Evolution of the rTCA cycle.

Variants in the form of the rTCA cycle used within the Aquificales provide insights into the evolutionary driving forces that have shaped this pathway. Recent studies in Hydrogenobacter thermophilus, which together with A. aeolicus belongs to the Aquificaceae family within Aquificales, identified novel enzymes for several rTCA reactions. The cleavage of citrate to acetyl-CoA and oxaloacetate, and the reductive carboxylation of -ketoglutarate to isocitrate, each conventionally recognized as single enzyme reactions, are in H. thermophilus both catalyzed in two steps by distinct enzymes also found in the A. aeolicus genome (see Fig. 2) [49]–[52]. This Aquificaceae variant of the rTCA cycle has an increased degree of symmetry, and reveals previously unrecognized homology relations in enzymes, which recapitulate the similarity in the local-group chemistry of their substrates.

The newly discovered citryl-CoA synthase and -ketoglutarate biotin carboxylase enzymes catalyze local functional group chemistry that is homologous to their counterparts succinyl-CoA synthase and pyruvate biotin carboxylase (reactions 1, 1′ and 3, 3′ in Fig. 2), respectively. Moreover, both sets of homologous enzymes have high sequence similarity, and were originally annotated as the same enzymes [49], [51]. Similarly, pyruvate and -ketoglutarate ferredoxin:oxidoreductase also catalyze homologous chemistry (reactions 2, 2′), and due to high sequence similarity were again annotated as the same enzyme in A. aeolicus [11].

These observations are striking in light of the discussion on metabolic innovations in the introduction. A hypothesis of a minimal ancestral metabolism with promiscuous enzymes catalyzing homologous reactions in parallel, is consistent with a highly symmetric rTCA variant used by members of the Aquificaceae. Aquificales are the deepest-branching clade using this pathway to fix carbon [53], and as mentioned their exclusive association with hydrothermal vents may result in some of the most conservative metabolic features [4].

Indeed, the new enzymes identified in the Aquificaceae have been argued to represent the ancestral rTCA enzymes [41], [50], [54]. The single step ATP citrate lyase found in most rTCA bacteria is suggested to have arisen through a gene fusion of the second sub-unit of citryl-CoA synthase with citryl-CoA lyase [41], [49], [50], while the combination of -ketoglutarate biotin carboxylase plus isocitrate dehydrogenase (ICDH) is suggested to have been replaced by a single ICDH with increased substrate specificity [54].

To further explore the evolutionary driving forces that underly the evolution of the rTCA cycle, we used H. thermophilus enzymes as a benchmark to survey the genomes of all Aquificales, as well as other clades using the rTCA cycle. Table 2 shows the distributions of each of the newly identified high homology rTCA enzymes across all Aquificale genomes.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Distribution of rTCA enzymes in Aquificales.

https://doi.org/10.1371/journal.pone.0087950.t002

Two clear groups of Aquificales are distinguished by the rTCA variants they use. All members of the Aquificaceae possess two biotin carboxylase (BC) and two CoA synthase (CS) enzymes. By contrast, all member of the Hydrogenothermaceae and Desulfurobacteriaceae possess only single copies of these enzyme types, except S. azorense, which possess two BC enzymes. In addition, in terms of both sequence similarity and length, both sub-units of the single BC enzyme in Hydrogenothermaceae and Desulfurobacteriaceae best match pyruvate (rather than -ketoglutarate) BC in Aquificaceae. Similarly, the single CS enzyme in these families best matches succinyl-CoA synthase in Aquificaceae. Finally, all Hydrogenothermaceae and Desulfurobacteriaceae are known to possess ATP citrate lyase, while all except S. azorense possess an ICDH enzyme that is significantly larger than the version found in Aquificaceae. This difference in length is consistent with the suggestion that an increased substrate specificity of the ICDH allowed it to supplant the combined function of the BC and more primitive ICDH enzymes [54]. This survey shows how sequence length of enzymes can be a useful secondary source of information, after sequence similarity, in functional annotation. Thus, we conclude that all Aquificaceae possess the highly symmetric rTCA variant shown in Fig. 2, while other Aquificale families mostly possess the more conventional rTCA variant.

Clear evolutionary driving forces can be identified that connect the different extant forms of the rTCA cycle. The gene fusion within ATP citrate lyase increases the effective concentration of citryl-CoA in the subsequent cleavage reaction, and can thus be understood to improve the kinetics of the rTCA cycle. This gene fusion would appear to have low cost, not requiring evolution of an additional compensatory pathway to release citryl-CoA as a free intermediate, since it is not used elsewhere in metabolism.

Detailed thermodynamic analyses have in turn identified two classes of rTCA reactions that in isolation would require ATP hydrolysis to proceed: carboxylation reactions and carboxyl reduction reactions [55]. These costs can be avoided in a number of ways. Coupling carboxyl reductions to subsequent carboxylations through thioester intermediates (e.g. succinyl- or acetyl-CoA), allows the combined sequence to be driven by a single ATP hydrolysis, while coupling endergonic to subsequent exergonic reactions can eliminate the ATP cost of the combined sequence [55]. The replacement of the combination of -ketoglutarate BC and the lower-specificity ICDH with a single high-specificity ICDH falls into this latter category. The -ketoglutarate BC reaction in Aquificaceae costs 1 ATP per carbon incorporated [51], while the subsequent reduction to isocitrate is highly exergonic, resulting in a nearly-reversible sequence without ATP cost when the reactions are coupled [42], [55], [56]. This adaptation can thus be seen as driven by improving the thermodynamic efficiency of the rTCA cycle.

An additional factor that could have further increased the ATP savings of using a single higher-specificity ICDH to generate isocitrate from -ketoglutarate is that like other -ketoacids oxalosuccinate is subject to spontaneous decarboxylation. The lifetime of oxaloacetate (also a -ketoacid) in water may be estimated at about 3 min at (Figure based on and from Ref. [57].), and oxalosuccinate is more unstable. Metal ions and amines are known, at least for oxaloacetate, to further enhance the rate of spontaneous decarboxylation [57]. The decarboxylation of oxalosuccinate back to its precursor -ketoglutarate would introduce a futile cycle that may have raised the average cost of the forward carboxylation reaction above 1 ATP.

Surveys in the genomes of other clades using the rTCA cycle (data not shown) are consistent with these suggested adaptations. Chlorobiales, a later branching photoautotrophic clade in which the rTCA cycle was originally described [58], “Candidatus Nitrospira defluvii”, a member of the Nitrospirae in which the rTCA cycle was recently discovered through metagenome analysis [59], as well as Sulfurimonas, an -proteobacterial family that uses the rTCA cycle [60], all use the less-symmetric rTCA variant. Their genomic features associated with the rTCA cycle are similar to those of the Hydrogenothermaceae and Desulfurobacteriaceae. All of these species possess a gene for ATP citrate lyase, have single copies of BC and CS enzymes, and have ICDH enzymes with a length of around 730–740 amino acid residues.

Together these observations can be used to reconstruct a phylometabolic tree of rTCA variants, which in turn represents a branch of increased resolution on a more general phylometabolic tree of carbon-fixation [2]. This tree is shown in Fig. 3. The hypothesized root node, shown on the left, consists of the symmetric rTCA variant, but catalyzed by a single set of of enzymes for both arcs. Duplication and divergence toward greater specificity enzymes through selection for improved kinetics then gives rise to the symmetric rTCA variant found in Aquificaceae, while gene fusion in ATP citrate lyase combined with an increased substrate specificity of ICDH gives rise to a second divergence that leads to the conventional rTCA cycle. A comparison of ATP citrate lyase genes suggested that Aquificale families using it obtained this enzyme through HGT from -proteobacteria, and that the Chlorobiale version represents the ancestral ATP citrate lyase [41]. Thus, the initial divergence between rTCA variants may have been between Aquificales and Chlorobiales, with two of the Aquificale families later joining the conventional rTCA group through HGT, as a result of ecological association with -proteobacteria.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Phylometabolic tree showing the evolution of the rTCA cycle.

A combination of improving kinetics (which increases growth rate) and improving thermodynamics (which increases growth efficiency) explains both divergences. For the first divergence, duplication and divergence toward higher substrate specificity of enzymes improves kinetics. For the second divergence, replacing enzymes 3′ +4′ by a higher specificity version of enzyme 4 removes a cost of 1 ATP hydrolysis in fixing , while fusion of one sub-unit of enzyme 1 with enzyme for the subsequent cleavage reaction improves kinetics. Green boxes represent homologous reactions catalyzed by enzymes with high sequence similarity, while purple boxes represent homologous reactions catalyzed by members of the same enzyme families. Reactions 5′ and 5* are catalyzed by the same enzyme. Differences in sequence divergence between green and purple enzymes may reflect differences in complexity of the reactions, see text for further discussion. The yellow node represents acetyl-CoA, the blue node represents oxaloacetate, and the red node represents succinyl-CoA. Dark blue arrows indicate the direction of mass through pathways.

https://doi.org/10.1371/journal.pone.0087950.g003

While not all five homologous enzymes in the symmetric Aquificaceae rTCA variant show the same sequence similarity as the first three reactions (1–3), we argue this is consistent with the suggested evolutionary scenario. Reactions 1–3 represent thermodynamic bottleneck reaction that in isolation would require ATP hydrolysis [55], and are catalyzed by elaborate multi-subunit enzymes that contain complex metal-centers and/or complex organic cofactors, and belong to small and highly conserved enzyme families. By contrast, the hydrogenation and dehydration reactions (4, 5) that follow these complex reactions are catalyzed by simpler enzymes belonging to highly diversified enzyme families used throughout metabolism. The selection pressure to conserve enzyme sequence similarity is therefore likely to have been lower, and either (or both) of these enzymes could also have been more easily replaced by other members of this family. This breakdown between “easy” and “hard” chemistries has been recognized as a general constraint in the early evolution of carbon-fixation pathways, with innovations primarily occurring through the emergence of new sequences of easier chemistries that connect harder bottleneck reactions [1].

Having both arcs catalyzed by single sets of promiscuous enzymes in the root node would likely have resulted in slower growth for organisms using such a strategy, but it would also have considerably lowered the overhead for early regulatory machinery. If both rTCA arcs could have been catalyzed by the same set of single enzymes, the whole cycle would required only 7 total enzymes for catalysis: the 5 homologous reactions plus the reduction of malate to fumarate, and the cleavage of citryl-CoA, both of which do not have analogs in the opposite arc. It has previously been argued that a fully connected rTCA+WL network was selected as the root of carbon-fixation, because its topology would have provided the most robust form of network autocatalysis for earlier eras of life [1], [2]. The present analysis suggests that an additional reason the rTCA cycle could have been privileged as a carbon fixation pathway capable of initiating the first cellular life, over alternate autocatalytic carbon-fixation pathways observed today [53], [61], is that its greater symmetry permitted it to emerge with simpler catalytic support and regulatory structure.

Each of the divergences in the tree of rTCA variants can be understood in terms of a cost-benefit tradeoff in metabolic innovations. Duplication and divergence toward greater substrate specificity of the enzymes catalyzing the two arcs would have improved the kinetics of the pathway, and thus the growth rate of organisms using it. While the genome expansion due to these duplications would have had increased cost associated to it, this was apparently more than offset by the benefit of increasing the kinetics of a pathway through which most cellular carbon is incorporated. We will discuss a different case in the biosynthesis of branched chain amino acids in the next section, where a lower mass flux appears to have shifted this balance away from favoring duplication and divergence.

For the second divergence, the fusion of the citryl-CoA synthase and citryl-CoA lyase would have improved the kinetics of the rTCA cycle, while the elimination of the ATP dependent -ketoglutarate BC would have improved the thermodynamic efficiency of the rTCA cycle. A secondary effect of this adaptation is that it replaces a carboxylation reaction based on with one based on , potentially allowing the new phenotype to thrive in less alkaline environments. It is interesting to note that a similar replacement of pyruvate BC and malate dehydrogenase, between whih oxaloacetate is the intermediate, in the opposite rTCA arc is not observed in any organism using rTCA to fix . This may simply be because oxaloacetate is the starting precursor to a wide range of anabolic pathways, and is most easily accessed if it is released into solution, and because it is moreover produced as a free intermediate during the cleavage of citrate.

The distribution of the different innovations governing the evolution of the two extant forms of the rTCA cycle may give some insights into the tradeoffs between optimizing growth rate and growth efficiency. The combined fitness advantage of improving kinetics by fusing the citrate cleavage sequence and improving thermodynamic efficiency through the emergence of the higher-specificity ICDH appears to be significant under most conditions. Only one family (Aquificaceae) within one clade (Aquificales) still uses the ancestral rTCA variant, while the conventional form is distributed across a wide range of bacterial clades. It is further interesting to note that in the vast majority of cases the two innovations occur together. In only one known case (S. azorense) has the fusion of the citrate cleavage reaction taken place without elimination of the -ketoglutarate BC. Characterization of additional Aquificales could help us disentangle the ordering and relative advantage of the two innovations.

Branched chain amino acids.

It has long been thought that most organisms synthesize -ketobutyrate, a central precursor to isoleucine, by deaminating threonine [64]. However, an increasing number of species have been found to instead derive -ketobutyrate from pyruvate and acetyl-CoA through what is known as the “citramalate” pathway (see Fig. 4). Originally discovered, and later described in detail, in the Spirochetes [65], [66], this pathway was subsequently discovered in a range of species across the bacterial and archaeal domains. It was found to be used in members of both the Euryarcheota [67]–[70] and Crenarcheota [71], [72], as well as Firmicutes (Clostridia) [73], [74], Chloroflexi [75], Chlorobia [76], Cyanobacteria [77], and several Proteobacteria [78]–[80].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 4. Branched chain amino acid biosynthesis.

The chemical sequences show the parallels in terms of local functional group chemistry within the reconstructed ancestral pathways to valine, leucine and isoleucine. The blue box highlight the citramalate pathway of -ketobutyrate synthesis, reconstructed here to represent the ancestral sequence to this compound. The molecules highlighted in orange and green in turn show the compact interconnectedness of the ancestral pathways to the branched chain amino acids. Parallels to substrate sequences within the oxidative TCA are also highlighted, as well as the alternate route to -ketobutyrate from threonine.

https://doi.org/10.1371/journal.pone.0087950.g004

Like the observation that the ancestral form of rTCA had an increased degree of symmetry that could have allowed a smaller regulatory structure, the sequence of local functional group transformations in the citramalate pathway is repeated in the synthesis of leucine from -ketoisovalerate (see Fig. 4). In contrast to the symmetric rTCA variant where parallel reactions are catalyzed by homologous enzymes, however, in this case the parallel reactions are in fact catalyzed by the same enzymes in both pathways [66]. The only exceptions are the pair of acetyl-CoA addition reactions (5, 5′) that initiate the two sequences, which are catalyzed by separate, though still homologous enzymes. Together with the broad observed distribution, this leads us to propose that the citramalate pathway represents the ancestral pathway to isoleucine, with threonine deaminase (often called threonine dehydratase, TDH) a more recent innovation.

The genome of A. aeolicus is consistent with this hypothesis, as it lacks the gene for threonine deaminase and instead contains two genes annotated as 2-isopropyl malate synthase (IPMS, reaction 5″) [11]. To confirm the presence of the citramalate pathway, and to place this use in evolutionary context, we performed broad genomic surveys for threonine deaminase (TDH) and citramalate synthase (CMS, reaction 5′). Presence of either gene satisfies the constraint of pathway completeness in PMA, as the other genes necessary are shared between both pathways. Distributions of the two genes/pathways are shown in Table 3.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Distribution of Isoleucine biosynthesis pathways.

https://doi.org/10.1371/journal.pone.0087950.t003

We note briefly that there is some uncertainty in the annotation of CMS, because of the sequence similarity not only to IPMS, but also to homocitrate synthase of the AAA lysine synthesis pathways [70]. However, the pooling of evidence in PMA lowers these uncertainties. For example, if a strain has two genes annotated as IPMS, possesses no gene for TDH, nor any of the other genes of the AAA lysine pathway to which homocitrate synthase might connect, one of the IPMS copies most likely instead codes for CMS. In addition, BLAST search comparisons among these genes were used for cross-validation. For most species with multiple genes annotated as IPMS, BLAST searches show that at the clade level these genes separate into clear groups based on sequence similarity. For most clades, genes within one of these groups were identified in the various experimental studies mentioned above as likely encoding CMS, thus anchoring the group as a whole. Combining all these different lines of evidence leads to the distributions shown in Table 3.

Among the archaea in our study, a vast majority (98/122, or 80%) posses a gene for CMS. By contrast, less than half (58/122, or 47.5%) possess a gene for TDH. Moreover, when TDH is found it is mostly co-present with CMS, while in many cases CMS represents the exclusive pathway to isoleucine. This is relevant, because two versions of TDH are known to exist: an `anabolic' version whose activity is obligatory for cell growth, and a `catabolic' version that only becomes active during salvage of excess threonine [64]. It has previously been noted that the archaeal TDH best matches the catabolic version of this enzyme [81]. This is consistent with experimental observation that for archaea which have both TDH and CMS, the citramalate pathway represent either a major or the exclusive pathway used in synthesis of isoleucine [68], [71]. Thus, we conclude that the citramalate pathway represents the ancestral pathway to isoleucine in archaea, with TDH representing a later innovation that initially emerged for salvage purposes.

Bacteria present a more complex picture than archaea regarding the synthesis of isoleucine. Both TDH and CMS are common in deep-branching bacterial clades, with TDH occurring with higher frequency. However, this balance is dominated by Firmicutes, and among other deep-branching bacteria, the citramalate pathway appears to be the most abundant route to isoleucine. Strikingly, a significant number of deep-branching clades whose members include (hyper)thermophiles, autotrophs, or both, show near exclusive use of the citramalate pathway: Aquificales, Chlorobiales, Nitrospirae, Planctomycetes, and Verrucomicrobia.

A closer look within clades where TDH represents the majority pathway provides further context for evolutionary reconstruction. For example, within the large and diverse Firmicute phylum, the aerobic Bacillales nearly all possess the threonine pathway, while the anaerobic Clostridiales exhibit the threonine and citramalate pathways in about equal frequencies. Within the Clostridiales in turn, the Clostridiaceae family, which contains many pathogenic strains [82], make nearly exclusive use of the threonine pathway, while the Thermoanaerobacteriales family, which like the Aquificales are hyperthermophiles, make nearly exclusive use of the citramalate pathway.

Within the Cyanobacteria in turn, the majority of strains using the threonine pathway belong to the genera Synechococcus and Prochlorococcus. Many of the former and all of the latter are divergent cyanobacterial lineages highly adapted to oligotrophic regions of the world's oceans, often dominating those environments [83]–[85]. Among all other cyanobacteria, the citramalate pathway represents the major route to isoleucine. Finally, within Deinococcales-Thermales [86], which have approximately equal numbers of both pathways, the Deinococcales use mainly the threonine pathway, while the hyperthermophilic Thermales use mainly the citramalate pathway.

Lastly, it has previously been observed that many deep-branching bacteria that possess both TDH and CMS, appear to possess the catabolic version of TDH [81]. In our sample of Thermotogales and Chloroflexi, both of which show frequent occurrence of TDH, BLAST searches show that many of these genes are better matched to the catabolic than the anabolic TDH of E. coli. Thus, all evidence combined leads us to conclude that the citramalate pathway is the ancestral pathway also in bacteria, and thus for all life, with the threonine pathway initially emerging as a salvage pathway.

This conclusion is important relative to the previous discussion of a minimal ancestral rTCA cycle. In addition to the enzymes shared between the citramalate pathway and the final sequence in leucine synthesis (reactions 5–7), the enzymes catalyzing the homologous sequences in valine and isoleucine synthesis (reactions 1–3) are similarly shared across pathways, while the final amination reaction (4) is in all three pathways performed by the same enzyme [87]. While some organisms have additional copies of the first thiamin catalyzed reaction (1) for purposes of regulation [88], [89], A. aeolicus possesses only a single, two-subunit enzyme for both reactions [11]. As mentioned, the lone exception to this pattern of promiscuous catalysis is the acetyl-CoA addition reaction (5, 5′) initiating both sequences in leucine/isoleucine synthesis. However, if these reaction had not been mediated by thioester intermediates, they would represent carboxyl reduction and carbon-carbon bond forming reactions that in carbon-fixation pathways are associated with ATP hydrolysis [55]. Similar to those more constrained carbon-fixation reactions, the high sequence similarity of the two enzymes caused them to be originally annotated as the same enzyme [11], [66]. The more complex, constrained nature of these reactions may thus have increased the kinetic advantage of duplication and divergence toward greater enzyme specificity. Thus, if as before we assume an ancestral enzyme with broader substrate affinity for reactions 5/5′, then all 21 total reactions in the synthesis of valine, leucine and isoleucine starting from pyruvate and acetyl-CoA would have required only 7 enzymes for catalysis (see also Fig. 4).

In contrast to the evolution of the rTCA cycle, most enzymes catalyzing homologous reaction sequences in this case have not duplicated and diverged, except for CMS and IMPS. Why this difference? The difference in mass flux between the two pathways would appear to a offer a straightforward explanation. While most cellular carbon passes through the rTCA cycle in autotrophs using that pathway, only 3 out of 20 amino acids are generated through the pathways in Fig. 4. Thus, even if we simplistically assume roughly equal biomass partitioning between amino acids, nucleotides and lipids, the mass flux difference between the two sets of pathways is about an order of magnitude. The only enzyme for which selection pressure to improve kinetics appears to have led to duplication and divergence within the synthesis of branched chain amino acids represents a more complex, possible thermodynamic bottleneck reaction.

Emergence, and combined regulation with the rTCA cycle.

Additional overlaps between the rTCA cycle and the branched chain amino acid biosynthetic pathways may provide insights into how the latter emerged. The dehydration/hydration isomerization and dehydrogenation sequences (reactions 6a, 6b, 7) in the reconstructed ancestral pathways to leucine and isoleucine is homologous to the sequence of local group transformations occurring in the opposite direction in the large-molecule half of rTCA (reactions 4′, 5′ and 5* in Fig. 2). The associated enzymes in the two sub-systems may also have a common origin. In A. aeolicus (which uses the ancestral rTCA variant), the dehydration/hydration-isomerization reaction is catalyzed by a single subunit aconitase enzyme (ACO) in rTCA, and a two subunit enzyme (LeuC/D) in the branched chain amino acid pathways. However, the combined length of LeuC and LeuD is similar to that of ACO, while LeuC and LeuD also have high sequence similarity to consecutive, adjacent portions within ACO (data not shown). This suggest the possibility that following duplication and divergence from a common ancestral enzyme the two subunits became fused within the rTCA cycle but not the branched chain amino acid pathways, due to the differences in mass flux density. Similarly, the homologous (de)hydrogenation reactions in the two sub-systems are in A. aeolicus also catalyzed by enzymes with high sequence similarity. The enzyme homologies do not extend to the (de)carboxylation reactions across the two sub-systems, which in the direction of decarboxylation is facile, and in the direction of carboxylation requires complex cofactors and ATP hydrolysis (see previous discussions). Phylogenetic reconstructions of the lineages of these enzymes could shed additional light on these hypotheses.

These observation may thus suggest that the existence of a sequence of substrate chemistry operating in the opposite direction within the ancestral rTCA cycle could have facilitated the emergence of the ancestral pathways to leucine and isoleucine. The only truly new form of chemistry in the citramalate pathway and its homologous sequence in leucine synthesis is the initiating reaction involving ligation of acetyl-CoA (reactions 5, 5′ in Fig. 4). Moreover, the facile decarboxylation of -carboxylic acids, which in the reductive direction of the rTCA cycle may have increased the cost of an already complex reaction, may have created an advantage when used in the opposite direction, possibly further facilitating the emergence of the ancestral citramalate/leucine sequence.

The pattern of pathway diversification by innovation of the initiating reaction, followed by re-use of similar or identical downstream enzymes to catalyze homologous reaction sequences, was also the primary mode of diversification proposed for carbon-fixation pathways in Ref. [1]. The difference to those previous proposals is that here also the direction of the re-used sequence chemistry is suggested to have changed. We should also note that in comparing these innovations some caution should be exercised because the establishment of the first pathway to a set of amino acids is an innovation occurring in the era prior to LUCA, while diversification of carbon-fixation pathways occurred after LUCA. Nonetheless the parallels in the suggested modes of innovations is striking and may suggest a general principle of early metabolic evolution.

If true, this hypothesis of pathway evolution may have allowed the LUCA to regulate the combined sub-systems with an even smaller genome than we have suggested for them separately. If the homologous enzymes in the two sub-systems arose through duplication and divergence from a common ancestor, then the two sequences could have potentially been catalyzed by the same enzymes in an earlier era. We suggested above that the complete ancestral forms of rTCA and branched chain amino acid biosynthesis could in an era of more promiscuous enzymes each have been catalyzed by only 7 enzymes total. The added observations here suggest that the entire connected network of rTCA plus the pathways to the branched chain amino acids could have been catalyzed by only 12 total enzymes.

Relation to the reversal of the TCA cycle.

The key innovation of the acetyl-CoA ligation reaction that initiates the homologous sequences within leucine and isoleucine synthesis, and possibly governed their emergence, may have also facilitated the later emergence of other pathways within metabolism, in particular the reversal of direction of the TCA cycle. The entire homologous reaction sequences (5–7) in the citramalate and leucine pathways are also known as “keto acid elongation” sequences [19], and are further repeated within the oxidative TCA cycle (as well as the AAA lysine synthesis pathway). It is becoming clear that the TCA cycle originally operated in the reductive direction (see Ref. [1] and references therein for discussion), which means that keto acid elongation likely occurred within branched chain amino acid synthesis prior to its use in the oxidative TCA cycle. The thiamin-facilitated decarboxylation of pyruvate (see Fig. 4) thus similarly appears to have been used first in the synthesis of branched chain amino acids then in its use within the oxidative TCA cycle. Finally, lipoic acid is used only in the oxidative (and not the reductive) direction of the TCA cycle, while its likely ancestral function was in the synthesis of glycine through the glycine cycle [2] (see also the section on lipoic acid below for additional supporting evidence).

Thus, we suggest that at the substrate level the prior existence of the biosynthetic pathways leading to valine/leucine/isoleucine facilitated the reversal of the TCA cycle from the reductive to the oxidative direction. The additional key innovation appears to have been the recruitment of lipoic acid from the glycine cycle to its interaction with thiamin in the production of acetyl-CoA from pyruvate. Broad affinity of earlier enzymes would have aided the emergence of this novel pathway as promiscuous activity could have allowed this pathway to proceed at lower rates, with duplication and divergence later being favored as respiration came fully online and mass flux through this pathway increased.

Lipoic acid.

Lipoic acid is a cofactor with very limited, but key metabolic roles. Lipoic acid is central to the “Glycine Cleavage System” (GCS) [91], which connects glycine and serine metabolism to folate one-carbon chemistry, and is also used (as previously mentioned) in the ferredoxin:oxidoreductase decarboxylation reactions in the oxidative TCA cycle and the degradation of branched chain amino acids. The GCS is known to be reversible [91], has nearly neutral thermodynamics [55], and likely originally operated in the reductive (i.e. biosynthetic, not degradative) direction as part of the ancestral pathway leading to glycine and serine [2]. For these reasons the GCS together with serine methyl transferase (SMT) has also been called the “Glycine Cycle” [2]. The phylometabolic analysis that places reductive glycine synthesis at the base of the tree of life, as part of the phenotype of the last common ancestor, suggests that the function of lipoic acid in the glycine cycle preceded its use in either the oxidative TCA cycle or the degradation of branched chain amino acids. This provides important context for interpreting the distribution of lipoic acid biosynthesis genes that we discuss next.

Three pathways are known for the biosynthesis of lipoic acid (see Fig. 5). The conventional pathway, first described in E. coli [92], involves transfer of an octanoyl moiety from the Acyl Carrier Protein (ACP) to one of the Lipoyl Dependent (LD) enzymes, followed by sequential sulfuration of the octanoyl moiety to produce the final lipoated enzyme [93]. The first step is catalyzed by octanoyl transferase (LipB), while the second is catalyzed by Lipoyl Synthase (LipA). A variant of this scheme was recently discovered in B. subtillis, which involves the same basic chemistry, but distinct set of enzymes and an additional intermediate in the transfer of octanoate. In B. subtillis, the distinct octanoyl transferase LipM transfers octanoate from ACP to the H-protein of the GCS, followed by a second transfer (catalyzed by LipL) to the E2 subunit of pyruvate dehydrogenase [94]–[96]. Both LipM and LipL had previously been obscured due to their sequence similarities to LplA of E. coli [95]. A third distinct pathway was recently discovered in an E. coli mutant in which LipB had been deactivated. In this mutant, lipoate protein ligase (LplA), normally used in the attachment of free lipoic acid to LD enzymes, is recruited in the transfer of free octanoate to an LD enzyme through an AMP-bound intermediate [93], [97]–[99]. In light of these variations, it is noteworthy that A. aeolicus lacks a gene for LipB and is annotated as having LipA and LplA [11]. This raises the question, which of the pathway alternates does A. aeolicus use, and what does this teach us about how the biosynthetic pathways and the functionality of lipoic acid evolved?

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 5. Lipoic acid biosynthesis and lipoyl-protein assembly.

In E. coli (green sequence), octanoate is transfered from ACP to the E2 subunit of pyruvate dehydrogenase (PDH) by LipB, followed by sulfuration to lipoic acid by LipA. In E. coli mutants lacking LipB, octanoate is transfered through an alternate route with an AMP-bound intermediate by LplA, normally used for incorporation of free lipoic acid. In B. subtillis (blue sequence), octanoate is transfered from ACP first to the H-protein of GCS by LipM, followed by a second transfer to the E2 subunit of PDH by LipL. B. subtillis also uses LplJ instead of LplA for incorporation of free lipoic acid. In red is the suggested ancestral biosynthesis of lipoic acid (see main text).

https://doi.org/10.1371/journal.pone.0087950.g005

To examine this question, we performed broad genomic surveys for each of the enzymes used in lipoic acid metabolism, shown in Table 4. In addition to LipA, LipB, LplA, LipM/LipL, our survey also includes LplJ, a distinct lipoate protein ligase found in B. subtillis [95]. It can immediately be seen that the conventional LipA+LipB combination is not widely distributed across deep-branching clades. Only 6 archaeal strains, nearly all (5/6) in the Thermoproteale family within the Crenarcheota, appear to possess this pathway. Among deep-branching bacteria, only the Deinococcales, Cyanobacteria and Clostridiale family within Firmicutes show significant abundance of the LipA+LipB combination. In contrast, LplA (and its combination with LipA) is widely and more evenly distributed across both archaea and bacteria. However, BLAST searches indicate that in most of these cases LplA is in fact a better match to LipM of B. subtillis than to LplA of either E. coli or the euryarchaeota T. acidophilum [100]. While this suggests that the B. subtillis variant may be the ancestral pathway to lipoic acid, in most cases we could not find a LipL gene to accompany the putative LipM gene, presenting a puzzle as this gene is absolutely required in lipoic acid synthesis in B. subtillis [95].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Distribution of lipoic acid biosynthesis genes.

https://doi.org/10.1371/journal.pone.0087950.t004

This is where the functional roles of lipoic acid provides critical evolutionary context. As explained, the likely ancestral function of lipoic acid is its role in connecting glycine/serine metabolism to folate- chemistry through the glycine cycle, for which it remains (nearly universally) essential to this day. In contrast, the role of lipoic acid in the oxidative TCA cycle or the degradation of branched chain amino acids likely arose later, and is not essential to many organisms. For example, many autotrophs, including A. aeolicus, do not use the oxidative TCA cycle nor do they degrade branched chain amino acids. Such organisms should thus not need LipL to transfer octanoate from the H-protein of GCS to other LD enzymes, because they do not possess them. Instead, direct sulfuration of octanoate bound to the H-protein of GCS, a reaction known to be catalyzed by LipA [101], is sufficient to allow the sole function of lipoic acid in these organisms. This proposed sequence of enzyme functions is supported by the observation that in deep-branching clades that do not possess the glycine cycle – including many Clostridia, all methanogenic families within the Euryarcheota or the Desulfurobacteriaceae family within Aquificales [2] – we do not find any genes associated with lipoic acid biosynthesis or uptake. Thus, we suggest that in many cases where a putative LipM is found but LipL is absent, a pathway is used in which LipM is directly followed by LipA (see Fig. 5), and we also propose that this represents the ancestral pathway for de novo lipoic acid biosynthesis. Note that this proposal further reinforces the conclusion that the reductive TCA cycle preceded the oxidative version.

In this scenario, an additional octanoyl transferase (LipL) emerged for the second subsequent transfer to the E2 domain of PDH, possibly through duplication and divergence of LipM as seen in B. subtillis. This would have introduced a redundancy into the lipoic acid system by producing two dedicated octanoyl transferase enzymes. This redundancy then appears to have been removed by replacing the two distinct transferase enzymes with a single all-purpose LipB transferase, as seen in E. coli. The high sequence similarity among LipM, LipL, LplA and LpIJ further suggests that the environmental uptake genes likewise arose through duplication and divergence from octanoyl transferase genes, but that some of this ancestral function was maintained in LplA, making possible its recruitment in E. coli mutants lacking LipB.

Vitamin B6.

In contrast to the narrow functionality of lipoic acid, vitamin B6, which refers to pyroxidal 5-phosphate and its substitutes, is one of the most functionally diverse cofactors, with its different forms facilitating a very wide range of reaction classes [102]. Its diverse functionality and relatively simple chemistry, plausibly accessible under abiotic conditions, has led to the suggestion that it may have been one of the earliest cofactors [103], [104].

There are two known biosynthetic pathways leading to vitamin B6 in modern metabolism. In the first recognized pathway, described in E. coli, pyridoxine phosphate is derived from 4-erythrophosphate [102]. This pathway is known as the `DXP dependent' pathway, because 1-deoxy-D-xylulose-5-phosphate (DXP) is the secondary input to the final condensation reaction that produces the pyridine ring of pyridoxine. In the second (`DXP independent') pathway, first described in B. subtillis, pyridoxal phosphate is synthesized through the direct condensation of ribulose phosphate and glyceraldehyde phosphate [105], [106].

Previous analyses found genes for the DXP-independent pathway to be highly conserved and distributed across both archaeal and bacterial domains, while genes for the DXP-dependent pathway were found mainly in the -proteobacteria, suggesting this latter pathway emerged later in evolution [107], [108]. However, A. aeolicus possesses the DXP-dependent pathway, prompting us to further examine this hypothesis. Table 5 shows the distribution of the key enzymes involved in the condensation steps in both sequences – PdxA/J in the DXP-dependent pathway and PdxS/T in the DXP-independent pathway – across bacteria and archaea.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Distribution of pyroxidal phosphate synthesis genes.

https://doi.org/10.1371/journal.pone.0087950.t005

These distributions show some striking patterns. It had previously been noted that the two pathways are mutually exclusive within organisms, which use only one or the other [107]. Our analysis shows that the pathways are in fact mutually exclusive at the clade level, much more so than we have seen for pathway variants in other sub-systems we have previously analyzed. PdxA is found in a few species within both archaea and Firmicutes, but that enzyme catalyzes a hydrogenation reaction, with the enzyme catalyzing the actual ring condensation reaction (PdxJ) completely absent in those cases. Our analysis further appears to confirm that the DXP-independent pathway represents the ancestral pathway. In addition to nearly all archaea, several deep-branching bacterial clades (Thermotogales, Chloroflexi, Deinococcales, Firmicutes) also use this pathway.

It was previously suggested that the DXP-dependent pathway arose within proteobacteria [108], but the fact that all members of several deep-branching clades use this pathway suggests it may actually have been an earlier innovation. The observed distribution of both pathways is difficult to explain, however. That the pathways are mutually exclusive at the clade level requires either early HGT to progenitors of clades, extensive gene transfer between select clades after they had diverged, or extensive transfer within clades that can take the appearance of genes sweeping through the population [17]. In any of these cases there is no obvious explanation for why transfer would have been restricted to occurring only between select clades, nor is a selective advantage apparent.

An alternative explanation would be (possibly repeated) convergent evolution early in the divergences of clades. This explanation has some appeal, as the key enzymes in both pathways, PdxA/J and PdxS/T, are in fact very similar in their 3-D structure and in the sequence of local functional group transformations that make up the respective condensation reactions [102]. However, even for convergent evolution we are lacking a good explanation for why only some clades would pervasively adopt this new strategy, while others did not.

If the evolutionary sequences and driving forces are not clear, we can at least identify features of the two pathways that would have facilitated the transition between them. The shared fold structure and similarity in reaction mechanisms, but low sequence similarity, between PdxA/J and PdxS/T have been interpreted to mean that they represent convergent discoveries [102]. However, it is also possible that PdxA/J emerged from PdxS/T and that both have been under strong selection pressure, causing their sequence to diverge strongly.

Another common feature between pathways is they both start from intermediates within the pentose phosphate pathway. The key exception is DXP itself, which is the product of the first committed reaction in the DOXP pathway of terpenoid backbone synthesis. Archaea exclusively use the alternate mevalonate (MVA) pathway to synthesize terpenoids, while bacteria use both the MVA and DOXP pathways [109], possibly providing a partial explanation for why the DXP-dependent pathway to vitamin B6 emerged only within bacteria and not archaea.

Finally, the other reactions in the DXP-dependent pathway that lead up to the condensation sequence catalyzed by PdxA/J represent common and widely used metabolic reactions catalyzed by members of highly diversified enzyme families. The reaction sequence connecting 4-erythrophosphate to phospho-4-hydroxy-threonine, the input to the PdxA/J-catalyzed ring condensation sequence, consists of a hydration/reduction of an aldehyde to a carboxyl group, a subsequent dehydrogenation of an alcohol to a carbonyl group, and finally the reductive amination of that carbonyl group. Especially in an earlier era of more promiscuous catalysts, this pathway could thus well have been recruited en bloc into the emergent DXP-dependent pathway. The selective advantage of this adaptation, and the way it might have led to the peculiar distribution of both pathways, remains to be explained, however.

Quinones.

The main component of the quinone pool in A. aeolicus was determined to be 2-demethylmenaquinone-7 (DMK-7) [110]. Other Aquificales, including its close relative H. thermophilus, had previously been found to use 2-methylthiomenaquinone-7 [111]–[113]. Menaquinone (MK) has significantly lower redox potential than ubiquinone (UQ), and, based on distributions of these two quinone types both across the tree of life [114], [115] and within clades known to bridge the anaerobic-aerobic domains, UQ was suggested to have emerged with the rise of atmospheric oxygen [116]. Membrane-dissolved quinones exchange electrons directly with the fumarate/succinate redox couple, which is respectively an electron acceptor or donor depending on the direction of this reaction [114], [117]. The possible emergence of the higher potential UQ with the rise of oxygen may thus have allowed the fumarate-succinate interconversion to reverse to the oxidative direction, further facilitating reversal of the TCA cycle as a whole. Generally, whereas reduced MK is easily oxidized in the presence of oxygen, disrupting electron flow into biosynthesis, reduced UQ is stable in the presence of oxygen [116]. The redox potential of the de-methylated version of menaquinone, DMK, lies between that of MK and UQ, possibly reflecting the microaerophilic character of A. aeolicus [110].

Lipid biosynthesis.

Lipid metabolism represents the single largest sub-system within the reconstructed network of A. aeolicus, containing nearly 200 out of ∼760 reactions. As mentioned this is partly due to the fact that we explicitly represent each reaction within this sub-system, and partly due to the fact that A. aeolicus has a complex and diverse lipid composition (see Supplementary Table I) [126]. However, the elongation of all fatty acid chains is a polymerization sequence in which 2-carbon units (from acetyl-CoA) are added, and then reduced, through a repeated sequence of the same 7 reactions catalyzed by the same 8 enzymes, with only the length of the fatty acid tail away from the reaction site varying [38]. Much of the size of this sub-network thus reflects representation in the model rather than the associated genome content.

In addition, the fatty acid sub-network is further expanded due to the fact that A. aeolicus uses fatty acid chains that contain methyl groups, propyl rings, and unsaturated bonds at different positions [126]. Each of these different substitutions is introduced at a different point during the elongation process, resulting in an expanding set of intermediates that is tracked in the network prior to the output of chains of different lengths and substitutions into the final lipid assembly process.

Finally, the lipid content of A. aeolicus is unusual among bacteria for containing both phospho-ester and phospho-ether lipids [126]. In general, most bacteria use fatty acid-based ester lipids, while archaea use isoprenoid-based ether lipids [127]. The additional use of fatty acid ether lipids by A. aeolicus thus represent a sort of intermediary strategy.

Altogether lipid biosynthesis can be thought of as a compact and highly modular system that distributes 2-carbon units over a set of states of different lengths and substitution patterns that can be varied depending on environmental context. Methyl group side-chains, unsaturated bonds or cyclopropane rings can be used to modify the fluidity of the membrane, while cyclopropane rings may also be used to adapt to lower pH [128]. The linkage of fatty acids to the the glycerol backbone can in turn be varied between ether or ester bonds to modify the permeability of the membrane [129]. Apart from basic inputs and final assembly (isoprenoids vs. fatty acids, ethers vs. esters), the regulation of lengths and substitution patterns appears to be the main factors permitting wide variability in lipid composition. The diverse and varied composition of A. aeolicus lipids, including both ester and ether linkages, appears to reflect the “stressed” hyperthermophilic conditions of the hydrothermal vents and springs where it lives.

Aquifex aeolicus as a model system for early metabolic evolution.

We have used phylometabolic analysis to reconstruct the whole-genome metabolic network of A. aeolicus, and have shown that it uses the likely ancestral pathways within many sub-systems. Perhaps even more important than whether this organism represents the “most ancestral” known metabolism or not, is that comparing the evolutionary trajectories of its sub-systems to those of other later branching bacteria often identifies the direction of evolutionary change going form the earliest cells to modern forms. This can help us infer properties of those earliest cells, even if the evolutionary change has already proceeded in A. aeolicus.

For example, our analyses have shown that A. aeolicus uses variants of both the rTCA cycle and the branched chain amino acid biosynthesis pathways reconstructed to be more ancestral than those used by other organism, which make greater use of parallel sequences of homologous local functional group chemistry. Combined with a general observation that selection appears to optimize the kinetics of enzymes by increasing their substrate specificity, this allows us to to hypothesize that ancestral life forms could have regulated those sub-networks with more promiscuous enzymes and a smaller genome, with evolution then producing the forms used by A. aeolicus.

There are indications that evolution may have modified some A. aeolicus metabolic sub-networks to such a degree that they are not closer to the ancestral form of those sub-network than other extant forms. Particularly notable among these is the apparent obligatory use of oxygen as the terminal electron acceptor, despite the fact that it is generally accepted that oxygen did not rise significantly in Earth's atmosphere until around 2.5 GYa [150], [151]. Rather than using this single observation to place the Aquificales as a late-branching clade that arose during or after the rise of oxygen, we instead argue that all evidence combined suggests that the use of oxygen is a later adaptation within energy metabolism that emerged in a sub-set of Aquificales.

Several observations support this interpretation. Many Aquificales do not use oxygen, instead using a fully anoxic metabolism relying on nitrate, elemental sulfur or thiosulfate as their terminal electron acceptors, while yet others can switch between anaerobic and microaerophile lifestyles [16], [136]. Aquifex pyrophilus, the closest known relative to A. aeolicus, can use both nitrate and oxygen as its terminal electron acceptor [152], which as discussed before remains a possibility in A. aeolicus as well. As also discussed, A. aeolicus can use various sulfur compounds as intermediate electron acceptors, but not as the terminal acceptor, possibly because of the tight coupling between the components of its respiratory super-complexes. Finally, the rTCA cycle has several oxygen-sensitive components, and all organisms using this carbon-fixation pathway (including all members of the Aquificales, Chlorobiales, -proteobacteria and Nitrospirae) are anaerobes or microaerophiles [53].

Altogether these observations lead to a plausible evolutionary scenario of Aquificale energy metabolism, from a fully anaerobic ancestral state toward diverse forms that cover the anaerobe-microaerophile spectrum. In this scenario, A. aeolicus could have sequentially added sub-components, with the oxygen handling parts added later, essentially encapsulating an otherwise anoxic chain within an oxygen-using chain. However, unlike most other Aquificales that also use oxygen, A. aeolicus seems to have lost the ability to use other intermediates within its electron transport chains as the terminal acceptor, possibly because of the tight integration of the respiratory chains (or simply because the right combination of experimental parameters have not yet been found). If this hypothesis is correct, then comparative genomic and experimental analyses of Aquificale energy metabolisms remains a key area of research for understanding the sequences of innovations that drove their evolutionary diversification.

There is likely no single organism today that best represents the earliest cells on earth, because all cells alive today are the product of a several billion year evolutionary process. However, the discussion about oxygen use in A. aeolicus was partly meant to highlight that different sub-components and sub-networks within bacterial (or archaeal) lineages may evolve independently from the rest of the system in response to specific selective pressures in the environment. Thus, some organisms may give more insights into the earliest forms of life, because their environments have changed less over the course of Earth history, and because they retain a larger number of sub-components that resemble reconstructed ancestral forms of those sub-components. In this regard we suggest that A. aeolicus is as good as any, and better than most, as a model system for studying the earliest metabolic evolution of life on Earth.

General principles of metabolic evolution.

By reconstructing the evolutionary sequences among pathways in A. aeolicus, we identified several general driving forces of metabolic evolution. We have highlighted throughout the way selection for improved kinetics and/or improved thermodynamics can shape a metabolic network. Comparing different sub-networks further reveals tradeoffs in the costs versus the benefits of these innovations, which appears to depend strongly on the relative mass flux density of sub-networks. Extending these analyses to other metabolic sub-networks, and to the evolutionary history of other organisms will improve our understanding of how tradeoffs between performance gains and their associated costs generally contributed to fitness in the earliest stages of cellular life.