Normal ligand binding and signaling by CD47 (integrin-associated protein) requires a long range disulfide bond between the extracellular and membrane-spanning domains.

CD47 is a unique member of the Ig superfamily with a single extracellular Ig domain followed by a multiply membrane-spanning (MMS) domain with five transmembrane segments, implicated in both integrin-dependent and -independent signaling cascades. Essentially all functions of CD47 require both the Ig and MMS domains, raising the possibility that interaction between the two domains is required for normal function. Conservation of Cys residues among CD47 homologues suggested the existence of a disulfide bond between the Ig and MMS domains that was confirmed by chemical digestion and mapped to Cys(33) and Cys(263). Subtle changes in CD47 conformation in the absence of the disulfide were suggested by decreased binding of two anti-Ig domain monoclonal antibodies, decreased SIRPalpha1 binding, and reduced CD47/SIRPalpha1-mediated cell adhesion. Mutagenesis to prevent formation of this disulfide completely disrupted CD47 signaling independent of effects on ligand binding, as assessed by T cell interleukin-2 secretion and Ca(2+) responses. Loss of the disulfide did not affect membrane raft localization of CD47 or its association with alpha(v)beta(3) integrin. Thus, a disulfide bond between the Ig and MMS domains of CD47 is required for normal ligand binding and signal transduction.

A cell responds to its environment through cues arising from binding of soluble mediators of cell-cell communication and from interaction with insoluble molecules in the extracellular matrix (ECM) 1 or on adjoining cells. A detailed understanding of how specific plasma membrane molecules transduce information from the extracellular milieu to the cytoplasm and how these are spatially and temporally integrated is required to begin to understand, predict, and potentially regulate these responses in normal and pathologic conditions. Two major models of transmembrane signal transduction in response to ligand binding have been advanced. For many Ig family and growth factor receptors, signaling is initiated by receptor clustering. Other receptors, like the heptaspanin family of heterotrimeric G protein-coupled receptors, seem to signal ligand binding through conformational changes in the membrane-spanning domain that lead ultimately to the activation of effector cascades.
CD47 is an Ig superfamily member involved in signaling from both cell-cell and cell-ECM interactions. Coligation of CD47 and the T cell antigen receptor (TCR) is a synergistic signal for activation of T lymphocytes (1-3) that requires CD47-induced association of protein kinase C with cytoskeleton (3). In addition, CD47 can exist in a plasma membrane complex with the integrin ␣ v ␤ 3 or ␣ 2 ␤ 1 that both modulates integrin function and has signal transduction properties distinct from either CD47 or the integrin in isolation (4 -7). CD47 binds directly to thrombospondin, a protein of the provisional ECM at sites of inflammation (8), and to SIRP␣1, a broadly expressed plasma membrane molecule most highly represented on neurons, macrophages, and dendritic cells (9,10). Interaction with these ligands can lead to cell adhesion to ECM (6,7,11), to cell-cell aggregation (12,13), or to alterations in cell behavior via heterotrimeric G protein-dependent and -independent mechanisms (3,6,14). CD47-deficient mice or cells show a variety of abnormalities consistent with a significant role for this molecule in modulating cell responses to adhesive stimuli. Since CD47 is an Ig family member that can in at least some circumstances signal via heterotrimeric G proteins, the paradigm for signal transduction through this interesting molecule is uncertain. CD47 is an unusual member of the Ig superfamily because, in addition to a single Ig domain, it has a highly hydrophobic, multiply membrane-spanning (MMS) domain that is thought to contain five transmembrane segments (15). Structure-function studies have demonstrated that both the Ig domain and the MMS domain of CD47 are essential for its role in signal transduction as well as for localization of CD47 to the cholesterol-rich plasma membrane domains known as glycosphingolipid-enriched membranes (gems) or rafts (1,2,4). Chimeric molecules in which the CD8 Ig domain or the FLAG epitope was substituted for CD47's Ig were mislocalized and nonfunctional, even when the new extracellular domain was ligated by appropriate antibodies (3). This requirement for its specific Ig domain is unlike those of more conventional Ig superfamily signaling molecules (16) and suggests that CD47 aggregation is insufficient to initiate signaling. Thus, the Ig domain plays a fundamental role in CD47 signal transduction in addition to its binding of the ligands thrombospondin and * This work was supported by National Institutes of Health Grants GM38330 and AI24674 and a grant from the Sandler Family Supporting Foundation. 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. ‡ Present address: Protein Design Labs, Inc., Fremont, CA 94555. § To whom correspondence should be addressed: Program in Microbial Pathogenesis and Host Defense, University of California, San Francisco, Box 0654, 513 Parnassus Ave., San Francisco, CA, 94143. Tel.: 415-514-0167; Fax: 415-514-0169; E-mail: ebrown@medicine. ucsf.edu. 1 The abbreviations used are: ECM, extracellular matrix; MMS, multiply membrane-spanning; TCR, T cell receptor; SIRP, signal-regulatory protein; M␤CD, methyl-␤-cyclodextrin; FITC, fluorescein isothiocyanate; CHAPS, 3-[(3-cholamidopropyl)dimethylammonio]-1-propanesulfonic acid; Tricine, N-[2-hydroxy-1,1-bis(hydroxymethyl)ethyl]glycine; PBS, phosphate-buffered saline; PAGE, polyacrylamide gel electrophoresis; mAb, monoclonal antibody; GPCR, G-protein-coupled receptor; IL-2 BNPS-skatole, 2-(2Ј-nitrophenylsulfonyl)-3-methyl-3bromoindolenine. SIRP␣1, which suggests the possibility that the CD47 Ig domain is required for a signaling-competent conformation of the molecule. The purpose of the present study was to determine the reason for the requirement for the CD47 Ig domain in its signaling function. We have found that there is a long range disulfide bond between Cys 33 and Cys 263 in human CD47 that is required for signal transduction as well as for normal SIRP␣1 binding. Moreover, the cysteines involved in this long range disulfide are conserved in all species' CD47 paralogs and in the poxvirus molecules of unknown function with structural homology to CD47. The unusual requirement for the CD47 Ig domain in signaling function can be explained at least in part by its interaction with the MMS domain, leading to appropriate conformation not only for ligand binding but for association with intracellular signaling cascades as well.
CD47 Chimera and Mutants, SIRP␣1-Fc Fusion Protein, and SIRP␣1/CD7 Chimera-Chimeric forms of CD47 were made with standard molecular techniques and have been described previously (1,2). Briefly, CD47/CD7 is composed of the Ig domain of CD47 linked to the transmembrane domain of CD7. CD47/GPI contains the CD47 Ig domain with a glycosylphosphatidylinositol anchor. FLAG-MMS and CD8-MMS include a FLAG epitope tag or murine CD8␣ Ig domain, respectively, linked to the MMS domain of CD47. Fig. 1A shows a schematic model of CD47 indicating the sites of designed MMS domain mutations. All mutations were constructed using standard molecular techniques. All products from PCR were confirmed by sequencing in both directions. The mutant, termed first ICL (1 st ICL), replaces 163 KTLKYRSGGMDEK 175 (all amino acid numbering based on immature CD47) with KAAAAAKAAAAAK. The second ICL mutation (2 nd ICL) replaces 234 LTS 236 with KKK. The third transmembrane mutation (3 rd TM) replaces 224 HYY 226 with AAA. The Cys/Ser mutants replace Cys residues at 33, 259, and/or 263 with Ser, as indicated. cDNA encoding the extracellular three Ig domains of human SIRP␣1 (SHPS-1) was obtained as IMAGE clone number 2017171. It was modified and transferred by standard molecular techniques into pCDM8 containing the coding sequence of the hinge and constant regions of the heavy chain of human IgG1 (20). SIRP␣1-Fc fusion protein was produced by transient transfection of COS cells or stable transfection of 293 cells using lipofectamine/PLUS reagent (Life Technologies, Inc.). Protein was harvested from culture supernatants using Protein A-Sepharose and confirmed to exist as a dimer by Western blot with anti-human IgG-Fc (apparent molecular mass of ϳ130 kDa). SIRP␣1/CD7 was constructed from the modified IMAGE clone and pIAP323, which contains the CD7 transmembrane domain (5). It consists of the three extracellular Ig domains of SIRP␣1 linked to the transmembrane domain of CD7.
Jurkat and JinB8 cells were transfected by electroporation at 280 V, 1000 microfarads, and 25°C in RPMI and immediately placed on ice. After 10 min, cells were placed in nonselective medium for 24 h and then transferred to medium containing 1.5 mg/ml Geneticin. OV10 cells were transfected using LipofectAMINE reagent. After 24 h, cells were passed into medium containing 400 g/ml Geneticin. None of the mutations prevented surface expression, and all cells were sorted by fluo-rescence-activated cell sorting or magnetic bead methods to isolate positive populations of equal expression level (Fig. 1B).
Chemical and Enzymatic Digests of CD47-CD47 was isolated from human placenta as previously described (17). Briefly, tissue was minced and homogenized in 50 mM Tris, 0.25 M sucrose, 5 mM iodoacetamide, 2 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 g/ml leupeptin, 10 g/ml aprotinin (pH 7.4). Membrane fractions were prepared from the cleared homogenate, dissolved in homogenization buffer with 80 mM ␤-D-octyl glucoside, and incubated with anti-CD47 Sepharose. Anti-CD47-Sepharose was washed with low and high salt CHAPS buffers (150 or 500 mM NaCl, 25 mM Hepes, 10 mM CHAPS, 3 mM Ca 2ϩ , 1 mM Mg 2ϩ , 25 M paranitrophenyl paraguanidinobenzoate, pH 7.4), and bound material was eluted with 10 mM CHAPS in 500 mM NaCl, 100 mM glycine, pH 3.3, and neutralized with one-tenth volume of 1 M Tris (pH 9.0). Eluted material was then dialyzed against low salt CHAPS buffer, reapplied to the anti-CD47 Sepharose, eluted, and dialyzed as described. Aliquots of isolated CD47 were lyophilized and redissolved in 1% acetic acid. Three volumes of 1 mg/ml BNPS-skatole in glacial acetic acid were added, and the mixture was incubated in the dark at 25°C for 24 h. The digest was diluted 1:1 with water and centrifuged at 10,000 ϫ g for 5 min to remove precipitated BNPS-skatole. The supernatant was concentrated in a Speedvac evaporator (Savant) to reduce volume and remove residual acetic acid and treated with Tricine sample buffer (0.1 M Tris, 24% glycerol, 8% SDS, 0.02% Coomassie Blue G-250) with or without 200 mM dithiothreitol for 40 min at 40°C. Samples were run on Tris-Tricine gels, blotted to Immobilon-P-SQ membranes (Millipore Corp.), and probed with mAb 131.
A similar method of affinity isolation of JinB8 transfectant CD47 was performed for endoproteinase Arg-C digests. Briefly, cells were treated with 20 mM iodoacetamide in PBS in the dark at 0°C for 45 min and then lysed in 10 mM CHAPS, 10 mM iodoacetamide, 10 g/ml aprotinin, 10 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride in PBS (pH 7.4) at 4°C for 30 min. Lysates were cleared by centrifugation at 14,000 ϫ g, and CD47 protein was immunoprecipitated with anti-CD47-Sepharose, washed, eluted, and dialyzed as described (17). Digests were performed in 10 mM CHAPS, 0.1 M Hepes (pH 8.2), with 20 g/ml protein sequencing grade endoproteinase Arg-C (Sigma) for 1 h at 37°C. Samples were then treated with Laemmli sample buffer with or without ␤-mercaptoethanol at 60°C for 15 min, run on 4 -20% gradient SDS-PAGE gels, blotted to polyvinylidene difluoride, and probed with mAb 151.
For evaluation of the trypsin sensitivity of the CD47 mutants, cells were pretreated with 10 mM iodoacetamide in PBS for 15 min on ice, lysed in 1% Triton X-100, 20 mM Tris, 140 mM NaCl, 10 mM iodoacetamide, 2 mM EDTA, pH 8.2, for 30 min and centrifuged at 14,000 ϫ g for 5 min. The cleared whole cell lysates were then digested with trypsin at 2.5 mg/ml for 30 min at 37°C, and the reaction was stopped by the addition of Laemmli sample buffer and incubation at 60°C for 15 min. Samples were then Western blotted with anti-CD47 Ig domain-specific mAb B6H12.
SIRP␣1-Fc Binding and SIRP␣1/CD47-mediated Cell Aggregation Assay-For assessment of binding of SIRP␣1-Fc protein, cells were incubated with SIRP␣1-Fc fusion protein, anti-CD47, or control mAb in PBS with 1% bovine serum albumin for 30 min on ice, washed, and labeled with anti-human IgG-Fc-FITC or anti-mouse IgG-FITC. Samples were then analyzed on a flow cytometer, and the mean fluorescence was determined, from which a ratio of SIRP␣1/CD47 fluorescence was calculated. For examination of mAb 10G2 binding, cells were pretreated with 100 ng/ml phorbol 12-myristate 13-acetate plus 2 M ionomycin for 18 h prior to analysis, and the percentage of positive staining was calculated. For treatment with methyl-␤-cyclodextrin (M␤CD), cells were harvested in PBS with 5 mM EDTA, washed once with PBS and once with Iscove's modified Dulbecco's medium plus 0.1% fatty acid-free bovine serum albumin. Cells were then resuspended at 2.5 ϫ 10 5 cells/ml with or without 10 mM M␤CD, incubated at 37°C for 10 -15 min, and washed with Iscove's modified Dulbecco's medium plus fatty acid-free bovine serum albumin. Samples were then processed for flow cytometry as described.
IL-2 Assays-Human CD47-transfected murine 3.L2 T cells were stimulated by murine CH27 B cells presenting peptide antigen at varied doses as previously described (1). Briefly, 1 ϫ 10 5 3.L2 cells, 1 ϫ 10 4 CH27 cells, and 0.03-1 M antigenic peptide (amino acids 64 -76 of hemoglobin d (21)) were added in RPMI medium to the wells of 96-well plates and incubated for 18 -24 h at 37°C. An IL-2 enzyme-linked immunosorbent assay was performed on harvested supernatants according to the manufacturer's instructions (Pharmingen). Duplicate samples were assayed, each in at least three replicates of each experiment.
Measurement of Ca 2ϩ Flux following Receptor Cross-linking-Jurkat or JinB8 cells at 2 ϫ 10 7 cells/ml in RPMI complete medium were incubated with 3 M fura-2-AM at 37°C for 20 min and then diluted 10-fold and incubated an additional 20 min. Cells were then washed and incubated with the indicated concentrations of mAb on ice for 20 min. After labeling, cells were washed once with RPMI medium and twice with calcium buffer (25 mM Hepes, 125 mM NaCl, 5 mM KCl, 1 mM Na 2 HPO 4 , 0.5 mM MgCl 2 , 1 mM CaCl 2 , pH 7.4) and resuspended at 2.5 ϫ 10 6 cells/ml in calcium buffer on ice. Fluorescence changes of a 2-ml stirred cell suspension warmed to 37°C were monitored with a F-2000 or F-4500 spectrofluorimeter (Hitachi Instruments, Danbury, CT) using 340-and 380-nm excitation wavelengths and 510-nm emission wavelength following the addition of 10 g/ml secondary antibody. Calcium concentrations were calculated as described by Grynkiewicz et al. (22). Calcium flux from CD47 cross-linking was identical when cells were coated with 1 or 10 g/ml anti-CD47 mAb, but no flux occurred when 0.1 g/ml anti-CD47 was used or when the cross-linking antibody was omitted, confirming that CD47 aggregation was required for the rise in [Ca 2ϩ ] i . The addition of 2 M ionomycin served as a positive control for Ca 2ϩ flux.
Isolation of Membrane Rafts-The location of cell surface proteins in sucrose density gradients was evaluated using tracer 125 I-labeled antibodies as described (4). Cells were incubated with 5 g/ml 125 I-mAb in growth medium at 4°C; washed; and lysed in 20 mM Tris-HCl, pH 8.2, 140 mM NaCl, 2 mM EDTA, 25 g/ml aprotinin, 25 g/ml leupeptin, 1 mM phenylmethylsulfonyl fluoride, 0.1% Brij 58. Sucrose solution was added to a final concentration of 40% using a stock of 60% sucrose in 20 mM Tris-HCl, pH 8.2, 140 mM NaCl, 2 mM EDTA, and this mixture was layered over a volume of 60% sucrose. 25 and 5% sucrose layers were added to form a step gradient, which was centrifuged at 170,000 ϫ g for 18 h at 4°C. Fractions of 0.5 ml were collected from the top of the gradient, and radioactivity in each fraction as well as the pellet was assessed.
Isolation of ␣ v ␤ 3 Integrin-containing Protein Complexes-Complexes were isolated as described (4). Briefly, OV10 cells expressing CD47 or mutants were mixed slowly in HBSS with anti-␤ 3 integrin-coated magnetic beads for 15 min at 37°C. Adherent cells were separated with a magnet and lysed in CHAPS buffer with mixing for 10 min. Beads were reisolated, bead-associated protein complexes were eluted with Laemmli sample buffer, and ␤ 3 -associated CD47 was quantitated as described (4). Values are expressed in relative units based on densitometry of chemiluminscence-exposed bands on x-ray films.
Statistical Analysis-All experiments were repeated at least three times. Error bars in graphs depict S.E. The statistical significance of each set of results was evaluated by performing a one-way analysis of variance followed by Dunnett or individual t tests as appropriate. A p value of Ͻ0.05 was considered significant. showing the location of tryptophan residues and composition and molecular weight of theorized BNPS cleavage products in the presence or absence of the potential disulfide bond. B, CD47 was isolated from human placenta and digested with BNPS-skatole as described. The resulting peptides were resolved on Tris-Tricine gels and detected by Western blotting using mAb 131 specific for the CD47 cytoplasmic tail. C, schematic diagram of CD47 showing the location of arginine residues and composition and molecular weight of theorized endoproteinase (Endo) Arg-C cleavage products in the presence or absence of the potential disulfide bond. D, CD47 was isolated from placenta or Jurkat transfectants and digested with endoproteinase Arg-C. The resulting peptides were resolved by SDS-PAGE and Western blotted using mAb 151 specific for the CD47 cytoplasmic tail.

A Long Range Disulfide Bond
Exists between the Ig and MMS Domains of CD47-Previous studies have indicated a requirement for both Ig and MMS domains of CD47 to mediate virtually all described functions of CD47 and for its efficient localization to membrane rafts. This suggests that an interaction between these domains may be important for function and subcellular localization. The Ig domain contains conserved cysteine residues at positions 41 and 114 that are necessary for the Ig domain formation as well as a potentially free Cys 33 (Fig. 1,  A and C). Analysis of the structurally unusual MMS domain predicts a membrane topology with five membrane-spanning segments and residues Cys 259 and Cys 263 on the ectoplasmic face of the membrane (15) (Fig. 1A). We postulated that one of these cysteines could be involved in a disulfide bridge between the Ig and MMS domains potentially important for an Ig/MMS interaction. We isolated CD47 from human placenta (17) or JinB8 transfectants and performed chemical and enzymatic digests to evaluate disulfide linkage of the Ig and MMS domains. Attempts to determine the presence of a long range disulfide bond via classic methods of enzymatic digestion, high pressure liquid chromatography peptide separation, and mass spectrometry yielded only Ig domain peptides, as in previous studies using Edman degradation instead of mass spectrometry (17). It is likely that MMS domain peptides were lost during processing due to their hydrophobic character. Instead, digestion with either BNPS-skatole or endoproteinase Arg-C followed by Western blotting with a mAb that recognizes the COOH terminus of CD47 was used to identify the fragmentation pattern with or without reduction. BNPS-skatole, which cleaves carboxyl-terminal to tryptophan residues, could determine the presence or absence of the putative long range disulfide, because an ϳ15-kDa carboxyl-terminal fragment resulting from cleavage after Trp 157 in the first transmembrane segment would be present without reduction only in the ab-sence of a disulfide between the Ig and MMS domains ( Fig. 2A). As shown in Fig. 2B, the 15.5-kDa fragment was not observed on Western blotting of nonreduced samples of placental CD47 after treatment with BNPS-skatole but was easily detected in digested samples that had been reduced prior to separation on SDS-PAGE. The smaller 5-kDa band seen in reduced digested samples must represent cleavage by BNPS at another residue. Since secondary reaction sites for BNPS are Cys and Met, the most likely site for this secondary BNPS cleavage is Cys 259 (yielding a 4.9-kDa fragment). If Cys 259 was involved in a disulfide bond with the Ig domain and the cleavage site was Cys 263 , BNPS cleavage would have released the C terminus even without sample reduction. The closest potential methionyl cleavage site (Met 172 ) would have yielded a significantly larger fragment (14 kDa) and is therefore unlikely to account for this small fragment. Whatever the identity of the cleavage site, it is within 5 kDa of the C terminus yet is covalently linked to the amino terminus of the protein in the absence of reduction, demonstrating that a cysteine carboxyl-terminal to this cleavage site is involved in a disulfide bond.
CD47 point mutants C33S, C259S, and C263S were transfected into the CD47-deficient Jurkat clone JinB8 (2) to identify the location of the cysteines involved in disulfide bonding. Unfortunately, it proved impossible to obtain sufficient quantities of mutant CD47 for BNPS digestion analysis. Digestion with endoproteinase Arg-C, however, yielded sufficient material for analysis. Theoretical sites of cleavage are shown in Fig.  2C. While the anticipated ϳ14-kDa fragment was observed in nearly 100% yield with reduction of digested placental CD47, nonreduced fragments reactive with the carboxyl-terminal mAb migrated at ϳ50 kDa, similar to undigested CD47 (Fig.  2D). Thus, all of the C-terminal fragments remained linked to additional fragments under nonreducing conditions. Analysis of Jurkat-expressed wild type CD47 showed that digestion yielded Ͼ95% release of a 14-kDa fragment with reduction, but ϳ75% of the 14-kDa fragment was retained in an ϳ50-kDa band under nonreducing conditions. Digestion of the C259S mutant CD47 did not increase the amount of free 14-kDa fragment without reduction. In contrast, digestion of both the C33S and C263S mutants led to quantitative release of the 14-kDa carboxyl-terminal fragment without reduction. Thus, Cys 33 and Cys 263 of CD47 are required for disulfide-dependent retention of the 14-kDa band with other fragments after Arg-C digestion. These data demonstrate that there is a disulfide bond between Cys 33 and Cys 263 in CD47 isolated from Jurkat T cells, and a similar disulfide between Ig and MMS domains is present in placental CD47.
Cys 33 and Cys 263 , but not Cys 259 , are conserved in all members of the CD47 family, including several poxvirus proteins (Fig. 1C). Based on conservation of these Cys residues, this long range disulfide bond is probably preserved in all members of the CD47 family.
Trypsin Digestion of CD47 Mutants-Whole cell lysates were digested with trypsin to examine differences in protease sensitivity of wild type or mutant CD47 molecules. Total cellular protein was Ͼ95% digested as determined by loss of Coomassie Blue staining of the digested lysate (data not shown). The digested lysate was separated by SDS-PAGE, and remaining CD47 was detected using an anti-Ig domain mAb, B6H12. Wild type CD47 was resistant to trypsin digestion, while a molecule in which the MMS domain was replaced by the CD7 transmembrane domain (CD47/CD7) was sensitive, suggesting that the presence of the CD47 MMS domain decreased trypsin sensitivity of the B6H12 epitope in the Ig domain. Like wild type CD47 (17), CD47/CD7 can form multimers on SDS-PAGE, and both monomeric and dimeric forms were detected. Although the C259S mutant showed trypsin resistance similar to wild type CD47, the C33S and C263S mutants showed much greater trypsin sensitivity (Fig. 3), suggesting increased accessibility of trypsin cleavage sites in the Ig domain in the absence of a disulfide between these two cysteines. Thus, loss of the long range disulfide increases the susceptibility the CD47 Ig domain to trypsin digestion.
In order to further evaluate the effects of these Cys/Ser point mutations on the accessibility of Ig domain epitopes under more physiologic conditions, binding of a panel of mAbs that recognize the CD47 Ig domain was assessed by flow cytometry.
All of the antibodies shown in Table I  The mAb B6H12 binds a temperature-sensitive epitope (23) and is inhibitory in integrin-dependent assays of CD47 function (5-8). The mAb 10G2 recognizes a site that contributes to soluble CD23 binding by the CD47-␣ v ␤ 3 complex (19). Expression of the 10G2 epitope is influenced by interaction of the MMS domain with cholesterol (4). Thus, loss of the Cys 33 -Cys 263 disulfide slightly alters mAb recognition and markedly increases trypsin sensitivity of the Ig domain, consistent with a conformational change in CD47 in the absence of the long range disulfide.

Loss of the Cys 33 -Cys 263 Disulfide Affects SIRP␣1 Ligand Binding and CD47-SIRP␣1-mediated Cell Aggregation-The
likelihood that there were subtle conformational effects from the loss of the Cys 33 -Cys 263 disulfide raised the possibility that loss of the disulfide might affect ligand binding and/or cell-cell interaction. To determine the importance of the disulfide bond in binding of SIRP␣1, a ligand for CD47 expressed prominently on macrophages and dendritic cells, the binding of soluble dimeric SIRP␣1 was assessed. When binding was assessed at a SIRP␣1-Fc concentration within the linear range of the binding curve, there was an ϳ50% reduction in binding of SIRP␣1-Fc to CD47 mutants that could not form the long range disulfide (Fig. 4A). The single point mutant C263S showed as great a defect in ligand binding as the double CC259/263SS mutant, and the C259S showed no defect in ligand binding. The absence of the MMS altogether in the CD47/CD7 chimera (5) showed a similar reduction in ligand binding, indicating that the principal contribution of the MMS to ligand binding is the disulfide between Cys 263 and Cys 33 . The CC259/263SS mutation also reduced SIRP␣1 binding in an unrelated cell type, the ovarian carcinoma OV-10 (Fig. 4A, right). Binding of SIRP␣1-Fc was clearly reduced in the Cys/Ser mutants over a broad concentration range (Fig. 4B).
To determine whether this loss of ligand binding had functional consequences, CD47-SIRP␣1-mediated cell-cell adhesion was evaluated (Fig. 5). Compared with JinB8 cells expressing wild type CD47, cells transfected with the CD47/CD7 chimera adhered to SIRP␣1-expressing cells less well, demonstrating 2 E. J. Brown, unpublished data.

FIG. 4. SIRP␣1 binding is reduced in Cys/Ser mutants of CD47.
A, SIRP␣1-Fc binding to JinB8 or OV10 transfectants. Cells were incubated with 0.5-1 g/ml SIRP␣1-Fc, washed, and stained with goat anti-human Fc-FITC. Binding was standardized to CD47 expression determined by mAb 2D3 binding. The ratio SIRP␣1-Fc/2D3 for normal CD47 was 1. *, p Ͻ 0.05 versus normal CD47. There was no significant binding to vector-transfected JinB8 cells. Results are mean Ϯ S.E. from three or four experiments. B, binding curve with JinB8 cells. Cells were incubated with increasing concentrations of SIRP␣1-Fc or a saturating concentration of mAb 2D3 (10 g/ml), followed by FITC-labeled antihuman Fc or anti-mouse IgG secondary Abs. Ratios of mean fluorescence of SIRP␣1 to anti-CD47 were calculated and shown from a representative experiment of three performed for normal CD47 (f) and the C263S mutant (Ⅺ).  that the MMS domain is required for optimal CD47-mediated cell aggregation. The CD47 point mutants lacking the long range disulfide showed as large a defect in adhesion as the CD47/CD7 chimera, suggesting that the disulfide is a critical component of the contribution of the MMS domain to cell-cell adhesion. In this assay, similar to soluble SIRP␣1 binding, the C259S mutant was normal, supporting the conclusion that Cys 263 is the relevant MMS domain amino acid for the disulfide bond. Other MMS domain mutants showed no defect in cell-cell adhesion, supporting the specificity of the requirement for the Cys 33 -Cys 263 disulfide for optimal SIRP␣1 binding (Fig. 5, data  not shown).
Loss of the Cys 33 -Cys 263 Disulfide Affects CD47 Synergy with TCR-While loss of the long range disulfide bond reduced ligand binding activity, it was unclear if the signaling activity of CD47 was similarly affected. To determine the role for the disulfide bond in CD47 signaling, functional assays using mAb that bound wild type and mutant CD47 identically were employed. Murine 3.L2 hybridomas (21) or CD47-deficient JinB8 cells expressing CD47 molecules containing point mutations that prevent disulfide formation were compared with cells expressing wild type CD47 or several other mutations in the MMS domain (Fig. 1). All CD47 mutants were expressed at levels comparable with or greater than wild type CD47 as assessed by flow cytometry or Western blotting (Fig. 1B). Several studies have demonstrated CD47 synergy with TCR ligation in T cell activation leading to IL2 synthesis (1, 2). 3.L2 cells transfected with normal human CD47 show an augmented response to antigenic peptides when anti-CD47 is allowed to bind to Fc␥ receptors on the antigen-presenting cells (1) (Fig.  6A). As previously reported (1), some, but not all, antibodies to the CD47 Ig domain promote synergy with antigenic peptide for 3.L2 activation, since 2D3 demonstrates synergy, but B6H12 does not (Fig. 6A). However, 2D3 did not show synergy with peptide in 3.L2 cells expressing either C33S or CC259/ 263SS CD47, which cannot form the long range disulfide bond (Fig. 6, C and D). Ligation of wild type CD47 showed synergy with antigen at mAb concentrations between 0.03 and 20 g/  5. CD47-SIRP␣1-mediated cellular adhesion is reduced in CD47 mutants lacking the long range disulfide. A, adhesion mediated by wild type CD47 on normal Jurkat cells to vector-or SIRP␣1/CD7-transfected JinB8 cells. Fluorescent dye-loaded cells were mixed, and cell-cell adhesion was assessed as described under "Materials and Methods." *, p Ͻ 0.0001 versus vector transfectants. B, adhesion mediated by wild type or mutant CD47-expressing JinB8 to SIRP␣1/ CD7-expressing JinB8 cells. Data represent CD47-SIRP␣1-mediated adhesion, standardized to cell-cell adhesion obtained with wild type CD47 from three experiments (mean Ϯ S.E.), each performed with triplicate samples. Both the second ICL and C259S mutants showed comparable aggregation with wild type CD47. CD47/CD7 and C263S and C33S mutants showed reduced aggregation (p Ͻ 0.05). ml, while the CD47 mutants lacking the disulfide showed no augmentation of the antigen dose response at any concentration in this almost 1000-fold range (data not shown). This difference in function occurred despite equal binding of the mAb to wild type and mutant CD47 (Table I). Other mutations in the MMS domain, such as major alteration in the first intracytoplasmic loop (Fig. 6B), did not inhibit synergy, demonstrating that the effects of the cysteine mutations on CD47 function are specific.
Ca 2ϩ Signaling Is Defective in Cys/Ser Mutants-Crosslinking intact CD47 on normal Jurkat cells induced a rise in [Ca 2ϩ ] i (Fig. 7A (24)). Cross-linking of CD47 on JinB8 reconstituted with CD47 induced a calcium response with similar kinetics to that of normal Jurkats (Fig. 7B) but of lower absolute magnitude. This reduction in magnitude was probably due to the reduced responsiveness (perhaps related to decreased expression of CD3 on the JinB8 clone), since the CD3-stimulated calcium response also was reduced in these cells. A role for interaction between the Ig and MMS domains in this response was suggested by the inability of CD47 mutants lacking either domain to activate a Ca 2ϩ response (Fig. 7, A-C). In contrast to wild type CD47, the C33S and C259S/C263S mutants failed to stimulate a Ca 2ϩ response in JinB8 cells (Fig.  7C). This was confirmed with multiple clones for both mutants. Analysis of single point mutants C259S and C263S indicated that Cys 263 was critical for the increase in [Ca 2ϩ ] i , while the C259S mutation did not affect Ca 2ϩ signaling (Fig. 7D). Mutations in the first ICL, second ICL and third transmembrane domain failed to affect the rise in Ca 2ϩ activated by CD47 cross-linking (Fig. 7C).
Changes in Membrane Raft Localization of Cys/Ser CD47 Are Not Sufficient to Explain Loss of Function-Membrane raft localization is thought to be necessary for CD47 function (3). Therefore, we evaluated if loss of the disulfide bond affected raft localization of CD47. Sucrose gradient fractionation studies showed a modest decrease in raft localization of Cys/Ser mutants relative to wild-type CD47 (Fig. 8A). A similar modest decrease was observed for the second ICL mutant that retains completely normal function in all assays, indicating that this extent of alter- ation in raft association cannot explain the functional deficiency of the Cys/Ser mutants. Thus, Cys/Ser mutants localize to rafts nearly normally but fail to function properly in this membrane compartment. This is consistent with our prior studies demonstrating that membrane raft localization is necessary but not sufficient for CD47 function (3) (Fig. 7B).
Previous studies demonstrated that reduction of membrane cholesterol increased recognition of the CD47 Ig domain by mAb 10G2 and suggested that cholesterol may associate with the CD47 MMS domain to modulate Ig domain conformation (4). Decreasing membrane cholesterol with M␤CD reduced SIRP␣1-Fc binding by CD47 about 50% in both OV10 and Jurkat cells (Fig. 8B, data not shown), implicating cholesterol in optimal SIRP␣1-binding by CD47. Binding to CD47 mutants unable to form the disulfide also was decreased by M␤CD, suggesting that an inability to associate with cholesterol was not the reason for reduced SIRP␣1 binding by these mutants.
Formation of CD47/␣ v ␤ 3 Integrin Complexes Is Not Inhibited by the Absence of the Disulfide Bond-CD47 association with ␣ v ␤ 3 or other integrins can be important in some of its signaling functions (6,7,11,15,17,25). ␣ v ␤ 3 integrin association of CD47 mutants was assessed in OV10 transfectants by coprecipitation, as previously described (4). While the absence of the MMS domain prevented formation of a stable protein complex ( Fig. 9 (4)), the absence of the disulfide bond did not inhibit stable CD47/␣ v ␤ 3 integrin association. Thus, the presence of the nondisulfide-linked MMS domain is sufficient for stable interaction of CD47 with ␣ v ␤ 3 .

DISCUSSION
The dual requirement for Ig and MMS domains of CD47 to mediate its functions and to localize to membrane rafts suggested that an interdomain interaction could be critical to CD47's role in cell biology. The presence of cysteines in the extracellular portions of the Ig and MMS domains indicated that formation of a disulfide bond between these domains was possible. Such a covalent link could significantly constrain CD47 structure with potential consequences for ligand binding and signaling. Previous protease digestion of erythrocyte-derived CD47 by Mawby et al. (26) suggested the possible occurrence of a disulfide bond between the MMS and Ig domains. However, these studies suggested the existence of the disulfide only in a fraction of the isolated CD47, and neither the site nor the consequences of the putative interaction was defined. We now have shown that the disulfide occurs between Cys 33 and Cys 263 of human CD47 in Jurkat T cells. Although failure to recover relevant peptides in the absence of reduction precluded mass spectrographic proof that the disulfide between Ig and MMS domains in placental CD47 is identical, this is the most likely possibility. CD47 with this disulfide is the only detectable form in placenta. Like erythrocytes, Jurkat T cells have a small amount of CD47 apparently lacking this disulfide; whether this means that there is some cellular regulation of disulfide formation is not yet known. While Cys 259 is conserved in mammalian CD47 molecules, poxvirus homologues have a Ser at this position. In contrast, all members of this protein family conserve both the Ig domain Cys and the MMS domain Cys involved in the long range disulfide (Fig. 1C). While little is known of the function of the poxvirus CD47 homologues, conservation of the relevant Cys residues suggests that the disulfide and the resulting conformational constraints are critical for function.
Consistent with this hypothesis, CD47 mutants lacking the long range disulfide were defective for multiple known functions of the molecule. Binding of CD47's cell ligand, SIRP␣1, was diminished, demonstrating that the long range disulfide constrains CD47 into a conformation optimal for ligand binding. Since SIRP␣1 binds to the Ig domain of CD47 (10,27), the additional disulfide most likely optimizes the presentation of the SIRP␣1-binding face of CD47. Consistent with this mode of action, loss of the long range disulfide modestly diminished the binding of two mAbs to the CD47 Ig domain. One of those two mAbs (B6H12) blocks CD47-SIRP␣1 interaction, but it is of interest that several other mAbs that block SIRP␣1 interaction with CD47 bind equivalently to wild type CD47 and the point mutants lacking the disulfide bond. The fact that one mAb showing diminished binding (10G2) does not affect SIRP␣1 interaction demonstrates that the ligand binding site is not the only aspect of the Ig domain affected by the constraints imposed by the long range disulfide.
Signal transduction by CD47 also depends on the long range disulfide, independent of ligand binding. The impairment of signaling by Cys/Ser mutations of CD47 was demonstrated in T cell assays in which the CD47 Ig domain was ligated by mAbs that bind equally to the Cys/Ser mutants and wild type CD47. This suggests that the Cys 33 -Cys 263 disulfide affects not only the presentation of Ig domain epitopes to ligand but the interaction of the MMS domain with potential cytosolic signaling cascades as well. Ligation of CD47 can induce both actin polymerization and protein kinase C translocation in T cells (3); while the precise molecular mechanisms for these effects are currently unknown, failure of CD47 synergy in the absence of the long range disulfide strongly suggests that it is required for CD47-induced activation of these cascades. In this regard, it is noteworthy that major mutations in two intracytoplasmic loops and in charged or polar residues in the MMS transmembrane segments had little effect on CD47 function as we have been able to test it. Thus, the molecular mechanisms by which CD47 links to signaling cascades remain undiscovered.
These data are consistent with a model for CD47 structure and function in which the Ig and MMS domains act interdependently. The restrictive effect of long range disulfide linkage between Cys 33 and Cys 263 would place a face of the Ig domain in close apposition to the MMS domain and could lead to further noncovalent interactions between the domains. Loss of the disulfide does not disrupt raft localization of the molecule, the cholesterol dependence of ligand binding, or interaction with ␣ v ␤ 3 integrin. Since these functions also depend on both the Ig and MMS domains, there may be interdomain interac- FIG. 9. Loss of the long range disulfide does not alter CD47 association with ␣ v ␤ 3 integrin. Association of CD47 with ␣ v ␤ 3 was determined as described under "Materials and Methods." The amount of CD47 coprecipitated with ␣ v ␤ 3 is depicted, with the extent of association of wild type CD47 set to equal 1 unit. Data are a summary of three independent experiments. tions that form independent of the Cys 33 -Cys 263 bond. Clearly, disulfide-mediated interaction between the domains apparently is required for CD47 signaling after ligand binding. It is possible to envision a mechanism by which ligand-induced changes in the extracellular domain of CD47 are transmitted to intracytoplasmic enzymatic cascades via this covalent interaction between Ig and MMS domains. Additionally, this orientation of the Ig domain could influence protein/protein or protein/ lipid interactions by CD47 within the plasma membrane.
Long range disulfide bonds have been identified in several members of the G-protein-coupled receptor family of heptaspanins. These include the angiotensin receptor AT1, ATP receptor P2Y1, Duffy antigen receptor, Kaposi's sarcoma-associated herpesvirus GPCR, and several chemokine receptors (CXCR1, CCR2, and CCR5) (28 -34). Like CD47, these GPCRs have a disulfide between their extracellular amino termini and the final extracellular loops in their MMS domains. Studies of these disulfides indicate that they often contribute to both ligand binding and signal transduction, presumably via conformational constraints on the MMS domain. It is intriguing that CD47 not only has a disulfide bond reminiscent of some GPCR family members but in addition can sometimes interact with heterotrimeric G proteins. Recent studies have shown that mutants of CCR5 and CXCR4 lacking the first two membrane-spanning segments retain ligand binding and signaling function, indicating that five transmembrane segments (as present in CD47) are sufficient for G-protein coupling (35). Perhaps CD47 shares an evolutionary origin with this family of signaling receptors. Although a member of the long N-terminal extracellular region family-B GPCRs (36) with Ig domains at its N terminus has been described (37), the Ig domains are not disulfide-bonded to the MMS domain.
In summary, a long range disulfide bond between the Ig domain and the MMS domain of CD47 is required for optimal ligand binding and at least some important aspects of CD47 signal transduction, and it is conserved in the CD47 family. We hypothesize that interactions between the Ig and MMS domain created by the constraints induced by this disulfide are critical for the function of these membrane receptors for cell-matrix and cell-cell interactions.