Characterization of the sialic acid-binding site in sialoadhesin by site-directed mutagenesis.

The sialoadhesins are a distinct subgroup of the immunoglobulin superfamily, comprising sialoadhesin, CD22, the myelin-associated glycoprotein, and CD33. They can all mediate sialic acid-dependent binding to cells with distinct specificities. Sialoadhesin is a murine macrophage-restricted cell-surface molecule with 17 extracellular immunoglobulin-like domains that recognizes NeuAc alpha 2-3Gal in N- and O-glycans and interacts preferentially with cells of the granulocytic lineage. Its sialic acid-binding site is located within the NH2-terminal (membrane-distal) V-set domain. Here we have carried out site-directed mutagenesis in an attempt to identify the binding site of sialoadhesin. A subset of nonconservative mutations disrupted sialic acid-dependent binding without affecting binding of three monoclonal antibodies directed to two distinct epitopes of sialoadhesin. A CD8 alpha-based molecular model predicts that these residues form a contiguous binding site on the GFCC'C" beta-sheet of the V-set domain centered around an arginine in the F strand. A conservative mutation of this arginine to lysine also abolished binding. This amino acid is conserved among all members of the sialoadhesin family and is therefore likely to be a key residue in mediating sialic acid-dependent binding of sialoadhesins to cells.

The sialoadhesins are a distinct subgroup of the immunoglobulin superfamily, comprising sialoadhesin, CD22, the myelin-associated glycoprotein, and CD33. They can all mediate sialic acid-dependent binding to cells with distinct specificities. Sialoadhesin is a murine macrophage-restricted cell-surface molecule with 17 extracellular immunoglobulin-like domains that recognizes NeuAc␣2-3Gal in N-and O-glycans and interacts preferentially with cells of the granulocytic lineage. Its sialic acid-binding site is located within the NH 2 -terminal (membrane-distal) V-set domain. Here we have carried out site-directed mutagenesis in an attempt to identify the binding site of sialoadhesin. A subset of nonconservative mutations disrupted sialic acid-dependent binding without affecting binding of three monoclonal antibodies directed to two distinct epitopes of sialoadhesin. A CD8␣-based molecular model predicts that these residues form a contiguous binding site on the GFCCC؆ ␤-sheet of the V-set domain centered around an arginine in the F strand. A conservative mutation of this arginine to lysine also abolished binding. This amino acid is conserved among all members of the sialoadhesin family and is therefore likely to be a key residue in mediating sialic acid-dependent binding of sialoadhesins to cells.
Carbohydrate-binding proteins mediate diverse biological functions in animals (reviewed in Ref. 1). The mammalian lectins so far identified can be divided into groups according to their structural similarities. These include the C-type lectins, the galectins, the P-type lectins, and the recently characterized immunoglobulin-type or I-type lectins (reviewed in Ref. 2). The best characterized I-type lectins are the sialoadhesins, a homologous group of cell-surface glycoproteins that recognize distinct sialylated glycans (3,4). Members of the sialoadhesin family include the eponymous member, sialoadhesin, which is expressed only on macrophage subpopulations (5); CD22, a B-cell-restricted antigen (6,7); the myelin-associated glycoprotein expressed by oligodendrocytes and Schwann cells (8); and CD33, an antigen restricted to myelomonocytic cells (9).
Each member of the sialoadhesin family contains a single NH 2 -terminal (membrane-distal) V-set domain followed by differing numbers of C2-set domains, ranging from 1 in CD33 to 16 in sialoadhesin. The greatest sequence similarity between sialoadhesins is found in the NH 2 -terminal two to four domains (5). In contrast, their cytoplasmic regions show little homology, indicating distinct intracellular functions of these proteins. Sialoadhesin specifically recognizes NeuAc␣2-3Gal as a minimal oligosaccharide in O-and N-linked glycans and glycolipids (10,11). In bone marrow, sialoadhesin is selectively concentrated at regions of membrane contact between macrophages and developing granulocytes (12), and the isolated molecule binds preferentially to granulocytes (13). These observations indicate that sialoadhesin is involved in the specific recognition of neutrophils by macrophages during hemopoiesis.
Previous work using a series of domain deletion constructs has shown that the sialic acid-binding site on sialoadhesin is located within the V-set domain (14). V-set domains consist of nine ␤-strands, designated A-G, making up two ␤-sheets, the GFCCЈCЉ sheet, and the ABED sheet (reviewed in Ref. 15). All members of the sialoadhesin family have unusual structural characteristics within their V-set domains that are unique within the Ig superfamily. Instead of the canonical intersheet disulfide bridge between the B and F ␤-strands, they possess an unusual intrasheet disulfide between the B and E strands. Furthermore, an interdomain disulfide bond is thought to bridge the first two NH 2 -terminal domains (16,17). 1 In this study, we have undertaken a site-directed mutagenesis screen of the V-set domain of sialoadhesin in an attempt to locate and characterize the sialic acid-binding site. Drastic changes to a subset of amino acids predicted to be on the surface of the V-set domain resulted in complete abrogation of sialic acid-dependent binding without affecting binding to antisialoadhesin monoclonal antibodies (mAbs). 2 When superimposed onto a CD8␣-based model of the V-set domain of sialoadhesin, these residues are seen to constitute a contiguous binding site clustered around an arginine on the F strand. This residue is conserved in all members of the sialoadhesin family and appears to be essential for sialic acid-dependent binding.

EXPERIMENTAL PROCEDURES
Materials-Unless specified otherwise, all reagents and chemicals were purchased from Sigma (Poole, United Kingdom or St. Louis, MO).
Protein A-Sepharose was purchased from Pharmacia (St. Albans, UK). Vibrio cholerae sialidase was purchased from Calbiochem, and Immulon 3 microtiter plates were from Dynatech Laboratories Inc. (Chantilly, VA). COS-1 cells were provided by the Imperial Cancer Research Fund Cell Bank (Clare Hall, UK).

Construction and Analysis of Mutant Forms of Sn(d1-3)Fc
Chimera-Sn(d1-3)Fc chimera was obtained by polymerase chain reaction amplification of DNA encoding domains 1-3 and subcloning into the Fc expression vector pIG1 as described (3). The entire Sn(d1-3)Fc fragment was excised using HindIII and NotI and subcloned by bluntend ligation into the XbaI restriction site of the phagemid expression vector pEF-BOS (18). Mutagenesis was performed directly on Sn(d1-3)Fc/pEF-BOS by the method of Kunkel (19) using the Muta-Gene phagemid in vitro mutagenesis kit (Version 2, Bio-Rad) according to the manufacturer's instructions. In all cases, the presence of the desired mutation was confirmed by dideoxy sequencing. The V-set domain of mutants that showed altered binding properties was sequenced in its entirety. The nomenclature used for mutant sialoadhesin proteins reflects the residue number for which the mutant amino acid was substituted (e.g. for R97D, arginine 97 was replaced by aspartic acid).
COS Cell Transfection and Mutant Protein Purification-Wild-type and mutant forms of Sn(d1-3)Fc protein and NCAM-Fc (used to control for nonspecific binding in solid-phase and enzyme-linked immunosorbent assays (ELISAs)) were produced by transient transfection of COS-1 cells using the DEAE-dextran method (20). The COS cell supernatants were harvested 7-10 days post-transfection, and the Fc chimeras were purified from tissue culture supernatants by protein A-Sepharose chromatography as described (20). Proteins were dialyzed against 20 mM Tris, pH 8.0; concentrated to ϳ1 mg/ml; and filter-sterilized. Protein concentrations were determined using the BCA protein assay kit (Pierce) with bovine serum albumin as a reference standard.
Monoclonal Antibodies and ELISAs-To determine whether mutant proteins were correctly folded, ELISAs were carried out. The antisialoadhesin mAbs 1C2 and 3D6 recognize the same epitope on domain 1, whereas SER-4 binds a distinct epitope on domains 2 and/or 3 (13,14). Fc chimeras at varying concentrations were adsorbed to wells of microtiter plates, and nonspecific binding sites were blocked as described (3). Wells were incubated with 15 g/ml purified 3D6 or SER-4 IgG for 1 h at 4°C. After washing, wells were incubated for 1 h at room temperature with peroxidase-conjugated goat anti-rat IgG (Sigma A5795; 1:50,000 dilution). The bound peroxidase activity was determined spectrophotometrically as described (3). Specific binding was determined by subtracting the values obtained with NCAM-Fc from values obtained with wild-type and mutant proteins.
Solid-phase Binding Assays-Solid-phase binding assays of Fc chimeras and human erythrocytes were performed as described (3,14). Human erythrocytes were either untreated (native) or derivatized to contain sialic acid in different linkages, NeuAc␣2-3Gal␤1-3GalNAc, NeuAc␣2-3Gal␤1-3(4)GlcNAc, or NeuAc␣2-6Gal␤1-4GlcNAc, as described (3). Briefly, Fc chimeras were bound at varying dilutions to wells of 96-well ELISA plates that had previously been coated with goat anti-human IgG. Erythrocyte binding to immobilized Fc chimeras was measured using the peroxidase activity of hemoglobin (3). Specific binding was determined by subtracting the values obtained with NCAM-Fc from the values obtained with wild-type and mutant proteins.
Molecular Modeling-By identifying characteristic immunoglobulin superfamily patterns in sialoadhesin, the sequence was aligned with that of CD8␣ and substituted into the CD8␣ crystal structure using the program MUTATE. 3 Major insertions and/or deletions to the CD8␣ framework were required in the B-C and C-CЈ loops. These loop regions were modeled to be stereochemically reasonable and positioned so as to minimize any steric clashes using the program FRODO (21) on an Evans & Sutherland ESV-10 graphics workstation; the CD8␣ mainchain conformation was retained for the rest of the model. Since the purpose of the model was to indicate the global positioning of mutations, no energy minimization was applied to attempt to fine tune the detailed local conformation.

RESULTS
Mutagenesis Strategy-Previous work has shown that the V-set domain of sialoadhesin is both necessary and sufficient for sialic acid-dependent binding of the correct specificity (14). We therefore targeted single amino acid substitutions to this domain and expressed mutants as Fc chimeras containing the first three domains of sialoadhesin fused to the Fc portion of human IgG1 (Sn(d1-3)Fc). To aid selection of amino acids for mutagenesis, the V-set domains of sialoadhesin family members were aligned with the V-set domain of CD8␣, whose structure is known (22). The alignment was particularly good on the B, C, E, and F strands (Fig. 1), where surface residues of sialoadhesin could be predicted with reasonable confidence.
The approach we used to localize the binding site was to make drastic changes to residues predicted to lie on the surface of the V-set domain and to examine the effects on sialic aciddependent binding. We made drastic changes because it has been shown that more conventional changes to alanine may only identify a fraction (25-40%) of the residues in the structural binding site (23,24). Those residues implicated by drastic mutations were subsequently mutated to alanine to determine whether they are important for sialic acid recognition.
The drastic mutations fell into two classes regarding their effects on sialic acid binding. Eleven mutants bound human erythrocytes to the same level as the wild-type Sn(d1-3)Fc protein and maintained specificity for NeuAc␣2-3Gal␤1-3GalNAc/NeuAc␣2-3Gal␤1-3(4)GlcNAc over NeuAc␣2-6Gal␤1-4GlcNAc (Table I). Figs. 2A and 3A show the results of erythrocyte binding to two mutants of this class, D29K and T93D. Six mutants were unable to mediate sialic acid-dependent binding, but bound both mAbs 3D6 and SER-4 (Table I). Fig. 4 shows binding data for R97D as an example of this class of mutant.
Interestingly, the mutants K8E, T93D, and K110E showed greatly reduced binding to mAb 3D6, with little or no effect on binding to mAb SER-4 or to erythrocytes (Table I). These observations suggest that Lys-8, Thr-93, and Lys-110 form part of the epitope recognized by mAb 3D6. Results obtained using T93D, which is typical of these mutants, are shown in Fig. 3.
Four of the six residues implicated in sialic acid recognition by drastic changes were studied further by mutating them to alanine. All alanine mutants bound to both mAbs (Table I), and the majority had no effect on erythrocyte binding. R97A was the only alanine mutation that abolished erythrocyte binding. To investigate the effect of a much more conservative substitution, Arg-97 was also changed to lysine. Similar to the R97D and R97A mutants, the R97K mutant showed greatly reduced binding to erythrocytes, but normal binding to both mAbs (Table I). Arg-97 therefore appears to be a key residue in mediating sialic acid-dependent binding of sialoadhesin.
Molecular Modeling of the Sialic Acid-binding Site on Sialoadhesin- Fig. 5 shows a molecular model of the sialoadhesin V-set domain based on the known structure of the V-set domain of CD8␣ (22). The six residues found to be involved in sialic acid-dependent binding (Ile-39, Tyr-41, Asn-95, Arg-97, Arg-105, and Asp-108) are predicted to lie close together, forming a well defined contiguous cluster of residues on the GFCCЈCЉ  (9), and human CD8␣ (44). Numbering of amino acids corresponds to the mature protein sequence of sialoadhesin (5). The ␤-strand assignments (solid bars) were based on the structure of CD8␣. Broken lines instead of bars are shown where there are no grounds for making precise assignments to ␤-strands. It should be noted that for the assignment of residues that are identical between members of the sialoadhesin family but are not characteristic of other V-set domains within the Ig superfamily, the other species homologues are taken into account, namely rat and human myelin-associated glycoproteins (45,46), human CD22 (7), and mouse CD33 (47). Thus, Trp-2 in sialoadhesin is replaced by Gln in mouse CD33, and Gly-66 in sialoadhesin is replaced by Lys in human CD22.
Binding of mutants to mAbs was assessed by ELISA in which mutants expressed as Sn(d1-3)Fc proteins were immobilized on plastic wells precoated with anti-human IgG, followed by incubation with the mAbs at saturation (15 g/ml) and peroxidase-conjugated anti-rat IgG. Binding data are expressed relative to the wild-type protein as follows: ϩϩ, Ͼ65%; ϩ, 35-65%; Ϫ, Ͻ35%. Each binding assay was repeated at least three times with similar results.
b Binding of mutants to erythrocytes was assessed in solid-phase binding assays in which mutants were immobilized to plastic wells as described above, and native human erythrocytes or erythrocytes derivatized with NeuAc␣2-3Gal␤1-3GalNAc, NeuAc␣2-3Gal␤1-3(4)GlcNAc, or NeuAc␣2-6Gal␤1-4GlcNAc were allowed to adhere for 30 min, followed by washing off the unbound cells. Bound cells were quantified using the peroxidase activity of hemoglobin. For native erythrocytes, the data show binding of mutants relative to the wild-type protein, using the same scheme as described above, at Fc chimera coating concentrations of 0.2 g/ml.
␤-sheet centered on Arg-97 in the middle of the F strand. The binding site is surrounded by residues that when mutated have no effect on binding. These include residues in the lower portion of the C (Asp-43), F (Thr-93), and G (Lys-110) ␤-strands; a mutation in the upper portion of the F ␤-strand (Glu-99); and mutations in the CЈ and CЉ ␤-strands lateral to the binding site (Fig. 5). Finally, those residues that were identified as being part of the 3D6 epitope (Lys-8, Thr-93, and Lys-110) are well separated in the primary sequence (in the A, F, and G strands), but the model predicts that they lie close together, COOHterminal to the sialic acid-binding site, where they are likely to form a conformationally sensitive epitope. DISCUSSION We have used site-directed mutagenesis in an attempt to locate the binding site for sialic acid within the V-set domain of sialoadhesin. Drastic mutations of six residues abolished erythrocyte binding with little or no effect on protein folding, as assessed by mAb binding. A model of this domain predicts that these residues form a discrete cluster on the G, F, and C strands, surrounded by residues that when mutated have no effect on erythrocyte binding. Of the alanine mutants, only R97A led to complete abolition of binding. A more conservative mutation of Arg-97 to lysine also disrupted sialic acid-dependent binding without affecting mAb binding. Arg-97 is conserved in all sialoadhesins, and mutation of the equivalent residue to alanine or lysine in CD22 (49) and to alanine in CD33 4 abolishes the binding activity of these proteins, suggesting that this residue is critical in sialic acid recognition by sialoadhesins.
The molecular model of sialoadhesin based on the CD8␣ structure (Fig. 5) is used primarily to depict the approximate positions of the mutants, and some aspects are likely to be inaccurate, particularly in the loop regions and in the CЈ and CЉ ␤-strands. However, the B, C, E, and F ␤-strands of Ig superfamily domains are structurally highly conserved. These strands also have characteristic sequence patterns that allow accurate alignments of superfamily proteins of unknown structure with those of known structure. For example, the central portions of the C and F strands have the patterns X⌽XCX and DXGXYX, respectively. The X residues will invariably be surface residues and will always be in well defined positions within the ␤-sheets. Thus, the positions of the F and C strand mutants are likely to be accurate in the model. The other mutations affecting binding were predicted to lie on the G strand, but their precise positions cannot be determined with 4 S. D. Freeman and P. R. Crocker, unpublished data.

FIG. 2. Binding assays of the sialoadhesin mutant D29K, typical of mutants that bind normally to erythrocytes and mAbs.
Wells were precoated with anti-human IgG and then incubated with varying concentrations of wild-type or mutant Sn(d1-3)Fc or with NCAM-Fc protein as a control. A, for erythrocyte binding assays, human erythrocytes were allowed to adhere for 30 min to the coated wells, and unbound cells were removed by washing. q, wild-type protein; E, D29K; ϫ, NCAM-Fc. B, for mAb binding assays, wells were incubated with anti-sialoadhesin mAbs 3D6 and SER-4, followed by peroxidaseconjugated goat anti-rat IgG. SER-4: qOOq, wild-type protein; EOOE, D29K; ϫOOϫ, NCAM-Fc. 3D6: q---q, wild-type protein; E---E, D29K; ϫ---ϫ, NCAM-Fc. Binding for both assays was quantified using o-phenylenediamine dihydrochloride as substrate, followed by optical density measurement of the reaction product at 450 nm. Results are expressed as mean values of four wells from single experiments. The standard deviations were consistently within 10% of the mean values and are not shown for clarity. Similar results were obtained in at least three independent experiments.
Although it could be argued that the effect of the mutations in disrupting sialic acid-dependent binding is due to changes in the overall folded structure of sialoadhesin, several arguments suggest that this is unlikely. First, all mutants that were unable to bind erythrocytes were able to bind three anti-sialoadhesin mAbs that recognize two distinct epitopes. It is known that the majority of mAbs bind discontinuous, conformationally sensitive epitopes (26), and our results demonstrate that this is the case for 1C2 and 3D6, mAbs directed to the V-set domain of sialoadhesin. Second, of 23 mutations made, only 8 disrupted erythrocyte binding, and these lie in a single contiguous region. Finally, mutations in the equivalent region of the V-set domain of CD22 also disrupt sialic acid-dependent binding of CD45 ligand without disrupting the binding of a conformationally sensitive mAb directed to this domain (49).
A combination of x-ray crystallography, molecular modeling, and mutagenesis has resulted in detailed descriptions of the binding sites of other sialic acid-binding proteins. Those studied so far include influenza virus hemagglutinin (27)(28)(29), viral and bacterial sialidases (30 -33), the polyoma virus VP1 protein (34), wheat germ agglutinin (35), and the mammalian cell adhesion molecule E-selectin (36 -38). Thus, sialic acid-binding sites can occur in a variety of distinct protein folds, including ␤-propellor (sialidases), "jelly roll" (polyoma VP1), and carbohydrate recognition domains typical of C-type Ca 2ϩ -dependent lectins (selectins). The present study, together with the accom-panying report on CD22 (49), is the first description of the putative sialic acid-binding site within adhesion molecules that belong to the Ig superfamily. Interestingly, the GFCCЈCЉ face that appears to be used by sialoadhesin and CD22 to interact with sialic acid is commonly used by other members of the Ig superfamily to bind protein ligands, as discussed in the accompanying paper (49).
For both influenza hemagglutinin and wheat germ agglutinin, interactions with sialic acid require diverse uncharged side chains, such as tryptophan and tyrosine (27,35). For several other sialic acid-binding proteins, positively charged amino acids (usually arginine) are also required. In a low resolution crystal structure of the polyoma virus VP1 protein, the guanidinium group of an arginine is thought to form a salt bridge with the carboxylate of sialic acid (34). Viral and bacterial sialidases have three conserved arginine residues thought to stabilize the carboxylate group of sialic acid (33). The E-selectin-sialic acid interaction is very sensitive to substitution of an arginine, and binding is abolished even when changed to lysine, indicating the critical role of this residue (36 -38). Similarly, for sialoadhesin and CD22, we demonstrated that substitution of a conserved arginine to alanine or lysine abolishes sialic acid-dependent binding.

FIG. 5. Ribbon diagram of the CD8␣-based model of the sialoadhesin V-set domain.
Residues that abolished sialic acid binding when mutated are shown in black. Those that had no effect are shown in white. Those that resulted in a reduction in binding to mAb 3D6 are shown in gray. Residues that occur in regions of the model where sequence alignment is uncertain are shown only as C-␣ spheres, whereas those that occur in regions of greater certainty are represented with side chains. This figure was prepared using MOLSCRIPT Version 1.4 (48) with modifications by R. Esnouf (see Footnote 3). Minimal modifications were made to the CD8␣ framework in the CЈ-D region. Sequence considerations (see Fig. 1) would imply greater structural differences between sialoadhesin and CD8␣ in this region, and therefore, the model only represents a general guide for amino acid positions.
shows a strict requirement for NeuAc␣2-6Gal (3, 4, 10, 39 -41). In the present study, none of the sialoadhesin mutants showed altered sialic acid linkage preference, so no insight was gained into how sialoadhesin discriminates between ␣2-3linked and ␣2-6-linked sialic acids. Structural studies with the polyoma virus VP1 protein have shown that the specificity for particular sialic acid linkage is determined by direct interactions of amino acid side chains with the sialic acid and galactose (34). A similar situation may occur with members of the sialoadhesin family since in recent studies with CD22, recognition of oligosaccharides terminating in NeuAc␣2-6Gal appeared to require an interaction of the CD22 binding site with both the sialic acid and galactose moieties (41). Further studies will be required to understand the molecular basis of sialic acid linkage discrimination between members of the sialoadhesin family.