Structural Basis for Catalysis and Inhibition ofN-Glycan Processing Class I α1,2-Mannosidases*

Endoplasmic reticulum (ER) class I α1,2-mannosidase (also known as ER α-mannosidase I) is a critical enzyme in the maturation of N-linked oligosaccharides and ER-associated degradation. Trimming of a single mannose residue acts as a signal to target misfolded glycoproteins for degradation by the proteasome. Crystal structures of the catalytic domain of human ER class I α1,2-mannosidase have been determined both in the presence and absence of the potent inhibitors kifunensine and 1-deoxymannojirimycin. Both inhibitors bind to the protein at the bottom of the active-site cavity, with the essential calcium ion coordinating the O-2′ and O-3′ hydroxyls and stabilizing the six-membered rings of both inhibitors in a 1C4conformation. This is the first direct evidence of the role of the calcium ion. The lack of major conformational changes upon inhibitor binding and structural comparisons with the yeast α1,2-mannosidase enzyme-product complex suggest that this class of inverting enzymes has a novel catalytic mechanism. The structures also provide insight into the specificity of this class of enzymes and provide a blueprint for the future design of novel inhibitors that prevent degradation of misfolded proteins in genetic diseases.

been shown to be responsible, in part, for the specificity of yeast ER class I ␣1,2-mannosidase (31). In mammalian Golgi ␣1,2mannosidases that trim Man 9 GlcNAc 2 to Man 5 GlcNAc 2 , this arginine residue is typically a leucine. Replacement of Arg 273 with leucine in yeast ER class I ␣1,2-mannosidase results in an enzyme that is capable of cleaving all four ␣1,2-linked mannose residues rather than just the single terminal residue of the middle arm of Man 9 GlcNAc 2 .
Although the observed protein-carbohydrate interactions in yeast ER class I ␣1,2-mannosidase have provided a wealth of information regarding substrate specificity, the N-glycan lacks the middle-arm terminal mannose residue that would be specifically cleaved during the enzymatic reaction. This prevented unambiguous identification of the residues involved in catalysis and provided no evidence for the role of the calcium ion. We present here the first structures of a class I ␣1,2-mannosidase in complex with the potent inhibitors kifunensine (KIF) (32) and 1-deoxymannojirimycin (dMNJ) (33). The structural results provide clear evidence for the role of calcium in substrate stabilization and suggest that this class of inverting enzymes has a novel catalytic mechanism. This work also provides insight into the specificity of these enzymes for ␣1,2linked mannose residues and the inhibitors kifunensine and 1-deoxymannojirimycin.
The importance of ER class I ␣1,2-mannosidases in ERassociated degradation of misfolded glycoproteins has been demonstrated in both yeast and mammalian cells. In yeast, it has been shown that a misfolded mutant of carboxypeptidase Y is rapidly degraded in wild-type cells, whereas it is stabilized in the mns1 mutant lacking the ER processing ␣1,2-mannosidase (8,11). In mammalian cells, treatment with the ␣1,2-mannosidase inhibitors 1-deoxymannojirimycin and kifunensine has been shown to block the degradation of the T cell receptor subunit CD3-␦ (34), tyrosinase (35), ␣ 2 -plasmin inhibitor (36), and a misfolded variant form of ␣ 1 -antitrypsin (10). In the case of ␣ 1 -antitrypsin, increased secretion of ␣ 1 -antitrypsin was also observed (37). Since the aggregation of misfolded ␣ 1 -antitrypsin in the ER leads to emphysema, understanding the structural basis of inhibition of class I ␣1,2-mannosidase is therefore the first step toward the structure-based design of novel therapeutic agents for this and other genetic diseases characterized by rapid degradation of misfolded glycoproteins.

EXPERIMENTAL PROCEDURES
Expression and Purification-The cloning and characterization of the soluble catalytic domain of human ER class I ␣1,2-mannosidase as a protein A fusion (pPROTA-ERManI) were described previously (13). The portion of the cDNA encoding the catalytic domain (amino acids 172-689) was excised from the pPROTA expression vector by digestion with EcoRI and ligated into the EcoRI site of the Pichia expression vector pPICZ␣A (Invitrogen, La Jolla, CA). The final construct in the expression vector (10 g) was linearized by digestion with HindIII and transformed into Pichia pastoris host strain X-33 by the lithium chloride transformation method as described in the Pichia expression manual (Invitrogen). Transformants were selected on yeast extract, peptone, dextrose (YPD)/Zeocin (100 g/ml) plates and screened for expression of the recombinant human ER ␣1,2-mannosidase enzyme activity as described below following methanol induction in small-scale liquid cultures (24, 38 -41). The Pichia transformant expressing the highest level of the soluble enzyme activity was used to produce the recombinant enzyme in large shaker flask and fermentor cultures. After optimizing the expression and purification of human ER ␣1,2mannosidase from 1-liter shaker flask cultures, enzyme expression was scaled-up by growth of the culture in a 100-liter fermentor (New Brunswick Scientific, Edison, NJ) at the Fermentation Research Facility of the University of Georgia. The fermentor, containing 100 liters of BMGY medium (400 g of biotin, 0.5% methanol, 1% glycerol, 1% yeast extract, 2% peptone, 0.1 M potassium phosphate (pH 7), and 1.34% yeast nitrogen base), was inoculated with a 5-liter overnight culture of the human ER ␣1,2-mannosidase Pichia transformant in BMGY medium. The culture was maintained at 30°C with agitation at 180 rpm and an air flow of 100 liters/min for 48 h until the glycerol in the culture was consumed. Human ER ␣1,2-mannosidase enzyme expression was then induced by the daily addition of methanol (50% (v/v) stock solution) to a final concentration of 0.5%. After 5 days of induction, the medium was harvested using a Sharples Model AS-16P continuous centrifuge at 15,000 rpm for 5 h, followed by filtration using a 0.45-m filter. Attempts to concentrate the culture medium by ultrafiltration resulted in Ͼ50% loss of enzyme activity and the appearance of human ER ␣1,2mannosidase as a precipitate on the ultrafiltration filter. As a result, the clarified culture medium was not concentrated, but was applied directly in 20-liter batches to an SP-Sepharose column (5 ϫ 13 cm; Amersham Pharmacia Biotech) at a flow rate of 16 ml/min. The column was washed with 500 ml of column buffer containing 10 mM sodium succinate (pH 6.0) and 1 mM CaCl 2 and eluted with a one-liter linear gradient of 0 -0.5 M NaCl in the same column buffer at a flow rate of 6 ml/min. The protein eluted as a sharp peak at ϳ400 mM NaCl. The fractions containing enzyme were pooled and concentrated by ultrafiltration through a YM-10 membrane (Amicon, Inc., Beverly, MA). The solubility of the concentrated protein was maintained by the addition of NDSB201 (Calbiochem) (42-44) to 1.0 M. After concentration, an EDTA-free protease inhibitor mixture (Roche Molecular Biochemicals, Mannheim, Germany) was added to prevent protease degradation. The concentrated enzyme preparation, in 5-7-ml batches, was further purified by loading onto a Superdex 75 gel filtration column (1.6 ϫ 65 cm; Amersham Pharmacia Biotech) pre-equilibrated with 20 mM MES (pH 7.0), 150 mM NaCl, 5 mM CaCl 2 , and 0.25 M NDSB201. Fractions containing human ER ␣1,2-mannosidase were pooled and concentrated to 1.8 mg/ml using a YM-10 membrane. The concentrated enzyme was stored at 4°C prior to use in crystallization trials. An aliquot of the purified enzyme (50 g) was subjected to SDS-polyacrylamide gel electrophoresis (45) and transferred to a polyvinylidene difluoride membrane to determine the N-terminal sequence as described previously (24). The protein (157 mg) was purified with a 16.7% yield from 100 liters of fermentor medium.
Solubility of Purified Recombinant Enzyme in the Presence of Nondetergent Sulfobetaines (NDSBs)-During purification, the recombinant protein was found to rapidly precipitate when stored at 4°C, even at low protein concentrations (Ͻ0.1 mg/ml). A series of solubility tests were therefore performed to investigate the effect of salts, detergents, glycerol, and NDSBs (43,44) on the solubility of the enzyme (data not shown). To test conditions for maintaining soluble enzyme during the crystallization trials, the purified enzyme (0.14 mg/ml) was concentrated in the presence of either 0.75 or 0.25 M NDSB256, NDSB201, or NDSB195 (Calbiochem) (43,44) or 1% glycerol using a Centricon-10 concentrator (Amicon, Inc.) to obtain a final protein concentration of ϳ4 mg/ml. The concentrated samples were stored at 4°C for 5 days. Aliquots were removed daily; and after centrifugation at 16,000 ϫ g for 1 min, the protein concentration of the supernatant was determined as described below. Glycerol and nonionic detergents were found to have a minimal effect on the solubility of the enzyme, with only 20 -30% of the protein remaining soluble after concentration and 10 -20% remaining soluble after 5 days. In contrast, samples containing the NDSB compounds resulted in ϳ65-81% of the enzyme remaining soluble after concentration and 25-40% remaining soluble after 5 days (data not shown). The most effective solubilizing agent was NDSB201. In NDSB201, 81% of the protein remained soluble after concentration, and 30% remained soluble after 5 days at 4°C. As a result of the solubility studies, the pooled enzyme preparation after the SP-Sepharose step was concentrated in the presence of 1.0 M NDSB201 prior to its application to the Superdex 75 column (as described above), and the gel filtration column was run with a buffer containing 0.25 M NDSB201.
Enzyme and Protein Assays-Human ER ␣1,2-mannosidase enzyme activity was determined by the addition of the enzyme sample to a final reaction volume of 40 l containing 20 mM MES (pH 7.0), 150 mM NaCl, 5 mM CaCl 2 and pyridylamine-tagged (PA) Man 9 GlcNAc 2 as an oligosaccharide substrate. The assays were incubated for 10 -20 min at 37°C, and the reactions were stopped by heating to 100°C for 5 min. The human ER ␣1,2-mannosidase product, Man 8 GlcNAc 2 -PA, was resolved from Man 9 GlcNAc 2 -PA on a Hypersil APS-2 NH 2 high-pressure liquid chromatography column (13,38). One unit of enzyme activity is defined as the amount of enzyme that generates 1 mol of Man 8 GlcNAc 2 from Man 9 GlcNAc 2 in 1 min at 37°C and pH 7.0. Protein concentration was determined using the BCA protein assay reagent (Pierce) as described by the manufacturer.
Crystallization and Initial Data Collection-Crystals of human ER ␣1,2-mannosidase were grown using the hanging-drop vapor diffusion method from equal volumes of protein solution (10 mg/ml in 20 mM MES (pH 7), 150 mM NaCl, 5 mM CaCl 2 , and 0.25 M NDSB201) and precipi-tating solution (1.6 -1.7 M (NH 4 ) 2 SO 4 ) suspended over a 1-ml reservoir containing the same precipitating solution. Rod-shaped crystals grew within ϳ5 days to a maximum size of 0.4 ϫ 0.2 ϫ 0.2 mm. The crystals are trigonal and belong to space group P3 1 21 with unit dimensions of a ϭ b ϭ 95.8 Å, c ϭ 136.8 Å, and ␥ ϭ 120°. One molecule is present in the asymmetric unit. The complexes of human ER ␣1,2-mannosidase with 1-deoxymannojirimycin and kifunensine were prepared by cocrystallization. The conditions described above for the uncomplexed protein were used with the inhibitor added to the precipitating solution. The concentration of inhibitor (0.5-25 mM) was screened to determine which concentration of inhibitor produced the best crystals. X-ray diffraction data were collected on crystals containing 0.5 mM kifunensine or 6.5 mM 1-deoxymannojirimycin. The protein-inhibitor crystals have the same space group and unit cell dimensions as the native uncomplexed crystals (see Table I). An initial set of data for the uncomplexed protein was collected from a single crystal at room temperature using CuK ␣ X-radiation (Rigaku Rotaflex RU200 rotating anode generator) on a Mar Research image plate detector (345-mm diameter). The crystal diffracted to a minimum d-spacing of 2.8 Å. High-resolution data for the native and inhibitor complexed crystals were collected at Beamline X8-C at the National Synchrotron Light Source (Brookhaven National Laboratory, Upton, NY) using a Quantum4 CCD detector and flashfrozen crystals. The crystals were cryoprotected prior to flash-freezing by soaking them in a 20% (v/v) glycerol/artificial mother liquor solution for ϳ3 min. All data were processed using DENZO/SCALEPACK (46). Final data reduction statistics are presented in Table I.
Structure Determination and Refinement-The uncomplexed human ER ␣1,2-mannosidase structure was solved by molecular replacement using the AMoRe program package (47) and the 2.8-Å room temperature data set (see Table I). The coordinates of yeast ␣1,2-mannosidase (Protein Data Bank code 1DL2) (29) were used as the search model with the following modifications: amino acid residues not conserved between the human and yeast sequences were truncated to alanine, and loops not present in the human enzyme were omitted. Rotation functions were calculated over a resolution range of 15 to 4 Å with a Patterson radius of 20 Å. The initial molecular replacement solution gave R cryst ϭ 46.4% and a correlation coefficient of 35.9%. The structure was refined using the CNS program suite (48). A maximum likelihood target (49) with a flat bulk solvent correction and no low resolution or cutoff applied to the data was used in the refinement protocol. Ten percent of the structure factors were randomly selected, excluded from the refinement, and used to compute R free (50). Refinement of the model using the simulated annealing slow-cooling protocol (51,52) was alternated with manual inspection and rebuilding of the model using TURBO-FRODO (53). After 16 cycles of refinement and manual rebuilding, 455 residues of the protein had been modeled. The structure comprises residues 241-388 and 391-697. Residues 226 -240, 389 -390, and 698 -699 could not be located owing to weak electron density in these regions. This model was subsequently used to refine the high-resolution native human ␣1,2-mannosidase (1.9 Å), HM⅐KIF (1.75 Å), and HM⅐dMNJ (2.38 Å) data (see Table I). Initial a -weighted difference (F o Ϫ F c ) electron density maps (see Fig. 1, a and b) were calculated using the native human ␣1,2-mannosidase model without any solvent molecules and allowed unambiguous location and orientation of the inhibitors and calcium ion. Models for 1-deoxymannojirimycin and kifunensine were generated and energy-minimized using the program SYBYL (Tripos Associates). The HM⅐dMNJ and HM⅐KIF structures were refined using the same protocol described above for the native protein. The progress of all three refinements was monitored by reductions in R cryst and R free . The final refinement statistics for all three molecules are presented in Table I. Each structure comprises residues 241-388 and 391-697. In addition to the protein, solvent, and inhibitor molecules (see Table I), each structure also contains one calcium and three sulfate ions. Analysis of the three structures using PROCHECK (54) showed that none of the non-glycine residues lie in disallowed regions of the Ramachandran plot.
Structure Alignment and Comparison-Structural alignment was carried out by superimposing the yeast ␣1,2-mannosidase, HM⅐dMNJ, or HM⅐KIF structure on the human ER ␣1,2-mannosidase structure using the RIGID option in TURBO-FRODO (53). The initial rigid body alignment used the coordinates of 28 structurally equivalent atoms located at the N and C termini of each helix of the (␣␣) 7 -barrel. An iterative least-squares fitting procedure was then carried out between all the C-␣ atoms lying at a distance progressively decreasing from 10 to 0.1 Å.

RESULTS AND DISCUSSION
Expression and Purification of Recombinant Human ER ␣1,2-Mannosidase-A soluble, enzymatically active form of the catalytic domain of human ER ␣1,2-mannosidase has been expressed in P. pastoris and purified to homogeneity (see "Experimental Procedures"). N-terminal sequencing of the purified protein yielded a single amino acid sequence of AEVP. This corresponds to an N terminus at amino acid 226 and indicates that, in addition to the cleavage of the vector-encoded ␣-factor signal sequence, an additional 54 N-terminal amino acids (residues 172-225) were excised by proteolytic cleavage during either the expression or purification of the protein. The purified protein encompasses amino acids 226 -699.
Structure Determinations-The purified recombinant human ER ␣1,2-mannosidase catalytic domain was crystallized using the hanging-drop vapor diffusion method, and its structure was determined using the molecular replacement technique. The coordinates of yeast class I ␣1,2-mannosidase were used as the probe molecule (29). The structure has been refined at 1.9-Å resolution to R cryst ϭ 22.2% and R free ϭ 25.0% (Table  I). The final model of the uncomplexed protein consists of residues 241-388 and 391-697 with overall dimensions of ϳ50 ϫ 50 ϫ 50 Å. The structures of the human ER ␣1,2mannosidase catalytic domain complexed with the inhibitors kifunensine and 1-deoxymannojirimycin have also been determined at 1.75-and 2.4-Å resolution, respectively. In each case, difference electron density clearly defined the position and orientation of the inhibitor molecule ( Fig. 1, a and b). The HM⅐KIF and HM⅐dMNJ structures have been refined to R cryst ϭ 21.9% and R free ϭ 24.1% and R cryst ϭ 19.0% and R free ϭ 23.9%, respectively (Table I).
Overall Structure of the Molecule-The core of the human ER ␣1,2-mannosidase structure is an (␣␣) 7 -barrel composed of 14 consecutive helices alternating from outside to inside the barrel (Fig. 2, A-C). The barrel has an approximate internal 7-fold symmetry. This arrangement of helices results in a topology of seven parallel inner helices (␣2, ␣4, ␣6, ␣8, ␣10, ␣12, and ␣14) and seven parallel outer helices (␣1, ␣3, ␣5, ␣7, ␣9, ␣11, and ␣13), concentric to the inner helices and anti-parallel to them. The structure is stabilized by a unique disulfide bond, Cys 527 -Cys 556 , which forms a bridge between helices 3 10 b and ␣11 (Fig.  2C). This disulfide bridge occurs between residues conserved in all members of the class I ␣1,2-mannosidase family and has been shown to be essential for activity (74).
The two ends of the (␣␣) 7 -barrel are structurally distinct (Fig. 2B). On one side, the short connection (SC side), the pairs of inner and outer helices are connected by short loops of up to four residues. The opposite side, the long connection (LC side), consists of a complex array of ␤-strands. The active site is on the long connection side of the protein. The C-terminal of the protein consists of a ␤-hairpin protruding back into the center of the inner barrel from the short connection side. This ␤-hairpin plugs the inner barrel and prevents the core of the protein from being an open channel. The ␤-hairpin, the inner helices, and the ␤-sheets on the long connection side of the barrel form a cavity ϳ15 Å deep, with an upper diameter of ϳ25 Å decreasing to ϳ10 Å at the top of the ␤-hairpin.
Calcium Binding in Human ␣1,2-Mannosidase-Calcium is essential for the activity of class I ␣-mannosidases (1)(2)(3)(4)(5)17) and has been shown to protect the yeast enzyme against thermal denaturation (30). In the human ␣1,2-mannosidase structure, the calcium ion binds to the carbonyl oxygen and O-␥ atoms of Thr 688 located at the top of the ␤-hairpin and to four water molecules, which are, in turn, hydrogen-bonded to one of the carboxylate groups of Glu 467 , Glu 599 , Glu 602 , and Glu 663 . Two additional water molecules complete the 8-fold pentagonal bipy-FIG. 1. Initial F o ؊ F c map for HM⅐KIF (a) and HM⅐dMNJ (b) complexes. The maps were calculated in the absence of the inhibitor or calcium ion. The maps are contoured at ϩ4 (thin lines) and ϩ20 (thick lines) for HM⅐KIF and at ϩ3.5 (thin lines) and ϩ15 (thick lines) for HM⅐dMNJ. The 1-deoxymannojirimycin in b is viewed at ϳ90°to the orientation of kifunensine. Please note that to facilitate the comparison of different inhibitors and the discussion of the catalytic mechanism, kifunensine has been numbered so that its six-membered ring has similar atom numbering to 1-deoxymannojirimycin and mannose. (This is not the standard IUPAC nomenclature.) The values given in parentheses are the completeness and R sym for the last resolution shell.
where F o and F c are the observed and calculated structure factors, respectively. For R free , the sum is extended over a subset of reflections (10%) excluded from all stages of refinement. R sym ϭ ⌺⌺͉I i Ϫ ͗I͉͘/⌺I i , where ͗I͘ is the average of equivalent reflections and the sum is extended over all measured observations for all unique reflections. c Rmsd, root mean square deviation; NSLS, National Synchrotron Light Source.
ramidal coordination (Table II). These water molecules, Wat 2 and Wat 3 , are located on the more accessible side of the calcium ion and are not hydrogen-bonded to any amino acid residues.
Kifunensine and 1-Deoxymannojirimycin Binding-Both kifunensine and 1-deoxymannojirimycin bind to the protein at the top of the C-terminal ␤-hairpin, at the bottom of the activesite cavity. Fig. 3A shows the location of kifunensine in the active-site cavity; 1-deoxymannojirimycin is not shown in Fig.  3A, but binds in the same location. The planes of both inhibitors' six-membered rings lie approximately parallel to the axis of the barrel. A key feature of the interaction between the protein and the inhibitors is the coordination of the calcium ion by the 2Ј-and 3Ј-hydroxyl groups of the six-membered rings of both inhibitors (Fig. 3, B and C). Whereas in various glycosylhydrolases (for examples, see Refs. 55 and 56), calcium has been found close to the active site, helping to stabilize the enzyme structure and imparting resistance to proteolysis and thermal denaturation (57), human class I ␣1,2-mannosidase is the first example of a glycosylhydrolase in which calcium is directly involved in inhibitor/substrate binding.
The calcium is 8-fold coordinated with a pentagonal bipyramidal geometry in contrast to the more common 7-fold coordination (58). Examples of 8-fold coordinated calcium ions include proteinase K (59) and C-type lectins such as the mannose-binding protein (60). One of the apices of the pyramid is occupied by the O-2Ј and O-3Ј hydroxyl groups of the inhibitors. In the absence of inhibitor, the 8-fold coordination of the calcium is maintained by two water molecules, Wat 2 and Wat 3 (Table II). Inhibitor binding appears to stabilize the calcium ion. If the average temperature factor for the protein and calcium ion are considered (Table I), a significant decrease in the calcium ion's temperature factor is observed in both HM⅐KIF (ϳ17 Å 2 ) and HM⅐dMNJ complexes (ϳ33 Å 2 ) ( Table I). The temperature factor for the calcium ion in the uncomplexed human ␣1,2-mannosidase structure is 40 Å 2 . This stabilization is related to a ϳ0.4-Å shift of the calcium ion in HM⅐KIF and HM⅐dMNJ structures toward the least accessible water molecule, Wat 1 . This water molecule is at the second apex of the bipyramidal coordination; the first apex is bisected by the O-2Ј and O-3Ј hydroxyls.
The hydroxyl groups of the six-membered rings of both 1-deresidues 389 -390. The secondary structure of the protein (see below) was assigned with the use of the program PROMOTIF (69) 7 -barrel axis from the long connection (LC side) (A) and at 90°to the first orientation (B). The calcium ion (Ca) is represented as a dark blue sphere. SC side, short connection. Also shown is a topological two-dimensional representation of the human ␣1,2-mannosidase structure (C). The figure is oriented as described for A. ␣-Helices are represented as circles, and ␤-strands as shown as arrows. For simplicity, helices ␣6a and ␣6b and helices ␣14a and ␣14b are represented as single helices. The essential disulfide bridge at Cys 527 -Cys 556 (S1) is also shown. The color scheme for the ␣-helices is conserved in all three panels. The ␤-hairpin that plugs the barrel is pink, and the 3 10 helices are black. The dashed line represents the discontinuity in the model at  3. Binding of 1-deoxymannojirimycin and kifunensine to human ER class I ␣1,2-mannosidase. A, location of kifunensine in the center of the (␣␣) 7 -barrel. 1-Deoxymannojirimycin superimposes with the six-membered ring of kifunensine, but for clarity, it is not represented in this panel. The color scheme is the same as described in the legend to Fig. 2. B and C, schematic representation of the interactions between human ␣1,2-mannosidase and kifunensine and 1-deoxymannojirimycin, respectively. Short and long dashed lines represent hydrogen bond interactions and van der Waals contacts, respectively. For simplicity, only hydrogen bonds between the protein, water, and inhibitor molecules are represented. Water-water hydrogen bonds are not represented. D, surface representation of the catalytic cavity of human ␣1,2-mannosidase in the vicinity of the kifunensine-binding site. The surface is colored according to its electrostatic potential. Kifunensine is shown in stick representation. The contour level is at Ϯ20 kT. A was prepared using MOLSCRIPT (70), and D was prepared with GRASP (72). oxymannojirimycin and kifunensine hydrogen bond directly to Arg 597 , Glu 599 , Glu 663 , and Glu 689 and via water molecules to Glu 330 , Glu 602 , Glu 467 , and Asp 463 (Fig. 3, B and C). These acidic residues define the bottom of the active-site cavity and have been shown to be important for catalysis (30). The O-6Ј hydroxyl group of the inhibitor is completely buried in a small side recess of the active-site cavity (Fig. 3D). The O-6Ј hydroxyl forms a short hydrogen bond with O-⑀1 of Glu 599 (ϳ2.5 Å) and a longer one with N-1 of Arg 597 (ϳ2.9 Å). In the case of kifunensine, additional interactions with the protein are made with atoms in the fused five-membered ring. The O-7 carbonyl and N-9 hydrogen bond directly to N-1 of Arg 597 and O-␦2 of Asp 463 , respectively. Additional hydrogen bond interactions are made via water molecules. N-9 interacts via water with O-␥ of Ser 464 , the O-8 carbonyl with the amide nitrogen of Arg 461 , and the O-9 carbonyl with N-2 of Arg 597 (Fig. 3B). Most of the contacts between the protein and inhibitor molecules are electrostatic interactions. The only van der Waals interactions found are between inhibitor atoms C-4, C-5, and C-6 and Phe 659 and C-7 and C-8 in kifunensine and Leu 525 (Fig. 3, B and C).
The protein does not undergo any large global conformation changes upon inhibitor binding. A superposition of all C-␣ atoms of human ␣1,2-mannosidase and the HM⅐KIF or HM⅐dMNJ complex yielded a root mean square deviation of 0.25 Å for both structures. Although small variations in some side chains do occur, the largest conformational change seen upon inhibitor binding is a movement of ϳ1.7 Å in the side chain of Arg 597 . In addition to the interactions of Arg 597 with the inhibitor (Fig. 3, B and C), N-2 and N-⑀ of Arg 597 now hydrogen bond with O-␥1 and O-␥2 of Glu 570 , respectively. The similarities between all three structures extend beyond the protein to the solvent. The solvent structure is essentially identical in all three protein structures, with the exception of the water molecules that are displaced from the active site upon inhibitor binding. In the HM⅐KIF and HM⅐dMNJ structures, each of the four hydroxyl oxygen atoms of the inhibitor replace a water molecule. An additional water molecule is displaced in the HM⅐KIF structure by N-9. The active-site cavity therefore appears to be preformed in the human ␣1,2mannosidase structure prior to inhibitor/substrate binding. Only one residue, Arg 597 , undergoes a significant conformational change upon inhibitor binding, and ordered water molecules form hydrogen bonds with the protein in a pattern that closely mimics the hydrogen bond network found in the presence of the inhibitor. This is comparable to the active sites of other saccharide-binding proteins where ordered water molecules have been observed to mimic the positions of the inhibitor/substrate hydroxyl atoms (61).
Conformation of Kifunensine and 1-Deoxymannojirimycin-The electron density of kifunensine and 1-deoxymannojirimycin reveals that the six-membered rings of both inhibitors have a non-standard 1 C 4 conformation when bound at the active site of the protein. This "all-axial" conformation of kifunensine's six-membered ring is probably the consequence of the fused five-membered ring. The energy-minimized structure of kifunensine built using the program SYBYL fit the initial 1.7-Å a -weighted F o Ϫ F c difference electron density maps without any conformational rearrangement (Fig. 1a) or ambiguity in the ring pucker. In contrast, the energy-minimized 1-deoxymannojirimycin molecule, built in the 4 C 1 conformation, required deformation to the 1 C 4 conformation in order for the molecule to fit the electron density (Fig. 1b). The conformation of the six-membered ring of 1-deoxymannojirimycin is almost certainly defined by its interactions with the protein (see above). The lack of conformational rearrangement necessary for kifunensine upon binding to the protein explains, in part, why this inhibitor has a higher affinity for the protein than 1-deoxymannojirimycin (32). The additional hydrogen bond interactions between the fused five-membered ring and the protein also contribute to the increased affinity of kifunensine.
Comparison between Human and Yeast ␣1,2-Mannosidase Structures-Pairwise superimposition of all C-␣ atoms of the yeast and human ␣1,2-mannosidase structures yielded a root mean square deviation of 1.44 Å, indicating that the overall structures of both proteins are essentially the same (Fig. 4, A-C). Although differences are observed between the two structures (Fig. 4A), these do not affect the positions of either of the critical active-site residues that are conserved in all class I ␣1,2-mannosidases or the calcium ion (Fig. 5B). Given the similarities in the structures and the lack of conformational changes seen in human ␣1,2-mannosidase upon inhibitor binding, it is reasonable to assume that both inhibitors would bind to yeast ␣1,2-mannosidase in a manner similar to that found for the human protein. In the yeast ␣1,2-mannosidase crystal structure, an N-glycan from one protein molecule interacts with a symmetry-related protein molecule in what is believed to be an enzyme-product complex (29). Comparison of the yeast ␣1,2-mannosidase and HM⅐dMNJ or HM⅐KIF structures reveals that 1-deoxymannojirimycin should bind to yeast ␣1,2mannosidase, whereas steric hindrance would occur between the five-membered ring of the kifunensine molecule and the middle-arm Man 7 of the N-glycan. These observations support the experimental data. Yeast ␣1,2-mannosidase crystals dissolve immediately when low concentrations of kifunensine are added to the mother liquor, but remain stable when soaked in high concentrations of 1-deoxymannojirimycin. Although a structure of yeast ␣1,2-mannosidase complexed with 1-deoxymannojirimycin has recently been determined, 2 attempts to co-crystallize yeast ␣1,2-mannosidase with kifunensine have systematically failed.
In the yeast ␣1,2-mannosidase structure, a glycerol molecule was found at the bottom of the active-site cavity (29). The structural comparison reveals that the glycerol molecule superimposes with the O-6, C-6, C-5, C4, and O-4 atoms of the inhibitor molecules (Fig. 5B). This observation reinforces the hypothesis that glycerol mimics saccharide binding (62) and supports the previous conclusion that the glycerol molecule partly occupies the putative binding site for Man 10 (29) (Fig. 5A).
Taken together, these observations strongly suggest that the positions of the six-membered rings of the inhibitor molecules mimic the location of the mannose residue cleaved during the catalytic mechanism. The interaction of the O-2Ј and O-3Ј hydroxyl groups of the mannose residue with the calcium ion, the interaction of the O-6Ј hydroxyl with the strictly conserved residues Arg 597 and Glu 599 , and the non-bonded interactions of C-4, C-5, and the C-6 methylene with Phe 659 would almost certainly restrict the orientation of the terminal mannose at the bottom of the active site. The excellent steric complementarity between the inhibitors and the active-site cavity (Fig.  3D) also supports the proposal and suggests that the C-1 atoms of kifunensine and 1-deoxymannojirimycin are in positions homologous to C-1 of Man 10 . The C-1 atoms of the HM⅐KIF and HM⅐dMNJ structures are ϳ2.18 and 2.26 Å, respectively, from the O-2Ј hydroxyl of Man 7 found in the yeast ␣1,2-mannosidase structure. A flattening of the Man 10 ring structure at C-1 into a skewed-boat conformation and/or distortion of the glycosidic linkage would enable the covalent linkage between C-1 of Man 10 and O-2 of Man 7 to be made without any other significant rearrangements being necessary (Fig. 5C).
In other inverting glycosylhydrolases such as cellulase EG1, it has been suggested that the protein has evolved to optimally bind the transition state (63). The close complementarity found between inhibitor and protein, the lack of conformational change in the protein upon inhibitor binding, and the similarity of the water molecule positions to the hydroxyls of the inhibitors all seem to suggest that the class I ␣1,2-mannosidases have also evolved to optimally bind the transition state rather than the substrate. The distortion of the substrate to the 1 C 4 conformation would appear to be dictated by its interactions with the protein and the shape of the active-site cleft, which is preformed prior to saccharide binding.
Catalytic Residues and Role of Calcium in the Catalytic Mechanism-The ␣1,2-mannosidase is an inverting glycosylhydrolase that specifically cleaves the ␣1,2-oligosaccharide linkage between C-1 of Man 10 and O-2 of Man 7 (Fig. 5A) with inversion of the anomeric configuration at C-1 and formation of a hydroxyl group at O-2. The catalytic mechanism usually involves two carboxylic acids separated by a distance of ϳ9.5 Å, one acting as a general base removing a proton from water and the other acting as a general acid donating a proton to the leaving group. Given the geometrical constraints expected for suitably positioned catalytic residues in inverting enzymes, the yeast ␣1,2-mannosidase structure suggests that the only carboxylic acid groups that could be involved in the catalytic mechanism are Glu 132 , Asp 275 , and Glu 435 in the yeast enzyme. These correspond to Glu 330 , Asp 463 , and Glu 599 in the human structure (29). These residues are also the only candidates for catalytic residues in the human protein (Fig. 5B). From the positions of the inhibitor molecules in the HM⅐KIF and HM⅐dMNJ structures, it is clear that Asp 293 , Glu 397 , Glu 602 , FIG. 5. Catalytic mechanism. A, schematic representation of the high-mannose Man 9 GlcNAc oligosaccharide showing the ␣1,2-linkage that is cleaved and numbering of the saccharide units. B, structural superimposition of the active-site region of HM⅐dMNJ, HM⅐KIF, and yeast class I ␣1,2-mannosidases. The Man 7 (M7) residue of the middle-arm branch of the N-glycan, the amino acid residues, and the glycerol molecule of yeast ␣1,2-mannosidases are yellow. The HM⅐KIF and HM⅐dMNJ structures are blue and green, respectively. Calcium, whose position is invariant in the three structures, is shown in dark blue. C, close-up of the putative linkage between the O-2Ј atom of Man 7 in yeast ␣1,2-mannosidases and the C-1 atom of 1-deoxymannojirimycin in the HM⅐dMNJ structure. The atom colored red lies in the plane defined by the nitrogen, C-2, C-3, and C-5 atoms and represents the putative deformation of the ring during catalysis. D, proposed catalytic mechanism as described under "Results and Discussion." W, water. Glu 663 , Glu 657 , and Glu 689 cannot be directly involved in catalysis. These residues either are completely buried at the bottom of the catalytic site or interact with atoms on the inhibitor at some distance from the anomeric C-1 atom (Fig. 3, B and C).
The structure of yeast ␣1,2-mannosidase enabled two hypotheses regarding the catalytic mechanism to be proposed (29). Please note that the numbering used through out this report is that of human ER class I ␣1,2-mannosidase (14). Please see the legend of Fig. 4A for the equivalent yeast residues. In the first hypothesis, Glu 330 was suggested to be the catalytic base, abstracting a proton from a water molecule, which, in turn, attacked the C-1 atom. The identification of the catalytic acid was not possible due to the lack of direct interactions between the O-2Ј hydroxyl of Man 7 and any protein residue. Asp 463 and Glu 599 were the most likely candidates, as they are on the opposite side of the glycosidic linkage to be cleaved to Glu 330 at distances of 9.5 and 9.6 Å, respectively. Asp 463 was suggested to be the more attractive candidate for the catalytic acid, as it is closer to O-2 of Man 7 than Glu 599 . The second, less favored hypothesis implicated Glu 599 as the catalytic base. Glu 599 is hydrogen-bonded to a water molecule, which, in turn, is hydrogen-bonded to the glycerol molecule and calcium ion. In this instance, Glu 330 was suggested to be the most likely candidate for the catalytic acid, as Asp 463 is on the same side of the oligosaccharide as Glu 599 .
Inspection of the HM⅐KIF and HM⅐dMNJ models shows that the six-membered ring structures of both inhibitors are in a 1 C 4 conformation. Given this ring pucker, if Glu 330 is the catalytic base, then attack by a water molecule would result in retention, not inversion, of the anomeric configuration at C-1. Consequently, the attack by water must come from the other side of the inhibitor/substrate molecule, making Glu 599 and Asp 463 the only candidates for the general base. If Glu 599 acts as the base, the only appropriately positioned water molecule on this side of the inhibitor to act as the nucleophile would be Wat 5 (Fig. 5B). This water molecule is hydrogen-bonded not only to Glu 599 , but also to the calcium ion (Fig. 3, B and C), and is at a distance of 3.4 Å from the anomeric C-1 atom. This distance could, however, be longer if the Man 10 saccharide undergoes distortion to a skewed-boat conformation and/or distortion of the glycosidic linkage, as could occur when the Man 10 saccharide is covalently linked to Man 7 . Given Glu 599 as the base, the most likely candidate for the catalytic acid is Glu 330 , as Asp 463 is on the same side of the inhibitor/ substrate as Glu 599 . Glu 330 is ϳ4.4 Å from O-2 of Man 7 and therefore too far away from O-2 for direct attack. A water molecule, Wat 8 , however, bridges between the carboxylate of Glu 330 and O-2 of Man 7 . Given that no significant conformational change is seen upon inhibitor binding, it is hard to envisage that a large conformational change occurs during catalysis. This would suggest that a second water molecule is involved in the catalytic mechanism acting as the acid (Fig. 5D). Although unlikely because it is on the same side of the inhibitor/substrate as Glu 599 , Asp 463 could alternatively act as the catalytic acid. This residue, like Glu 330 , does not directly hydrogen bond to O-2 of Man 7 , but interacts via a water molecule.
An alternative mechanism could be envisaged with Asp 463 as the catalytic base. A water molecule, Wat 9 , is seen in almost the same position in both inhibitor structures to hydrogen bond to Asp 463 and the O-2Ј hydroxyl of the inhibitor molecule (Figs. 3 (B and C) and 5B). In the HM⅐KIF structure, this water molecule also hydrogen bonds to N-9. This water molecule is 3.44 and 3.78 Å from the anomeric C-1 atom in the HM⅐dMNJ and HM⅐KIF structures, respectively. This distance could potentially decrease if the proposed distortion of Man 10 to the skewed-boat conformation and/or deformation of the glycosidic linkage occurred. If Asp 463 were the catalytic base, Glu 330 would be the best candidate for the catalytic acid, as it is on the opposite side of inhibitor/substrate to Asp 463 . As described above, if Glu 330 were the catalytic acid, an additional water molecule would have to be involved in the mechanism, as Glu 330 is not within hydrogen-bonding distance of O-2 of Man 7 , and no large conformational changes are expected to occur during catalysis. In this mechanism, calcium is not directly involved in the catalytic mechanism, but would solely be involved in stabilizing the conformation of the Man 10 saccharide through its interactions with the O-2Ј and O-3Ј hydroxyls. Similarly, in this mechanism, Glu 599 would only be involved in stabilizing the O-6Ј hydroxyl of the inhibitor rather than having a dual role and also acting as the base.
Although there remains some ambiguity in the identity of the catalytic acid/base, it is obvious from the inhibitor structures and the yeast ␣1,2-mannosidase enzyme-product complex (29) that the catalytic mechanism must deviate from the classical inverting enzyme mechanism. As no large conformational changes in the protein are expected during catalysis, and no acidic group is within hydrogen-bonding distance of O-2 of Man 7 , it would appear that a water molecule must play the role of the acid (Fig. 5D). The first of the two hypotheses also invokes a direct role for the calcium ion in the catalytic mechanism, as the water molecule activated by the putative catalytic base is coordinated by the calcium ion (Fig. 3, B and C).
Although not common, calcium has been suggested to activate a water nucleophile in the catalytic mechanisms of inosineuridine N-ribohydrolase (64) and staphylococcal nuclease from Staphylococcus aureus (65). Although there is no direct experimental evidence that the mode of catalysis for this glycosylhydrolase is different from that of other inverting enzymes, deviations in the classical mechanism have been observed for other enzymes (66).
Specificity and Inhibition of Class I ␣1,2-Mannosidase-Calcium is essential for the activity of class I ␣1,2-mannosidases (1)(2)(3)(4)(5)17). The HM⅐KIF and HM⅐dMNJ structures provide the first direct evidence of the role of calcium in substrate binding. The calcium ion coordinates to the O-2Ј and O-3Ј hydroxyls of the inhibitors and appears to help stabilize the 1 C 4 conformation of 1-deoxymannojirimycin. In addition, the equatorial conformation of the C-5-C-6 linkage is stabilized by the association of the O-6Ј hydroxyl with Glu 599 and Arg 597 in a side pocket of the active-site cleft (Fig. 3D) and by the van der Waals interactions that the C-6 methylene and C-5 and C-4 atoms make with Phe 659 . These interactions appear to be the most critical for inhibitor/substrate binding and help define the specificity of this class of enzymes.
Class I ␣1,2-mannosidases do not cleave ␣1,3or ␣1,6-linked mannose residues (1)(2)(3)(4)(5). From the yeast ␣1,2-mannosidase enzyme-product complex, it is obvious that an ␣1,3or ␣1,6-linked mannose would not be able to bind to the protein in the same location as kifunensine and 1-deoxymannojirimycin. The difference in the orientation of the saccharide ring when linked to the M7 mannose Man 7 through an ␣1,3or ␣1,6-linkage would result in numerous steric clashes between the protein and the oligosaccharide. The inability to cleave ␣1,3or ␣1,6-linked mannose residues will also be determined by the oligosaccharide's ability to interact with all the protein's saccharide-binding sites. The enzymes are also specifically inhibited by 1-deoxymannojirimycin and kifunensine, but not by 1-deoxynojirimycin, castanospermine, swainsonine, or 1,4dideoxy-1,4-imino-D-mannitol. Since the only difference between 1-deoxymannojirimycin and 1-deoxynojirimycin is the orientation of the O-2Ј hydroxyl (Fig. 6), the specificity of the enzyme for these inhibitors would appear to be dictated by the inhibitor's ability to coordinate to the calcium ion. Calcium would be unable to coordinate the O-2Ј hydroxyl in 1-deoxynojirimycin and therefore cannot stabilize the inhibitor in a 1 C 4 conformation. The O-2Ј and O-3Ј hydroxyls of castanospermine are in the same orientation as those of 1-deoxynojirimycin, suggesting that this inhibitor is also unable to coordinate the calcium ion. These observations confirm the critical role of the O-2Ј hydroxyl.
The necessity for the substrate or inhibitor to coordinate the calcium ion and the geometry of the active site, especially the side pocket for the C-6 -O-6Ј atoms, also explain the lack of inhibition of class I ␣1,2-mannosidases by swainsonine and the poor inhibition by 1,4-dideoxy-1,4-imino-D-mannitol (67) (Table  III). Swainsonine and 1,4-dideoxy-1,4-imino-D-mannitol are potent inhibitors of the class II mannosidases such as Golgi ␣-mannosidase II and lysosomal ␣-mannosidase (68). In contrast to kifunensine, 1-deoxymannojirimycin, castanospermine, and 1-deoxynojirimycin, which are pyranose mannose and glucose substrate mimics, the class II mannosidase inhibitors swainsonine and 1,4-dideoxy-1,4-imino-D-mannitol are thought to mimic the ring-flattened transition state mannosyl cation of these retaining enzymes as furanose analogs of mannose (68) (Fig. 6). The energy-minimized structures of swainsonine and 1,4-dideoxy-1,4-imino-D-mannitol show that although the furanose ring will have a different pucker to the 1 C 4 conformation of the pyranose ring of 1-deoxymannojirimycin and kifunensine, the O-2Ј and O-3Ј hydroxyls are in the same orientation in all four inhibitors (Fig. 6). This suggests that both swainsonine and 1,4-dideoxy-1,4-imino-D-mannitol should be able to coordinate to the calcium ion. The difference in the structure of 1,4-dideoxy-1,4-imino-D-mannitol and swainsonine explains why the enzyme is inhibited poorly by 1,4-dideoxy-1,4imino-D-mannitol (67) and insensitive to swainsonine (Table  III). Modeling of swainsonine in the active site of human ER class I ␣1,2-mannosidase such that the O-2Ј and O-3Ј hydroxyls of swainsonine coordinate the calcium ion indicates numerous steric clashes of the pyranose ring with the protein, especially with the strictly conserved Phe 659 . Unlike the six-membered ring of swainsonine, the C-5/C-6/O-6Ј arm of 1,4-dideoxy-1,4imino-D-mannitol is conformationally flexible. The flexibility of this C-5/C-6/O-6Ј arm would allow the inhibitor to bind to the protein, although the interactions may be less than optimal. This would explain the poor inhibitory properties of this inhib- a The IC 50 values were determined as described under "Experimental Procedures" using a constant 3 ng/l enzyme and 4 M Man 9 GlcNAc 2 -PA substrate and varying the concentrations of the inhibitor.
FIG. 6. Chemical structures of the class I ␣1,2-mannosidases inhibitors 1-deoxymannojirimycin and kifunensine, the class II ␣-mannosidase inhibitors swainsonine and 1,4dideoxy-1,4-imino-D-mannitol, as well as the specific ER ␣-glucosidase I and II inhibitors castanospermine and 1-deoxynojirimycin. Please note that to facilitate the comparison of different inhibitors and the discussion of the catalytic mechanism, all of the inhibitors have been numbered so that they have similar atom numbering to 1-deoxymannojirimycin and mannose. (This is not the standard IUPAC numbering.) itor. These data are also consistent with the fact that class II ␣-mannosidases are calcium-independent retaining enzymes whose catalytic acidic residues do not share the same geometrical constraints encountered in the class I inverting enzymes.
The ability of inhibitors to coordinate to calcium appears to be the first and most essential criterion to be considered when designing new inhibitors for class I ␣1,2-mannosidases. The preference for 1-deoxymannojirimycin over 1-deoxynojirimycin and castanospermine, the lack of inhibition of the enzyme by swainsonine, and the weak inhibition by 1,4-dideoxy-1,4-imino-D-mannitol all point to the critical role of correctly orientated O-2Ј and O-3Ј hydroxyls in a pyranose ring structure, not only for substrate specificity, but also for inhibitor specificity. Other factors that need to be considered include the ability to bind to the protein in the 1 C 4 conformation rather than the lower free energy 4 C 1 conformation and the presence of substituents at C-5 capable of forming non-bonded interactions with Phe 659 and hydrogen bonding to Arg 597 and Glu 599 in the small side recess of the active site. The pyranose rings of kifunensine and 1-deoxymannojirimycin are identical (Fig. 6). However, kifunensine binds to ␣1,2-mannosidases with higher affinity than 1-deoxymannojirimycin, indicating that the five-membered ring and its substituents are also important components of inhibitor binding and therefore design. The inhibitor structures presented in this study therefore provide the first structural data regarding the specificity of this class of enzymes for ␣1,2mannose residues and a blueprint for future inhibitor design of novel therapeutic agents for the treatment of genetic diseases that are characterized by rapid degradation of misfolded glycoproteins.