Different sites in a protein evolve at different rates (Kimura & Ohta, 1974; Perutz, Kendrew & Watson, 1965), and these rate differences are shaped by the interplay of functional and structural constraints each site experiences (Echave, Spielman & Wilke, 2016). For example, protein surface sites tend to evolve faster than interior sites of a protein (Franzosa & Xia, 2009; Shahmoradi et al., 2014; Yeh et al., 2014b; Yeh et al., 2014a; Huang et al., 2014; Ramsey et al., 2011; Dean et al., 2002; Scherrer, Meyer & Wilke, 2012; Mirny & Shakhnovich, 1999; Zhou, Drummond & Wilke, 2008). Active sites in enzymes tend to be highly conserved (Jack et al., 2016; Dean et al., 2002; Kimura & Ohta, 1973; Huang et al., 2015), and sites involved in protein–protein interactions are somewhat more conserved than other surface sites (Mintseris & Weng, 2005; Kim et al., 2006; Franzosa & Xia, 2009; Jack et al., 2016).
Analyses of sequence variation in a structural context frequently make use of site-specific evolutionary rate estimates, and a wide variety of different methods exist to infer such rates from either codon or amino-acid sequences (Nielsen & Yang, 1998; Yang & Nielsen, 2002; Kosakovsky Pond, Frost & Muse, 2005; Kosakovsky Pond & Muse, 2005; Yang et al., 2000; Murrell et al., 2012; Lemey et al., 2012; Pupko et al., 2002; Fernandes & Atchley, 2008; Huang & Golding, 2014; Huang & Golding, 2015; Mayrose et al., 2004). The most widely applied methods for codon sequences are based on dN∕dS, the rate of non-synonymous substitutions per non-synonymous site dN divided by the rate of synonymous substitutions per synonymous site dS. The dN∕dS ratio is commonly used to infer purifying (dN∕dS < 1) or positive selection (dN∕dS > 1) in protein-coding genes (Nielsen & Yang, 1998; Goldman & Yang, 1994). The most popular method for rate inference in amino-acid sequences is Rate4Site (Pupko et al., 2002; Mayrose et al., 2004). Rate4Site assigns a score to a site as a proxy for the rate of evolution at that site. Rate4Site is typically used to locate active sites on the protein structure, such as protein–protein interaction or protein–ligand interaction sites and catalytic sites (Mousson et al., 2005; Fischer, Mayer & Söding, 2008; Tuncbag, Gursoy & Keskin, 2009; Bradford et al., 2006; Guney et al., 2008). How dN∕dS inference methods relate to Rate4Site scores is not known.
The relationship between protein structure and evolutionary variation has been investigated in many different protein families and many different datasets of varying divergence levels and taxonomic origin, and some studies have used codon sequences and codon-related methods to infer the rate of evolution (Franzosa & Xia, 2009; Shahmoradi et al., 2014; Scherrer, Meyer & Wilke, 2012; Zhou, Drummond & Wilke, 2008; Kim et al., 2006) while others have used amino-acid sequences (Ramsey et al., 2011; Yeh et al., 2014b; Yeh et al., 2014a; Huang et al., 2014; Huang et al., 2015; Jack et al., 2016; Mirny & Shakhnovich, 1999). Because of these differences in datasets and analysis approaches, it is not obvious to what extent results from different studies can be compared. To the extent that different studies produce contradictory results, and they frequently do (Jackson et al., 2016; Echave, Spielman & Wilke, 2016), are these contradictions due to fundamental differences in the analyzed datasets (e.g., highly diverged sequences from many taxonomic groups vs. weakly diverged sequences from a single population) or in the employed methods to infer evolutionary rates (e.g., inference based on amino-acid sequences vs. on codon sequences)?
Here we address the second question, to what extent analyses at the codon level are comparable to analyses at the amino-acid level. Specifically, we use extensive simulations to ask how similar the site-specific Rate4Site scores are to site-specific dN∕dS values. We simulate sequence divergence both under dN∕dS models and under mutation–selection models, and we then ask how inferred Rate4Site scores for these simulated alignments compare to (i) the true simulated dN∕dS values at each site and (ii) the inferred dN∕dS values obtained from the simulated alignments. We find that Rate4Site scores generally correlate well with dN∕dS, in particular if both quantities are inferred from sequence data. We verify this observation on rates inferred from empirical datasets, and we conclude that amino-acid level and codon-level analysis of rate variation will generally yield comparable results.
Generation of simulated alignments
Our simulation approach was similar to the one employed by Spielman, Wan & Wilke (2016). In brief, we first generated a set of balanced, binary trees with different branch lengths and numbers of taxa, using the R package ape (Paradis, Claude & Strimmer, 2004). We then simulated sequence evolution along these trees using the python library pyvolve (Spielman & Wilke, 2015a).
We generated a total of 40 trees, using all pairwise combinations of five different branch lengths and eight different numbers of taxa. The branch lengths we used were 0.0025, 0.01, 0.04, 0.16, and 0.64. These numbers indicate the divergence in mutations per site between two nodes in a tree. The numbers of taxa we used were 16, 32, 64, 128, 256, 512, 1,024, and 2,048.
To generate alignments with site-specific dN∕dS values, we simulated sequences with 100 codon sites using a site-specific Muse–Gaut model (Muse & Gaut, 1994). To simulate sequences with constant dS, we set dS = 1 at all sites and set dN at each site to a different value randomly drawn from a uniform distribution between 0.1 and 1.6. To simulate sequences with variable dS, we assigned each site a distinct dN and dS value, by first choosing a randomly drawn dN∕dS value, then choosing a randomly drawn dS value, and then setting dN = dS × (dN∕dS). The dN∕dS values were drawn from a uniform distribution between 0.1 and 1.6, and the dS values were drawn from a uniform distribution between 0.5 and 2. We generated 50 replicate sequence alignments for each combination of branch length, number of taxa (128–2,048), and choice of dS (constant or variable), for a total of 2,500 sequence alignments.
We also generated alignments with gamma-distributed site-specific dN∕dS values. We simulated sequences with 100 codon sites using a site-specific Muse–Gaut model. We set dN∕dS to a randomly drawn value from a gamma distribution. We ran simulations for six distinct gamma distributions, using shape parameters α and rate parameters β previously estimated for six HIV-1 proteins (Table 2 in Meyer & Wilke, 2015b). For each protein (i.e., distinct gamma distribution), we generated 50 replicate sequence alignments for each combination of branch length and number of taxa (16–256), for a total of 1,500 sequence alignments per protein.
For sequences simulated according to MutSel models, we used sequence alignments previously published in Spielman, Wan & Wilke (2016), specifically the alignments simulated with unequal nucleotide frequencies. These sequences were simulated using the Halpern and Bruno model (HB98) (Halpern & Bruno, 1998), and we had alignments for the same tree parameters, dS variation (constant/variable), and replicate numbers as our simulations of the dN∕dS model, for a total of 2,500 sequence alignments.
To acquire the Rate4Site scores, the simulated sequences were translated into amino acids using biopython. The translated sequences were inputted into Rate4Site along with their corresponding trees. We ran Rate4Site with the following options:
rate4site -s aln_file -t tree_file -o norm_rates_file \ -y orig_rates_file
Here, aln_file is the input fasta file with aligned sequences. The file tree_file contains the phylogenetic tree. The file norm_rates_file is the output file into which Rate4Site writes z-normalized rate scores, and orig_rates_file is the output file into which Rate4Site writes original rate scores. The option -y causes Rate4Site to output original scores. (By default, Rate4Site only outputs z-transformed scores.) In our analysis we used only the original scores, renormalized such that they had a mean of 1.
We inferred site-specific dN∕dS using the one-parameter fixed-effects likelihood method (FEL1) implemented in HyPhy (Kosakovsky Pond, Frost & Muse, 2005). We ran HyPhy using the FEL1 script provided in Spielman, Wan & Wilke (2016). After running the dN∕dS inference, we explicitly set dN∕dS = 0 at all sites that did not experience any amino-acid changes. We did so because the FEL1 method assigns a site-wise dN∕dS of 1 to completely conserved sites that contain no synonymous and no non-synoymous mutations. However, dN∕dS should equal 0 at such sites in a one-parameter model, which implicitly assumes that dS is the same at all sites and hence will be non-zero even at completely conserved sites.
Analysis of empirical datasets
For analysis of empirical datasets, we used data from Spielman & Wilke (2013) and Meyer & Wilke (2015b). From Spielman & Wilke (2013), we acquired unaligned sequence data for six arbitrarily chosen membrane proteins: Mannose-6-phosphate receptor M6PR (Ensembl transcript ID: ENST00000000412), CD74 (Ensembl transcript ID: ENST00000009530), CD4 (Ensembl transcript ID: ENST00000011653), G protein-coupled receptor class C, GPRC5A (Ensembl transcript ID: ENST00000014914), Gamma-aminobutyric acid type A receptor, GABRA1 (Ensembl transcript ID: ENST00000023897), and TNF receptor superfamily member 17, TNFRSF17 (Ensembl transcript ID: ENST00000053243). We aligned the amino-acid sequences using MAFFT 7.305b (Multiple Alignment using Fast Fourier Transform) (Katoh & Standley, 2013). We ran MAFFT using default options with:
mafft input_fasta_file > output_fasta_file
Here, input_fasta_file is the fasta file containing sequences to be aligned and output_fasta_file is the output file into which the alignment is written. The aligned amino-acid sequences were subsequently back-translated to codon sequences using the unaligned nucleotide sequences.
For sequence data from Meyer & Wilke (2015b), we used the aligned amino and nucleotide sequence files for all of the proteins used in the paper.
We inferred phylogenetic trees from amino-acid sequences using RAxML (Stamatakis, 2014). We ran RAxML with the following options:
raxmlHPC-PTHREADS-SSE3 -T 48 -s fasta_file -w output_directory \ -n tree_name -m PROTCATLG -p 1
Here, fasta_file is the input file containing the aligned sequences. RAxML outputs all output files into the directory indicated by output_directory, and tree_name is the name for the output tree files. The option -m PROTCATLG picks a protein CAT model with LG matrix for the tree inference, and the option -p 1 generates a random number seed for parsimony inference.
Finally, for all empirical datasets, we inferred Rate4Site scores and per-site dN∕dS values as described in the subsection “Rate inference.”
Both Rate4Site scores and per-site dN∕dS values are measures of the extent to which selection acts on individual protein sites. The Rate4Site model decomposes evolutionary distances in amino-acid alignments into a site-specific component rk and a branch-specific component ti, such that the total divergence at site k along branch i can be written as rkti. Here, rk is the Rate4Site score at site k and ti is the branch length of branch i in the phylogenetic tree. Importantly, rk is the same at all branches in the tree and ti is the same at all sites for each branch i. Because the rate decomposition is invariant under a rescaling of and , Rate4Site scores are not unique unless an additional normalization condition is specified as well. The Rate4Site software solves this uniqueness problem by turning the rk into z-scores. However, the more natural normalization is to divide all rk by their mean, , where the sum runs over all sites in the protein. These normalized scores have the simple interpretation of providing the relative increase or decrease in substitution rate at site k compared to the average rate of substitution in the rest of the protein.
In contrast to Rate4Site scores, which are calculated from amino-acid alignments, dN∕dS ratios are calculated on nucleotide alignments. They estimate the rate of non-synonymous divergence relative to the rate of synonymous divergence. However, just like in Rate4Site, in a site-specific dN∕dS model evolutionary divergence is decomposed into a site-specific dN∕dS value and a site-independent branch length. Thus, Rate4Site and site-specific dN∕dS measure fundamentally the same quantity. The main difference is the input data (amino-acid sequences vs. codon sequences) and the normalization (relative to mean across sites vs. relative to the synonymous divergence rate dS).
Relationship between Rate4Site scores and true dN∕dS
To determine the relationship between Rate4Site and dN∕dS models, we began by simulating sequence evolution with known, site-specific dN∕dS values and then comparing these true dN∕dS values to Rate4Site scores inferred from the simulated alignments (Fig. 1). We first considered the case of constant dS among all sites. Thus, for each site in each alignment, we randomly drew a dN from a uniform distribution ranging from 0.1 to 1.6. We set dS = 1 for all sites, such that the dN∕dS ratios similarly varied from 0.1 to 1.6. We ran simulations along a set of 25 balanced trees with different branch lengths and numbers of taxa, as used previously in a study of dN∕dS inference (Spielman, Wan & Wilke, 2016). Simulated sequences were 100 codon sites long, and we generated 50 replicate simulations for each simulation condition.
We calculated the correlation between each site’s true dN∕dS and its inferred Rate4Site score to assess how well the Rate4Site scores agreed with the simulated rates. We then plotted the mean correlation strengths in replicate simulations against the simulations’ branch lengths and number of taxa. We found that correlation strengths systematically increased with both increasing branch lengths and number of taxa (Fig. 2A). While correlations were low to moderate for the least-diverged and smallest alignments, for larger and/or more diverged alignments correlations approached values ranging from 0.8 to 1.0.
We also performed a comparison of the magnitude of Rate4Site scores and dN∕dS scores, by calculating root-mean-square deviations (RMSD) between these scores. Because these two types of scores are not measured in the same units, this comparison may not seem meaningful. However, we can convert both types of scores into normalized, relative scores by dividing them by their mean score. These normalized scores have comparable interpretations and RMSDs between them are meaningful quantities.
We found that RMSD values were generally moderate, between 0.1 and 0.6 (Fig. 2B). They declined with both increasing number of taxa and increasing sequence divergence. However, overall RMSD depended more strongly on branch length than on the number of taxa. Visual inspection of normalized Rate4Site scores plotted against normalized dN∕dS scores revealed no major systematic differences between these scores (Fig. 3). Differences seemed to be driven primarily by the sampling noise inherent in estimating site-specific evolutionary rates.
We repeated the same analysis but now using simulations in which dS was allowed to vary among sites as well. The dN∕dS range was kept the same as before (0.1–1.6), but now each site had its own unique dS, randomly chosen from a uniform distribution ranging from 0.5 to 2. Overall, we found similar patterns in the variable dS case as we had seen for constant dS (compare Figs. 2C, 2D to Figs. 2A, 2B. However, correlations were generally somewhat weaker (Fig. 2C) and RMSD values somewhat higher (Fig. 2D) than what we had observed for constant dS. These results were to be expected, since Rate4Site as an amino-acid based metric does not take synonymous variation into account, and thus the dS variation acts simply as added random noise on the dN∕dS scores compared to Rate4Site scores.
Relationship between Rate4Site scores and scaled selection coefficients
The dN∕dS model is not a particularly realistic model of sequence evolution, because it does not have the notion of an underlying fitness landscape. A mutation increasing fitness should fix much more rapidly than the reverse mutation decreasing fitness. However, in a dN∕dS model, both mutations fix at the same rate. To increase realism in our analysis, we next investigated Rate4Site in the context of sequences simulated with mutation–selection (MutSel) models. MutSel models are specified by scaled selection coefficients, which describe the relative fitness of different amino acids (or codons) at each site in a sequence. We can derive expected dN∕dS values from these scaled selection coefficients (Spielman & Wilke, 2015b; Dos Reis, 2015) and hence we can ask how Rate4Site scores compare to the predicted dN∕dS values in MutSel models.
For this analysis, we employed previously published sequence alignments from Spielman, Wan & Wilke (2016). These alignments had been simulated using the Halpern and Bruno model (HB98) (Halpern & Bruno, 1998) along the same 25 phylogenetic trees we employed in our previous analysis (five branch lengths in all pairwise combinations with five datasets with different numbers of taxa). As before, there were 50 replicates per simulation condition, and we again had one dataset with constant dS and one with variable dS. In the dataset with constant dS, all synonymous codons have the same fitness (neutral synonymous codons). In the dataset with variable dS, there are fitness differences among synonymous codons (non-neutral synonymous codons). See Spielman, Wan & Wilke (2016) for details of parameter choices.
Our results for sequences simulated with MutSel models were broadly similar to our results for sequences simulated with dN∕dS models (Fig. 4). As before, correlations increased and approached 1 with increasing branch lengths and numbers of taxa, and RMSDs commensurately decreased. However, correlation strengths were consistently lower and RMSD values higher for the MutSel datasets than for the dN∕dS datasets at the same sequence divergence and number of taxa (compare Figs. 4 to 2). As before, differences between normalized Rate4Site scores and normalized true dN∕dS values seemed to be driven primarily by the sampling noise inherent in estimating site-specific evolutionary rates (Fig. 5). Finally, we found only minor differences between simulations with neutral synonymous codons and simulations with non-neutral synonymous codons. However, in the latter case, correlations were generally slightly lower and RMSD values somewhat higher (compare Figs. 4C and 4D to 4A and 4B).
Relationship between Rate4Site scores and inferred dN∕dS
The preceding analyses asked to what extent Rate4Site scores reflect the known underlying parameters used to generate the sequence alignments. An alternative question, possibly more applicable to practical sequence analysis, is to what extent Rate4Site scores mirror dN∕dS values inferred on the same sequence data. To address this second question, we inferred site-wise dN∕dS values for all sites in all alignments studied in the previous two subsections. The dN∕dS values were inferred using the one-rate fixed-effects likelihood method (FEL1) implemented in HyPhy (Kosakovsky Pond, Frost & Muse, 2005). The FEL1 method assigns one dN value per site and one dS value across all sites in the sequence (Spielman, Wan & Wilke, 2016). Therefore, the variation in the inferred dN∕dS values is captured entirely by dN.
We found that Rate4Site scores were very highly correlated to inferred dN∕dS across all datasets and simulation conditions (Figs. 6 and 7). For sequences simulated with the dN∕dS model, correlations for all branch lengths exceeded 0.8 (Figs. 6A and 6C). Correlations generally increased and approached 1 for sequence alignments with more divergent sequences or more taxa. RMSD values were large for the smallest and least-diverged alignments but declined rapidly as either branch length or number of taxa increased (Figs. 6B and 6D). There was little difference between alignments simulated with constant dS and with variable dS (Figs. 6A vs. 6C and 6B vs. 6D). Results for sequences simulated with MutSel models were similar (Fig. 7). The main difference was that correlation coefficients were more sensitive to the number of taxa. For the lowest number of taxa (128) correlation coefficients were systematically lower when sequences were simulated with MutSel models rather than with dN∕dS models. The pattern reversed for the highest number of taxa (2,048). RMSDs, however, were systematically higher for sequences simulated with MutSel models rather than with dN∕dS models. In all cases, there was little difference between sequences simulated with neutral synonymous codons and with non-neutral synonymous codons (Figs. 7A vs. 7D and 7B vs. 7D).
Analysis of more realistic parameter settings and of empirical datasets
The above simulations utilized rates that were uniformly distributed across sites. In empirical datasets, however, the majority of sites evolves slowly and only very few sites evolve rapidly, with a resulting distribution that is approximately gamma (see e.g., Meyer & Wilke, 2015b). Therefore, we wanted to assess if the observed relationships between Rate4Site scores and dN∕dS hold for simulations with gamma-distributed true rates. We simulated sequences with gamma-distributed true dN∕dS values, using the shape and rate parameters obtained by Meyer & Wilke (2015b) for six different HIV-1 proteins. For these simulations, we also included a range of smaller numbers of taxa (16, 32, and 64) than before, to increase realism in the simulations.
We found a wide range of correlations between Rate4Site and true dN∕dS (Fig. 8A and Parts A of Figs. S1–S5), but the overall pattern was similar to what we had previously observed for uniformly distributed rates: Correlations increased and approached 1 as either sequence divergence or the number of taxa increased. Also note that the lowest number of taxa now was 16 vs. 128 in Fig. 2, and thus at identical numbers of taxa and branch lengths, the correlations for gamma-distributed rates were generally higher than the correlations for uniformly distributed rates. On the other hand, the RMSD values between Rate4Site scores and true dN∕dS were higher than in our previous analysis (compare Figs. 8B to 2B).
We also inferred dN∕dS from the simulated sequences and compared them to the Rate4Site scores. Unlike in our previous analysis on inferred dN∕dS (Fig. 6), the correlations between Rate4Site and inferred dN∕dS now spanned a wide range of values (Figs. 8C and Parts C of Figs. S1–S5) but were generally higher than the corresponding correlations between Rate4Site and true dN∕dS. On the flip side, RMSD values were larger when calculated for inferred dN∕dS than when calculated for true dN∕dS, and there was a notable uptick in RMSD at the highest branch lengths (Figs. 8D and Parts D of Figs. S1–S5). These observations can be explained with imprecise point estimates produced by both Rate4Site and dN∕dS inference at both low and high sequence divergence: In general, both methods correctly assess which sites in an alignment evolve more rapidly than which other sites, and hence correlations are high, but the specific rate estimates assigned to individual sites are poor at both low and high sequence divergence, and hence RMSDs are high as well.
Finally, we asked to what extent the results found for simulated sequences carry over to empirical datasets. We inferred both Rate4Site scores and site-wise dN∕dS in two distinct datasets, one consisting of six membrane proteins in mammals (taken from Spielman & Wilke, 2013) and one consisting of six alignments of HIV-1 genes (taken from Meyer & Wilke, 2015b). For both datasets, we measured the correlation between the original (non-normalized) dN∕dS and normalized Rate4Sites scores. We found that Rate4Site scores and dN∕dS were highly correlated, with correlation coefficients exceeding 0.8 in all cases (Fig. 9). Thus, we can conclude that our simulation results carry over to empirical datasets, and that Rate4Site scores are generally comparable to inferred dN∕dS values.
Note that we plotted raw (non-normalized) dN∕dS in Fig. 9, and as a result points are not expected to fall onto the x = y line. We plotted the data in this way because dN∕dS values are not typically normalized to their mean, and we wanted to assess how similar in magnitude raw dN∕dS scores are to Rate4Site scores. We found that for proteins under strong positive selection (such as HIV-1 gp120), raw dN∕dS is virtually identical to Rate4Site. However, for more typical proteins that experience little positive selection, raw dN∕dS values tend to be up to a factor 10 smaller than the corresponding Rate4Site scores.
We have compared codon-level site-specific evolutionary rates estimated via dN∕dS to amino-acid level site-specific rates estimated via Rate4Site. We have found that Rate4Site scores correlate well with the known true dN∕dS values both in sequences simulated with a dN∕dS model and in sequences simulated with a MutSel model. Correlations generally increase and approach 1 for more diverged sequence alignments and for alignments with more taxa. Correlations are generally somewhat stronger when there is no variation in dS among sites, though this effect is minor. We have also compared Rate4Site scores to inferred dN∕dS values and have found high correlations between the two measures, even for less diverged and smaller alignments. Finally, we have verified the relationship between Rate4Site scores and dN∕dS in a set of empirical datasets, and have found correlations close to 1 between the two rate estimates.
Surprisingly, even in scenarios of low sequence divergence or few taxa, when Rate4Site scores are only weakly correlated with the true dN∕dS, we have found that they nevertheless correlate highly with inferred dN∕dS. For all levels of sequence divergence and numbers of taxa, Rate4Site scores always correlate more strongly with inferred dN∕dS than with true dN∕dS. While this observation may seem counterintuitive, there is a simple explanation. The evolutionary process is stochastic, and by random chance some sites will experience substantially more or fewer mutations than expected given their true rate. At those sites, all rate inference methods are expected to over/under-estimate the true rate, since their only input are the mutations that actually happened. As a result, we would generally expect different inference methods to produce results that are more similar to each other than they are to the true, underlying parameters. In agreement with this reasoning, a strong relationship between Rate4Site and inferred dN∕dS was also evident in empirical datasets. For the majority of alignments that we considered, correlation coefficients were in excess of 0.9. For the two alignments with the lowest correlation coefficients, of 0.83 for both alignments, the number of sequences were 19 and 22 and sequence divergence was low. Thus, unless alignments are very small and/or have very little divergence, Rate4Site scores and site-specific dN∕dS can be expected to correlate strongly in all cases. These findings demonstrate that Rate4Site and dN∕dS approaches have comparable ability to infer rates from sequence alignments. For sufficiently diverged alignments, both methods accurately recover the true underlying rates. And for alignments with less divergence, they mis-estimate the underlying rates in a similar fashion.
In several cases, we saw high RMSDs despite high correlations between Rate4Site and inferred dN∕dS. This observation suggests that both approaches rank sites similarly in terms of whether they are more rapidly or more slowly evolving, even when the point estimates for rates differ. One major factor causing differences in point estimates is the specific statistical approach used for rate inference in Rate4Site and in our dN∕dS estimation. In particular, Rate4Site uses a random-effects likelihood (REL) approach in a Bayesian framework, while for dN∕dS we employed a fixed-effects likelihood (FEL) approach in a maximum-likelihood framework. Prior work has shown that REL and FEL inference for the same model (e.g., dN∕dS) tends to produce comparable estimates (Meyer, Dawson & Wilke, 2013; Kosakovsky Pond & Frost, 2005). However, the dynamic range for REL models tends to be reduced relative to that of FEL models, and in particular, REL models tend to overestimate the rates of sites with no or very few mutations. This pattern is evident for example in the left-most panel of Fig. 3A, where the Rate4Site scores do not fall below 0.5. Because of the tendency of REL models to overestimate rates of completely conserved sites, we generally prefer FEL approaches for site-specific rate estimation, and we believe that an FEL version of the Rate4Site model would be a useful addition to the toolbox of rate-inference approaches.
The dN∕dS metric is frequently used to identify sites under positive selection in viruses (Vijaykrishna et al., 2008; Wood et al., 2009; Demogines et al., 2013; Meyer & Wilke, 2015a). By contrast, Rate4Site has been mostly applied to identify conserved sites that correspond to protein–protein interaction sites or active sites in enzyme (Mousson et al., 2005; Fischer, Mayer & Söding, 2008; Tuncbag, Gursoy & Keskin, 2009; Bradford et al., 2006; Guney et al., 2008). Our results here show that for purposes of finding the most conserved or most rapidly varying sites in a sequence alignments, both methods would likely identify similar sites. One advantage of the dN∕dS approach, of course, is the ability to test whether dN∕dS is significantly above 1. When using Rate4Site scores, one can identify the most rapidly varying sites but one cannot run a statistical test that would determine whether the site is positively selected or not.
Recently, there has been considerable interest in linking site-specific rate variation to structural features of proteins (Echave, Spielman & Wilke, 2016). Studies addressing this topic have considered both dN∕dS-based methods (Scherrer, Meyer & Wilke, 2012; Franzosa & Xia, 2009; Shahmoradi et al., 2014; Kim et al., 2006; Meyer & Wilke, 2015b; Meyer & Wilke, 2015a) and Rate4Site scores (Huang et al., 2014; Yeh et al., 2014b; Yeh et al., 2014a; Jack et al., 2016; Huang et al., 2015), though these studies have generally been done on disparate datasets. The extent to which results found with dN∕dS carry over to Rate4Site and vice versa has not been clear. Our findings here show that since the two methods infer rates that correlate strongly with each other, either type of inferred rate should produce comparable correlation strengths with structural features such as solvent accessibility.
We note several caveats to our conclusions. First, our simulated alignments were generally large and diverged, even for the smallest number of taxa and lowest branch lengths. Even smaller and/or less diverged alignments will yield more noisy, less reliable Rate4Site inferences. Second, all our simulated alignments were obtained under the assumption that sites evolve independently from each other and that the rate of evolution does not change over time. These assumptions will generally increase the congruence between the true, simulated dN∕dS and the inferred Rate4Site score. However, the strong correlations we observed between Rate4Site scores and dN∕dS in several empirical datasets demonstrate that neither of these assumptions and limitations fundamentally invalidate our main findings. Amino-acid level and codon-level analyses of sequence data will generally yield comparable estimates of site-specific rates of evolution.
Relationship between Rate4Site scores and site-specific dN∕dS for simulations performed with a dN∕dS model with gamma-distributed true rates
We used the gamma distribution observed for the HIV-1 capsid protein, with shape parameter α = 0.230 and rate parameter β = 0.659 (Meyer & Wilke, 2015b). Each point represents the mean over 50 replicate simulations. The error bars represent the standard error. In nearly all cases, error bars are smaller than the symbol size. (A) Correlations and (B) RMSD values between Rate4Site scores and true dN∕dS. (C) Correlations and (D) RMSD values between Rate4Site scores and inferred dN∕dS. All simulations were performed without codon bias (neutral synonymous codons).
Relationship between Rate4Site scores and site-specific dN∕dS for simulations performed with a dN∕dS model with gamma-distributed true rates
We used the gamma distribution observed for the HIV-1 gp120 protein, with shape parameter α = 0.350 and rate parameter β = 0.249 (Meyer & Wilke, 2015b). Each point represents the mean over 50 replicate simulations. The error bars represent the standard error. In nearly all cases, error bars are smaller than the symbol size. (A) Correlations and (B) RMSD values between Rate4Site scores and true dN∕dS. (C) Correlations and (D) RMSD values between Rate4Site scores and inferred dN∕dS. All simulations were performed without codon bias (neutral synonymous codons).
Relationship between Rate4Site scores and site-specific dN∕dS for simulations performed with a dN∕dS model with gamma-distributed true rates
We used the gamma distribution observed for the HIV-1 matrix protein, with shape parameter α = 0.647 and rate parameter β = 0.861 (Meyer & Wilke, 2015b). Each point represents the mean over 50 replicate simulations. The error bars represent the standard error. In nearly all cases, error bars are smaller than the symbol size. (A) Correlations and (B) RMSD values between Rate4Site scores and true dN∕dS. (C) Correlations and (D) RMSD values between Rate4Site scores and inferred dN/dS. All simulations were performed without codon bias (neutral synonymous codons).
We used the gamma distribution observed for the HIV-1 protease protein, with shape parameter α = 0.378 and rate parameter β = 1.016 (Meyer & Wilke, 2015b). Each point represents the mean over 50 replicate simulations. The error bars represent the standard error. In nearly all cases, error bars are smaller than the symbol size. (A) Correlations and (B) RMSD values between Rate4Site scores and true dN∕dS. (C) Correlations and (D) RMSD values between Rate4Site scores and inferred dN∕dS. All simulations were performed without codon bias (neutral synonymous codons).
We used the gamma distribution observed for the HIV-1 reverse transcriptase protein, with shape parameter α = 0.238 and rate parameter β = 0.627 (Meyer & Wilke, 2015b). Each point represents the mean over 50 replicate simulations. The error bars represent the standard error. In nearly all cases, error bars are smaller than the symbol size. (A) Correlations and (B) RMSD values between Rate4Site scores and true dN∕dS. (C) Correlations and (D) RMSD values between Rate4Site scores and inferred dN∕dS. All simulations were performed without codon bias (neutral synonymous codons).