Structure of the Methanococcus jannaschii mevalonate kinase, a member of the GHMP kinase superfamily.

The mevalonate-dependent pathway is used by many organisms to synthesize isopentenyl pyrophosphate, the building block for the biosynthesis of many biologically important compounds, including farnesyl pyrophosphate, dolichol, and many sterols. Mevalonate kinase (MVK) catalyzes a critical phosphoryl transfer step, producing mevalonate 5'-phosphate. The crystal structure of thermostable MVK from Methanococcus jannaschii has been determined at 2.4 A, revealing an overall fold similar to the homoserine kinase from M. jannaschii. In addition, the enzyme shows structural similarity with mevalonate 5-diphosphate decarboxylase and domain IV of elongation factor G. The active site of MVK is in the cleft between its N- and C-terminal domains. Several structural motifs conserved among species, including a phosphate-binding loop, have been found in this cavity. Asp(155), an invariant residue among MVK sequences, is located close to the putative phosphate-binding site and has been assumed to play the catalytic role. Analysis of the MVK model in the context of the other members of the GHMP kinase family offers the opportunity to understand both the mechanism of these enzymes and the structural details that may lead to the design of novel drugs.

The ability to make isopentenyl pyrophosphate is essential for all organisms, because this molecule is the building block for synthesizing many biologically important compounds, such as ubiquinone, cholesterol, dolichol, steroid hormones, and some vitamins (1). Two metabolic pathways, either mevalonatedependent or -independent, are available for the production of isopentenyl pyrophosphate. For mevalonate-dependent isoprenoid synthesis, assembly begins with the condensation of three acetyl-CoAs to generate a ␤-hydroxy-␤-methylglutaryl-CoA molecule, which is reduced to mevalonate. Mevalonate kinase transfers the ␥-phosphoryl group from ATP to the 5-hydroxyl oxygen in mevalonate to produce phosphomevalonate (Fig. 1). A second phosphoryl group is then added by phosphomevalonate kinase and the product, 5-pyrophosphomevalonate, is decarboxylated to isopentenyl pyrophosphate (1,2). Despite the existence of a mevalonate-independent pathway in some microorganisms, the process that generates isopentenyl pyrophosphate via mevalonate is critical in most eukaryotes and some microorganisms.
Kinases have been classified mainly into three distinct families according to their sequence similarities and structural folds (3). The classical P-loop-containing kinases share a common loop between a ␤-strand and an ␣-helix, with the consensus sequence Gly-Xaa 2 -Gly-Xaa-Gly-Lys (3,4). A second group, composed primarily of protein kinases (4,5), has a conserved loop connecting two antiparallel ␤-strands that are involved in phosphate binding. The third group, which includes actin, Hsp70, and sugar kinases (4,6), uses two ␤-hairpins to bind the phosphate in ATP.
Mevalonate kinase is a member of the distinct GHMP superfamily, which includes galactokinase (G), homoserine kinase (H), mevalonate kinase (M), and phosphomevalonate kinase (P) (4,7). 4-Diphosphocytidyl-2C-methylerythritol kinase, mevalonate 5Ј-diphosphate decarboxylase, and archaeal shikimate kinase have also been assigned as members of this family (4,8). It is important to note that shikimate kinases classified in yeast and prokaryotes have the P-loop kinase fold (8,9). These proteins share several structural features. A common motif, Pro-Xaa 3 -Gly-Leu-Gly-Ser-Ser-Ala-Ala, is highly conserved and is thought to be involved in the binding of ATP, functioning as a phosphate-binding loop. The crystal structure of the homoserine kinase, which shares about 20% sequence identity with the mevalonate kinase, has recently been solved, both as the apoprotein and as complexed forms (4,10). Its structure shows a novel left-handed ␤␣␤ fold similar to domain IV of elongation factor G (4). Similar to classic P-loop kinases, the phosphate binding site is composed of a conserved loop between a ␤-strand and an ␣-helix, with the consensus sequence Pro-Xaa 3 -Gly-Leu-Gly-Ser-Ser-Ala-Ala. Unexpectedly, the HSK 1substrate complex structures suggest that the orientation of the bound ATP to HSK is different from classical P-loop kinases, with the purine base located at what would be the phosphoryl acceptor substance binding site in a classic P-loop kinase (3)(4)(5)10). Moreover, ATP seems to adopt an energeti-* Use of the Argonne National Laboratory Structural Biology Center beamlines at the Advanced Photon Source was supported by the Office of Energy Research, United States Department of Energy, under Contract W-31-109-ENG-38. This work was supported by the Robert A. Welch Foundation and by a grant from the National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The cally less favorable syn conformation, which is not found in classic P-loop kinases (4,10). The high degree of sequence similarity in GHMP proteins suggests that other members may share at least some of the structural features of HSK (4,7).

EXPERIMENTAL PROCEDURES
Protein Expression and Purification-The Methanococcus jannaschii MVK sequence was initially cloned into pET28a vector with an Nterminal His tag and a T7 promoter, and the construct was used to transform B834(DE3) Escherichia coli strain (2). Cells were grown in LB medium containing 100 g/ml kanamycin at 37°C until A 600 was around 0.8. Cells were harvested, washed with M9 minimal medium (with 100 g/ml kanamycin and 50 g/ml 19 amino acids except Met), and then grown in this medium for 1 h at 37°C. Selenomethionine was added to 50 g/ml and isopropyl ␤-D-thiogalactopyranoside to 1 mM concentration. Cells were grown at 18°C for 20 h and then harvested and frozen at Ϫ20°C. Cells were thawed, re-suspended in 20 mM Tris-HCl, 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride, and 20 mM imidazole (pH 8.0), and then lysed by French press and sonication. The extract was heated to 70°C for 15 min to precipitate most E. coli proteins. The enzyme was purified to homogeneity, as judged by SDS-PAGE, on a nickel affinity column (AP Biotech) following the instructions of the supplier.
Crystallization and Data Collection-Crystallization was accomplished by the hanging drop vapor diffusion method. The MVK (10 to 15 mg/ml) in 5 mM Tris, 10 mM dithiothreitol (pH 8.0) was mixed with an equal volume of the precipitation solution (28% dioxane and 10 mM dithiothreitol) and equilibrated against 100 -200 l of reservoir solution in which components were the same as the precipitation solution. Single crystals grew overnight at 16°C but were generally very small. Well formed crystals were picked and transferred to newly set protein drops in the same precipitation solution except with the addition of 10 mM MgCl 2 . Larger crystals grew to a size of about 0.2-0.3 mm in three dimensions. All crystals belong to the space group P2 1 with cell dimensions a ϭ 56.138 Å, b ϭ 44.028 Å, c ϭ 64.845 Å, ␣ ϭ 90°, ␤ ϭ 102.71°, ␥ ϭ 90°, with one molecule in each asymmetric unit.
Data were collected at 121 K using a cryoprotection solution containing the original precipitant with additional 20% 2-methyl-2,4-pentanediol. Data were collected at beamline 14-BM-D at the Advanced Photon Source in Argonne National Laboratory. A four-wavelength (peak, edge, high energy remote, and low energy remote) MAD data set was collected from a single crystal (Table I). 1°oscillation widths were collected through a range of 360°for each wavelength. Data were integrated and reduced using DENZO (11) and SCALEPACK (11).
Phasing and Refinement-The structure was solved by the multiple wavelength anomalous dispersion (MAD) method. SOLVE (12-17) was used to determine the heavy atom position and calculate the initial phases. All six selenium sites were found by this method. Phases were improved by using density modification with RESOLVE (18,19). The map quality was sufficiently good to enable the complete model to be built in the program SPOCK (20). CCP4 programs, including FFTBIG and PROCHECK, have been used to generate the density map and check the final structure quality (21). Refinement was carried out with CNS (crystallographic ϩ NMR systems) (22). The current refinement statistics are listed in Table II.
Structure Analysis-DALI (23) was used to search the Protein Data Bank for proteins having folds similar to the MVK N-and C-domains. SwissPDB Viewer (24) was used to make structural alignments. The mevalonate molecule was generated and docked into the MVK cavity Green, the completely conserved residues; yellow, identical residues; cyan, similar residues. Three conserved motifs are highlighted. The residues labeled by asterisks are the equivalents of those indicated as important by mutagenesis studies in human or rat MVK.   using the AFFINITY module in the INSIGHT program (25,26). The ATP was initially modeled into the MVK cavity by superimposing the HSK-AMPPNP-homoserine structure with the apo-MVK structure and changing AMPPNP to ATP (9). The conformation of ATP is the same as that of AMPPNP; however, the ATP molecule was rotated slightly and translated to achieve a low energy between the ATP and the protein, calculated using INSIGHT. All other analysis of the structure and all figures shown were produced using SPOCK (20).

RESULTS AND DISCUSSION
Overall structure of MVK-Three very conserved motifs, designated motif I, II, and III, have been identified in GHMP proteins (4). The multiple sequence alignment of MVK from different species and the M. jannaschii HSK is shown in Fig. 2. Motif I is close to the N terminus of the protein, including residues corresponding to Lys 8 and Glu 14 of the M. jannaschii MVK. Motif II corresponds to the region between Pro 104 and Ser 114 of the M. jannaschii MVK, with an internal Gly-Leu-Gly-Ser-Ser-Ala sequence that is invariant, and forms the phosphate-binding loop. Motif III is characterized by Lys 272 -Cys 281 of the Methanococcus jannaschii MVK and has an invariant sequence Thr-Gly-Ala-Gly-Gly-Gly-Gly among various mevalonate kinases. In HSK, this region corresponds to the sequence Thr-Ile-Ser-Gly-Ser-Gly-Pro, a loop in close contact with the phosphate-binding loop, which is presumed to stabilize the conformation of the latter (4).
The structure of MVK contains all 312 residues of the wildtype MVK as well as five residues from the N-terminal histidine tag fusion. It is a monomer in crystal, although the dimer was reported in the aqueous solution (2). The R-factor has been refined to 0.20 with an accompanying R free value of 0.28 (Table  II). Nearly 88% of the non-glycine residues are in the most favorable conformation.
The C ␣ of MVK is shown in Fig. 3a. Similar to HSK, the MVK monomer contains two structural domains. At the domain interface there is a deep cavity presumed to be the active site (Fig. 3, a and b). Two molecules of dioxane, used as the crystallization precipitant, have been found to occupy this cavity (Fig. 5b). The structure of MVK is readily superimposable with that of HSK (Fig. 3c), indicating the structural conservation among the GHMP kinases. The root mean square deviation is 1.6 Å between the 155 superimposed C ␣ atoms.
The Structure of the N-terminal Domain and the Fold Comparison-The N-terminal domain of MVK comprises a large mixed five-stranded ␤-sheet (formed by strand a, b, c, d, and e) and a smaller anti-parallel four-stranded ␤-sheet (formed by strand a, b, f, and g), which are packed with four long ␣-helices (Fig. 4a).
The coordinates for the N-terminal domain (defined as residues 1-176) were used to search for similar structures by DALI (23). The best match was with 1fwk, the structure of HSK (4), with a Z-score 16.6. The second highest hit was 1fi4, the mevalonate 5-diphosphate decarboxylase (MDD) from Saccharomyces cerevisiae (27), with a Z-score of 9.7. Other top hits include 1dar, the elongation factor G (Z-score 5.4); 1b63, the Mutl protein (Z-score 5.0); and 1e3h, the guanosine pentaphosphate synthetase (Z-score 4.2).
Although the N-terminal domains of MVK, HSK, and MDD (shown in Fig. 4, a-c) are similar, there are substantial differences around the region corresponding to ␤-strand c in MVK. Compared with HSK, there is an insertion between ␤-strand c and ␣-helix D in MVK, which includes strand d and most of the following loop. MDD lacks the equivalent ␤-strands c and d of MVK, but instead it has a very long loop that links ␤-strand f (equivalent to strand b in MVK) with ␣-helix A (corresponding to helix D in MVK).
The Structure of the C-terminal Domain and the Fold Comparison-The structure of the C-terminal domain has a core unit that adopts a ␤␣␤␤␣␤ fold, formed with a four-strand ␤-sheet (strand h, i, j, and k) packed with two ␣ helices (L and M). There are three additional helices (I, J, and K) that are considered to be an insert between strand h and helix L (4).
DALI searches were performed using the coordinates of the C-terminal domain (defined as residues 177-312). The top match corresponds to 1fwk, the code for homoserine kinase (HSK), with a Z-score of 9.9, followed by 1fi4, the mevalonate 5-diphosphate decarboxylase, with Z-score 9.4. All other hits possessed Z-scores less than 5.0, including 1dar (elongation factor G, Z score 4.2), 1ha1 (a heterogeneous nuclear ribonu- FIG. 3. Overall structure of MVK. a, stereoscopic view of C ␣ of MVK labeled every 10 residues. Residue 1 is Met 1 , before which is a part of the His tag (five residues). b, MVK structure and its secondary structures. The N-domain is colored cyan and the C-domain green. The residues from the His tag are not shown. The helices are represented by capital letters and the strands by lowercase letters. c, stereoscopic view of C ␣ for the aligned structure of MVK (gray) and HSK (green; PDB code 1fwk). There is an ADP (cyan) molecule complexed with HSK. The His tag is not shown. cleoprotein, Z-score 4.0), and many others.
The C-terminal domains of MVK, HSK, and MDD are shown in Fig. 4, d-f. Each possesses a similar core unit with several insertions composed mainly of ␣-helices. Compared with those in HSK and MDD, the C-terminal tail in MVK is truncated and lacks the C-terminal ␤-strand (strands l and n in HSK and MDD, respectively) that folds toward the N-terminal domain and joins the large mixed ␤-sheet.
Homology Model of the Active Site of MVK-Mutations such as E19D, S145A, S146A, E193Q, D204A, and A334T in human MVK and K13M in rat MVK have been shown to decrease the activity of the enzymes (28 -31). In M. jannaschii MVK, Glu 14 , Ser 111 , Ser 112 , Glu 144 , Asp 155 , Ala 276 , and Lys 8 are identical to the equivalent residues of the human and rat protein, respectively (Fig. 2). Among the mutations studied, D204A elicits the most pronounced effects on catalysis by decreasing V m ϳ10,000-fold (28). These results helped to establish the probable mechanism of the enzyme and facilitated our analysis of the active cavity.
All of these conserved residues are located in a region that defines the active site cavity (Fig. 5a). For example, the residues corresponding to motif III form a Gly-rich loop very close to the phosphate-binding loop corresponding to motif II. Cys 281 in the Gly-rich loop is ϳ2.0 Å from Cys 107 in the phosphatebinding loop. The relative orientation of these two residues as well as a continuous electron density connecting their side chains suggest a disulfide bond, despite the existence of dithiothreitol in the crystallization condition. Formation of the disulfide linkage may stabilize the conformation of the putative phosphate-binding loop. In addition, Lys 272 in the Gly-rich loop is 3.1 Å away from Glu 14 , a motif I residue, which indicates electrostatic interactions.
The positions of the phosphate-binding loops of MVK and HSK (Fig. 5, b and c) are nearly identical when the proteins are superimposed. A dioxane molecule is bound about 3.0 Å from the main chain amides of Ser 111 and Ser 112 (which are both invariant residues on the phosphate-binding loop) and probably forms hydrogen bonds with these two amino acids. In various HSK-substrate complex structures, this region is occupied by phosphate groups and hydrogen bonds with the main chain amides on the phosphate-binding loop are present. For example, one HSK-ADP structure shows that the ␤-phosphate group occupies this position and forms hydrogen bonds with the main chain amides of Ser 97 and Ser 98 (equivalents of Ser 111 and Ser 112 ). Close to the phosphate-binding loop is the invariant residue Glu 144 , which aligns with Glu 130 in HSK, a residue that coordinates the Mg 2ϩ ion (10). These results reveal roughly the region in MVK where the phosphate tail of ATP may bind. In HSK, the purine ring is bound in a pocket close to the phosphate-binding site. A similar pocket is apparent in MVK that likely serves the same function. The backbone of Lys 74 -Cys 76 of MVK can be superimposed to the backbone of Asn 62 -Ala 64 of HSK. In HSK, these residues form van der Waals contacts with the purine ring (within 5 Å), and the main chain amide of Val 63 forms a hydrogen bond with N 7 in the purine. Although the exact amino acids are somewhat different, the similarity of the molecular surfaces and the constellation of protein atoms in this region suggest that it is likely that in MVK these corresponding residues may form similar interactions with the purine ring.
Based on the alignment of the structures of apo-MVK and HSK-AMPPNP-homoserine, an ATP molecule and a mevalonate molecule were docked into the cavity (Fig. 5d). The phosphate tail of the docked ATP molecule closely interacts with the phosphate-binding loop (within 5 Å) and forms hydrogen bonds with several main chain amides, including that of Ser 112 . The purine ring fits well into the putative purine-binding pocket, consistent with the analysis described above. The carboxyl group of the docked mevalonate molecule is near a large positively charged patch in the C-domain part of the cavity and forms hydrogen bonds with the main chain amide of Ala 276 and the side chain amide of Arg 196 . The 3Ј-methyl group of the mevalonate fits into a hydrophobic patch at the N-domain side of the cavity. In the model, the 3Ј-OH group is within hydrogen bond distance of the side chain of Lys 8 , a conserved residue in motif I. The 5Ј-OH group in the docked mevalonate is less than 3.0 Å from the ␥-phosphorus of the docked ATP molecule. The 5Ј-OH group of the modeled mevalonate is only about 4.0 Å from Asp 155 . The model indicates a possible reaction mechanism that is consistent with previous enzymological studies on MVK.
Possible Reaction Mechanism-The reaction is likely to be initiated by the nucleophilic attack to the ␥-phosphorus of ATP by the 5Ј-OH of mevalonate (Fig. 6). Based on the analysis of the active cavity, the best candidate for the catalytic residue is Asp 155 (located about 4.0 Å from the 5Ј-OH of the docked mevalonate), which is consistent with the mutation data available for human MVK (28). It is likely that Asp 155 functions as a general base with its side carboxyl group in position to abstract a proton from the 5Ј-OH of mevalonate, thereby facilitating the nucleophilic attack. Notably, this residue is struc- turally aligned with Asn 141 in HSK, which has been shown to interact directly with the ␦-OH of the homoserine (10). Lys 8 , Ala 276 , and Arg 196 appear to be key to the binding of meval-onate (Fig. 6). The phosphate-binding loop might be the main force to bind and stabilize the phosphate tail of ATP. The N terminus of ␣-helix E is pointing toward the phosphate groups and may offer some stabilizing effect during phosphoryl transfer, as is the case in many other kinases (3,4). The Mg 2ϩ that might be coordinated by Glu 144 is likely to activate the ␥-phosphate. This mechanism is consistent with the model proposed by the Miziorko laboratory (29).
In conclusion, our work has defined the structure of M. jannaschii mevalonate kinase and revealed the structural conservation among GHMP proteins. As more GHMP protein structures are solved, the detailed comparison and analysis of these proteins with substrates and inhibitors will further our understanding of the enzymatic mechanisms and offer us opportunities to design potential drugs against various diseases.