Structural basis of trimannoside recognition by concanavalin A.

Despite the fact that complex saccharides play an important role in many biological recognition processes, molecular level descriptions of protein-carbohydrate interactions are sparse. The legume lectin concanavalin A (con A), from Canavalia ensiformis, specifically recognizes the trimannoside core of many complex glycans. We have determined the crystal structure of a con A-trimannoside complex at 2.3-Å resolution and now describe the trimannoside interaction with con A. All three sugar residues are in well defined difference electron density. The 1,6-linked mannose residue is bound at the previously reported monosaccharide binding site; the other two sugars bind in an extended cleft formed by residues Tyr-12, Pro-13, Asn-14, Thr-15, and Asp-16. Hydrogen bonds are formed between the protein and all three sugar residues. In particular, the 1,3-linked mannose residue makes a strong hydrogen bond with the main chain of the protein. In addition, a water molecule, which is conserved in other con A structures, plays an important role in anchoring the reducing sugar unit to the protein. The complex is further stabilized by van der Waals interactions. The structure provides a rationale for the high affinity of con A for N-linked glycans.

It is well established that carbohydrates play a role in a myriad of important biological recognition processes; infection, the immune response, cell differentiation, and neuronal development may all be regulated to some extent by protein-carbohydrate interactions (1)(2)(3)(4). One area of therapeutic interest in carbohydrate recognition has relied on the their role as cell surface receptors enabling adherence of bacteria, parasites, and viruses in the early stages of infection (5,6). The abnormal structure and levels of certain tumor cell surface glycans may also present opportunities for therapeutic intervention (7). The notion of using oligosaccharide analogues to disrupt cell-cell recognition is an appealing one, and is the focus of considerable current activity in relation to the development of anti-inflammatory agents (8). However, the ubiquitous use of carbohydrates in nature potentially poses serious specificity problems. Understanding the molecular basis of carbohydrate recognition might provide the necessary basis on which to rationally design biologically active saccharide analogues.
Although highly homologous, and often sharing monosaccharide selectivity, plant lectins exhibit exquisite oligosaccharide specificity (9). While the function of these proteins is unknown, their interaction with saccharides has proved a valuable source of fundamental information. Structures of protein-saccharide complexes have been reported for lectins from Erythina coroallodendron (EcorL) (10), Griffonia simplicifolia (GS4) (11), Lathyrus ochrus (LOL1) (12,(42)(43)(44), pea (13), and lentil (preliminary data only) (14). The GS4 and EcorL lectins are galactose-specific, while LOL1, pea, and con A are mannosespecific. Interestingly, the overall organization of the monosaccharide binding site is conserved among the lectins (9). In GS4 and EcorL, the galactose residue is rotated relative to mannose in the binding site with only subtle changes in the precise side chain organization observed. Oligosaccharide complexes of EcorL, GS4, and LOL1 have provided structural insights into carbohydrate recognition, and, in particular, these structures have provided experimental evidence for the importance of water molecules in mediating carbohydrate recognition (12).
Concanavalin A (con A) 1 is the most extensively studied member of the lectin family, and was first isolated and crystallized in 1919 (15). Although the structure of the protein was determined in the early 1970s (16), it was not until 1989 that the 2.9-Å structure of con A-methyl ␣-D-mannopyranoside complex was determined (17). This represented the first structure of any lectin carbohydrate complex, and it explained the socalled "Goldstein rules" for con A monosaccharide specificity (18). The sugar was reported to be anchored to the protein by several direct hydrogen bonds and by van der Waals interactions. Subsequent extension of the resolution to 2.0 Å (19) permitted a more detailed description of the contacts between the protein and the monosaccharide. However, the same depth of understanding is not available to explain the oligosaccharide specificity of lectins. The precise contributions of hydrogen bonding, van der Waals interactions, and rearrangement of bound and bulk water to the specificity of the lectin-oligosaccharide interactions continues to be a subject of interest (20).
The oligosaccharide specificity of con A is well documented (9). Interactions are centered on the so-called trimannoside core ( Fig. 1) found in all N-linked glycans, and it is this specific interaction that forms the basis of con A's use as a tool in histochemical staining (21). Although there are several reports of modeling studies on the con A-oligosaccharide interaction (22,45,46), no crystal structure has been reported for con A complexed to any oligosaccharide. We now report the 2.3-Å resolution structure of con A bound to the N-linked glycan core trimannoside Man␣1-6(Man␣1-3)Man.

MATERIALS AND METHODS
Con A and Man␣1-6(Man␣1-3)Man were purchased from Sigma (Poole, United Kingdom) and Dextra Laboratories (Reading, United Kingdom), respectively. Crystals (dimensions: 0.3 mm ϫ 0.4 mm ϫ 1.2 mm) of the proteincarbohydrate complex were obtained, after 2 weeks, from a hanging drop of 8 mg/ml protein and 7 mM trimannoside equilibrated against 20% polyethylene glycol (M r 6000), pH 9.0. All diffraction data were * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The atomic coordinates and structure factors (codes 5CNA, 1 CVN, and R1CVNSF) have been deposited in the Protein Data Bank, Brookhaven National Laboratory, Upton, NY.
collected at room temperature on a crystal mounted in a glass capillary using the Enraf-Nonius/MacScience DIP2000 dual image plate. X-rays were generated using an Enraf-Nonius FR591 rotating anode generator and focused using the MacScience mirror system. The fresh crystal diffracted to 2.1 Å; however, crystal decay limited the effective resolution to 2.3 Å. Data were recorded as 252 non-overlapping 12.5-min 0.5°o scillations and processed using DENZO and SCALEPACK (23). The crystal had a primitive unit cell of dimensions a ϭ 81.65 Å, b ϭ 66.68, c ϭ 108.32, ␣ ϭ ␥ ϭ 90.0°, ␤ ϭ 97.79°. The asymmetric unit contains a tetramer; Matthew's number 2.9 Da Å Ϫ3 , approximately 52% solvent. A k 2n systematic absence was visible, and the space group was assigned as P2 1 . A summary of the data is given in Table I. The structure was determined using the molecular replacement procedure AMORE (24) as implemented in the CCP4 package (25). The 2-Å structure of the methyl ␣-D-mannopyranoside-con A complex (Protein Data Bank code 5CNA) was used as the search model (with metal ions, sugars, and waters removed). A random subset of data (10%) was omitted from all refinement calculations in order to provide an unbiased assessment of the refinement (26). The raw molecular replacement solution in P2 1 was rigid body refined to 2.7 Å and gave a free R-factor of 34.0% and an R-factor of 34.6%. An F o Ϫ F c electron density map was generated with phases calculated from the rigid refined model, strong density (Ͼ5) was observed for all 8 metal ions and for 8 of the 12 possible sugar units, with weak but convincing density for the remaining 4. This was taken as confirmation of the correctness of the space group assignment and the molecular replacement solution. Refinement proceeded with X-PLOR (27) (restrained positional and thermal factor) alternating with manual intervention using O (28). Non-crystallographic positional and thermal factor restraints were maintained throughout the refinement. The metal ions were included in the refinement with zero electrostatic charge, and the trimannoside molecules were included when a difference F o Ϫ F c map showed 3 density for all 12 mannose residues (Fig.  2). Water molecules were added to the model, if (a) they corresponded to peaks with magnitudes greater than 3.5 in the F o Ϫ F c map, (b) they made physically reasonable hydrogen bonds with the protein (or other ordered water molecules), (c) they subsequently reappeared in peaks of 1.0 in the 2F o Ϫ F c map, and (d) a drop in the free R-factor was observed. The final model comprises 948 residues and has an R-factor of 20.5% and a free R-factor of 25.5% for data between 8 and 2.3 Å with F Ͼ 2.5F, for all data in the range from 8 to 2.3 Å, the R-factor is 21.7%. Table II gives a summary of refinement results. A Ramachandran plot (not shown) revealed 1 residue in a generously allowed region and no residues in disallowed regions, 86.5% of the residues were in the most favored regions and the remaining 13.5% in additionally allowed regions. All stereochemical parameters measured by PROCHECK version 3.3 (29) were better than average for a structure at 2.3 Å (data not shown). Co-ordinates (code 1CVN) and structure factors (code R1CVNSF) have been deposited with Protein Data Bank (30). Coordinates will be available 6 months after the date of this publication, and structure factors after 1 year.

RESULTS AND DISCUSSION
Overall Structure of the Protein in Con A-Trimannoside Complex-In the crystal of the con A-trimannoside complex, con A is a tetramer with each monomer consisting of a sandwich of two ␤ sheets (Fig. 3). The overall fold of the protein is identical to the native structure (16). Residues 118 -123 are almost completely disordered in this structure; the loops at Ser-161 and Ser-204 are also in poor density. The remainder of the structure is well ordered, and apart from the loops mentioned above, the 2F o Ϫ F c electron density is unbroken for the backbone at a contour level of 1.0. Comparing this structure to the 1.75-Å native structure (31) shows an average root mean square deviation for all c␣ atoms of 0.47 Å; excluding flexible and disordered loops reduces this value by more than half. Comparing the trimannoside complex reported here to the 2.0-Å methyl ␣-mannoside-con A structure (19) shows an average root mean square deviation for all c␣ atoms of 0.28 Å; again, excluding flexible and disordered loops reduces this value.
Structure of Bound Trimannoside in Con A-Trimannoside Complex-The con A-trimannoside complex displays well defined difference electron density for all three sugar residues. In contrast, the previously reported trimannoside-pea lectin co- FIG. 1. Biantennary N-linked glycan. The trimannoside core is shown boxed.   complex (13) only one mannose residue was visible and the remaining two sugars units disordered (in effect directly analogous to structures of the con A and LOL1 monosaccharide complexes). The direct recognition of all three sugar units of trimannoside by con A is in marked contrast to the extensive water network employed by LOL1 (12) in binding Man␣1-3Man␤1-4GlcNAc. Perhaps due to the resolution of the current study, an extensive water network surrounding the sugar was not observed.
Conformation of Bound Trimannoside-NMR studies on Nlinked glycans indicate two principle conformations for the ␣-1,6-antenna of the trimannoside component, with the ␣-1,3antenna being conformationally invariant (except where the ␤-Man unit is substituted) (32,33). In solution, the ␣-1,6antenna exists predominantly in an extended "forward" form. This arrangement is also evident from the crystal structure of con A-bound trimannoside, and is one of the conformations predicted by Imberty and Perez (46) from molecular modeling studies on the binding of GlcNAc␤1-2Man␣1-6Man to con A. In effect, con A binds the trimannoside in its solution conformation without substantial conformational perturbation of the saccharide.
Since O4 of the reducing sugar unit is in contact with Tyr-12, substitution at this position is expected to abolish ligand binding in the mode observed for the unsubstituted trimannoside. Such a steric clash accounts for the reduced binding (34) of con A by N-linked glycans possessing a bisecting GlcNAc residue at O-4 of the trimannoside unit.
Specific Protein-Carbohydrate Contacts-All three sugar residues make contacts with the protein by way of hydrogen bonds and/or van der Waals interactions (Tables III and IV, Fig. 4).
1,6-Linked Mannose-The 1,6-linked mannose residue sits in the monosaccharide binding site of con A, and interactions between the protein and this sugar unit are essentially those observed in the con A-methyl ␣-mannoside complex (17).
Reducing Mannose-The binding site for the reducing man-nose residue is bounded by Tyr-12, which forms a hydrogen bond to mannose O-4. In addition, O-2 is hydrogen-bonded to a structurally conserved bridging water molecule, which is also present in both native con A and the methyl ␣-mannoside-con A complex. While the trimannoside used in this study contains a free reducing terminus, which can equilibrate between ␣and ␤-anomers, well defined density is observed for the ␣-configured hemiacetal. The anomeric hydroxyl group makes no contact with protein. A ␤-configuration, as present in the N-linked oligosaccharide ligands for con A (Fig. 1), would also not make any contact with protein.
1,3-Linked Mannose-The 2-and 6-hydroxyl groups of the 1,3-linked mannose residue are not in contact with either protein or bridging water. OH-3 forms a strong hydrogen bonded to the to the protein backbone (Thr-15) and is also hydrogenbonded to the side chain of Thr-15; O-4 is also hydrogen-bonded to the side chain of the same residue. Work reported by Mandal and co-workers (35), which employed a systematically modified series of monodeoxytrimannosides, demonstrated that the 2,4and 6-hydroxyl groups of the 1,3-linked mannose unit of trimannoside are not essential for inhibition of con A-mediated hemeagglutination of rabbit erythrocytes. In contrast the corresponding 3-deoxy sugar was 10-fold less effective. The apparent recognition recorded here of OH-4 of the 1,3-linked mannose unit of trimannoside by Thr-15 was therefore somewhat unexpected. It should be noted, however, that the observed Thr-15OG1-O-4 heteroatom distance of 3.1 Å is at the longer end of what one might consider a strong hydrogen bond.
His-205 and Trimannoside Recognition-Carver and coworkers (22) have suggested that His-205 is located near to the 1,3-arm of the bound trimannoside, and may contribute to its recognition. This model has been further commented on by Chervenak and Toone (36). Data reported here do not support a role for His-205 in trimannoside recognition, since it is located at least 6.5 Å from the bound ligand.
Overall Picture-Unlike most other mannose-specific lectins, con A specifically recognizes all three sugar units of the Nlinked glycan trimannoside core. Data reported here confirm a single high affinity site on con A for the 1,6-linked mannose unit of the trimannoside, together with an extended binding  site that forms specific direct and indirect (via structural water) contacts with both the reducing and 1,3-linked mannose units.
Structural Basis of the High Affinity for Trimannoside by Con A and Dioclea grandiflora Lectin, but Not Pea Lectin or LOL1-The extended con A binding site for trimannoside results in specific interactions with all three sugar units. As noted above, the 1,6-linked sugar is recognized by the single high affinity monosaccharide binding site of con A (17). Similar monosaccharide recognition sites are found on both pea lectin and LOL1 (42,13). However, neither of the latter two lectins bind trimannoside with high affinity, indicating that the specificity of con A for trimannoside resides in the extended binding site, which recognizes both the reducing unit and 1,3-linked mannose residue.
The reducing mannose residue makes good hydrogen bonds to both Tyr-12 and a bridging water molecule, which is in turn ligated by Asn-14, Asp-16, and Arg-228. In both pea lectin and LOL1 (42) the con A Tyr-12 is replaced by Phe, which is incapable of making the hydrogen bond to the reducing mannose O-4 observed in the con A-trimannoside complex. In addition, two of the three residues that act as ligands to the structural water are replaced by non-polar residues: Asp-16 by Ala (37) and Arg-228 by Gly (42). Although a water is found in the same position in LOL1 (but not in pea lectin), it is only ligated by one Asn residue. The two key recognition elements for the reducing unit of trimannoside are therefore substantially different in pea lectin and LOL1. The 1,3-linked mannose residue forms a strong hydrogen bond to the main chain of the protein, with additional hydrogen bonds to O-3 and O-4 from Thr-15. This  4. The extended trimannoside binding site of con A. Panel A, a close-up view of the saccharide in the binding site. The amino acids are shown as space-filling spheres with side chains colored by residue type; the trisaccharide is shown in stick format. (Figure was generated by RASTER3D (41,47).). Panel B, a schematic representation of the hydrogen bonds between the sugar and protein.
The distances for these hydrogen bonds are given in Table III. residue is conserved in LOL1, but is replaced by Ala in pea lectin (37). We have already mentioned that the contribution of Thr-15 to trimannoside recognition warrants further investigation.
We note that the amino acids we have identified as being important for recognition of trimannoside by con A (Tables III  and IV, Fig. 4B) are conserved in the Dioclea grandiflora lectin (38), which also has a high affinity for trimannoside (36).
Summary-The con A-trimannoside structure reveals a highly specific interaction based on hydrogen bonds to the main chain and/or side chain of the protein from all three sugar residues. A structurally conserved water molecule appears to play a crucial role in recognition.
A comparison between the native and mannoside complexes of con A at 2.0-Å resolution shows that trimannoside binds to what appears to be a relatively rigid and preformed extended binding site from which water molecules are displaced. Specific direct and indirect interactions with amino acids, which are not conserved in LOL1 and pea lectin, account for the high affinity and specificity of con A for the trimannosyl core of N-linked glycans.