The amino-terminal immunoglobulin-like domain of sialoadhesin contains the sialic acid binding site. Comparison with CD22.

Sialoadhesin and CD22 are members of a recently characterized family of sialic acid-dependent adhesion molecules belonging to the immunoglobulin superfamily. Sialoadhesin is a macrophage-restricted receptor containing 17 extracellular Ig-like domains which recognizes oligosaccharides terminating in NeuAcα2-3Gal in N- and O-linked glycans. CD22 is a B cell-restricted receptor with seven Ig-like domains which selectively recognizes oligosaccharides terminating in NeuAcα2-6Gal in N-glycans. Sequence similarity between these proteins is highest within their first four amino-terminal Ig-like domains. Here we identify the domain(s) containing the binding sites of both molecules by generating a series of extracellular domain deletion mutants fused to the Fc portion of human IgG1. Binding activity was analyzed by solid phase cell adhesion assays and also by surface plasmon resonance using purified glycophorin and CD45 as ligands for sialoadhesin and CD22, respectively. For sialoadhesin, the amino-terminal V-set Ig-like domain was both necessary and sufficient to mediate sialic acid-dependent adhesion of the correct specificity. In contrast, for murine CD22, only constructs containing both the V-set domain and the adjacent C2-set domain were able to mediate sialic acid-dependent binding. These results are consistent with the sialic acid binding site for both proteins residing in the membrane distal V-set domain, but for CD22 a direct contribution in binding from the neighboring C2-set domain cannot be excluded.

Sialoadhesin and CD22 are members of a recently characterized family of sialic acid-dependent adhesion molecules belonging to the immunoglobulin superfamily. Sialoadhesin is a macrophage-restricted receptor containing 17 extracellular Ig-like domains which recognizes oligosaccharides terminating in NeuAc␣2-3Gal in N-and O-linked glycans. CD22 is a B cell-restricted receptor with seven Ig-like domains which selectively recognizes oligosaccharides terminating in NeuAc␣2-6Gal in N-glycans. Sequence similarity between these proteins is highest within their first four amino-terminal Ig-like domains. Here we identify the domain(s) containing the binding sites of both molecules by generating a series of extracellular domain deletion mutants fused to the Fc portion of human IgG1. Binding activity was analyzed by solid phase cell adhesion assays and also by surface plasmon resonance using purified glycophorin and CD45 as ligands for sialoadhesin and CD22, respectively. For sialoadhesin, the amino-terminal V-set Ig-like domain was both necessary and sufficient to mediate sialic acid-dependent adhesion of the correct specificity. In contrast, for murine CD22, only constructs containing both the V-set domain and the adjacent C2set domain were able to mediate sialic acid-dependent binding. These results are consistent with the sialic acid binding site for both proteins residing in the membrane distal V-set domain, but for CD22 a direct contribution in binding from the neighboring C2-set domain cannot be excluded.
Most immunoglobulin (Ig) superfamily members mediate intercellular recognition through protein-protein interactions (for review see Ref. 1). In contrast, sialoadhesin and CD22 are members of a new subgroup of the Ig superfamily which mediate cellular interactions by recognizing specific sialylated glycans present in cell surface glycoconjugates (2)(3)(4)(5). Other members of this subgroup (recently termed the Sialoadhesin family) include the myelin-associated glycoprotein (MAG) 1 and CD33 (6,7).
Members of the Sialoadhesin family comprise a single amino-terminal V-set Ig domain (8) followed by varying numbers of C2-set domains. The V-set domain has an unusual arrangement of conserved cysteines that are predicted to form an intra-␤-sheet disulfide bond (9). In addition, all members of the Sialoadhesin family have conserved cysteines that are likely to create a disulfide bond between the V-and the adjacent C2-set domains (9,10). Sialoadhesin is the largest cell surface member of the Ig superfamily identified to date with 17 extracellular Ig-like domains (2). In comparison, the predominant form of CD22 contains seven Ig-like domains (11)(12)(13)(14)(15). The first four amino-terminal Ig-like domains of sialoadhesin and CD22 share Ϸ50% sequence similarity (2).
A striking feature of members of the Sialoadhesin family is their highly restricted expression pattern which suggests that these receptors are likely to mediate discrete biological functions. Sialoadhesin is expressed by macrophages in hemopoietic and secondary lymphoid tissues, MAG is present on myelinating oligodendrocytes and Schwann cells, CD33 is found on myeloid cells and CD22 is B cell-restricted. In vivo and in vitro studies have shown that sialoadhesin interacts selectively with cells of the granulocytic lineage, whereas CD22 interacts preferentially with T and B lymphocytes (6,16,17). In addition, CD22 has been implicated in modulating B cell signaling pathways via its association with the surface IgM-B cell receptor complex (18 -21).
Although sialoadhesin and CD22 are both sialic acid-binding proteins they have distinct preferences for sialylated glycoconjugates. Thus, sialoadhesin recognizes oligosaccharides terminating in NeuAc␣2-3Gal in N-glycans, O-glycans or glycolipids, whereas CD22 recognizes oligosaccharides terminating in NeuAc␣2-6Gal␤1-4GlcNAc, predominantly in N-glycans (3,4,6,(22)(23)(24). A key step toward understanding the molecular basis for this sialic acid-dependent adhesion is the identification of binding regions for members of the Sialoadhesin family. Previous studies using truncation mutants fused to the Fc portion of human IgG1 have shown that the first three Ig-like domains of sialoadhesin and CD22 are sufficient for mediating sialic aciddependent binding to a variety of cells (3,4,6,13,17,23,24). In this report, we have extended this analysis and shown that for sialoadhesin, the amino-terminal V-set Ig domain is both necessary and sufficient for sialic acid-dependent binding to cellular and molecular ligands. In contrast, murine CD22 appears to require both the V-set Ig domain and the adjacent C2-set domain as a "minimal unit" for mediating sialic acid-dependent adhesion. These results are consistent with the sialic acid binding site for both proteins residing in the membrane distal V-set domain, but for CD22 a direct contribution in binding from the neighboring C2-set domain cannot be excluded.
The NCAM-Fc consisted of the entire extracellular domain (five Ig-like domains and two type III fibronectin-related domains) of neural cell adhesion molecule fused to the Fc portion of human IgG1. This protein was used as a negative control in solid phase binding assays. cDNA encoding this construct was kindly provided by Dr. D. Simmons (Imperial Cancer Research Fund, University of Oxford).
The Sn(d1) construct was made by PCR amplification as described above, using a reverse primer (5Ј-3Ј), GTCTAATCCGTTGTCACAGT-GACCGTGGT, which resulted in the insertion of a stop codon at the end of domain 1, corresponding to Pro-120 of the mature polypeptide (2). The PCR product was sequenced and cloned into the mammalian expression vector pEF-BOS (34).
Production of Recombinant Proteins-COS-1 cells were transfected at 70 -80% confluence by the DEAE-dextran method using 20 g of plasmid DNA/15-cm diameter plate. 24 h post-transfection, the medium was changed to Dulbecco's modified Eagle's medium with 0.5% fetal calf serum that had been depleted of IgG with protein A-Sepharose. In the case of Sn(d1), medium was changed to Dulbecco's modified Eagle's medium ϩ 1% fetal calf serum. The COS cell supernatants were harvested 7-10 days post-transfection, and insoluble material was removed by centrifugation and filtration. The Fc chimeras were purified on protein A-Sepharose as described (32). The Sn(d1) protein was purified by affinity chromatography using the 3D6 mAb as described (2).
Preparation of Cells-Human erythrocytes were freshly prepared and stored in Alsever's solution. Mouse bone marrow cells were flushed from excised femora and dissociated into a single cell suspension by gentle pipetting in RPMI 1640 ϩ 20 mM HEPES followed by passage through a 25-gauge needle. Mouse mesenteric lymph node cells were obtained by gentle dissociation of the tissue in RPMI 1640 ϩ 20 mM HEPES. Sialidase treatment of bone marrow cells, mesenteric lymphocytes, and erythrocytes was carried out with 0.1 unit/ml V. cholerae sialidase in RPMI 1640 ϩ 20 mM HEPES for 3 h at 37°C followed by three washes in phosphate-buffered saline (PBS). Human erythrocytes were derivatized to contain sialic acid in different linkages as described previously (23).
Solid Phase Binding Assays-Fc chimeras at varying concentrations were adsorbed for 3 h at 37°C to wells of microtiter plates that had been coated overnight at 4°C with goat anti-human IgG at 15 g/ml in 0.1 M bicarbonate buffer, pH 9.6. For binding assays with Sn(d1) protein, microtiter plates were coated as above with 40 g/ml F(abЈ) 2 fragments of purified 3D6 mAb followed by varying amounts of purified Sn(d1). 10 7 mouse bone marrow cells or mesenteric lymphocytes were labeled at 37°C for 30 min with 10 g/ml BCECF-AM dye in 1 ml of RPMI 1640 ϩ 20 mM HEPES and washed three times in PBS ϩ 0.25% BSA. Immediately prior to binding assays, cells were resuspended in PBS ϩ 5 mM EDTA ϩ 0.25% BSA to a concentration of 8 ϫ 10 6 cells/ml or 0.4% volume/volume for erythrocytes, and 50-l aliquots were added to wells of the microtiter plate containing 50 l of the same buffer. After 1 h at 37°C, 50 l of 0.25% glutaraldehyde in PBS was added, and bound cells were fixed for 5 min at room temperature with gentle shaking. The nonadherent cells were removed by three washes with PBS, and fluorescence was measured using the Cytofluor TM 1350 (Millipore) system, setting the excitation wavelength at 485 nm and the emission wavelength at 530 nm. In some experiments the percentage of cells binding in each well was determined. Erythrocyte binding was measured using the pseudoperoxidase activity of hemoglobin as described (2).
Single Cell Analysis and Immunocytochemistry-Bone marrow cells that had adhered to Fc chimeras in solid phase assays on 96-well plates were fixed in 0.125% glutaraldehyde. Detection of cell surface antigens was performed by immunoperoxidase labeling using a Vectastain ABC kit (Vector Laboratories) as described (17). At least 300 cells were counted per well.
BIAcore Surface Plasmon Resonance Measurements-Receptor-ligand interactions were evaluated at a molecular level using the BIAcore TM biosensor (Pharmacia Biotech AB, Uppsala, Sweden) at 25°C. The running buffer used was HBS (150 mM NaCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, pH 7.4) and 0.005% Surfactant P-20 (Pharmacia). The mAb R10Z8E9 was immobilized onto the carboxymethylated dextran CM5 research grade sensor chip using the standard amine coupling kit (Pharmacia). For coupling, purified R10Z8E9 was injected in 10 mM sodium acetate, pH 4.5, at a concentration of 28 g/ml. Following coupling, noncovalently bound antibody was removed by sequential washing with 0.1 M glycine HCl, pH 2.5, and 50 mM NaOH. Specific interactions of Sn-Fc and CD22-Fc constructs with their respective ligands were carried out at a continuous flow rate of 1 l/min in HBS buffer. All injected proteins were diluted in HBS. Ligand binding to the sensor surface is recorded as an increase in response units where 1,000 response units corresponds approximately to the binding of 1 ng of protein/mm 2 .

Production of Truncated Sn-Fc and CD22-Fc Chimeras-
Previous studies have shown that recombinant forms of sialoadhesin and CD22 containing the first three amino-terminal Ig-like domains (d1-3) fused to the Fc portion of human IgG1 mediate cell adhesion in a sialic acid-dependent manner (3,4,6,13,17,23,24). To localize the binding site further, we made a set of three additional domain truncation mutants. For both sialoadhesin and CD22 the chimeric domain constructs made were (d1)Fc, (d1-2)Fc, (d2-3)Fc, and (d1-3)Fc (Fig. 1A). In addition, Sn(d1-17)Fc and CD22(d1-7)Fc are the wild type constructs containing all of the extracellular Ig-like domains. Purified preparations of each form were obtained by protein A-Sepharose affinity chromatography using supernatants of transiently transfected COS cells (Fig. 1B).

Comparison of the Binding Properties of Sn-Fc Chimeras with Human Erythrocytes and CD22-Fc Chimeras with Mouse
Lymphocytes-It was shown previously that human erythrocytes and mouse mesenteric lymphocytes bind at high levels to sialoadhesin and CD22, respectively (6,17,22). These cellular ligands were used in solid phase adhesion assays to characterize the binding activities of Sn-Fc and CD22-Fc domain truncation mutants. Sn(d1-17)Fc, Sn(d1-3)Fc, Sn(d1-2)Fc, and Sn(d1)Fc exhibited concentration-dependent adhesion to human erythrocytes ( Fig. 2A). Adhesion to all constructs was sialic acid-dependent as pretreatment of erythrocytes with sialidase completely abolished binding (data not shown). Both Sn(d1)Fc and Sn(d1-3)Fc were able to mediate similar levels of binding. A lower level of binding was observed with Sn(d1-2)Fc and Sn(d1-17)Fc. In contrast, no binding was observed with Sn(d2-3)Fc. Since this construct was engineered to contain the CD33 leader peptide, we also made a form of Sn(d1-3)Fc with the same leader peptide, designated Sn(d1-3L)Fc (Fig. 1B). The binding activity of this construct was identical to that of Sn(d1-3)Fc, showing that introduction of the CD33 leader peptide does not affect subsequent binding activity (data not shown). These results therefore suggest that for sialoadhesin, domain 1 is both necessary and sufficient for sialic acid-dependent binding.
For CD22, mesenteric lymphocytes bound at high levels to CD22(d1-7)Fc and CD22(d1-3)Fc with intermediate binding to CD22(d1-2)Fc, but no binding was observed with CD22(d1)Fc or CD22(d2-3)Fc (Fig. 2B). To exclude the possibility that the failure of CD22(d1)Fc to mediate binding was due to denaturation during purification (see BIAcore experiments below), binding assays were also carried out with unpurified Fc chimeras in the form of tissue culture supernatants. Under these conditions, no binding of lymphocytes to CD22(d1)Fc was observed, whereas high and equivalent levels of binding were seen with CD22(d1-2)Fc and CD22(d1-3)Fc (data not shown). Binding of lymphocytes to CD22 was sialic acid-dependent as pretreatment of cells with sialidase abolished binding with all constructs (data not shown). Since the CD22(d1)Fc construct may be incorrectly folded, these results are not inconsistent with the possibility that, like sialoadhesin, the binding site of CD22 lies in domain 1.
Binding of Bone Marrow Cells to Truncated Sn-Fc and CD22-Fc Chimeras-Previous single cell analyses with bone marrow cells have shown that sialoadhesin binds preferentially to developing granulocytes, whereas CD22 binds selectively to B lymphocytes (17). To test the cellular specificity of the truncation mutants, solid phase binding assays were performed with bone marrow cells, and the bound cells were characterized by immunocytochemistry using anti-mouse myeloid cell mAb 5C6 (anti-CD11b) and the anti-mouse B lymphocyte mAb TIB146 (anti-CD45 ABC/B220). The rank order of binding observed for Sn-Fc and CD22-Fc domain constructs with bone marrow cells (Table I) was similar to that observed with human erythrocytes and mouse lymphocytes, respectively (Fig. 2). No binding was observed with NCAM-Fc used as a negative control. Single cell analysis showed that the Sn-Fc constructs selectively bound cells of the myeloid lineage, whereas CD22-Fc constructs showed a strong preference for B lymphocytes (Table I). These results demonstrate that the cellular binding preferences of the domain constructs were preserved. Sialic Acid Linkage Specificity of Sialoadhesin-To confirm that domain 1 of sialoadhesin preferred NeuAc␣2-3Gal in Nand O-linked glycans over NeuAc␣2-6Gal in N-glycans, we compared the binding of Sn(d1)Fc and Sn(d1-3)Fc to a panel of derivatized human erythrocytes carrying a single type of sialylated glycan. These cells are generated by removing all terminal sialic acids and then reconstituting only specific sialy-lated glycans using purified sialyltransferases (35). This strategy has been used successfully to characterize the binding specificity of several sialic acid-dependent receptors including sialoadhesin, CD22, MAG, and CD33 (6,7,22,23). Sialidasetreated erythrocytes were reconstituted to give exclusively 3-O, 3-N, or 6-N. Both Sn(d1)-Fc and Sn(d1-3)-Fc bound erythrocytes containing either 3-O or 3-N, but not the 6-N structure (Fig. 3). In a previous study full-length sialoadhesin exhibited a similar sialic acid specificity when expressed on the surface of transiently transfected COS cells (5). Thus, the sialic acid specificity of domain 1 is determined independently of domains 2 and 3.
To show unequivocally that domain 1 of sialoadhesin contains the sialic acid binding site we engineered a soluble form of domain 1, Sn(d1), lacking the Fc portion. The purified recombinant protein migrated on SDS-polyacrylamide gel electrophoresis with an apparent M r of 12,500 (Fig. 4), close to the calculated M r of 13,302. Solid phase binding assays were carried out by immobilizing Sn(d1) onto microtiter plates coated with (FabЈ) 2 fragments of 3D6 mAb. Sn(d1) was able to bind human erythrocytes at high levels (Fig. 4). Binding was sialic acid-dependent as sialidase treatment completely abolished binding (data not shown).
Analysis of Binding Using Surface Plasmon Resonance-Previous studies have shown that sialoadhesin binds selectively to glycophorin on human erythrocytes, whereas CD22 binds to CD45 on lymphocytes (13,22,36). We next carried out binding assays with these defined molecular ligands for sialoadhesin and CD22 using the BIAcore biosensor, which employs surface plasmon resonance to detect protein interactions in real time (37). In a BIAcore experiment one of the proteins is immobilized onto a dextran matrix, and the other protein ligand is injected over the matrix. Protein binding is detected optically and measured as arbitrary units (response units, as defined earlier). In all BIAcore experiments the Fc chimeras were immobilized indirectly onto the sensor chip via a covalently coupled mAb (R10Z8E9), which is directed specifically against the Fc portion of human IgG. For both sialoadhesin and CD22, only those constructs that bound cellular ligands in solid phase binding assays were selected for the BIAcore experiments.
For sialoadhesin, we used BSA and asialoglycophorin to control for nonspecific and non-sialic acid-dependent binding to glycophorin, respectively. As shown in Fig. 5, A, C, and E, when the protein samples were injected before the binding of Sn(d1-3)Fc to the surface, the response rose quickly to a plateau level and remained at that level until the injection ended, after   3. Solid phase binding assays of Sn(d1)Fc and to Sn(d1-3)Fc to derivatized human erythrocytes. Human erythrocytes were either untreated (native) or treated with sialidase (asialo). Sialylated glycans on asialo erythrocytes were then reconstituted using purified sialyltransferases to give exclusively 3-O, 3-N, or 6-N. Binding assays were performed as described in the legend to Fig. 2. which it dropped quickly to the same level as before the injection. In contrast, when glycophorin was injected following the immobilization of Sn(d1-3)Fc, the response did not rise to a plateau level but instead continued to increase during the injection, indicating that glycophorin was binding to the Sn(d1-3)Fc protein. Following the injection, the response initially dropped quickly as the injected sample was washed out of the flow cell. However, the response remained elevated after the injection, indicating that glycophorin remained bound to the Sn(d1-3)Fc and slowly dissociated (see Fig. 5A, inset). In comparison, following immobilization of the Sn(d1-3)Fc protein, no changes in response were seen with the control proteins, asialoglycophorin and BSA.
Because of the variability in absolute responses between experiments, binding to glycophorin of CD22(d1-3)Fc, Sn(d1-17)Fc, and Sn(d1)Fc was compared with Sn(d1-3)Fc in separate experiments (Fig. 5). Sn(d1)Fc and Sn(d1-17)Fc bound glycophorin as indicated by a small but clearly increased response during and immediately after glycophorin injection, whereas the response after BSA and asialoglycophorin injection remained unchanged. Binding of Sn(d1)Fc and Sn(d1-17)Fc to glycophorin occurred at levels similar to that of Sn(d1-3)Fc. As might be expected, no binding was observed with CD22(d1-3)Fc.
Similar binding experiments were performed with the truncated CD22-Fc chimeras to CD45 purified from rat thymus (Fig. 6). BSA and CD45 were injected before and after injection of the CD22-Fc chimeras. CD22(d1-3)Fc bound as well as CD22(d1-7)Fc. In comparison, CD22(d1-2)Fc bound poorly to CD45 (Fig. 6A). No binding of CD45 was observed with FIG. 4. Solid phase binding assay of Sn(d1) with human erythrocytes. Sn(d1) protein at varying concentrations was adsorbed onto microtiter plates that had been coated with 40 g/ml F(abЈ) 2 fragments of 3D6 mAb. Binding assays were carried out with human erythrocytes as described in the legend to Fig. 2. Data show the mean values of duplicate well at each point; similar results were obtained in three independent experiments. The inset shows SDS-polyacrylamide gel electrophoresis of 5 g of purified Sn(d1) protein under reducing conditions, stained with Coomassie Blue.

FIG. 5. Analysis of glycophorin binding to truncated Sn-Fc chimeras.
Purified forms of the indicated Fc chimeras were injected into a flow cell in which an anti-human IgG mAb had been immobilized onto the sensor surface. At the end of the injection the response is higher than the base line, which represents bound Fc chimera. BSA (B, 0.5 mg/ml), asialoglycophorin (aG, 0.5 mg/ml), and glycophorin (G, 0.5 mg/ml) were all injected for 3 min (short bars) before and after the Fc chimera was bound. An increase in the response both during and following the injection of glycophorin reflects binding to that particular Fc chimera. The inset in panel A shows a magnified view of glycophorin binding before (dotted lines) and after (solid lines) injection of Sn(d1-3)Fc. Results of three experiments are shown. In each experiment binding of Sn(d1-3)Fc is compared with CD22(d1-3)Fc (panels A and B), Sn(d1-17)Fc (panels C and D), and Sn(d1)Fc (panels E and F). Similar results were obtained for each construct in at least three independent experiments.
FIG. 6. Analysis of CD45 binding to truncated CD22-Fc chimeras. Panel A, purified forms of the indicated Fc chimeras were injected (long bars) into a flow cell in which an anti-human IgG mAb had been immobilized onto the sensor surface. BSA (0.5 mg/ml) and purified rat thymus CD45 (30 g/ml) were injected before and after (short bars) the Fc chimera was bound. Panel B, tissue culture supernatants (TCS) containing the indicated Fc chimeras were injected (long bars) into a flow cell in which an anti-human IgG mAb had been immobilized onto the sensor surface. BSA (10 g/ml) and CD45 (10 g/ml) were injected before and after (short bars) the Fc chimera was bound. An increase in the response both during and following the injection of CD45 reflects binding to that particular Fc chimera. Similar results were obtained for each construct in at least three independent experiments. Sn(d1-3)Fc.
One explanation for why CD22(d1-2)Fc bound weakly to CD45 might be that a large fraction of the protein had been denatured during affinity purification of the protein which involves elution at pH 3.0. To test this possibility, we compared binding of unpurified CD22(d1-2)Fc and CD22(d1-3)Fc present in COS cell supernatants. As shown in Fig. 6B, unpurified CD22(d1-2)Fc and CD22(d1-3)Fc bound at equivalent levels to CD45, suggesting that CD22(d1-2)Fc had undergone selective denaturation during low pH elution. Binding of the CD22 constructs to CD45 was sialic acid-dependent because sialidase treatment of immobilized CD45 prevented subsequent binding (data not shown).
Epitope Mapping of mAbs Directed against Sialoadhesin and CD22-The availability of a panel of truncated Sn-Fc and CD22-Fc chimeras allowed us to map the epitopes recognized by mAbs to each protein using the BIAcore. For sialoadhesin, we have recently described a panel of mAbs which define at least two distinct epitopes, one recognized by the cross-competing mAbs 3D6 and 1C2 and another recognized by SER-4 (17). Both 3D6 and 1C2 bound to Sn(d1)Fc, Sn(d1-2)Fc, and Sn(d1-3)Fc (Fig. 7), showing that the epitope recognized by these antibodies is present within domain 1. However, binding of both antibodies to Sn(d1-2)Fc was lower than to the other two proteins, suggesting that a fraction of Sn(d1-2)Fc may have been partially denatured. SER-4 mAb bound to Sn(d1-3)Fc but not to Sn(d1)Fc or Sn(d1-2)Fc (Fig. 7A). Therefore the epitope of SER-4 may lie in domain 3 or in a region shared between domains 2 and 3.
Two mAbs, CY34 (30) and NIMR6 (15), have been shown to recognize mouse CD22. Using the BIAcore with unpurified forms of the CD22-Fc constructs, CY34 mAb bound to CD22(d1-2)Fc, CD22(d1-3)Fc, and CD22(d1-7)Fc but not to CD22(d1)Fc (Fig. 8). Since it is possible that the CD22(d1)Fc construct was incorrectly folded, these results suggest that the epitope recognized by CY34 mAb may lie in either domain 1 or domain 2 or in a region shared between both of these domains. In contrast, NIMR6 mAb bound CD22(d1-3)Fc and CD22(d1-7)Fc but not CD22(d1-2)Fc (Fig. 8). This shows that the epitope recognized by this antibody is either in domain 3 or is in a region shared between domains 2 and 3.

DISCUSSION
By using a series of domain deletion constructs we have mapped the sialic acid binding sites of sialoadhesin and CD22  -4 (panels B, E, and H), or 1C2 (panels C, F, and I) were then injected at a concentration of 10 g/ml (short bars). An increase in response both during and after the injection of mAbs reflects binding to that particular Fc chimera.
as well as epitopes recognized by mAbs specific to each molecule. We demonstrate that for sialoadhesin, the amino-terminal V-set Ig-like domain (domain 1) is both necessary and sufficient for binding in all assays. Domains 2 and 3 do not appear to contribute significantly to the adhesive functions since binding of Sn(d1)Fc was at levels comparable to that of Sn(d1-3)Fc. However, a recombinant construct that contained the first two domains displayed reduced binding as compared with the construct containing domain 1 on its own or domains 1ϩ2ϩ3. Since mAbs 3D6 and 1C2 bound to Sn(d1-2)Fc at reduced levels compared with Sn(d1)Fc and Sn(d1-3)Fc, this suggests that a fraction of Sn(d1-2)Fc was partially denatured, which could therefore explain its reduced binding activity. Similar observations were made with CD22(d1-2)Fc, and in this case it was found that full activity could be restored if binding assays and epitope mapping were carried out with unpurified forms of the Fc proteins. Since the routine purification of Fc proteins involves an elution step at pH 3.0, these results indicate that the Sn-and CD22(d1-2)Fc constructs are unusually susceptible to acid denaturation. In this respect it is interesting that our attempts to generate purified MAG(d1-2)Fc have been unsuccessful so far. 2 In contrast to sialoadhesin, our results show that murine CD22 requires at least the first two domains for binding to cellular and molecular ligands. These observations are consistent with a recent report (38) demonstrating that for human CD22 expressed in COS cells the presence of the first two amino-terminal domains together is essential for sialic acid-dependent adhesion. In view of the similarities in sequence and domain organization between the amino-terminal domains of sialoadhesin and CD22 it might be expected that, similar to sialoadhesin, the binding site of CD22 would reside in the first amino-terminal Ig-like domain. Indeed, recent mutagenesis screens of sialoadhesin and CD22 have shown that domain 1 of both molecules contains residues that are critical for sialic acid-dependent binding. 3 We therefore favor the idea that the sialic acid binding site of CD22 lies in domain 1 and that domain 2 is required for correct folding of domain 1. Indirect support for this interpretation is provided by the observation that human CD22 constructs lacking domain 2 show reduced reactivity with mAbs that are directed against domain 1 and/or domain 2 (13,38), thereby suggesting that domain 2 is critical for the correct conformation of domain 1. Likewise, in the present study it is possible that the failure of the CY34 mAb to bind CD22(d1)Fc is a result of incorrect folding. Generation of more mAbs and/or production of chimeric constructs between appropriate mouse strains (12) will be required to resolve this issue. Unpublished observations in our laboratory with MAG and CD33 show that, like CD22, domain 1 on its own is unable to mediate sialic-acid dependent adhesion. 4 Within the Sialoadhesin family, therefore, it appears that sialoadhesin is unusual in that domain 1 can function independently of domain 2. It is possible that formation of a disulfide bond between domains 1 and 2 (9, 10) could be important for the correct folding of domain 1 in the cases of CD22, MAG, and CD33, but not in the case of sialoadhesin.
For the majority of Ig superfamily adhesion molecules involved in heterophilic interactions, structure-function studies have shown that the dominant binding sites are localized to the first amino-terminal Ig-like domain (reviewed in Ref. 1). In general, it can be expected that the amino-terminal domain is more accessible for mediating interactions with corresponding counterreceptors. The binding activities of the membrane prox-imal domains are less defined, but they may be important in making the binding site more accessible. Thus, the binding site on ICAM-1 for lymphocyte function-associated molecule-1, rhinovirus, and malaria-infected erythrocytes is localized in the first amino-terminal domain, but removal of the membrane proximal domains results in decreased ligand binding when ICAM-1 is expressed in monkey COS cells (39,40). In certain cases where domain 1 has been shown to contain the binding site, it has also been shown that there is a conformational dependence on domain 2. With ICAM-1 and CD4, it has not been possible to express domain 1 in the absence of domain 2 (39,40). This conformational dependence is thought to be a result of the close packing of the first two domains as shown in the crystal structure of CD4 (41,42). Thus, it appears that for several adhesion molecules belonging to the Ig superfamily, the ligand binding site is located in the first amino-terminal domain, but an adjacent domain is required for correct conformation.
Several cell adhesion molecules of the Ig superfamily exhibit a dual homophilic/heterophilic adhesion activity, including NCAM, CD31, and CEA family proteins (1). In these molecules, the homophilic and heterophilic binding sites appear to be distinct. For example, homophilic NCAM-NCAM binding interactions in trans appear to be mediated by a 10-amino acid sequence within the third Ig-like domain (43). This is distinct from its heterophilic glycosaminoglycan binding region in the second amino-terminal domain. NCAM can also mediate carbohydrate-dependent cis interactions with L1 via a region located in its fourth Ig-like domain (44). CD31-CD31 homophilic interactions involve a more extensive area over multiple domains than NCAM homophilic interactions. It is proposed that domains 2 and 3 on one CD31 molecule interact in an antiparallel fashion with domains 5 and 6 on an apposing CD31 molecule (45). A similar mechanism has been proposed for CEA homotypic interactions, which involve reciprocal interactions between antiparallel CEA molecules aligned in trans (46). It is therefore evident that cell adhesion molecules containing several Ig-like domains can bind multiple ligands using distinct domains.
The Sialoadhesin family represents a new class of carbohydrate-binding proteins in which an Ig fold has acquired the capacity to mediate sialic acid-dependent adhesion. The overall molecular organization of sialoadhesins bears several similarities to selectins, arguably the best characterized mammalian lectins involved in cell-cell interactions (reviewed in Ref. 47). Selectins contain an amino-terminal C-type lectin domain that functions in carbohydrate binding. Our results show that for sialoadhesin, the lectin region is within the amino-terminal V-set domain. We anticipate that the same region is used by CD22, MAG, and CD33. The localization of the carbohydrate binding sites at the amino terminus of selectins and sialoadhesins could be important in allowing their respective lectin domains to extend out of the cell glycocalyx to interact with carbohydrate ligands on apposing cells (7). Selectins use varying numbers of complement control protein domains to extend the lectin domain from the cell surface, whereas sialoadhesins use variable numbers of C2-set Ig domains. In selectins, correct folding of the lectin domain may require an adjacent epidermal growth factor-like domain (48 -51). This is analogous to the apparent requirement of the second Ig domain in CD22 (and possibly MAG and CD33) for correct folding of the V-set domain.
The V-set and adjacent C2-set domains of sialoadhesins contain structural features that are not present within other adhesion molecules of the Ig superfamily and could therefore be involved in sialic acid-dependent binding. These include an inter-␤-sheet disulfide between ␤-strands B and E of the V-set domain (9) and a predicted disulfide between the V-and C2-set domains 2 (10). Since the V-set domain of sialoadhesin can function independently of the C2-set domain, a direct role of the interdomain disulfide for binding of sialoadhesin to sialic acid is unlikely, but it may be important for correct folding of the V-set domain in the cases of CD22, MAG, and CD33. The role in sialic acid-dependent binding of the conserved inter-␤sheet disulfide within the V-set domain of sialoadhesins awaits further investigation.