Structural Basis for Recognition of High Mannose Type Glycoproteins by Mammalian Transport Lectin VIP36*

VIP36 functions as a transport lectin for trafficking certain high mannose type glycoproteins in the secretory pathway. Here we report the crystal structure of VIP36 exoplasmic/luminal domain comprising a carbohydrate recognition domain and a stalk domain. The structures of VIP36 in complex with Ca2+ and mannosyl ligands are also described. The carbohydrate recognition domain is composed of a 17-stranded antiparallel β-sandwich and binds one Ca2+ adjoining the carbohydrate-binding site. The structure reveals that a coordinated Ca2+ ion orients the side chains of Asp131, Asn166, and His190 for carbohydrate binding. This result explains the Ca2+-dependent carbohydrate binding of this protein. The Man-α-1,2-Man-α-1,2-Man, which corresponds to the D1 arm of high mannose type glycan, is recognized by eight residues through extensive hydrogen bonds. The complex structures reveal the structural basis for high mannose type glycoprotein recognition by VIP36 in a Ca2+-dependent and D1 arm-specific manner.

In eukaryotic cells, post-translational modification of secreted proteins and intracellular protein transport between organelles are ubiquitous features. One of the most studied systems is the N-linked glycosylation pathway in the synthesis of secreted glycoproteins (1)(2)(3). The N-linked glycoproteins are subjected to diverse modifications and are transported through the endoplasmic reticulum (ER) 3 via the Golgi apparatus to their final destinations inside and outside of the cell. Incorporation of the cargo glycoproteins into the transport vesicles is mediated by transmembrane cargo receptors, which have been identified as intracellular lectins. For example, mannose 6-phosphate receptor (4) functions as a cargo receptor for lysosomal proteins in the trans-Golgi network, whereas ER-Golgi intermediate compartment (ERGIC)-53 (5,6) and its yeast orthologs Emp46/47p (7) are transport lectins for glycoproteins that are transported out of the ER.
VIP36, vesicular-integral protein of 36 kDa, was originally isolated from Madin-Darby canine kidney cells as a component of detergent-insoluble, glycolipid-enriched complexes containing apical marker (8). Confocal and immunoelectron microscopic experiments have suggested that VIP36 is distributed by either the pre-Golgi secretory pathway (9 -11) or post-Golgi pathway (8,12). Furthermore we showed that VIP36 is involved in intracellular transport, in the secretion of glycoproteins (e.g. clusterin) from polarized Madin-Darby canine kidney cells (13), and in the secretion of ␣-amylase from rat parotid glands (14). Taken together, VIP36 appears to play significant roles not only in vesicular transport from the ER to the Golgi complex but also from the trans-Golgi network to the plasma membrane.
The exoplasmic/luminal domain of VIP36 as well as the luminal domain of ERGIC-53 and Emp46/47p share homology with L (leguminous)-type lectins and are thus identified as carbohydrate recognition domains (CRDs). It has been shown that ERGIC-53 interacts with glycoproteins carrying high mannose type glycan by endo-␤-N-acetylglucosaminidase H treatment (15)(16)(17) and binds glycoproteins in a Ca 2ϩ -and pH-dependent manner (18). We have previously found that VIP36 has high affinity for high mannose type glycans containing Man-␣-1,2-Man residues in Man 7-9 (GlcNAc) 2 -Asn peptides (19). Kamiya et al. (20) recently reported in detail the carbohydrate binding properties of VIP36 assayed by frontal affinity chromatography. This work suggested the Ca 2ϩ dependence of carbohydrate binding and the specificity for the D1 arm, Man-␣-1,2-Man-␣-1,2-Man residues, of high mannose type glycans (corresponding to Man(D1)-Man(C)-Man(4); Fig. 1). In addition, using a flow cytometry-based method, it was also demonstrated that VIP36 binds glycoproteins carrying high mannose type glycans (21). These observations suggested that VIP36 is involved in the transport of glycoproteins via high mannose type glycans.
Crystal structures of the CRD of rat ERGIC-53 in the absence and presence of Ca 2ϩ have been determined, confirming its structural similarity to the L-type lectins (22,23). In these reports, it was shown that the putative ligand-binding site of ERGIC-53 is similar to the mannose-binding site of the L-type lectins. Very recently, we reported the crystal structures of the CRD of Ca 2ϩ -independent K ϩ -bound Emp46p and the metalfree form of Emp47p (24). To date, however, no structures of transport lectins in complex with high mannose type glycans have been determined. To investigate the structural basis of the mechanism of high mannose type glycoprotein recognition by VIP36, we determined crystal structures of the exoplasmic/luminal domain of VIP36 alone and in complex with Ca 2ϩ and mannose, Man-␣-1,2-Man (termed Man 2 , which corresponds to Man(D1)-Man(C), Man(C)-Man (4)
Protein Expression and Purification-The DNA fragment for residues 51-301, which correspond to the CRD and part of the stalk domain of canine VIP36, was cloned into the BamHI and EcoRI sites of pGEX4T-1 plasmid (GE Healthcare). The recombinant VIP36 was expressed in Escherichia coli BL21(DE3). Cells were harvested after induction with 0.1 mM isopropyl ␤-D-thiogalactoside (Wako) for 15 h at 20°C and lysed by sonication in phosphate buffered saline buffer containing 2 mM CaCl 2 . The cell lysate was loaded on a glutathione-Sepharose 4B column (GE Healthcare). The glutathione S-transferase (GST) fusion protein was eluted by glutathione (Wako) and cleaved by thrombin protease (GE Healthcare). The cleaved proteins were passed through a glutathione-Sepharose 4B column to remove GST protein and further purified by a benzamidine-Sepharose 4FF column (GE Healthcare) to remove the thrombin protease. The purified protein was dialyzed against 10 mM MES (pH 6.5) and 2 mM CaCl 2 .
Crystallization and X-ray Data Collection-Initial crystallization conditions were screened using the Large Scale Protein Crystallization and Monitoring System (PXS) (33). The crystallization conditions of VIP36 in its Ca 2ϩ -bound form were obtained in a buffer containing 18 mg ml Ϫ1 protein, 15% (w/v) polyethylene glycol 4000, 1.5 M NaCl, and 0.1 M MES (pH 6.5) with incubation at 277 K for 4 days. For the metal free-form, the Ca 2ϩ -bound crystal was soaked with this buffer containing 10 mM EDTA to remove Ca 2ϩ . The crystal of Ca 2ϩ .Man-bound VIP36 was obtained by soaking the Ca 2ϩ -bound crystal with the buffer containing 50 mM D-mannose (Sigma). The Ca 2ϩ .Man 2bound VIP36 was co-crystallized in a buffer containing 10 mg ml Ϫ1 protein, 3.4 mM 2␣-mannobiose (Sigma), 5% (w/v) polyethylene glycol 4000, 0.3 M MgCl 2 , and 0.1 M MES (pH 6.5) with incubation at 277 K for 1 week. On the other hand, the Ca 2ϩ .Man 3 GlcNAc-bound VIP36 was co-crystallized in a buffer containing 10 mg ml Ϫ1 protein, 3.4 mM Man 3 GlcNAc, 10% (w/v) polyethylene glycol 4000, and 0.4 M imidazole malate (pH 6.0) with incubation at 277 K for 3 weeks. Despite the extensive co-crystallization with Man 3 GlcNAc the electron density map did not show any interaction with the GlcNAc moiety; only the Man 3 portion was recognized by VIP36. Data sets of the metal-free and Ca 2ϩ .Man 2 -and Ca 2ϩ .Man 3 GlcNAcbound forms were collected under cryogenic conditions with crystals soaked with a cryoprotectant buffer containing 20% (v/v) glycerol. The Ca 2ϩ -, Ca 2ϩ .Man-bound crystals were soaked with a buffer containing 2.5 M LiCl instead of 1.5 M NaCl for data collection under cryogenic conditions. The diffraction data were processed using HKL2000 (34). The metal-free and Ca 2ϩ -, Ca 2ϩ .Man-, and Ca 2ϩ .Man 2 -bound crystals belong to space group C2 with two molecules per asymmetric unit. In contrast, the Ca 2ϩ .Man 3 GlcNAc-bound crystal belongs to space group P2 1 2 1 2 1 with five molecules per asymmetric unit. The crystallographic parameters of VIP36 are shown in Table 1.
Structure Determination and Refinement-The crystal structure of VIP36 was solved by the molecular replacement method using the program MOLREP (35) with the Ca 2ϩ -bound ERGIC-53 (Protein Data Bank code 1R1Z) (23) as a search model. The refinement procedures were carried out with Crystallography and NMR System (CNS) (36) and REFMAC5 (37). Model fitting to the electron density maps was performed manually using Coot (38). The stereochemical quality of the final models was assessed by PROCHECK (39). Final refinement statistics are summarized in Table 1. Figures were prepared using GRASP (40), LIGPLOT, (41), and PyMOL (42).

RESULTS
Crystallization and Overall Structure of Exoplasmic/Luminal Domain of VIP36-The exoplasmic/luminal domain (residues 51-301) of VIP36, corresponding to the CRD and part of the short stalk domain, was crystallized. Despite extensive crystallization screening, diffraction quality crystals of the CRD (residues 51-278) alone could not be produced. The crystal structure of the exoplasmic/luminal domain of Ca 2ϩ -bound VIP36 was solved by the molecular replacement method using the Ca 2ϩ -bound ERGIC-53 CRD (Protein Data Bank code 1R1Z) (23) as a search model. The crystal belongs to space group C2 with two molecules (A and B) per asymmetric unit. Both VIP36 molecules are related by ϳ2-fold symmetry, forming a pseudo-homodimer. However, gel filtration data demonstrate that this protein is monomeric in solution (supplemental Fig. 1).
To obtain crystals of mannose-bound VIP36, the Ca 2ϩbound crystals were soaked in a solution containing mannose in molar excess. The Man 2 -bound form was obtained by co-crystallization. To obtain the metal-free form, the Ca 2ϩ -bound crystal was soaked with buffer containing 10 mM EDTA. Crystallization of VIP36 in the absence of Ca 2ϩ was not successful. Following treatment with EDTA the Ca 2ϩ is completely absent from molecule B, whereas approximately half of the Ca 2ϩ ions are removed from molecule A. In the C2 crystal form, crystallization of VIP36 in complex with longer oligomannoses was not successful due to the crystal packing around the ligand-binding site. To find other crystal forms, we therefore carried out crystallization screening in the presence of Ca 2ϩ and longer oligomannoses, Man-␣-1,2-Man-␣-1,2-Man, Man-␣-1,2-Man-␣-1,3-Man, Man 3 GlcNAc, and Man 6 (GlcNAc) 2 -Asn. Diffraction quality crystals were obtained from co-crystallization with the Man 3 GlcNAc alone. The crystal belongs to space group P2 1 2 1 2 1 with five molecules (A, B, C, D, and E) per asymmetric unit. The structure has the Man-␣-1,2-Man-␣-1,3-Man moiety in molecules A and B, whereas the GlcNAc moiety is disordered. In contrast, only one mannose residue is ordered in molecule C, and all the carbohydrate residues are disordered in molecules D and E. The dimer interfaces of the P2 1 2 1 2 1 crystal form are different from that of the C2 crystal form, suggesting that VIP36 is monomeric.
Upon Ca 2ϩ binding, significant conformational changes occur around Loops 1 and 2 (Fig. 3). Large movements of the Ca 2ϩ -coordinating and neighboring atoms are observed for the O␦-2 of Asp 131 , O␦-1 of Asn 166 , and N␦-1 of His 190 . The distances are 1.7, 1.6, and 4.0 Å, respectively. The O␦-2 of Asp 131 and N␦-1 atom of His 190 form hydrogen bonds with Ca 2ϩ -coordinating water molecules W1 and W2, respectively, whereas the O␦-1 of Asn 166 is directly coordinated with Ca 2ϩ . As we will describe further below, these residues bind the carbohydrate moiety in the presence of Ca 2ϩ suggesting a mechanism for the Ca 2ϩ -dependent carbohydrate binding of VIP36.
Structure of VIP36 in Complex with Ca 2ϩ and Man-In the structure of VIP36 in complex with Ca 2ϩ and mannose, the mannose is located in a pocket neighboring the Ca 2ϩ -binding site on the concave ␤-sheet and has well defined electron density (Fig. 5A). The mannose-binding site (also called the "primary binding site" hereafter) comprises ␤7 and Loops 1, 2, and 3. A number of specific contacts can be seen between the protein and the ligand. The mannose is bound by Asp 131 (O␦-1 and O␦-2), Asn 166 (N␦-2), His 190 (N⑀-2), Gly 260 (N), Asp 261 (N), and Leu 262 (N) through hydrogen bonds in the complex (Fig. 5D). Incidentally the side-chain positions of Asp 131 , Asn 166 , and His 190 in the primary binding site are stabilized by Ca 2ϩ .
Structure of VIP36 in Complex with Ca 2ϩ and Man 2 -The Man-␣-1,2-Man residues have extremely well defined electron density in the structure of VIP36⅐Man 2 complex (Fig. 5B). The 4-OH and 6-OH groups of the non-reducing mannose residue make hydrogen bonds with Ser 96 (O␥) and Asp 261 (O␦-1),  respectively (Fig. 5E). In contrast, the hydrogen bond between the 6-OH group and Asp 261 (O␦-1) is not observed in molecule B (supplemental Fig. 4A). The binding site of the reducing-end mannose residue of Man 2 is almost the same as the primary binding site (Fig. 5, D and E).
Structure of VIP36 in Complex with Ca 2ϩ and Man 3 GlcNAc-In the structure of VIP36 with Man 3 GlcNAc, the Man 3 moiety is ordered, whereas the GlcNAc moiety is disordered. The Man-␣-1,2-Man-␣-1,3-Man residues have defined electron density (Fig. 5C). The binding site of the non-reducing mannose residue of Man 3 GlcNAc is almost the same as the primary binding site (Fig. 5, D and F). The 6-OH group of the ␣1-2and ␣1-3-linked mannose residue makes hydrogen bonds with Tyr 164 (O) and Asn 166 (O) (Fig. 5F). The ␣1-3and ␤1-4linked mannose residue is recognized by Asp 261 (O␦-1) through a hydrogen bond, whereas the hydrogen bond is not observed in molecule B (supplemental Fig. 4B).

DISCUSSION
Many lectins, such as mannose-binding proteins and the asialoglycoprotein receptors, achieve higher affinity and selectivity through oligomerization of their CRDs (43). For instance, ERGIC-53 and Emp46/47p are known to form oligomeric complexes through the putative coiled-coil region in the stalk domain (15,44,45). On the other hand, it has been shown that  no disulfide-linked or stable non-covalent oligomers of recombinant exoplasmic/luminal domain (residues 45-322) or endogenous VIP36 could be detected by cross-linking or sedimentation analysis (46). Furthermore we confirmed that the exoplasmic/luminal domain of VIP36 (residues 51-301 and 51-322) is monomeric in physiological solution by gel filtration analyses (supplemental Fig. 1). Indeed the stalk domain (residues 279 -322) of VIP36 is 95-162 residues shorter than those of ERGIC-53, Emp46p, and Emp47p. The portion of the stalk domain of VIP36 included in our construct (residues 279 -301) does not form coiled-coil structure ( Fig. 2A and supplemental  Fig. 2). The short stalk domain and the absence of coiled-coil domain suggest that VIP36 may likely function as a monomer.
It is known that leguminous lectins coordinate Mn 2ϩ and Ca 2ϩ ions, termed S1 and S2, respectively, in their ␤-sandwich structures (Fig. 6, A and B) (47). The S1 ion stabilizes the S2-binding site, and the S2 ion fixes the positions of the amino acids that interact with the oligosaccharide ligands. In this study, we showed that VIP36 has a single Ca 2ϩ in the CRD structures and that the Ca 2ϩ fixes the positions of Asp 131 , Asn 166 , and His 190 that interact with carbohydrate ligands in the primary binding site (Figs. 3 and 5). Specifically significant conformational changes upon Ca 2ϩ binding take place around the Ca 2ϩ and primary carbohydrate-binding site of VIP36 (Fig. 3). Similar but more pronounced structural changes of the corresponding site upon metal binding were also observed in concanavalin A (ConA) (48) and ERGIC-53 (23). The Ca 2ϩ of ConA induces large conformational changes to stabilize the correct geometry of the Ca 2ϩ -binding site that comprise the trans to cis isomerization of the Ala 207 -Asp 208 peptide bond accompanied by the formation of the carbohydratebinding site (48). In VIP36 structures, such isomerization changes were not observed. We also observed that Ca 2ϩ is required for interaction between VIP36 and Man 6 (GlcNAc) 2 -Asn by SPR experiments (supplemental Fig. 3). These results explain the Ca 2ϩ -dependent carbohydrate binding of VIP36.
It was shown that ERGIC-53 contains two Ca 2ϩ ions termed M1 and M2 (Fig. 6, C and D) and that the M1 ion does not lie at the same site as the S1 ion, although M2 is located at the corresponding S2 site (23). The Ca 2ϩ of VIP36 corresponds to the M2 site of ERGIC-53. When the VIP36 and ERGIC-53 structures are superimposed, the M2 metal ions overlay with a separation that is less than 0.2 Å. Although the Ca 2ϩ of VIP36 is equivalent to the M2 site of ERGIC-53, the electron density maps of VIP36 show no peak that could be assigned as a metal ion either at the corresponding M1 site or at any other sites within the structure. In the Ca 2ϩ -binding site (M2 and S2), the Ca 2ϩ -coordinating residues are structurally very well conserved except for Asp 193 in VIP36 (Fig. 6, B and D). The corresponding Asp 19 (ConA) and Asp 189 (ERGIC-53) residues are coordinated by the two metal ions. In VIP36, the O␦-1 of Asp 193 forms a hydrogen bond with the main-chain nitrogen atom of Asp 167 to stabilize the Ca 2ϩ -binding site. As a result, only one Ca 2ϩ ion fixes the ligand binding residues in VIP36. Our crystallographic studies also suggest that VIP36 did not bind other divalent cations, neither Mn 2ϩ nor Mg 2ϩ (data not shown). Loop 1 of VIP36 is two residues longer than that of ERGIC-53. Likewise the residues coordinating the M1 ion in ERGIC-53 (Asp 163 , Asp 165 ,  (Fig. 7). The carbohydrate-binding site is located in a negatively charged pocket (Fig.  2B). The extended carbohydrate-binding site comprises ␤5, ␤7, and Loops 1, 2, and 3. The mannose residue in the primary binding site corresponds to the Man(C) moiety at the middle of the D1 arm. In addition, Asp 261 is flexible in mannose recognition and can flip between Man(3) and Man(D1) depending on the local environment (supplemental Fig. 4). The crystallo-graphic results correlate well with the SPR experiments (Fig. 4) and the previous results (19,20) that VIP36 recognizes Man-␣-1,2-Man-␣-1,2-Man residues of the D1 arm of high mannose type glycans.
Next the carbohydrate-binding site of VIP36 was compared with that of ConA in complex with Man-␣-1,3-(Man-␣-1,6-)Man (corresponding to Man(3)-(Man(4Ј))-Man(4)) (49). Although the carbohydrate binding specificity of VIP36 and ConA is essentially different, the structural conservation of ligand-binding sites between them is observed in not only the Man(C)-binding site but also the Man(D1)-and Man(4)-binding sites of VIP36 (Fig. 7C). In the Man(C)binding site of VIP36, the ligand binding residues are structurally very well conserved except for His 190 in VIP36 (Fig. 7C). The corresponding mainchain nitrogen atom of Arg 228 in ConA is bound to only the 3-OH group of the mannose residue. In contrast, N⑀-2 of His 190 in VIP36 is bound to the 3-OH and 4-OH groups and acts simultaneously as hydrogen bond donor and acceptor (Fig. 7B). In the Man(4)-binding site, although the carbohydrate binding loop conformation of VIP36 is largely different than that of ConA, the side-chain position of Tyr 164 is very similar to that of Tyr 12 in ConA.
When the carbohydrate-binding site of VIP36 was compared with the corresponding site of ERGIC-53 in complex with Ca 2ϩ (23), the Man 4 binding residues of VIP36 are largely identical to the corresponding residues of ERGIC-53 ( Fig. 7D and supplemental Fig. 2). The structural conservation suggests that ERGIC-53 also binds the D1 arm moiety of high mannose type glycoproteins, which is consistent with the function of ERGIC-53 as a transport lectin for high mannose type glycoproteins. These contain Man 8 (GlcNAc) 2 with an intact D1 arm and are transported from the ER by the ER quality control mechanism (2, 3). However, there are some structural differences in the ␤5, ␤7, and Loops 1 and 3 regions: (i) in ␤5 the side-chain orientation of Ser 96 of ERGIC-53 is dissimilar to that of VIP36, (ii) the side-chain orientation of Asp 129 in ␤7 of ERGIC-53 is different than that of the corresponding Asp 131 of VIP36, (iii) Phe 162 of ERGIC-53 is replaced by Tyr 164 in VIP36, and (iv) the Loop 3 of ERGIC-53 is positioned further away from the ligand when compared with VIP36, and Asp 261 of VIP36 is replaced by Gly 260 in ERGIC-53. Most significantly, Tyr 164 and Asp 261 in VIP36 are better suited than Phe 162 and Gly 260 in ERGIC-53 for binding to the Man 4 . In ERGIC-53, the presumptive weak interactions between ERGIC-53 and carbohydrate ligands might be compensated by the oligomerization of the CRDs.
To demonstrate that the crystal structure strictly represents the complex formed in solution, we simulated a complex model of Man 8 (GlcNAc) 2 -Asn and VIP36 (Fig. 8A). In this model, there are no significant steric clashes between high mannose type N-glycan and VIP36. The monomeric VIP36 seems to accommodate the glycan along an extended ligand-binding site. Kamiya et al. (20) have suggested that VIP36 recognizes the D1 arm and showed that mannose trimming and monoglucosylation of the D1 arm resulted in significant reduction in affinity for VIP36 CRD using frontal affinity chromatography analysis. When a glucose residue is modeled into the VIP36⅐Man 4 structure at the Man(D1) position through ␣1-3 linkage, a steric hindrance occurs between the glucose and Glu 98 of VIP36 (data not shown). In addition, we have shown that VIP36 recognizes the D1 arm, Man-␣-1,2-Man-␣-1,2-Man, using SPR analysis (Fig. 4) and that Asp 131 of VIP36 plays an essential role in binding 35 S-labeled secretory glycoproteins (13). Kawasaki et al. (21) have also shown that Asp 131 of VIP36 was involved in ligand binding using a flow cytometry-based method. On the other hand, it was shown that Asp 121 , Asn 156 , and His 178 of human ERGIC-53, which correspond to Asp 131 , Asn 166 , and His 190 of VIP36, are involved with binding to mannose and its cargo glycoprotein, cathepsin Z-related protein (15,18,50). From these observations, we conclude that the interaction between VIP36 and high mannose type glycans in solution is also achieved through interactions between these amino acid residues and the D1 arm.
It has been shown that VIP36 recycles between the ER and the Golgi complex (9 -11). To date, however, there is no obvious evidence that VIP36 is involved in retrograde transport of glycoproteins from the Golgi complex to the ER. On the other hand, we revealed that VIP36 localizes in the trans-Golgi net-work (12) and is involved in secretion of high mannose type glycoproteins clusterin and ␣-amylase (13,14). It has been generally known that the D1 arm is trimmed by cis-Golgi mannosidase I to form Man 5 (GlcNAc) 2 in the cis-Golgi. The carbohydrate structure has a lower affinity for VIP36 (19,20). In this study, we have shown that VIP36 specifically binds the Man-␣-1,2-Man-␣-1,2-Man residues of the D1 arm of high mannose type glycan. Taken together, VIP36 might be involved in anterograde transport of certain glycoproteins carrying high mannose type glycan with the D1 arm from the ERGIC via the Golgi complex to the plasma membrane by protecting the D1 arm against trimming by cis-Golgi mannosidase I. Although it is not known whether or not high mannose type glycan of rat salivary ␣-amylase has the D1 arm, a possible model for binding between VIP36 and salivary ␣-amylase carrying high mannose type glycan (Man 8 (GlcNAc) 2 ) in rat parotid acinar cells is shown (Fig. 8B).
In summary, we determined the first complex structure of the exoplasmic/luminal domain of the transport lectin VIP36 and Ca 2ϩ and Man, Man 2 , and Man 3 GlcNAc, which are part of the D1 arm of high mannose type glycans. Our results provide structural insights into the mechanism of recognition of high mannose type glycoproteins by VIP36 in a Ca 2ϩ -dependent and D1 arm-specific manner. Further biochemical analysis such as subcellular localization of VIP36 on a wide variety of cells and identification of its cargo glycoproteins together with the detailed carbohydrate structures will provide further insight into the mechanism of high mannose type glycoprotein transport by VIP36.