Data

We downloaded from the KEGG (release version 50) (29) and MetaCyc (release version 15.1) (30) databases the list of proteins with annotated catalytic activity. Enzymes annotated with more than one EC number, a basic definition of enzyme promiscuity, correspond to ∼2% of the total (Table 1). Annotations corresponding to partial EC numbers were not considered in the study. Redundancy in the KEGG database was removed within each set at a threshold of 90% of sequence similarity. In our study, we were interested in those sequences from the data set belonging to organisms in the reference tree of life (see definition below). Additionally, in order to apply our analysis of chemical similarity, the information in KEGG about the biochemical reactions that catalyze the enzymes was collected. Finally, we obtained 152,041 enzyme sequences from KEGG verifying simultaneously these specifications (Table 1). In the case of MetaCyc, we were able to identify 1080 entries verifying these specifications (Table 1).

We used as the reference tree of life the phylogenetic tree reconstructed by Ciccarelliet al. (31) from a multiple alignment of 31 DNA orthologs occurring in 191 organisms with sequenced genome. In this set, data were curated by these authors in order to avoid horizontal gene transfer events, and phylogenetic inferences were made by maximum likelihood and the JTT substitution matrix by using PHYML (32). In our study, we grouped together branches of the same genus in this reference tree, resulting in 108 species (seesupplemental Table S1). The apparent speed of sequence evolution can be measured as the cumulative branch length distance from a species to the root, being the root placed through automatic midpoint rooting. The choice of such a rooting is not a problem for the purpose of our study because midpoint rooting falls somewhere between the three domains, and our results are actually based on what happens within each of them. In the present paper, we refer only to relative divergence times (i.e. the ordering of divergence times of taxonomic groups), not to absolute ones. Divergence times of broad taxonomic groups, when considered, are those given in Ref.33.

To link enzyme and phylogenetic information, organisms in the databases were mapped into the tree of life. Different strains from the same species were assigned to the same organism. When several organisms were matched to the same taxon in the tree of life, values were averaged. We were able to assign 108 species in the tree of life to 356 organisms in KEGG (33%) and 120 in MetaCyc (11%) (SeeTable 2 andsupplemental Tables S1 and S2 for details).

Evaluation of Enzyme Promiscuity

The term “enzyme promiscuity” is used throughout to refer to the general concept of the ability of an enzyme to process multiple substrates or reactions. Furthermore, we introduce more specific definitions of enzyme promiscuity inTable 3 at the enzyme level by looking at the chemical diversity of the reactions that the enzyme can process, the enzyme chemical diversity, and at the catalytic activity level by looking at the list of reactions that can be simultaneously processed by enzymes annotated for this activity, the enzyme latent promiscuity.

First, we evaluated enzyme chemical diversity by computing the average chemical dissimilarity between the reactions in the list of biochemical reactions annotated for the enzymes in the databases. Chemical similarity was evaluated for pairs of reactionsR_A andR_B using the well known Tanimoto coefficientT_c(R_A,R_B), a number ranging between 0 (dissimilar reactions) and 1 (identical reactions). In this study,T_c was evaluated by using the molecular signature of the reactions (28). Molecular signatures are vectors whose components correspond to canonical representations of the graph associated with molecular structures (27). IfG = (V,E) is a molecular graph, where verticesV correspond to atoms and edgesE correspond to bonds, then the molecular signature ofG is given byEquation 1,

where σ(x) represents the atomic signature ofG rooted at atomx. The molecular signature of reactionR,

wheres_i andp_i are the stoichiometric coefficients of substratesS_i and productsP_j, is defined by the difference between the signatures of products and substrates,

Based on the previous definition, reaction chemical similarity between the pair of reactionsR_A andR_B is given byEquation 3,

where the chemical points represent the dot product between the signatures.

Second, our definition of latent enzyme promiscuity for one EC number was given by the number of different activities where enzymes annotated with this activity have been observed to participate. It was computed as follows. We listed all enzymes that were annotated with a given EC number, and we counted the total number of different EC numbers where these enzymes have been annotated.

Evaluation of Phylogenetic Diversity

The phylogenetic distance between two specieso_i ando_j was computed as the intertaxa branch length distance δ(o_i,o_j) of the two taxa within the tree of life (31); this is generally called a patristic distance. This value corresponds to their average sequence divergence in the multiple alignment of 31 orthologs (31). Phylogenetic diversity of a set of organisms was computed as in (34), by averaging the patristic or taxonomic distance between individual organisms. Phylogenetic diversity of a reaction or EC number was estimated by computing the phylogenetic diversity of organisms annotated with this reaction or catalytic function.

Catalytic Function Classification and Efficiency Estimation

We used two types of functional classification for enzymes, one based on pathway modules and another on ortholog groups. KEGG pathway modules for each catalytic function (EC number) were downloaded from the KEGG site. Cluster of ortholog groups inEscherichia coli were downloaded from the NCBI, National Institutes of Health, Web site. Catalytic efficiency was computed from the BRENDA database (35) (seeTable 4). For each EC number, we stored the values ofK_m andK_cat of those reactions involving native enzymes (i.e. excluding mutants) where we were able to match substrate information with KEGG by means of the following procedure. For each entry in Brenda corresponding to an enzyme, we went through the list of substrates processed by this enzyme, flagging the entry as consistent if the same substrate was annotated in KEGG for the EC number that corresponds to the enzyme. Finally, we collected all available catalytic efficiency values in the consistent entries.

Reaction and Annotation Level Enzyme Promiscuity

To characterize enzyme promiscuity, our basic definition is given by the chemical diversity found among the reactions that an enzyme can process. Typically, the number of biochemical reactions annotated for an enzyme is higher than the number of catalytic functions according to their EC classification because enzymes can use the same catalytic activity to process more than one substrate (supplemental Fig. S1). Therefore, in order to get a more accurate estimate of the enzyme catalytic capabilities, enzyme promiscuity at the reaction level was measured by computing the average chemical dissimilarity (seeTable 3) between the reactions in the list of biochemical reactions annotated for each enzyme in the KEGG (29) or MetaCyc database (30). Contrarily to the two previous parameters (number of annotated reactions and EC numbers), our definition based on reaction chemical diversity provides an overall estimate of enzymatic capabilities and a quantitative characterization of enzyme promiscuity, which is less likely to be affected by the bias arising from redundant annotations or arbitrary/inconsistent classifications (36). In fact, it is not always the case that an enzyme annotated with a large number of reactions is able to process chemically more diverse substrates (seesupplemental Fig. S2). The number of substrates, however, might be underestimated when they are labeled uniquely in generic terms, such as “alcohol” or “dialkyl ketone,” an issue that was found in 33 EC numbers (4.3% of total ECs in the data set) in our data set. Of these, only 15 EC numbers (1.9% of total ECs in the data set) were annotated uniquely with one generic substrate. The effect of the presence of this small set of enzymes mislabeled as non-promiscuous, thus, is not going to alter significantly any general trend observed in the results.

To further characterize enzyme promiscuity, our second definition is at the EC number annotation level, which might be more properly defined aslatent enzyme promiscuity because its definition is the number of different activities where enzymes have been annotated simultaneously with a given EC number. Therefore, this value is a measure of the total number of potential catalytic activities that an enzyme can process, based on the different functions that have been observed within its orthologs across the species.

Both definitions of promiscuity at the reaction and annotation activity level are interrelated because enzymes with the ability of catalyzing a greater chemical diversity are potentially more likely to be involved in a greater number of catalytic functions across organisms. On average, we observed this trend (r = 0.77,p = 2.6 × 10⁻²) (seeFig. 1A). In some cases, however, enzymes annotated with one EC number, and therefore with no latent promiscuity, are actually able to process a significant number of chemically diverse substrates, a fact that could be related to the EC number definition. For example, aspartate aminotransferases (EC 2.6.1.1), which are enzymes with no latent promiscuity according to KEGG annotation, can potentially process tyrosine, phenylanine, and trytophan as well (37), making its reaction chemical diversity 0.51. This value is one of the highest in the group of EC numbers with no latent promiscuity. Therefore, reaction level chemical diversity is in general a more accurate promiscuity measure than EC number annotation level.

To examine how both of our definitions of promiscuity relate to the definition in Ref.18, which is based on EC difference in digits, we plot inFig. 1B for each enzyme the relationship between EC number dissimilarity and latent promiscuity. As it is shown in the plot, higher values of latent enzyme promiscuity usually correspond to a greater dissimilarity in digits, whereas lower values appear more scattered according to their EC number dissimilarity. Nevertheless, both measures are highly correlated (r = 0.85,p = 3.5 × 10⁻³).

Phylogenetic Distance in Trees of Life and Enzyme Promiscuity

To perform our study of origins of enzyme promiscuity, we need an unbiased and comprehensive data set of phylogenetic and metabolic data. Phylogenetic diversity of any subset in a list of species might be estimated once a reference tree of life is established. As described under “Experimental Procedures,” we took here as a reference the tree in Ref.31, where genes in 191 species (seeTable 1) with completely annotated genomes were carefully selected in order to minimize horizontal gene transfer effects. The number of species was further reduced to 108 when branches of the same genus were grouped together (seesupplemental Tables S1 and S2). In order to estimate phylogenetic diversity, we took two approaches: tree distances calculated from sequence data following the procedure in the reference tree (31) and a tree based solely on evolutionary relationships through the use of phylogenetic systematics (38) (see dendrograms insupplementary Fig. S3).

The reason for using two reference trees in this study is motivated by the fact that patristic distances (sum of the length of the branches) extracted from sequence alignment data should be used with caution. We hope that such pairwise distances will reflect phylogenetic diversity; however, some species with a much higher rate of DNA change than others will bias the estimation by having a higher terminal branch length. To limit the risk of facing such bias, results are presented in this section comparatively for the estimation of phylogenetic diversity using pairwise patristic distances on the reference tree based on sequence data and for the estimation of phylogenetic diversity using a purely taxonomic (or topological) approach by counting the number of nodes separating each pair of species in the reference tree based on evolutionary relationships (i.e. by looking at the tree without its branch lengths).

To investigate how effects associated with the sequence data selection, such as the unequal rate of change among selected orthologs, the effects from horizontal gene transfer events (39,40), and species sampling (41), might be introducing some bias in the tree of life, we evaluated the distribution of phylogenetic distances, finding that some taxa such as “prokaryotes” were overrepresented in our reference tree due to their greater availability of sequenced genomes. In fact, we observed that the distribution of phylogenetic distances (patristic distances from a maximum likelihood tree calculated from aligned DNA sequences) between organisms in the reference tree (see “Experimental Procedures” for definitions) follows a bimodal distribution (seesupplemental Fig. S4), illustrating the fact that some phyla are overrepresented in the tree, as is the case for proteobacteria and firmicutes (seeTable 1). From this form of distribution, we might expect that phylogenetic distances between pairs in a random set of species would not monotonically increase with sample size (seesupplemental Fig. S5). Therefore, we can assume for this study that obtaining higher average phylogenetic distances can be properly interpreted as greater phylogenetic diversity or spread of species across the tree of life rather than a bias coming from a tree artifact.

To further evaluate the effect of investigation bias, we relied on the BRENDA database (35), which is one of the most comprehensive resources on experimental enzymatic data. One can hypothesize that enzymes present in many organisms might have been investigated in more detail, which could have led consequently to its annotation with multiple reactions. We assumed that the number of entries in Brenda for a given EC number provides a good estimate of the extent of the overall number of experimental studies performed on the enzyme. On the other hand, the presence of the enzyme in multiple organisms can be quantified, as previously, by computing the average phylogenetic distance between all organisms annotated with this enzyme. In our tests, the comparison between these two measures revealed no significant correlation, suggesting that the number of experimental studies performed on enzymes is not directly related to their phylogenetic extent in the tree of life, as can be seen insupplemental Fig. S6,A andB. This result rules out a potential artifact in our observations due to investigation bias.

To test the hypothesis that those catalytic functions in enzymes with greater chemical diversity are to be found and shared across more distant species in the tree of life and therefore come from deeper ancestral enzymes in the tree of life, we measured the chemical diversity of an enzyme by using the promiscuity definitions previously introduced. That hypothesis is preferring an ACCTRAN optimization of homoplasies, favoring reversions over convergences; indeed, part of the tested hypothesis is that losing chemical diversity by progressive specialization is easier than having greater (twice or more) chemical diversity formed by ancestral enzymatic reactions. Conversely, enzymatic reactions that process specific substrates might be expected to be found usually confined within groups of more closely related species. Because catalytic specialization is an emergent property driven by evolution, based on our hypothesis, it should be observed at both enzyme and catalytic activity levels of promiscuity.

Similarly to the chemical distance between the substrates for a reaction pair, we measured the phylogenetic diversity of the reaction pair, which we define as the average phylogenetic distance between all of the organisms that can process that pair of reactions. InFig. 2A, we plot the relationship between the reaction chemical diversity and reaction phylogenetic diversity of enzymes in our set. Interestingly, we found a significant high correlation. Therefore, this result suggests that the capability of an enzyme to process more than one substrate is intimately related to its evolvability. Reactions that are catalyzed by a highly specific enzyme are shared by poorly divergent species (i.e. specific taxa in the tree of life), suggesting a later emergence of this function during evolution. In contrast, promiscuous enzymes are more likely to appear uniformly distributed across species in the tree of life.

The conclusion does not change significantly according to the type of distance used to calculate species pairwise distances in the reaction phylogenetic diversity (Fig. 2,A andB); the correlation coefficient isr = 0.86,p = 1.3 × 10⁻³ when using patristic distances of a tree based on sequence data andr = 0.83,p = 3.0 × 10⁻³ when using taxonomic distances in the tree, showing that the first estimation of the correlation was not affected by too strong inequalities in DNA rate of change among lineages.

Similar results were obtained inFig. 2C at annotation level (r = 0.81,p = 1.4 × 10⁻²). In this case, we studied the relationship between all potential activities and phylogenetic distance of EC numbers, which was computed as the average distance between the species annotated with this particular EC number. This common trend reveals that not only are biochemical reactions found across more diverse species more likely to be processed by promiscuous enzymes, but also catalytic functions present in a greater diversity of species are potentially more promiscuous.

To evaluate the consistency of our previous results, which were based on KEGG, we recomputed them on MetaCyc, which has a higher curation level in terms of enzyme function annotation. For instance, although latent enzyme promiscuity values were found similar for both databases (correlation between promiscuity valuesr = 0.88,p = 7.1 × 10⁻⁴, shown insupplemental Fig. S7), there were nevertheless specific cases where EC numbers that appeared in KEGG as non-promiscuous were found in MetaCyc annotated as promiscuous. This result is due to the fact that KEGG annotations are performed, in their majority, by automated methods. Therefore, KEGG annotations might have overlooked some of the species-specific enzymatic functions of the multifunctional proteins present in the tree of life, a problem that could lead our study to underestimate the actual distribution of enzymatic functions across species in the tree. Thus, to evaluate any bias arising from the way KEGG is annotated, our previous tests were repeated on the MetaCyc database. The data set that we were able to process from MetaCyc was, in turn, considerably smaller than the one from KEGG. This difference in size is due to the fact that MetaCyc does not contain complete genomes as KEGG does but rather a small number of highly curated reference enzymes for each biochemical activity. The results nonetheless followed the same trend as before, despite the way enzyme promiscuity was measured either from the chemical diversity of their reactions (r = 0.74,p = 1.5 × 10⁻²;supplemental Fig. S8) or as latent promiscuity (r = 0.94,p = 4.7 × 10⁻⁴;supplemental Fig. S9). The fact that the results from the highly curated MetaCyc database closely mirror the results from KEGG provides some confirmation that the larger but much less accurate KEGG data set is providing meaningful results.

Chemical Diversity and Latent Enzyme Promiscuity across Species

Our previous findings show a close relationship between reaction chemical diversity and evolutionary time as measured through phylogenetic divergence. Based on this result, we might expect to see some distinctive shift in species with highly specialized catalytic activities. In fact, using our measure of reaction chemical diversity, it is possible to parallel successive divergence times of groups in the tree of life of species with metabolic annotations. InFig. 3,A andB, reaction chemical diversity of enzymes was averaged and plotted for each organism and taxonomic group.Fig. 3,C andD, shows the results obtained by using latent enzyme promiscuity. Interestingly, taxonomic groups going from less to more specialized catalytic function appear in the following order archaea, bacteria, fungi, plants, and animals, which is the order of divergence times of each group (33), illustrating the fact that groups of later divergence in evolution are in general more specialized in their catalytic functions. We found that the separation between phylogenetic groups Archaea, Eubacteria, and Eukaryota, represented inFig. 4,A andB, is significant in both latent enzyme promiscuity and reaction chemical diversity (p < 2.2 × 10⁻¹⁶ for a multivariate analysis of variance test).

Reaction chemical diversity and latent enzyme promiscuity together provide an account of how catalytic capabilities of organisms have evolved and adapted to their environments. For instance, euryarchaeota, which can both process many substrates and contain high latent promiscuity, are an example of extremophiles, which can survive under extreme conditions and therefore need a highly adaptable metabolic system. The same goes for deinococcus-thermus, which are a small group of bacteria highly resistant to environmental hazards. To generalize an evolutionary interpretation of the plot (Fig. 4,A andB), we must keep in mind that all species here are today's products of genetic lineages that do have the same evolutionary time since the origins of life. However, those species do not face the same chemical changes in their environments, and their position in the plot could also reflect past adaptive pressures. We infer that organisms showing high latent enzyme promiscuity and high reaction chemical diversity (Fig. 4,A andB,top right) must be pioneers that have to face strong chemical variations in their recently colonized environments. Organisms showing low latent enzyme promiscuity and high reaction chemical diversity (Fig. 4,A andB,top left) must be “old warriors” that have been accustomed to facing strong environmental changes since long ago. If the data set can be considered as “transfer-free,” the spirochaetes and tenericutes are “old warriors” in the sense that their ancestors might have faced very strong environmental changes in a recent past and therefore they still kept a high reaction chemical diversity but do not have to face strong environmental pressure anymore. In other words, they are not in a pioneering situation anymore or are less so than in their recent past. Similarly, organisms exhibiting high latent enzyme promiscuity and low reaction chemical diversity (Fig. 4,A andB,bottom right) must be organisms that recently arrived in stable chemical conditions, like, for instance, recent parasites, commensals, or symbionts, such as those observed in actinobacteria, a bacteria with mutualistic tendencies (42). Last, organisms with low latent enzyme promiscuity and low reaction chemical diversity (Fig. 4,A andB,bottom left) must be highly integrated organisms. Pluricellularity, for instance, is buffering the genetic/enzymatic effects of environmental chemical changes. Moreover, multicellular mobile organisms have the choice to rapidly escape from chemical hazards and actively reach chemical optima. It is not surprising to find animals in that area of the plot. Plants are multicellular organisms; however, they cannot move and escape chemical aggression. This may be why we find them at the middle of the plot, with moderate latent enzyme promiscuity and moderate reaction chemical diversity.

Metabolic Functions of Promiscuous Enzymes

We did not observe a link between enzyme promiscuity and gene essentiality (3,43) (i.e. with genes that are absolutely required for cell survival). However, analysis of metabolic functions found in promiscuous enzymes, given inTable 5, revealed that catalytic activities with a higher degree of latent promiscuity (≥5) are mainly involved in amino acid (p = 1.7 × 10⁻²) and lipid metabolism (p = 1.9 × 10⁻²). This fact was seen both in the whole data set using the functional classification of pathways where they are involved (KEGG modules) (44) and using the NCBI clusters of ortholog groups for enzyme sequences inE. coli, shown inFig. 5,A andB), respectively. Based on our previous finding of a greater phylogenetic extent of promiscuous enzymes, amino acid and lipid metabolisms might be associated here with the earliest form of biochemical reactions, which still are being processed by multifunctional enzymes (45,46). It is not surprising; free amino acids are found in abiotic environments like interstellar meteorites and ice (47,48). They must have been one of the very first complex organic compounds (i.e. with all main atomic species of life: carbon, hydrogen, oxygen, nitrogen, and sulfur) already available to the earliest protobionts and enzymatic functions). Enzymes performing amino acid catabolism and anabolism must have been the very first ones in life. Conversely, metabolisms of glycans (p = 7.4 × 10⁻³) and secondary metabolites (p = 4.4 × 10⁻¹) are catalyzed by enzymes with a low degree of latent promiscuity, probably therefore of more recent origin. Their greater specificity might be linked to the fact that secondary metabolism is generally associated with defense mechanisms specific to organisms, and glycan pathways are usually associated with multicellular developmental processes and restricted to the metazoa (43).

Catalytic Efficiency of Promiscuous Enzymes

We used our definition of latent promiscuity in order to investigate the previously reported experimental observation that enhancement of latent catalytic activities does not seem often to trade off with parallel decreases in the original function (16,24), in contradiction with the commonly accepted assumption that broad substrate acceptance generally comes at the price of low reaction turnover numbers (19). In our results, catalytic efficiency, estimated from experimental data as described under “Experimental Procedures,” does not seem to be related in general to the latent promiscuity of enzymes because the trend that we observed remains flat for most of the values of latent enzyme promiscuity (r = 0.02,p = 0.27; seesupplemental Fig. S10). This result, thus, corroborates the observed fact (16) that enzyme promiscuity can be in many cases achieved without compromising efficiency, suggesting that although there is a trend in evolution toward specialization, catalytic efficiency has remained on average at the same level during the history of life, perhaps due to energy constraints.