Localization of the putative sialic acid-binding site on the immunoglobulin superfamily cell-surface molecule CD22.

B-lymphocyte antigen CD22 is a member of the recently described sialoadhesin family of immunoglobulin-like cell-surface glycoproteins that bind glycoconjugates terminating in sialic acid. One prominent ligand for CD22 is the highly glycosylated leukocyte surface protein CD45. Using surface plasmon resonance spectroscopy, we characterized the interaction of recombinant mouse CD22 with native CD45 purified from rat thymus (CD45-thy). By in situ desialylation and resialylation of immobilized CD45-thy, we show that mouse CD22 binds to the sialoglycoconjugate NeuGc alpha 2-6Gal beta 1-4GlcNAc carried on CD45-thy N-glycans. Previous studies have shown that the sialic acid-binding site lies within the two membrane-distal domains of CD22 (domains 1 and 2), which are V-set and C2-set immunoglobulin superfamily domains, respectively. To further localize the binding site, we have made 42 single amino acid substitutions throughout both domains. All 12 mutations that abrogated binding to CD45-thy without disrupting antibody binding were of residues within the GFCC'C" beta-sheet of domain 1. These residues are predicted to form a contiguous binding site centered around an arginine residue in the F strand that is conserved in all members of the sialoadhesin family. Our results provide further evidence that immunoglobulin superfamily cell adhesion molecules use the GFCC'C" beta-sheet of membrane-distal V-set domains to bind structurally diverse ligands, suggesting that this surface is favored for cell-cell recognition.

Immunoglobulin superfamily (IgSF) 1 domains are probably the commonest domain type involved in cell-surface recognition, being present in ϳ40% of all proteins identified on the surface of leukocytes (1). One possible reason for this is that IgSF domains provide a stable, but versatile, recognition plat-form, capable of binding to structurally diverse ligands (2). Typically, IgSF cell adhesion molecules bind either to other IgSF molecules or to integrins (3,4), but recent reports indicate that some IgSF cell adhesion molecules bind carbohydrate ligands (reviewed in Ref. 5). The best characterized of these lectin-like IgSF proteins are a group of homologous proteins (termed the sialoadhesin family) that bind carbohydrate structures terminating in sialic acid (5)(6)(7). Members of this family include the leukocyte proteins CD22, sialoadhesin, and CD33 as well as myelin-associated glycoprotein and Schwann cell myelin protein (5)(6)(7).
CD22 is expressed on a subpopulation of mature B-cells and has been implicated in cell adhesion as well as in modulating signaling through the B-cell antigen receptor (BCR) (reviewed in Ref. 8). CD22 associates loosely with the BCR (9, 10) and is tyrosine-phosphorylated following BCR ligation (11). This leads to association with and activation of the tyrosine phosphatase SHP (12), which can inhibit signaling through the BCR (13)(14)(15). The binding of anti-CD22 antibody-coated beads to B-cells decreases the activation threshold of the BCR, presumably by removing CD22 (and associated SHP) from the vicinity of the BCR (12). Together, these findings suggest that physiological interactions between CD22 and natural cell-surface ligands may function to modulate signaling through the BCR (12).
With the exception of antibody-carbohydrate interactions, little is known about carbohydrate recognition by IgSF molecules (5). As a first step toward understanding the structural basis of sialic acid recognition, we undertook to identify the sialic acid-binding site on CD22. Previous studies on human (27) and mouse (22,28) CD22 have shown that the sialic acidbinding site lies within domains 1 and 2. In the present study, we extend this work by making single amino acid substitutions of surface residues throughout domains 1 and 2 of mouse CD22. Our results suggest that the CD22 sialic acid-binding site is situated on the GFCCЈCЉ ␤-sheet of domain 1 centered on an arginine residue in the F strand that appears to be essential for sialic acid recognition.
Surface Plasmon Resonance Spectroscopy-All BIAcore experiments were performed on a BIAcore TM biosensor (Pharmacia Biosensor, Uppsala) at 25°C in the running buffer HBS, which contains 150 mM NaCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 10 mM HEPES, pH 7.4, and 0.005% Surfactant P-20 (Pharmacia Biosensor). Proteins were covalently coupled via amine groups onto the carboxymethylated dextran surface of CM5 (research-grade) sensor chips (Pharmacia Biosensor) using the standard amine coupling kit (Pharmacia Biosensor) as recommended (36), with the following modifications. During coupling, CD45-thy and sCD45ABC-CHO were injected for 7 min at 20 -40 g/ml in 10 mM sodium formate, pH 3, and in 10 mM sodium acetate, pH 4, respectively. Both proteins were regenerated by injecting 100 mM HCl for 3 min. The anti-human Fc antibody was injected at 28 g/ml in 10 mM sodium acetate, pH 4.5, and regenerated by sequential 3-min injections of 0.1 M glycine HCl, pH 2.5, and 5 mM NaOH. All experiments were performed at a flow rate of 1 l/min, except for the amine coupling reactions, which were performed at a flow rate of 5 l/min.
Expression of Mutant Proteins-The DNA fragment encoding domains 1-3 of mouse CD22 (C57Bl allele; see "Results and Discussion") had previously been cloned into the EcoRI site of the expression vector pIG (28,40), yielding a chimeric protein comprising domains 1-3 of CD22 fused to the Fc portion of human IgG1 (CD22Fc). The entire CD22Fc fragment was excised with HindIII and NotI, blunt-ended, and subcloned by blunt-end ligation into the XbaI site of the phagemid expression vector pEF-BOS (41). CD22 mutants were generated directly in CD22Fc/pEF-BOS as described (42) using the Muta-Gene phagemid mutagenesis kit (Version 2, Bio-Rad). All mutations were confirmed by DNA sequencing. When CD22 mutations disrupted CD45 or CY34 binding, the DNA encoding domains 1 and 2 was sequenced in order to exclude spurious mutations. Mutant CD22Fc chimeras were expressed by transient transfection of COS-7 cells as described previously (42). Tissue culture supernatants (TCS) were concentrated 3-4fold before analysis using Centricon-10 concentrators (Amicon, Inc.).

Mouse CD22 Binds to NeuGc on CD45 N-Glycans-Previous
studies have shown that only the two NH 2 -terminal domains of human (27) and mouse (22, 28) CD22 (domains 1 and 2) are required for sialic acid binding. However, we have found that a mouse CD22 construct containing domains 1 and 2, but lacking domain 3, was somewhat unstable (28). We therefore used a construct containing domains 1-3 of mouse CD22 fused to the Fc portion of human IgG1 (CD22Fc) (6). Using surface plasmon resonance spectroscopy, as implemented in the BIAcore instrument (43), we have shown that CD22Fc binds to native CD45 (CD45-thy) purified from rat thymus (28). To further characterize this interaction, we modified the sialoglycoconjugates present on CD45-thy and examined the effect on CD22Fc binding (Fig. 1).
Analysis of the Sialic Acid Composition of CD45-thy-Since mouse CD22 requires ␣2-6-linked NeuGc for binding, biologically relevant ligands for CD22 should contain this sialic acid rather than NeuAc. Normal human tissues do not contain NeuGc, but this sialic acid is common in rodents (45,46). However, the relative amount of NeuGc differs between cell types and is developmentally regulated (47)(48)(49)(50). Previous stud-ies of mouse lymphocytes found that NeuGc constituted 40 -50% of the sialic acid in glycolipids (51,52). However, no analysis of mouse or rat lymphocyte glycoproteins has been reported. We therefore analyzed the sialic acid composition of glycoproteins isolated from rat thymus (Table I). For comparison, we also studied a rat serum protein and rat CD45 that had been expressed in CHO cells (Table I). This analysis revealed that most (Ͼ98.8%) of the sialic acid in the thymic proteins CD45-thy and thy-1 is NeuGc. In contrast, the serum protein ␣ 1 -acid glycoprotein, which is synthesized by hepatocytes, contains mainly (Ͼ89%) NeuAc (Table I). NeuAc constituted 98% of the sialic acid in sCD45ABC-CHO (Table I), which is in agreement with other studies of glycoproteins expressed in CHO cells (53). Taken together, these results demonstrate that CD45-thy is a suitable ligand for murine CD22 since it contains abundant ␣2-6-linked NeuGc. In support of a physiological role for this interaction, Law et al. (22) recently demonstrated that CD45 is prominent among the glycoproteins that are immunoprecipitated from mouse B-cell lines using mouse CD22Fc.
Sequence Alignments and Mutagenesis Strategy-Two mouse CD22 alleles have been isolated from BALB/c and DBA/2J mice, respectively (54,55). While sequencing the CD22 construct used in the present study (which originated from C57Bl mice (6)), it emerged that it encodes a third allele (Fig. 2,  CD22 C57Bl). This allele is identical to the BALB/c allele in the region encoding domains 1-3, with the exception of the codons for residues 79 (Val instead of Cys, numbered from the initiation codon), 247 (Arg instead of Cys), and 250 (Arg instead of His), in which the DNA sequence is identical to the DBA/2J allele (Fig. 2). These changes result in the loss of an unusually FIG. 1. Mouse CD22Fc binds NeuGc␣2-6Gal␤1-4GlcNAc carried on CD45 N-glycans. A-D, CD45-thy was covalently coupled to the BIAcore sensor surface. CD22Fc and the sialic acid-binding lectins MAA and SNA were then injected at 0.5 mg/ml for 4 min each (bars) over unmodified thymic CD45 (A), sialidase-treated CD45 (B), sialidase-treated CD45 resialylated with NeuAc using Gal␤1-4GlcNAc ␣2-6-sialyltransferase (C), or sialidase-treated CD45 resialylated with NeuGc using Gal␤1-4GlcNAc ␣2-6-sialyltransferase (D). Following each injection, bound protein was eluted with a 4-min injection of 100 mM HCl (arrows mark the beginning of these injections). E, sCD45ABC-CHO was coupled to the sensor surface. Mouse CD22Fc, MAA, and SNA were injected (0.5 mg/ml for 4 min) first over unmodified sCD45ABC-CHO and then after the indicated desialylation and resialylation steps. The binding response during each injection was measured 20 s after the injection (to eliminate the bulk phase effect) and is expressed as a percentage of the maximal response seen for each ligand during the experiment, which was 1380, 3640, and 7300 response units for CD22Fc, MAA, and SNA, respectively. ST6N, Gal␤1-4GlcNAc ␣2-6-sialyltransferase; ST3N, Gal␤1-3(4)GlcNAc ␣2-3-sialyltransferase. positioned pair of cysteine residues that are present in the CD22 BALB/c allele, but not in any of the other sialoadhesin family members (56).
To aid in the selection of residues to mutate, domains 1 and 2 of CD22 were aligned with IgSF domain sequences for which there are structural data available (Fig. 2). Domain 1 of CD22 was aligned with the V-set domain (domain 1) of rat CD2 (57, 58) (Fig. 2), whereas domain 2 was aligned with domain 2 of VCAM-1 (59). CD22 residues in domains 1 and 2 could be assigned accurately to the structurally conserved B, C, E, and F ␤-strands (Fig. 2) by aligning residues characteristic of IgSF domains (2,60,61). In a similar manner, residues in CD22 domain 1 could be assigned to the beginning of the D strand and to the end of the G strand, and residues in domain 2 could be assigned to the A strand and to the end of the G strand (Fig.  2). In contrast, CD22 residues could not reliably be assigned to the CЈ and CЉ strands of domain 1 or to the CЈ/D strand of domain 2 (Fig. 2). The assignment of residues to the loop regions was tentative except for the E-F loop, which is structurally conserved in V-set and C2-set IgSF domains (2,60,61).
The sialic acid-binding site on sialoadhesin has been definitively localized to its V-set domain (domain 1 (28)), but in the case of CD22, a contribution from domain 2 has not been ruled out (27,28). To further localize the sialic acid-binding site on CD22, we mutated residues predicted to lie on the surface of domain 1 or 2. We introduced drastic changes rather than mutating to alanine because our primary aim was to delineate the structural binding site. It has been shown that alanine mutagenesis may only identify a fraction (ϳ25-40%) of the residues within the binding site (62, 63). We have previously used this approach of making drastic mutations to identify the interacting surfaces of the cell adhesion molecules CD2 and CD48 (42) 2 and obtained results that agree well with structural studies (58,64).
Identification of the Sialic Acid-binding Site on CD22-Mutant CD22Fc chimeras were expressed by transient transfection of COS-7 cells and then analyzed for ligand and antibody binding by surface plasmon resonance spectroscopy using the approach outlined schematically in Fig. 3A (upper left). TCS containing wild-type or mutant CD22Fc was injected over a sensor surface to which an anti-Fc mAb had been covalently coupled (Fig. 3, A and B, long bars). The initial rapid increase is due to the high bulk refractive index of the injected TCS ("bulk phase effect"), whereas the slower, more sustained increase reflects the binding of CD22Fc to the anti-Fc mAb on the sensor surface (Fig. 3, A and B, long bars). The contribution from the bulk phase effect ends when the injection of the TCS is completed and the flow of the running buffer resumes. The response then drops rapidly to a new, elevated base line, the level of which is proportional to the mass of bound CD22Fc, with 1000 response units representing ϳ1 ng/mm 2 of bound protein (43). The control protein BSA and CD45-thy (both at 26 g/ml) were injected over the sensor surface both before (to control for a bulk phase effect) and after the binding of wildtype or mutant CD22Fc to the sensor surface. A substantially increased response is seen when CD45 is injected over immobilized wild-type CD22Fc, reflecting binding, whereas the response to the injection of BSA is unchanged (Fig. 3, A and B, Wild type). The mutant CD22Fc constructs were analyzed in the same way and compared with wild-type CD22Fc (Fig. 3).
Initially, nine mutations were made in each of domains 1 and FIG. 2. Alignment of CD22 with IgSF molecules of known structure. The predicted protein sequences of domains 1 and 2 of the mouse CD22 alleles C57Bl (this study), BALB/c (GenBank TM /EMBL accession number L02844), and DBA/2J (GenBank TM /EMBL accession number L16928) were manually aligned with human CD22 (GenBank TM /EMBL accession number X59350), mouse sialoadhesin (GenBank TM /EMBL accession number Z36293), and either rat CD2 (domain 1 (d1)) (SwissProt accession number P08921) or VCAM-1 (domain 2 (d2)) (SwissProt accession number P19320). The ␤-strand assignments (solid bars) were based on the structures of CD2 (57, 64) and VCAM-1 (59) as well as on structural data from other IgSF domains (2,60,61). The division between domains 1 and 2 of mouse CD22 is made at the junction of exons 4 and 5 (55). Dashed lines instead of bars are shown where there are no grounds for making precise assignments to ␤-strands. Boxed mouse CD22 residues were mutated in the present study, whereas boxed sialoadhesin residues were mutated in the accompanying study (65).
2 (Table II). Of these, only two mutations, both in domain 1, led to a decrease in CD45 binding (R130E and E140K) (Fig. 3A and Table II). Both mutants bound normally to mAb CY34 (Fig. 4A and Table II). The sequence alignment (Fig. 2) places Arg-130 and Glu-140 on adjacent F and G ␤-strands in domain 1 (Fig.  5). Interestingly, Arg-130 is one of only five residues in domain 1 (apart from residues characteristic of IgSF domains) that are completely conserved within the sialoadhesin family (indicated by ¦ in Fig. 2) (2, 56, 60, 61), suggesting that it may play an important role in sialic acid recognition. To provide stronger evidence for this, we made the substitutions R130A and R130K, which are less likely to abrogate binding by introducing unfavorable effects. Both mutations abolished CD45 binding (Fig. 3B) without affecting the binding of mAb CY34 (Table II), suggesting that Arg-130 is critical for sialic acid recognition.
The binding site was further defined with 22 additional mutations in and around the GFCCЈCЉ ␤-sheet of domain 1 (Fig. 2 and Table II). Of a total of 42 mutations made (Table II), 30 had little or no effect on CD45 binding (examples include K74E, R120D, K149D, and K185E (Fig. 3)), 10 completely abolished CD45 binding (examples include R130E, R130K, R130A, and W138R (Fig. 3)), and 2 substantially decreased, but did not abolish, CD45 binding (E140K and K73E (Fig. 3)). The partial effect of the latter mutants suggests that they lie on the periphery of the binding site. According to the alignment shown ϩϩ ϩϩ a The 18 mutations made initially are in boldface. Not included in the analysis are six CD22 mutants that either were not expressed (R108E) or were expressed at very low levels (H30D, F63D, D181R, E189K, and K234D) and bound neither CY34 nor CD45, strongly suggesting that they were not correctly folded.
b The ability of the CD22 mutants to bind purified CD45 was assayed as shown in Fig. 3. ϩϩ, binding Ͼ80% of that seen with wild-type CD22; ϩ, binding detected but clearly decreased (10 -30% of wild-type level); Ϫ, no binding detected (Ͻ5% of wild-type level).
c The binding of mAb CY34 to the CD22 mutants was assayed as shown in Fig. 4. ϩϩ, binding indistinguishable from that of wild-type CD22; Ϫ, no binding detected.
in Fig. 2, the mutations that abrogate CD45 binding fall within the GFCCЈCЉ ␤-sheet and are predicted to form a well defined contiguous region centered around Arg-130 in the F strand (Figs. 2 and 5). The positioning of the F and C strand mutations is likely to be correct because the alignment of CD22 with CD2 in both these regions is excellent (Fig. 2), and these strands form part of the structurally conserved core of IgSF domains (60,61). Because of a poor alignment with CD2, the positioning of the G and CЈ strand mutants is more tentative (Fig. 2). However, it is clear that residues in the F-G loop and/or the beginning of the G strand contribute to the CD45-binding site.
Our finding that none of the nine mutations in domain 2 affect CD45 binding (Fig. 3 and Table II) suggests that domain 2 does not contribute directly to sialic acid recognition. This is consistent with the observation that domain 1 of sialoadhesin is sufficient for sialic acid binding (28). Furthermore, mutagenesis of sialoadhesin (65) suggests that its sialic acid-binding site is also localized to the GFCCЈCЉ ␤-sheet of domain 1, centered around the same conserved F strand arginine (Figs. 2 and 6). Taken together, these data suggest that sialic acid recognition by CD22 and sialoadhesin involves only domain 1. Prior observations that domain 1 of CD22 binds poorly (22) or not at all (27,28) to ligand when expressed in the absence of domain 2 may be explained by an inability of domain 1 to fold correctly in the absence of domain 2. Support for this is provided by two lines of evidence that suggest that domains 1 and 2 of CD22 are intimately associated. First, the conserved cysteines present in domains 1 (A-B loop) and 2 (B-C loop) of sialoadhesin family members appear to form an interdomain disulfide bridge (66). 3 Second, residues in both domains 1 and 2 contribute to the CY34 epitope (see below and Table II).
A potential source of artifact in the present study is the possibility that some or all of the mutants do not lie within the sialic acid-binding site, but instead disrupt the overall folded structure of CD22. While this possibility cannot be eliminated, several considerations suggest that this is unlikely. First, all mutants that did not bind CD45 still bound mAb CY34. Our mutagenesis studies suggest that CY34 binds to a "discontinuous" or "conformational" epitope on CD22 (see below), which requires the correct folding of domain 1 and 2. This is supported by our observation that several, widely spaced mutants that were expressed only at very low levels bound neither CD45 not CY34 (see Footnote a to Table II). Second, the 12 mutations FIG. 4. A, analysis of CY34 binding to CD22 mutants. TCS containing the indicated CD22Fc mutants was injected for 12 or 40 min over a sensor surface to which an anti-Fc mAb had been covalently coupled (see Fig. 3). The anti-CD22 mAb CY34 (in TCS) was then injected over the immobilized CD22 mutants for 4 -6 min. An elevated base line following the injection of CY34 TCS indicates binding. B, CY34 and CD45 bind to different regions of CD22. In this experiment, an anti-Fc mAb was immobilized to the sensor surface. Left, BSA and CD45 (65 g/ml each) were injected for 4 min before and after the binding of wild-type CD22Fc (30 g/ml, 10 min) to the sensor surface. Middle, purified CY34 (30 g/ml, 6 min) was injected before and immediately after the binding of CD22Fc to the sensor surface, followed by injection of BSA and CD45. Right, after the binding of CD22Fc, BSA and CD45 are injected, followed by CY34. that decrease CD45 binding (without affecting CY34 binding) lie within a single contiguous area, with the mutations that have a partial effect (K73E and E140K) situated on the edge of this area. And finally, mutations in the equivalent region of sialoadhesin also disrupt sialic acid binding without disrupting the binding of mAbs directed to this domain 1 (65).
The CY34 Epitope Includes Portions of Domains 1 and 2-CY34 is an allele-specific mouse anti-mouse CD22 mAb (35) that has been reported to bind the CD22 BALB/c allele, but not the DBA/2J allele (55). The CD22 C57Bl allele identified in the present study also binds CY34 (28). Using truncation mutants, it has been shown that the CY34-binding site lies within domains 1 and 2 of CD22 (22,28). Of the 42 mutants in domains 1 and 2, three (R120D, K149D, and K185E) abolished CY34 binding (Figs. 4A and 5 and Table II). None of these three mutations affected CD45 binding, suggesting that they do not disrupt the overall structure of CD22 (and Fig. 4A and Table  II). The mutated residues are widely distributed in the primary sequence, with Arg-120 in the E-F loop of domain 1, Lys-149 at the junction of domains 1 and 2, and Lys-185 in the C strand of domain 2. Although distant in the primary sequence (Fig. 2), Arg-120, Lys-149, and Lys-185 are likely to lie in close proximity in the folded structure (see Fig. 5 for the predicted positions of Lys-149 and Arg-120). Thus, as with the majority of monoclonal antibodies (67,68), CY34 binds a discontinuous (and therefore conformationally sensitive) epitope that includes portions of domains 1 and 2 and is some distance from the putative sialic acid-binding site (Fig. 5). In agreement with the latter, CD45 binding to immobilized CD22 is not inhibited by bound CY34, nor is CY34 binding inhibited by bound CD45 (Fig. 4B), demonstrating that their binding sites on CD22 do not overlap.
IgSF Molecules Involved in Cell-Cell Recognition Bind Structurally Diverse Ligands Using the Same ␤-Sheet-This analysis of CD22 and the accompanying study on sialoadhesin (65) suggest that both these proteins bind sialoglycoconjugates through the GFCCЈCЉ ␤-sheet of their membrane-distal V-set  (72) with CD28 or CTLA-4. All the mutations lie within membrane-distal V-set IgSF domains of these molecules. The ribbon drawings of human CD2 and VCAM-1 are based on their crystal structures (58,59). A ribbon drawing of domain 1 of CD2 is used as a template to display the CD22, sialoadhesin, and CD80 mutants. The positioning of the residues was guided by the ␤-strand assignments in Fig. 2 and Ref. 72. This figure was drawn using MOLSCRIPT (73). domain (Fig. 6). As discussed in the accompanying paper (65), it seems likely that other members of the sialoadhesin family bind sialoglycoconjugates through the same site. The ligandbinding sites of several cell-surface IgSF molecules involved in cell-cell recognition have recently been characterized (Fig. 6). These include the T-cell surface molecule CD2 (58, 69 -71), which binds to the closely related IgSF molecules CD48 and CD58; VCAM-1 (59), which binds to the integrin VLA-4; and the B-lymphocyte molecule CD80 (72), which binds to the T-cell surface molecules CD28 and CTLA-4. In each instance, the binding sites have been localized to different portions of the GFCCЈCЉ (CD2 and CD80) or GFC (VCAM-1) ␤-sheet of the membrane-distal domain (Fig. 6). This ␤-sheet appears to be favored for interactions mediating cell-cell recognition, presumably because of its membrane-distal location and because, as shown for CD2 (58,64) and VCAM-1 (59), it is well exposed at the top of these molecules, making it accessible to ligands on the opposing cell surface.
The variable IgSF domains in B-and T-cell antigen receptors are capable of binding to an enormous variety of structures. However, antigen recognition involves loop regions between the ␤-sheets, which are known to display considerable structural diversity. In contrast, the ␤-sheets show far less structural diversity. Indeed, the central portion of each ␤-sheet (comprising the B, C, E, and F ␤-strands) forms the structurally conserved core of the IgSF fold (60,61). The observation that GFC(CЈCЉ) ␤-sheets bind to structures as diverse as integrins, IgSF molecules, and sialoglycoconjugates provides impressive evidence of the versatility of IgSF domains (2).