The endosymbionts include Blochmannia and Buchnera, obligate intracellular mutualists of ants and aphids, respectively. Non-endosymbiotic species include the related enteric bacteria Escherichia coli and Salmonella enterica serovar Typhimurium (called Salmonella typhimurium here), and the more distantly related gamma-Proteobacteria Shewanella sp., Acinetobacter sp., and Pseudomonas putida. Although the non-endosymbiotic species can be host-associated, they do not rely on hosts for reproduction.
However promising, Dr/Dc is not without controversy. For example, its use for detecting positive selection is questionable . Although I use Dr/Dc to measure the effects of purifying (not positive) selection, I must still consider that mutational parameters and base composition can influence estimates of Dr and Dc and their ratio , . Compared to Ka/Ks, far less effort has been devoted to mitigating estimation biases for Dr/Dc. Smith  found that sequence composition can influence Dr/Dc and developed modifications implemented by in-house scripts to reduce that bias; however, the modifications are not readily accessible. In the current study, I use simulations to evaluate estimation biases and to confirm that they cannot explain patterns observed in genome comparisons.
As detailed below, both insect endosymbionts show a consistent increase in Dr/Dc compared to their free-living relatives. Simulations confirm a significant effect of base composition on Dr/Dc but in the opposite direction of the patterns observed from genome comparisons, thereby making the Dr/Dc comparisons more conservative. The persistent elevation of Dr/Dc in endosymbionts further supports reduced selective constraint in these species. While other explanations exist, I argue this result is best explained by reduced efficacy of purifying selection against radical amino acid changes in bacteria with small Ne, as predicted by the nearly neutral theory. Although I focused on a few bacterial lineages, this study offers proof in concept that Dr/Dc is useful for assessing selective constraint and may offer a valuable tool when high sequence divergences preclude reliable estimates of Ks. Broader application of Dr/Dc to diverse bacterial genomes would benefit from improved estimation algorithms so that values are less sensitive to base composition and mutational parameters.
Pairwise genome divergence summaries
Schematic relationships among the ten genomes compared are shown in Figure 1 and genome information listed in Table 1. The broader phylogenetic position of such AT-rich endosymbionts is difficult to resolve and is left ambiguous in this tree. However, a recent study of the gamma-Proteobacteria  suggests that Blochmannia and Buchnera represent two independent origins of endosymbiosis. Because Blochmannia and Buchnera have coevolved with their respective insect hosts, sequence divergence within each reflects changes that have occurred in the context of an endosymbiotic lifestyle. Taxa for pairwise comparisons were selected to include phylogenetically independent pairs (boldface, Table 2) that show comparable levels of amino acid divergence for shared orthologs. In addition, comparisons between E. coli vs. S. typhimurium and between B. floridanus vs. B. vafer were included but are not phylogenetically independent from other pairs in the dataset.
Blochmannia and Buchnera have coevolved with their insect hosts, and sequence divergence within each pair reflects changes that have occurred in the context of their endosymbiotic lifestyle. The six pairwise comparisons made across these bacterial lineages are listed in Table 2.
The Reciprocal Sequence Distance (RSD) algorithm  was used to identify orthologs shared within each of the six genome pairs and those shared across all ten genomes. The number of orthologs detected within each genome pair (Table 3, column 2) is based on conservative, stringent detection parameters may exclude some divergent genes. Nonsynonymous and synonymous divergences (dN and dS, respectively) were estimated using a likelihood-based approach  that accounts for the distinct base composition and codon structure of these species (see Methods). The median dS values illustrate that synonymous divergences far exceed saturation for all pairs, except E. coli vs. S. typhimurium where the median dS value still exceeds one (Table 3).
Dr/Dc is associated with nonsynonymous divergence
I initially applied four criteria for categorizing radical substitutions: charge, polarity, the amino acid classification by Miyata et al.  (noted as Dr/Dc-MY) that considers polarity and volume, and the classification scheme of Hanada et al.  (noted as Dr/Dc-HAN) found to correlate with Ka/Ks values in mammalian genes. In order to identify which of the four classifications considered (charge, polarity, MY, and HAN) offers the most promising metric for selective constraint, I tested for an association between dN and Dr/Dc under the four classifications. This follows the general approach of Hanada et al. , who identified a classification scheme that maximized the correlation between Dr/Dc and Ka/Ks. In the present study, Dr/Dc calculated under each classification showed a significant, positive association with dN for several genome pairs (Table 4). Of the six pairs considered, Dr/Dc -polarity showed the strongest association for two pairs, Dr/Dc -HAN for one pair, and Dr/Dc-MY for three pairs (two of which are phylogenetically independent). Based on its overall strong association with dN, Dr/Dc -MY was selected for subsequent analyses.
Analysis of Dr/Dc-MY
Among exceptionally conserved proteins, very low values for Dr and/or Dc (such as zero or slightly above zero) made estimates of Dr/Dc unreliable and resulted in high standard errors. In addition, very rarely, Dr or Dc was exceptionally high and uncorrected values exceeded 0.75, thus precluding application of the Jukes Cantor correction. The “high s.e. filter” omits orthologs with zero radical or zero conservative amino acid substitutions, orthologs for which the standard error of Dr/Dc exceeded 50%, and those for which uncorrected Dr or Dc exceeded 0.75. This filter removed relatively few (<6%) of the orthologs from most pairs (Table 3). However, the filtering of 24.3% of orthologs from the E. coli vs. S. typhimurium comparison reflects the fact that many genes had low amino acid divergences and thus high standard errors of Dr/Dc. Regardless of classification, within each dataset Dr was typically less than Dc (illustrated for Dr/Dc-MY, Table 3), as expected if negative fitness consequences are more severe for radical amino acid changes than for conservative ones.
In addition to its association with dN in some genome pairs, Dr/Dc -MY also shows weak though significant associations with other protein features, such as %GC, aromaticity, and GRAVY (Table S1). Dr/Dc -MY shows a strong association with Dr/Dc based on charge, Hanada's classification, and for some genome pairs, polarity. In Blochmannia and Buchnera, Dr/Dc showed a negative association with %GC content. This may reflect the underlying association between Dr/Dc and dN, since dN and %GC are negatively associated in endosymbionts. (Relatively conserved proteins retain signatures of the ancestral, moderate base composition, whereas divergent proteins show greater impacts of AT compositional biases.) An alternative explanation for the negative association between Dr/Dc and %GC is that elevated AT-content inflates Dr/Dc estimates; however, just the opposite is supported by simulation results below.
Higher Dr/Dc in endosymbionts suggests lower selective constraint
For direct comparisons among the genome pairs, analyses were restricted to the 276 orthologs detected across all ten genomes (Table S2). I further filtered this data to removed orthologs that did not survive the “high s.e. filter” for any one of the five species pairs i–v (Table 2). The resulting dataset retained a common set of 256 orthologs that span a range of metabolic functions. Application of the same filter to the E. coli-S. typhimurium pair retained just 119 of the 276 shared orthologs. For this genome pair, summary statistics for those 119 genes, as well as 221 genes retained by a less stringent filter, are presented. Notably, genome pairs i–v have comparable dN levels for the 256 shared genes (Table 5, column 6).
Among the 256 shared orthologs, Dr/Dc is higher for pairs of endosymbionts than for pairs of free-living gamma-Proteobacteria (Figure 2). Within Buchnera and within Blochmannia, 66–79% of genes show higher Dr/Dc compared to pairs of free-living bacteria (Table 6). This elevation of Dr/Dc in endosymbiont pairs is highly significant by sign and Wilcoxon tests (Table 6). While the elevation of Dr/Dc in Buchnera compared to Blochmannia is significant, the difference is not nearly as extreme as that observed between endosymbiont vs. free-living bacterial pairs.
Among 256 orthologs shared across all six genome pairs, (a) mean values (with bars showing standard error) and (b) median values of Dr/Dc were substantially higher within pairs of endosymbionts (orange) than free-living species (blue). This elevation of Dr/Dc was significant by a sign test and Wilcoxon test (see text and Table 6). Dr/Dc was calculated under the classification scheme of Miyata et al. . A less stringent filter was used for E. coli-S. typhimurium, in order to retain a comparable number of genes for pair. For this pair, the 221 genes analyzed consisted of the 256 orthologs surviving the “high s.e. filter” (see text) in the five other genome pairs, minus 35 genes that contain zero radical and/or conservative substitutions between E. coli and S. typhimurium.
In order to compare Dr/Dc between two genome pairs on a gene-by-gene basis, I calculated the ratio of the two Dr/Dc values (one value for each pair), for each of the 256 genes examined. I refer to this ratio of Dr/Dc ratios as the ‘elevation index’ for brevity. If the two genomes show no difference in Dr/Dc, the expected value of the elevation index is one. Figure 3 shows the distribution of this index for select comparisons, presented on a log scale for ease of visualization. The mean value and 99% confidence intervals are provided for each comparison. For the majority of genes, Dr/Dc in endosymbionts is higher than the corresponding ortholog in free-living pairs (Figure 3 a–d; elevation index >1 for most genes). In each comparison of endosymbionts vs. free-living bacteria, the lower bound of the 99% confidence interval of the elevation index is substantially greater than one (Table 6; Figure 3 a–d), consistent with a substantial elevation of Dr/Dc in the endosymbionts. By contrast, the index shows a more even distribution around one for Buchnera vs. Blochmannia (Figure 3e), and for E. coli-Shewanella sp. vs. Acinetobacter sp.-P.putida (Figure 3f), reflecting comparable Dr/Dc values within these comparisons. While Buchnera showed a higher Dr/Dc ratio than Blochmannia, this is only marginally significant (p<0.017) by the sign test and has a relatively low ChiSquare value in the Wilcoxon test (Table 6). In the comparison between the free-living species pairs, the 99% confidence interval of the elevation index spans one (Figure 3f) and the sign and Wilcoxon test results are nonsignificant. For all possible comparisons of two genome pairs, the above statistics are listed in Table 6.
The ratio of two Dr/Dc values (or, the ‘elevation index’ described in the text) is presented along the x-axis on a log scale for ease of comparison. Mean values and 99% confidence intervals are listed for each comparison. Equality of the two Dr/Dc values would give a ratio of one, indicated by the red reference line. (a–d) Dr/Dc is substantially elevated within Buchnera and Blochmannia when either is compared to related free-living bacteria. Considered in the framework of this figure, significant sign test results for these comparisons (Table 6) indicate that significantly more genes fall above the value of one (i.e., higher Dr/Dc in endosymbionts) than fall below this value. Differences were modest between (e) Buchnera and Blochmannia and (f) free-living bacterial pairs. Dr/Dc was calculated under the classification scheme of Miyata et al. .
Importantly, the 256 shared orthologs span a range of functional categories that include fundamental processes such as information transfer (DNA replication, transcription, and translation) and cell processes (cell division and cell cycle physiology). Across these various categories, endosymbiont genes show a consistent elevation in Dr/Dc (Figure 4; Figure S1; Table S3). The significance of this pattern was tested in three ways. First, to determine if the elevation index (here, (Dr/Dc for endosymbionts)/(Dr/Dc for free-living bacteria)) was significantly greater than one, we estimated its 95% and 99% confidence intervals. The lower bound of the 95% confidence interval exceeded one for all comparisons, and the lower bound of the 99% confidence interval exceeded one for most comparisons (asterisks in Figure 3; confidence intervals listed in Table S3). In addition, for most functional categories, both the sign test and Wilcoxon test were highly significant (Table S3), indicating elevation of Dr/Dc in endosymbionts. The few non-significant results were functional categories with relatively few genes and therefore small sample sizes in the statistical tests. The pervasive nature of this Dr/Dc elevation in endosymbionts across diverse functional groups suggests that the underlying mechanism operates genome-wide (see Discussion).
The ratio of Dr/Dc values for endosymbiont pairs divided by Dr/Dc for free-living bacterial pairs (or, the ‘elevation index’ described in the text) consistently exceeds one, indicating higher Dr/Dc in the endosymbionts. This pattern holds across diverse functional categories, including core cellular processes such as information transfer. Median (dark grey) and mean values (light grey, with bars showing 95% confidence intervals) are based on 256 shared orthologs shared among genomes. Asterisks indicate that the lower bound of the 95% (*) or 99% (**) confidence interval of the ‘elevation index’ exceeds one, which is marked with a hashed reference line. Functional categories are based on MultiFun assignments to E. coli orthologs , . When E. coli genes were assigned to more than one broad category, that ortholog is represented more than once in the summary data presented. The numbers of orthologs analyzed for each functional category are as follows: Cell processes, 23; Cell structure, 73; Information transfer, 122; Metabolism, 112; Regulation, 19; Transport, 22; Unclassified, 14. Figure S1 shows the data points underlying the mean and median values presented here. Table S3 provides 99% confidence intervals and the results of sign and Wilcoxon tests, which also support significant elevation in endosymbiont Dr/Dc values across functional groups.
Sensitivity of Dr/Dc estimates in simulated datasets
Potential estimation biases are critical when comparing bacteria representing different lifestyles. Namely, genomes of host-dependent species often have low %GC compared to free-living relatives (Table 1). Given that sequence composition and mutational parameters can influence Dr/Dc , , I explored whether estimation biases could explain the observed elevation of Dr/Dc in endosymbionts.
DNA sequence datasets were simulated under a codon model that varied in four factors. (Please see the Methods for rationale behind parameter values.) (i) Codon frequencies were especially important to consider here, as this is the factor varies among the genomes compared. Datasets were generated under equilibrium codon frequencies constrained to match the genomic codon composition of Buchnera APS, Blochmannia floridanus, or E. coli. This use of codon tables accounted for differences among genomes in %GC content, relative synonymous codon usage and relative amino acid usage. (ii) dN/dS was set to 0.3 or 0.6. (iii) The distance between sequences, t, measured as mean substitutions per site, was set to 0.5, 1, or 4. (iv) The transition/transversion ratio (kappa) was set to 1, 2, 5, or 10. All combinations of these parameters generated 72 datasets of 500 sequence pairs each. For all simulated datasets the expected Dr/Dc is one, because the codon model does not distinguish radical versus conservative substitutions.
Visual inspection of the median Dr/Dc values shows that many values deviate from one, reflecting biases in the estimate (Figure 5). Codon frequencies have a consistent and large effect (Figure 6). In particular, sequences simulated under the AT-rich codon frequencies of endosymbionts showed a considerable reduction in Dr/Dc estimates. This bias was exacerbated as kappa increased. Varying t or dN/dS had less of an effect. To quantify the sensitivity of Dr/Dc estimates to the four simulation parameters, I used a multi-way analysis of variance (ANOVA). F values (Table 7) reflect the magnitude of the effects and illustrate a high sensitivity of Dr/Dc to codon frequencies (F value of 5,323). ANOVA also points to secondary but highly significant effects of kappa and t.
The medians shown here are based on 500 sequence pairs for each of the 72 simulated datasets. The expected Dr/Dc is one (marked with hashed line) because the codon model used in simulation did not distinguish radical versus conservative substitutions. Four parameters were varied in the simulation: codon frequencies (constrained to match those of Buchnera APS, Blochmannia floridanus, or E. coli codon tables), the dN/dS ratio, the distance between sequences (t), and the transition/transversion ratio (kappa). Dr/Dc estimates were substantially suppressed in datasets simulated under endosymbiont codon frequencies, compared to datasets simulated under E. coli codon frequencies.
The chart represents a simplified version of Figure 5, pooling across other simulation parameters (kappa, t, and dN/dS) to illustrate the significant effect of codon frequencies. (a) Mean values (with bars indicating standard errors) and (b) median values of Dr/Dc are suppressed in datasets simulated under endosymbiont codon frequencies. Notably, these estimation biases occur in the opposite direction as the patterns observed in empirical genome comparisons, thereby making the inference elevated Dr/Dc in endosymbionts more conservative.
Under most circumstances, the high sensitivity of Dr/Dc estimates to codon frequencies would caution against comparing Dr/Dc across sequences with compositional differences. However, in this particular study, the trend observed in simulations is just the opposite of patterns observed in genome comparisons. That is, sequences simulated under AT-rich endosymbiont codon frequencies show a significant deflation of Dr/Dc estimates, whereas empirical datasets point to elevated Dr/Dc in Blochmannia and Buchnera. In this sense, estimation biases make the conclusion of higher Dr/Dc in endosymbionts more conservative.
Numerous empirical studies support the prediction of the nearly neutral model: the efficiency with which selection can remove harmful variants depends on Ne. Across vertebrates, invertebrate, plants, fungi, and bacteria, a reduction in Ne is often coupled with reduced selective constraint and accelerated accumulation of deleterious changes. These inferences rely on measures of selective constraint, most commonly Ka/Ks or related indices. However, Ka/Ks is unreliable across deep sequence divergences where synonymous substitution rates are saturated.
Here, I explored the utility of Dr/Dc, an alternative index for selective constraint that is based on physiochemical effects of amino acid changes. Dr/Dc offers a tool to compare selective constraint across species that are highly divergent and therefore not amenable to accurate estimates of synonymous divergences. Dr/Dc may be especially useful for genome sequence comparisons, since genomes are often selected to span broad phylogenetic and metabolic diversity rather than to facilitate comparisons among closely-related taxa. Although Dr and Dc (and their ratio) are prone to estimation biases, earlier work shows that Dr/Dc behaves as expected for an index of selective constraint. For example, measures of Dr/Dc show a positive association with Ka/Ks , , suggesting that proteins under relaxed selective constraint accumulate more nonsynonymous changes and, of these changes, a greater proportion are radical. Earlier work also shows that an increase in radical substitutions corresponds to other signatures of reduced selective constraint, such as a higher rate of indels and increased proportion of variable sites . Moreover, several studies have documented higher rates of radical substitutions in primates than in rodents, suggesting that purifying selection against radical changes is less effective in relatively small primate populations , , , .
The present study shows that Dr/Dc behaves similarly in bacterial species and therefore offers a promising gauge for selective constraint. First, within genome pairs, Dr/Dc shows a modest but significant, positive association with rates of protein evolution (measured here as dN) in most genome pairs considered here. Although exceptions exist, the overall trend suggests that proteins under relaxed functional constraint experience a greater proportion of radical substitutions.
Second, Dr/Dc is higher in the obligate endosymbionts Blochmannia and Buchnera, compared to orthologous genes in related free-living bacteria. This elevation of Dr/Dc was highly significant by Wilcoxon and sign tests. Data simulations highlight biases in the estimation of Dr/Dc; however, these biases do not explain the primary result of elevated Dr/Dc in endosymbionts. That is, simulated datasets constrained to match the AT-rich, endosymbiont codon frequencies show lower, rather than higher, Dr/Dc values. In light of the simulation results, it is likely that Dr/Dc in endosymbionts is even higher than estimated here.
The observed elevation of Dr/Dc in the two endosymbiont groups is consistent with other genomic features pointing to the fixation of slightly deleterious mutations in these lineages. Such features include: rapid rates of DNA sequence and protein evolution, a concentration of this rate increase at nonsynonymous sites, and reduced structural stability of proteins and of ribosomal RNA , , . Furthermore, population genetic studies point to reduced intraspecific variation, an excess of nonsynonymous polymorphisms, and non-neutral distributions of allele frequencies all consistent with deleterious evolution in small populations , . These results from population-level studies are not predicted nor explained by the alternative hypothesis of relaxed purifying selection; the results are, however, expected consequences of genetic drift in small populations.
Combined with these other genome features, increased Dr/Dc in Buchnera and Blochmannia further supports the hypothesis that obligate endosymbionts are vulnerable to stochastic effects of drift due to small Ne. Mutation pressure alone cannot account for the observed elevation in the relative frequency of radical amino acid substitutions, as simulation results indicate that strong AT compositional bias, in and of itself, does not elevate Dr/Dc. While relaxed purifying selection in endosymbionts may contribute to higher Dr/Dc, shifts in selection coefficients are typically expected to affect specific genes or functional groups and are unlikely to fully explain a pervasive elevation in Dr/Dc across multiple functional categories. Although alternative processes may play contributing roles as discussed in more depth below, I view elevated Dr/Dc as best explained by a genome-level process such as genetic drift in small endosymbiont populations.
Alternative explanations for elevated Dr/Dc
Additional processes may contribute to the distinct genomic features of endosymbionts that are often attributed to genetic drift (e.g., rapid evolutionary rates, elevated Ka/Ks, and elevated Dr/Dc shown here). First, changes in population size often correspond to shifts in ecological niche, and consequently, changes in selection coefficients. Any study testing the effects of Ne differences in natural populations, including studies of mammals, insects and bacteria, faces the limitation of potential coincident shifts in selective pressures. This is certainly true in comparisons of endosymbiotic vs. free-living bacteria. Endosymbiotic bacteria may experience relaxed purifying selection, perhaps due to the relatively stable intracellular environment they occupy. Relaxed selection on gene function played a role in early gene loss and metabolic streamlining of small endosymbiont genomes . Among genes that are retained, relaxed purifying selection on particular functional categories may contribute to the observed elevation in evolutionary rates, Ka/Ks, and Dr/Dc.
This said, relaxed purifying selection is unsatisfying as a sole explanation for the pervasive, genome-wide patterns in endosymbionts. Importantly, relaxed selection is expected to affect specific genes or functional groups, rather than have a genome-wide effect across multiple functional categories. Following this logic, previous studies have argued that the pervasive rate acceleration and elevated Ka/Ks are unlikely due to relaxed selection alone, but rather reflect the genome-wide process of genetic drift in small populations –. Similarly, genome-wide mechanisms such as genetic drift best explain the consistent elevation in Dr/Dc observed here. The 256 orthologs analyzed here were selected because they are shared across the ten genomes considered. As expected for persistent genes, encoded functions include core processes such as information transfer (transcription, translation and replication) and cell processes (which includes cell division and cell cycle physiology), along with a range of other functional categories. One might imagine relaxed selection on particular metabolic functions with limited importance in host cellular environment. For example, relaxed selection may contribute to the more severe elevation in Dr/Dc among the small number (fourteen) of unclassified orthologs included (Figure 4). However, it is more difficult to imagine that an endosymbiotic lifestyle reduces selection coefficients across core cellular processes.
As a second alternative explanation of elevated Dr/Dc in endosymbionts, constitutive overexpression of the chaperonin GroEL , might offer compensatory mechanism to cope with deleterious substitutions and therefore facilitate the accumulation of such changes in endosymbiont genomes. That is, GroEL and other molecular chaperones may act as buffers (i.e., capacitors) for evolution by allowing substrate proteins to accumulate mutations but still fold correctly , , . A recent study indicates an important role of chaperonins on the evolution of interacting proteins , although other work found no evidence that GroEL acts as a capacitor for evolutionary change . While intriguing, the potential role of GroEL as a capacitor in endosymbiont evolution remains speculative .
Third, it is possible that linkage to positively selected sites may contribute to elevated Dr/Dc and other genome features of endosymbionts. Because Buchnera and Blochmannia are strictly asexual, the entire genome is linked. In this situation, linkage to selected sites may have profound and pervasive consequences. For instance, the elimination of strongly deleterious mutations may reduce variation elsewhere in the genome and contribute to the accumulation of slightly deleterious mutations. In addition, growing evidence suggests that positive selection influences many endosymbiont proteins , . Due to linkage, positive selection anywhere in the genome would purge variation elsewhere and any slightly deleterious changes would become fixed along with the beneficial mutation. This effect of genetic draft  is potentially significant in endosymbionts but largely unexplored, perhaps because this force is difficult to distinguish from genetic drift in small populations.
In sum, this study shows that Dr/Dc offers a promising gauge of selective constraint in bacteria. Specifically, the ratio shows the predicted elevation in species that are prone to the accumulation of deleterious changes. Previous work has shown that endosymbionts show a pervasive, genome-wide rate acceleration and elevated Ka/Ks. Likewise, we found that Dr/Dc is consistently elevated in endosymbionts and affects numerous functional categories, including fundamental cellular processes. Relaxed purifying selection on particular functions may contribute to the observed elevation in Dr/Dc, but relaxed selection alone is not satisfying as a sole explanation for this pervasive phenomenon. Rather, the results are more consistent with genome-wide processes such as genetic drift in small populations.
The taxonomic scope of the present study was limited by the to need to simulate datasets that mimic the codon frequencies as the genomes compared here. These simulations eliminate the possibility that estimation biases explain the observed elevation of Dr/Dc in endosymbionts. However, the observation of biases highlights important challenges for the broader use of Dr/Dc across genomes, particularly those that vary in sequence composition. In future work, the development of new methods to estimate Dr/Dc more accurately will broaden the utility of this index, in the same way that gradual improvements to Ka and Ks estimates have made Ka/Ks a broadly useful metric.
Chromosomal sequences and annotations were downloaded from NCBI Genbank on February 1, 2010–January 1, 2011 for the ten genomes listed in Table 1. The Reciprocal Sequence Distance (RSD) algorithm  was used to identify the reciprocal best BLAST hits (rbh) between translated ORFs. This program uses blastp to identify potential matches of a given translated gene, aligns all potential matches using ClustalW , and calculates a maximum likelihood estimation of amino acid substitutions between proteins by invoking PAML 4.2 . Protein divergences were based on an empirical amino acid substitution rate matrix  and accounted for variation in evolutionary rates among protein sites using a gamma distribution with shape parameter alpha = 1.53 (as recommended by Smith ). The protein with the lowest divergence was then compared to the first genome using blastp, followed by the alignment and divergence calculations. If the protein match with the lowest divergence was the same as the original query sequence, the pair was considered orthologous and the divergence was retained in the output. I used stringent criteria in RSD of retaining only blastp matches of E<10−10 (with the RSD parameter thresh = 1e-10), and requiring that the alignable region of two sequences exceed 80% of the alignment's total length (RSD parameter div = 0.8). These stringent criteria may generate false negatives, by excluding true orthologs that are highly divergent. Such comparisons were performed within each of the pairs listed in Table 2. All genomes were then compared to E. coli using RSD as outlined above, in order to identify orthologs shared across the ten genomes.
Molecular evolutionary analysis.
Within each genome pair, shared orthologs were aligned using transalign.pl , a script that invokes ClustalW  to perform a amino-acid based alignment and back-translates to the nucleotide sequence. Each pair of aligned nucleotide sequences was analyzed in the codeml package of PAML 4.2  using pairwise analysis (runmode = −2) to obtain ML estimations of dS and dN and their standard errors, the transition/transversion ratio (kappa), and the total distance (t) between sequences. Estimates of dS and dN accounted for the distinct base compositions and codon structures of the genomes considered. Specifically, the equilibrium frequencies of codons were calculated from the nucleotide frequencies at the three codon positions (CodonFreq = 2). CodonW (http://codonw.sourceforge.net/) was used to quantify the %GC content, aromaticity, and GRAVY for each gene.
The per-site rate of radical and conservative substitutions (Dr and Dc) and their standard errors were estimated with the Hon-New program, which applies the method of Zhang  to account for the transition/transversion bias. Dr and Dc was calculated based on four classification schemes: charge, polarity, the amino acid classification by Miyata et al.  that considers polarity and volume, and the classification scheme of Hanada et al. . Amino acid groups under each classification scheme are shown in Table S4. For each gene pair, information about kappa (as estimated above in codeml) was considered in the Dr and Dc estimations. Dr and Dc values were adjusted with a Jukes Cantor correction , using the formula Dcorr = −0.75*LN(1−(4/3*Dorig)), where Dorig is the uncorrected divergence value and Dcorr is the corrected value.
Under the circumstances listed below, I considered Dr/Dc estimates as unreliable and excluded them under the “high s.e. filter.” First, this filter excludes particularly conserved genes that had zero radical and/or conservative amino acid changes, or so few amino acid substitutions that the percent standard error of Dr/Dc exceeded 50% of the Dr/Dc ratio itself. In addition, the Jukes Cantor correction cannot be applied if Dorig>0.75 since the natural log of zero or a negative number is undefined. The very few genes with such high divergence were also excluded as part of the “high s.e.” filter.
The ortholog detection methods detailed above identified 276 orthologs shared across the ten genomes considered. Twenty of those orthologs were excluded because they did not pass the “high s.e. filter” in one or more of the genome pairs i–iv (Table 2). Subsequent comparisons among bacterial genome pairs focused primarily upon the remaining 256 shared orthologs that survived the filter in each of the five pairs. The recently-diverged E. coli vs. S. typhimurium pair had numerous genes with very low Dr and/or Dc values for which the standard error of Dr/Dc exceeded 50%. This pair was filtered less stringently in order to retain a comparable number of genes (see text and Table 5 legend).
Assignment of functional categories.
The 256 shared orthologs were assigned broad functional categories based on the MultiFun assignment of E. coli orthologs , , downloaded from http://genprotec.mbl.edu/. Multifun provides detailed classification system for the physiological and cellular roles of gene products. For the purpose of this broad analysis, I considered only the highest level category, represented by the first number in the Multifun assignment, and I did not consider cellular location (represented under Multifun category 7). Broad functional classes considered here included: Cell processes, Cell structure, Information transfer, Metabolism, Regulation, Transport, and Unclassified (no Multifun assignment other than Location in some instances). Several E. coli genes are assigned to more than one category. The specific functions within each broad category are listed on the Multifun website, at http://genprotec.mbl.edu/files/MultiFun.txt.
I used simulations to explore the sensitivity of Dr/Dc estimates to differences in sequence composition and patterns of nucleotide change. Following the general approach of Smith  and Dagan , I simulated sequence datasets that varied in their equilibrium codon composition frequencies and various mutational parameters. These datasets were generated with the evolver program of PAML using a codon model of substitution. Since the simulations do not distinguish between radical versus conservative amino acid substitutions, the expected Dr/Dc for these datasets is one, and deviations from one suggest a bias in the estimate.
In more detail, the following four factors were varied in the simulation. (i) Datasets were generated under three equilibrium codon frequencies. Three codon tables were developed from codon frequencies of all coding regions of the Buchnera APS, Blochmannia floridanus, or E. coli genomes. Raw codon usage number for all loci were calculated in codonW, and the number of each codon was converted to its relative frequency among the 61 possible codons. (Stop codons are excluded in this analysis.) This representation of codon usage accounts for differences in %GC, amino acid frequencies, and relative synonymous codon usage. (ii) The dN/dS ratio was set to 0.3 or 0.6, the values also selected by Smith . The lower value (0.3) falls within the range of empirical Ka/Ks estimates for endosymbionts genes . (iii) The distance between sequences, t, measured as mean substitutions per site, was set to 0.5, 1, or 4, and (iv) the transition/transversion ratio (kappa) was set to 1, 2, 5, or the exceptionally high value of 10. For both t and kappa, the lower parameter values span the medians estimated from the empirical datasets used (Tables 3 and 5). The higher parameter values allowed us to test the effect of more extreme values on simulation results.
The combinations of all parameter settings resulted in 72 total datasets (3×2×3×4). Each simulated dataset contained 500 pairs of sequences that were 1,200 nucleotides long. In total, 36,000 sequence pairs were generated. Each sequence pair was analyzed using codeml and Hon-New, as described above for empirical genome comparisons. The sensitivity of Dr/Dc-MY estimates to the four simulation parameters was tested with a multi-way analysis of variance (ANOVA).
Elevated Dr/Dc in endosymbionts is consistent across functional categories (individual data points). Data points underlie the median and mean values presented in Figure 4. The y-axis (on log scale) shows the ratio of Dr/Dc values for endosymbiont pairs and free-living bacterial pairs, for individual genes. This value typically exceeds one, indicating higher Dr/Dc in the endosymbionts across diverse functional categories that include core cellular processes. Data include the 256 shared orthologs shared among the genomes considered. Functional categories are based MultiFun classification of the E. coli ortholog , . When a given gene is assigned to more than one broad category, it is represented more than once in the data points shown here. The numbers of orthologs within each functional category are listed in the legend of Figure 4.
Nonparametric tests of association between Dr/Dc and various sequence features.
Gene IDs for the orthologs shared among the genomes considered.
Comparison of Dr/Dc between endosymbiotic vs. free-living bacterial species, across seven functional categories.
I thank Rich Fox (MBL, Woods Hole) and Peyton Vaughn (Duke University) for help with computational tools used here. I am grateful to David Rand for discussions that helped to frame this study, and thank two anonymous reviewers for thoughtful comments.
Conceived and designed the experiments: JJW. Performed the experiments: JJW. Analyzed the data: JJW. Contributed reagents/materials/analysis tools: JJW. Wrote the paper: JJW.
- 1. Ohta T, Kimura M (1971) On the constancy of the evolutionary rate of cistrons. J Mol Evol 1: 18–25.
- 2. Ohta T (1973) Slightly deleterious mutant substitutions in evolution. Nature 246: 96–98.
- 3. Li WH (1997) Molecular Evolution. Sunderland, Massachusetts: Sinauer Associates, Inc. 487 p.
- 4. Hughes AL, Friedman R (2009) More radical amino acid replacements in primates than in rodents: Support for the evolutionary role of effective population size. Gene 440: 50–56.
- 5. Lynch M, Blanchard J, Houle D, Kibota T, Schultz S, et al. (1999) Perspective: Spontaneous deleterious mutation. Evolution 53: 645–663.
- 6. Lynch M, Burger R, Butcher D, Gabriel W (1993) The mutational meltdown in asexual populations. J Hered 84: 339–344.
- 7. Barton NH, Charlesworth B (1998) Why sex and recombination? Science 281: 1986–1990.
- 8. Muller J (1964) The relation of recombination to mutational advance. Mutat Res 1: 2–9.
- 9. Charlesworth B (1990) Mutation-selection balance and the evolutionary advantage of sex and recombination. Genet Res 55: 199–221.
- 10. Xu J (2004) The prevalence and evolution of sex in microorganisms. Genome 47: 775–780.
- 11. Lynch M (1995) Mutation accumulation and the extinction of small populations. Amer Nat 146: 489–518.
- 12. Eyre-Walker A, Keightley PD (1999) High genomic deleterious mutation rates in hominids. Nature 397: 344–347.
- 13. Li WH, Tanimura M, Sharp PM (1987) An evaluation of the molecular clock hypothesis using mammalian DNA sequences. J Mol Evol 25: 330–342.
- 14. Ohta T (1995) Synonymous and nonsynonymous substitutions in mammalian genes and the nearly neutral theory. J Mol Evol 40: 56–63.
- 15. Keightley PD, Lercher MJ, Eyre-Walker A (2005) Evidence for widespread degradation of gene control regions in hominid genomes. PloS Biology 3: 282–288.
- 16. Eyre-Walker A, Keightley PD, Smith NG, Gaffney D (2002) Quantifying the slightly deleterious mutation model of molecular evolution. Mol Biol Evol 19: 2142–2149.
- 17. Moran NA (1996) Accelerated evolution and Muller's rachet in endosymbiotic bacteria. Proc Natl Acad Sci U S A 93: 2873–2878.
- 18. Woolfit M, Bromham L (2003) Increased rates of sequence evolution in endosymbiotic bacteria and fungi with small effective population sizes. Mol Biol Evol 20: 1545–1555.
- 19. Keightley PD, Eyre-Walker A (2000) Deleterious mutations and the evolution of sex. Science 290: 331–333.
- 20. Ohta T (1972) Evolutionary rate of cistrons and DNA divergence. J Mol Evol 1: 150–157.
- 21. Goto H, Inomata N, Szmidt AE, Yamazaki T (2004) Nucleotide variation of seven genes in Drosophila kikkawai. Genes Genet Syst 79: 19–26.
- 22. Woolfit M, Bromham L (2005) Population size and molecular evolution on islands. Proc Biol Sci 272: 2277–2282.
- 23. Ohta T (1976) Role of very slightly deleterious mutations in molecular evolution and polymorphism. Theor Popul Biol 10: 254–275.
- 24. Johnson KP, Seger J (2001) Elevated rates of nonsynonymous substitution in island birds. Mol Biol Evol 18: 874–881.
- 25. Nachman MW, Boyer SN, Aquadro CF (1994) Nonneutral evolution at the mitochondrial NADH dehydrogenase subunit 3 gene in mice. Proc Natl Acad Sci U S A 91: 6364–6368.
- 26. Rand DM, Kann LM (1996) Excess amino acid polymorphism in mitochondrial DNA: contrasts among genes from Drosophila, mice, and humans. Mol Biol Evol 13: 735–748.
- 27. Ballard JW, Whitlock MC (2004) The incomplete natural history of mitochondria. Mol Ecol 13: 729–744.
- 28. Charlesworth B, Morgan MT, Charlesworth D (1993) The effect of deleterious mutations on neutral molecular variation. Genetics 134: 1289–1303.
- 29. Charlesworth B (1994) The effect of background selection against deleterious mutations on weakly selected, linked variants. Genet Res 63: 213–227.
- 30. Maynard Smith J, Haigh J (1974) The hitch-hiking effect of a favorable gene. Genet Res 23: 23–35.
- 31. Gillespie JH (2001) Is the population size of a species relevant to its evolution? Evolution 55: 2161–2169.
- 32. Gillespie JH (2000) The neutral theory in an infinite population. Gene 261: 11–18.
- 33. Gillespie JH (1999) The role of population size in molecular evolution. Theor Popul Biol 55: 145–156.
- 34. Hudson RR, Turelli M (2003) Stochasticity overrules the “three-times rule”: Genetic drift, genetic draft, and coalescence times for nuclear loci versus mitochondrial DNA. Evolution 57: 182–190.
- 35. Gerrard DT, Filatov DA (2005) Positive and negative selection on mammalian Y chromosomes. Mol Biol Evol 22: 1423–1432.
- 36. Lu J, Tang T, Tang H, Huang J, Shi S, et al. (2006) The accumulation of deleterious mutations in rice genomes: a hypothesis on the cost of domestication. Trends Genet 22(3): 126–31.
- 37. Warnecke T, Rocha EP (2011) Function-specific accelerations in rates of sequence evolution suggest predictable epistatic responses to reduced effective population size. Mol Biol Evol 28: 2339–49.
- 38. Moran NA, Plague GR (2004) Genomic changes following host restriction in bacteria. Curr Opin Genet Dev 14: 627–633.
- 39. Moran NA, Wernegreen JJ (2000) Lifestyle evolution in symbiotic bacteria: insights from genomics. Trends Ecol Evol 15: 321–326.
- 40. Andersson SG, Kurland CG (1998) Reductive evolution of resident genomes. Trends Microbiol 6: 263–268.
- 41. Abbot P, Moran NA (2002) Extremely low levels of genetic polymorphism in endosymbionts (Buchnera) of aphids (Pemphigus). Mol Ecol 11: 2649–2660.
- 42. Funk DJ, Wernegreen JJ, Moran NA (2001) Intraspecific variation in symbiont genomes: bottlenecks and the aphid-buchnera association. Genetics 157: 477–489.
- 43. Buchner P (1965) Endosymbiosis of animals with plant microorganisms. New York: Interscience Publishers, Inc.
- 44. Moran NA, McCutcheon JP, Nakabachi A (2008) Genomics and evolution of heritable bacterial symbionts. Annu Rev Genet 42: 165–190.
- 45. Kuo CH, Moran NA, Ochman H (2009) The consequences of genetic drift for bacterial genome complexity. Genome Res 19(8): 1450–4.
- 46. Wernegreen JJ, Moran NA (1999) Evidence for genetic drift in endosymbionts (Buchnera): Analyses of protein-coding genes. Mol Biol Evol 16: 83–97.
- 47. Clark MA, Moran NA, Baumann P (1999) Sequence evolution in bacterial endosymbionts having extreme base compositions. Mol Biol Evol 16: 1586–1598.
- 48. Smith NGC (2003) Are radical and conservative substitution rates useful statistics in molecular evolution? J Mol Evol 57: 467–478.
- 49. Dagan T, Talmor Y, Graur D (2002) Ratios of radical to conservative amino acid replacement are affected by mutational and compositional factors and may not be indicative of positive Darwinian selection. Mol Biol Evol 19: 1022–1025.
- 50. Rand DM, Weinreich DM, Cezairliyan BO (2000) Neutrality tests of conservative-radical amino acid changes in nuclear- and mitochondrially-encoded proteins. Gene 261: 115–125.
- 51. Pupko T, Sharan R, Hasegawa M, Shamir R, Graur D (2003) Detecting excess radical replacements in phylogenetic trees. Gene 319: 127–135.
- 52. Zhang J (2000) Rates of conservative and radical nonsynonymous nucleotide substitutions in mammalian nuclear genes. J Mol Evol 50: 56–68.
- 53. Williams KP, Gillespie JJ, Sobral BW, Nordberg EK, Snyder EE, et al. (2010) Phylogeny of gammaproteobacteria. J Bacteriol 192: 2305–2314.
- 54. Wall DP, Fraser HB, Hirsh AE (2003) Detecting putative orthologs. Bioinformatics 19: 1710–1711.
- 55. Yang Z (1997) PAML: a program package for phylogenetic analysis by maximum likelihood. Comput Appl Biosci 13: 555–556.
- 56. Miyata T, Miyazawa S, Yasunaga T (1979) Two types of amino acid substitutions in protein evolution. J Mol Evol 12: 219–236.
- 57. Hanada K, Shiu SH, Li WH (2007) The nonsynonymous/synonymous substitution rate ratio versus the radical/conservative replacement rate ratio in the evolution of mammalian genes. Mol Biol Evol 24: 2235–2241.
- 58. Lockhart P, Novis P, Milligan BG, Riden J, Rambaut A, et al. (2006) Heterotachy and tree building: A case study with plastids and eubacteria. Mol Biol Evol 23: 40–45.
- 59. Bastolla U, Moya A, Viguera E, van Ham RC (2004) Genomic determinants of protein folding thermodynamics in prokaryotic organisms. J Mol Biol 343: 1451–1466.
- 60. Moran NA (2003) Tracing the evolution of gene loss in obligate bacterial symbionts. Curr Opin Microbiol 6: 512–518.
- 61. Ishikawa H (1984) Alteration with age of symbiosis of gene expression in aphid endosymbionts. Biosystems 17: 127–134.
- 62. Fares MA, Ruiz-Gonzalez MX, Moya A, Elena SF, Barrio E (2002) Endosymbiotic bacteria: groEL buffers against deleterious mutations. Nature 417: 398.
- 63. Fares MA, Moya A, Barrio E (2004) GroEL and the maintenance of bacterial endosymbiosis. Trends Genet 20: 413–416.
- 64. Williams TA, Fares MA (2010) The effect of chaperonin buffering on protein evolution. Genome Biol Evol 2: 609–619.
- 65. Raineri E, Ribeca P, Serrano L, Maier T (2010) A more precise characterization of chaperonin substrates. Bioinformatics 26: 1685–1689.
- 66. Fares MA, Barrio E, Sabater-Munoz B, Moya A (2002) The evolution of the heat-shock protein GroEL from Buchnera, the primary endosymbiont of aphids, is governed by positive selection. Mol Biol Evol 19: 1162–1170.
- 67. Toft C, Fares MA (2010) Structural calibration of the rates of amino acid evolution: In a search for Darwin in drifting biological systems. Mol Biol Evol (10): 2375–85.
- 68. Chenna R, Sugawara H, Koike T, Lopez R, Gibson TJ, et al. (2003) Multiple sequence alignment with the Clustal series of programs. Nucleic Acids Res 31: 3497–3500.
- 69. Jones DT, Taylor WR, Thornton JM (1992) The rapid generation of mutation data matrices from protein sequences. Comput Appl Biosci 8: 275–282.
- 70. Bininda-Emonds OR (2005) TransAlign: using amino acids to facilitate the multiple alignment of protein-coding DNA sequences. BMC Bioinformatics 6: 156.
- 71. Jukes T, Cantor C (1969) Evolution of protein molecules. In: Munro H, editor. Mammalian Protein Metabolism. New York: Academic Press. pp. 21–132.
- 72. Serres MH, Goswami S, Riley M (2004) GenProtEC: An updated and improved analysis of functions of Escherichia coli K-12 proteins. Nucleic Acids Res 32 Database issue: D300–302.
- 73. Serres MH, Riley M (2000) MultiFun, a multifunctional classification scheme for Escherichia coli K-12 gene products. Microb Comp Genomics 5: 205–222.