Crystal Structures of N-Acetylglucosamine-phosphate Mutase, a Member of the α-d-Phosphohexomutase Superfamily, and Its Substrate and Product Complexes*

N-Acetylglucosamine-phosphate mutase (AGM1) is an essential enzyme in the synthetic process of UDP-N-acetylglucosamine (UDP-GlcNAc). UDP-GlcNAc is a UDP sugar that serves as a biosynthetic precursor of glycoproteins, mucopolysaccharides, and the cell wall of bacteria. Thus, a specific inhibitor of AGM1 from pathogenetic fungi could be a new candidate for an antifungal reagent that inhibits cell wall synthesis. AGM1 catalyzes the conversion of N-acetylglucosamine 6-phosphate (GlcNAc-6-P) into N-acetylglucosamine 1-phosphate (GlcNAc-1-P). This enzyme is a member of the α-d-phosphohexomutase superfamily, which catalyzes the intramolecular phosphoryl transfer of sugar substrates. Here we report the crystal structures of AGM1 from Candida albicans for the first time, both in the apoform and in the complex forms with the substrate and the product, and discuss its catalytic mechanism. The structure of AGM1 consists of four domains, of which three domains have essentially the same fold. The overall structure is similar to those of phosphohexomutases; however, there are two additional β-strands in domain 4, and a circular permutation occurs in domain 1. The catalytic cleft is formed by four loops from each domain. The N-acetyl group of the substrate is recognized by Val-370 and Asn-389 in domain 3, from which the substrate specificity arises. By comparing the substrate and product complexes, it is suggested that the substrate rotates about 180° on the axis linking C-4 and the midpoint of the C-5—O-5 bond in the reaction.

cell wall is synthesized from UDP-GlcNAc. A specific inhibitor of AGM1 from pathogenetic fungi could be a new candidate for an antifungal reagent that inhibits cell wall synthesis. Such a reagent would be expected to have fewer side effects because it acts on the cell wall, which does not exist in humans. In this article, we report the crystal structures of AGM1 from C. albicans (CaAGM1; 544 residues, 60 kDa) for the first time, both in the apoform and in complexes with substrates (Zn 2ϩ /PO 4 3Ϫ /GlcNAc-6-P) and products (Zn 2ϩ /PO 4 3Ϫ /GlcNAc-1-P), and discuss the catalytic mechanism. Although the reaction catalyzed by AGM1 is reversible, in this study we assigned GlcNAc-6-P and GlcNAc-1-P as the substrate and product, respectively, to avoid confusion.

EXPERIMENTAL PROCEDURES
Purification, Crystallization, and Data Collection-CaAGM1 was overproduced, purified, and crystallized as described previously (6,15). CaAGM1 was expressed as a glutathione S-transferase fusion and purified with a glutathione-Sepharose 4 FastFlow column (Amersham Biosciences) and a DEAE-Toyopearl 650 column (Tosoh, Tokyo). Crystals were obtained within a week using a reservoir solution (pH 4.6) containing 200 mM NH 4 H 2 PO 4 and 14 -20% (w/v) polyethylene gly- FIGURE 1. Schematic illustration of the role of AGM1 in the biosynthetic pathway of UDP-GlcNAc. In eukaryotes, UDP-GlcNAc is synthesized from fructose 6-phosphate by four successive reactions: (i) the conversion of Fru-6-P into GlcN-6-P; (ii) the acetylation of GlcN-6-P into GlcNAc-6-P; (iii) the interconversion of GlcNAc-6-P and GlcNAc-1-P; and (iv) the uridylation of GlcNAc-1-P into UDP-GlcNAc. AGM1 is catalyzed in step iii. In prokaryotes, the intramolecular phosphoryl transfer (step iii) occurs before acetylation (step ii), and GlcN-1-P is generated as the intermediate. The substrate and product complexes were prepared by soaking in cryoprotectant 1 with each ligand (25 mM) and 5 mM ZnCl 2 for 1 min. All crystals were cryocooled in an N 2 gas stream at 95 K. Diffraction data were collected using synchrotron radiation at SPring-8 (BL44B2 with a Mar CCD detector) and at the Photon Factory (BL5 with an ADSC Quantum 315 CCD detector and BL6A with an ADSC Quantum 4R CCD detector) and also using Cu-K␣ radiation from a rotating anode generator with RAXIS-IV and RAXIS-VII imaging plate detectors. All data sets were processed and scaled using the programs DENZO and SCALEPACK from the HKL2000 package (16).
Phasing, Model Building, and Refinement-The crystal structure of the apoenzyme was solved by the multiple isomorphous replacement method using the programs SOLVE (17) and RESOLVE (18). The structural model was constructed using the program O (19). The crystal structures of the substrate and product complexes were determined by the molecular replacement method using the apoenzyme structure as a search model by the program MOLREP (20) in CCP4i (21,22). All structures were refined using the program CNS (crystallography NMR software) (23). The atomic coordinates of CaAGM1 have been deposited in the Protein Data Bank under accession numbers 2DKA (apoenzyme), 2DKC (substrate complex), and 2DKD (product complex). The accessible surface area was calculated using the program SURFACE (24,25) in CCP4 (21). Structures were superimposed using the program LSQMAN (26). Figures of structures were prepared with the program MOLSCRIPT (27) and Raster3D (28).

RESULTS AND DISCUSSION
Overall Structure of AGM1-The crystal structure of CaAGM1 was determined at 1.93 Å resolution. Final statistics for all data sets are shown in Table 1. In all structures, approximately 90% of the residues are in the most favored region of the Ramachandran plot defined by PROCHECK (22,29). It is suggested that the two molecules in the asymmetric unit of the crystal, which are related by noncrystallographic 2-fold symmetry, do not form a biological dimer, because each molecule of this pair buries only 4% of its total solvent-accessible surface in intermolecular contacts, an area outside the range expected for oligomeric proteins (30). Gel-filtration chromatography analysis (data not shown) also supported this hypothesis.
AGM1 contains domains 1 (residues 1-191), 2 (residues 192-311), 3 (residues 312-456), and 4 (residues 457-544). They are arranged in a "heart shape" (Fig. 2, A and B). Domains 3Ϫ ion are shown as balland-stick models, where carbon, oxygen, nitrogen, and phosphorus atoms are presented in black, red, blue, and magenta, respectively. A Zn 2ϩ ion is shown as a purple sphere. C, topology diagram of CaAGM1 with the same coloring scheme as in A. Helices are indicated by rectangles and ␤-strands by arrows. The active-site regions are marked with asterisks. In domains 1-3, the four common strands are numbered, and helices and strands that are not part of the common structure are shown in lighter colors.
1-3 fold similarly, with four ␤-strands located between two ␣-helices (Fig. 2C). Domain 4 has a completely different fold, which contains two antiparallel ␤-sheets. It has less interaction with the other domains and thus is more mobile. In the electron density map of the apoenzyme, domain 4 in one of two noncrystallographic symmetric molecules is disordered. However, in the structures of the substrate and product complexes, the electron density of domain 4 is clearly observed because of the binding of the substrate.
Comparison with Phosphomannomutase/Phosphoglucomutase (PMM/PGM)-PMM/PGM is also a member of the ␣-Dphosphohexomutase superfamily and catalyzes the reversible conversion of mannose 6-phosphate into mannose 1-phosphate and glucose 6-phosphate into glucose 1-phosphate (12,13). PMM/PGM from Pseudomonas aeruginosa (PaPMM/ PGM) has 463 residues and shares 20.7% sequence identity with CaAGM1 (Fig. 3) as calculated by the program FASTA (31, 32). The superposition of CaAGM1 (the product complex) with PaPMM/PGM (the complex with mannose 1-phosphate; Protein Data Bank code 1PCJ) is shown in Fig. 4A. These two structures, which are essentially identical, are composed of four domains arranged in a heart shape. In particular, the activesite residues are well superimposed (Fig. 4B). For AGM1, domain 4 has two additional ␤-strands (the red region in Fig. 4A), which were not observed in the structures of the superfamily. These strands are superimposed with an ␣-helix in PMM/PGM. Circular Permutation-For AGM1 and PMM/PGM, domains 1-3 are structurally similar to each other, each having four ␤-strands located between two ␣-helices. The topology The topology diagram is drawn as described for Fig. 2C. First, the core structure of PMM/PGM is selected. Next, the N and C termini are fused. The bold red line is cleaved to form the new N and C termini. As a result, the structure obtained is the same as the core structure of AGM1. C␣ traces of the apoform, the substrate complex, and the product complex are shown in green, red, and blue, respectively. The flexible active-site loop observed only in the substrate and product complexes is shown by a red dot circle. GlcNAc-6-P (the substrate) and GlcNAc-1-P (the product) are shown in magenta and white, respectively. PO 4 3Ϫ and Zn 2ϩ ions are shown as described for Fig. 2A. B, a macrograph about the substrate and the product. The N-acetyl groups are indicated by blue circles. The axis of rotation is indicated by the green line. diagrams for domains 2 and 3 are essentially the same between these proteins. The ␤-strands are arranged in the order of 2-1-3-4, and strand 4 is antiparallel to the other three strands. However, the diagram for domain 1 is completely different, as shown in Fig. 4C. In this domain, the strands are arranged in different order. For AGM1, the ␤-strands are arranged in the order of 4 -3-1-2, and strand 2 is antiparallel to the other three strands. For all of the other proteins in the superfamily, the order is the same as in PMM/ PGM (11). The active-site residues are well superimposed, and the sequences around the active-site residues are conserved between AGM1 and PMM/PGM (Fig. 3). It is known that circular permutation can occur by the spatial adjacency of the N and C termini of the polypeptide chain (33,34). This mutation involves the fusion of the N and C termini and cleavage at another site. Such a mutation is observed in domain 1 of this enzyme (Fig. 4C). The distance on the amino acid sequence between the active-site residues of domains 1 and 2 is 224 residues in AGM1, whereas it is only 134 residues in PMM/PGM. This large gap suggests that AGM1 evolutionarily diverged from the superfamily in the earliest period (35). Although it has been suggested previously that an insert sequence probably exists (36), our structural analysis demonstrates that the difference lies in the strand order.
Active Site-The crystal structures of the complexes with substrates (Zn 2ϩ /PO 4 3Ϫ /GlcNAc-6-P) and products (Zn 2ϩ / PO 4 3Ϫ /GlcNAc-1-P) were also determined at 2.20 and 2.10 Å resolution, respectively (Fig. 5A). The PO 4 3Ϫ ion was found in these complexes but not in the apoform. The substrate (GlcNAc-6-P) and product (GlcNAc-1-P) molecules were observed in the active site. The position of the active site in this enzyme is similar to those found in other enzymes belonging to the ␣-Dphosphohexomutase superfamily (35,36). The catalytic cleft is formed by four loops from each domain (Fig. 6): the active serine loop (Thr-64 -Glu-71 in domain 1), the metal-binding loop (Asp-290 -Arg-295 in domain 2), the sugar-binding loop (Glu-387-His-391 in domain 3), and the phosphate-binding loop (Arg-512-Ala-519 in domain 4). The residues identified as affecting the activity by previous mutation analysis are located on all of these loops except the sugar-binding loop (6).
The active serine loop contains the catalytic residue Ser-66 (Fig. 7, A and  B). This residue is observed as the phosphorylated form in the superfamily (11,12). However, judging from the electron density maps, it is not phosphorylated in all structures of the apoform and the substrate and product complexes. In the substrate and product complexes, the O␥ atom of Ser-66 is not covalently bonded to the PO 4 3Ϫ ion. In the apoenzyme, Ser-66 is disordered in one of two noncrystallographic symmetric molecules. On the other hand, in the substrate and product complexes, it is clearly observed but is in the disallowed region of the Ramachandran plot defined by PROCHECK (22,29) under the influence of the Zn 2ϩ ion. The active serine loop is fixed by the Zn 2ϩ ion through Ser-66.
The metal-binding loop plays an important role (Fig. 7, A and  B). A metal ion is essential for the activity of AGM1. In the apoenzyme, no metal ion is observed at this position, and the side chains of three aspartic acid residues (Asp-290, -292, and -294) on this loop point to various positions. On the other hand, in the substrate and product complexes, the Zn 2ϩ ion is chelated to the side chains of these residues. In both cases, the above features of the aspartic acid residues are clearly observed in the electron density maps. The charge of the metal ion becomes more positive in the transition state, and probably the PO 4 3Ϫ ion is smoothly transferred between the O␥ atom of Ser-66 and the O-6 or O-1 atom of the substrate (12).
The sugar-binding loop binds to the hydroxyl groups of the substrate (Fig. 7, C and D). In the substrate and product complexes, the side chain of Glu-387 is hydrogen-bonded to the O-3 and O-4 atoms of the sugar ring. In addition, one water molecule is hydrogen-bonded to the O-4 atom in the substrate complex, whereas the side chain of Asn-389 is hydrogen-bonded to the O-3 atom in the product complex. Compared with the substrate and product complexes, the sugar rings of the substrate (GlcNAc-6-P) and product (GlcNAc-1-P) molecules are related by a 180°rotation around an axis linking the C-4 atom and the midpoint of the C-5-O-5 bond (Fig. 5B). It is suggested that the substrate (the reaction intermediate) rotates 180°around the axis in the reaction.
The phosphate-binding loop, located at the opposite side of the active serine loop, interacts with the phosphoryl group of the substrate (Fig. 7, E and F ). This loop is particularly flexible, because in the apoenzyme this loop is not visible in the electron density map. In the substrate and product complexes, however, it is clearly observed (the red dot circle in Fig. 5A). It also plays the role of covering the active-site cleft. The phosphoryl group of the substrate is hydrogen-bonded to the side chains of Arg-512, Ser-514, Gly-515, Thr-516, and Arg-521. The binding modes in the substrate and product complexes are very similar to each other. In addition, two water molecules are hydrogen-bonded to the phosphoryl group in the substrate complex. But only one water molecule binds to the phosphoryl group in the product complex. The N-acetyl group of the substrate in the product complex is replaced by the water molecule in the substrate complex.
Substrate Specificity-The N-acetyl group of the substrate is responsible for the substrate specificity (Fig. 7, G and H). In the substrate complex, the carbonyl oxygen atom of the N-acetyl group is hydrogen-bonded to the main chain N atom of Lys-371  through a water molecule. In addition, the N-acetyl group is located near the side chain of Val-370 at a distance of 3.24 Å. A sufficiently large space is indispensable for the recognition of the substrate. If Val-370 were much larger, steric hindrance would occur. Val-370 of CaAGM1 is replaced by the histidine residue of PMM/PGM, the aspartic acid residue of phosphoglucosamine mutase, and the tryptophan residue of PGM. Phosphoglucosamine mutase and PGM are also members of the ␣-D-phosphohexomutase superfamily. It is difficult for such large residues of other proteins in the superfamily to accept the N-acetyl group. Consequently, it is suggested that a small side chain as the valine residue is suitable to recognize the N-acetyl group of the substrate.
On the other hand, in the product complex, the carbonyl oxygen atom of the N-acetyl group is hydrogen-bonded to a water molecule and the side chain of Asn-389. Interestingly, Asn-389 of CaAGM1 is replaced by the serine residues of all the other proteins in the superfamily (35,36). It is assumed that this residue is important to the recognition of the N-acetyl group of the substrate.
Catalytic Mechanism-The catalytic mechanism of AGM1 can be understood in the same way as those proposed for other proteins in the superfamily (12,13). It is considered that the catalytic mechanism requires two phosphoryl transfer reactions: the nucleophilic substitution on the phosphorus atom by the S n 2 mechanism and the cleavage of the phosphoanhydride bond (Fig.  8). First, the phosphoryl group from the phosphorylated Ser-66 binds to the substrate. In the present structure, Ser-66 is not phosphorylated. But it is assumed that this reaction is not done without the phosphorylation of this serine residue, because a hydroxyl group of a PO 4 3Ϫ ion by itself is too basic to be substituted for the hydroxyl group of the substrate at C-6 or C-1. The substrate is converted into a bis-phosphorylated intermediate. This hypothetical pathway via the intermediate could be supported by the measurement of the activity with sugar 1,6-diphosphates and the experiment detecting the intermediate (10). Then, this intermediate rotates 180°. Consequently, the phosphoryl group at C-6 changes positions with another phosphoryl group at C-1. Finally, the phosphoryl group, which has moved near the metal ion, dissociates from the intermediate and binds to Ser-66.
The active-site cleft is too narrow for this intermediate to rotate. The active-site cleft could be expanded by the movement of domain 4. The bis-phosphorylated intermediate is released from the active site and then rotates 180°and is accommodated again to the active-site cleft. In this event, the metal ion plays an important role. As seen in the apoenzyme, when a Zn 2ϩ ion is absent, a PO 4 3Ϫ ion is also unobserved. This indicates that the metal ion is indispensable for the holding of the PO 4 3Ϫ ion. Furthermore, as mentioned above, the positive charge from the metal ion could contribute to the transfer of the PO 4 3Ϫ ion (38,39).