In Situ trans Ligands of CD22 Identified by Glycan-Protein Photocross-linking-enabled Proteomics*

CD22, a regulator of B-cell signaling, is a siglec that recognizes the sequence NeuAcα2–6Gal on glycoprotein glycans as ligands. CD22 interactions with glycoproteins on the same cell (in cis) and apposing cells (in trans) modulate its activity in B-cell receptor signaling. Although CD22 predominantly recognizes neighboring CD22 molecules as cis ligands on B-cells, little is known about the trans ligands on apposing cells. We conducted a proteomics scale study to identify candidate trans ligands of CD22 on B-cells by UV photocross-linking CD22-Fc chimera bound to B-cell glycoproteins engineered to carry sialic acids with a 9-aryl azide moiety. Using mass spectrometry-based quantitative proteomics to analyze the cross-linked products, 27 glycoproteins were identified as candidate trans ligands. Next, CD22 expressed on the surface of one cell was photocross-linked to glycoproteins on apposing B-cells followed by immunochemical analysis of the products with antibodies to the candidate ligands. Of the many candidate ligands, only the B-cell receptor IgM was found to be a major in situ trans ligand of CD22 that is selectively redistributed to the site of cell contact upon interaction with CD22 on the apposing cell.

Glycan-binding proteins (GBPs) 1 mediate diverse aspects of cell communication through their interactions with their counter-receptors comprising glycan ligands carried on cell surface glycoproteins and glycolipids. Identification of the in situ counter-receptors of glycan-binding proteins is problem-atic due to the fact that the vast majority of the glycoproteins of a cell will carry highly related glycan structures because they share the same secretory pathway that elaborates their glycans post-translationally en route to the cell surface. Thus, although many glycoproteins will carry the glycan structure recognized by a GBP, the challenge is to determine whether one, several, or all of these cell surface glycoproteins (and glycolipids) are recognized in situ as physiologically relevant counter-receptors (1)(2)(3)(4). Standard in vitro methods, such as co-precipitation from cell lysates or Western blotting using binding protein probes, are useful for identifying glycoproteins that contain the glycan structure recognized by the GBP. However, these may not be relevant ligands in situ due to constraints imposed by their microdomain localization and the geometric arrangement of their glycans relative to the GBP presented on the apposing cell.
In this report, we examine the in situ ligands of CD22 (Siglec-2), a member of the siglec family and a regulator of B-cell receptor (BCR) signaling that recognizes glycans containing the sequence NeuAc␣2-6Gal as ligands (2,5,6). Regulation of BCR signaling by CD22 is effected by its proximity to the BCR through recruitment of a tyrosine phosphatase, SHP-1, which is in turn influenced by CD22 binding to its glycan ligands (6). Glycoproteins bearing CD22 ligands are abundantly expressed on B-cells and bind to CD22 in cis (on the same cell) (7), regulating BCR signaling (2,5,6). Although binding to cis ligands has been shown to "mask" CD22 from binding low avidity synthetic sialoside probes (2,7), CD22 can also interact with ligands on apposing immune cells in trans (8 -10). Interactions of CD22 with trans ligands influence T-cell signaling in vitro (11,12), mediate B-cell homing via binding to sinusoidal endothelial cells in the bone marrow (13), and aid in "self"-recognition (14). Thus, interactions with both cis and trans ligands modulate CD22 function in immune homeostasis.
Several groups have demonstrated that recombinant CD22-Fc chimera is capable of binding and precipitating the majority of glycoproteins from B-and T-cell lysates whose glycans contain the sequence NeuAc␣2-6Gal (15)(16)(17)(18). Among them, CD45, IgM, and CD22 itself were identified as specific B-cell binding partners and were postulated to have functional significance as in situ cis ligands of CD22 in regulation of BCR signaling (11, 16, 18 -20). Several reports have also documented in situ interactions of CD22 with IgM and CD45, but these interactions were found to be of low stoichiometry and sialic acid-independent (19 -21), leaving open the question of which glycoproteins served as in situ cis ligands of CD22 on B-cells that masked the glycan ligand binding site of CD22 (7). Subsequently, using metabolically labeled B-cells with sialic acids containing a photoactivatable 9-aryl azide moiety, we demonstrated that CD22 could be photocrosslinked to its cis ligands, effectively tagging the in situ cis ligands with CD22 (15). Notably, there was no cross-linking observed to IgM or CD45, demonstrating that they are not significant in situ cis ligands of CD22 (15). Instead, only glycans of neighboring CD22 molecules interacted significantly with CD22, resulting in photocross-linking of homomultimeric complexes of CD22. Thus, despite the fact that most B-cell glycoproteins are recognized in vitro, CD22 selectively recognizes glycans of neighboring CD22 molecules as cis ligands in situ.
With the perspective gained from analysis of cis ligands, we wished to determine whether CD22 was also selective in recognition of trans ligands upon cell contact. We have previously demonstrated that CD22 is redistributed to sites of cell contact of interacting B-cells and T-cells and that redistribution is mediated by the interaction of CD22 with sialic acidcontaining trans ligands on the apposing cell (8). Stamenkovic et al. (22) had previously demonstrated that binding of T-cells to CD22-expressing COS cells was blocked by an anti-CD45RO antibody, suggesting that CD45 was a functional trans ligand of CD22 on T-cells. However, we found that redistribution of CD22 to sites of cell contact was also observed with CD45-deficient B-cells (8), indicating that, at a minimum, other glycoproteins must also serve as trans ligands of CD22 on B-cells.
To assess whether CD22 recognizes all or a subset of glycoproteins as trans ligands on an apposing cell, we initiated an unbiased analysis of the trans ligands of CD22 on apposing B-cells using our protein-glycan cross-linking strategy (15). By cross-linking CD22-Fc to intact B-cells, we identified 27 candidate trans ligands of CD22 by quantitative mass spectrometry-based proteomics. We then looked at the in situ trans interactions of CD22 in the physiologically relevant cellular context by cross-linking CD22 expressed on one cell to the trans ligands with photoreactive sialic acids on the apposing cell. Our results indicate that only a subset of cell surface glycoproteins, including IgM and, to a lesser extent, CD45 and Basigin, are selectively recognized in trans by CD22. Indeed, IgM in particular is a preferred trans ligand that is selectively redistributed to the sites of cell contact on apposing B-cells in a CD22-and sialic acid-dependent manner despite a vast excess of cell surface glycoproteins that carry a glycan recognized by CD22. The results support the view that factors other than glycan sequence are critical for the in situ engagement of glycan-binding proteins with glycan ligand bearing counter-receptors on the same cell (in cis) or apposing cell (in trans).

EXPERIMENTAL PROCEDURES
Cells-CHO-FlpIn cells and stably transfected CHO-FlpIn cells expressing human CD22 with a C-terminal V5 tag (23) were maintained in Dulbecco's modified Eagle's medium/F-12 medium supplemented with 10% fetal bovine serum and either 100 g/ml Zeocin (Invitrogen) or 500 g/ml hygromycin B (Roche Applied Science), respectively. CD22-Fc was purified from culture supernatants of COS-1 cells by Protein A affinity chromatography 3-7 days after transfection with CD22-Fc-pCDM8 using Lipofectamine TM 2000 (24). BJA-B K20 cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum. Human lymphocytes were enriched from human blood with Ficoll-Hypaque density gradient centrifugation. Human and murine B-cells were enriched from blood and spleen, respectively, using B-cell negative isolation kits (Dynal, Invitrogen).
Photoaffinity Cross-linking and Cell Lysis-9AAz NeuAc-K20 B-cells or human lymphocytes (1 ϫ 10 7 cells/ml) were incubated with 20 g/ml CD22-Fc in PBS on ice for 30 min and then exposed to a hand-held 254 nm UV source for 30 min. Human IgG (Fc fragment) was used as a control. In the case of photocross-linking to CD22 expressed on CHO cells, 9AAz NeuAc-K20 cells resuspended in PBS were overlaid onto CHO cell monolayers in 6-well plates at 1 ϫ 10 7 cells/ml and cross-linked as described above. Unbound CD22-Fc or IgG (Fc fragment) and K20 cells, respectively, were washed away with PBS and/or 0.2 mM glycine in PBS, pH 2.2. Cells were lysed at 1 ϫ 10 7 cells/ml in 50 mM Tris, pH 8.0, 150 mM NaCl, 5 mM EDTA, 1% Nonidet P-40 with a protease mixture (Calbiochem).
Immunoprecipitation and Western Analysis-Lysates were incubated overnight with Protein G-agarose (Invitrogen) or anti-V5-agarose (Sigma) at 4°C with end-over-end rotation. The immunoprecipitates were washed four times with lysis buffer and eluted with 4ϫ lithium dodecyl sulfate buffer (Invitrogen). Immunoprecipitates were resolved on 10 or 4 -12% NuPAGE gels (Invitrogen), transferred to nitrocellulose, and blocked for 1 h with 5% milk in Tris-buffered saline with 0.05% Tween 20 (TTBS). Blots were incubated with antibodies diluted in 5% milk, TTBS; washed with TTBS; and visualized with chemiluminescence.
One-dimensional Gel Electrophoresis and Mass Spectrometry (MS/ MS)-Protein G immunoprecipitates from lysates of 9AAz NeuAc-K20 cells incubated with CD22-Fc with or without photocross-linking were resolved by gel electrophoresis and stained with EZ-Blue (Sigma). The high molecular mass region was excised into the following slices: band 1, greater than ϳ385 kDa; band 2, ϳ385-165 kDa; band 3, ϳ165-95; and band 4, ϳ95-70 kDa. Gel slices were subjected to in-gel digestion as reported previously (27). Peptides from in-gel trypsin digests were pressure-loaded onto a 100-m (inner diameter) fused silica capillary column containing 10 cm of C 18 resin. Peptides were eluted from the column using a 2-h gradient with a flow rate of 0.25 l/min directly into an LTQ ion trap mass spectrometer (Thermo Fisher Scientific). The LTQ was operated in data-dependent scanning mode with one full MS scan followed by seven MS/MS scans of the most abundant ions with dynamic exclusion enabled. Raw MS/MS data were searched using the SEQUEST TM algorithm (28) using a concatenated target/decoy variant of the human IPI database (version 3.33, released on September 13, 2007). No enzyme specificity was stated, and a static modification of ϩ57 Da was specified on cysteine to account for iodoacetamide alkylation. Mass tolerance for precursor ions and fragment ions was set to 1.00 Da. SEQUEST data from each band were filtered and sorted with DTASelect version 1.9 requiring a minimum ⌬CN of 0.8 and XCorr values of 1.8, 2.5, and 3.5 for 1ϩ, 2ϩ, and 3ϩ charge states. The false positive rates estimated by the program from the number and quality of spectral matches to the decoy database were less than 1.0%. When combining peptides into "proteins identified," all isoforms/individual members of a protein family supported by the data were considered, and the entries with the most annotation are reported here.

Spectral Count Analysis of One-dimensional Gel Electrophoresis-MS/MS Data for Identification of trans Ligand
Candidates-Two replicate data sets, each containing a ϩUV sample and a ϪUV sample, were subjected to spectral count analysis. Spectral count values of peptides derived from proteins from gel slices corresponding to molecular mass Ͼ95 kDa (gel band 1, Ͼ385 kDa; gel band 2, ϳ385-165 kDa; and gel band 3, ϳ165-95 kDa) were analyzed in both ϪUV and ϩUV samples. A protein hit was considered to be a potential CD22 trans ligand if its peptides were represented by higher spectral count values in the gel bands from the ϩUV sample as compared with those from the ϪUV sample. To identify potential CD22 trans ligands, a "normalized trans ligand score" was calculated for every protein hit as follows. First, the difference between the summed spectral counts (across bands 1-3) in the ϩUV sample and the ϪUV sample was calculated. Next, the spectral counts (across bands 1-3) of both the ϩUV and ϪUV samples were summed to obtain the total spectral counts. The normalized trans ligand score was then calculated as the ratio of the difference to the total spectral counts. Following analysis of spectral count values at the peptide level, the normalized trans ligand scores of various peptides corresponding to a protein ID were averaged to obtain a normalized trans ligand score for that protein.
The mean and S.D. of the trans ligand scores of the two replicate data sets were also calculated. Data were filtered to remove protein IDs that were not represented in both replicate data sets and proteins that did not have Ͼ4 spectral counts in at least one replicate. Environmental contaminants were detected by high spectral counts across not only gel bands 1-3 but also band 4 (ϳ95-70 kDa) and control bands corresponding to lower molecular masses and were removed. Mean trans ligand scores of the remaining protein hits varied from 1 to Ϫ1. Proteins with mean trans ligand scores Ͼ0.33 were considered CD22 ligand candidates. Standard deviations between the normalized trans ligand scores of replicate data sets were used to assess the statistical significance of the spectral count analysis. In addition, the standard deviation of the normalized trans ligand scores obtained for the various peptides corresponding to a protein using spectral count values at the peptide level was used to measure the statistical significance of the trans ligand score at the protein level. Several of the candidate ligands identified by spectral count analysis were independently verified by Western analysis.
Stable Isotope Labeling by Amino Acids in Cell Culture (SILAC)-MuDPIT-Stable, heavy isotopes of lysine ([ 13 C]Lys) and arginine ([ 13 C, 15 N]Arg) differing from their light amino acid counterparts by 6 and 10 Da, respectively, were used to enable extensive coverage of peptides following tryptic digestion. K20 cells were maintained in RPMI 1640 medium supplemented with dialyzed serum and light or heavy lysine and arginine (Invitrogen) for 2 weeks prior to performing the SILAC experiment. Four replicates (5 ϫ 10 8 cells) each of light and heavy K20 cells were desialylated by treatment with 50 milliunits/ml A. ureafaciens sialidase and sialylated with 9AAz NeuAc by enzymatic engineering. Cells were incubated with 37.5 g/ml CD22-Fc on ice for 30 min, and two replicates each of light and heavy K20 cells were exposed to UV irradiation for photoaffinity cross-linking. All samples were washed extensively and lysed. Lysates from samples of light K20 cells exposed to UV irradiation were mixed 1:1 with those of heavy K20 cells not exposed to UV irradiation, and lysates of heavy K20 cells exposed to UV irradiation were mixed 1:1 with those of light K20 cells not exposed to UV irradiation to yield two pairs of swapped replicates. The combined lysates were subjected to immunoprecipitation with rProtein G-agarose. Reduction, alkylation of cysteines, and protein digestion with trypsin were performed on-bead, and tryptic peptides were resolved by MuDPIT as described previously (29) using an LTQ-Orbitrap mass spectrometer (Thermo Fisher Scientific). Tandem mass spectra were analyzed using the following software analysis protocol. MS/MS spectra were searched with the ProLuCID algorithm (30) against the human IPI database (version 3.44, released on May 20, 2008 with 71,884 protein entries) concatenated to a decoy database in which the sequence for each entry in the original database was reversed (31). The search parameters were as follows: precursor ion mass tolerance was 4.5 Da, fragmentation ion mass tolerance was 0.6 Da, maximum number of missed cleavages was 2, and static cysteine modification was set to 57.02146. For the heavy search, arginine and lysine were set to a static modification of 10.00827 and 6.020127, respectively. No enzyme specificity was considered for the searches. All searches were parallelized and performed on a Beowulf computer cluster consisting of 100 1.2-GHz Athlon CPUs (32). The results were assembled and filtered using the DTASelect (version 2.0) program (33,34). DTASelect 2.0 uses a linear discriminant analysis to dynamically set XCorr and ⌬CN thresholds for the entire data set to achieve a user-specified false positive rate (less than 1% at the peptide level in this analysis). The false positive rates are estimated by the program from the number and quality of spectral matches to the decoy database. When combining peptides into proteins identified, all isoforms/individual members of a protein family supported by the data were considered, and the entries with the most annotation are reported here.
Analysis of SILAC-MuDPIT Data for trans Ligand Candidates-Four data sets were analyzed. Two replicate data sets contained proteins from a sample containing light cells photocross-linked to CD22-Fc (LϩUV) together with control heavy cells (HϪUV), and the remaining two replicate data sets contained swapped samples wherein heavy cells were photocross-linked to CD22-Fc (HϩUV) and mixed with control light cells (LϪUV). SILAC ratios were computed by the program Census using a linear least squares correlation algorithm (35). The ratio between the light versus heavy version of each peptide was calculated as the slope of the correlation line, whereas the closeness of fit was indicated as the correlation coefficient (r). Only peptide ratios with r 2 Ͼ 0.5 were accepted to remove poor quality data. The quantified proteins were further filtered to remove proteins that did not appear in more than one replicate data set. Depending on the sample (e.g. LϪUV mixed with HϩUV), reciprocals of SILAC (light/ heavy) ratios were calculated if required to obtain ϩUV/ϪUV ratios. Environmental contaminants were detected by high standard deviations between ϩUV/ϪUV ratios in swapped replicates and removed. Average ϩUV/ϪUV ratios of the remaining protein IDs ranged from 0.46 to 37.06. A higher cutoff of 2.17 (reciprocal of 0.46) was chosen to factor in the distribution of SILAC ratios. Protein IDs with ϩUV/ϪUV ratios Ͼ2.17 were considered CD22 trans ligand candidates. Standard deviations between ϩUV/ϪUV ratios at the peptide level and between replicate data sets at the protein level were used to assess the statistical significance of the measurements. Several of the candidate ligands identified were independently verified by Western analysis.

Incorporation of the Photocross-linker 9AAz
NeuAc onto Bcell Surface Glycoproteins-With the goal of identifying B-cell glycoproteins that could serve as trans ligands of CD22, we opted to use our approach of labeling total B-cell glycoproteins with 9AAz NeuAc, which does not impair CD22 binding to its ligands and allows efficient in situ photoaffinity crosslinking to CD22 (15). Then, following immunoprecipitation of the complex using CD22 as a tag, the ligands could be identified by proteomics and immunochemical techniques. Accordingly, to label cell surface glycoproteins of B-cells, 9AAz NeuAc was installed by either metabolic labeling (15,36) or enzymatic engineering with sialyltransferase, ST6GalI (25) (Fig. 1A). For optimal incorporation of 9AAz NeuAc, we used K20 cells cultured in serum-free medium because BJA-B K20 is a hyposialylated cell line that cannot synthesize its own sialic acids due to a deficiency in a key enzyme, UDP-GlcNAc-2Ј-epimerase, but will readily incorporate sialic acids added to the culture medium (37). Both metabolic incorporation and enzymatic engineering yielded sialylation profiles similar to that of "fully sialylated" cells cultured in medium supplemented with serum as detected with the sialic acid-specific lectin SNA (Fig. 1, B and C, and supplemental Fig. S1). trans Ligand Candidates of CD22-Fc on B-cells Identified by Proteomics-The ideal experiment for identifying in situ ligands of CD22 is to perform the UV cross-linking on a cell containing membrane-bound CD22 engaged in cell contact with 9AAz NeuAc-labeled B-cells. However, in preliminary experiments with this approach, although trans ligand crosslinking could be readily demonstrated by immunochemical techniques (see below), the amount of cross-linked complex that could be obtained was insufficient to identify the captured ligands by proteomics techniques. Therefore, we opted to first identify a subset of the glycoproteins as candidate ligands using intact 9AAz NeuAc-labeled B-cells as a source of in situ ligands and CD22-Fc chimera as the trans interacting receptor (Fig. 1D). In principle, the yield of cross-linked protein could be increased 10 -100-fold because B-cells could be prepared in large quantities in a single cell suspension, and CD22-Fc had access to the entire surface of the cell, not just the point of cell contact. To test this approach, CD22-Fc was UV cross-linked to 9AAz NeuAc-K20 cells. Uncross-linked CD22-Fc was then washed away by low pH, and cell lysates were subjected to immunoaffinity capture with Protein Gagarose to remove non-cross-linked proteins and high molecular weight cis cross-linked CD22 from 9AAz NeuAc-K20 cells. Western analysis yielded anti-CD22-and anti-Fc-reactive protein species of higher molecular mass than CD22-Fc that were not seen in the absence of UV exposure (Fig. 1E) or 9AAz NeuAc labeling (supplemental Fig. S2A). Similar results were obtained with 9AAz NeuAc-resialylated lymphocytes from human blood (supplemental Fig. S2B).
To identify the cross-linked ligands resolved as high molecular weight species, we used a mass spectrometry-based proteomics approach. Gel sections containing the high molecular weight region of the gel were subjected to in-gel trypsin digestion and nano-LC-MS/MS. Analysis of spectral count values of peptides identified a total of 24 proteins that were found in the high molecular weight region of the UV irradiation-exposed samples relative to the non-UV irradiation-exposed samples. Database analysis of these revealed that 18 were known glycoproteins, and six were annotated as predicted proteins or nuclear proteins. Only the known glycoproteins were considered candidate trans ligands in the subsequent analyses (Table I). Among them, CD22 was established as a trans ligand candidate based on spectral hits for C-terminal CD22 peptides (amino acid 327 and above) that are not found in the recombinant CD22-Fc protein.
As an alternative approach, we exploited SILAC to identify candidate trans ligand candidates cross-linked to the CD22-Fc chimera ( Fig. 2A) (38). SILAC experiments were performed by culturing the BJA-B K20 B-cell line in growth medium with light ([ 12 C]lysine and [ 12 C, 14 N]arginine) or heavy ([ 13 C]lysine and [ 13 C, 15 N]arginine) stable isotopes of lysine and arginine. Cell surface glycoproteins were then modified with 9AAz NeuAc by enzymatic engineering and allowed to interact with CD22-Fc chimera. Cells differentially labeled with light or heavy amino acids were either exposed to UV irradiation to induce photocross-linking or left untreated, subjected to extensive washing, and lysed. Lysates were then combined and subjected to immunoaffinity enrichment and on-bead trypsin digestion, peptides were analyzed by MuDPIT-MS/ MS, and the ratios of peak intensities of light and heavy peptides in the mass spectra were used to identify proteins more abundantly co-immunoprecipitated with CD22-Fc upon photocross-linking. A total of 19 proteins with SILAC-derived ϩUV/ϪUV ratios greater than 2.2 were identified as candidate trans ligands, all of which were annotated as membrane glycoproteins in the UniProtKB/Swiss-Prot database (Fig. 2B, Table  II, supplemental Table S1 and annexure to Table S1) (39).
In all, the two proteomics approaches yielded 27 glycoproteins as CD22 trans ligand candidates, 10 of which were identified in both approaches (supplemental Fig. S3A and Table S2). Functional analyses using the Swiss-Prot database (39) and the on-line DAVID knowledgebase (40) and Fatigo tools (41,42) indicated that a majority of the CD22 ligand candidates had a role in immune response, and all 27 hits were clustered by their functions in lymphocyte activation, signal transduction, cell adhesion, and transport (supplemental Fig. S3, B and C, and Table S3).
The candidate ligands identified by the proteomics approaches above were validated by immunochemical analysis of CD22-Fc photocross-linked to glycoproteins on 9AAz NeuAc-K20 cells. Antibodies suitable for Western analysis were obtained for eight of the 10 candidate glycoproteins identified by both approaches and 11 of the 19 identified by SILAC analysis (supplemental Table S2). Western analysis revealed that all were photocross-linked to CD22-Fc with the appearance of higher molecular mass form(s) corresponding to the sum of the molecular mass of one or more molecules of CD22-Fc and the glycoprotein ( Fig. 3 and supplemental Table S2). That cross-linking was CD22-specific was verified by performing control experiments in which 9AAz NeuAc-K20 cells were incubated with IgG (Fc fragment) protein instead of CD22-Fc and exposed to UV irradiation. No background was introduced in the Western analysis, confirming that cross-linking was CD22specfiic. Representative Western blots with antibodies to CD45, IgM, Basigin, 4F2, HLA I, HLA II, and MCAM are illustrated in Fig. 3.

In Situ CD22 trans Ligands on B-cells in Cell
Contacts-Armed with the identities of the membrane-bound B-cell glycoproteins that are photocross-linked to CD22-Fc, we next sought to determine whether they would also be detected as in situ trans ligands in the context of cell to cell contact. To examine this, we used a CHO cell line expressing CD22 with a C-terminal V5 tag for capture of glycoprotein ligands on the B-cell line BJA-B (subclone K20) labeled with 9AAz NeuAc (Fig.  4A). The 9AAz NeuAc-labeled or asialo K20 cells were overlaid onto CHO cells expressing CD22-V5, subjected to UV exposure, and washed with low pH to remove non-covalently trans Ligands of CD22

TABLE I Identification of CD22 trans ligand candidate proteins by spectral count analysis
Spectral count values of peptides released by trypsin digestion from gel slices corresponding to the molecular mass of photocross-linked CD22-Fc were analyzed in CD22-Fc-cross-linked (ϩUV) replicate samples as well as control (ϪUV) replicate samples. Normalized trans ligand scores (see "Experimental Procedures") were calculated from the spectral count values of the "ϩUV" and "ϪUV" samples for every peptide and averaged across all peptides corresponding to a protein ID to obtain the "trans ligand score" for that protein. The average of the "trans ligand scores" of replicates is reported as the "mean score." Proteins with mean scores Ͼ0.33 were considered candidate ligands. Listed in the table are details of candidate CD22 ligands identified by this procedure. CD22 itself was established as a trans ligand candidate based on analysis of spectral count values of C-terminal CD22 peptides (amino acids 327-847) that are not found in the recombinant CD22-Fc protein. PEPTIDE COUNT, number of unique peptides; % SEQ COVERAGE, percent sequence coverage; STDEV, standard deviation. bound cells. Extensive cross-linking of cells was observed only for the 9AAz NeuAc-engineered K20 cells overlaid on CD22-V5-CHO cells, whereas control experiments that were not subjected to UV treatment or that used K20 cells lacking 9AAz NeuAc or CHO cells lacking CD22-V5 expression showed few residual K20 cells bound to the CHO monolayer (Fig. 4B).
Western analysis of anti-V5 immunoprecipitates confirmed that UV irradiation resulted in formation of anti-CD22-reactive protein species of higher molecular mass than CD22-V5 (Fig.  4C). To identify which glycoproteins accounted for the crosslinked species, representing in situ trans ligands of CD22, anti-V5 immunoprecipitates were subjected to Western analysis with antibodies to 11 of the 27 glycoproteins identified as trans ligand candidates in the spectral count and SILAC analyses of the CD22-Fc-ligand complexes. Of these, only three, IgM, CD45, and Basigin, showed cross-linking to CD22 (Fig.  4D). Although we did not use CHO cells expressing other V5-tagged proteins as controls, no significant background was observed upon Western analysis of immunoprecipitates from controls comprising 9AAz NeuAc-K20 cells overlaid on CHO cells (not expressing CD22; Fig. 4D). Notably, crosslinking of CD22 to IgM was particularly robust with little crosslinking to CD45 and Basigin (Fig. 4D and supplemental Table S2). Moreover, no cross-linking of CD22 was detected to HLA I, HLA II, 4F2, Plexin-B2, Siglec-10, CD33, SLC1A5, or MCAM (supplemental Table S2) despite their robust cross-linking to CD22-Fc (Fig. 3) even though several of these glycoproteins are known to be as or more abundant than IgM on B-cells (43,44). The results show that the in situ recognition of trans ligands at the site of cell to cell contact is highly selective with IgM as the preferred trans ligand of CD22.

CD22-and Sialic Acid-dependent Redistribution of IgM to Sites of Cell Contact of Apposing B-cells-We
previously demonstrated that CD22 is redistributed to sites of cell contact between B-cells as a result of trans ligand interactions on the apposing cell (9). Given the robust cross-linking of IgM as a trans ligand of CD22 in cell to cell contacts, we investigated the localization of CD22 and IgM on interacting primary murine and human B-cells by immunofluorescence and microscopic analysis. As observed previously, CD22 was redistributed to sites of cell contact of interacting murine B-cells (Fig.  5A) and also was redistributed to the site of cell contact on primary human B-cells (supplemental Fig. S4). Interestingly, IgM was also localized preferentially at the sites of cell contact of interacting murine and human B-cells ( Fig. 5A and supplemental Fig. S4). This localization of IgM was also seen in samples stained with the Fab fragment of anti-IgM antibody and hence was not caused by antibody-mediated cross-linking of IgM on apposing cells. In contrast, CD45 was not significantly enriched at sites of cell contact (Fig. 5A). Similarly, there was diffuse staining with FITC-labeled S. nigra agglutinin, which detects all glycoproteins bearing the NeuAc␣2-6Gal sequence recognized by CD22 (Fig. 5A).
To assess the possibility that the IgM localization was a result of trans interactions with CD22, we compared localization of IgM on 300 pairs of contacting B-cells from wild type and CD22 knock-out mice (representative micrographs are shown in Fig. 5, A and B). Remarkably, IgM was recruited to sites of cell contact in 35 Ϯ 5% (average Ϯ S.E. of four independent experiments) of the interacting pairs of B-cells from wild type mice but in only 9 Ϯ 3% (average Ϯ S.E. of four independent experiments) of interacting B-cell pairs from CD22 KO mice (Fig. 5D) (p ϭ 0.0047 in a twotailed, unpaired Student's t test). CD45 and SNA localization remained unaltered (Fig. 5, compare A and B). The results strongly suggest that trans interactions of CD22 are responsible for IgM redistribution to the site of cell contact of interacting B-cells.
To investigate whether these trans interactions were mediated by CD22 binding to sialic acids on IgM, we also assessed CD22 and IgM localization on 300 pairs of contacting B-cells from ST6GalI knock-out mice that lack the FIG. 2. CD22 trans ligand candidates identified by SILAC analysis. A, scheme for identifying CD22 trans ligand candidates by SILAC analysis of samples subjected to photocross-linking, immunoaffinity isolation, on-bead trypsin digestion, and MuDPIT. B, spectral counts of protein hits plotted against average SILAC ratio upon UV cross-linking. Red and blue data points indicate protein hits considered as candidate trans ligands and non-ligands, respectively. IP, immunoprecipitation.
trans Ligands of CD22

SILAC (light/heavy) ratios were computed for all peptides identified after on-bead trypsin digestion and MuDPIT analysis of samples containing either light cells photocross-linked to CD22-Fc (LϩUV) and control heavy cells (HϪUV) or heavy cells photocross-linked to CD22-Fc (HϩUV) and control light cells (LϪUV)
. SILAC ratios of proteins, obtained by averaging the ratios of the constituent peptides, were converted to ϩUV/ϪUV ratios. Proteins with ϩUV/ϪUV ratios Ͼ2.17 were considered candidate ligands of CD22. Listed in the table are details of candidate CD22 ligands identified by this procedure. Standard deviations of the ϩUV/ϪUV ratios are not listed for single peptide spectrum-based protein identifications. PEPTIDE COUNT, number of unique peptides; %SEQ COVERAGE, percent sequence coverage; STDEV, standard deviation. sialyltransferase required for synthesis of the sialylated CD22 ligands (representative micrographs are shown in Fig.  5C). As expected, B-cells from ST6GalI knock-out mice were not stained by SNA, and CD22 was not redistributed to the sites of cell contact in interacting B-cells from ST6GalI knock-out mice (Fig. 5C). Notably, IgM was concentrated at sites of cell contact in only ϳ7 Ϯ 2% (average Ϯ S.E. of four independent experiments) of the interacting B-cell pairs from ST6GalI knock-out mice, 5-fold lower than contacting B-cell pairs from wild type mice (Fig. 5D) (p ϭ 0.0026 in a two-tailed, unpaired Student's t test). The results demonstrate that sialic acid-mediated trans interactions of CD22 are responsible for IgM redistribution to the site of contact of interacting B-cells.  4. trans ligands of CD22 on apposing B-cells. A, scheme for identifying CD22 trans ligands on apposing B-cells using an immunochemical screen. B, phase and fluorescence microscopy of 5-chloromethylfluorescein diacetate-stained asialo or 9AAz NeuAc-K20 cells (green) overlaid onto CHO or CD22-CHO cell monolayers with or without UV cross-linking and subjected to a low pH wash. C, anti-CD22 Western analysis of anti-V5 immunoprecipitates conducted on lysates from 9AAz NeuAc-K20 cells overlaid onto CHO or CD22-(V5)-CHO cells with or without UV cross-linking. The band corresponding to CD22 is marked with a blue arrow, and cross-linked bands are indicated in purple. An asterisk denotes the position of CD22 dimer. D, Western analysis of anti-V5 immunoprecipitates of 9AAz NeuAc-K20 cells overlaid onto CHO or CD22-(V5)-CHO cells with or without UV cross-linking. Red and purple arrows indicate uncross-linked and CD22-V5-cross-linked glycoproteins, respectively. Also seen in the anti-Basigin Western blot is the Ig heavy chain marked with a green arrow. Uncross-linked Basigin glycoforms range from ϳ33 to 50 kDa and are not appreciably seen in anti-V5 immunoprecipitates. IP, immunoprecipitation.

DISCUSSION
A major function ascribed to the glycan ligands of CD22 is to modulate the activity of CD22 in BCR signaling by influencing its proximity to the BCR complex (2,5,6,45). As a co-receptor of the BCR, CD22 regulates signaling by recruitment of the phosphatase SHP-1 and other Src homology 2 domain regulatory proteins via tyrosine motifs in its cytoplasmic domain where proximity to the BCR is required for maximal effect. Early support of this idea came from the observation that sequestration of CD22 with antibody-coated beads enhances activation through BCR ligation (46). Although IgM and CD22 are predominantly in distinct microdomains in resting murine B-cells, ablation of cis ligands causes increased co-localization of CD22 and the BCR in raft-clathrin domains, resulting in suppression of BCR signaling (47)(48)(49). Inclusion of CD22 ligands in a polymeric T-independent antigen also suppresses BCR activation by physically ligating CD22 to the BCR complex (50,51). Similarly, co-expression of CD22 trans ligands on antigen-presenting cells dampens Bcell activation, presumably by recruitment of CD22 to sites of cell contact in close proximity to the BCR synapse (8,14). Such observations suggest that trans ligands of CD22 could participate in recognition of self and establish a threshold for B-cell activation and maintenance of peripheral tolerance (14,50,51).
In addition to the activation of B-cells, CD22-ligand interactions have been proposed to mediate other aspects of B-cell biology including differentiation, antigen presentation, and trafficking to bone marrow that involve interactions with trans ligands on other cell types (5,13,52,53). Recent reports demonstrate that high affinity cis ligands of CD22 are downregulated on B-cells during differentiation in germinal centers (54,55), favoring binding of CD22 on activated B-cells to trans ligands on other contacting cells. trans ligands of CD22 have been documented to exist on B-cells, T-cells, monocytes, endothelial cells, and dendritic cells (9,10,13,18,56). Previously, we demonstrated that CD22 is redistributed to sites of B/B-cell contact mediated by sialic acid-dependent interactions with trans ligands on the apposing B-cell (8). Homotypic B-cell interactions occur in lymphoid organs and bone marrow whenever B-cells are in close contact and have been demonstrated to be important in the regulation of IgE and IgG synthesis (57,58), polyclonal B-cell activation (59), maintenance of peripheral tolerance (60), differentiation of pre-Bcells into mature B-cells in the bone marrow (61,62), and bystander transfer of antigens from one B-cell to another (63). Because CD22 plays a role in establishing the threshold for B-cell receptor signaling, identifying the ligands in B/B-cell contacts that regulate CD22 redistribution to sites of cell contact is of significant interest.
In this study, we investigated the nature of the trans ligands of CD22 on B-cells that mediate redistribution of CD22 to sites of cell contact in homotypic B-cell interactions. Because CD22 is known to recognize the majority of B-cell glycoproteins that carry the NeuAc␣2-6Gal sequence in cell lysates (15,17,18), a major objective was to establish whether CD22 recognizes all B-cell glycoproteins as trans ligands in situ or preferentially recognizes a subset of them. We first identified 27 trans glycoprotein ligand candidates by proteomics analysis of complexes formed by UV cross-linking CD22-Fc chimera to B-cells labeled with 9-aryl azide-NeuAc. This expands the number of identified sialic acid-dependent binding partners of CD22-Fc chimera from four (CD45, IgM, CD19, and CD22) in previous reports (15,16,18,20) to 27 (see Tables I and II). The validity of the "hits" was assessed by confirming that the protein hits formed UV irradiation-induced cross-linked complexes with CD22 by Western blot analysis using antibodies available to 11 of these glycoproteins, including eight of 10 glycoproteins identified by both the spectral count and SILAC analyses. The ability of CD22 to use these glycoproteins as trans ligands at the site of cell contact was then assessed by cross-linking B-cells labeled with 9-aryl azide-NeuAc to CD22-expressing CHO cells followed by analysis of the cross-linked products by Western blot. Remarkably, only three of these 11 glycoproteins, IgM, Basigin, and CD45, were detected as trans ligands in the context of B-cells bound to CD22-CHO cells. We do not suggest that they are the only in situ trans ligands of CD22, only that of the 11 analyzed, these three are the only three detected with IgM appearing to be selectively recognized based on the robust detection of CD22-IgM complexes relative to those of CD45 and Basigin (e.g. compare Figs. 3 and 4D).
The inference that IgM is preferentially recognized by CD22 as an in situ trans ligand was further supported by the demonstration that IgM and CD22 were co-distributed to the site of contact of interacting primary murine and human B-cells and that redistribution of IgM to the site of contact was both CD22-and sialic acid-dependent ( Fig. 5 and supplemental Fig. S4). The conclusion that this interaction is selective is underscored by the fact that surface IgM (43) and CD22 (64) are expressed at only ϳ13,000 and ϳ25,000 molecules per cell, respectively, whereas CD45, a major cell surface glycoprotein, is expressed at ϳ150,000 molecules per cell (65,66). More remarkably, the total sialic acid content of a lymphocyte is estimated to be in excess of 10 9 molecules per cell (67) of which a major fraction (Ͼ25%) would be found in sequences recognized by CD22. Thus, the interaction of CD22 with IgM at the site of cell contact is highly selective, occurring in the presence of a large excess of other glycoproteins that carry the glycan sequence recognized by CD22. Although the functional significance of the redistribution of IgM and CD22 to the site of cell contact in instances of homotypic B-cell interactions is not clear (57)(58)(59)(60)(61)(62)(63), the result is that CD22 and IgM are brought together in close proximity on both cells, which would have the expected consequence of down-regulation of signaling from BCRs at the site of cell contact.
The preferential recognition of IgM as a trans ligand strongly suggests that factors other than glycan sequence determine the ability of a glycoprotein on an apposing cell to serve as a trans ligand of CD22. Such factors may include the geometric distribution and projection of glycans from the surface of the cell as well as microdomain localization and/or interaction with the cytoskeleton of the cell. It is remarkable that IgM is selectively recognized by CD22 in trans but is not significantly recognized as a cis ligand of CD22 on B-cells (15). Thus, the factors that make IgM a preferred in situ trans ligand do not make IgM a suitable ligand for CD22 in cis.
We anticipate that the approach used here can also be used to investigate trans ligands of CD22 for other cells (e.g. T-cells, epithelial cells, dendritic cells, etc.) known to contact B-cells in physiologically relevant contexts (5, 6, 9 -14, 53, 66, 68). Stamenkovic et al. (22) provided evidence that CD45RO is a physiologically relevant in situ trans ligand of CD22 on T-cells by virtue of blocking binding of T-cells to CD22 with anti-CD45 but concluded that other ligands are also likely involved (11,69). In principle, this approach can also be used to examine the cis and trans ligands of other siglecs, provided that they can accommodate a substituent on sialic acid that allows covalent cross-linking. In this regard, Kohler and coworker (70) have demonstrated that N-acyl diazirine-modified sialic acids are an effective alternative for photocross-linking of glycoprotein ligands to CD22. Although this report has focused on sialic acid-dependent ligands of CD22, it should be noted that some siglecs, including CD22, also selectively bind trans ligands that are not sialic acid-dependent, presumably via an alternative binding site(s) (56,71,72). Such observations emphasize the need to conduct in situ analyses to distinguish physiologically relevant receptor-ligand interactions from in vitro interactions that are missing the cellular context.
Finally, the results presented here have broader significance to the general problem of identifying the in situ ligands of glycan-binding proteins. Most glycan-binding proteins in the major glycan-binding protein families (e.g. siglecs, C-type lectins, and galectins) interact primarily with the glycan moiety of glycoproteins, which is sufficient for driving ligand interactions in vitro (e.g. from cell lysates). However, it is clear that the cellular context can significantly restrict the utilization of a glycoprotein as an in situ ligand even if it carries the glycan epitope recognized as a ligand in vitro. Thus, glycoproteins observed to be in vitro binding partners are best viewed as ligand candidates until verified as physiologically relevant ligands in situ. This article contains supplemental Figs. S1-S4, Tables S1-S3,   and an annexure to Table S1.