The Crystal Structure of Phosphoglucose Isomerase/Autocrine Motility Factor/Neuroleukin Complexed with Its Carbohydrate Phosphate Inhibitors Suggests Its Substrate/Receptor Recognition*

Phosphoglucose isomerase catalyzes the reversible isomerization of glucose 6-phosphate to fructose 6-phosphate. In addition, phosphoglucose isomerase has been shown to have functions equivalent to neuroleukin, autocrine motility factor, and maturation factor. Here we present the crystal structures of phosphoglucose isomerase complexed with 5-phospho-d-arabinonate andN-bromoacetylethanolamine phosphate at 2.5- and 2.3-Å resolution, respectively. The inhibitors bind to a region within the domains' interface and interact with a histidine residue (His306) from the other subunit. We also demonstrated that the inhibitors not only affect the enzymatic activity of phosphoglucose isomerase, but can also inhibit the autocrine motility factor-induced cell motility of CT-26 mouse colon tumor cells. These results indicate that the substrate and the receptor binding sites of phosphoglucose isomerase and autocrine motility factor are located within close proximity to each other. Based on these two complex structures, together with biological and biochemical results, we propose a possible isomerization mechanism for phosphoglucose isomerase.

Phosphoglucose isomerase (PGI) 1 (EC 5.3.1.9), a glycolytic enzyme, is an essential enzyme in all tissues. It interconverts glucose 6-phosphate and fructose 6-phosphate, hence plays a central role in both the glycolysis and the gluconeogenesis pathways. PGI deficiency in humans is an autosomal recessive genetic disorder that has the typical manifestation of nonspherocytic hemolytic anemia of variable clinical expression (1,2). The serum activity of human PGI serves as a tumor marker in cancer patients (3,4) and elevation in PGI activity is closely correlated with metastasis (5,6). In addition to its essential role in carbohydrate metabolism in the cytoplasm, PGI also functions as neuroleukin (NLK) (7)(8)(9), autocrine motility factor (AMF) (10,11), and maturation factor (MF) (12).
NLK is a neurotrophic growth factor that promotes the survival of spinal and sensory neurons (13). Interestingly, PGI and NLK have previously been reported to have differential induction capabilities for certain neurons (9). AMF, a new class of cytokines, can stimulate cell migration in vitro and metastasis in vivo (10, 14 -16). Amino acid sequencing and immunological cross-reactivity experiments suggest that mouse AMF is identical or closely related to PGI/NLK (10). MF is capable of mediating the differentiation of human myeloid leukemic HL-60 cells to terminal monocytic cells, and apparently a high degree of homology exists between the MF of myeloid leukemia cells and PGI or NLK (12).
It was previously suggested that PGI may recognize a sugarcontaining molecule(s) at the cell surface (7) and that NLK binds to the cell surface in a carbohydrate-dependent manner utilizing a PGI-like structure. Watanabe (10,16) also pointed out that AMF (PGI/NLK/MF) may contain a PGI-like structure and initiates signal transduction by interacting with the carbohydrate side chains of the extracellular domain of the AMF receptor (AMFR). Carbohydrate phosphates can act as AMF inhibitors probably by binding and blocking the receptor recognition site on AMF. These novel functions of PGI may depend directly on the catabolism of phosphosugar, making an understanding of the molecular catalytic mechanism of the enzyme potentially significant.
We previously reported the substrate-free structure of PGI/ AMF/NLK and confirmed that PGI can function as AMF and NLK (17). Nonetheless, the nature of the substrate-active site or the receptor binding site of PGI/AMF/NLK remains poorly understood. In this report, we tested the inhibitory effect of two carbohydrates containing phosphate, 5-phospho-D-arabinonate (5PA), and N-bromoacetylethanolamine phosphate (BAP), on the PGI enzymatic activity and the AMF-induced cell motility. 5PA, the five-carbon homologue of 6-phosphogluconate, is the equivalent analogue of the cis-1,2-enediolate intermediate in the reaction of phosphoglucose isomerase (18), and BAP is characterized as an active-site directed inhibitor to mammalian phosphoglucose isomerase in an effort to identify activesite residues (19). We then presented the crystal structures of PGI from Bacillus stearothermophilus complexed with 5PA and BAP at 2.5-and 2.3-Å resolution, respectively.

MATERIALS AND METHODS
Preparation of the Wild Type and N-Bromoacetylethanolamine Phosphate-linked Phosphoglucose Isomerase-The isolation and purification of PGI have been previously reported (20). The complete inactivation of PGI was achieved by incubating the enzyme (0.6 mg/ml) at 30°C overnight with 4 mM N-bromoacetylethanolamine phosphate (BrAc-NHEtOP) in a phosphate buffer (20 mM, pH 7.0). BrAcNHEtOP was synthesized according to the protocol described by Hartman et al. (21). Free inhibitors were removed by passing the reaction mixture through a desalting column.
Cell Motility Assay-The cell motility assay of mouse CT-26 cells was measured by the method of Lin et al. (22) with minor modifications. Polyvinylpyrrolidone-free polycarbonate filters (Nuclepore, 8-m pore size) were soaked in 0.5 M acetic acid overnight then washed with distilled water and coated with 50 l of Matrigel for 2 h. Cells were placed on the top part of the Boyden chamber at a cell dose of 5 ϫ 10 4 per 200 l in Dulbecco's modified Eagle's medium (DMEM) with 10% phosphate-buffered saline. The bottom part of the chamber was filled with or without PGI in DMEM for positive and control experiments. For the BAP inhibition assay, BAP-linked PGI was used instead of PGI in the bottom chamber. For the 5PA inhibition assay, the PGI and 5PA were preincubated for 2 h before filling the bottom chamber together with DMEM. Incubation was carried out at 37°C for 16 h. The filters were removed and fixed in 4% paraformaldehyde for 15 min at room temperature. Cells on the upper filter surface were removed with a cotton swab. The filters were stained with hematoxylin for 40 min, and the cells on the lower filter surface were counted under a light microscope.
Enzymatic Assay-The phosphoglucose isomerase activity was determined at 30°C by the coupled glucose-6-phosphate dehydrogenase method (23). The standard assay mixture (1 ml) contains 20 mM potassium phosphate (pH 7.0), 10 units of glucose-6-phosphate dehydrogenase, 2 mM ␤-NAD ϩ , 4 mM fructose 6-phosphate, and 0.5 g of PGI or PGI covalently linked with BAP. To assess the inhibitory effect, 5-phosphoarabinoate was included in the assay mixture to a final concentration of 0.1 mM.
Crystallization and Data Collection-The PGI⅐BAP crystals were grown at room temperature by hanging-drop vapor diffusion of a 2-l protein solution ( All data sets were taken at room temperature on a Rigaku R-AXIS II imaging plate system using double-mirror-focused CuK ␣ x-ray radiation generated from a RigakuRU-300 rotating anode operating at 50 kV and 80 mA. Data were indexed, integrated, and scaled using DENZO and SCALEPACK software packages (24) (Table I).
Structure Determination and Refinement-The AMoRe program package (25) was used in the rotation and translation search. The previous PGI crystal structure (17) was used as the search model (Protein Data Bank (PDB) accession code: 2PGI). The search model was placed in a P1 cell with a ϭ b ϭ c ϭ 90 Å and ␣ ϭ ␤ ϭ ␥ ϭ 90°. Data between 8.0 and 4.5 Å and a Patterson radius of 20 Å were used for all rotation and translation function calculations. The data sets from PGI⅐5PA and PGI⅐BAP both gave significant solutions. After rigid body refinement, the optimal solution was determined for PGI⅐BAP data at Euler angle ␣ ϭ 71.32°, ␤ ϭ 76.29°, and ␥ ϭ 305.05°; x ϭ 0.482, y ϭ 0.139, and z ϭ 0.150. When compare with the non-normalized structure factors between search model and target crystal in the definition region, the correlation coefficient and R-factor are 84.0 and 23.5%, respectively. The solution of PGI⅐5PA was at ␣ ϭ 71.71°, ␤ ϭ 76.34°and ␥ ϭ 305.77°, x ϭ 0.483, y ϭ 0.137, z ϭ 0.151 with a correlation coefficient of 77.6 and an R-factor of 28.3%. Structure refinement and model building were carried out using the XPLOR (26) and XTALVIEW programs (27). The rigid body refinement was used first, and the model was treated as one group to optimize the orientational and translational parameters. Simulated annealing omit maps were then used to reduce the model bias. A bulk solvent mask was calculated to improve the reflection data. The final models contain 3661 non-hydrogen atoms, including 134 oxygen atoms from water molecules for PGI⅐BAP (PDB accession code: 1C7Q); and 3,539 non-hydrogen atoms, including 94 oxygen atoms from water molecules for PGI⅐5PA (PDB accession code: 1C7R). The first and the last two amino acids are presumably disordered in both crystal structures and residues 96 to 105 of PGI⅐5PA were not visible in the electron density map. The final refinement statistics are summarized in Table I

RESULTS AND DISCUSSION
Cell Motility and Enzyme Inactivation-Previously, we demonstrated by cell migratory stimulation assay that PGI from B. stearothermophilus exhibits AMF cell motility activity (17). To directly address the question whether the isomerase and the cell motility functions of the protein share the same substrate binding site, we tested the effect of two carbohydrates containing phosphates, 5PA and BAP, on these functions. The PGI activities were monitored by the rate of isomerizing fructose 6-phosphate (28). In the presence of 0.1 mM 5PA, the isomerase activity of the enzyme was decreased by 86%. PGI that has been conjugated with BAP was essentially inactive. A similar result with the BAP-linked PGI has been reported by Gibson et al. (19).
Results from the motility assay and the stimulatory effect on CT-26 cells in the absence ( Fig. 2A) or presence (Fig. 2B) of PGI are presented. In the presence of 5PA (C) and BAP (D), the inhibitory effect on cell motility can be readily observed. The current data indicate that 5PA and BAP not only inhibit PGI enzymatic activity but also constrain the induced cell motility of PGI with no effect on basal migration. The results confirm that the substrate and the receptor binding site of PGI/AMF are overlapped.
General Structure-The orthorhombic crystals of PGI⅐BAP and PGI⅐5PA are isomorphous to the substrate-free PGI (17) crystals. The folding topology of these two inhibitor complexes is structurally identical to the substrate-free form as shown in Fig. 3a. Briefly, the structure is composed of two globular domains (designated as the large and small domain) and an "arm-like" C-terminal tail. Each domain has a ␤-sheet core surrounded by ␣-helices that link the ␤-strands. The active site of substrate/inhibitor binding is located within the two globular domains, the C-terminal tail, and the interface between the two subunits. The large domain also contains a protruding loop region opposite to the C-terminal tail. The structure can be represented by a prolate ellipsoid with an "arm-like" structural feature on each side. The dimensions of the monomer are about 71 x 75 x 33 Å. Even though there is one subunit in the crystallographic asymmetric unit, the active form of the enzyme is a dimer. The monomer-monomer associate in an arm-to-arm hug fashion with intimate contacts and form a hydrophilic channel that coincides with the crystallographic 2-fold axis, which runs through the dimer as indicated by the arrow in Fig. 3b. The solvent-accessible surface (29) of the subunit is 19505 Å 2 . The intermolecular contact area is 10,751 Å 2 , with a summation of 3192 and 7559 Å 2 of the hydrophilic and hydrophobic surface areas, respectively. Dimer formation causes a burial of 55% of the monomer surface area. Fig. 4 shows the structural comparisons of substrate-free PGI with those of the PGI⅐5PA and PGI⅐BAP complexes. Both inhibitors gave similar, but not identical, results. Their binding sites lie in the slight cleft located within the large domain, the small domain, and the C-terminal tail of the monomer as predicted . The cleft is close to the subunit interface, which is formed by the association of the two subunits. Least-squares superposition (C␣) of the PGI⅐BAP and PGI⅐5PA complexes with the substrate-free structure gave root mean square deviations of 0.25 and 0.46 Å, respectively. This small difference indicates that the overall conformations of these three structures are very similar, except for local conformational changes in the loop and substrate binding regions. We will discuss some important local structural variations later to shed light on the action of the inhibitor binding.
Inhibitor Binding Site of Transition State Analogue 5PA-5-Phosphoarabinonate (5PA) is a stable analogue of the cis-1,2enediolate intermediate that is believed to occur transiently in the phosphoglucose isomerase reaction (18). 5PA is also one of the strongest known competitive inhibitors of PGI. Therefore, it possibly represents a transition state analogue and the 5PAbound form of the protein may resemble a catalytic intermediate or a transition state. In the PGI⅐5PA complex, a total of 16 hydrogen bonds, 7 from the large domain, 6 from the small domain, and 2 from the C-terminal region, are observed between the protein and the ligand. Residues participate in 5PA binding are shown in Table  II and include charged, polar, and nonpolar side chains, as well as atoms from the peptide backbone. It is also interesting to point out that the His 306 positioned in the binding site is contributed by the other subunit of the dimer. This positioning may explain why the active form of the isomerase is a dimer.
Inhibitor Binding Site of BAP-Even though BAP may not represent a transition state analogue, it can modify PGI stoichiometrically, resulting in complete and rapid inactivation of the enzyme (19). In addition, data from mutagenesis (30) suggest that His 306 is an active-site residue of PGI. The imidazole of His 306 could be the nucleophile that attacks BrAcNHEtOP. As shown in Fig. 1b, BAP formed a covalent bond with His 306 . Therefore, the complex structure of PGI⅐BAP can confirm that BAP is indeed as an active site inhibitor and react with His 306 .
In contrast to the fairly deep binding location of 5PA (Fig.  5A), BAP is situated near the entrance of the binding pocket (Fig. 5B). The noteworthy difference between 5PA and BAP is the orientation of the phosphate moieties on these inhibitors. The phosphate group on 5PA is pointed toward the bottom of the pocket. Conversely, the phosphate moiety on BAP swings about 180°and orients in the opposite direction. As shown in Fig. 5B, the binding pocket of BAP is lined with Ser 140 , Thr 142 , Thr 143 , Glu 145 , His 306 , Glu 417 , and Lys 420 . The His 306 residue was previously suggested to act as a general base to initiate the isomerization reaction (31). As expected, the PGI enzyme is alkylated by BAP specifically at position N ⑀1 of His 306 contributed from the other subunit of the dimer. In addition to a water molecule, the phosphate moiety of BAP constitutes hydrogen bonds with the hydroxyl group of Ser 140 and Thr 143 , whereas the side chain of Lys 420 neutralizes the opposite charges and forms one salt bridge with the phosphate group.
Inhibitor-induced Local Conformational Changes-Achari et al. (32,33) have proposed an occurrence of local conformational changes upon inhibitor binding. We compared the structures of the inhibitor-bound complexes with that of the uncomplexed PGI (17) and observed significant conformational changes in the active site of either PGI⅐5PA or PGI⅐BAP. The comparison of the PGI structures with and without 5PA bound is shown in Fig. 5A. The most significant difference appears at the region between residues Gly 200 and Val 205 . This variation may be due partially to the enhanced interaction between Arg 202 and the phosphate group on 5PA. The quanidino group of Arg 202 in the PGI⅐5PA complex structure moved 3-4 Å toward the inhibitor from its original position in the unbound protein.
Because the phosphate moiety of BAP points to the entrance of the binding pocket, a local conformational change upon BAP binding is found at the region between residues Lys 139 and Thr 144 as shown in Fig. 5B. Ser 140 is the key residue responsible for this structural variation. Upon inhibitor binding, the side-chain hydroxyl group of Ser 140 moves into proximity with BAP and forms two hydrogen bonds with an oxygen atom of the phosphate group and the backbone oxygen O 4 .
Proposed Catalytic Mechanism of Phosphoglucose Isomerase-Previous results suggest that the availability of the active site on PGI is required for stimulating the migration of tumor cells. Therefore, the understanding of the molecular catalytic mechanism of the enzyme should be physiologically significant.
It has long been proposed that the catalytic mechanism of phosphoglucose isomerase should include the following steps: 1) binding of the cyclic form of the substrate to the enzyme, 2) ring opening of the substrate, 3) base-catalyzed isomerization via a cis-endiol intermediate, 4) ring closure of the product, and 5) release of product (31, 34 -35). Isotope effect studies suggest that the isomerization step is rate-limiting (34). The transition state should have a structure similar to cis-enediolate, because both 5PA (18) and 5-phosphoarabinohydroxamate (36) behave as transition state inhibitors. The isomerization activity is pH-dependent and follows a bell-shaped curve and, together with the temperature dependence of the pK a values, suggests that histidine and lysine residues participate in the catalysis (31). Mutagenic studies indicate that one of the conserved lysine (Lys 420 in PGI) and arginine (Arg 202 in PGI) residues play indispensable roles in catalysis (30). Affinity labeling of the conserved histidine (His 306 in PGI) by BrAcNHEtOP, an active site-directed inhibitor of phosphoglucose isomerase, suggests that the conserved histidine functions as a general base in the isomerization step (30).
Considering all of the above data from the literature in combination with the active-site structure of phosphoglucose isomerase unveiled in the current study (Fig. 5), we propose the following molecular catalytic mechanism for PGI: (i) The reaction is initiated by binding the substrate to the enzyme. A network of hydrogen bonds between the substrate and the active site amino acids stabilizes this binding. It is worth noting that the amino acids comprising the active site, especially those directly contacting the substrate, are conserved throughout the PGI family (17). (ii) Lys 420 is involved in the opening of the phosphoglucopyranose ring by functioning as a general base. The side-chain amino group of Lys 420 is within 3.0 Å of the C1 and C2 hydroxyl groups of the inhibitor (Fig. 6a) and the importance of Lys 420 in catalysis has been confirmed by mutagenic studies (30). (iii) His 306 , acting as a general base, extracts the proton from C2. Simultaneously, Glu 285 acts as a general acid and donates a proton to the C1 carbonyl oxygen (Fig. 6b). The concerted action of these two residues transforms the substrate to the cis-enediol intermediate and inaugurates the isomerization step. By reversing their roles in the subsequent step, the carboxylate of Glu 285 , acting as the base, and the imidazolium of His 306 , acting as the acid, work on the intermediate to complete the isomerization reaction. The substrate at the transition state of isomerization presumably has a structure similar to the cis-enediolate (Fig. 6). Through electrostatic interaction, Glu 145 stabilizes the developing partial positive charge on the side chain of His 306 , while Arg 202 stabilizes the partial developing negative charge on Glu 285 . Arg 202 also contributes to the interaction of the phosphate group and may lock the substrate in an optimum position for isomerization. The ammonium of Lys 420 provides charged hydrogen bonds to the C1 and/or C2 oxygen atoms of the cis-enediolate. (iv) After isomerization, Lys 420 performs an acid-catalyzed ring closure reaction, and the product is released from the enzyme.
The crystal structure of rabbit skeletal muscle PGI (rPGI) with a competitive inhibitor D-gluconate 6-phosphate was published by Jeffery et al. (37) during the preparation of the present manuscript. The topology of rPGI is very similar to that of PGI from B. stearothermophilus, but the orientation of the bound D-gluconate 6-phosphate is different from that of 5PA in the present study. Consequently, the catalytic mechanism proposed by Jeffery et al. (37) has features that differ from that proposed in this study. In rPGI, a His 388 -Glu 216 dyad and Lys 518 (corresponding to His 306 , Glu 145 , and Lys 420 of the B. stearothermophilus PGI) were proposed as the general baseacid pair to initiate the isomerization step. The function of the His-Glu dyad is in close agreement with our proposed mechanism. However, a function of providing charged hydrogen bonds to the C1 and C2 oxygen atoms of the transition sate is suggested for the active-site Lys in the present study.
A conserved Glu (Glu 285 in this study) is thought to be the residue that donates a proton to the substrate in the initial isomerization step (Fig. 6). The proposed role of Glu 285 is supported by the following reasons: 1) the closeness of the O 1 carboxylate oxygen of 5PA and the O ␦1 carboxylate oxygen of Glu 285 ; 2) the indispensable role of Glu 285 in catalysis suggested by our unpublished mutational data; and 3) the involvement of glutamate in catalysis of PGI suggested by affinity labeling by 1,2-anhydrohexitol 6-phosphate (38). No glutamate was assigned in the proposed mechanism of rPGI. In our catalytic model, all three residues (His 306 , Glu 285 , and Lys 420 ) play significant roles in the isomerization step; mutation of any one of them would drastically impair the catalytic ability of PGI. The function of Arg 272 in rPGI (corresponding to Arg 202 in this study) was proposed to make the overall electrostatic potential at the substrate positive and provide stabilization to the developing negative charge on the enediolate-like transition state. A similar function is fulfilled by Lys 420 by providing charged hydrogen bonds to the transition state in our model. Upon binding of 5PA, the quanidino group of Arg 202 moved toward the carboxylate group of Glu 285 by 0.6 Å, and as a result, the pK a of Glu 285 could be lowered and the acidic role of Glu 285 in the isomerization step could be enhanced. Due to the close contact of Arg 202 and the phosphate group of 5PA, we believe  that Arg 202 also plays a key role for the recognition of phosphosugar during isomerization. It should be noted that our crystal structure dose not rule out the possibility of Lys 420 being the role of a general acid during isomerization as suggested by the previous rPGI study. The actual function of each active-site residue may be defined in more precise fashion in the future after more biochemical data are available.
In summary, we have elucidated the crystal structures of PGI complexed with the transition state analogue 5-phospho-D-arabinonate (5PA) and N-bromoacetylethanolamine phosphate (BAP) at 2.5-and 2.3-Å resolution, respectively. We also demonstrate that the inhibitors, 5PA and BAP, not only inhibit the enzymatic activity of PGI but also inhibit the AMF-induced cell motility of CT-26 mouse colon tumor cells. Thereby, the present study provides the first view that the locations of the substrate binding site for phosphoglucose isomerase and the receptor binding site for autocrine motility factor are overlapped. We also propose a possible isomerization mechanism for PGI.