GshA sequences fall into three distinct groups

We assembled an initial set of GshA sequences by searching the NCBI Protein database [15] using either GshA or glutamate-cysteine ligase as query words. The set was expanded using the output of BLAST [16] searches with the sequences from Escherichia coli, Saccharomyces cerevisiae and Arabidopsis thaliana. Some of the sequences found in the BLAST searches correspond to hypothetical proteins whose functions have not been experimentally verified. For organisms that are known to synthesize GSH, these sequences are likely to encode GshA. However, the assignment of function is uncertain for the sequences from Mycobacterium tuberculosis, Streptomyces coelicolor and Clostridium acetobutylicum because these organisms are not known to synthesize GSH [4,17].

The lack of significant overall sequence identity between the members of these groups could indicate that these proteins arrived at a common function by convergent evolution from different progenitors, or that the sequences have diverged so far that evolutionary relationships are no longer readily apparent. For truly related but very divergent proteins, it is often possible to identify locally conserved regions that are important for structure and/or function. Therefore, our next approach was to search for such motifs using Block Maker [19,20]. (It would be ideal to search for motifs with a divergent set of proteins in which the pairwise identities are less than about 40%, so that the motifs found are highly conserved because they are important for structure and/or function. However, the high pairwise identities between plant and alpha-proteobacterial sequences in group 3 made this impossible.) Using a query set containing six group 1 sequences, seven group 2 sequences, seven group 3 sequences, and four cyanobacterial sequences, Block Maker identified three blocks of conserved sequence (Figure 3). A search of the non-redundant database using a PSSM (position-specific scoring matrix) generated from these blocks retrieved a total of 81 sequences with E-values less than 10. No proteins with identified functions other than GshA were found with E-values less than 7. Thus, the PSSM was very effective at discriminating between GshA and non-GshA sequences in the database. Many of the sequences were duplicate entries of a single sequence or sequences from multiple strains of the same organism. Taking this into account, the search retrieved 44 sequences of known or putative GshAs with E-values less than 1. A subset of these sequences representing all three groups and cyanobacteria is shown in Table 2. The p-values for each occurrence of the three motifs, as well as the E-values for each sequence, are very low in all three groups. These data provide support for a real but distant relationship between the three groups of sequences, as well as the cyanobacterial sequences.

It is difficult to generate reliable phylogenetic trees in cases such as this in which only small blocks of conserved sequence can be identified. Figure 4 shows a tree generated from the three conserved blocks using the parsimony method in PAUP 4.0b [21]. The tree topology supports the findings reported above in that the sequences cluster into three distinct groups, with the cyanobacterial sequences being most closely associated with the group 3 (plant and alpha-proteobacteria) sequences. The available information does not provide reliable information about the order of branching within each of these groups. Similar findings were obtained using the neighbor-joining method in PAUP. A general association of eukaryotic and alpha-proteobacterial sequences that would support the hypothesis that the eukaryotic gene arose by transfer from the mitochondrial progenitor is not seen. The striking association of plant sequences with alpha-proteobacterial sequences will be addressed below.

Possible explanations for the distribution of GshA: lateral transfer versus massive gene loss

Figure 5 shows the distribution of known and putative GshA genes mapped onto a universal tree of life generated using rRNA sequences. Two possible scenarios could account for the observed pattern of distribution. First, a GshA gene might have arisen early in either the bacterial or eukaryotic lineage and undergone a lateral transfer to the other domain. Alternatively, a GshA gene could have been present in the last common ancestor, but have been lost in nearly all Archaea, the deepest-branching eukaryotes and many bacteria. Although the possibility of massive loss of GshA at first glance seems unlikely, careful perusal of Figures 1 and 5 reveals that most of the genera that lack GSH and/or GshA lack aerobic metabolism and therefore do not need GSH for protection against reactive oxygen species. Many Archaea (all of the methanogens, as well as Archaeoglobus, Pyrodictium, Thermoproteus and Thermofilum) are anaerobes. Thermoplasma, although facultatively aerobic, respires sulfur rather than oxygen. The deepest-branching eukaryotes (Giardia, Trichomonas, Vairimorpha and Encephalitozoon) lack mitochondria, and therefore aerobic metabolism. Entamoeba histolytica has apparently lost its mitochondria and now lacks aerobic respiratory pathways. Bacteria such as Aquifex and Thermotoga are anaerobes. Thus, the possibility of massive gene loss in these lineages is quite plausible. However, a problem with this hypothesis is that it requires that the GshA gene, which appears to be important only to aerobes, must have originated in the last common ancestor before the emergence of O_2, which occurred only after the evolution of cyanobacteria about 2.6 billion years ago [22].

The possibility that GshA arose early in the bacterial lineage and then was transferred to the eukaryotic lineage and into at least one archaeon (Halobacterium) is perhaps more appealing. GshA may have arisen in cyanobacteria concomitant with the origin of oxygenic photosynthesis, a particularly attractive hypothesis because this would have been the first time protection against reactive oxygen species would have been important. Later, as proteobacteria developed aerobic metabolic processes that took advantage of the growing O₂ concentration in the atmosphere, the consequent production of reactive oxygen species would have given a selective advantage to microbes that acquired a GshA gene by lateral gene transfer from cyanobacteria. GshA would have been advantageous to aerobic eukaryotes as well, providing selective pressure for the acquisition and retention of a bacterial GshA gene. Transfer of a GshA gene from the alpha-proteobacterial progenitor of mitochondria, as postulated previously [8,11,12], is one possible mechanism for the spread of GshA into eukaryotes. Our analysis does not show a significant association between eukaryotic and alpha-proteobacterial sequences that would support this mechanism (Figure 4). However, the extreme divergence of the sequences limits our ability to resolve phylogenetic relationships, so this mechanism remains a possibility. There are other possible mechanisms for an early trans-domain lateral gene transfer, as well. For example, Doolittle has proposed that protists that consume bacteria as food may incorporate bacterial genes into their genomes [23].

Lateral transfer of GshA genes

Lateral gene transfer from alpha-proteobacteria to plants is particularly intriguing because transfer and retention of a foreign gene in a sophisticated multicellular organism is more difficult than in bacteria. Bacteria can acquire DNA from their environment in multiple ways (transformation, transduction and conjugation) [26]. Furthermore, a transferred gene can be easily transmitted to progeny after recombination into genomic or plasmid DNA. However, known mechanisms for transfer of DNA into plants are more limited. The best understood mechanism is the transfer of T-strand DNA from the Ti-plasmid of Agrobacterium tumefaciens into wounded plant tissues, a process resulting in the formation of tumors [30]. It is not known whether foreign genes can be transferred into plants by this mechanism in nature, but such a process is plausible. Perpetuation of a transferred gene is also not as easily achieved in seed plants as it is in bacteria, because the gene must be incorporated into genomic DNA in apical meristem cells, undifferentiated stem cells that produce new organs, including the cones or flowers that generate male and female gametes. An interesting issue is whether the group 3 alpha-proteobacterial gene displaced an ancestral group 2 eukaryotic gene, or whether the ancestral gene was first lost, allowing the alpha-proteobacterial gene to fill the functional gap.

Additional potential cases of lateral gene transfer are also suggested by our data. GshA from Xylella fastidiosa, a gamma-proteobacterium, clusters with group 3, rather than with the other gamma-proteobacterial sequences in group 1. This organism is a plant pathogen, and its physical association with plants has apparently provided an opportunity for gene transfer, either from a plant or, more likely, from an associated alpha-proteobacterium. Sequences from Halobacterium sp. NRC-1 (archaeon), Mycobacterium tuberculosis and Streptomyces coelicolor (high-GC Gram-positive bacteria), and Streptococcus pneumoniae (low-GC Gram-positive bacterium) cluster with the plant and alpha-proteobacterial sequences in group 3, and a sequence from Clostridium acetobutylicum, a low-GC Gram-positive bacterium, clusters with the gamma-proteobacterial sequences in Group 1. (Note that synthesis of γ-glutamyl cysteine has been demonstrated in Halobacterium sp. NRC-1 [31], but not in Mycobacterium tuberculosis, Streptomyces coelicolor or Streptococcus pneumoniae.) The occurrence of GshA homologs in these organisms could reflect persistence of an ancestral GshA in only some genera in the Archaea and in the Gram-positive bacteria, or could be the result of lateral transfer into a limited number of organisms. As discussed above, it is difficult to distinguish between these two explanations, although the lateral gene transfer hypothesis is most appealing.

Why are GshA sequences so divergent?

The three groups of GshA sequences are so divergent that it was difficult to demonstrate an evolutionary relationship between them. This level of sequence divergence is unexpected, and warrants some thought. Three obvious factors contribute to divergence of sequence in orthologs. First, the organisms being compared may be very distant. This explanation is probably not sufficient to explain the sequence divergence in the GshAs. The gamma-proteobacteria represented in group 1 are reasonably closely related to the alpha-proteobacteria in group 3, and the crown eukaryotes in group 2 are reasonably closely related to the plants in group 3. Orthologous relationships between these groups can often be identified. Of the 1,923 COGs (clusters of orthologous groups) [32,33,34] identified in P. aeruginosa (a gamma-proteobacterium), 80% are also found in the combined group containing Mesorhizobium loti and Caulobacter crescentus (both alpha-proteobacteria). As a specific example, the GshB sequences from E. coli (a gamma-proteobacterium) and C. crescentus have 41% identity, and those from Arabidopsis and human have 43% identity. Detection of orthologs in even more distantly related organisms is also possible in many cases. We have used PSI-BLAST to find orthologs of enzymes that use glutathione (including glutaredoxins, glutathione-S-transferases, glutathione reductases and glutathione peroxidases) in both bacteria and eukaryotes (data not shown).

A second reason that homologous sequences may be very divergent is that little selective pressure has been required to maintain function at the level required for the organism to succeed. This situation may occur if the reaction being catalyzed is not very demanding, or if the product of the reaction does not contribute to the fitness of the organism in an important way. An example of the first scenario is o-succinylbenzoate synthase, which catalyzes the dehydration of 2-hydroxy-6-succinyl-2,4-cyclohexadiene carboxylate to form o-succinylbenzoate synthase in the biosynthetic pathway for menaquinone synthesis. o-Succinylbenzoate synthases from various organisms have very low pairwise sequence identities compared to those seen for other members of the enolase superfamily. The low sequence identities in these enzymes have been interpreted as reflecting relatively low constraints upon the sequence because the reaction is quite facile even in the absence of the enzyme (because it forms an aromatic product), and thus the enzyme is not required to provide a great deal of assistance [35]. This scenario is unlikely to account for the divergence of the GshA genes, as formation of a peptide bond is a quite difficult reaction. In fact, members of the ATP-grasp superfamily that catalyze comparable reactions (that is, GshBs (see further below), ribosomal S6 modification enzymes, D-Ala-D-Ala ligases) are sufficiently well conserved to be easily detected by PSI-BLAST, even though they utilize different substrates [36]. With respect to the second scenario, it is clear that the ability to synthesize GSH provides a significant advantage. Bacteria and yeast that lack functional GshA are viable, but are hypersensitive to oxidative damage [37,38]. GshA-deficient strains of A. thaliana are viable, but are hypersensitive to cadmium [39]. Mice in which γ-Glu-Cys ligase has been knocked out die before gestational day 13 [40]. Thus, low selection pressure cannot account for the high levels of divergence among the GshA sequences.

Finally, sequences of homologs may diverge if they are subject to different selective pressures in different lineages. This might occur if the protein has a second function (a 'moonlighting' function) in some lineages and is subject to selective pressure that alters regions of the protein involved in that function. Alternatively, if a protein interacts with other proteins, then differences in those partner proteins will drive changes in the regions of the protein involved in the interaction. The possibility that the high level of divergence in GshA proteins is due to one of these factors is intriguing and worth experimental exploration.

GshB: a different story

A set of GshBs was assembled by searching the NCBI protein database using either GshB or glutathione synthetase as query words, and from the outputs of BLAST searches with the E. coli and S. cerevisiae proteins as query sequences. The sequences in the GshB set fall into two distinct groups, corresponding to bacteria and eukaryotes, that have no significant relationship to each other on the basis of pairwise sequence identities. As seen with GshA, PSI-BLAST searches with GshB sequences from either group did not find GshB sequences from the other group. A PSI-BLAST search with the human GshB converged after two iterations. The output contained only eukaryotic GshBs, a few putative homoglutathione synthestases, and a few eukaryotic proteins of unknown function. No bacterial GshBs were found in the output. After multiple iterations, a PSI-BLAST search using the E. coli GshB (gi121663) as an initial query sequence found 509 sequences of enzymes in the ATP-grasp superfamily, to which the bacterial enzyme is known to belong, but no eukaryotic GshB sequences. The ATP-grasp superfamily [41] includes at least 15 families of enzymes that catalyze formation of a bond between a carboxylate group of one substrate and an amino, imino or thiol group of a second substrate. The bacterial GshBs are most closely related to ribosomal S6 modification enzymes, which catalyze the addition of glutamate to the carboxyl terminus of ribosomal protein S6. Other members of the superfamily include carbamoyl phosphate synthases, cyanophycin synthetases and D-Ala-D-Ala ligases. Notably, many members of the ATP-grasp superfamily were found in eukaryotes. For example, A. thaliana has at least three superfamily members in the PSI-BLAST output, including carbamoyl phosphate synthetase, 3-methylcrotonyl-CoA carboxylase, and acetyl CoA carboxylase. However, the Arabidopsis GshB is not found in the output. Thus, ATP-grasp superfamily members in eukaryotes are more closely related to bacterial GshBs than to eukaryotic GshBs.

Crystal structures of GshBs from E. coli [42] and human [43] are available, so we are able to consider the evolutionary relationship between the two groups of sequences in the context of the structural information (Figure 6). The E. coli and human enzymes have the same overall structural fold, which is typical of the ATP-grasp superfamily [36]. The human enzyme has a number of insertions, and in addition, has a circular permutation of the sequence that results in the movement of part of the carboxy-terminal domain to the amino-terminal domain. The sequence identity between the E. coli and human enzymes is strikingly low. Even in the regions that can be structurally superimposed, it is only 10%. Furthermore, only 2 out of 11 residues involved in binding GSH are conserved between the two enzymes [43].

The structural similarity between the E. coli and human GshBs indicates that the bacterial and eukaryotic enzymes could be related to each other, but is not sufficient to prove that they are related to each other because the common structure might have arisen by convergent evolution from different progenitors. This consideration is especially worth noting in light of the very low conservation of residues in the active site. Consequently, we looked for evidence of an evolutionary relationship by attempting to find proteins that might be distantly related to the bacterial and eukaryotic GshB proteins and therefore might bridge the sequence gap between them using the Shotgun algorithm [44], which was designed to facilitate searches for distant relations between proteins. The algorithm performs a BLAST search with each of a set of query sequences. It then sorts the hits found by all of the proteins, and for each hit, identifies the query sequences that found that hit. A hit that is found by multiple members of two distinct groups of proteins, even with low BLAST scores, can be examined closely to determine whether it provides a link between the groups.

We ran Shotgun with a query set containing six eukaryotic GshBs, eight bacterial GshBs, eight S6 modification enzymes (four from bacteria and four from archaea), and four proteins of unknown function (three from bacteria and one from the archaeon Methanococcus jannaschii). Sample outputs are shown in Figure 7. Generally, sequences were found by either bacterial GshBs and the S6 modification enzymes, or by eukaryotic GshB proteins, but not by both groups. There was only one exception, which was a protein (a putative GshB from Medicago truncatula, gi 4808539) found by all six of the eukaryotic GshBs with BLAST scores above 194 (BLAST probabilities less than 5.2e-31), and one bacterial GshB, with a BLAST score of 63 (BLAST probability of 1.0). Given the number of bacterial GshB sequences in the query set, the finding of one bacterial sequence with an insignificant p-value can be regarded as a chance occurrence. Thus, we were unable to establish any relationship between the two groups of GshB sequences using Shotgun.

Demonstration of conserved sequence motifs would provide supporting evidence for an evolutionary relationship between eukaryotic and bacterial GshBs. In this case, the Block Maker algorithm used to analyze the GshA proteins did not perform well. This algorithm begins by constructing a ClustalW [45] multiple sequence alignment. The circular permutation of the eukaryotic sequences with respect to the bacterial sequences, as well as the extremely low sequence identities, make alignment difficult and preclude the identification of blocks in the permuted region. Therefore, a different motif-finding algorithm, MEME 3.0 [46,47], was used to analyze a divergent set of GshBs, along with several of the proteins related to the bacterial GshBs. Twelve strongly conserved motifs were identified (Table 4). The pattern of distribution of these motifs in eukaryotic GshBs, bacterial GshBs and S6 modification enzymes (the enzymes most closely related to the bacterial GshBs) is illustrated in Figure 8. Motifs 11 and 12 are peculiar to the bacterial GshB proteins and map to the substrate-binding region of the protein. These motifs are replaced by others in the S6 modification enzymes and in D-Ala-D-Ala ligases (data not shown), while the motifs involved in the ATP-binding site (1, 4, 6 and 5) are conserved. This protein scaffold clearly provides a modular structure that is easily adapted to bind different substrates in proximity to bound ATP. Two motifs (1 and 5) that contribute to the ATP-binding pocket were found in both eukaryotic and bacterial GshBs, as well as other members of the ATP-grasp super-family. These motifs were present in different orders in the bacterial and eukaryotic GshBs (see Figure 8) because motif 5 is involved in the circular permutation that moves part of the carboxy-terminal domain to the amino-terminal domain of the eukaryotic enzymes. The lack of conserved motifs in the substrate-binding region is consistent with the comparison of the E. coli and human GshB structures, which shows remarkably little similarity in this region of the protein.

In a case such as this one, it can be difficult to determine whether common sequence motifs have arisen by divergence from a common progenitor, or by convergent evolution driven by a common function. The two common motifs found in the bacterial and eukaryotic GshBs correspond to the ATP-binding pocket. Thus, we examined the possibility that motifs 1 and 5 arose by convergent evolution driven by the need to bind ATP by looking for these motifs in other proteins that bind ATP. We searched the non-redundant database with motifs 1 and 5 using the MAST algorithm. Searches with motif 1 and 5 retrieved 145 and 40 sequences with E-values less than 10, respectively. Nearly all of the proteins with known functions in the output were GshBs or other members of the ATP-grasp superfamily. All of the 145 proteins found by motif 1 were members of the ATP-grasp superfamily except for two transcriptional regulators with E-values of 4.8 (gi15613287) and 7.8 (gi15224768), dihydroorotate dehydrogenases with E-values greater than 7 from several organisms, and a proline/betaine transporter (gi1589297) with an E-value of 9.9. All of the proteins with known functions found by motif 5 were members of the ATP-grasp superfamily except the LIM-containing protein kinase 2t (gi3273207), which had an E-value of 6.2. Thus, motifs 1 and 5 are characteristic of the ATP-binding region of the ATP-grasp superfamily enzymes.

The question of whether eukaryotic GshBs are members of the ATP-grasp superfamily is difficult to answer with certainty because the eukaryotic GshBs are so dramatically different from the other superfamily members. The two conserved sequence motifs involved in ATP binding do provide a link between eukaryotic GshBs and the ATP-grasp superfamily, but it is rather tenuous, as it is possible that these sequences provide the best way to bind ATP within the context of this structural fold and have evolved by convergent evolution in eukaryotic GshBs and the ATP-grasp superfamily members. The lack of conservation in the glutathione-binding region of bacterial and eukaryotic GshBs is also an important consideration. We feel that the evidence for a true evolutionary relationship between the eukaryotic GshBs and the ATP-grasp superfamily is rather weak given the evidence available at this time. For our purposes here, it is sufficient to conclude, in agreement with Polekhina et al. [43], that eukaryotic GshBs did not evolve directly from bacterial GshBs, but rather that both evolved from ancestors that had the characteristic fold of the ATP-grasp superfamily.

A different twist? A fused GSHA-ATP-grasp superfamily homolog in an odd collection of bacteria

In most organisms, GshA and GshB are encoded by separate genes that are not in close proximity. An interesting variation on this theme may occur in a small number of bacteria. Clostridium perfringens, Listeria monocytogenes, Listeria innocuans and Pasteurella multocidans have an open reading frame (ORF) that could encode a GshA homolog fused to an ATP-grasp superfamily member, raising the possibility that this protein might combine the two activities required for synthesis of GSH. C. perfringens and L. monocytogenes contain low levels of GSH (0.25 μmol/g residual dry weight) that are about 20-fold lower than those found in E. coli, but it is not known whether they synthesize GSH or simply import it from the medium [4]. Experimental determination of the function of these fused proteins is clearly needed.

The occurrence of an ORF for the fusion protein in this cluster of bacteria is curious because C. perfringens, L. monocytogenes and L. innocuans are low-GC Gram-positive bacteria, while P. multocida is a gamma-proteobacterium. C. perfringens is found in soil and sewage and is often part of the normal intestinal flora of animals and humans. It causes gangrene and food poisoning in humans [48]. L. monocytogenes and L. innocuans are ubiquitous contaminants of soil and water, and L. monocytogenes causes listeriosis, a serious food-borne illness [48]. P. multocida colonizes the nasopharynx and gastrointestinal tract of many animals and birds, and causes a wide range of illnesses [49]. Human infections are most often caused by dog or cat bites. Thus, the association of these bacteria with animals as either commensal or pathogenic organisms has apparently provided an opportunity for lateral transfer of the gene encoding the fusion protein.

The ORF for the putative fusion protein in P. multocida is particularly intriguing because gamma-proteobacteria typically have a group 1 GshA and a typical bacterial GshB. P. multocida has neither of these, and neither does its close relative, Haemophilus influenzae. The lineage leading to P. multocida and H. influenzae diverged from other gamma-proteobacteria approximately 270 million years ago [50]. P. multocida and H. influenzae have considerably smaller genomes (2,014 and 1,743 predicted coding regions, respectively [50,51]) than E. coli (4,288 predicted coding regions [52]), suggesting that this lineage has undergone substantial genome reduction. It is possible that the lineage leading to P. multocida and H. influenzae lost GshA, and P. multocida subsequently acquired the fused GshA-ATP-grasp superfamily homolog in its place.

Putting together the pieces: thoughts on the evolution of the pathway

The pathway for GSH biosynthesis involves two enzymes, and it is of interest to consider which of these evolved first. Horowitz has postulated that biosynthetic pathways evolve in a retrograde fashion, beginning with the last enzyme in the pathway [53]. This hypothesis rests on the assumption that organisms had, at one time, access to a supply of precursors for biological polymers such as DNA, RNA, proteins and polysaccharides. As the supply of a given precursor dwindled, the most successful organisms would be those that 'invented' an enzyme with which to catalyze formation of that precursor from compounds present in the environment. Thus, there would be continuous selective pressure to add enzymes in the retrograde direction to catalyze synthesis of precursors from ever more simple constituents. Evolution of a biosynthetic pathway in the forward direction was deemed unlikely, as there would be no selective pressure for evolution of enzymes to produce intermediates of no further use to the organism. Horowitz's proposal is logical and appealing. There are cases, however, in which forward evolution of a pathway seems more likely. For example, many organisms make complex natural products whose roles generally involve killing or manipulating other organisms. The pathways for building these complex structures have probably evolved in a forward direction by addition of enzymes capable of adding to the complexity of a pre-existing molecule and thereby contributing to its biological potency.

The GSH biosynthesis pathway is most likely to have evolved in a forward direction. If the pathway had evolved in a retrograde direction, the Horowitz theory would postulate that GshB arose to take advantage of γ-Glu-Cys present in the environment. It is unlikely that γ-Glu-Cys would have been available because formation of the high-energy amide bond would be unlikely to occur abiotically. Furthermore, this molecule would be unstable to oxidation in aerobic environments. However, evolution of the GSH biosynthesis pathway in the forward direction makes considerable sense. γ-Glu-Cys can serve some of the functions of GSH, and therefore could be advantageous to an organism even in the absence of GshB. Indeed, halobacteria contain millimolar levels of γ-Glu-Cys, but do not convert it further to GSH [2]. However, γ-Glu-Cys is not an ideal solution, as it is more easily oxidized than GSH [31]. Furthermore, the reactivity of a thiol depends upon its pKa, as thiolates are orders of magnitude more nucleophilic than thiols [54]. The nucleophilicity of the thiol in γ-Glu-Cys should be diminished by the proximity of the negatively charged carboxylate. Further reaction of γ-Glu-Cys with Gly to form GSH would improve its properties with respect to both oxidation and nucleophilicity, thus, providing selective pressure for evolution of a GSH synthetase (GshB).

One possible source for an enzyme to catalyze the next step in a pathway evolving in the forward direction is the enzyme that catalyzed the last step, as this enzyme has a binding site that accommodates the product of the last reaction, and that product is the substrate for the next reaction. A similar situation occurs for pathways evolving in a retrograde direction. This type of enzyme recruitment, which takes advantage of already existing substrate-specificity determinants, but requires changes in catalytic groups, appears to occur rather infrequently [55]. For example, among 510 proteins involved in the small-molecule metabolic pathways in E. coli, homology between consecutive enzymes in a pathway occurs only six times [56]. Most often, enzymes are recruited to catalyze new reactions by virtue of the catalytic abilities of their active sites, and interactions required for substrate binding are then optimized. The GSH biosynthesis genes, however, appear to be an optimal case for recruitment of one enzyme to catalyze a subsequent reaction. As GshA has a binding site that accommodates γ-Glu-Cys, it would appear to be an ideal progenitor of GshB, which uses γ-Glu-Cys as a substrate and also catalyzes the ATP-dependent formation of an amide bond. However, GshA and GshB appear to be structurally distinct. There are no experimental structures for GshAs, but recent work suggests that GshAs are homologs of glutamine synthetases [57]. GshBs have a different structural fold, characteristic of the ATP-grasp superfamily. Thus, the data support a scenario in which emergence of GshA was followed, in most organisms, by the recruitment of a different protein to serve as the progenitor of GshB. It is particularly interesting that, in both the bacterial and eukaryotic lineages, the ATP-grasp structural fold provided the starting point for the evolution of GshB.