Cobalamin-Independent Methionine Synthase (MetE): A Face-to-Face Double Barrel That Evolved by Gene Duplication

Cobalamin-independent methionine synthase (MetE) catalyzes the transfer of a methyl group from methyltetrahydrofolate to L-homocysteine (Hcy) without using an intermediate methyl carrier. Although MetE displays no detectable sequence homology with cobalamin-dependent methionine synthase (MetH), both enzymes require zinc for activation and binding of Hcy. Crystallographic analyses of MetE from T. maritima reveal an unusual dual-barrel structure in which the active site lies between the tops of the two (βα)8 barrels. The fold of the N-terminal barrel confirms that it has evolved from the C-terminal polypeptide by gene duplication; comparisons of the barrels provide an intriguing example of homologous domain evolution in which binding sites are obliterated. The C-terminal barrel incorporates the zinc ion that binds and activates Hcy. The zinc-binding site in MetE is distinguished from the (Cys)3Zn site in the related enzymes, MetH and betaine–homocysteine methyltransferase, by its position in the barrel and by the metal ligands, which are histidine, cysteine, glutamate, and cysteine in the resting form of MetE. Hcy associates at the face of the metal opposite glutamate, which moves away from the zinc in the binary E·Hcy complex. The folate substrate is not intimately associated with the N-terminal barrel; instead, elements from both barrels contribute binding determinants in a binary complex in which the folate substrate is incorrectly oriented for methyl transfer. Atypical locations of the Hcy and folate sites in the C-terminal barrel presumably permit direct interaction of the substrates in a ternary complex. Structures of the binary substrate complexes imply that rearrangement of folate, perhaps accompanied by domain rearrangement, must occur before formation of a ternary complex that is competent for methyl transfer.


Introduction
Methionine synthases catalyze the transfer of a methyl group from N5-methyl-5,6,7,8-tetrahydrofolate (CH 3 -H 4 folate) to L-homocysteine (Hcy), the terminal step in the biosynthesis of methionine. Two apparently unrelated families of proteins catalyze this reaction: cobalamin-dependent methionine synthase (MetH; EC 2.1.1.13) and cobalaminindependent methionine synthase (MetE; 5-methyltetrahydropteroyltriglutamate-homocysteine methyltransferase; EC 2.1.1.14) Organisms that synthesize or transport B 12 encode the cobalamin-dependent enzyme whereas organisms that cannot obtain B 12 encode only the cobalamin-independent enzyme. Escherichia coli and many other species of bacteria express both enzymes, but mammals utilize only cobalamindependent methionine synthase while plants and yeasts utilize only the cobalamin-independent enzyme.
MetH and MetE both face the same mechanistic challenge. They must catalyze the transfer of a very poor leaving group from the tertiary amine, CH 3 -H 4 folate, to a relatively poor nucleophile, the sulfur of Hcy. MetH facilitates this transfer by using cobalamin as an intermediate methyl carrier [1]. Cobalamin accepts a methyl group from CH 3 -H 4 folate at one active site and donates it to Hcy at a second site [2]. In contrast, MetE appears to catalyze the direct transfer of the methyl group from CH 3 -H 4 folate to Hcy [3]. This latter strategy seems to offer a less satisfactory answer to the mechanistic problems: measured k cat values for MetE are smaller than those for MetH by a factor of approximately 50-100.
MetE and MetH both activate Hcy by binding the thiolate form of the substrate to Zn þ2 [4]. A similar strategy for alkylation of thiol groups is employed in protein farnesyltransferase [5], geranylgeranyltransferase [6], methanol:CoM methyltransferase (MtaA) [7], the E. coli DNA repair Ada protein [8], and betaine-Hcy methyltransferase (BHMT) [9]. However, the sets of zinc ligands and the structures that house the zinc-binding sites are not conserved within this functional family. In particular, the metal ligands and their positions in the sequence are not the same in MetH and MetE. Three cysteines bind the essential zinc in MetH; the first cysteine ligand resides at the end of strand 6 of a (ba) 8 barrel, and the remaining vicinal cysteine ligands follow strand 8. A histidine and two cysteines have been identified as metal ligands in E. coli MetE by a combination of mutagenesis experiments [10,11] and extended X-ray absorption fine structure (EXAFS) measurements [12]. The relative positions of these residues in the sequence led to the prediction that in a (ba) 8 MetE barrel the histidine and cysteine ligands would reside at the ends of strands 5 and 8 [4]. In contrast, the sequences of MetE enzymes give few if any clues to the strategy for binding and activation of folate by MetE. Thus, the mode of folate binding is a key question to be addressed by structure analysis. In both MetE and MetH, activation of the leaving group is thought to involve the protonation of CH 3 -H 4 folate in a ternary complex, E Hcy CH 3 -H 4 folate in MetE, or E cob(I)alamin CH 3 -H 4 folate in MetH [4]. However, the residues that may facilitate protonation have not been identified for either enzyme.
MetE appears to have evolved through gene duplication of a sequence encoding a domain of approximately 340 residues that binds and activates Hcy. Within the family of MetE enzymes (Figure 1), the N-and C-terminal halves exhibit significant sequence homology. The C-terminal half is more highly conserved than the N-terminal half and has homologs in archae and elsewhere. Among these thiol methyltransferases are several enzymes that are approximately half the size of MetE and utilize corrinoid proteins, rather than folates, as methyl donors. Taken together, these observations suggested that the MetE gene arose as the result of a primordial gene duplication event followed by loss of zincand Hcy-binding determinants from the duplicated sequence [11]. If this hypothesis is correct, the two halves of the MetE sequence should display structural homology, and the Nterminal domain should be more closely related to the Cterminal domain than to any other protein in the database.
To determine how MetE has assembled an active site for catalysis of direct methyl transfer from CH 3 -H 4 folate to Hcy, we have solved the crystal structure of Thermotoga maritima MetE at 2.0 Å resolution, along with structures of the binary substrate complexes with Hcy and folate. Difficulties in crystallization of the E. coli enzyme were circumvented by analyzing the MetE from T. maritima. This thermophilic bacterium encodes orthologs of E. coli MetH and E. coli MetE. T. maritima MetE (TM1286) is 41% identical to the E. coli enzyme and is only 19 residues shorter than E. coli MetE (Figure 1), making it an excellent prototype for the MetE family. MetE comprises two (ba) 8 barrels. To our knowledge, it is the first example of a dual-(ba) 8 barrel enzyme in which the active site is located between barrels arranged in a headto-head orientation. MetE also provides a rare example of a catalytic zinc site in which four residues serve as metal ligands. Repetition of features within the structure supports the idea that MetE evolved through gene duplication of a primordial zinc/Hcy (ba) 8 barrel.

Description of the Fold and Its Evolution
We have determined the structures of several forms of MetE from T. maritima (Table 1), including the zinc-replete binary substrate complexes with folate (the substrate in these experiments was N5-methyl-5,6,7,8-tetrahydropteroyl-(tri)-c-L -glutamate [CH 3 -H 4 PteGlu 3 ]) at 2.59 Å resolution and with Hcy at 2.20 Å resolution. The two (ba) 8 barrels are formed by residues 1-351 (folate barrel) and 387-734 (Hcy barrel) and joined by an extended inter-domain linker (Figure 2A). An unusual feature is that many of the binding determinants for folate actually reside in the C-terminal barrel that binds Hcy.
In (ba) 8 (triose phosphate isomerase [TIM]) barrel enzymes, the active site is usually located near the C-termini of the inner barrel strands, with catalytic residues contributed by the b-a segments (loops) that join these strands to the outer helices. These barrels are topologically polar with their ''tops'' decorated by insertions that extend the b-a loop segments. MetE is the first example of a dual-(ba) 8 barrel in which the decorated tops of the two barrels face each other to form a single active site that lies between the domains. A deep cleft between the barrels permits entry of the substrates (Figure 2A). As a result of the arrangement of the barrel domains, residues 352-386 of the inter-barrel linker must span approximately 65 Å to connect the bottoms of the two barrels ( Figure 2A).
The N-and C-terminal barrels share a number of strikingly similar features that provide structural evidence for gene duplication. Both barrels incorporate long extensions in the first four b-a loops but, with the exception of a8AF ( Figure  2B), lack insertions in the last four loops. A pseudo 2-fold axis superimposes these similar extensions ( Figure 2). The helical extension at the start of b1-a1, a1AH, augments the side of the C-terminal barrel and is repeated in the N-terminal barrel. The b2-a2 loops that appear in both barrels are the longest extensions in the structure, and we refer to them as the ''long hairpin loops.'' In the C-terminal barrel that binds Hcy and Zn þ2 , the long b2-a2 loop begins with helix a2AH and then forms an antiparallel excursion that harbors a number of conserved residues, some of which are involved in binding folate. The b3-a3 loops both include a short helix, a3A, that carries folate-binding determinants in the Hcy (Cterminal) barrel. The b4-a4A segments of both barrels incorporate a conserved sequence, identified by asterisks in Figure 1. In E. coli MetE, this sequence, Gln-Ile-Asp-Glu-Pro-Ala, is identical in both barrels.
Despite the pseudosymmetry that relates the b-a loops of the two barrels, there are significant differences in the sequences and conformations of these connecting loops that distinguish the functional roles of the two barrels. The major binding determinants for the folate substrate lie primarily in the second, third, and fourth b-a loops of the C-terminal barrel. The equivalent binding site in the duplicated Nterminal barrel has been obliterated. Although the long b2-a2 hairpin of the N-terminal barrel resembles the corresponding hairpin of the C-terminal (Hcy) barrel, the potential binding groove for the folate tail is closed by extensive hydrophobic contacts between a1AF, a1F, and a2CF and by interaction with the long hairpin of the C-terminal barrel. The positioning the long b2 hairpin significantly closer to the barrel top than in the C-terminal barrel occludes the pterinbinding site ( Figure 2B).
Several other adaptations enhance the barrel-barrel interface and appear to influence the relative orientations of the barrels. The b8-a8 loop containing helix a8AF is 16 residues longer than the corresponding loop in the C-terminal domain and makes numerous inter-barrel contacts. Likewise, loop b4-a4 in the C-terminal domain is extended by six residues, increasing contacts with the N-terminal barrel. The single-domain archaeal Hcy methyltransferases that are homologous to the C-terminal domain are missing these six residues, supporting the notion that this insert evolved to enlarge the inter-domain interface.

The Zinc Site
The zinc-binding site of MetE is distinguished from most catalytic zinc sites by the presence of four protein ligands to zinc. Mutagenesis of E. coli MetE had previously shown that zinc is bound by a histidine and two cysteines [11] that are equivalent to residues His618, Cys620, and Cys704 in T. maritima MetE. EXAFS studies of the E. coli enzyme indicated a fourth oxygen or nitrogen coordinated to zinc that seemed likely to be a water oxygen. However, the structure of zincreplete enzyme with folate bound (Table 1) clearly reveals the presence of a fourth protein ligand to zinc. A carboxylate oxygen of the invariant Glu642, which had not previously been identified as a metal ligand, is coordinated to zinc. In the zinc-replete complexes of MetE with folate that provide resting-state structures of the zinc site, the metalligand cluster adopts tetrahedral geometry. EXAFS measurements on substrate-free E. coli MetE are also consistent with tetrahedral coordination [3,11] with bond lengths of 2.31 Å for two Zn-S bonds and 2.04 Å for two nitrogen or oxygen ligands. The observed Zn-S bond lengths in our structure are 2.30 Å , Zn-N is 2.07 Å , and Zn-O is 2.14 Å , in good agreement with the EXAFS measurements.
The residues that bind zinc are located near the ends of barrel strands 5, 6, and 8 ( Figure 3). His618 is the Cterminal residue of b5 and Cys620 is located on the following b-a loop. Glu642 is at the C-terminus of b6, and Cys704 resides on the loop following strand b8. The ba loops that contain the two cysteine residues are drawn together to form the metal-binding site, distorting the barrel (see Discussion).

The Binary Complex with Hcy
In the complex of Hcy with zinc-replete MetE (Table 1), Hcy is positioned by numerous interactions with conserved protein residues ( Figure 4). The amino group is coordinated by hydrogen bonds to Asp577, Glu462, and the carbonyl of Ile409. The carboxyl group of Hcy is bound by the backbone amide and the side chain hydroxyl of Ser411, and hydrophobic contact to the Hcy sulfur is provided by Met468. These interactions with the Hcy substrate are reminiscent of those observed in MetH [2] (see Discussion).
Hcy binding induces significant changes at the metal site ( Figure 4). Binding of the substrate sulfur to zinc does not proceed by a simple dissociative reaction in which sulfur is substituted for the oxygen of Glu642. Instead, Hcy approaches the metal ion from the side opposite the Glu642 ligand. This mode of association is unusual; in most catalytic  8 barrels. The N-terminal barrel (aqua) is joined to the C-terminal barrel (yellow) by a 35-residue inter-domain linker (gray) that spans 65 Å . Except for the a1 helix (at the right), the linker residues are in extended conformations. This view is along the approximate 2-fold axis that relates the two barrels. The drawing is based on coordinates for zinc-replete MetE in complex with CH 3 -H 4 folate ( Table 1). The zinc ligands, zinc, and CH 3 -H 4 folate are shown in ball-and-stick representation. This figure and all subsequent figures were prepared using RIBBONS [43]. See Figure 1 for the nomenclature used to describe secondary structures. (B) A side-by-side view of the barrels of MetE, arranged to show the similarities of the b-a loop extensions. Major extensions that follow the first four b strands of the barrels are shown in cyan and gold for the N-terminal and C-terminal barrels, respectively. The drawing is based on coordinates for the zinc-replete binary complex with folate (not shown). The active site is located in the C-terminal barrel (on the right) between the extra-barrel b hairpin of the b2-a2 loop and the C-termini of the barrel strands. Zinc is gray and the zinc ligands, His618, Cys620, Glu642, and Cys704, are shown in ball-and-stick mode. DOI: 10.1371/journal.pbio.0030031.g002 zinc sites the incoming substrate replaces a dissociable fourth ligand without inversion [13]. In the structure of the binary complex determined at pH 5.2, displacement of Glu642 and inversion at Zn þ2 are incomplete. Zinc coordination changes from tetrahedral to distorted trigonal bipyramidal geometry (Figure 4), and the substrate sulfur and glutamate oxygen both exhibit unusually long ligand-metal distances: 3.15 Å and 2.9 Å for Zn-S and Zn-O, respectively. Comparison with the zinc-replete complex with folate ( Figure 4B) shows that zinc and His618 have moved 0.76 Å and 0.98 Å toward the substrate while Cys620, Glu642, and Cys704 remain essentially fixed. The Zn-S(Cys) and Zn-N distances are not significantly altered, but the zinc and these three ligands become more nearly coplanar ( Figure 4B). Complete inversion would result in geometry resembling the Hcy complex of MetH ( Figure 4C).
The geometry at the metal site in the Hcy complex has been confirmed by omit refinement and by tests with restrained models. The density around the zinc atom in omit maps is well resolved from Hcy but continuous with that of the cysteine ligands ( Figure 4D), consistent with a Zn-Hcy distance that is significantly longer than the Zn-Cys bond distances. Imposing tetrahedral restraints in refinements with Hcy as the fourth ligand results in large difference Fourier peaks that also indicate a long Zn-Hcy distance.
Cysteinyl tRNA-synthetase incorporates an active-site zinc with the same set of ligands as MetE and provides a precedent for an inversion at the metal center induced by substrate binding. Cysteinyl tRNA-synthetase uses this zinc ion not for catalysis, but to discriminate against serine, exploiting the strong zinc-thiolate interaction with its substrate [14,15]. In the absence of substrate the zinc site displays geometry that is intermediate between tetrahedral and trigonal bipyramidal. As in MetE, the cysteine substrate binds opposite a glutamate residue and does not displace the glutamate ligand directly. Upon cysteine binding, zinc moves away from glutamate and forms a 2.5 Å bond to cysteine.

Folate Binary Complex
Structures of CH 3 -H 4 folate bound to MetE from T. maritima have been determined in both the reduced (zincreplete) enzyme and the oxidized (disulfide-bonded) form (Table 1). CH 3 -H 4 folate is bound in identical fashion in both structures. The novel feature of these structures is that the MetE folate complex fails to comply with the classic picture of substrate binding in a TIM barrel. The folate substrate binds in a deep cleft between the two barrels, with its glutamate tail accommodated by a groove in the enzyme surface ( Figure 5). The pterin is displaced from the top of the N-terminal barrel, and simultaneously shifted away from the axis of the C-terminal barrel that binds zinc and Hcy (see Figure 2A). It is thus inappropriate to call the Nterminal barrel the ''folate-binding domain.'' An animation, in which Figure 5 is rotated about its vertical axis, provides a more complete view of the structure and its bound ligands (Video S1).
The pterin ring of CH 3 -H 4 folate is positioned by stacking and by hydrogen bonding with conserved residues ( Figure 5). Glu583 makes a bidentate interaction with the 2-NH 2 and N3 groups of CH 3 -H 4 folate, an arrangement found in several other folate-binding sites. Interaction of N3 with an acidic group has been shown to be important for catalytic activity in dihydrofolate reductase [16], thymidylate synthase [17], and MetH [18]. In the binary complexes with MetE that we report here, the pterin ring of the folate is stacked against Trp539, Lys104 hydrogen bonds to O4, and the folate N5 is hydrogen bonded to a water molecule. However the N8 and N1 positions of the pterin are exposed to solvent.
The subsite that binds the glutamate tail is a groove lined by conserved basic residues ( Figure 5). Arg15, Lys18, Arg493, and Arg496 interact with the first glutamyl residue, which is the only one of the three c-linked glutamate residues that is ordered in the binary complex. Weak binding of the other tail  residues is surprising for an enzyme that exhibits an absolute requirement for polyglutamylated folate, but glutamate tails have displayed disorder in other structures where they also contribute to strength of binding [19]. The orientation of bound folate in the binary complexes does not allow transfer of the methyl group to Hcy. As can be seen in Figure 5, the N5-methyl carbon faces away from zinc and Hcy; it is 11 Å from the sulfur of Hcy when the binary complexes are superimposed. A rotation about the folate N10-C49 dihedral angle, with the interactions of the paraamino benzoyl moiety and glutamate acting to anchor the substrate, would position the methyl group correctly with respect to Hcy, decreasing the distance between groups that react to 6.0 Å . This distance is still long, and additional protein or substrate rearrangements would be necessary to close the gap between the sulfur of Hcy and the methyl carbon. An alternative route to a ternary complex that supports methyl transfer would be complete dissociation and reassociation of the CH 3 -H 4 folate. However, because so many interactions with conserved residues are observed in this binary complex, it seems likely to represent an initial intermediate rather than a dead-end complex (see Discussion).

The 467 Asp-Met-Val Sequence Mediates Interaction between the Substrate-Binding Sites
One of the fascinating features of the binary substrate complexes is the evidence for communication between the substrate-binding sites, mediated by the invariant 467 Asp-Met-Val (DMV) sequence, which forms the N-terminal turn of helix a2AH. When Hcy binds, the side chain of Met468 alters its position in concert with backbone displacements of the aspartic acid and methionine residues that start the helix.
Comparison of the substrate-free structure with the Hcy binary complex shows how the DMV region moves toward the zinc center when Hcy binds ( Figure 6). These changes in turn affect the interactions and orientation of Trp539, favoring the conformation in which Trp539 can stack against the pterin ring. In the absence of substrates, Trp539 can adopt another conformation that would overlap the binding site for the pterin ring.
In the binary complex with CH 3 -H 4 folate, the DMV loop is also recruited to the position it occupies when Hcy is bound. The changes that are induced by binding of folate are reproduced in the complex of folate with the oxidized enzyme. Thus, binding of either substrate favors a conformation that would be expected to increase the affinity of MetE for the other substrate. These small but significant conformation changes observed in the binary complexes precede larger rearrangements that must be induced by binding of both substrates to form a competent ternary complex. Cooperativity in substrate binding could increase the concentrations of the ternary complex and thereby increase turnover in a system that is already plagued by slow chemistry [4].

Comparisons with MetE from Arabidopsis thaliana
A very recent paper, which appeared after the submission of our manuscript, has described the structure of MetE from A. thaliana [20]. Although the folds of the enzymes from A. thaliana and T. maritima are obviously similar, there are some significant differences in the reported features of the zincbinding sites. In the complexes of A. thaliana MetE with Hcy or methionine, the distances between zinc and substrate or product sulfur are long, as is the case in the Hcy complex of MetE from T. maritima, but water rather than glutamate has been assigned as the ligand opposite to Hcy or Met. In contrast, the electron density in omit maps of the Hcy complex of T. maritima MetE shows no evidence for a water intervening between glutamate and zinc (see Figure 4D). In substrate-free A. thaliana MetE, the metal-ligand bonds are all very long and the geometry is highly distorted, suggesting some disordering or partial oxidation of the metal site under the conditions used for crystallization.
To study folate binding, PteGlu 5 and CH 3 -H 4 PteGlu 5 were added to crystals of the Hcy or methionine complexes of A. thaliana MetE [20]. In the resulting structures, the pterin ring is flipped relative to its position in the binary folate complex of T. maritima MetE, and adopts an orientation that is similar to what we anticipated from model building. Although the occupancy of the reduced folate appears to be low, it is estimated that the methyl group of the CH 3 -H 4 PteGlu 5 is about 7 Å from the sulfur of Hcy, too distant for transfer to Hcy. Thus, both structure analyses suggest that additional conformation changes must occur to form a reactive ternary complex.

Comparisons of MetE with MetH
MetE and MetH display no detectable sequence homology and have different sets of zinc ligands. Comparison of the barrels from MetE with the corresponding domains of MetH that bind folate or Hcy reveal that the N-terminal barrel of MetE, which carries some folate-binding determinants, differs in significant ways from the folate barrel of MetH, whereas the Hcy barrels share many similar features.
Two other (ba) 8 barrels that bind CH 3 -H 4 folate have been described: methyltetrahydrofolate corrinoid/iron-sulfur protein methyltransferase [21] and the folate-binding module of MetH [2]. These homologous barrels both bind the CH 3 -H 4 folate substrate at the top of the folate barrel and use similar interactions with residues contributed by the Ctermini of the inner barrel strands. In MetE, CH 3 -H 4 folate is displaced from the N-terminal barrel and bound primarily by residues in the long extra-barrel b hairpin of the C-terminal Hcy domain (see Figures 2A and 5). Dissimilarities of the decorating loops in the N-terminal barrel of MetE and the folate barrel of MetH are documented by poor statistics for sequence matches and for alignments with the structures of corrinoid/iron-sulfur protein methyltransferase or MetH.
The Hcy barrels of MetE and MetH are compared in Figure  7, which shows how each structure accommodates metal binding. Strand distortions in the Hcy barrels, which have been associated with construction of a metal-binding site [2,9], are related but not identical in MetE and MetH (or BHMT [9]). In MetH, strand b7 is extruded from the barrel and strands b6 and b8 are pinched together to bind the zinc [2]. In MetE, strands b6 and b7 are displaced relative to their positions in MetH, allowing strands b5 and b8 to approach one another. In both enzymes, distortion of b8 is accompanied by a splay in strand b1; only one classic hydrogen bond is made between strands b1 and b2. Conserved residues in MetH, MetE, and BHMT stabilize inter-strand interactions by forming side-chain-to-main-chain hydrogen bonds.
Major differences in the connecting loops b1-a1 and b8-a8 displace the zinc and Hcy sites in MetE relative to MetH by approximately 6 Å so that the sulfur of Hcy is no longer on the barrel axis but is shifted toward one wall of the barrel (Figure 7). The altered positions of the ligands and the binding of zinc by Glu642 lead to inversion of the zinc center relative to its configuration in MetH. Displacement of the Zn þ2 /Hcy site and the unusual mode of folate binding seem to have evolved to allow methyl transfer in a ternary E Hcy CH 3 -H 4 folate complex.
Despite the translation and inversion of the metal and its ligands, the orientation and local interactions of Hcy are almost identical in MetE and MetH (Figure 7). Both enzymes use conserved carboxylate residues to interact with the amino group of Hcy, forming salt bridges (see Figure 4). In both MetH and BHMT, the carboxyl group of Hcy is bound by a pair of backbone amides located at the beginning of the b1-a1 extension. In MetE the corresponding interactions of COO À are made by the backbone amide and the side chain hydroxyl of Ser411, again located on the extension following b1.
Curiously, it appears that distortions of the barrel strands and connecting loops need not be undone when metal binding is lost through evolution. Distortions of the barrel strands and their downstream loops are retained in the Nterminal barrel of MetE despite loss of metal and Hcy binding, and similar distortions were first observed in uroporphyrinogen decarboxylase [22], which is also not a metalloenzyme. Despite the lack of detectable sequence homology, uroporphyrinogen decarboxylase is the closest structural relative of the Hcy domain of MetE in the current protein database: alignment using DALI [23] matches the two folds with a similarity score that is higher than that for the Hcy barrels from MetE and MetH. Uroporphyrinogen decarboxylase may have evolved from a Zn þ2 /Hcy barrel, despite the fact that it is no longer a zinc-dependent thiol alkyltransferase.

Substrate Binding and Activation: Inferences from the Structures
In the structure of the MetE Hcy complex, the zinc site adopts distorted trigonal bipyramidal geometry with long bonds from the metal ion to glutamate and Hcy (see Figure 4). In contrast, EXAFS measurements on the Se-Hcy complex of E. coli MetE at pH 7.2 are best fit to a tetrahedral ligand environment with two sulfurs (2.33 Å ), one nitrogen or oxygen (2.02 Å ), and one selenium (2.433 Å ), in which the longer Zn-Se distance reflects the increased radius of selenium relative to sulfur [3]. EXAFS studies of the related methyltransferase MT2-A, which has the same set of zinc ligands as MetE, also indicate tetrahedral geometry in the binary complex with coenzyme M (2-mercaptoethanesulfonic acid) [7]. Both of these studies concluded that the geometry at zinc does not change appreciably upon binding of the thiolate substrate.
Because the crystals of the Hcy binary complex of MetE were equilibrated at pH 5.2, it is possible that Hcy may be protonated in the X-ray structure. Thus, the long S-Zn bond and partial inversion at zinc might be explained by the relatively weak interaction between the thiol and the metal ion, as documented for thiol and thioether ligands in model compounds [24,25,26]. X-ray studies of Hcy binding at neutral pH and EXAFS measurements at lower pH will be necessary to determine whether pH is a key parameter that affects the metal-ligand geometry. It is possible that complete inversion of zinc geometry will be observed in the structure of the MetE Hcy complex at neutral pH.
A long Zn-S À bond may be functionally important. It has been suggested that a long bond and/or distorted geometry would optimize the reactivity of zinc-dependent thiol alkyltranferases [27] by increasing the charge on the thiolate sulfur, and would avoid trapping of a lower-energy tetrahedral species. Since zinc is known to have flexible coordination geometry [28], a five-coordinate state seems plausible, and the observed structure of MetE Hcy may indeed afford a glimpse of an intermediate or transition-like state that is poised to attack the N5 methyl group of the folate substrate.
The motion of Zn þ2 that accompanies Hcy binding to MetE is unique among the Hcy methyltransferases with known structures. There is no evidence for inversion of configuration at zinc in MetH, and inversion is precluded in BHMT, where a leucine occupies the position analogous Glu642 of MetE. Zinc motion provides a novel way to alter the distribution of charge among the zinc and its ligands, thereby modulating thiolate reactivity. The effects on the electronic structure could be larger than those resulting from the changes in bond lengths observed in other zinc-dependent alkyltransferases [29,30]. By analogy with a proposal for the reaction cycle of protein farnesyltransferase [30], reassociation of Glu642 could promote dissociation of the methionine product following methyl transfer.
An unexpected feature of the MetE structures is the binding mode of CH 3 -H 4 folate, with the pterin ring incorrectly oriented for methyl transfer. It is possible that folate binds initially in this manner to avoid blocking access to the Hcy binding site, which lies between zinc and CH 3 -H 4 folate. Space-filling models show that the Hcy-binding site remains accessible in the MetE CH 3 -H 4 folate binary complex. The observed binding mode thus permits random addition of substrates but does not rule out a kinetically preferred order of binding.
In both MetH and MetE, protonation of methyltetrahydrofolate is thought to occur in the ternary complexes but not in the binary folate complexes [4]. The immediate proton donor has not yet been identified by enzymatic or biochemical studies of E. coli MetE. Formation of the binary Hcy complex results in release of a proton to solvent (Z. S. Zhou and R. G. Matthews, unpublished data). Thus Hcy is not a likely proton donor. It is more likely that an active-site residue may serve as an acid catalyst. Structures of the binary complexes do not implicate a particular residue as a general acid catalyst, but suggest possible candidates. His111 could serve as a proton donor through water if protonation were to occur before folate rearrangement. His672, part of a conserved 670 Asp-Ile-His-Ser-Pro sequence, or Asp467, located in the DMV loop, may be positioned to serve as donors if protonation occurs after folate rearrangement. All three of these candidate residues are invariant in multiple sequence alignments.

Gene Duplication of a Sequence Encoding the Hcy Barrel
Earlier comparisons of sequences had suggested that MetE evolved by gene duplication. The two domains of E. coli MetE share a conserved Gln-Ile-Asp-Glu-Pro-Ala repeat [31], and the sequences display 39% identity (50% similarity) within the regions now seen to span b3 to a4A of the two barrels. The pseudosymmetry of the structural features decorating the barrels provides compelling evidence for a close relationship between the two halves of MetE, verifying the inferences based on sequence alignments. The cores of (ba) 8 barrels, though they display characteristic distortions and can be classified into subgroups [32], do not provide as strong evidence for relatedness as do the similarities of regions inserted in intervening loops. It is difficult to ascertain whether the regions responsible for folate binding were inserted before or after gene duplication. The long b2-a2 loop in the N-terminal barrel and corresponding sequences in archaeal relatives of MetE favor the notion that precursors of folate-binding segments were present before duplication. In any case, folate-binding determinants have developed or been retained primarily in the C-terminal Hcy barrel but not in its N-terminal replicate.
The idea that the zinc/Hcy barrel is the ancestral fold is based on several lines of evidence. This barrel shows significant structural homology to a broad family of zincdependent thiol methyltransferases, including not only MetH and BHMT but also the single-domain archaeal transferase enzymes that react with methylcobalamin. Indeed it seems likely that the three enzymes that convert Hcy to methionine, MetE, MetH, and BHMT, are all descended from a primordial zinc/Hcy barrel. In contrast, DALI searches that assess structural similarities [23] reveal that the N-terminal barrel is more similar to the C-terminal barrel of MetE than to any other known protein, suggesting that its immediate precursor is the Hcy barrel of MetE.
Gene duplication of a Zn þ2 /Hcy barrel would have replicated the sites that bind Zn þ2 and Hcy, but these sites have been disabled in the N-terminal barrel of contemporary MetE. Disruption of zinc and Hcy binding is effected by both residue mutations and backbone conformational changes ( Figure 8). Mutation of the equivalent of Cys620 in the Hcy barrel to Tyr232 leads to stacking with the adjacent Tyr233 (phenylalanine in most MetE sequences) and results in a major backbone conformational change. The hydroxyl group of the Tyr232 forms a hydrogen bond to Asn199 of the conserved Leu-Val-Asn-Glu-Pro-Ala sequence at the b4-a4 loop and thus removes a crucial Hcy-binding determinant. Although Cys309, the equivalent of Cys704 of the C-terminal barrel, is retained in most MetE sequences, mutation of the other zinc ligands and many of the Hcy-binding residues leads to a complete overhaul of the binding site for Hcy.

Why Is a Second Barrel Recruited for a Ternary Complex Mechanism?
Although an entire barrel domain is recruited to elicit direct methyl transfer from folate to Hcy, the structure reveals that the functional groups of this domain are mostly disabled. The binding determinants and potential catalytic residues are located primarily in extensions and excursions of the C-terminal Zn þ2 /Hcy barrel rather than in the ''new'' Nterminal domain. Duplication and divergence of an entire barrel is an elaborate strategy to accommodate a second relatively large substrate at a single active site, and MetE provides the first instance of the face-to-face barrel construction that is required to build such an active site. A more common strategy to accommodate a ternary complex is exemplified by the related Hcy methyltransferase, BHMT. In this enzyme the site for the rather small second substrate, betaine, is constructed in part from b-a barrel extensions and in part from a dimerization arm that is appended to the barrel and contributes to the active site of the partner chain of the functional dimer.
The N-terminal domain of MetE may nevertheless be essential for ternary complex formation and catalysis. In the family of thiol alkyltransferases, quenching of opposite charges on Hcy and the alkyl donor is believed to drive the reactions, and a hydrophobic environment [33] and desolvation [34] of reactants may be critical for reactivity. Rearrangement of folate to form a viable ternary complex may be accompanied by rearrangement or closure of domains around the substrate-binding cleft that would reposition key residues and isolate the active site from solvent. Hcy binding is accompanied by a contraction of the top of the Cterminal barrel that alters the relative positions of the N-and C-terminal barrels. This small domain shift clearly gives hints of inter-domain flexibility. On the other hand, the commissioning of the N-terminal domain may reflect an evolutionary strategy in which gene duplication is utilized as the most facile way to recruit additional sequences. Structures of ternary complexes, obtained using mutant enzymes and/or substrate analogs, should provide further insights into the methyl transfer reaction and the functional roles of the Nterminal domain.
Team (DND-CAT) Synchrotron Research Center located at Sector 5 of the APS. DND-CAT is supported by E. I. DuPont de Nemours, the Dow Chemical Company, the US National Science Foundation through grant DMR-9304725, and the State of Illinois through the Department of Commerce and the Board of Higher Education grant IBHE HECA NWU 96. Assistance with data collection at the APS synchrotron was provided by Zdzislaw Wawrzak (DND-CAT) and Joseph Brunzelle (LS-CAT). We thank Rowena G. Matthews, Kather-ine A. Pattridge, and Rebecca E. Taurog for critical evaluation of the manuscript.
Competing interests. The authors have declared that no competing interests exist.
Author contributions. RP and MLL conceived and designed the experiments. RP performed the experiments. RP analyzed the data. RP contributed reagents/materials/analysis tools. RP and MLL wrote the paper.