Crystal Structures of Complexes of N-Butyl- and N-Nonyl-Deoxynojirimycin Bound to Acid β-Glucosidase

Gaucher disease is caused by mutations in the gene encoding acid β-glucosidase (GlcCerase), resulting in glucosylceramide (GlcCer) accumulation. The only currently available orally administered treatment for Gaucher disease is N-butyl-deoxynojirimycin (Zavesca™, NB-DNJ), which partially inhibits GlcCer synthesis, thus reducing levels of GlcCer accumulation. NB-DNJ also acts as a chemical chaperone for GlcCerase, although at a different concentration than that required to completely inhibit GlcCer synthesis. We now report the crystal structures, at 2Å resolution, of complexes of NB-DNJ and N-nonyl-deoxynojirimycin (NN-DNJ) with recombinant human GlcCerase, expressed in cultured plant cells. Both inhibitors bind at the active site of GlcCerase, with the imino sugar moiety making hydrogen bonds to side chains of active site residues. The alkyl chains of NB-DNJ and NN-DNJ are oriented toward the entrance of the active site where they undergo hydrophobic interactions. Based on these structures, we make a number of predictions concerning (i) involvement of loops adjacent to the active site in the catalytic process, (ii) the nature of nucleophilic attack by Glu-340, and (iii) the role of a conserved water molecule located in a solvent cavity adjacent to the active site. Together, these results have significance for understanding the mechanism of action of GlcCerase and the mode of GlcCerase chaperoning by imino sugars.

Subsequent structural studies led to increased understanding of the GlcCerase structure (8) and to solution of GlcCerase structures (Table 1) to which small molecules were bound, either covalently, i.e. conduritol-B-epoxide (1,2-anhydro-myoinositol; CBE) (5), or non-covalently, i.e. isofagomine (IFG) (9) (Fig. 1). No structures of mutant enzymes are yet available, but recent work has demonstrated that some mutants cause a loss of protein stability during biosynthesis, whereas other mutants result in loss of catalytic activity (10,11).
At present, two treatments are available for Gaucher disease patients. The first is enzyme replacement therapy using recombinant human GlcCerase (Cerezyme TM ) expressed in Chinese hamster ovary cells (12)(13)(14). The second is substrate reduction therapy, in which partial inhibition of glycosphingolipid synthesis by N-butyl-deoxynojirimycin (NB-DNJ; Zavesca TM ) (15) decreases accumulation of glycosphingolipids, including GlcCer. A third therapeutic option is on the horizon, namely chaperone therapy, in which active site-directed inhibitors are used to stabilize mutant forms of GlcCerase as they pass through the secretory pathway (16,17). Interestingly, NB-DNJ and N-nonyl-deoxynojirimycin (NN-DNJ) also act as chemical chaperones of GlcCerase (18), since elevated GlcCerase activity occurs upon incubation of cultured cells expressing various GlcCerase mutants with NN-DNJ or NB-DNJ (19,20). Thus, NB-DNJ and NN-DNJ inhibit both GlcCer synthase (15) and GlcCerase, although at different concentrations (18,19). The mechanism of inhibition, or of binding to GlcCerase, is not known.
We now report the crystal structures of complexes of NB-DNJ and NN-DNJ with recombinant human GlcCerase, expressed in cultured plant cells (pGlcCerase) (21); the structure of pGlcCerase is highly homologous to that of Cerezyme TM (21). We compare the structures of pGlcCerase complexed to NB-DNJ and NN-DNJ with that of a complex of Cerezyme TM with another potential chaperone, IFG, and demonstrate some important differences in their mode of binding that have implications for understanding the catalytic cycle of GlcCerase and the mode of its chaperoning by imino sugars.

EXPERIMENTAL PROCEDURES
Crystallization-pGlcCerase was produced as described (21). The enzyme was diluted in crystallization buffer (0.02% NaN 3 /10 mM sodium citrate, pH 5.5, containing 7% (v/v) ethanol, washed three times, and concentrated to 4 -5 mg/ml in a Centricon device using a filter with a 30-kDa cut-off. NB-DNJ and NN-DNJ were dissolved in water to yield 0.1-M stock solutions and added to the pGlcCerase solution to a final concentration of 13 mM. Cocrystallization was performed using the micro-batch technique under oil. Protein and crystallization solutions were dispensed into micro-batch crystallization plates under oil (1:1 v/v silicone and paraffin oils), such that the final solution contained 50% protein solution and 50% crystallization liquor. The best diffracting crystals of NB-DNJ/ pGlcCerase were obtained in 0.2 M (NH 4 ) 2 SO 4 /0.1 M Tris, pH 6.5, 25% (w/v) polyethylene glycol 3350, and the best crystals of NN-DNJ/pGlcCerase were obtained using 0.2 M NH 4 COOCH 3 /0.1 M Hepes, pH 7.5, 25% (w/v) polyethylene glycol 3350. Crystals were cryo-protected with 20% (v/v) ethylene glycol for x-ray data collection. Data Collection and Refinement-Data were collected on the BM14 beamline of the European Synchrotron Radiation Facility (Grenoble, France). Images were indexed with the HKL2000 software package and scaled with SCALEPACK (22). The structures of NB-DNJ/pGlcCerase and NN-DNJ/pGlcCerase were solved using rigid body refinement and refined with Refmac5 (23) ( Table 2). The crystal structure of pGlcCerase (21) (PDB code 2v3f) was used as a starting model, and FreeR flags were taken from the corresponding structure factors file. Model manipulation and water editing was performed using Coot graphics software (24). Images were created with PyMol and LIGPLOT (25). Structures and structure factors were deposited in the Protein Data Bank (code 2v3d for NB-DNJ/pGl-cCerase and code 2v3e for NN-DNJ/ pGlcCerase-NN-DNJ) ( Table 1).

Modeling of Michaelis-Menten Complexes-
The imino sugar of NN-DNJ was used to build a model of GlcCer in the pGlcCerase active site. Because the aliphatic chain of NN-DNJ does not resemble the sphingoid base or the N-acylated fatty acid of GlcCer, the ceramide moiety was modeled on the basis of the crystal structure of the complex of ganglioside GM3 with endo-glycoceramidase II (endo-GCase II) (PDB 2OSX) (26). The catalytic residues of endo-GCase II were aligned with those of GlcCerase (with Glu-233 of endo-GCase II corresponding to Glu-235 of GlcCerase, and Ser-351 of endo-GCase II corresponding to Glu-340 of GlcCerase; in addition, His-304 of endo-GCase II was aligned with His-311 of GlcCerase). The glucose moiety of GlcCer did not exactly overlay the imino sugar of either NNor NB-DNJ; therefore, the structure of GlcCer was modeled using the torsion angles of the GlcCer moiety of GM3 bound to endo-GCase II. In addition, the torsion angles of the glucoside bond and of some bonds of the N-acyl chain were modified to avoid steric clashes.

RESULTS AND DISCUSSION
Binding of NB-DNJ and NN-DNJ to the Active Site of pGlcCerase-Crystals of NB-DNJ/pGlcCerase and NN-DNJ/ pGlcCerase were obtained in space group P2 1 (Table 2), with two protein molecules in the asymmetric unit. Superimposition of the native pGlcCerase structure (21) on those of NB-DNJ/ pGlcCerase or NN-DNJ/pGlcCerase resulted in root mean square deviations of only 0.2 and 0.3 Å, respectively, showing that neither of these ligands produces a global structural change upon binding to the enzyme.
Both inhibitors bind to the active site of pGlcCerase, with the imino sugar making hydrogen bonds with side chains of active site residues (Figs. 2 and 3). Hydrogen bond distances between the imino sugar and the active site residues are similar for NB-DNJ and NN-DNJ. The alkyl chains of NB-DNJ and NN-DNJ  are oriented toward the entrance of the active site. The interactions of pGlcCerase with these alkyl chains are via hydrophobic interactions and are, therefore, less specific than the hydrogen bond interactions formed with the pyranose ring, consistent with the ability of imino sugars with alkyl chains of varying lengths to bind to GlcCerase (18). The alkyl chains of both inhibitors are stabilized by interaction with Tyr-313 near the active site, and the alkyl chain of NN-DNJ makes an additional contact with Leu-314, near the entrance to the active site ( Fig.  2A). Interestingly, there is a Ͼ300-fold difference in the K i values of the two inhibitors (116 M for NB-DNJ and 0.3 M for NN-DNJ) (27), which cannot be accounted for by this one additional contact. Rather, the lower K i of NN-DNJ is most likely a reflection of the increased overall hydrophobicity of NN-DNJ compared with NB-DNJ (Table 3), with the hydrophobic surface at the entrance to the active site favoring the more hydrophobic ligand.
Comparison of the structures of NB-DNJ/pGlcCerase and NN-DNJ/pGlcCerase with that of IFG/DG-Cerezyme (9) demonstrates that the pyranose-like ring makes a similar number of hydrogen bonds with the enzyme in all three cases ( Fig. 2 and Table 3). Where these structures differ significantly is in the interaction of their nitrogen atoms with the active site of GlcCerase. The secondary amine in IFG is structurally homologous to the anomeric carbon of GlcCer and is located between the two catalytic glutamic acid residues, Glu-325 and Glu-340, forming hydrogen bonds with each of their carboxyl moieties. In contrast, the tertiary amines of NN-DNJ and NB-DNJ are located in a position similar to that of the ring oxygen of GlcCer and do not, therefore, form any hydrogen bonds or salt bridges with GlcCerase.
Loops at the Entrance to the Active Site-The active site of GlcCerase is formed on one side by three loops, which have been observed in a number of conformations (Figs. 3 and 4) (5,7,8,28). However, in all GlcCerase structures to which an inhibitor was bound, only one set of conformations was observed (5,9), which was the same as that observed in pGlcCerase crystallized in a P2 1 space group (21). The conformation of the active site-associated loops presumably facilitates substrate binding. Thus, the side chain of Asn-396 on loop 2 forms a hydrogen bond with the inhibitor hydroxyl group that corresponds to the same group on the non-chiral pyranose carbon atom of GlcCer (Fig. 2). Loop 3 shows even more flexibility, in some cases adopting a helical structure, whereas in others it assumes a coil.
One residue on loop 3 that appears relevant to catalysis is Tyr-313. In the three structures to which a non-covalent inhibitor is bound (NB-DNJ/pGlcCerase, NN-DNJ/pGlcCerase, and IFG/DG-Cerezyme), the hydroxyl group of Tyr-313 hydrogen bonds to Glu-340. This requires loop 3 to be in a helical conformation. In the structure with a covalent inhibitor (CBE-DG-Cerezyme), in which CBE is covalently attached to Glu-340, Tyr-313O forms a hydrogen bond with Glu-235 and loop 3 assumes a coil conformation. If the two observed interactions of Tyr-313 are adopted by the enzyme during catalysis, then a conformational change in loop 3 may be an integral element of the GlcCerase reaction mechanism.
Loop 1 does not display major changes in its backbone angles or in its secondary structure. However, it appears to be affected by crystal contacts (8), and hence some movements in this loop are observed. Specifically, the aromatic residues on loop 1, Trp-348 and Phe-347, are involved in crystal contacts, and it is therefore possible that it is more mobile in solution. The mobility of these hydrophobic residues may be relevant to the association of GlcCerase with the lipid membrane (8).
Catalytic Mechanism of GlcCerase-Based on the structures reported herein, and on that of GM3 bound to the active site of endo-GCase II (26), we modeled GlcCer in the active site of GlcCerase (Fig. 5), allowing us to examine the reaction mechanism of GlcCerase. Three issues arose out of this analysis. First, while there is significant evidence to support a nucleophilic role for Glu-340 in the GlCerase catalytic cycle, several recent studies (e.g. Ref. 29) have implied direct attack of the carboxylate oxygen of Glu-340 on the anomeric carbon of GlcCer. The NN-DNJ/pGlcCerase and NN-DNJ/pGlcCerase struc-

TABLE 3 Number and type of bonds in structures of GlcCerase complexes
Side chains of residues with van der Waals distances Յ4 Å to the ligand were considered as non-polar contacts.

Structure
Number tures do not support such a direct attack, because the apical hydrogen on the anomeric carbon is positioned between the carbon atom and the attacking oxygen atom in such a way that it would block nucleophilic attack by steric hindrance (Fig. 6). If, however, there is an intermediate involving a planar carbon atom, attack by Glu-340 would be possible (30,31). In the case of CBE, direct nucleophilic attack on the epoxide carbon by Glu-340 is possible, because the hydrogen atom is not apical, rendering the carbon susceptible to such attack. The second issue concerns the protonation state of Glu-235. This residue corresponds to Glu-35 in lysozyme, which supplies a proton to the leaving group in the initial stage of the reaction and thus must be protonated in the resting state of the enzyme (30,31). The basic limb of the pH/activity profile of lysozyme, with a pK of ϳ6.5, is attributed to deprotonation of Glu-35. The elevated pK a of this residue in lysozyme has been explained by it being partially buried. GlcCerase has a similar pH/activity profile, with pKs of 4.5 and 6.5 (32,33). However, in Cerezyme (8), Glu-235 is near His-311 N ␦1 (3.2 Å), Asn-234 (3.3 Å), and Gln-284 (3.6 Å). His-311 is part of a hydrogen bond network involving Asp-282, Arg-120, and the catalytic residue Glu-340 (Fig. 7). Given its proximity to Asp-282, it is presumably protonated, despite being buried in the active site. The close proximity of polar residues, particularly if His-311 is charged, should only serve to lower the pK a of Glu-235 rather than raising it to pH 6.5. Further investigation is thus required to understand the protonation states of these residues.
The third issue concerns the role of water in the catalytic mechanism; a water molecule is required for hydrolysis of the covalent intermediate. In all GlcCerase structures of sufficient resolution to see water molecules (1OGS, 1Y7V, 2NSX, 2NT0, 2NT1, 2V3F, 2V3D, and 2V3E), a cavity near the catalytic residue Glu-235 contains water molecules. One particular water molecule at an equivalent location is visible in other glycosidase structures, such as endo-GCase II (26) and xylanase (34). These conserved water molecules may be involved in the catalytic cycle because the solvent cavity is located next to the catalytic residues and is accessible from the active site (Fig. 8). 4 CONCLUSIONS NB-DNJ is the only currently available orally administered drug for treatment of Gaucher disease. Its primary effect is via inhibition of GlcCer synthase; however, the compound also has a coincidental chaperone effect on GlcCerase. The chaperoning effect of NN-DNJ was The loops in GlcCerase occur in a number of conformations, but only two are shown for clarity, in yellow and green; these conformations give the most pronounced changes in the entrance to the active site. Tyr-313, which may play a role in the catalytic mechanism, is indicated. The catalytic residues are shown as red sticks.  . Proposed modified catalytic mechanism of GlcCerase. In scheme A, the anomeric carbon of glucose would not be susceptible to nucleophilic attack by Glu-340 due to steric clashes with the hydrogen atom of the glucose anomeric carbon. Hydrolysis of the glycoside bond must, therefore, proceed through a carbenium ion intermediate. B, the commonly accepted mechanism assumes a direct nucleophilic attack on the anomeric carbon, without formation of a carbenium ion intermediate. SEPTEMBER 28, 2007 • VOLUME 282 • NUMBER 39 shown to be specific for GlcCerase with mutations in the catalytic domain, domain III (11,19,35 ). The crystal structures of NB-DNJ/ pGlcCerase and NN-DNJ/pGlcCerase reported herein are consistent with these observations, because the inhibitor binds at the active site. The pK a of NB-DNJ also appears to be advantageous for fulfilling its role as a chaperone. In the endoplasmic reticulum, where the protein is synthesized, the neutral pH is close to the pK a of NB-DNJ. Under these conditions, NB-DNJ would be neutral (Table 3), and so less soluble, favoring binding to the active site. However, at the acidic pH of the lysosome, NB-DNJ would be charged, thus increasing its solubility and lowering the free energy of binding, thus competing less with GlcCer.