Binding Specificities of the Sialoadhesin Family of I-type Lectins

The carbohydrate binding specificities of three sialoadhesins, a subgroup of I-type lectins (immunoglobulin superfamily lectins), were compared by measuring lectin-transfected COS cell adhesion to natural and synthetic gangliosides. The neural sialoadhesins, myelin-associated glycoprotein (MAG) and Schwann cell myelin protein (SMP), had similar and stringent binding specificities. Each required an α2,3-linked sialic acid on the terminal galactose of a neutral saccharide core, and they shared the following rank-order potency of binding: GQ1bα ≫ GD1a = GT1b ≫ GM3 = GM4 ≫ GM1, GD1b, GD3, GQ1b(nonbinders). In contrast, sialoadhesin had less exacting specificity, binding to gangliosides that bear either terminal α2,3- or α2,8-linked sialic acids with the following rank-order potency of binding: GQ1bα > GD1a = GD1b = GT1b = GM3 = GM4 > GD3= GQ1b ≫ GM1 (nonbinder). CD22 did not bind to any ganglioside tested. Binding of MAG, SMP, and sialoadhesin was abrogated by chemical modification of either the sialic acid carboxylic acid group or glycerol side chain on a target ganglioside. Synthetic ganglioside GM3 derivatives further distinguished lectin binding specificities. Deoxy and/or methoxy derivatives of the 4-, 7-, 8-, or 9-position of sialic acid attenuated or eliminated binding of MAG, as did replacement of the sialic acid acetamido group with a hydroxyl. In contrast, the 4- and 7-deoxysialic acid derivatives supported sialoadhesin binding at near control levels (the other derivatives did not support binding). These data are consistent with sialoadhesin binding to one face of the sialic acid moiety, whereas MAG (and SMP) may have more complex binding sites or may bind sialic acids only in the context of more restricted oligosaccharide conformations.

Sialoadhesins (1) are a structurally and functionally related family consisting of five immunoglobulin superfamily lectins (I-type lectins) (2) including myelin-associated glycoprotein (MAG), 1 Schwann cell myelin protein (SMP), CD22, CD33, and sialoadhesin. MAG and SMP are found on oligodendroglia and Schwann cells in the nervous system (3,4), CD22 is expressed on a subset of B lymphocytes, sialoadhesin on a subset of macrophages, and CD33 on cells of myelomonocytic lineage. Sialoadhesins have been proposed to mediate cell-cell recognition, perhaps via their carbohydrate binding activities (5)(6)(7). Each sialoadhesin family member has two or more Ig-like domains: an amino-terminal V-set domain followed by one or more (up to 16) C2-set domains (8). Domain deletion and sitedirected mutagenesis of sialoadhesin and CD22 localize their carbohydrate-binding sites to the amino-terminal V-set domain, with contributions (for CD22) from the adjoining C2-set domain. These first two domains share very high amino acid sequence similarity between MAG and SMP (Ͼ70%) and significant similarity across all I-type lectins (Ͼ30% in pairwise comparisons) (2,8,9).
Each I-type lectin binds to carbohydrate structures bearing a nonreducing terminal sialic acid (1,6,10). Sialic acids are a common nonreducing terminus of vertebrate glycoconjugates and appear to play uniquely important roles in recognition phenomena. Because sialic acids may be linked to Gal, GalNAc, or other sialic acid residues at various positions and because they may carry different substituents on their 9-carbon base structure, the sialic acids represent a diverse family of carbohydrate determinants (11). In certain sialic acid-dependent recognition systems, determinant stringency is low. For example, selectins bind to oligosaccharides bearing truncated sialic acids (12) or appropriately placed anionic groups (sulfates, carboxylic acids) otherwise unrelated to the sialic acid structure (13)(14)(15)(16). In contrast, sialoadhesins appear to have more stringent sialic acid specificities (see "Discussion") (9). In this study, we used cells expressing different sialoadhesins to explore and compare the fine structural preferences of their binding to target sialylated glycoconjugates.

EXPERIMENTAL PROCEDURES
Gangliosides-The ganglioside structures used in this study are shown schematically in Fig. 3. Purified bovine brain G M1 , G D1a , G D1b , G D3 , and G T1b were from EY Laboratories (San Mateo, CA) or Matreya, Inc. (Pleasant Gap, PA), and G Q1b was from Accurate Chemical & * This work was supported by National Science Foundation Grant IBN-9631745, and by grants from the National Multiple Sclerosis Society and the Medical Research Council (to M. B. T. and J. C. R.). 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.
Plasmids were propagated in Escherichia coli MC1061/p3 and purified by polyethylene glycol precipitation. COS-1 cells, routinely maintained in 10% fetal calf serum in Dulbecco's modified Eagle's medium at 37°C in a humidified atmosphere of 90% air and 10% CO 2 , were transiently transfected with lectin-expressing plasmids via a high efficiency procedure (using 40 g/ml DEAE-dextran) (25). Transfected cells were returned to culture for 40 -50 h to allow lectin expression to proceed and then were detached from plates for adhesion experiments (see below). Lectin expression was confirmed by flow cytometry and/or immunocytochemistry using the following monoclonal antibodies: mAb 513 (MAG/SMP cross-reactive) (4, 7), SER-4 (sialoadhesin) (26), and Chemicon 2112 (CD22; Chemicon International, Inc., Temecula, CA).
Microplate Cell Adhesion to Adsorbed Glycolipids-Adhesion was performed as reported previously (20,22,27). Aliquots (50 l) of ethanol/water (1:1) containing phosphatidylcholine (0.5 M), cholesterol (2.0 M), and gangliosides (concentrations as indicated) were added to microwells (96-well Serocluster, Costar Corp., Cambridge, MA). Plates were incubated for 90 min uncovered at ambient temperature to allow partial evaporation and lipid adsorption (28,29), after which the wells were washed with water. Wells were preblocked by addition of 100 l/well Hepes-buffered Dulbecco's modified Eagle's medium containing 1.5 mg/ml bovine serum albumin. Plates were covered and incubated for 10 min at 37°C prior to cell addition (see below).
Transfected COS cells were harvested using hypertonic Ca 2ϩ /Mg 2ϩfree phosphate-buffered saline containing 1 mM EDTA as described (22), collected by centrifugation, and resuspended at 10 7 cells/ml in Dulbecco's phosphate-buffered saline containing 2 mg/ml bovine serum albumin. Transfected cells were pretreated with neuraminidase, which enhances cell adhesion without changing carbohydrate binding specificity (20), as follows. Aliquots of cells (500 l) were placed in 1.5-ml microcentrifuge tubes, and 10 milliunits of Vibrio cholerae neuraminidase (Calbiochem) were added. Suspensions were incubated for 1.5-2 h at 37°C with end-over-end mixing. Cells were collected by centrifugation, washed twice with Dulbecco's phosphate-buffered saline containing 2 mg/ml bovine serum albumin, and resuspended at 250,000 cells/ml in Hepes-buffered Dulbecco's modified Eagle's medium containing 1.5 mg/ml bovine serum albumin. Cell viability was determined by trypan blue exclusion on representative transfected cells. Prior to pretreatment, cells were 84% viable. After neuraminidase or control pretreatment, viability ranged from 81 to 85%, essentially unchanged from the freshly collected cells. Quantitation of cell adhesion was via an enzyme assay (see below) that measured only viable cells.
Aliquots of the cell suspension (200 l) were added to preblocked, lipid-adsorbed microwells and incubated at 4°C for 10 min to allow the cells to settle and then at 37°C for 45 min. To gently remove nonadherent cells after the incubations, plates were immersed in phosphatebuffered saline, inverted, and placed in an immersed Plexiglas box that was sealed with a gasket to exclude air (27). The inverted plate in its fluid-filled chamber was placed in a centrifuge carrier and centrifuged at 110 ϫ g. The box was again immersed in phosphate-buffered saline; the plate was removed and righted (while immersed); and excess surface buffer was removed by aspiration, leaving 300 l/well. Adherent cells were lysed by addition of 20 l of 10% Triton X-100 to each well, and 80 l were removed to a fresh 96-well plate for quantitation. Cell adhesion was quantitated by measuring lactate dehydrogenase activity in the cell lysate after addition of 120 l of phosphate-buffered saline containing 0.7 mM NADH and 4.7 mM pyruvate. The decrease in absorbance at 340 nm as a function of time was measured simultaneously in each well using a Molecular Devices UV multiwell kinetic plate reader. This method is amenable to testing large numbers of samples. The data presented are compiled from Ϸ4000 individual data points and are presented as the mean Ϯ S.E. of the mean for 3-103 replicate determinations. Where indicated, the statistical significance of adhesion to ganglioside-adsorbed surfaces compared with control surfaces (adsorbed with phosphatidylcholine and cholesterol, but no ganglioside) was determined using a two-tailed Student's t test.

Ganglioside Binding Specificities of Sialoadhesins-MAG-,
SMP-, and sialoadhesin-transfected COS cells bound specifically to ganglioside-adsorbed surfaces (Figs. 1-3). Adhesion to the most potent target gangliosides was typically very high (Ͼ80% of the cells added), whereas background adhesion to surfaces adsorbed with phosphatidylcholine and cholesterol without ganglioside was low. COS cells transfected with CD22 failed to adhere to any ganglioside tested (G D1a , G D1b , G D3 , G T1b , G Q1b , and G Q1b␣ ). COS cells transfected with either of the FIG. 1. Adhesion of COS cells expressing sialoadhesins to adsorbed gangliosides. COS cells transiently transfected to express MAG (A), SMP (B), or sialoadhesin (C) were collected from culture dishes; pretreated with neuraminidase to enhance adhesion; and placed in microwells previously adsorbed with phosphatidylcholine, cholesterol, and the indicated gangliosides. After incubation, nonadherent cells were removed by centrifugation, and adherent cells were quantitated enzymatically (see "Experimental Procedures"). Adhesion is expressed relative to the total number of cells added to each well and represents the mean Ϯ S.E. of 3-103 replicate determinations. Background adhesion, represented by a horizontal line in each panel, was determined on wells adsorbed with phosphatidylcholine and cholesterol without gangliosides (7.5 Ϯ 0.6% for MAG, 11.0 Ϯ 0.8% for SMP, and 5.9 Ϯ 0.6% for sialoadhesin).
two splice variants of MAG (L-MAG and S-MAG) demonstrated the same extent and specificity of adhesion to a representative set of ganglioside-adsorbed surfaces (G M1 , G D1a , G D1b , G T1b , and G Q1b␣ ) (data not shown). Therefore, L-MAG-transfected COS cells were used in subsequent experiments, and all data presented on MAG-mediated adhesion refer to the long splice variant.
The two neural sialoadhesins, MAG and SMP, had similar ganglioside binding specificities (Figs. 1-3). The abundant brain gangliosides G D1a (at Ն12.5 pmol/well) and G T1b (at Ն25 pmol/well) supported highly significant adhesion (p Ͻ 0.0002) of both MAG-and SMP-transfected COS cells (Fig. 1, A and B). Other gangliosides including G M3 and G M4 also supported significant adhesion of both lectins, although only at Ն10-fold higher ganglioside concentrations compared with G D1a . In contrast, neither MAG nor SMP bound to G M1 , G D1b , or G D3 , indicating that both lectins require a terminal ␣2,3-linked sialic acid. All gangliosides that supported statistically significant adhesion of SMP contained the NeuAc␣2,3Gal terminal structure (see Fig. 3), whereas all nonsupportive gangliosides lacked this terminal structure. MAG supported adhesion to the same gangliosides, although typically with higher efficiency (greater number of adherent cells). This may be due to more efficient transfection with the MAG plasmid, higher expression of the transfected MAG, and/or more effective ganglioside binding by MAG. Flow cytometry using a MAG/SMP cross-reactive antibody (mAb 513) indicated that more MAG-transfected cells (48.2%) expressed the highest level of lectin compared with SMP-transfected cells (28.3%). Within these highest expressing populations, the mean fluorescence intensities were similar (496 and 441 relative units for MAG and SMP, respectively).
In addition to gangliosides bearing the NeuAc␣2,3Gal terminus, G Q1b (which bears only ␣2,8-linked sialic acid termini) supported a low amount of adhesion by MAG-transfected cells. This preparation of G Q1b , however, was contaminated with a small amount of G T1b (20). We conclude that MAG and SMP bind with similar rank-order potency to gangliosides terminated with NeuAc␣2,3Gal (see Fig. 3).
In contrast to MAG and SMP, sialoadhesin had a distinctly broader binding specificity. Several gangliosides with terminal NeuAc␣2,3Gal structures (G D1a , G T1b , G M3 , and G M4 ) as well as G D1b (which bears only a terminal NeuAc␣2,8NeuAc structure) supported nearly equivalent sialoadhesin-mediated adhesion (Fig. 1C). G D3 and G Q1b , which also bear only NeuAc␣2,8NeuAc termini, supported sialoadhesin binding with moderate potency. Binding was structurally specific in that G M1 did not support sialoadhesin-mediated adhesion.
Prior studies indicated that MAG bound with markedly high affinity to one of the "Chol-1" gangliosides (22). These minor brain gangliosides bear a sialic acid linked ␣2,6 to the GalNAc(III) of the gangliotetraose core (structures in Fig. 3) (30). Fig. 2 presents a comparison of adhesion of MAG-, SMP-, and sialoadhesin-transfected COS cells to synthetic Chol-1 and related gangliosides. MAG and SMP again had markedly similar binding specificities (Fig. 2, A and B). G T1␤ was equipotent to G T1b in supporting MAG and SMP binding, whereas G Q1b␣ was 10-fold more potent. G M1␣ , which contains a single ␣2,6linked sialic acid, failed to support adhesion of either lectin. Therefore, the terminal NeuAc␣2,3Gal structure is required for both SMP-and MAG-mediated cell adhesion, and additional sialic acids on the internal GalNAc(III) and Gal(II) of the gangliotetraose core enhance binding of MAG and SMP to a similar extent. In contrast, G Q1b␣ was only modestly (Ͻ3-fold) more potent than G T1b in supporting sialoadhesin-mediated adhesion. Binding potencies for all gangliosides tested using MAG-, SMP-, and sialoadhesin-mediated cell adhesion are summarized in Fig. 3.
The MAG/SMP cross-reactive antibody mAb 513 (4, 7), shown previously to block MAG binding to neurons (31) and gangliosides (22), demonstrated the carbohydrate-binding site structural similarity between MAG and SMP and their difference from sialoadhesin. As shown in Fig. 4, mAb 513 eliminated or markedly reduced binding of MAG and SMP to G T1b , whereas binding of sialoadhesin was unaffected. The anti-sialoadhesin blocking mAb 3D6 (32) inhibited binding of sialoadhesin to G T1b (data not shown).
Sialic Acid Substructure Binding Specificities of I-type Lectins-Sialic acid is a complex monosaccharide, with a carboxylic acid, an N-acyl group, and a glycerol side chain within its structure (see Fig. 7). Chemically modified and synthetic gangliosides were used to determine which sialic acid substituent groups are required for binding by sialoadhesin family members.
Since G D1a supports highly significant adhesion of MAG, SMP, and sialoadhesin (Fig. 1), it was used as a basis for testing sialic acid chemical modifications. G D1a was selectively oxidized with periodate under conditions that cleave exclusively between C-7-C-8 and C-8 -C-9 on the sialic acid glycerol side chain. Mass spectrometry indicated equal conversion of G D1a sialic acids to their corresponding 7-and 8-carbon aldehydes (data not shown). A portion of the resulting G D1a aldehydes was reduced with sodium borohydride, resulting in conversion to the corresponding 7-and 8-carbon alcohols. As shown in Fig. 5, neither the 7/8-aldehyde nor 7/8-alcohol sialic acid derivatives of G D1a supported binding of any of the I-type lectins tested. Similarly, modifications of the carboxylic acids on G D1a abrogated binding. Conversion of both sialic acids on G D1a to the corresponding 1-ethyl esters, 1-amides, or 1-alcohols completely eliminated binding of MAG-, SMP-, and sialoadhesin-transfected COS cells (Fig. 5). The structures of all  and dark stippled bars), in medium supplemented with 20 g/ml mAb 513 (filled bars), or in medium supplemented with control isotype-matched mAb (light stippled bars). Cell suspensions were incubated for 1 h at 0°C to allow antibody binding, and then aliquots (300 l) were added to preblocked microwells that had been adsorbed with phosphatidylcholine and cholesterol alone (control; open bars) or with ganglioside G T1b (stippled and closed bars). Cells were incubated, and adhesion was quantitated as described under "Experimental Procedures." Adhesion is expressed relative to the total number of cells added to each well and represents the mean Ϯ S.D. of duplicate or triplicate determinations.
Since G M3 and G M4 (bearing a terminal N-acetylneuraminic acid) supported substantial adhesion mediated by both sialoadhesin and MAG, a series of synthetic analogs based on these structures (19) was used to determine the role of each sialic acid hydroxyl group and the sialic acid N-acyl group on adhesion (binding of SMP to G M3 and G M4 was insufficient to allow valid comparisons). Consistent with chemical modification studies, the 8-deoxy and 9-methoxy forms of G M3 failed to support adhesion mediated by either MAG or sialoadhesin (Fig. 6). In contrast, the 4-deoxy and 7-deoxy forms of G M3 were comparable to G M3 in supporting sialoadhesin-mediated adhesion, but failed to support substantial MAG-mediated adhesion. Furthermore, the sialic acid acetamido group appears to be involved in lectin binding. G M4 supported sialoadhesin and MAG binding, whereas a derivative bearing a 5-deaminated analog of neuraminic acid (KDN-G M4 ) failed to support binding by either lectin (Fig. 6). These data are consistent with the prior published observations that glycoconjugates bearing N-glycolylneuraminic acid fail to support MAG (20) or sialoadhesin (33) binding.

DISCUSSION
Sialoadhesins (1,8,9) are a functionally and structurally related subfamily of carbohydrate-binding immunoglobulin superfamily members (I-type lectins) (2). The sialoadhesin family consists of the eponymous member (sialoadhesin), MAG, SMP, CD22, and CD33. MAG and SMP are expressed on myelinating cells in the nervous system, sialoadhesin on a subset of macrophages, CD22 on certain B lymphocytes, and CD33 on cells of myelomonocytic lineage (9). Sialoadhesins mediate cell-cell interactions by binding to target sialylated glycoconjugates (1,6,10,32). They share the same general polypeptide domain structure: an amino-terminal V-set Ig-like domain followed by one or more C2-set Ig-like domains, a transmembrane domain, and a short cytoplasmic tail. The ligand recognition site has been localized to the amino-terminal V-set domain (sialoadhesin) (34) or the V-set domain with contributions from the adjacent C2-set domain (CD22) (34,35). Additionally, sialoadhesins have extensive sequence similarity. The first two amino-terminal Ig-like domains of MAG and SMP are 56% identical (72% similar, including conservative amino acid replacements), and other sialoadhesins range from 32 to 43% sequence similarity in pairwise comparisons. Site-directed mutagenesis (36,37) indicates that sialoadhesin and CD22 bind to sialylated glycoconjugates via amino acids on one surface of the V-set domain. This is consistent with sialoadhesin's sialic acid substituent group binding specificity (see below), whereas MAG's specificity indicates a more complex binding site.
Our prior studies demonstrated that (i) MAG bound to gangliosides with the specificity G Q1b␣ Ͼ G T1b ϭ G D1a Ͼ G M3 Ͼ Ͼ G M1 , G D1b , G Q1b , the latter of which did not support adhesion; and (ii) modification of the glycerol side chain, carboxylic acid, or N-acyl group abrogated MAG-mediated adhesion (20,22). This study confirms and extends those findings. MAG-medi- ated adhesion was repeated to the above gangliosides as well as to G M4 , KDN-G M4 , six synthetic derivatives of G M3 , and various gangliosides with ␣2,6-sialic acids linked to the GalNAc(III) of the gangliotetraose core. These new data were compared directly with adhesion of COS cells expressing CD22, sialoadhesin, SMP, and the short isoform of MAG.
Consistent with their extensive sequence similarity, the two neural sialoadhesins, SMP and MAG, were remarkably similar in their ganglioside binding (Figs. 1-3). Both bound only to structures bearing terminal ␣2,3-linked sialic acids (e.g. G T1b and G D1a ) and failed to bind to those terminated with ␣2,8linked structures (e.g. G D1b and G D3 ). Among glycoconjugates with ␣2,3-linked sialic acid termini, SMP and MAG distinguished sialic acid linkage patterns and neutral core variations. In contrast to sialoadhesin, di-and trisialogangliosides with the gangliotetraose core (G D1a and G T1b ) supported adhesion of SMP and MAG Ϸ10-fold better than did monosialogangliosides (G M3 and G M4 ). Furthermore, the Chol-1 ganglioside, G Q1b␣ , was 10-fold more potent than any other ganglioside tested (Figs. 1-3). Chol-1 gangliosides are quantitatively minor structures that are expressed exclusively on cholinergic neurons (30,41). The functional significance of their preferential binding to the neural sialoadhesins is not known. Although the terminal tetrasaccharide on G Q1b␣ is also found on O-linked glycoproteins (42,43), polyclonal antibodies against Chol-1 gangliosides do not cross-react with any glycoprotein (44), suggesting that the oligosaccharide on G Q1b␣ adopts a unique conformation that fits particularly well in the SMP and MAG binding pockets. In addition to having similar carbohydrate recognition specificities, the observation that both SMP-and MAG-mediated adhesion to gangliosides is inhibited by the same conformationally restricted monoclonal antibody (mAb 513) (45) confirms the similarity of their binding sites.
Sialic acids are unusual among monosaccharides in their complexity and diversity (11). They carry a carboxylic acid (C-1), an N-acyl group attached to C-5, and a glycerol side chain attached to C-6 ( Fig. 7), each of which is involved in molecular recognition by certain sialoadhesins. Blocking the carboxylic acid abrogates binding (Fig. 5), as does replacement of the acetamido group with a hydroxyl (compare G M4 with KDN-G M4 in Fig. 6) or truncation of the glycerol side chain (Fig. 5). These data are consistent with prior studies on the sensitivity of sialoadhesin and CD22 binding to modifications of the sialic acid residue (6, 33, 46 -48) and contrast with studies on selectins, in which extensive modifications of sialic acids have no effect (12,14,49). In fact, substitution of the entire sialic acid (e.g. on sialyl-Le X or sialyl-Le a ) with a sulfate ester results in retention of ligand binding by all selectins (13,14), but abrogates binding by CD22 (47).
The sialic acid substructural binding specificities of sialoadhesin and MAG have implications for ligand docking on the proteins. For sialoadhesin, modification of the C-8 or C-9 hydroxyl, the acetamido nitrogen or methyl group (33), or the C-1 carboxylic acid eliminated binding (Figs. 5 and 6), whereas removal of the C-4 or C-7 hydroxyl was without effect. This pattern is consistent with binding primarily to a single face of the sialic acid (top face in Fig. 7). Sialic acid binding to sialoadhesin can be compared with x-ray crystallography of sialic acid binding to the influenza virus hemagglutinin (50), in which a carboxylate oxygen, the acetamido nitrogen, and the 8-and 9hydroxyls face into a depression on the hemagglutinin surface, whereas the 7-hydroxyl faces the solvent. This model is consistent with Ig-domain studies and site-directed mutagenesis (34,37), which place the ligand-binding site of sialoadhesin on a contiguous cluster of residues on the surface of the GFCCЈCЉ ␤-sheet of the V-set Ig-like domain.
Sialic acid modifications that block sialoadhesin binding also block MAG binding. In addition, removal of either the 4-or 7-hydroxyl inhibits MAG binding (Fig. 6). Since the 7-hydroxyl and 8/9-hydroxyls extend in opposite directions (Fig. 7), a more complex model of MAG binding is implicated. One possibility is that the MAG binding site consists of a deep pocket or apposing polypeptide sheets. Alternatively, the 7-hydroxyl group may stabilize a conformation of the oligosaccharide that is preferentially bound by MAG at a single protein surface. To date, no direct evidence addresses whether one or more than one protein surface on MAG is responsible for sialic acid binding, although biophysical and electron microscopic studies suggest that MAG may have a bent rod configuration with apposed Ig-like domains (51,52). Studies using chimeric molecules indicate that the first three Ig domains of MAG are necessary and sufficient for binding to neurons (45) and sialoglycoconjugates (1), although the sialic acid substructure specificities of truncated forms of MAG have not been reported. Further protein structural and functional studies will be needed to establish the sialoglycoconjugate-binding site on MAG (and on SMP) and to determine the precise role each sialic acid hydroxyl group plays in protein binding.