Crystal Structure of Human 3-Hydroxy-3-methylglutaryl-CoA Lyase

3-Hydroxy-3-methylglutaryl-CoA (HMG-CoA) lyase is a key enzyme in the ketogenic pathway that supplies metabolic fuel to extrahepatic tissues. Enzyme deficiency may be due to a variety of human mutations and can be fatal. Diminished activity has been explained based on analyses of recombinant human mutant proteins or, more recently, in the context of structural models for the enzyme. We report the experimental determination of a crystal structure at 2.1 Å resolution of the recombinant human mitochondrial HMG-CoA lyase containing a bound activator cation and the dicarboxylic acid 3-hydroxyglutarate. The enzyme adopts a (βα)8 barrel fold, and the N-terminal barrel end is occluded. The structure of a physiologically relevant dimer suggests that substrate access to the active site involves binding across the cavity located at the C-terminal end of the barrel. An alternative hypothesis that involves substrate insertion through a pore proposed to extend through the barrel is not compatible with the observed structure. The activator cation ligands included Asn275, Asp42,His233, and His235; the latter three residues had been implicated previously as contributing to metal binding or enzyme activity. Arg41, previously shown to have a major effect on catalytic efficiency, is also located at the active site. In the observed structure, this residue interacts with a carboxyl group of 3-hydroxyglutarate, the hydrolysis product of the competitive inhibitor 3-hydroxyglutaryl-CoA required for crystallization of human enzyme. The structure provides a rationale for the decrease in enzyme activity due to clinical mutations, including H233R, R41Q, D42H, and D204N, that compromise active site function or enzyme stability.

The importance of the ketogenic cycle is underscored in hereditary HMG-CoA lyase deficiency, which can result in hypoketotic hypoglycemia and a marked increase in serum levels of several organic acids. Uncontrolled HMG-CoA lyase deficiency is lethal in about 20% of cases and can result in mental retardation, episodes of seizure, and coma (5). Several mutations in the HMG-CoA lyase gene correlating with deficiency have been identified (5), including the following missense mutations: H233R (6), R41Q, D42E, D42H, D42G (7), and recently E279K (8). These mutations have been studied by using our recombinant human lyase protein expression system, providing a more detailed biochemical explanation for the significant consequences of such mutations.
The cleavage of HMG-CoA requires the presence of a divalent cation such as Mg 2ϩ or Mn 2ϩ , and the reaction has been postulated to involve general acid/base catalysis (see next page Scheme 1). A number of active site residues, including divalent cation ligands, have been proposed on the basis of site-directed mutagenesis and characterization of mutants by kinetic and electron spin resonance studies. His 235 (9), Asp 42 , and Glu 72 (10) have been proposed to support divalent cation binding to protein. Asp 42 (10), His 233 (6), and Cys 266 (11) are important for catalytic efficiency. For D42A, V max is diminished by 1.3 ϫ 10 5 -fold compared with wild type enzyme. Estimated V max values for H233A and C266A indicate reductions of 6.4 ϫ 10 3 -and 1.3 ϫ 10 4 -fold, respectively. An additional important catalytic residue is Arg 41 . R41Q is reduced in V max by 2.7 ϫ 10 5 -fold relative to the wild type enzyme; the enzyme activity for R41M is so low that it could not be determined accurately by a sensitive radioactive assay. A role for Arg 41 in stabilization or protonation of the enolate form of the acetyl-CoA has been suggested (12). These residues are conserved in all HMG-CoA lyase sequences reported. Molecular sieve chromatography results and sedimentation equilibrium experiments indicate that the native enzyme is a dimer in solution (13). A reactive Cys 323 has been implicated in regulation of activity that involves the formation of an intersubunit disulfide crosslink (14).
Recently, a three-dimensional model proposing a substrate binding cavity for human HMG-CoA lyase has been reported (15). This model is interpreted to suggest that HMG-CoA binds through a central pore of the (␤␣) 8 barrel of the enzyme. However, most (␤␣) 8 barrel enzymes do not have a pore running through the middle of the barrel; instead at least one end of the barrel is blocked. The crystal structure of the bifunctional Pseudomonas enzyme 4-hydroxy-2-ketovalerate aldolase/acylating acetaldehyde dehydratase was recently published. The structure reveals a bound metal ion and an oxalate molecule (16). Sequence similarities to HMG-CoA lyase prompted use of this information to produce a homology model for HMG-CoA lyase (12). However, a detailed understanding of human clinical mutations and hypotheses concerning enzyme regulation and catalytic activity require experimental determination of a human HMG-CoA lyase structure. We now describe the complete three-dimensional structure of human HMG-CoA lyase determined at 2.1 Å resolution.

EXPERIMENTAL PROCEDURES
Expression and Purification of Wild Type HMG-CoA Lyase-The expression plasmid for human HMG-CoA lyase (pTRC-HL1) codes for the mature mitochondrial isoform lacking the transit peptide, except for additional Met-Gly residues at its N terminus (17). The N terminus of the mature human enzyme is threonine 28 (Fig. 1). The recombinant protein retains a Met-Gly sequence, remaining from the transit peptide, encoding an NcoI cleavage site used for cloning the construct into a pTrc99A vector. However, Edman degradation of isolated recombinant human enzyme indicated that the N-terminal methionine has been cleaved from ϳ80% of the protein. Therefore, Gly-Thr would be the expected N terminus in the lyase enzyme x-ray structure. The molecular mass for the mature 298amino acid peptide deduced from the cDNA is 32 kDa (18).
Human HMG-CoA lyase was expressed and purified as described previously by Roberts et al. (17). Briefly, the enzyme was purified from the crude Escherichia coli extract using three chromatographic steps as follows: a Q-Sepharose anion exchange column, a phenyl-agarose column, and a Superose 12 column. An ammonium sulfate precipitation step occurs between the first and second columns. Enzyme purity was assessed by SDS-PAGE and Coomassie Blue staining (19). The protein concentration was determined by the method of Bradford (20) using the protein assay reagent from Bio-Rad and bovine serum albumin as a standard. A subunit molecular mass of 32 kDa, the apparent subunit molecular weight determined on the basis of mobility upon SDS-PAGE and in agreement with the calculated mass deduced from the cDNA sequence, was used to calculate the molar concentration of the enzyme. HMG-CoA lyase enzymatic activity was determined by a citrate synthase-coupled assay (1) as modified by Kramer and Miziorko (21). The purified enzyme was concentrated to 8 -10 mg/ml and stored at Ϫ80°C (in 20 mM potassium phosphate, pH 7.8, 20% glycerol, 1 mM dithiothreitol, and 100 mM NaCl) until used for crystallization.
Crystallization and X-ray Data Collection-For crystallization experiments, the enzyme was diluted to the desired concentration, usually 5-6 mg/ml, using filtered water and filtered glycerol. The competitive inhibitor hydroxyglutaryl-CoA (HG-CoA) was added to the diluted protein at an ϳ1 mM concentration. The inhibitor was necessary for generation of uniform diffraction quality crystals. Inhibitor studies suggest that HG-CoA is not an efficient substrate for HMG-CoA lyase (22). Crystals sufficient for x-ray studies were obtained using an equilibration buffer of 0.1 M Hepes, pH 7.5, 60 mM MgCl 2 , and 15% polyethylene glycol 8K. The enzyme was mixed 1:1 with the equilibration buffer using sitting drop trays at 19°C. Crystals in the shape of a distorted cube with approximate dimensions of 0.2 ϫ 0.1 ϫ 0.1 mm formed overnight and continued to grow in size for an additional day.
X-ray diffraction data were collected from a single human lyase crystal to 2.1 Å resolution using synchrotron radiation at the Advance Photon Source, BioCARS beamline 14BM-C, at the Argonne National Laboratory. After soaking for about 2 min in a cryoprotectant solution containing 20% glycerol in mother liquor, the crystals were frozen in liquid nitrogen. The crystal-to-detector distance was 200 mm, and for each 0.5°oscillation, a 5-s exposure time was utilized. Diffraction data processing was carried out using HKL2000 (23). The crystals belong to the monoclinic space group C2 with unit cell parameters a ϭ 197.0 Å, b ϭ 117.1 Å, c ϭ 86.8 Å, and ␤ ϭ 112.5°. Six monomers were found in the asymmetric unit, with a corresponding Mathews coefficient of V m ϭ 2.2 Å 3 /Da (24), corresponding to a solvent content of 42%. The data collection statistics are given in Table 1.
Structure Determination and Refinement-The structure was determined by molecular replacement with MOLREP within the CCP4 program suite (25), with the monomer structure of Brucella melitensis lyase (Protein Data Bank code 1YDN) as the search model. An initial solution locating all six monomers yielded a correlation coefficient of 0.40 and an R-factor of 0.49 in the resolution range 20.0 -4.0 Å. At this point, the bacterial enzyme amino acid residues were replaced with those corresponding to the human enzyme. The structure was further refined using the program CNS (26) with manual adjustments between refinement cycles. In general, each cycle of refinement consisted of rigid body refinement, positional refinement, simulated annealing, and temperature factor refinement. After each cycle of CNS refinement, both 2F o Ϫ F c and F o Ϫ F c difference Fourier maps were calculated and used for manual fitting and rebuilding of the model using the program TURBO-FRODO (27). At later stages of refinement, water molecules were added at electron densities greater than 3 in the F o Ϫ F c map that were within at least 3.3 Å of a potential hydrogen-bonding partner. The final model gave a crystallographic R-factor of 0.226 and R-free of 0.265. The refinement statistics are given in Table 1.

RESULTS AND DISCUSSION
Overview of the Structure-Six monomers, designated as A-F, are present in the asymmetric unit of the crystal. These six monomers SCHEME 1

TABLE 1 Data collection and refinement statistics
The numbers in the parentheses are values for the highest resolution shell.

Data collection
Resolution/highest resolution shell, Å 50-2.1 (2.14-2.10) form three dimers (A and B; C and D; and E and F) and can be superimposed onto one another with an r.m.s. deviation of less than 0.5 Å for all main chain atoms. The overall G-factor calculated by Procheck (28), as a measure of stereochemical quality of the model, was 0.11. At the N terminus of the polypeptide, the first residue Gly 27 , a remainder from the transit peptide, was not observed in any of the six monomers. However, Thr 28 , the N terminus of the mature eukaryotic enzyme, was observed in all six monomers of the asymmetric unit. Residues Lys 137 -Glu 144 and Gly 265 -Gly 271 in monomer D as well as Ile 139 -Asn 140 and Cys 266 -Gly 271 in monomer F are disordered. The last two C-terminal residues Lys 324 and Leu 325 are disordered and not visible in all polypeptide chains. The Ramachandran plot reveals that more than 88.9% of the non-glycine and non-proline residues in all six monomers fall within the most favored region, and none were in a disallowed region. Monomer Structure-The monomer of human lyase includes 9 ␤-strands and 12 ␣-helices ( Fig. 1), with the overall polypeptide fold similar to that of the Brucella lyase (42) (r.m.s. deviation, 1.0 Å for 292 C-␣ atoms). The human monomer structure consists of a (␤␣) 8 TIM barrel motif and an additional polypeptide region made of residues 290 -323, which includes ␤9, ␣11, and ␣12 (Fig. 2). The internal cavity, seen from the C-terminal side of the barrel, was constructed by residues from the 8 ␤-strands (␤1-␤8) on the inner side of the TIM barrel and was enhanced in size by residues from ␣1, ␣5, and the three loops (between ␤2-␣3, ␤5-␣7, and ␤6-␣8). These latter structural elements were around the C-terminal end of the barrel and make the internal cavity deeper and wider at its entrance. The internal cavity is funnel-shaped with approximate dimensions of 10 Å in depth and 9 Å in diameter at its entrance. This cavity is larger than the internal cavities found in many TIM barrel enzymes whose substrates are smaller molecules than a CoA derivative (e.g. triose-phosphate isomerase (29)) but similar to those of methylmalonyl-CoA mutase (30) and malate synthase (31), both of which utilize a CoA derivative as substrate.

No. total reflections
Dimer Structure-For the six monomers in the asymmetric unit there are two different contact areas. Arranged as three dimers, the buried surface area calculated by CNS showed about 2400 Å 2 buried between two monomers, corresponding to 10% of the total surface area. Alternatively, two of the dimers can be arranged as a square-shaped tetramer using symmetry operations. For this second contact area, the buried surface area calculated between two monomers was 1022 Å 2 , less than 5% of the total surface area. These results indicated that the dimer formed by the first contact area corresponds to the physiological form of HMG-CoA lyase, which solution studies suggest exists as a dimer (13,17). Two different views of the physiological dimer are shown in Fig. 2C with the molecular dyad shown in the left panel. The two monomers ((␤␣) 8 barrels) in the dimer are arranged such that the axes of the two barrels are perpendicular to each other (Fig. 2C, right panel), as seen in the dimeric structures of triose-phosphate isomerase (29) and flavocytochrome b (32). The dimer contacts are made largely by residues in ␤9, ␣11, and ␣12 that are not part of the barrel (colored yellow in Fig. 2, A and B, and magenta and blue in Fig. 2C). In the lyase structure, each monomer forms its own barrel. In contrast, in the structure of TIM, a loop of one monomer crosses over to the other monomer forming part of the entrance to the other barrel.
Pseudomonas mevalonii HMG-CoA lyase, which has 52% sequence identity with human lyase, is inactivated by the affinity label 2-butynoyl-CoA (33) and is protected by the competitive inhibitor hydroxyglutaryl-CoA. These observations have been interpreted to indicate that Cys 266 , the target of the reagent, maps in the active site. Additional mutagenesis studies on Cys 266 (11) indicate that this residue influences catalytic efficiency (V max of C266A is 10 Ϫ4 of wild type) and are consistent with an active site assignment. In the human lyase structure, Cys 266 is situated in a loop between ␤8 and ␣10 located at the C-terminal side of the TIM barrel and forms a part of the barrel entrance. Residues in the N-terminal half of this loop (residues 259 -278) containing Cys 266 have the motif G(L/A)GGCP(Y/F), which is found in 47 known or putative HMG-CoA lyase proteins (data not shown). Because this loop is disordered in two of the six monomers in the asymmetric unit, it is most likely a flexible loop in solution. Perhaps, loop movement could occur upon substrate binding and could position Cys 266 to influence the catalytic efficiency of the enzyme.
Cys 323 , located at the C terminus, has been suggested to play a role in mammalian HMG-CoA lyase regulation involving a thiol/disulfide exchange mechanism, based on studies of avian and human lyase using the bifunctional cross-linking reagents dibromopropanone and o-phenylenedimaleimide (13,14). The regulatory sulfhydryls are proposed to form a disulfide bond that cross-links adjacent subunits and diminishes activity. Elimination of the Cys 323 sulfhydryl by mutagenesis blocks covalent dimer formation, supporting the involvement of this residue. Additional support derives from the observation that the P. mevalonii enzyme lacks a corresponding cysteine and is not sensitive to oxidation and covalent dimer formation. However, in the x-ray structure of the human lyase dimer, the distance between two Cys 323 residues on adjacent monomers of the dimer is 29 Å, too large to form a disulfide linkage between them. The inter-subunit distance between Cys 266 and Cys 323 of the adjacent subunit is 14 Å, which also appears to be too distant to support disulfide bond formation. On the other hand, the flexibility of the Cys 266 loop as observed in the crystal structure and the possible movement of the C-terminal tail may allow closer juxtaposition of either of these two cysteines in solution. Further studies are required to distinguish between these possibilities. Metal-binding Site-Electron density indicated a bound metal ion, most likely a Mg 2ϩ ion (Fig. 3A), as the crystallization medium included 60 mM MgCl 2 . The metal-binding site is found in the internal bottom of the (␤␣) 8 TIM barrel. Enzyme activity is absolutely dependent upon the presence of metals such as Mg 2ϩ or Mn 2ϩ (21). The metal ligand-bind-ing site displays octahedral coordination utilizing four conserved amino acid side chains and two water molecules (Fig. 3A), although not all six monomers in the asymmetric unit have both water molecules. The side chain ligands are the imidazole nitrogens of His 233 and His 235 , the carboxylate of Asp 42 , and the side chain amide oxygen of Asn 275 , with distances of about 2.2 Å for all coordinating atoms. In the crystal structure of 4-hydroxy-2-ketovalerate aldolase, a member of the HMG-CoA lyase family and a class II metal requiring aldolase, the ligands to bound manganese include His 200 , His 202 , Asp 18 , and a water molecule, liganding similar to that now found in human HMG-CoA lyase (16). The x-ray structure of isopropylmalate synthase, another member of the HMG-CoA lyase family exhibiting a TIM barrel structure, also reveals two histidines and an aspartic acid as metal-binding ligands (34). Metalbinding solution studies utilizing the lyase mutants H235A and D42N concluded that His 235 and Asp 42 are likely metal-binding residues (9, 10), consistent with the observed structure. Additionally, amino acids His 233 and Asp 42 have been identified in clinical mutations (6,35). In mutagenesis studies, H233A and D42A exhibit catalytic rates diminished by 6.4 ϫ 10 3 -and 1.3 ϫ 10 5 -fold, respectively, and thus, these metal liganding residues are clearly important for enzyme activity.
Substrate-binding Site-The active site is accessible only from the C-terminal side of the TIM barrel. The solvent-accessible surface of human HMG-CoA lyase shows that the N-terminal face is occluded (Fig. 3B, left), although the C-terminal face is accessible to the active site cavity (Fig. 3B, right). Residues found at the bottom of the barrel, closing off the N-terminal side of the barrel, include Val 36 , Val 38 , Glu 72 , Leu 106 , Val 125 , Arg 165 , Tyr 167 , Ser 201 , and Asp 257 . The substrate-binding cavity does not extend through the monomer as proposed previously by modeling studies (15). In malate synthase, a TIM barrel enzyme that catalyzes a homologous Claisen reaction, the acetyl-CoA-binding site is on the C-terminal side of the barrel (31). However, for methylmalonyl-CoA mutase, the methylmalonyl-CoA substrate accesses the active site, which is located at the C-terminal end of the barrel, from the N-terminal side of the barrel through a narrow tunnel along the axis of the TIM barrel domain (30). In the model for HMG-CoA lyase proposed by Casals et al. (15), the adenosine moiety of CoA substrate is positioned at the N-terminal side of the pore, which extends through the TIM barrel, although the HMG moiety of the substrate is on the C-terminal side of the barrel. However, our current structure of HMG-CoA lyase precludes this model, because the N-terminal side of the cavity is blocked as is shown in Fig. 3B, left. It is highly improbable that the enzyme adopts a different conformation, upon binding of the substrate, to form a pore through the TIM barrel. Such a change would require drastic structural refolding. Thus, in the current human lyase structure, the substrate can only access the active site from the C-terminal side (Fig. 2C and Fig. 3,  B and C).
Inside the active site cavity, in addition to the metal ion electron density, extra electron density (Ͼ3 level in the F o Ϫ F c map) in the vicinity of the metal ion site was observed and could be fitted with a model of 3-hydroxyglutaric acid (Fig. 3A). Hydroxyglutaric acid is presumably a hydrolysis product of the inhibitor 3-hydroxyglutaryl-CoA, which was included in the crystallization medium. Hydrolysis of the thioester linkage of the CoA derivative may have occurred during crystallization at pH 7.5 or during subsequent data collection. Arg 41 makes a salt bridge to the carboxylate of the hydroxyglutaric acid closest to the metal ion. This interaction suggests that Arg 41 might have a role in substrate binding or orientation, although the K m for R41Q (62 M) is not very different from that of wild type (48 M).
Mutation of Arg 41 results in the most drastic change in catalytic activity among all the lyase mutants studied so far. R41Q, which represents a This extra density (contoured at 3 level in the F o Ϫ F c map) was fit with 3-hydroxyglutaric acid, a hydrolysis product of the inhibitor 3-HG-CoA. A possible binding mode for the CoA ligand is modeled into the active site as discussed in the text. The adenosine moiety would be exposed to the solvent and is not shown. The ␣and ␥-carboxylates of the modeled HG-CoA are marked as C␣ and C␥, respectively. B, the solvent-accessible surface of human HMG-CoA lyase is shown, colored with the electrostatic potentials, blue for positive and red for negative. Left, the external view of the N-terminal side of the barrel is shown. The absence of a cavity on this side is indicated by the dashed yellow circle. Right, a view of the C-terminal face of the TIM barrel is depicted. The entrance to the active site cavity is marked with a red dashed circle. The negative potential (red) seen at the base of the cavity is because of the presence of Asp 42 . C, a cross-section of the putative substrate/inhibitor binding cavity with a ball-and-stick model of HG-CoA. Mg 2ϩ ion (blue sphere) and the observed hydroxyglutaric acid (thick yellow bonds) are shown. The cavity is large enough to accommodate a molecule of HG-CoA (modeled in as indicated by thin bonds). The surface figures were prepared using the program GRASP (41). clinical mutation, was found to be impaired by 2.7 ϫ 10 5 -fold in V max relative to the wild type enzyme (12). This extreme diminution of catalytic efficiency upon mutation of Arg 41 strongly suggests that this residue is directly involved in the chemistry of the catalytic reaction. Determination of the exact role of Arg 41 in catalysis requires the structure of the lyase complexed with a substrate analog or inhibitor. In addition to the salt bridge to hydroxyglutaric acid, Arg 41 makes a salt bridge/hydrogen bond with Glu 72 . Mutant E72A has both weakened substrate binding (ϳ10-fold) and metal binding properties (26-fold for Mg 2ϩ ; Ͼ200fold for Mn 2ϩ) and was postulated to be a metal ligand (10). Instead, the carboxylate of Glu 72 makes a 2.8-Å ionic interaction with the NH 2 of Arg 41 and a 2.7-Å interaction with the N ⑀ of Arg 41 . Thus, the role for Glu-72 includes charge balance for Arg 41 and orientation of Arg 41 for proper positioning in substrate binding and proper positioning of Arg 41 in a catalytic role. The 3-hydroxyl group of the bound hydroxyglutaric acid is making a hydrogen bond to a water molecule, which in turn hydrogen bonds to Gln 45 and Ser 75 . Whether this water molecule would act as the general base for the deprotonation reaction to form the corresponding C-3 oxyanion of the HMG-CoA requires further studies. Proper positioning of HMG-CoA may require contributions from the binding of the CoA moiety to the enzyme.
If indeed the HMG moiety of the substrate binds to the enzyme at the same position as observed with the hydroxyglutaric acid molecule, two binding modes for the CoA moiety can be envisioned, because either one of the two carboxyl groups of the glutaric acid can form a thioester bond. As shown in Fig. 3A, however, only one carboxylate (labeled as C␣ in Fig. 3A), the one distant from the metal ion, can form the thioester bond so that the CoA moiety can be accommodated in the cavity. It is not possible to model the CoA moiety as extending inward from the interior carboxylate (labeled as C␥ in Fig. 3A) of hydroxyglutaric acid toward the metal (there is not enough room). The CoA moiety has two common conformations, an extended conformation and a bent U-shaped conformation, which is seen in citrate synthase (37). In the bent conformation, the CoA molecule is bent at the pyrophosphate group such that the adenine group atoms are in van der Waals contact with atoms in the pantetheine moiety. The observed active site cavity is not wide enough to accommodate this bent CoA conformation (Fig. 3C), so the likely conformation is extended. The 5Ј-pyrophosphate and 3Ј-phosphate groups of the CoA moiety would be exposed to the surface of the enzyme, as seen in many CoA-binding enzymes, including two known CoA-binding TIM barrel enzymes, methyl malonyl-CoA mutase (30) and malate synthase (31,36). In this conformation of CoA with the thioester linkage to the C-␣ atom, not only are the phosphates exposed to the solvent surrounding the exterior surface of the enzyme but also 3-4 backbone atoms of the pantothenate moiety are exposed to the solvent, which is unlikely. One other possibility for HMG-CoA binding is to bring the substrate carboxylate closer to the metal from the observed hydroxyglutaric acid-binding position. In our current structure, the closest distance from the glutarate carboxylate (labeled C␥ in Fig. 3A) to the Mg ϩ2 ion is 7.1 and 5.5 Å to one of the two water molecules liganded to the Mg ϩ2 ion (W1 in Fig. 3A). There is no amino acid or ordered water bridging between the carboxylate and the liganded water molecule. Thus, a carboxylate oxygen from the ␥-carboxylate of the HMG-CoA substrate might replace a water ligand-binding site in the productive enzyme-metal-HMG-CoA complex. Alternatively, the carboxylate oxygens could hydrogenbond to the metal water ligands. Similar situations have been observed in the structures of 4-hydroxy-2-ketovalerate aldolase/dehydrogenase and isopropyl malate synthase, both members of the HMG-CoA lyase family. In the aldolase structure, the 1-carboxylate and 2-keto moieties of the substrate 4-hydroxy-2-ketovalerate bind the Mn 2ϩ ion as oxygen ligands (16). Perhaps HMG-CoA lyase may follow these models in which the carboxy-late of the substrate may bind the metal ion, as suggested by Tuinstra and Miziorko (10) from the metal binding studies with the H235A and D42N mutants. In such an arrangement with the carboxylate moiety of the substrate binding the metal ligand, the pantothenate moiety would be less exposed to solvent. Alternatively, because the entrance of the barrel is wider than the width of the straight chain of the pantetheine moiety, the CoA portion of HMG-CoA might adopt a somewhat less extended conformation so that the entire pantothenic moiety would fit inside the cavity. Resolution of these alternatives and the role of residues in catalysis must await a structure of the lyase with substrate or analog bound in the barrel cavity.
Clinical Mutations-A number of missense mutations for human HMG-CoA lyase have been characterized using the following recombinant mutant proteins: H233R, R41Q, D42E, D42H, and D42G (6,35). His 233 and Asp 42 are magnesium ion ligands in the active site of the lyase x-ray structure (Fig. 3A). Mutations of these metal liganding residues would likely impair the lyase reaction significantly, and biochemical and modeling studies of these mutant enzymes have demonstrated previously such impairment (6,10). Arg 41 is located in the active site of the lyase structure making a salt bridge to the C-␥ carboxyl group of the Hg-CoA hydrolysis product, hydroxyglutaric acid (Fig. 3A). Mutagenesis studies of Arg 41 identify it as a critical residue, and substitution of the basic arginine with either glutamine or methionine leads to a significant reduction in enzyme activity (12). Arg 41 has been implicated in stabilization or protonation of an enolized form of acetyl-CoA product. It is entirely reasonable, based on the current structure, that the carbonyl/ enoyl group of HMG-CoA substrate/acetyl-CoA product would interact with Arg 41 , when these CoA ligands bind to the enzyme.
Recently carboxylic acid mutations D204N (15) and E279K (8) have been implicated in human disease. In our structure, Asp 204 makes hydrogen bonds to amide nitrogens of Gly 207 , Val 208 , and Gly 209 , keeping the loop between ␤6 and ␣8 in its proper conformation. However, it is not obvious how changes caused by this mutation are structurally communicated into the active site. Because D204N could also make the same hydrogen bonds and no role for the negative charge of Asp-204 is apparent in the structure, it is more puzzling that D204N has diminished activity (14% of wild type V m (15)). Glu 279 is found on ␣-helix 10, which is not part of the (␤␣) 8 barrel. Rather this helix is at an interface between the external surface of the barrel and the extra domain, which consists of a ␤-strand and two ␣-helices found at the C terminus of the enzyme (Fig. 2). Its carboxylate side chain makes a short hydrogen bond (Ͻ2.5 Å) with the main chain carbonyl groups of Val 260 . The E279K mutation would disrupt this tight hydrogen bond and might also cause a steric effect, in addition to an ionic effect as the charge is changing by ϩ2. Thus E279K is likely a folding mutant, consistent with results obtained from the mutagenesis studies. When the E279K mutant was expressed in the in vitro system, an unstable recombinant protein resulted (10).
Two serine mutations, S75R and S201Y, have been identified, and the recombinant proteins were expressed (15). Neither of these enzymes was found to be active in the soluble fraction of crude extracts. Ser 75 is on ␤2 of the (␤␣) 8 barrel and is hydrogen-bonded to a water molecule, which in turn makes a hydrogen bond to the hydroxyl group of the organic acid, the hydrolysis product of HG-CoA. A change from serine to arginine eliminates the water molecule that might be important for catalysis. Alternatively, an arginine might interfere with substrate binding because of steric hindrance. Additionally, a positive charge is added to the active site. Ser 201 is located on ␤6, and for the mutation S201Y, there is no room for a phenolic side chain. Residues Arg 165 , Tyr 167 , Ala 231 , His 233 , and Asp 257 are present in the vicinity of Ser 201 . If this serine mutates to a tyrosine, a disruption of local folding would be expected. In the HMG-CoA lyase model of Casals et al. (15), HMG-CoA is modeled as passing through the N-terminal end of the TIM barrel and extending through the inner cavity to the C-terminal side of the barrel. In this model the S75R and S201Y are predicted to occlude this inner cavity, because of the longer length of the side chains, at positions where the CoA tail is predicted to bind (15). In our x-ray structure the N-terminal side of the TIM barrel is occluded.
Another reported clinical mutation is L263P (38). Assay for HMG-CoA lyase activity using patient skin fibroblasts with this mutation indicated a reduction in enzyme activity to 12% of wild type. Leu 263 is a conserved residue found in a loop between ␤8 and ␣10, residues 259 -278 (Fig. 2B). This loop contains the Cys 266 motif and the metal ligand Asn 275 . Leu 263 is far from the active site and exterior to the (␤␣) 8 barrel. Along with Leu 263 , there are a number of residues forming a hydrophobic patch, Leu 44 , Leu 59 , Val 260 , Ala 261 , Ile 306 , and Leu 310 . An amino acid change from leucine to proline could change the signature Cys loop conformation, distorting the local conformation and possibly creating an unstable protein.
Conclusions-The structure of human lyase with a bound metal ion and a molecule of 3-hydroxyglutaric acid (a hydrolysis product of the competitive inhibitor, 3-hydroxygluraryl-CoA) reveals the roles of amino acid residues that have been extensively studied by mutagenesis. In particular, His 233 , His 235 , Asp 42 , and Asn 275 have been identified as metal-binding ligands. Arg 41 is likely to be involved in binding of the substrate, stabilization of the anionic reaction intermediate, or perhaps directly involved in the chemistry of the catalytic reaction, as the general acid involved in the protonation of the C-2 enolate of acetyl-CoA after the cleavage of the C-2-C-3 bond. However, the exact identity of the general base that abstracts the proton from the C-3 hydroxyl group of HMG-CoA and the general acid that protonates the C-2 atom must await further structural studies. The current structure also reveals the structural implications of human mutations that are found in patients with lyase deficiency. Many of the clinical mutation sites are located at or very near the active site. His 233 and Asp 42 are liganded to the metal ion, and Arg 41 makes a salt bridge with a carboxylate of 3-hydroxyglutaric acid, suggesting an important role in substrate binding or orientation. On the other hand, others are located away from the active site, indicating that they are not directly involved in the catalysis. These mutants are most likely folding mutants, in which the mutation causes one or more conformational changes resulting in a less efficient enzyme (D204N) or an unstable protein.