A Novel Enzymatic Function on an Ancient Metal Binding Scaffold

View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by Elsevier - Publisher Connector

Structure, Vol. 9, 637–646, July, 2001, 2001 Elsevier Science Ltd. All rights reserved. PII S0969-2126(01)00622-0 Crystal Structure of Methylmalonyl-Coenzyme A Epimerase from P. shermanii: a Novel Enzymatic Function on an Ancient Metal Binding Scaffold

Andrew A. McCarthy,1 Heather M. Baker,1 Steven tabolite propionyl-CoA, which must then be converted C. Shewry,1 Mark L. Patchett,3 and Edward N. Baker1,2,4 to the 4-carbon succinyl-CoA via methylmalonyl-CoA. 1 School of Biological Sciences Defects in this pathway, which include the B12-depen- 2 Department of Chemistry dent enzyme methylmalonyl-CoA mutase, can result in University of Auckland a buildup of methylmalonic acid, which causes severe Auckland acidosis and also damages the central nervous system New Zealand [2]. Methylmalonyl-CoA is also an intermediate in the 3 Institute of Molecular Biosciences breakdown of the branched amino acids valine, leucine, Massey University and isoleucine, again through the pathway leading from Palmerston North propionyl-CoA to succinyl-CoA. This pathway can also New Zealand operate in reverse, however; for example, the Propioni- bacteria can derive energy by the conversion of lactate through succinyl-CoA to propionyl-CoA, ultimately ex- Summary creting propionate. Methylmalonate is also used in the biosynthesis of important polyketide antibiotics such as Background: Methylmalonyl-CoA epimerase (MMCE) erythromycin, which requires propionate and methyl- is an essential enzyme in the breakdown of odd-num- malonate units [4–5]. bered fatty acids and of the amino acids valine, isoleu- As a chiral molecule, methylmalonyl-CoA must be in cine, and methionine. Present in many bacteria and in the correct isomeric form for use by the enzymes of this animals, it catalyzes the conversion of (2R)-methylmalo- pathway. Thus, in the conversion of propionyl-CoA to nyl-CoA to (2S)-methylmalonyl-CoA, the substrate for succinyl-CoA, the first (carboxylation) step produces the the B12-dependent enzyme, methylmalonyl-CoA mutase. S epimer of methylmalonyl-CoA, whereas the third step, Defects in this pathway can result in severe acidosis catalyzed by methylmalonyl-CoA mutase, requires the and cause damage to the central nervous system in R epimer as a substrate [1]. The key intervening epimeri- humans. zation reaction is catalyzed by the enzyme methylmalonyl-CoA epimerase (MMCE: E.C. number 5.1.99.1), Results: The crystal structure of MMCE from Propioni- which interconverts (2S)-methylmalonyl-CoA and (2R)- bacterium shermanii has been determined at 2.0 A˚ reso- methylmalonyl-CoA. Likewise, since the incorporation lution. The MMCE monomer is folded into two tandem of 2-methylmalonate extender units into polyketide anti- ␤␣␤␤␤ modules that pack edge-to-edge to generate an biotics is also stereospecific [4], MMCE may play a role 8-stranded ␤ sheet. Two monomers then pack back- in determining this stereospecficity, depending on the to-back to create a tightly associated dimer. In each source of the 2-methylmalonate units. monomer, the ␤ sheet curves around to create a deep MMCE has been identified in both bacteria and ani- cleft, in the floor of which His12, Gln65, His91, and mals [6–7]. A search through currently available ge- Glu141 provide a binding site for a divalent metal ion, as shown by the binding of Co2ϩ. Modeling 2-methylmal- nomes using the sequence of the MMCE from Propioni- onate into the active site identifies two glutamate resi- bacterium shermanii [8] shows that homologs are dues as the likely essential bases for the epimerization present in many (but not all) microbial species, including reaction. both archaea and eubacteria; we found 16 examples with pairwise sequence identities of 30%–40%. The best Conclusions: The ␤␣␤␤␤ modules of MMCE corre- characterized MMCE is that from Propionibacterium spond with those found in several other proteins, includ- shermanii [6], which is a highly heat-stable enzyme with ing bleomycin resistance protein, glyoxalase I, and a a subunit size of 16.5 kDa (147 amino acid residues). Its family of extradiol dioxygenases. Differences in connec- activity has been proposed to be dependent on the tivity are consistent with the evolution of these very presence of two active site functional groups that can different proteins from a common precursor by mecha- act as acid/base catalysts, one to abstract a proton nisms of gene duplication and domain swapping. The from the asymmetric carbon atom, creating a planar metal binding residues also align precisely, and striking carbanion intermediate, and the other to protonate the structural similarities between MMCE and glyoxalase I carbon from the opposite face to accomplish the inver- suggest common mechanisms in their respective epi- sion of the chiral center [9]. MMCE is also inactivated merization and isomerization reactions. by treatment with metal-chelating agents and can be activated by incubation with divalent ions, notably by Introduction Co2ϩ, but also by Ni2ϩ,Mn2ϩ, and Zn2ϩ in the P. shermanii enzyme [6] and by Fe2ϩ,Mn2ϩ, and Co3ϩ in rat liver Methylmalonyl-CoA is an important metabolic interme- MMCE [7]. diate that links several key degradative and biosynthetic Functionally, MMCE appears to be related to mande- pathways [1–3]. In animals, the breakdown of lipids with odd-numbered carbon chains leads to the 3-carbon me- Key words: metalloenzyme structure; epimerization; methylmalonyl-CoA; protein evolution; vicinal oxygen chelate superfamily; 4 Correspondence: [email protected] metal-assisted catalysis Structure 638

Table 1. Refinement Statistics

Se-Met Native CoSo4 Resolution range (A˚ ) 20.0–2.0 20.0–2.2 20.0–2.9 Number of reflections 35,299 (3,930) 44,991 (5,042) 24,327 (2,681)

Rcryst (Rfree) (%) 22.8 (26.1) 24.8 (29.3) 30.1 (32.1) Rmsd from Ideal Geometry [45] Bond lengths (A˚ ) 0.015 0.007 0.007 Bond angles (Њ) 1.33 1.4 1.3

Number of Atoms Protein 4596 6898 9208 Water 90 129 0 2Ϫ 2Ϫ 2ϩ Ions 30 (6 SO4 ) 45(9SO4 ) 8 (8 Co ) Average B factors (A˚ 2) Main chain 27.7 24.6 20.4 Side chain 30.4 26.2 26.5 Water 23.1 20.9 — Ions 36.8 40.0 19.0

late racemase, whose activity also depends on a diva- prises 4596 protein atoms, covering residues 3–147 of lent metal ion and two active site acid/base groups [10], each molecule (the N-terminal residues 1–2 are not but structurally, it has no obvious relationship with this clearly visible), together with 90 water molecules and 6 or any other racemase. Recent crystal structures for sulfate ions. This model has good geometry (Table 1), several other epimerases have revealed remarkable di- with only 1.3% of the residues in the Ramachandran versity in their folding patterns [11–12]. MMCE has, how- plot being outliers, as defined by Kleywegt and Jones ever, been suggested recently to be a member of a [20]. Pairwise root-mean-square (rms) differences in C␣ diverse enzyme superfamily referred to as the vicinal positions between the four monomers range between oxygen chelate (VOC) superfamily, in which the mem- 0.37 A˚ and 0.54 A˚ . bers are linked by common mechanistic chemistry Subsequent to the refinement of the SeMet-substi- [13–15]. tuted MMCE structure, the crystal structure of the native We undertook the determination of the three-dimen- enzyme was refined against the slightly lower-resolution sional structure of MMCE in order to see whether this (2.2 A˚ ) native data, in a different crystal form containing constituted a new class of epimerase and to understand six MMCE monomers in the asymmetric unit. Full refine- its metal ion dependence and catalytic mechanism, its ment details for this structure are in Table 1. In both the high thermostability, and its evolutionary relationships. SeMet and native crystal structures, the packing of the The crystal structure, determined at 2.0 A˚ resolution, molecules indicates the presence of MMCE dimers (see reveals the metal binding site and suggests a mode of below), with two dimers per asymmetric unit for the substrate binding that leads to the identification of two SeMet form, three dimers for the native protein, and no putative catalytic acid/base groups. It also reveals a evidence of higher oligomers. striking structural and functional homology with glyoxa- Comparison of the SeMet-substituted and native lase I (GLO) [16], confirming MMCE as a member of the MMCE structures shows that there are no differences VOC enzyme superfamily, with common elements in its that are attributable to the substitution of selenomethio- enzymatic mechanism. The mode of assembly of the nine for methionine. The rms difference in main chain MMCE monomer, from two tandem ␤␣␤␤␤ units, also positions is similar to that for the crystallographically links it to an intriguing evolutionary pathway in this family independent molecules in the two structures, and that includes gene duplication, gene fusion, and domain changes in side chain conformations are generally seen swapping from a common precursor [17]. only for surface residues. Only at the one solvent- exposed methionine residue, Met50, is there any differ- Results and Discussion ence in side chain conformation or in the conformations of neighboring residues. We therefore use the higher- Structure Determination and Model Quality resolution SeMet structure in the following description The structure of P. shermanii MMCE was solved by and discussion. multiwavelength anomalous diffraction (MAD) methods [18], using crystals of the selenomethionine-substituted Monomer Structure protein. An initial atomic model was built into the 2.4 A˚ The MMCE monomer (Figures 1a and 1c) has an ␣/␤ resolution electron density map calculated from the fold that contains a striking two-fold internal repeat. It SeMet crystal data. This model, which comprised the is made up of 2 modules of about 60 residues each, four independent molecules present in the crystal asym- with each module having a ␤␣␤␤␤ fold in which the metric unit, was refined at 2.0 A˚ resolution using CNS connectivity of the ␤ strands is 1-4-3-2 (Figure 1b). The

[19] to final R factors of 22.8% (Rcryst) and 26.1% (Rfree, ␤ sheets of the two modules pack edge-to-edge, with calculated with 10% of the data). The final model com- the first strand of the N-terminal module antiparallel to Methylmalonyl-CoA Epimerase Crystal Structure 639

Figure 1. The Monomer Fold in MMCE (a) The monomer fold represented as a ribbon diagram. (b) The monomer fold represented as a topol- ogy diagram, showing the two ␤␣␤␤␤ modules. (c) The monomer fold represented as a C␣ plot with every tenth residue labeled. The N-terminal module is in red, the C-terminal module is in blue, and the connecting helix is in green. In (b), the boundaries of the secondary structural elements and the topologi- cal location of the metal binding residues are also shown.

the first strand of the C-terminal module. The resulting (residues 74–84) that connects module 1 to module 2 8-stranded ␤ sheet forms the core of the monomer, (Figures 1a and 1c). curving around to create a large cleft that houses the The two ␤␣␤␤␤ modules of the MMCE monomer show active site. Forming the floor of the cleft are the two very little sequence identity but extremely close struc- central strands ␤1 and ␤5 (the first strands of the two tural identity. When the N-terminal module (residues modules). At the top of the cleft, the prominent loop 7–72) is superimposed onto the C-terminal module (resi-

␤6-␤7 (residues 121–125) extends out toward the ␤2-␤3 dues 86–146), 42 C␣ atoms can be matched with an rms loop (residues 41–44) on the other side (Figures 1a and difference in atomic positions of 1.52 A˚ ; these come 1c); these loops are somewhat flexible, having relatively primarily from the three innermost strands of each mod- high B factors, and differ in their conformations between ule (␤1, ␤4, and ␤3 and ␤5, ␤8, and ␤7) and the helix that the various molecules in our crystals. One side of the joins the first strand (␤1 or ␤5) to the second (␤2 or ␤6). cleft is also partially sealed off by the 3-turn ␣ helix Differences occur between the two edge strands, ␤2 and Structure 640

mer, which is at the high end of dimer interfaces ana- lyzed by Jones and Thornton [21], clearly accounting for the very high dimer stability. Much of the interface is hydrophobic, with the side chains of Phe7, Ile10, Phe31, Trp33, Pro53, Leu89, Met92, Trp94, and Val138 among those buried between the monomers. There are also a total of eight hydrogen bonds made by the buried side chains of Asp5, Tyr15, Tyr26, His34, and Arg109.

Evolutionary Relationships The three-dimensional structure of MMCE reveals relationships that are not clearly apparent at the level of the amino acid sequence. Comparisons against the current structural database, using DALI [22], showed that MMCE belongs to a superfamily of proteins that includes glyoxalase I [16], bleomycin resistance protein (BRP) [23], and several bacterial extradiol dioxygenases [24–27]. These proteins have diverse functions, but have a common fold in which ␤␣␤␤␤ modules closely resembling those in MMCE are combined in several different ways [17]. Many, but not all, are enzymes in which the fold provides a platform for binding an essential metal ion [15, 17]. In Figure 2, we highlight the variety of ways in which these modules are used. The functional unit of each protein can be thought of as a “cradle” created by the edge-to-edge packing of the ␤ sheets of two ␤␣␤␤␤ modules; the resulting 8-stranded ␤ sheet curves around Figure 2. The MMCE Dimer, Compared with Related Structures to enclose the substrate binding cleft. In MMCE, the (a) The MMCE dimer, showing the “back-to-back” packing of the “cradle” is formed by a single 16.5 kDa polypeptide two ␤ sheets. The monomers are shown in red and blue, and the containing two tandem ␤␣␤␤␤ units. Two “cradles” then bound Co2ϩ ion is shown with a purple sphere. associate to form a back-to-back dimer. In GLO and (b and c) The domain-swapped bleomycin resistance protein (BRP) BRP, in contrast, each “cradle” is formed by ␤␣␤␤␤ units and human glyoxalase (GLO) dimers (monomers shown in red and from two different polypeptides, in a domain-swapped blue), with the Zn2ϩ bound to GLO shown as a purple sphere. (d) The single-polypeptide, 2-domain dioxygenase (HPPD) structure, dimer. In the dioxygenases, amongst which 4-hydroxy- with one domain in red, one in blue, and the single Fe2ϩ ion in purple. phenylpyruvate dioxygenase (HPPD) [26] is the closest structural homolog (as shown by DALI), a single polypeptide of about twice the length (35 kDa) contains four ␤ ␤ 6 ( 6 is quite irregular), and in the connecting loops. ␤␣␤␤␤ units that are organized into two domains; each The latter are also the main source of differences be- domain forms one “cradle”, although only one of these tween the independent monomers of the asymmetric has a functional active site. unit and between the native and SeMet structures. When It has been suggested that this family has arisen from the sequences of the two modules are compared on the an ancestral single-module ␤␣␤␤␤ protein, through mul- basis of this structural alignment, only nine residues are tiple gene duplication and fusion events [17]. A pro- identical (about 15% identity), and only two of these posed early event was the fusion of two ␤␣␤␤␤ modules identities, His12/His91 and Ala14/Ala93, seem to have to give an edge-to-edge-fused 2-module monomer. The clear structural or functional significance. Both are im- MMCE monomer is such a species. A back-to-back di- portant for the metal binding site. Other pairs of identi- mer, such as that represented by MMCE, can then be ties are not conserved in the MMCE sequences from seen as a likely precursor to the domain-swapped di- other organisms. mers typified by GLO and BRP. Alternatively, fusion of the two MMCE-like monomers and mutations that inacti- Dimer Structure vate one of the active sites provide a route to the 4-module, In solution, MMCE forms an extremely stable dimer. The 2-domain dioxygenases. These evolutionary pathways, same dimer organization is seen in both the native and illustrated in Figure 3, were predicted on the basis of SeMet MMCE crystals, as shown in Figure 2a. In this sequence and structure comparisons [17], but without structure, the two monomers pack back-to-back, by the an example of the supposed MMCE-like intermediate. stacking of their ␤ sheets, with their active sites opening The structure of MMCE thus provides a present-day outward in opposite directions. Each monomer active example of this key intermediate in an intriguing evolu- site is thus completely independent. The dimer interface tionary pathway. is extensive, involving (from each monomer) the central The individual ␤␣␤␤␤ modules can be superimposed strands ␤4, ␤1, and ␤5, the loops ␤3-␤4 and ␤7-␤8, and the across these structures with a high degree of fidelity, module 1 ␣ helix. Dimerization buries a total of 1710 A˚ 2 except that, as noted by Bergdoll et al. [17], the odd- of solvent-accessible surface from each monomer. This numbered modules always superimpose better on each represents 20% of the total surface area of each mono- other than on the even-numbered modules (and vice Methylmalonyl-CoA Epimerase Crystal Structure 641

tion of this structural conservatism, since it can appar- ently interconvert between the domain-swapped dimer form characteristic of human GLO and a metastable, but enzymatically-active, monomer [28]. The latter should resemble the MMCE monomer. At the level of the dimer, the similarity between MMCE and GLO persists, and using the same C␣ atoms for superposition, 200 C␣ atoms from the MMCE dimer match the equivalent atoms in the GLO dimer with an rms positional difference of 1.65 A˚ . This is a slightly larger figure than for the individual monomers, implying some slight adjustment, and when one pair of monomers is superimposed, it takes a rotation of 8.2Њ to fully match the other pair. We conclude that, in these proteins, it is the dimer that provides the common stable structural unit. Interactions between the two ␤␣␤␤␤ modules that make up the active site “cradle” seem insufficient to give such close correspondence between conventional (MMCE) and domain- swapped (GLO) versions; rather it is the extensive interface between the backs of the two “cradles” (the MMCE dimer interface) that stabilizes the overall structure. A very stable MMCE-like dimer would provide the basis for domain swapping without disturbing structure and function. Sequence identity across these structurally related Figure 3. A Schematic Diagram Showing a Potential Evolutionary proteins is low. Between MMCE and GLO, a structure- Pathway for This Protein Family This traces the family from the original duplication of a ␤␣␤␤␤ mod- based alignment gives 17 identical residues for human Identities .(%16ف) and 23 for E. coli GLO (%12ف) ule, through a conventional dimer, to a domain-swapped dimer or GLO a two-domain monomer. Present-day representatives of these latter are much fewer in the other proteins, however (only 9 entities are MMCE, GLO, and the dioxygenases, as indicated. residues are identical between MMCE and BRP), indicating that MMCE is significantly closer to GLO. Certain features are conserved; the helix in module 1 is N-capped versa). For example, the first module of MMCE superim- with an Asp (Asp19 in MMCE) and terminates with a poses on the first modules of GLO and BRP with rms Gly (Gly32), for example, and other sequence identities differences in atomic positions of 1.10 A˚ (58 C␣) and noted by Bergdoll et al. [17] for GLO, BRP, and DHBD 1.09 A˚ (44 C␣), respectively, but it superimposes on their are mostly conserved in MMCE. These can generally be second modules with rms differences of 1.53 A˚ (36 C␣) traced to specific signatures of the fold. and 1.62 A˚ (36 C␣), respectively. Likewise the second modules superimpose more closely on each other than on the first modules. This is consistent with the evolu- Metal Binding tionary scheme in which the gene for a two-module MMCE is activated by divalent metal ions, with the most protein like MMCE, with distinctive first and second effective being Co2ϩ and Mn2ϩ, and the overall order Zn2ϩ Ͼ Cu2ϩ Ͼ ف modules, is a likely intermediate in the evolution of all being Co2ϩ Ͼ Mn2ϩ Ͼ Ni2ϩ Ͼ Mg 2ϩ modern family members. The first and second modules Cd2ϩ for the P. shermanii enzyme [6]. No bound metal differ in the regularity of their outer strands, in their ion is present in either the native or SeMet MMCE struc- connecting loops, and in certain characteristic se- ture, and we attributed this to the absence of suitable quence features [17]. divalent metal ions in the crystallization media and/or Most remarkably, the MMCE structure shows how the relatively low pH of crystallization (4.2–4.9), which little the domain swapping seen in GLO and BRP affects might tend to protonate potential histidine ligands. Ac- the overall form of the active site “cradle” or indeed the cordingly, we have also cocrystallized MMCE with 1 mM full dimer. When the entire MMCE monomer is superim- Co2ϩ. These crystals were in a different crystal form, posed onto a comparable hypothetical GLO monomer related to that of the SeMet crystals but with eight mole- made up of module 1 from chain A and module 2 from cules in the asymmetric unit (Table 1), and diffracted chain B, 100 C␣ atoms can be superimposed with an more poorly. Nevertheless, the solution of the structure rms difference of 1.28 A˚ ; this is not significantly different by molecular replacement showed that the overall struc- from the matching of the same atoms of the individually ture is essentially unchanged, except in some of the optimized modules (58 C␣ at 1.10 A˚ for the first modules, mobile external loops, but with a single strong peak of and42C␣ at 1.34 A˚ for the second modules). That is, additional electron density at the same position in each the same active site “cradle”, with the same curvature, is of the eight monomers (Figure 4). This peak, at a height created whether it is formed within a single polypeptide of 8 ␴ in an Fo-Fc map, and 5.5 ␴ in a 2Fo-Fc map, repre- or by two domain-swapped modules. The GLO from Pseu- sents the bound Co2ϩ ion. domonas putida, in fact, provides an excellent illustra- The Co2ϩ ion is bound deep in the active site cleft, Structure 642

Figure 4. The Metal Binding Site in MMCE (a) The difference electron density peak for the bound Co2ϩ ion is shown, in coordinating distance of the side chains of His12, Gln65, His91, and Glu141. Virtually no movement of these coordinating groups occurs on metal complexation to the apo-protein. (b) The metal sites of MMCE and GLO are superimposed using only the C␣ atoms of the two proteins for superposition. For MMCE, the polypeptide backbone is in gray, with side chains and the Co2ϩ ion in blue; for GLO, the polypeptide backbone is in black, with side chains in red (monomer A) and gold (monomer B), and the Zn2ϩ ion in red.

coordinated to the side chains of His12 (N⑀2), Gln65 bound Zn2ϩ ion is coordinated to Gln33, Glu99, His126, (O⑀1), His91 (N⑀2), and Glu141 (O⑀1) (Figure 4). These and Glu172 [16], and in E. coli GLO, the metal ion (Co2ϩ, ligands come from topologically equivalent positions in Ni2ϩ,orCd2ϩ) binds to His5, Glu56, His74, and Glu122 the two modules (Figure 1b); His12 and Gln65 from the [29]. When MMCE is superimposed on either of the GLO first and fourth strands of module 1 and His91 and structures, as described above, the Co2ϩ ion of MMCE Glu141 from equivalent positions in the first and fourth is only 0.2 A˚ from the position of the Zn2ϩ ion in human strands of module 2. Strikingly, these correspond pre- GLO, and 1.3 A˚ from that of the Co2ϩ ion in E. coli GLO. cisely to the metal binding ligand positions in glyoxalase The arrangement of the protein ligands around the I and the dioxygenases; indeed it has been suggested Co2ϩ ion is the same as for the Zn2ϩ ion of human GLO [17] that the formation of a symmetric protein with the (Figure 4b) and for the Co2ϩ,Ni2ϩ,orCd2ϩ ions bound ability to bind a metal ion via four ligands was a crucial to E. coli GLO. His12, Gln65, and Glu141 are arranged step in the evolution of this family of proteins. The MMCE in a plane around the Co2ϩ, with His91 as an axial ligand structure further supports this view. In human GLO, the in an octahedral coordination geometry that leaves two Methylmalonyl-CoA Epimerase Crystal Structure 643

Figure 5. A Stereo Diagram Showing the Modeled Binding of 2-Methylmalonate to the Co2ϩ Ion in MMCE The 2-methylmalonate moiety is in gold (ball- and-stick representation), and the Co2ϩ ion is shown by a pink sphere. In the (2R)-epimer (shown), the C2 atom of 2-methylmalonate 3A˚ from Glu48, and on inversion of itsف is -3A˚ from the noncoف configuration would be ordinated carboxyl oxygen of Glu141, sug- gesting that these two residues are the two essential bases in P. shermanii MMCE. Thr122, from the loop 121–125, is also shown; in the crystal structure, this loop binds a sulfate ion, and it may form a flexible lid that closes over the bound 2-methylmalonate moiety (see text).

cis positions to be filled by water molecules. In the lower- striking. Both bind similar substrates (a thioester of 2-meth- resolution Co2ϩ-MMCE structure, no water ligands can ylmalonate for MMCE, and a thiohemiacetal of 2-oxoal- be reliably placed, but E. coli GLO has two (for the Co2ϩ, dehyde for GLO), both have enzymatic reactions that Ni2ϩ, and Cd2ϩ complexes) [29], and human GLO has involve proton transfer in their respective racemization one [16]. This geometry seems to be essential for proper and isomerization reactions, and both are dependent functioning. In human GLO, a transition state analog on a divalent metal ion for activity. The structural and binds to the Zn2ϩ by occupying two cis coordination functional similarities between MMCE and GLO thus positions and displacing the water ligand [30]. In E. coli enable us to propose a plausible model for substrate GLO, active metal ions (Co2ϩ,Ni2ϩ, and Cd2ϩ) have two binding to MMCE, in the absence of crystals of an water molecules occupying cis octahedral sites, but MMCE-substrate complex. when an “incorrect” metal ion (Zn2ϩ) is used, the geome- We modeled 2-methylmalonate, in its R epimer, into try is changed to trigonal bipyramidal, and activity is lost the active site of MMCE by assuming that it chelates [29]. In the apo-MMCE structures (native and SeMet), the the Co2ϩ ion, occupying the two “empty” cis-octahedral ligand residues are identically arranged to those in the positions (Figure 5). For GLO, a bound transition state Co2ϩ-bound structure, indicating that the protein fold analog binds likewise [30], and the same binding mode provides a stable, prepared platform for the binding of an has been proposed for MMCE from mechanistic similari- appropriate divalent metal ion in the correct geometry. ties in the VOC superfamily [15]. For 2-methylmalonate binding to MMCE, the two coordinating oxygens are the Substrate Binding and Catalysis carbonyl oxygen on C3 and one of the carboxyl oxygens Mechanistic studies on MMCE in tritiated water [9, 31] on C1, giving a 6-membered chelate ring. The distance are consistent with a two-base mechanism, such as is between these two oxygens in the Co(II)-(2-methylmalo- used in proline racemase, mandelate racemase, and nate) complex used for modeling [32] is 2.77 A˚ , almost many other racemases [10]. In this mechanism, one base identical with that found in the GLO-substrate analog removes the C2 proton from one side of the 2-meth- complex (2.63 A˚ ), further validating the model. ylmalonyl-CoA substrate, the C2 configuration inverts, Our model of 2-methylmalonate binding has three no- and the conjugate acid of a second base donates a table features. First, the position of the noncoordinated proton to C2 from the opposite side. A feature of this C1 carboxyl oxygen coincides with that of a sulfate ion mechanism is that the base that withdraws the proton that is found in all three MMCE structures. This sulfate is protected and does not exchange with solvent until ion sits between the metal ion and the ␤6-␤7 loop (resi- the product dissociates. MMCE catalyzes the racemiza- dues 121–125) that projects into the active site. In GLO, tion reaction in both directions, and the negligible levels the equivalent loop (152–159) moves down over the sub- of exchange into remaining substrate for either direction strate as a lid [27]. In MMCE, this loop is also flexible, therefore imply that both bases are monoprotic. as judged from its variability in the different crystal struc- The structural similarities outlined above place MMCE tures, and we propose that it likewise folds over the firmly in the same protein superfamily as GLO and the carboxylate group of the 2-methylmalonate substrate in dioxygenases. This family has been termed the vicinal MMCE. oxygen chelate (VOC) superfamily [13–15] because of Second, we identify two likely residues for the cata- evident similarities in the mechanistic chemistry of en- lytic bases in MMCE. On one side of the (2R)-methylma- 3A˚ from the C2 atom, in positionف ,zyme members; in each case, proton abstraction and lonate group is Glu48 transfer is involved, and substrate binding to a metal to abstract the proton (Figure 5). On the other side, the ion stabilizes the anionic intermediate. The functional noncoordinated carboxyl oxygen of the metal ligand 3A˚ away uponف 4A˚ from C2, and would beف similarities between MMCE and GLO are particularly Glu141 is Structure 644

Table 2. Data Collection and Phasing Statistics Data Collection Monochromatic MAD (SeMet)

Data type Native CoSO4 SeMet Inflection Edge Remote Wavelength (A˚ ) 1.08 1.54 1.08 0.9793 0.9792 0.9184

Space group P212121 P21 P21 P21 P21 P21 Unit cell dimensions a ϭ 56.0 A˚ a ϭ 88.0 A˚ a ϭ 43.6 A˚ b ϭ 114.0 A˚ b ϭ 85.0 A˚ b ϭ 78.6 A˚ c ϭ 156.0 A˚ c ϭ 91.0 A˚ c ϭ 89.4 A˚ ␤ϭ108.6Њ␤ϭ92.0Њ Molecules/asymmetric unit 6 84444 Maximum resolution (A˚ ) 2.2 2.9 2.0 2.1 2.1 2.1 Total observations 201,176 82,653 204,757 203,195 202,731 201,341 Unique reflections 50,685 27,983 41,070 67,306 67,407 67,288 Completeness (%)a 97.8 (85.8) 99.4 (97.5) 99.4 (96.1) 95.5 (97.1) 95.6 (96.8) 95.1 (96.0) a,b Rmerge (%) 4.3 (14.3) 11.9 (27.4) 6.2 (35.4) 4.7 (26.0) 4.8 (30.1) 4.7 (30.4) a Average I/␴I 31.6 (8.9) 9.1 (4.1) 25.0 (4.0) 21.1 (4.0) 20.8 (3.3) 21.5 (3.4) Phasing Phasing powerc Centric 1.0 1.0 1.0 Acentric 1.4 1.4 1.6 d Rcullis 0.61 0.62 0.79 FOM before solvent flattening 0.61 FOM after solvent flattening 0.85 a Figures for the outermost shell are given in parentheses; 2.28–2.20 A˚ for native, 3.0–2.9 A˚ for CoSO4 cocrystals, and 2.17–2.10 A˚ for MAD data. b Rmerge ϭ⌺h⌺i|Ih,i ϪϽIϾh|/⌺|ϽIϾh|. c Phasing power ϭϽFhϾ/ϽLOCϾ, where ϽFhϾ and ϽLOCϾ are the rms of the heavy-atom structure factor and lack of closure, respectively. d Rcullis ϭ⌺|||Fph (obs)| Ϯ |Fp (obs)|| Ϫ |Fh (calc)||/⌺||Fph (obs)| Ϯ |Fp (obs)||, where Fph,Fp, and Fh are the structure-factor amplitudes for the heavy-atom derivative, the native protein, and the heavy-atom contribution, respectively.

inversion of the C2 configuration, in position to donate its S epimer from propionyl-CoA, but is required in its a proton. This may be associated with transient dissoci- R epimer for the conversion to succinyl-CoA by the B12- ation of Glu141 from the metal, as is the case for the dependent enzyme methylmalonyl-CoA mutase. The in- corresponding residue in GLO, Glu172, which is impli- terconversion between R and S forms requires a speci- cated as an essential acid/base group in catalysis by fic metal ion-dependent epimerase, methylmalonyl-CoA that enzyme [30]. Of the two putative catalytic bases in epimerase (MMCE). MMCE, Glu141 is conserved in all MMCE sequences The crystal structure of MMCE shows that it forms a available. Glu48 is not fully conserved, but in those se- tightly associated dimer in which each monomer com- quences that lack Glu48, the Gln ligand that is trans to prises two ␤␣␤␤␤ modules. The modules pack edge- Glu141 in P. shermanii MMCE (Gln65) is replaced by Glu. to-edge to create an 8-stranded ␤ sheet that curves This would allow the noncoordinated carboxyl oxygen of around to create a deep active site cleft at the bottom this residue to act as the base instead, in a similar way of which is a specific metal binding site. In MMCE, a ϩ to Glu141. Co2 ion binds at this site. Third, the environment of the bound 2-methylmalo- The MMCE fold places it within a superfamily of pro- nate moiety is relatively hydrophobic, as it is buried teins that includes glyoxalase I (GLO), bleomycin resis- between the two ␤ sheet modules at the bottom of the tance protein (BRP), and bacterial extradiol dioxygen- binding cleft; this would account for the lack of solvent ases. These have diverse activities but a common exchange once the substrate is bound. The hydrophobic scaffold that is built up in different ways from ␤␣␤␤␤ environment emphasizes the importance of the metal modules like those in MMCE. The structure of MMCE ion for assisting catalysis by coordinating the substrate indicates that it is representative of a likely key interme- and helping stabilize the generated carbanion intermediate in the evolution of the domain-swapped family diate. members GLO and BRP and the duplicated dioxygenases. Remarkably, the active site “cradle” is unchanged between conventional (MMCE) and domain-swapped Biological Implications (GLO) family members. MMCE and GLO share a remarkable congruence in Substrate stereospecificity is an important feature of activity and mechanism, despite little sequence identity. many biochemical reactions. Interconversions between Both have very similar folds, almost identical metal sites propionyl-CoA and succinyl-CoA, during lipid and amino with similar metal preferences, and a flexible loop that acid breakdown, and the biosynthesis of polyketide anti- can fold down over the substrate. Both reactions involve biotics by some organisms, occur through the intermedi- proton abstraction and donation. MMCE binds a thioes- ate methylmalonyl-CoA. This metabolite is produced in ter substrate, and GLO binds a thiohemiacetal. These Methylmalonyl-CoA Epimerase Crystal Structure 645

similarities support the view that the demands of mecha- into 2Fo-Fc and Fo-Fc maps. The NCS constraints were relaxed during nistic chemistry have guided the evolution of this super- later stages, and the final cycles were carried out with no restraints. family and lead to plausible models for metal-assisted Restrained individual isotropic B factors were used for all atoms; final rms differences between the B factors of bonded atoms were substrate binding and catalysis in MMCE. 1.5 A˚ 2 for main chain atoms and 2.2 A˚ 2 for side chain atoms. Water molecules were located using the water pick protocol in CNS, refined Experimental Procedures with unit occupancy, and visually inspected to ensure they were within hydrogen bonding distance of potential hydrogen bond part- Protein Production and Crystallization ners. Individual B factors were refined with restraints between Methylmalonyl-CoA epimerase was obtained by the expression of bonded atoms. For refinement of the native and Co2ϩ-substituted the recombinant protein in Escherichia coli. The pTEEX plasmid structures, MMCE dimers were positioned in the respective unit encoding the P. shermanii MMCE gene was transformed into E. cells with AMoRe [40]. The native structure was refined using the coli strain BL21 (DE3), and the expressed protein was purified as protocol described for the SeMet structure. The Co2ϩ-substituted described [33]. A key step was a 15 min heat treatment at 70ЊCin structure was refined using a similar protocol, but with strict NCS which most E. coli proteins were denatured and substantially pure constraints throughout because of the lower resolution. Full details MMCE was obtained. The protein was then further purified by am- of the three refinements are given in Table 1. monium sulfate fraction, gel filtration, and ion exchange chromatog- raphy and concentrated by ultrafiltration to a final concentration of Other Methods about 20 mg mlϪ1. Calculations of solvent-accessible surface were made with AREA- Selenomethionine-substituted MMCE was prepared similarly, ex- IMOL, using a 1.4 A˚ radius sphere as a probe, and structural super- cept that the gene was cloned into the expression vector pProEXHt positions were performed with LSQKAB; both programs are from (Life Technologies), which results in a fusion to an N-terminal His- the CCP4 program suite [41]. In the structural superpositions, the tag with a tobacco etch virus (TEV) protease cleavage site. This C␣ atoms of the secondary structural elements were first superim- plasmid was transformed into the methionine auxotrophic E. coli posed. Additional C␣ atoms were then progressively added to or strain BL41 (DE3), the transformed bacteria were grown in LeMaster deleted from this structural “core”, using a 3␴ cutoff. Binding of the medium supplemented with 0.5 mM selenomethionine (SeMet), and (2R)-methylmalonate moiety was modeled into the active site by the SeMet-substituted MMCE was purified as described [33]. The assuming regular octahedral geometry for the cobalt atom, and histidine tag was then removed by TEV protease digestion. Mass bidentate chelate coordination as in the glyoxalase-transition state spectroscopic analysis of the SeMet-MMCE indicated that all seven analog complex [30]. The metal-ligand bond lengths and 2-methyl- Met were substituted by SeMet. malonate conformation were taken from a small-molecule Co(II) The native, Co2ϩ-substituted and SeMet-substituted proteins all complex of this ligand [32]. Hydrogen bonds were assigned using crystallized under similar conditions, but in different crystal forms the criteria of Baker and Hubbard [42]. Figures, unless otherwise (Table 2). Native crystals were grown by hanging-drop vapor diffu- indicated, were drawn with MOLSCRIPT [43], and rendered with sion at 18ЊC with reservoir solutions containing 0.2 M ammonium RASTER3D [44]. sulfate, 0.1 M sodium acetate buffered at pH 4.9, and varying amounts of PEG 2000 monomethyl ether as precipitant [33]. The SeMet-substituted protein crystals, grown at lower pH (4.2–4.3), Acknowledgments were smaller but diffracted to higher resolution (2.0 A˚ ). The Co2ϩ- substituted protein was crystallized using the native conditions, but This work was supported by the Marsden Fund of New Zealand. We thank Professor Peter Leadlay for very helpful discussions and with the addition of 1 mM CoSO4. Ashley Deacon and other staff of the Stanford Synchrotron Radiation Laboratory for expert help and advice with data collection. Data Collection and Reduction Crystals were transferred to a synthetic mother liquor with an in- creased PEG 2000 concentration before data collection and were Received: April 16, 2001 flash-frozen at an optimal concentration of 40% PEG 2000. All X-ray Revised: June 7, 2001 diffraction data were collected at 110 K. Multiwavelength data were Accepted: June 11, 2001 collected on Beamline 9.2 at the Stanford Synchrotron Radiation Laboratory (SSRL) using a 2 ϫ 2 ADSC Quantum 4 CCD detector. References Three wavelengths around the Se K-edge were selected on the basis of an X-ray fluorescence spectrum collected directly from the 1. Kaziro, Y., and Ochoa, S. (1964). The metabolism of propionic crystal. Additional data sets were collected for native and SeMet- acid. Adv. Enzymol. 26, 283–387. substituted crystals on SSRL Beamline 7.1 using a Mar345 imaging 2. Fenton, W.A., and Rosenberg, L.E. (1995). Disorders of propio- 2ϩ plate detector, and data sets for Co -substituted crystals were nate and methylmalonate metabolism. In The Metabolic and collected on a Rigaku RU-300 X-ray generator equipped with focus- Molecular Bases of Inherited Disease, C.R. Scriver, A.L. ing mirrors and a Mar345 MAR Research imaging plate. Data were Beaudet, W.S. Sly, and D. Valle, eds. (New York: McGraw-Hill), processed and scaled using DENZO and SCALEPACK [34]. Full pp. 1423–1449. statistics are given in Table 2. 3. Voet, D., and Voet, J.G. (1995). Biochemistry, Second Edition (New York: John Wiley & Sons). Structure Solution and Refinement 4. Marsden, A.F.A., Caffrey, P., Aparicio, J.F., Loughran, M.S., The positions of 23 selenium sites (from a possible 28) were deter- Staunton, J., and Leadlay, P.F. (1994). Stereospecific acyl trans- mined using SnB, version 2.0 [35]. The selenium positions were fers on the erythromycin-producing polyketide synthase. Sci- refined, and phases were calculated to 2.1 A˚ resolution using ence 263, 378–380. MLPHARE [36], followed by solvent flattening with DM [37]. The 5. Pfeifer, B.A., Admiraal, S.J., Gramajo, H., Cane, D.E., and final figure of merit was 0.85; other phasing details are contained Khosla, C. (2001). Biosynthesis of complex polyketides in a in Table 2. One molecule was built into the resulting 2.1 A˚ resolution metabolically engineered strain of E. coli. Science 291, 1790– electron density map using skeletonized density, main chain and 1792. side chain databases, and baton-building methods, as implemented 6. Leadlay, P.F. (1981). Purification and characterization of meth- in O [38]. The other three molecules in the asymmetric unit were ylmalonyl-CoA epimerase from Propionibacterium shermanii. then generated by applying refined noncrystallographic symmetry Biochem. J. 197, 413–419. (NCS) operators. Structure refinement was carried out with CNS 7. Stabler, S.P., Marcell, P.D., and Allen, R.H. (1985). Isolation and

[19], monitored by the Rfree [39] value (10% of data) and with the characterization of DL-methylmalonyl-Coenzyme A racemase application of NCS constraints. Refinement of the SeMet data set from rat liver. Arch. Biochem. Biophys. 241, 252–264. involved cycles of refinement interspersed with manual rebuilding 8. Davis, N.K. (1986). The structural genes for methylmalonyl-CoA Structure 646

metabolism in Propionibacterium shermanii. PhD thesis, Cam- lase I suggests a structural basis for differential metal activation. bridge University, Cambridge, UK. Biochemistry 39, 8719–8727. 9. Leadlay, P.F., and Fuller, J.Q. (1983). Proton transfer in meth- 30. Cameron, A.D., Ridderstro¨ m, M., Olin, B., Kavarana, M.J., ylmalonyl-CoA epimerase from Propionibacterium shermanii. Creighton, D.J., and Mannervik, B. (1999). Reaction mechanism Studies with specifically tritiated (2R)-methylmalonyl-CoA as of glyoxalase I explored by an X-ray crystallographic analysis of substrate. Biochem. J. 213, 635–642. the human enzyme in complex with a transition state analogue. 10. Powers, V.M., Koo, C.W., Kenyon, G.L., Gerlt, J.A., and Kozarich, Biochemistry 38, 13480–13490. J.W. (1991). Mechanism of the reaction catalyzed by mandelate 31. Fuller, J.Q., and Leadlay, P.F. (1983). Proton transfer in meth- racemase. 1. Chemical and kinetic evidence for a two-base ylmalonyl-CoA epimerase from Propionibacterium shermanii. mechanism. Biochemistry 30, 9255–9263. The reaction of (2R)-methylmalonyl-CoA in tritiated water. Bio- 11. Giraud, M.F., Leonard, G.A., Field, R.A., Berlind, C., and Nai- chem. J. 213, 643–650. smith, J.H. (2000). RmlC, the third enzyme of dTDP-L-rhamnose 32. Perkins, C.M., Rose, N.J., and Stenkamp, R.E. (1990). The struc-

pathway, is a new class of epimerase. Nat. Struct. Biol. 7, ture of cobalt methylmalonate complexes, CoClN4O7C10H28 and

398–402. CoCl2N4O5.5C11H28, models for metal complexes of ␥-carboxyglu- 12. Deacon, A.M., Ni, Y.S., Coleman, W.G., Jr., and Ealick, S.E. tamic acid. Inorg. Chim. Acta 172, 119–125. (2000). The crystal structure of ADP-L-glycero-D-mannohep- 33. McCarthy, A.A., et al., and Baker, E.N. (2001). Expression, crys- tose-6-epimerase: catalysis with a twist. Structure 8, 453–462. tallization and preliminary characterization of methylmalonyl co- 13. Babbitt, P.C., and Gerlt, J.A. (1997). Understanding enzyme su- enzyme A epimerase from Propionibacterium shermanii. Acta perfamilies: chemistry as the fundamental determinant in the Crystallogr. D 57, 706–708. evolution of new catalytic activities. J. Biol. Chem. 272, 30591– 34. Otwinowski, Z., and Minor, W. (1997). Processing of X-ray dif- 30594. fraction data collected in oscillation mode. Methods Enzymol. 14. Bernat, B.A., Laughlin, L.T., and Armstrong, R.N. (1999). Elucida- 276, 307–326. tion of a monovalent cation dependence and characterization 35. Weeks, C.M., and Miller, R. (1999). The design and implementa- of the divalent cation binding site of the fosfomycin resistance tion of SnB version 2.0. J. Appl. Crystallogr. 32, 120–124. protein (FosA). Biochemistry 38, 7462–7469. 36. Otwinowski, Z. (1991). Maximum likelihood refinement of heavy 15. Armstrong, R.N. (2000). Mechanistic diversity in a metalloen- atom parameters. In Isomorphous Replacement and Anomalous zyme superfamily. Biochemistry 39, 13625–13632. Scattering, W. Wolf, P.R. Evans, and A.G.W. Leslie, eds. (War- 16. Cameron, A.D., Olin, B., Ridderstrom, M., Mannervik, B., and rington, England: Daresbury Laboratory), pp. 80–86. Jones, T.A. (1997). Crystal structure of human glyoxalase I - 37. Cowtan, K., and Main, P. (1998). Miscellaneous algorithms for evidence for gene duplication and 3D domain swapping. EMBO density modification. Acta Crystallogr. D 54, 487–493. J. 16, 3386–3395. 38. Jones, T.A., Zou, J.Y., Cowan, S.W., and Kjeldgaard, M. (1991). 17. Bergdoll, M., Eltis, L.D., Cameron, A.D., Dumas, P., and Bolin, Improved methods for building protein models in electron den- J.T. (1998). All in the family: structural and evolutionary relation- sity maps and the location of errors in these models. Acta Crys- ships among three modular proteins with diverse functions and tallogr. A 47, 110–119. variable assembly. Protein Sci. 7, 1661–1670. 39. Bru¨ nger, A.T. (1992). The free R value: a novel statistical quantity 18. Hendrickson, W.A. (1991). Determination of macromolecular for assessing the accuracy of crystal structures. Nature 355, structures from anomalous diffraction of synchrotron radiation. 472–474. Science 254, 51–58. 40. Navaza, J. (1994). AMoRe: an automated package for molecular 19. Bru¨ nger, A.T., et al., and Warren, G.L. (1998). Crystallography & replacement. Acta Crystallogr. D 50, 157–163. NMR system: a new software suite for macromolecular structure 41. CCP4 (1994). The CCP4 (Collaborative Computational Project determination. Acta Crystallogr. D 54, 905–921. Number 4) suite: programs for protein crystallography. Acta 20. Kleywegt, G.J., and Jones, T.A. (1996). Phi/psi-cology: rama- Crystallogr. D 50, 760–763. chandran revisited. Structure 4, 1395–1400. 42. Baker, E.N., and Hubbard, R.E. (1984). Hydrogen bonding in 21. Jones, S., and Thornton, J.M. (1995). Protein-protein interac- globular proteins. Prog. Biophys. Mol. Biol. 44, 97–179. tions: a review of protein dimer structures. Prog. Biophys. Mol. 43. Kraulis, P.J. (1991). MOLSCRIPT: a program to produce both Biol. 63, 31–65. detailed and schematic plots of protein structures. J. Appl. Crys- 22. Holm, L., and Sander, C. (1993). Protein structure comparison tallogr. 24, 946–950. by alignment of distance matrices. J. Mol. Biol. 233, 123–138. 44. Merritt, E.A., and Bacon, D.J. (1997). Raster3D: photorealistic 23. Dumas, P., Bergdoll, M., Cagnon, C., and Masson, J.H. (1994). molecular graphics. Methods Enzymol. 277, 505–524. Crystal structure and site-directed mutagenesis of a bleomycin 45. Engh, R.A., and Huber, R. (1991). Accurate bond and angle resistance protein and their significance for drug sequestering. parameters for X-ray protein structure refinement. Acta Crys- EMBO J. 13, 2483–2492. tallogr. A 47, 392–400. 24. Han, S., Eltis, L.D., Timmis, K.N., Muchmore, S.W., and Bolin, J.T. (1995). Crystal structure of the biphenyl-cleaving extradiol Accession Numbers dioxygenase from a PCB-degrading Pseudomonad. Science 270, 976–980. The refined coordinates and structure factor amplitudes for the na- 25. Senda, T., et al., and Mitsui, Y. (1996). Three-dimensional struc- tive and SeMet-substituted MMCE structures have been deposited tures of free form and two substrate complexes of an extradiol in the Protein Data Bank, with entry codes 1JC5 (structure factor ring-cleavage type dioxygenase, the BphC enzyme from Pseu- amplitudes R1JC5SF) and 1JC4 (structure factor amplitudes domonas sp. strain KKS102. J. Mol. Biol. 255, 735–752. R1JC4SF), respectively. 26. Serre, L., et al., and Cohen-Addad, C. (1999). Crystal structure of Pseudomonas fluorescens 4-hydroxyphenylpyruvate dioxygenase: an enzyme involved in the tyrosine degradation pathway. Structure 7, 977–988. 27. Kita, A., et al., and Miki, K. (1999). An archetypical extradiol- cleaving catecholic dioxygenase: the crystal structure of cate- chol 2,3-dioxygenase (metapyrocatechase) from Pseudomonas putida mt-2. Structure 7, 25–34. 28. Saint-Jean, A.P., Phillips, K.R., Creighton, D.J., and Stone, M.J. (1998). Active monomeric and dimeric forms of Pseudomonas putida glyoxalase I: evidence for 3D domain swapping. Bio- chemistry 29, 10345–10353. 29. He, M.M., Clugston, S.L., Honek, J.F., and Matthews, B.W. (2000). Determination of the structure of Escherichia coli glyoxa-