As proteins evolve from a common ancestor, their sequences and structures diverge from each other (Chothia and Lesk, 1986; Povolotskaya and Kondrashov, 2010). Multiple previous studies have investigated the relationship between the conservation of protein molecular function, sequence identity (Lee et al., 2007; Tian and Skolnick, 2003; Worth et al., 2009) and structural similarity (Chothia and Lesk, 1986; Wilson et al., 2000). For example, the likelihood that two proteins share the same molecular function, given their sequence (Tian and Skolnick, 2003) or structural (Wilson et al., 2000) similarity, has been used to investigate the emergence of new protein functions (Rost, 2002; Conant and Wolfe, 2008), and to perform functional annotations of protein sequences (Lee et al., 2007; Wilson et al., 2000). In this work, we focused on a different and currently unaddressed set of questions. Namely, how far can two sequences diverge while continuously maintaining the same molecular function? What are the temporal patterns of this divergence across billions of years of evolution? And how different protein sites contribute to the long-term divergence between orthologs with the same molecular function? We note that the requirement for the continuous conservation of molecular function is crucial in this context, as multiple examples of convergent evolution and protein engineering demonstrate that the same molecular function, such as catalysis of a specific chemical reaction, can in principle be accomplished by proteins with unrelated sequences and even different folds (Bork et al., 1993; Galperin et al., 1998; Omelchenko et al., 2010).

It was previously demonstrated that proteins with the same structural fold frequently diverge to very low (~10%) levels of sequence identity (Rost, 1997). These results suggest that conservation of protein fold, that is, the overall arrangement and topological connections of protein secondary structures (Murzin et al., 1995), exerts relatively minor constraints on how far protein sequences can diverge. In contrast to protein folds, it is possible that conservation of specific molecular functions will significantly limit the long-term divergence of protein orthologs. While only a relatively small fraction of protein residues (~3–5%) are usually directly involved in catalysis (Lehninger et al., 2013), recent analyses have demonstrated that even sites located far from catalytic residues may be significantly constrained in evolution. Because substitutions at these sites can have substantial effects on molecular function (Firnberg et al., 2016), it is likely that functionally-related sequence constraints extend far beyond catalytic residues.

In this study, we explored the long-term divergence patterns of protein orthologs by characterizing their pairwise sequence and structural similarity as a function of their divergence time. We used several models of molecular evolution to calculate the divergence rates, defined as the decrease in pairwise sequence identity or structural similarity per unit time, between orthologous proteins with the same molecular function. We also characterized the long-term divergence patterns at protein sites with different levels of evolutionary conservation, different locations in protein structures, and different experimentally measured fitness effects of amino acid substitutions. Finally, we explored how the limits of sequence and structural divergence after billions of years of evolution depend on the degree of functional conservation between orthologs.

To study the evolution of proteins with the same molecular function, we initially focused our analysis on enzymes because their molecular function is usually well defined. The Enzyme Commission (EC) classifies enzymatic functions using a hierarchical 4-digit code (Bairoch, 1999), such that two enzymes that share all four EC digits catalyze the same biochemical reaction. We used protein sequences representing 64 EC numbers from 22 diverse model organisms across the three domains of life (Supplementary file 1). The considered activities include members of all six major enzyme classes: oxidoreductases, transferases, hydrolases, lyases, isomerases and ligases.

To investigate whether the conservation of enzymatic function limits the divergence between orthologous sequences, we first calculated global pairwise sequence identities between orthologs as a function of their divergence times (Figure 1, Figure 1—figure supplement 1). The pairwise divergence times reported in the literature (Hedges et al., 2006) between the considered 22 species (Supplementary file 1) were used as a proxy for the divergence times between corresponding orthologous proteins. For each enzymatic activity, we constructed phylogenetic trees based on the orthologous protein sequences and compared them to the corresponding species’ trees. Protein sequences with clear differences to the species'phylogenetic tree topologies, suggesting cases of horizontal gene transfer, were excluded from the analysis (see Materials and methods).

Figure 1 with 7 supplements see all

Download asset Open asset

We next considered two simple models of long-term protein evolution, one without a long-term limit of sequence divergence and the other with an explicit divergence limit. The first model corresponds to sequence divergence with equal and independent amino acid substitution rates across all proteins sites (Dickerson, 1971; Zuckerkandl and Pauling, 1965); see Equation 1, where y represents global sequence identity, t represents divergence time, and R₀ represents the average substitution rate (Dickerson, 1971). In this model, back substitutions are not allowed, and sequence divergence slows down with time simply due to multiple substitutions at the same protein sites and progressively fewer non-mutated sites. The second model corresponds to sequence divergence where, in addition to sites with equal and independent substitution rates, there is a minimal fraction of identical sites at long divergence times; the fraction of identical sites is represented by Y₀ in Equation 2.

We applied the two models to fit the sequence divergence for each of the considered enzymatic functions. The best model fits for four representative metabolic activities are shown in Figure 1 (black for the first model and red for the second); the fits for the remaining metabolic activities are shown in Figure 1—figure supplement 1. In 62 of the 64 cases, the second model fits the divergence data significantly better than the first model (F-test p-value<0.05, Supplementary file 2a). Moreover, in 95% of the cases (61/64) the maximum likelihood value of the parameter Y₀ is significantly higher (Wald test p-value<0.05) than the average sequence identity between random protein sequences based on their optimal global alignment (~13.5%, shown in Figure 1 and Figure 1—figure supplement 1 by dashed black lines). The distribution of the fitted parameter Y₀ suggests a long-term sequence identity >25% (with average ~40%) between considered orthologs (Figure 2a); this demonstrates that conservation of a specific enzymatic function significantly limits long-term protein sequence divergence. Notably, model two is mathematically equivalent (see Materials and methods) to a divergence model with equal substitution rates across sites, a limited number of amino acid types accepted per site, and allowed back substitutions (Tajima and Nei, 1984; Gilson et al., 2017; Yang, 2006). In this model, the parameter Y₀ represents the inverse of the effective number of acceptable amino acid types per site during protein evolution. Our results thus suggest that, on average, only 2 to 4 amino acids are acceptable per site for proteins that strictly conserve their molecular function (Figure 2a, top blue X axis); we note that this low average number does not contradict the fact that more than four amino acid types could be observed at a given protein site at low frequencies (Breen et al., 2012).

Figure 2

Download asset Open asset

The two aforementioned models simplify the process of sequence divergence by considering the same substitution rates across protein sites. A more realistic and commonly used model of protein evolution assumes a gamma distribution (Yang et al., 2000) of substitution rates across protein sites; see Equation 3; (Ota and Nei, 1994), where α represents the shape parameter of the gamma distribution. The best fits of such a variable-rate model (blue in Figure 1 and Figure 1—figure supplement 1) showed that the rates of protein sequence divergence between orthologous enzymes have decreased by more than 10 times during ~4 billion years of evolution (see Materials and methods and Supplementary file 2b). Although the third model does not explicitly consider a long-term divergence limit, the obtained model fits also show that the vast majority of orthologous enzymes with the same function will remain above 25% sequence identity on the timescales when Earth environments will be hospitable to life (1–3 billion years from the present O'Malley-James et al., 2014) (Figure 1—figure supplement 2).

The observed divergence limit is not an artifact due to difficulty detecting remote protein homologs, as it occurs at relatively high sequence identities (Figure 1 and Figure 1—figure supplement 1), for which corresponding orthologs can be easily identified by computational sequence comparison methods. Furthermore, the results remained similar when we restricted the analysis to orthologous enzyme pairs with experimentally validated molecular functions (Figure 1—figure supplement 3), based on publications referenced in the BRENDA database (Chang et al., 2015). The results also remain robust towards the variance in the estimates of divergence times between considered species (see Materials and methods). We note that the divergence limit between orthologs with the same molecular function does not imply that the rates of molecular substitutions decrease in evolution. It is also not simply due to the curvilinear relationship between time and sequence identity caused by multiple mutations at the same sites; specifically, the observed decrease in divergence rates is substantially higher (by >10 fold) than the one expected in model one simply due to multiple substitutions at the same protein sites. Instead, the effective limit is reached when, due to a small number of amino acids types accepted per protein site and back substitutions, additional amino acid replacements do not lead to a substantial further increase in protein sequence and structural divergence (Meyer et al., 1986).

Interestingly, following the previously introduced metaphor of the expanding protein universe (Povolotskaya and Kondrashov, 2010; Dokholyan et al., 2002), we can use the third model (Equation 3) to express the divergence rate between orthologs as a function of protein distance (D = 1 – y, where y is the fractional sequence identity ranging from 0 to 1), see Equation 4. This equation, similarly to Hubble’s law of universe expansion (Hubble, 1929), describes how the divergence rate depends on the distance between protein orthologs. In contrast to the real universe, the expansion rate of the protein universe significantly decreases with divergence time and with distance between protein orthologs. For example, the divergence rate between protein orthologs drops, on average, to only ~2% sequene identity decrease per billion years when their mutual sequence identity reaches 30% (corresponding to protein distance of 70%; Figure 1—figure supplement 4).

The analyses described above focused on the divergence of enzymes with the same molecular function. In order to investigate whether the observed divergence patterns are not specific to enzymes, we repeated the analysis using non-enzymatic ancient orthologs (Figure 1—figure supplement 5, Supplementary file 2c). The set of analyzed 29 protein families included ribosomal proteins, heat shock proteins, membrane transporters, and electron transfer flavoproteins (Supplementary file 2d). Based on the same set of 22 species used in the analysis of enzyme families, we found that model two fitted the data significantly better than model one, and that the parameter Y₀ was >25% for the majority (23/29) of the non-enzymatic protein families (Figure 2b, Supplementary file 2c). Interestingly, we also identified 19 additional orthologous groups showing two clearly different divergence patterns (Figure 1—figure supplement 6), with pairs of eukaryotic orthologs diverging substantially faster and farther than prokaryotic orthologs in the same protein family. The orthologous groups with this behavior included mitochondrial ribosomal proteins and initiation factors of mitochondrial translation (Supplementary file 2e). It has been previously postulated that mitochondrial ribosomal proteins diverged significantly faster in eukaryotes, compared to the divergence between their bacterial orthologs, due to compensatory protein substitutions following the accumulation of slightly deleterious substitutions in the mitochondrial ribosomal RNA (Barreto and Burton, 2013).

Having established, in the first half of the manuscript that conservation of molecular function significantly limits long-term sequence evolution, we investigated, in the second half, how different protein sites contribute to the observed divergence constraints. Specifically, whether the same protein sites are conserved between ancient orthologs in different phylogenetic lineages, how sites with different fitness effects of amino acid substitutions contribute to the divergence limit, and how structural locations of protein sites affect their long-term divergence patterns. We also explored how different levels of functional specificity constrain sequence and structural divergence.

To investigate whether the same protein sites are usually conserved between orthologs in different phylogenetic lineages, we aligned the sequences of ancient enzyme orthologs with the same molecular function (see Materials and methods). We then calculated how often each protein site is occupied by identical amino acids across pairs of orthologs from phylogenetically independent linages (Figure 3—figure supplement 1). Orthologous protein pairs from independent lineages were obtained using species' pairs that do not share any edges in the corresponding phylogenetic tree (Arnold and Stadler, 2010) (Figure 3a); for example, in Figure 3a the pair D-H is independent of the pair A-B but not of the pair E-F. We performed the above analysis using 30 enzymatic activities for which at least 20 independent pairs of orthologs could be identified based on annotations in the KEGG database (Kanehisa et al., 2016) (see Materials and methods). The results demonstrated that only a relatively small fraction of protein sites (10–20%) are universally conserved, that is, they are identical in a majority (>90%) of independent lineages (Figure 3b). Therefore, the observed long-term divergence limit is not primarily due to sets of universally conserved protein sites; instead, different sites contribute to the limit in independent phylogenetic lineages. By comparing the fractions of universally conserved sites to the average sequence identity between distant orthologs (~40%, Figure 2a) we found that these sites account, on average, for only ~35% of the observed sequence identity at long divergence distances. The analysis also revealed that different protein families show different probability distributions of identical sites (Figure 3—figure supplement 1). This is likely a consequence of diverse structural and functional requirements across protein families, leading to protein-family specific constraints on protein sites.

Figure 3 with 1 supplement see all

Download asset Open asset

We next investigated the long-term divergence patterns at protein sites with different fitness effects of amino acid substitutions. To that end, we experimentally measured the fitness effects of all possible single amino acid substitutions in a representative enzyme, the Escherichia coli dihydrofolate reductase (FolA, EC 1.5.1.3). We selected FolA for the experiments due to its small size (159 amino acids) and essential role in the E. coli metabolism (Benkovic et al., 1988); also, the long-term protein sequence identity between FolA orthologs (~32%, see Figure 1a) is similar to other analyzed enzymes (Figure 2a). Following a recently described strategy (Kelsic et al., 2016), we used the Multiplex Automated Genome Engineering (MAGE) approach (Wang et al., 2009) to introduce every possible amino acid substitution at each FolA site in E. coli. To evaluate the relative fitness effects of protein substitutions we measured the growth rate of strains containing each protein variant compared to the ‘wild type’ (WT) strain into which substitutions were introduced. Relative growth rates were measured in parallel by performing growth competition experiments between the pooled mutants. Amplicon sequencing of the folA gene was then used to measure the relative changes of mutant and WT abundances as a function of time (see Materials and methods, Supplementary file 3).

Using the MAGE growth measurements in E. coli, we investigated the patterns of long-term sequence divergence at protein sites with different fitness effects of amino acid substitutions. Specifically, we sorted FolA protein sites into several groups according to their experimentally measured average fitness effects (Figure 4—figure supplement 1), and explored sequence divergence for sites within each fitness group (Figure 4a, different colors). We evaluated sequence identity between FolA orthologs across divergence times using all pairwise comparisons between ~300 orthologous sequences from the COG database (Galperin et al., 2015). Although, as expected, sites with stronger fitness effects diverged more slowly, our analysis revealed interesting differences in temporal divergence patterns for sites with small and large fitness effects. For sites in the least deleterious fitness group (Figure 4a, blue) we observed, similar to the global sequence identity, a substantial decrease (~10 fold, see Equation 5 in Materials and methods) in mutual divergence rates after ~1.5 billion years of evolution. Notably, even for FolA sites with mild fitness effects, sequence identity remains above 25% at long divergence distances. In contrast to sites with mild fitness effects, sites with the most deleterious mutations (Figure 4a, black) displayed a much slower, but approximately constant average divergence rate throughout evolutionary history. This pattern suggests that, in contrast to divergence at sites with small fitness effects, the divergence at sites with large effects is not yet close to saturation.

Figure 4 with 2 supplements see all

Download asset Open asset

To assess the generality of the FolA results we used another dataset (Kelsic et al., 2016), obtained using MAGE, of fitness values for all possible amino acids substitutions in the E. coli translation initiation factor InfA (Figure 4b). Consistent with the relatively higher level of sequence conservation of InfA, we observed lower average mutant growth rates and lower rates of sequence divergence in each fitness group. Nevertheless, the long-term divergence patterns were qualitatively similar between the two proteins. For sites in the least deleterious InfA fitness group (Figure 4b, blue), we observed a substantial decrease in the divergence rate after ~2 billon years of evolution. In contrast, sites with strongest fitness effects (Figure 4b, pink) displayed a slower but approximately constant divergence rate.

Because the fitness effects of mutations at a protein site may change in evolution (Lunzer et al., 2010; Chan et al., 2017), it is interesting to investigate how fitness effects measured in one species, such as E. coli, correlate with the site conservation in other species at the divergence limit. To explore this question, we calculated the probability that a protein site is occupied at long evolutionary times (~2 billion years for FolA and ~2.5 billion years for InfA) by the same amino acid in phylogenetically independent lineages (Figure 3a). We then investigated how this probability changes as a function of the average fitness effects of substitutions at the site measured in E. coli (Figure 4c). For both FolA and InfA, the probability that a protein site is identical at large divergence distances first increases, approximately linearly with increasing average fitness effects, and then begins to saturate for sites with large (>30% growth rate decrease) fitness effects. Thus, the fitness effects at a protein site correlate with the site’s conservation even after billions of years of evolution.

The sequence constraints revealed by our analysis likely arise due to the conservation of corresponding protein structures required for efficient catalysis and molecular function (Wilson et al., 2000; Watson et al., 2005). Therefore, in addition to sequence divergence we also investigated the long-term structural divergence of orthologous proteins with the same function. For this analysis we used >1000 orthologous pairs of enzymes sharing all 4 EC digits with known 3D structures (Berman et al., 2000) (see Materials and methods); structures of orthologous enzymes were aligned using the TM-align algorithm (Zhang and Skolnick, 2005). This analysis demonstrated that the average root mean square deviation (RMSD) between C-alpha atoms of the orthologous enzymes increases (Spearman’s r = 0.44, p-value<1e-20) with divergence time (Figure 5a). Nevertheless, the C-alpha RMSD between orthologs rarely increases beyond 3 Å, even at long evolutionary distances. Consistent with sequence evolution (Figure 1), we also observed a substantial decrease in the rate of structural divergence after ~1.5 billion years of divergent evolution.

Figure 5

Download asset Open asset

Figure 5—source data 1

RMSD versus divergence times for proteins with the same enzymatic function.

https://doi.org/10.7554/eLife.39705.021

Download elife-39705-fig5-data1-v2.txt

Only a small fraction of all enzyme residues forms an active site and directly participates in catalysis. Therefore, we investigated next how the sequence divergence depends on the spatial proximity of protein positions to active site residues. It was previously demonstrated that evolutionary rates of amino acid substitutions correlate with protein sites’ spatial distance to catalytic residues (Jack et al., 2016). Notably, differences in short-term evolutionary rates do not imply the existence of a divergence limit, nor do they inform how site-specific divergence patterns correlate with sites' structural locations. Thus, the main goal of our analysis was to investigate the temporal patterns of the long-term divergence, and the effective divergence limit for sites at various distances to the active site. We considered catalytic site annotations available from the Protein Data Bank (Berman et al., 2000), UniProt-KB (UniProt Consortium, 2015) and the Catalytic Site Atlas (Porter, 2004) and quantified the average sequence divergence for sites at various distances from catalytic residues (see Materials and methods, Figure 5b). We based this analysis on the same set of enzymatic activities used to study global sequence divergence (Figure 1 and Figure 1—figure supplement 1). Although, as expected, residues close to the active site were the most highly conserved (Jack et al., 2016; Halabi et al., 2009), even distant sites displayed a substantial divergence limit (~40%) at long evolutionary distances. This result suggests that the spatial constraints required for specific molecular function usually propagate throughout the entire protein structure and significantly limit the long-term divergence even at sites distant from catalytic residues.

Finally, we investigated how various degrees of functional conservation affect the long-term divergence between orthologs. To that end, we compared the long-term sequence and structural similarities of enzymes sharing their full EC classification to those sharing only the first three digits of their EC classification (Figure 6a and b); for this comparison we only used orthologs from species with divergence times > 2 billion years (see Materials and methods). In contrast to enzymes sharing all four EC digits, conservation of the first three digits indicates only a general class of substrates or cofactors (Bairoch, 1999). This analysis revealed significantly lower long-term sequence identities (27% vs. 37% identity, Mann-Whitney p-value<10⁻²⁰) and structural similarities (2.4 vs. 1.8 Å RMSD, P-value 2 × 10⁻¹⁸) between orthologs sharing only partial EC numbers. Notably, orthologs sharing only the first three EC digits are still substantially more conserved, both in sequence and structure (p-values<10⁻²⁰), than pairs of enzymes with the same structural fold, but unrelated enzyme activities (i.e. enzymes sharing no digits in the EC classification) (Figure 6a and b, Dawson et al., 2017).

Figure 6

Download asset Open asset

Figure 6—source data 1

Sequence identities between ancient orthologs sharing the same EC number and only the first three digits of their EC numbers.

https://doi.org/10.7554/eLife.39705.023

Download elife-39705-fig6-data1-v2.txt

We also investigated the sequence constraints at the same level of protein structural divergence for proteins with different degrees of functional conservation. To that end, we calculated the sequence identity between orthologs, sharing either their full or partial EC numbers, at different bins of long-term structural similarity (Figure 6c). Interestingly, we observed that even at the same level of C-alpha RMSD divergence, orthologs sharing full EC numbers usually have higher levels of sequence identity compared to orthologous pairs with the same level of structural divergence but sharing only three EC digits. This result indicates that conservation of molecular function constrains sequence divergence even beyond the requirement to maintain a specific spatial structure.

Our analysis demonstrates that, in contrast to proteins with the same fold (Rost, 1997), the requirement to strictly conserve the same molecular function significantly limits the long-term sequence and structural divergence of protein orthologs. Although we confirmed the result by Povolotskaya and Kondrashov (2010) that ancient protein orthologs are still diverging from each other, our study reveals that the rate of this divergence becomes increasingly slow for orthologs that strictly conserve their function. Even a slight relaxation of functional specificity, for example from full to partial EC conservation (Figure 6a and b), leads to substantially more pronounced long-term sequence and structural divergence. Similarly, a substantial sequence identity between homologous restriction endonucleases is usually limited to isoschizomers, that is, proteins specific to the same target DNA sequence (Pingoud et al., 2014).

We believe that the observed divergence patterns can be explained by the following mechanistic model. Proteins with the same molecular function usually conserve the identity of their chemical and biological substrates and interaction partners. This conservation leads to functional pressure to closely preserve the spatial positions and dynamics of key protein residues necessary for efficient catalysis and function (Lehninger et al., 2013). In turn, the requirement to continuously preserve structural properties and functional dynamics of key protein residues likely imposes a strict requirement to preserve the overall protein structure, in other words, structural optimality necessary for efficient and specific protein function. The evolutionary pressure to preserve structural optimality (to within <3 Å C-alpha RMSD) required for a given molecular function leads, in agreement with the results by Chothia and Lesk (1986) and others (Gilson et al., 2017), to substantial levels of overall sequence conservation and the observed divergence limit. Our analysis further demonstrates that less than four amino acid types are accepted, on average, per site for proteins strictly conserving their molecular function.

We note that the observed conservation may reflect both the impact of amino acid substitutions on protein activity, due to changes in equilibrium positions and dynamics of protein residues, and on protein abundance, due to changes in overall protein stability (Bershtein et al., 2015; Adkar et al., 2017). Nevertheless, direct and comprehensive biochemical experiments demonstrated that the deleterious effects of mutations primarily arise from changes in specific protein activity rather than decreases in protein stability and cellular abundance (Firnberg et al., 2016). Our results support this model, demonstrating that conservation of functional specificity imposes substantially more stringent long-term sequence constraints than simply conservation of protein folds and protein stability. Indeed, while every folded protein should maintain its stability, our results demonstrate that maintaining stable folds does not constrain long-term divergence to more than 10-15% sequence identity, which is substantially less than the requirement for maintaining specific molecular functions (~40% sequence identity). 

The presented results demonstrate that only about a third of the sequence conservation between distant orthologs with the same molecular function can be attributed to universally conserved protein sites, that is, sites occupied by identical amino acids in almost all lineages. In contrast, we found that different protein sites are usually identical between orthologs from different lineages. This result is likely due, at least in part, to the epistatic nature of protein sequence landscapes, where mutations that are neutral in one lineage are often prohibitively deleterious in another (Breen et al., 2012; Lunzer et al., 2010). In the context of the aforementioned divergence model, the evolution of mitochondrial ribosomal proteins in eukaryotes (Figure 1—figure supplement 6) provides an interesting example, suggesting that orthologs’ divergence can be substantially accelerated by co-evolution with their interaction partners or relaxation of selection pressures.

Our experimental and computational analyses also delineate two distinct stages of the long-term divergence of orthologs with the same molecular function. During the first 1–2 billion years of divergence, substitutions at protein sites with mild fitness effects lead to a substantial (40–60%) decrease in sequence identity. After the first stage, divergence at these sites effectively saturates. The saturation at sites with small fitness effects, combined with very slow divergence rate at sites with large fitness effects (Figure 4), leads to a substantially slower sequence and structural divergence during the second stage. Interestingly, as a consequence of this slowdown, for the past billion years there has not been a substantial decrease in sequence and structural similarity between ancient orthologs with the same molecular function. Further analyses of biochemical, biophysical and cellular constraints will reveal how various structural and functional properties influence proteins’ long-term evolution, and how protein functional efficiency may be compromised by deleterious mutations (Shendure and Akey, 2015).

Sequence reads were deposited to the SRA database with accession number: SRP152339. Source data files have been provided for figure 1, figure 5, figure 6 and figure 1 supplements 1, 3, 5 and 6.

1. 1. Konaté MM
   2. Plata G
   3. Park J
   4. Usmanova DR
   5. Wang HH
   6. Vitkup D

   (2018) SRA

   ID SRP152339. Molecular function limits divergent protein evolution on planetary timescales.

   https://www.ncbi.nlm.nih.gov/sra/SRP152339

1. 1. Eric D Kelsic
   2. Hattie Chung
   3. Niv Cohen
   4. Jimin Park
   5. Harris H Wang
   6. Roy Kishony

   (2016) Supplementary material of: https://doi.org/10.1016/j.cels.2016.11.004

   Data S1.

   https://doi.org/10.1016/j.cels.2016.11.004

1. 1. Adkar BV
   2. Manhart M
   3. Bhattacharyya S
   4. Tian J
   5. Musharbash M
   6. Shakhnovich EI

   (2017) Optimization of lag phase shapes the evolution of a bacterial enzyme

   Nature Ecology & Evolution 1:149.

   https://doi.org/10.1038/s41559-017-0149

   • PubMed
   • Google Scholar
2. 1. Altschul SF
   2. Madden TL
   3. Schäffer AA
   4. Zhang J
   5. Zhang Z
   6. Miller W
   7. Lipman DJ

   (1997) Gapped BLAST and PSI-BLAST: a new generation of protein database search programs

   Nucleic Acids Research 25:3389–3402.

   https://doi.org/10.1093/nar/25.17.3389

   • PubMed
   • Google Scholar
3. 1. Arnold C
   2. Stadler PF

   (2010) Polynomial algorithms for the maximal pairing problem: efficient phylogenetic targeting on arbitrary trees

   Algorithms for Molecular Biology 5:25.

   https://doi.org/10.1186/1748-7188-5-25

   • PubMed
   • Google Scholar
4. 1. Bairoch A

   (1999) The ENZYME data bank in 1999

   Nucleic Acids Research 27:310–311.

   https://doi.org/10.1093/nar/27.1.310

   • PubMed
   • Google Scholar
5. 1. Barreto FS
   2. Burton RS

   (2013) Evidence for compensatory evolution of ribosomal proteins in response to rapid divergence of mitochondrial rRNA

   Molecular Biology and Evolution 30:310–314.

   https://doi.org/10.1093/molbev/mss228

   • PubMed
   • Google Scholar
6. 1. Battistuzzi FU
   2. Feijao A
   3. Hedges SB

   (2004) A genomic timescale of prokaryote evolution: insights into the origin of methanogenesis, Phototrophy, and the colonization of land

   BMC Evolutionary Biology 4:44.

   https://doi.org/10.1186/1471-2148-4-44

   • PubMed
   • Google Scholar
7. 1. Bell EA
   2. Boehnke P
   3. Harrison TM
   4. Mao WL

   (2015) Potentially biogenic carbon preserved in a 4.1 billion-year-old zircon

   PNAS 112:14518–14521.

   https://doi.org/10.1073/pnas.1517557112

   • PubMed
   • Google Scholar
8. 1. Benkovic SJ
   2. Fierke CA
   3. Naylor AM

   (1988) Insights into enzyme function from studies on mutants of dihydrofolate reductase

   Science 239:1105–1110.

   https://doi.org/10.1126/science.3125607

   • PubMed
   • Google Scholar
9. 1. Berman HM
   2. Westbrook J
   3. Feng Z
   4. Gilliland G
   5. Bhat TN
   6. Weissig H
   7. Shindyalov IN
   8. Bourne PE

   (2000) The protein data bank

   Nucleic Acids Research 28:235–242.

   https://doi.org/10.1093/nar/28.1.235

   • PubMed
   • Google Scholar
10. 1. Bershtein S
    2. Serohijos AW
    3. Bhattacharyya S
    4. Manhart M
    5. Choi JM
    6. Mu W
    7. Zhou J
    8. Shakhnovich EI

    (2015) Protein homeostasis imposes a barrier on functional integration of horizontally transferred genes in Bacteria

    PLOS Genetics 11:e1005612.

    https://doi.org/10.1371/journal.pgen.1005612

    • PubMed
    • Google Scholar
11. 1. Betts HC
    2. Puttick MN
    3. Clark JW
    4. Williams TA
    5. Donoghue PCJ
    6. Pisani D

    (2018) Integrated genomic and fossil evidence illuminates life's early evolution and eukaryote origin

    Nature Ecology & Evolution 2:1556–1562.

    https://doi.org/10.1038/s41559-018-0644-x

    • PubMed
    • Google Scholar
12. 1. Bork P
    2. Sander C
    3. Valencia A

    (1993) Convergent evolution of similar enzymatic function on different protein folds: the hexokinase, ribokinase, and galactokinase families of sugar kinases

    Protein Science 2:31–40.

    https://doi.org/10.1002/pro.5560020104

    • PubMed
    • Google Scholar
13. 1. Breen MS
    2. Kemena C
    3. Vlasov PK
    4. Notredame C
    5. Kondrashov FA

    (2012) Epistasis as the primary factor in molecular evolution

    Nature 490:535–538.

    https://doi.org/10.1038/nature11510

    • PubMed
    • Google Scholar
14. 1. Chan YH
    2. Venev SV
    3. Zeldovich KB
    4. Matthews CR

    (2017) Correlation of fitness landscapes from three orthologous TIM barrels originates from sequence and structure constraints

    Nature Communications 8:14614.

    https://doi.org/10.1038/ncomms14614

    • PubMed
    • Google Scholar
15. 1. Chang A
    2. Schomburg I
    3. Placzek S
    4. Jeske L
    5. Ulbrich M
    6. Xiao M
    7. Sensen CW
    8. Schomburg D

    (2015) BRENDA in 2015: exciting developments in its 25th year of existence

    Nucleic Acids Research 43:D439–D446.

    https://doi.org/10.1093/nar/gku1068

    • PubMed
    • Google Scholar
16. 1. Chothia C
    2. Lesk AM

    (1986) The relation between the divergence of sequence and structure in proteins

    The EMBO Journal 5:823–826.

    https://doi.org/10.1002/j.1460-2075.1986.tb04288.x

    • Google Scholar
17. 1. Conant GC
    2. Wolfe KH

    (2008) Turning a hobby into a job: how duplicated genes find new functions

    Nature Reviews Genetics 9:938–950.

    https://doi.org/10.1038/nrg2482

    • PubMed
    • Google Scholar
18. 1. Dawson NL
    2. Lewis TE
    3. Das S
    4. Lees JG
    5. Lee D
    6. Ashford P
    7. Orengo CA
    8. Sillitoe I

    (2017) CATH: an expanded resource to predict protein function through structure and sequence

    Nucleic Acids Research 45:D289–D295.

    https://doi.org/10.1093/nar/gkw1098

    • Google Scholar
19. 1. DeSantis TZ
    2. Hugenholtz P
    3. Larsen N
    4. Rojas M
    5. Brodie EL
    6. Keller K
    7. Huber T
    8. Dalevi D
    9. Hu P
    10. Andersen GL

    (2006) Greengenes, a chimera-checked 16S rRNA gene database and workbench compatible with ARB

    Applied and Environmental Microbiology 72:5069–5072.

    https://doi.org/10.1128/AEM.03006-05

    • PubMed
    • Google Scholar
20. 1. Dickerson RE

    (1971) The structures of cytochrome c and the rates of molecular evolution

    Journal of Molecular Evolution 1:26–45.

    https://doi.org/10.1007/BF01659392

    • PubMed
    • Google Scholar
21. 1. Dokholyan NV
    2. Shakhnovich B
    3. Shakhnovich EI

    (2002) Expanding protein universe and its origin from the biological big bang

    PNAS 99:14132–14136.

    https://doi.org/10.1073/pnas.202497999

    • PubMed
    • Google Scholar
22. Software

    1. Felsenstein J

    (2005) PHYLIP (Phylogeny inference package) , version 3.6

    University of Washington, Seattle.

    http://evolution.genetics.washington.edu/phylip.html
23. 1. Firnberg E
    2. Labonte JW
    3. Gray JJ
    4. Ostermeier M

    (2016) A comprehensive, High-Resolution map of a gene's Fitness Landscape

    Molecular Biology and Evolution 33:1378.

    https://doi.org/10.1093/molbev/msw021

    • PubMed
    • Google Scholar
24. 1. Galperin MY
    2. Walker DR
    3. Koonin EV

    (1998) Analogous enzymes: independent inventions in enzyme evolution

    Genome Research 8:779–790.

    https://doi.org/10.1101/gr.8.8.779

    • PubMed
    • Google Scholar
25. 1. Galperin MY
    2. Makarova KS
    3. Wolf YI
    4. Koonin EV

    (2015) Expanded microbial genome coverage and improved protein family annotation in the COG database

    Nucleic Acids Research 43:D261–D269.

    https://doi.org/10.1093/nar/gku1223

    • PubMed
    • Google Scholar
26. 1. Gilson AI
    2. Marshall-Christensen A
    3. Choi JM
    4. Shakhnovich EI

    (2017) The role of evolutionary selection in the dynamics of protein structure evolution

    Biophysical Journal 112:1350–1365.

    https://doi.org/10.1016/j.bpj.2017.02.029

    • PubMed
    • Google Scholar
27. 1. Guy L
    2. Ettema TJ

    (2011) The archaeal 'TACK' superphylum and the origin of eukaryotes

    Trends in Microbiology 19:580–587.

    https://doi.org/10.1016/j.tim.2011.09.002

    • PubMed
    • Google Scholar
28. 1. Halabi N
    2. Rivoire O
    3. Leibler S
    4. Ranganathan R

    (2009) Protein sectors: evolutionary units of three-dimensional structure

    Cell 138:774–786.

    https://doi.org/10.1016/j.cell.2009.07.038

    • PubMed
    • Google Scholar
29. 1. Hedges SB
    2. Dudley J
    3. Kumar S

    (2006) TimeTree: a public knowledge-base of divergence times among organisms

    Bioinformatics 22:2971–2972.

    https://doi.org/10.1093/bioinformatics/btl505

    • PubMed
    • Google Scholar
30. 1. Hubble E

    (1929) A relation between distance and radial velocity among extra-galactic nebulae

    PNAS 15:168–173.

    https://doi.org/10.1073/pnas.15.3.168

    • PubMed
    • Google Scholar
31. 1. Jack BR
    2. Meyer AG
    3. Echave J
    4. Wilke CO

    (2016) Functional sites induce Long-Range evolutionary constraints in enzymes

    PLOS Biology 14:e1002452.

    https://doi.org/10.1371/journal.pbio.1002452

    • PubMed
    • Google Scholar
32. Book

    1. Jukes TH
    2. Cantor CR

    (1969)

    Evolution of Protein Molecules

    In: Munro H. N, editors. Mammalian Protein Metabolism. Academic Press. pp. 21–132.

    • Google Scholar
33. 1. Kanehisa M
    2. Sato Y
    3. Kawashima M
    4. Furumichi M
    5. Tanabe M

    (2016) KEGG as a reference resource for gene and protein annotation

    Nucleic Acids Research 44:D457–D462.

    https://doi.org/10.1093/nar/gkv1070

    • PubMed
    • Google Scholar
34. 1. Kelsic ED
    2. Chung H
    3. Cohen N
    4. Park J
    5. Wang HH
    6. Kishony R

    (2016) RNA structural determinants of optimal codons revealed by MAGE-Seq

    Cell Systems 3:563–571.

    https://doi.org/10.1016/j.cels.2016.11.004

    • PubMed
    • Google Scholar
35. 1. Kristensen DM
    2. Kannan L
    3. Coleman MK
    4. Wolf YI
    5. Sorokin A
    6. Koonin EV
    7. Mushegian A

    (2010) A low-polynomial algorithm for assembling clusters of orthologous groups from intergenomic symmetric best matches

    Bioinformatics 26:1481–1487.

    https://doi.org/10.1093/bioinformatics/btq229

    • PubMed
    • Google Scholar
36. 1. Ku C
    2. Nelson-Sathi S
    3. Roettger M
    4. Sousa FL
    5. Lockhart PJ
    6. Bryant D
    7. Hazkani-Covo E
    8. McInerney JO
    9. Landan G
    10. Martin WF

    (2015) Endosymbiotic origin and differential loss of eukaryotic genes

    Nature 524:427–432.

    https://doi.org/10.1038/nature14963

    • PubMed
    • Google Scholar
37. 1. Langmead B
    2. Salzberg SL

    (2012) Fast gapped-read alignment with bowtie 2

    Nature Methods 9:357–359.

    https://doi.org/10.1038/nmeth.1923

    • PubMed
    • Google Scholar
38. 1. Larkin MA
    2. Blackshields G
    3. Brown NP
    4. Chenna R
    5. McGettigan PA
    6. McWilliam H
    7. Valentin F
    8. Wallace IM
    9. Wilm A
    10. Lopez R
    11. Thompson JD
    12. Gibson TJ
    13. Higgins DG

    (2007) Clustal W and clustal X version 2.0

    Bioinformatics 23:2947–2948.

    https://doi.org/10.1093/bioinformatics/btm404

    • PubMed
    • Google Scholar
39. 1. Lee D
    2. Redfern O
    3. Orengo C

    (2007) Predicting protein function from sequence and structure

    Nature Reviews Molecular Cell Biology 8:995–1005.

    https://doi.org/10.1038/nrm2281

    • PubMed
    • Google Scholar
40. Book

    1. Lehninger AL
    2. Nelson DL
    3. Cox MM

    (2013)

    Lehninger Principles of Biochemistry (6th Edition)

    New York: W.H. Freeman.

    • Google Scholar
41. 1. Lunzer M
    2. Golding GB
    3. Dean AM

    (2010) Pervasive cryptic epistasis in molecular evolution

    PLOS Genetics 6:e1001162.

    https://doi.org/10.1371/journal.pgen.1001162

    • PubMed
    • Google Scholar
42. 1. Meyer TE
    2. Cusanovich MA
    3. Kamen MD

    (1986) Evidence against use of bacterial amino acid sequence data for construction of all-inclusive phylogenetic trees

    PNAS 83:217–220.

    https://doi.org/10.1073/pnas.83.2.217

    • Google Scholar
43. 1. Murzin AG
    2. Brenner SE
    3. Hubbard T
    4. Chothia C

    (1995) SCOP: a structural classification of proteins database for the investigation of sequences and structures

    Journal of Molecular Biology 247:536–540.

    https://doi.org/10.1016/S0022-2836(05)80134-2

    • PubMed
    • Google Scholar
44. 1. O'Malley-James JT
    2. Cockell CS
    3. Greaves JS
    4. Raven JA

    (2014) Swansong biospheres II: the final signs of life on terrestrial planets near the end of their habitable lifetimes

    International Journal of Astrobiology 13:229–243.

    https://doi.org/10.1017/S1473550413000426

    • Google Scholar
45. 1. Omelchenko MV
    2. Galperin MY
    3. Wolf YI
    4. Koonin EV

    (2010) Non-homologous isofunctional enzymes: A systematic analysis of alternative solutions in enzyme evolution

    Biology Direct 5:31.

    https://doi.org/10.1186/1745-6150-5-31

    • Google Scholar
46. 1. Ota T
    2. Nei M

    (1994) Estimation of the number of amino acid substitutions per site when the substitution rate varies among sites

    Journal of Molecular Evolution 38:642–643.

    https://doi.org/10.1007/BF00175885

    • Google Scholar
47. 1. Parfrey LW
    2. Lahr DJ
    3. Knoll AH
    4. Katz LA

    (2011) Estimating the timing of early eukaryotic diversification with multigene molecular clocks

    PNAS 108:13624–13629.

    https://doi.org/10.1073/pnas.1110633108

    • PubMed
    • Google Scholar
48. 1. Pingoud A
    2. Wilson GG
    3. Wende W

    (2014) Type II restriction endonucleases--a historical perspective and more

    Nucleic Acids Research 42:7489–7527.

    https://doi.org/10.1093/nar/gku447

    • PubMed
    • Google Scholar
49. Software

    1. Plata G

    (2018) Count-MAGE-Seq, version 4f827f6

    GitHub.

    https://github.com/platyias/count-MAGE-seq
50. 1. Porter CT

    (2004) The catalytic site atlas: a resource of catalytic sites and residues identified in enzymes using structural data

    Nucleic Acids Research 32:129D–133.

    https://doi.org/10.1093/nar/gkh028

    • Google Scholar
51. 1. Povolotskaya IS
    2. Kondrashov FA

    (2010) Sequence space and the ongoing expansion of the protein universe

    Nature 465:922–926.

    https://doi.org/10.1038/nature09105

    • PubMed
    • Google Scholar
52. 1. Rost B

    (1997) Protein structures sustain evolutionary drift

    Folding and Design 2:S19–S24.

    https://doi.org/10.1016/S1359-0278(97)00059-X

    • PubMed
    • Google Scholar
53. 1. Rost B

    (2002) Enzyme function less conserved than anticipated

    Journal of Molecular Biology 318:595–608.

    https://doi.org/10.1016/S0022-2836(02)00016-5

    • PubMed
    • Google Scholar
54. 1. Shendure J
    2. Akey JM

    (2015) The origins, determinants, and consequences of human mutations

    Science 349:1478–1483.

    https://doi.org/10.1126/science.aaa9119

    • PubMed
    • Google Scholar
55. 1. Sheridan PP
    2. Freeman KH
    3. Brenchley JE

    (2003) Estimated minimal divergence times of the major bacterial and archaeal phyla

    Geomicrobiology Journal 20:1–14.

    https://doi.org/10.1080/01490450303891

    • Google Scholar
56. 1. Spang A
    2. Saw JH
    3. Jørgensen SL
    4. Zaremba-Niedzwiedzka K
    5. Martijn J
    6. Lind AE
    7. van Eijk R
    8. Schleper C
    9. Guy L
    10. Ettema TJG

    (2015) Complex archaea that bridge the gap between prokaryotes and eukaryotes

    Nature 521:173–179.

    https://doi.org/10.1038/nature14447

    • PubMed
    • Google Scholar
57. 1. Stamatakis A

    (2014) RAxML version 8: a tool for phylogenetic analysis and post-analysis of large phylogenies

    Bioinformatics 30:1312–1313.

    https://doi.org/10.1093/bioinformatics/btu033

    • PubMed
    • Google Scholar
58. 1. Tajima F
    2. Nei M

    (1984) Estimation of evolutionary distance between nucleotide sequences

    Molecular Biology and Evolution 1:269–285.

    https://doi.org/10.1093/oxfordjournals.molbev.a040317

    • PubMed
    • Google Scholar
59. 1. Tatusov RL
    2. Koonin EV
    3. Lipman DJ

    (1997) A genomic perspective on protein families

    Science 278:631–637.

    https://doi.org/10.1126/science.278.5338.631

    • PubMed
    • Google Scholar
60. 1. Thiergart T
    2. Landan G
    3. Schenk M
    4. Dagan T
    5. Martin WF

    (2012) An evolutionary network of genes present in the eukaryote common ancestor polls genomes on eukaryotic and mitochondrial origin

    Genome Biology and Evolution 4:466–485.

    https://doi.org/10.1093/gbe/evs018

    • PubMed
    • Google Scholar
61. 1. Tian W
    2. Skolnick J

    (2003) How well is enzyme function conserved as a function of pairwise sequence identity?

    Journal of Molecular Biology 333:863–882.

    https://doi.org/10.1016/j.jmb.2003.08.057

    • PubMed
    • Google Scholar
62. 1. UniProt Consortium

    (2015) UniProt: a hub for protein information

    Nucleic Acids Research 43:D204–D212.

    https://doi.org/10.1093/nar/gku989

    • PubMed
    • Google Scholar
63. 1. Wang HH
    2. Isaacs FJ
    3. Carr PA
    4. Sun ZZ
    5. Xu G
    6. Forest CR
    7. Church GM

    (2009) Programming cells by multiplex genome engineering and accelerated evolution

    Nature 460:894–898.

    https://doi.org/10.1038/nature08187

    • PubMed
    • Google Scholar
64. 1. Watson JD
    2. Laskowski RA
    3. Thornton JM

    (2005) Predicting protein function from sequence and structural data

    Current Opinion in Structural Biology 15:275–284.

    https://doi.org/10.1016/j.sbi.2005.04.003

    • PubMed
    • Google Scholar
65. 1. Wilson CA
    2. Kreychman J
    3. Gerstein M

    (2000) Assessing annotation transfer for genomics: quantifying the relations between protein sequence, structure and function through traditional and probabilistic scores

    Journal of Molecular Biology 297:233–249.

    https://doi.org/10.1006/jmbi.2000.3550

    • PubMed
    • Google Scholar
66. 1. Worth CL
    2. Gong S
    3. Blundell TL

    (2009) Structural and functional constraints in the evolution of protein families

    Nature Reviews Molecular Cell Biology 10:709–720.

    https://doi.org/10.1038/nrm2762

    • PubMed
    • Google Scholar
67. 1. Wu M
    2. Scott AJ

    (2012) Phylogenomic analysis of bacterial and archaeal sequences with AMPHORA2

    Bioinformatics 28:1033–1034.

    https://doi.org/10.1093/bioinformatics/bts079

    • Google Scholar
68. 1. Yang Z
    2. Nielsen R
    3. Goldman N
    4. Pedersen AM

    (2000)

    Codon-substitution models for heterogeneous selection pressure at amino acid sites

    Genetics 155:431–449.

    • PubMed
    • Google Scholar
69. Book

    1. Yang Z

    (2006)

    Computational Molecular Evolution

    Oxford: Oxford University Press.

    • Google Scholar
70. 1. Ye Y
    2. Godzik A

    (2003) Flexible structure alignment by chaining aligned fragment pairs allowing twists

    Bioinformatics 19 Suppl 2:ii246–ii255.

    https://doi.org/10.1093/bioinformatics/btg1086

    • PubMed
    • Google Scholar
71. 1. Zhang Y
    2. Skolnick J

    (2005) TM-align: a protein structure alignment algorithm based on the TM-score

    Nucleic Acids Research 33:2302–2309.

    https://doi.org/10.1093/nar/gki524

    • PubMed
    • Google Scholar
72. Book

    1. Zuckerkandl E
    2. Pauling L

    (1965)

    Evolutionary Divergence and Convergence in Proteins

    In: Bryson V, Vogel H. J, editors. Evolving Genes and Proteins. Academic Press. pp. 97–166.

    • Google Scholar

Author details

1. Mariam M Konaté

   1. Department of Systems Biology, Columbia University, New York, United States
   2. Division of Cancer Treatment and Diagnosis, National Cancer Institute, Bethesda, United States

   Contribution

   Data curation, Software, Formal analysis, Investigation, Methodology, Writing—original draft

   Contributed equally with

   Germán Plata

   Competing interests

   No competing interests declared

2. Germán Plata

   Department of Systems Biology, Columbia University, New York, United States

   Contribution

   Data curation, Software, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Mariam M Konaté

   For correspondence

   gap2118@cumc.columbia.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-6470-7748

3. Jimin Park

   1. Department of Systems Biology, Columbia University, New York, United States
   2. Department of Pathology and Cell Biology, Columbia University, New York, United States

   Contribution

   Investigation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

4. Dinara R Usmanova

   Department of Systems Biology, Columbia University, New York, United States

   Contribution

   Formal analysis, Investigation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-5031-0013

5. Harris Wang

   1. Department of Systems Biology, Columbia University, New York, United States
   2. Department of Pathology and Cell Biology, Columbia University, New York, United States

   Contribution

   Supervision, Methodology, Project administration, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-2164-4318

6. Dennis Vitkup

   1. Department of Systems Biology, Columbia University, New York, United States
   2. Department of Biomedical Informatics, Columbia University, New York, United States

   Contribution

   Conceptualization, Supervision, Funding acquisition, Methodology, Writing—original draft, Project administration, Writing—review and editing

   For correspondence

   dv2121@cumc.columbia.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-4259-8162

• 7,018

  views

• 475

  downloads

• 24

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.