Phylogenetic Divergence of CD47 Interactions with Human Signal Regulatory Protein α Reveals Locus of Species Specificity

Cell-cell interactions between ubiquitously expressed integrin-associated protein (CD47) and its counterreceptor signal regulatory protein (SIRPα) on phagocytes regulate a wide range of adhesive signaling processes, including the inhibition of phagocytosis as documented in mice. We show that CD47-SIRPα binding interactions are different between mice and humans, and we exploit phylogenetic divergence to identify the species-specific binding locus on the immunoglobulin domain of human CD47. All of the studies are conducted in the physiological context of membrane protein display on Chinese hamster ovary (CHO) cells. Novel quantitative flow cytometry analyses with CD47-green fluorescent protein and soluble human SIRPα as a probe show that neither human CD47 nor SIRPα requires glycosylation for interaction. Human CD47-expressing CHO cells spread rapidly on SIRPα-coated glass surfaces, correlating well with the spreading of primary human T cells. In contrast, CHO cells expressing mouse CD47 spread minimally and show equally weak binding to soluble human SIRPα. Further phylogenetic analyses and multisite substitutions of the CD47 Ig domain show that human to cow mutation of a cluster of seven residues on adjacent strands near the middle of the domain decreases the association constant for human SIRPα to about one-third that of human CD47. Direct tests of cell-cell adhesion between human monocytes and CD47-displaying CHO cells affirm the species specificity as well as the importance of the newly identified binding locus in cell-cell interactions.

Cell-cell interactions are of course critical in immune function but can have important and revealing differences between species as well as nonlinear dependences on molecular parameters, such as affinity. We explore such general issues with integrin-associated protein (IAP), 2 or CD47, which is a ubiquitous but unique immunoglobulin superfamily receptor with a single Ig domain, a pentaspan transmembrane segment, and a variably spliced cytoplasmic tail (1). The known functional roles of CD47, originally limited to integrin activation (2), have expanded considerably since the identification of SIRP␣ as a physiological ligand (3)(4)(5)(6). SIRP␣ is another Ig superfamily member with three Ig domains and a single span transmembrane segment. SIRP␣ is also broadly expressed, present at high levels on myeloid lineages (monocytes, neutrophils) and neurons but absent from many if not most other cells (7). Both CD47 and SIRP␣ are found in many mammals, but the differences in sequence (Fig. 1A) across species have yet to be evaluated.
CD47-SIRP␣ interactions regulate transmigration of monocytes and neutrophils (8 -11) with severe impairment in neutrophil recruitment to sites of infection in CD47 knock-out mice (12). CD47 on "self " red cells, platelets, lymphocytes, and other nonapoptotic cells inhibits clearance by macrophages in mice by signaling through SIRP␣ (13)(14)(15)(16). However, in humans as opposed to mice, severe reductions of CD47 levels on red cells, as found on various Rh-deficient patient cells, can occur with little to no evidence of anemia or enhanced clearance by splenic macrophages (17,18) and also little to no reported evidence of enhanced RBC interactions with peripheral blood monocytes (19).
The single Ig domain of CD47 is sufficient for binding SIRP␣ at least for humans (20,21), but it may be significant that the human sequence of the lone Ig domain of CD47 differs from that of mice by 38% (Fig. 1A). On the SIRP␣ side, the N-terminal Ig domain is sufficient for the interaction with CD47 (21)(22)(23), but this also differs from humans to mice (by 34%). Although it appears that the CD47-SIRP␣ interaction is enhanced by domain orientation with a conserved "long range" disulfide between the Ig domain and one of the extracellular loops of the membrane-spanning domain (20), further information about the CD47 binding locus for SIRP␣ is limited to that of Liu et al. (24), who used random peptide libraries in phage display to generate a cyclic peptide (CERVIGTGWVRC) that competed with CD47 and bound SIRP␣ at millimolar affinity. A homology model of CD47 by Liu et al. (24) used rat CD2 as a template and led them to postulate a binding site for SIRP␣ involving human CD47 residues Thr 91 , Gly 92 , Arg 114 , and Val 115 . However, the cross-species variation in the residues comprising a putative binding site did not match recent findings on species specificity of human SIRP␣ binding that show that human SIRP␣ binds to CD47 on red cells from pig and man but does not show detectable binding with red cells from mice, rats, or cows (25). Both Thr 91 and Val 115 are not conserved in pig CD47, suggesting that these amino acids are not required for SIRP␣ binding. Here we exploit the broad phylogenetic variation in CD47 (Fig. 1A) to identify the binding locus for species-specific interactions with SIRP␣.

EXPERIMENTAL PROCEDURES
Chemicals-DPBS without Ca 2ϩ or Mg 2ϩ (Invitrogen) was used as is or supplemented with either 1% BSA (Roche Applied Science) or 1% BSA plus 0.05% Tween 20 (Sigma). Tris-buffered saline (TBS) (Bio-Rad) supplemented with Tween 20 (Sigma) and nonfat dry milk (Albertsons, Inc., Boise, ID) was used in Western blotting. Bordetella pertussis toxin (EMD Biosciences, San Diego, CA), cytochalasin D (Sigma), latrunculin A (Sigma), and formaldehyde (methanol-free; Polysciences, Warrington, PA) were used in certain experiments. PKH26 and PKH67 cell labeling kits (Sigma) were used to label monocytes and T cells, respectively. All other reagents used were from Sigma unless noted otherwise.
Expression Vectors and CD47 Mutagenesis-Human CD47 (hCD47; isoform 2) (26), mouse CD47 (mCD47; isoform 2) (26), mCD47 lacking the 21-amino acid insert (4,27), and other CD47 variants were typically PCR-amplified, digested with XhoI and BamHI (New England Biolabs, Beverly, MA), and ligated to similarly digested vector, pEGFP-N3 (Clontech), which results in an in-frame fusion of EGFP at the C terminus of full-length CD47. Single site mutants were prepared using the QuikChange mutagenesis protocol (Stratagene, La Jolla, CA), and multiple site mutants were prepared using sequential application of the QuikChange protocol or through gene synthesis by PCR. The CD47-encoding DNA segment was always sequenced prior to use. Sequence information for all oligonucleotides is available upon request.
Transient Expression of CD47-GFP-CHO cells were plated at 10 5 cells/cm 2 1 day prior to transfection. On the day of transfection, medium was replaced with 2 ml of Opti-MEM I (Invitrogen) per 25 cm 2 of surface area, and 10 -15 l of Lipofectamine 2000 and 5 g of plasmid DNA were diluted in 0.25 ml of Opti-MEM I separately and 5 min later were mixed together and incubated for at least 20 min at room temperature. Lipid-DNA complexes in a total volume of 0.5 ml of Opti-MEM I were then added to the flasks, incubated for 4 -6 h, and then replaced with 5 ml of serum-containing medium. Transfected cells were harvested using DPBS supplemented with 2 mM EDTA (Invitrogen) 1-2 days post-transfection for analysis.
Soluble Human SIRP␣ Production-The extracellular domain of human SIRP␣1 fused to GST (referred to as "soluble SIRP␣" or "SIRP␣" henceforth) was prepared by transfection of COS-1 cells using Lipofectamine 2000 (Invitrogen) and the expression plasmid (6), and the protein was purified from cellfree supernatant using glutathione-Sepharose 4B (GE Healthcare, Piscataway, NJ) as described previously (25). Aglycosylated soluble SIRP␣ was prepared by transfection in 10 g/ml tunicamycin. Core-glycosylated soluble SIRP␣ (high mannose) was prepared by transfecting Lec 1 cells. Monocytic THP-1 cells do not show any significant reduction in SIRP␣ levels (Ͼ90% of untreated level) when grown in the presence of deoxymannojirimycin, suggesting that core glycosylation is sufficient for export, whereas tunicamycin treatment reduces levels significantly (ϳ30% of untreated level; data not shown). Soluble SIRP␣ production in COS-1 cells is also reduced in the presence of tunicamycin, which could, however, be related to off target effects.
Fluorescent Labeling of Transfected CHO with Soluble SIRP␣ and CD47 Antibodies-Dynamic light scattering indicated that the GST-SIRP␣ plus anti-GST complex had a mean hydrodynamic radius, R h Ϸ 12 nm (by mass). Manufacturer-supplied measures of R h for two large proteins (IgG, 160 kDa, 7.1 nm; thyroglobulin, 650 kDa, 10.1 nm) yield a power law fit of R h ϭ 2.0 ϫ M 0.25 , and this correlation implies a molecular mass for the complex of M Ϸ 1.3 MDa, consistent with gel filtration (M Ն 1 MDa). Assuming two SIRP␣ (monomers) to two anti-GST, this molecular mass would imply approximately six SIRP␣ per complex. Five microliters of the soluble SIRP␣ (final concentration ϳ1 M); 5 l of 2 mg/ml AlexaFluor 647 rabbit anti-GST; 40 l of DPBS, 1% BSA; and ϳ2.5 ϫ 10 6 CHO cells were mixed and incubated at room temperature for at least 30 min. Cells were pelleted and resuspended in 1 ml of cold DPBS, 1% BSA and analyzed immediately. For blocking, saturating antibody was added prior to soluble SIRP␣. For antibody labeling, saturating levels of anti-CD47 antibody in 50 l of DPBS, 1% BSA and ϳ2.5 ϫ 10 6 CHO cells were mixed together and incubated as above. Cells were washed in 0.5 ml of DPBS, 1% BSA and then resuspended in 50 l of DPBS, 1% BSA containing 5 l of secondary antibody (2 mg/ml). After an incubation of at least 20 min at room temperature, cells were washed once in 0.5 ml of DPBS and resuspended in 1 ml of DPBS, 1% BSA and kept on ice. Measurement of CD47 Density on Primary Cells-Anticoagulated human blood was treated with RBC lysis buffer (Sigma) and labeled with saturating levels of B6H12 (BD Biosciences) or BRIC126 (IBGRL) for 1 h on ice. Cells were washed twice with DPBS, 1% BSA and incubated with the saturating levels of secondary antibody for at least 30 min on ice. Cells were washed and resuspended in DPBS, 1% BSA and stored on ice until flow cytometric analysis. Mean fluorescence intensities were calibrated against washed human RBC (prepared separately) labeled with saturating B6H12/BRIC126 levels and the same secondary antibody. Surface areas were determined microscopically for detached CHO cells (460 m 2 ) and human T cells (184 m 2 ) approximating these cells as spherical (surface area ϭ 4 ϫ projected area). The upper bound estimate of 50,000 CD47 molecules on the surface of the human RBC (surface area ϭ 128 m 2 ) was used (18).
Western Blotting-CHO cell lysates were prepared from frozen cell pellets lysed in DPBS containing 0.5% SDS and protease inhibitors (4-(2-aminoethyl)-benzenesulfonyl fluoride, E-64, bestatin, leupeptin, aprotinin, and Na-EDTA) and further homogenized by passing through a 21-gauge needle. Cell lysates were denatured for 10 min at 100°C after the addition of ␤-mercaptoethanol (Fisher) and then left untreated or incubated with peptide:N-glycanase (New England Biolabs, Beverly MA) or endoglycosidase H f (New England Biolabs) at 37°C overnight. Samples were then electrophoresed on BisTris gels (NuPAGE; Invitrogen) under reducing conditions in MOPS buffer and transferred to 0.2 M polyvinylidene difluoride (Bio-Rad). The membrane was blocked with 5% nonfat milk in TTBS (Tris-buffered saline with 0.1% Tween 20) by shaking for 1 h at room temperature or overnight at 4°C, washed with TTBS, and incubated with alkaline phosphatase-conjugated goat anti-GFP secondary antibody (Rockland Immunochemicals) in TTBS for at least 60 min. The membrane was washed in TTBS and then in TBS and incubated with the alkaline phosphatase substrate, nitro blue tetrazolium/5-bromo-4-chloro-3-indolyl phosphate (Sigma) until satisfactory color development.
Cell Spreading Assay-200 l of DPBS containing soluble SIRP␣ at 7 g/ml was used to coat 35-mm dishes with glass coverslip bottoms (MatTek Corp., Ashland, MA; available surface area ϳ1.6 cm 2 ) for 1 h at room temperature and then blocked with 200 l of DPBS, 1% BSA or DPBS, 1% BSA, 0.05% Tween 20 for at least 1 h at room temperature. Just before adding cells, the block solution was removed, and the dish was washed with 2 ml of serum-free medium (CD-CHO(A) for CHO and AIM-V for T cells; Invitrogen), and then 2 ml of fresh medium was added and incubated at 37°C/5% CO 2 until the experiment. CHO cells expressing GFP fusions did not require additional labeling. However, human T cells were labeled with PKH67 (Sigma) immediately prior to cell spreading experiments. The dish was maintained at 37°C using a heating stage. Cells were added, and images of the contact area under total internal reflection fluorescence (TIRF) illumination were taken 5, 10, and 20 min after the addition of cells. 10-fold lower SIRP␣ levels gave no spreading of any cells over this time period.
Cell-Cell Conjugate Assay-Human monocytes were labeled with PKH26 (Sigma) per instructions and mixed with CHO cells expressing GFP alone or GFP fusions of different CD47 variants incubated at room temperature in DPBS, 1% BSA for 30 min with intermittent mixing and analyzed by flow cytometry (after compensation using single color controls). The ratio of the number of clustered CHO cells to unclustered CHO cells is expressed as a percentage and denoted "percentage of CHO cells in clusters." Flow Cytometric Analysis and Sorting-For analysis, forward scatter, side scatter, and fluorescence (FL1, FL2, and FL3 channels in logarithmic mode) were acquired for at least 10,000 events using a FACScan or FACSCalibur (BD Immunocytometry Systems, San Jose, CA). A BD FACSVantage TM SE cell sorter (with BD FACSDiVa TM option; BD Immunocytometry Systems) was used for cell sorting. CHO cells expressing three different levels of hCD47-GFP were simultaneously sorted into Opti-MEM I (Invitrogen) supplemented with 30% fetal bovine serum and antibiotic-antimycotic. Approximately 6 ϫ 10 5 cells from each subpopulation were collected, spun, and resuspended in CD-CHO(A) (Invitrogen) prior to use in cell spreading experiments.
Microscopy-Images were acquired on a Nikon TE300 or Olympus IX71 inverted microscope with a ϫ60 (oil, 1.4 numerical aperture) objective using a Cascade CCD camera (Photometrics, Tuscon, AZ). Image acquisition was performed with Image Pro software (Media Cybernetics, Silver Spring, MD). All subsequent image analysis was done using ImageJ (28). In cell spreading experiments, the contact area was measured using a 488-nm laser that illuminated ϳ200 nm into the sample from the glass surface (TIRF mode).
Homology Modeling and Molecular Dynamics-The amino acid (aa) sequence of the extracellular domain of human CD47 was aligned to the rat myelin oligodendrocyte protein (MOG) sequence (Protein Data Bank code 1PK0) (29) using ClustalW version 1.82 (30). The alignment was submitted to Swiss Model (31) to obtain a homology model. The obtained structure was further refined by molecular dynamics simulation (ϳ200 ps) and energy minimization (10,000 steps) in a box of explicit water (32). The numbering of residues in CD47 neglects the first 18 amino acids that make up the signal peptide. A homology model of the N-terminal Ig domain of human SIRP␣1 was also constructed using coordinates of the light chain variable region (V L ) of a Fab structure (Protein Data Bank code 1RHH) (33) with ϳ22% sequence identity as template. Both Protein Data Bank structures are provided as supplemental files.
Fluorescence Recovery after Photobleaching (FRAP)-The MicroPoint laser (Photonic Instruments, St. Charles, IL) with a nanosecond pulse laser exciting the dye Coumarin 440 was used to photobleach CD47-GFP in adherent CHO cells. FRAP was performed by bleaching a small spot (diameter ϳ2 m) with multiple pulses of the laser, and the fluorescence recovery was determined by plotting the time profile of the backgroundcorrected fluorescence in the bleached region normalized to the background-corrected total cell fluorescence signal.

RESULTS
Display of Wild-type CD47 and CD47-GFP Fusion-Fulllength human CD47 (hCD47) was expressed with a C-terminal GFP tag in transiently transfected CHO cells (Fig. 1, A and B). The fusion protein localizes to the CHO cell membrane as is evident from the edge bright GFP fluorescence and also from binding of antibodies to the extracellular Ig domain of hCD47. Flow cytometry shows that expression levels vary by 1000-fold, with red cell and T cell densities near the midrange of CHO cell expression (Fig. 1C); the variation is usefully exploited here in a number of studies directed at clarifying the interaction.
A soluble fusion of the human SIRP␣1 triple Ig extracellular domain to GST was precomplexed with AlexaFluor 647-anti-GST, and this SIRP␣ was incubated with the CHO cells; SIRP␣ binds to single cells in proportion to the level of GFP fluorescence, whereas GST alone (plus anti-GST) does not show detectable binding. The anti-CD47 antibody B6H12 is known to inhibit binding (6,21) and indeed blocks binding here to background levels; the nonblocking antibody 2D3 (6, 21) does not perturb binding, and neither does the intracellular GFP tag on CD47 nor a range of CHO cell treatments, such as heating to physiological temperature, cooling to 4°C (to inhibit endocytosis), or disruption of the cytoskeleton (Fig. S1).
For quantitative analysis of soluble SIRP␣ binding by flow cytometry, the hCD47-GFP expression range was split into eight bins (128 channels each), and the bin average binding was replotted on linear scales (Fig. 1D). The "binding slope" of each regressed line is simply the bound fluorescence normalized by expression level (i.e. the bound fraction, ). By titrating soluble SIRP␣, should and does vary hyperbolically with SIRP␣ (Fig. 1E). The saturable and specific binding of normal hCD47 binding to SIRP␣ yields an apparent dissociation constant K d(app) ϭ 0.7 M and shows no cooperativity (Hill exponent ϭ 1). For brevity, we will more simply denote K d(app) by K d henceforth. This method of analyzing flow cytometry binding data appears both novel and simple for the quantitative determination of affinities with transiently expressing cells. Moreover, determinations of the binding slope made with sub-FIGURE 1. Phylogenetic tree for CD47 and GFP fusion display for assessing interactions with human SIRP␣. A, the percentage of identity in the Ig domain of CD47 between human and other species is depicted along with a schematic of the vector used to transfect the human CD47-GFP fusion. B, CHO cells expressing human CD47-GFP were labeled with 2D3, detected using an AlexaFluor 647 secondary antibody, showing that the fusion is targeted to the membrane and the Ig domain is accessible to antibodies. C, CHO cells expressing the fusion were probed with GST or SIRP␣ complexed with fluorescent secondary and analyzed by flow cytometry. The addition of B6H12 prevents labeling of cells, whereas 2D3 does not. For reference, densities of CD47 on human RBC are ϳ390 molecules/m 2 (18). D, the flow cytometry log-log data is replotted on linear scales to determine binding slopes (generally R 2 Ն 0.95) that accurately yield the bound fraction of SIRP␣ for the given amount of SIRP␣. At the highest expression levels, bound SIRP␣ fluorescence sometimes falls below the linear regression due presumably to molecular crowding effects; such data are not included in the analysis. E, saturable binding of soluble SIRP␣ to hCD47-GFPexpressing CHO cells based on binding slopes at each SIRP␣ concentration. The hyperbolic fit (line) gives the apparent K d ϭ 0.7 m. CMV, cytomegalovirus. saturating concentrations of [SIRP␣] Յ K d are clearly the most sensitive to structural perturbations as exploited below.
Glycosylation Is Not Required for CD47 Export or SIRP␣ Interactions-Both CD47 and SIRP␣ are highly glycosylated proteins (5,34). SIRP␣ glycosylation is already known to moderate its interactions with CD47 (25,35). CD47 glycosylation has unknown roles in function or even export in mammalian cells. With yeast, surface display of the human CD47 Ig as a fusion to the S. cerevisiae mating protein Aga2p required glycosylation for export, with expression levels reduced upon progressive mutation of N-linked sequons (36). Nonetheless, yeastdisplayed protein appeared folded based on thermal denaturation studies and antibody binding, even for Ig-CD47 with all five of the N-linked sequons knocked out. We expressed the same pentamutant here in CHO cells as a GFP fusion (hCD47N ⌬5 ).
Western blotting confirmed that the hCD47N ⌬5 -GFP lacked N-linked sugars, since the molecular weight was similar to that of peptide:N-glycanase-treated hCD47-GFP ( Fig. 2A). The sugars on hCD47 here are a combination of high mannose types (sensitive to both endoglycosidase H and peptide:N-glycanase) and complex types (peptide:N-glycanase only). hCD47N ⌬5 -GFP was also exported efficiently to the CHO cell surface with expression levels comparable with those of hCD47-GFP (Fig.  2B), in contrast to yeast display of the Ig domain fused to Aga2p. A wide range of antibodies raised against human CD47 (B6H12, 2D3, BRIC126, CIKm1, 6H9, and OVTL16) bound to the aglycosylated CD47 efficiently, suggesting again that the Ig domain was folded correctly in maintaining antibody epitopes. Binding of complex-glycosylated, core-glycosylated, and aglycosylated soluble SIRP␣ to hCD47N ⌬5 was also conserved, with measurably weaker binding seen with complex glycosylated soluble SIRP␣ (Fig. 2, B and C). Similar overall results were also obtained when hCD47-GFP was expressed aglycosylated through tunicamycin treatment (not shown).
A slightly tighter apparent K d ϭ 0.4 M is found in binding of soluble SIRP␣ to the deglycosylated hCD47N ⌬5 on CHO cells (Fig. 2D) (versus 0.7 M for wild type) (Fig. 1E). Sugars are obviously not required on either SIRP␣ or CD47 for their mutual interaction. Indeed, sugars inhibit interactions to a similar extent for either protein (1.5-2-fold), probably through steric mechanisms. The association of CD47 and SIRP␣ is therefore mediated by amino acids on these proteins.
CD47 Mediates Cell Adhesion and Spreading That Is Species-specific-hCD47 mediates cell-cell interactions, so a more physiologically relevant but idealized test of adhesive function was developed with SIRP␣-coated surfaces. These surfaces simplify the membranes of monocytes and other phagocytes in that our coated surfaces are flat and smooth, eliminate any diffusion and adhesion-reinforcing clusters of SIRP␣, and also lack a glycocalyx and other surface proteins that could modulate CD47-SIRP␣ interactions. A principal question addressed with these simple surfaces then is whether cell-surface adhesion shows similar or different trends from cell-cell adhesion.
CHO cells expressing hCD47 adhere and spread over minutes on SIRP␣ surfaces (Fig. 3A) but not on surfaces coated with either GST (shown) or BSA. TIRF imaging was used to visualize and measure the increasing lamellopodial contact area of the spreading cells (Fig. 3B), but TIRF also shows that the GFP chimera with CD47, also known as integrin-associated protein, is not aggregated or macroscopically clustered in any type of focal adhesion structure. Initially, the contact area increases linearly in time, suggesting diffusion-limited cell spreading (37). FRAP experiments on CD47-GFP indicate that the mobile fraction is greater than 90%, with a t1 ⁄ 2 for fluorescence recovery on the order of seconds, which yields a preliminary diffusion constant of D Ϸ 0.05-0.3 m 2 /s (supplemental Fig.  S2). This is smaller than D for mobile lipids (38), but it is in the physiological range for membrane proteins (39) and compares well to the cell spreading rates seen here. CD47 is thus highly mobile in this system, and its mobility as well as affinity for SIRP␣ contribute to cell spreading.
Pertussis toxin, which catalyzes ADP-ribosylation of the ␣-subunit heterotrimeric G proteins of the G i family and is known to inhibit CD47-mediated heterotrimeric G i signaling (1), does not inhibit CHO cell spreading. Surface display of the GFP-tagged Ig domain fused to the transmembrane/ cytoplasmic region of IL2R␣ (40) is also sufficient for cell spreading. These observations certainly suggest that the CD47-SIRP␣ interaction is sufficiently strong to provide cells with adequate traction. Interestingly, cell spreading even proceeds at 4°C, where signaling is expected to be suppressed.
To assess the implied role of CD47 density on the kinetics of cell spreading on SIRP␣ surfaces, fluorescence-activated cell sorting was used to separate CHO cells expressing low, medium, or high levels of CD47 (ϳ30, 300, and 3000 molecules/m 2 , respectively). The highest levels of CD47 (3000 molecules/ m 2 ) promote the fastest cell spreading ( Fig. 3C; normalized by total cell area), with no significant CHO cell spreading seen at the lowest levels of CD47 expression (30 molecules/m 2 ). Medium levels of CD47 (300 molecules/m 2 ) lead to an intermediate rate of spreading. These latter surface densities are similar to CD47 levels on primary human T cells (130 -280 molecules/ m 2 ), which show similar initial spreading rates (Fig. 3C). The spreading observed with T cells is also CD47-specific, since pretreatment with the SIRP␣-competing antibody B6H12 blocks spreading, whereas the noncompeting 2D3 antibody does not. T cells also do not spread on surfaces coated with BSA or GST.

. Surface adhesion of CD47-expressing CHO cells and human T cells on SIRP␣-coated surfaces.
A, CHO cells expressing hCD47-GFP spread on SIRP␣ surfaces but not on GST surfaces. Contact area is most clearly seen under TIRF microscopy. B, time evolution of cell spreading and increase in contact area shown for a single CHO cell expressing hCD47-GFP. C, kinetics of normalized contact area (contact area normalized to total cell area) increase for CHO cells expressing different levels of hCD47-GFP and human T cells. Error bars, S.E.
CHO cells expressing mCD47 also spread on SIRP␣-coated surfaces but not as readily as hCD47-expressing cells. Medium, physiological levels of mCD47 (300 molecules/m 2 ) show Ͻ20% of the contact areas compared with CHO cells expressing similar densities of hCD47 (Fig. 4A).
Human SIRP␣ Binds Weakly to mCD47-Soluble SIRP␣ binding to CHO cells is consistent with the cell adhesion results above, with mCD47-expressing CHO cells binding only 17 Ϯ 3% of what is bound to hCD47-expressing cells using intermediate SIRP␣ concentrations (Fig. 4B). Over the full range of SIRP␣ concentrations, the K d for binding to mCD47 is much weaker and is found to be 11.4 M, which is 16-fold higher than the apparent K d for hCD47 (Fig. 4C). To rule out glycosylation effects, core-glycosylated mCD47-GFP displayed on Lec1 cells allowed sufficient export to show that binding to human SIRP␣ has a similar affinity (not shown).
Interestingly, attempts to express mCD47-GFP without any N-linked sugars in CHO cells through tunicamycin treatment showed little to no export to the membrane. A variant of mCD47 with all six potential N-linked glycosylation sites knocked out genetically showed similar difficulty in export. Furthermore, lateral mobility of GFP-tagged mouse CD47 is similar to that of human CD47 (supplemental Fig. S2; data not shown), ruling out mobility as a factor contributing to differences in SIRP␣ binding.
CD47 Human-Mouse Chimeras and Multisite Mutants Reveal the Locus of Species Specificity-The reduced binding of soluble SIRP␣ to mouse CD47 combined with the 40% variation in the Ig domain primary sequence suggested an initial chimera strategy to coarsely identify the binding locus for SIRP␣. The entire Ig domain is a single structural entity, without subdomains or even exon boundaries, and so to minimize structural perturbations, chimeras were designed to contain just one discontinuity in primary sequence (i.e. human to mouse or mouse to human). Two highly conserved regions were chosen as junctions for parsing the Ig domain into three fragments of ϳ40 amino acids each (Fig. 5A). All chimeras were constructed with the membrane-spanning domain of human CD47. The expressed chimeras (denoted by H for human and M for mouse) were tested in binding to soluble SIRP␣ as well as several human and mouse anti-CD47 antibodies.
Binding of SIRP␣ depends on fragment II being human, since its replacement by mouse sequence decreases binding to 20% of the human level (Fig. 5, B and C). This is also seen for blocking antibody B6H12 and nonblocking antibody 2D3 (Fig. 5C, Table 2), implicating distinct epitopes within the ϳ40-amino acid fragment II of hCD47. In contrast, the mouse blocking antibody, mIAP301, binds fragment I (Table 2) so that one of the nested chimeras (M-H-H) contains both the mIAP301 and B6H12 epitopes, and both antibodies are able to effectively block human SIRP␣ binding. Steric inhibition with antibody binding to a proximal epitope is certainly possible in this system based on the glycosylation studies above (Fig. 2). One chimera created to try to confirm the importance of fragment II was H-M-H, but this contained two discontinuities in primary sequence and did not bind mIAP301 or B6H12, so it was excluded from further analysis.
Fragment II is clearly critical in the mouse versus human species specificity of SIRP␣ binding. It is possible that in addition to conserved amino acids in fragment II, conserved amino acids in fragments I and III also contribute to observed binding. We therefore exploited previously measured divergence in soluble SIRP␣ binding to pig and cow red blood cells (RBCs) that showed that pig RBCs bind to SIRP␣-like human RBCs, whereas cow RBCs show zero binding (25). These results with RBCs were reproduced here with CD47-displaying CHO cells; cow CD47 shows 1 Ϯ 1% of the binding seen with human CD47 (n ϭ 7).
Sequence comparisons of human, cow, and pig CD47 identify 13 key differences overall; eight aa differences in fragment II, four aa differences in fragment I, and a single aa difference in fragment III (Fig. 6A). These 13 changes were created in human CD47 (h⌬13cCD47) and showed essentially a cow CD47 phenotype: 5 Ϯ 2% of soluble SIRP␣ binding versus human CD47 (n ϭ 3). For technical reasons given below, we also made human CD47 with 12 of the above 13 changes (h⌬12cCD47; excluding K67E) and again found cowlike binding to SIRP␣: 3 Ϯ 1% of binding versus human CD47 (n ϭ 3). In contrast, constructs with either the single human-to-cow difference in fragment III or the four aa changes in fragment I were made and found to bind SIRP␣ the same as native human CD47 (Fig. 6, B and C).
Human-to-cow mutations within fragment II proved more interesting. First, however, two of the eight changes correspond to the removal or addition of N-linked sequons (N55R and D62N, respectively), so these changes were implemented in hCD47-GFP either individually or in combination (hFII ⌬2 cCD47 shown) and then tested for binding to soluble SIRP␣ and antibodies (Fig. 6, B and C; Table 3). As with the changes in fragment I and fragment III, neither of the two N-linked sequon changes in fragment II resulted in any decreased binding of SIRP␣ or antibody. Subsequently, in trying to assess all eight human residues in fragment II, we found that one of the eight mutations in fragment II (K67E) did not allow proper export, so a human-to-cow heptamutant that could be expressed, denoted hFII ⌬7 cCD47 (excludes K67E), was tested for binding to SIRP␣ (1 M). This key ⌬7 construct showed significantly reduced binding compared with the native human sequence (Fig. 6C); an apparent K a ϭ (2.3 M) Ϫ1 ϭ 0.43 M Ϫ1 obtained by inverting the dissociation constant is indeed less than one-third of the K a ϭ (0.7 M) Ϫ1 ϭ 1.4 M Ϫ1 for wild type. For the ⌬7 construct and all of the others, we verified that at least two anti-CD47 antibodies would show the same binding within error as wild type (e.g. 2D3 and 6H9 bind hFII ⌬7 cCD47), which is an accepted indicator of global folding (41). Proper folding is expected, because Ig domains tend to be relatively robust against multisite mutations (42) and because the mutations here are chosen based on the 70% human-cow identity (Fig. 1A).  In addition to binding soluble protein, cell spreading measurements show that contact areas with the hFII ⌬7 cCD47 mutant are again intermediate between those of hCD47 and mCD47 (Fig. 6D), consistent with an intermediate activity. For "medium" expression levels (300 molecules/m 2 ) of hFII ⌬7 cCD47, spreading is about one-third that of hCD47. The . Mutation of seven residues on human CD47 leads to large reduction in soluble SIRP␣ binding. A, alignment of human, pig, and cow CD47 Ig domains, with critical differences between pig and cow sequences marked with arrows. B, CHO cells expressing hCD47-GFP, hFI ⌬4 cCD47, hFIII ⌬1 cCD47, hFII ⌬2 cCD47, hFII ⌬7 cCD47, and mCD47-GFP were probed with soluble SIRP␣ complexed with fluorescent secondary antibody and analyzed by flow cytometry. C, relative binding of B6H12 and soluble human SIRP␣ to CHO-displayed hCD47-GFP, hFI ⌬4 cCD47, hFIII ⌬1 cCD47, hFII ⌬2 cCD47, hFII ⌬7 cCD47, and mCD47-GFP. Error bars, S.D. from three independent determinations. D, normalized contact area of CHO cells expressing high or medium levels of hCD47-GFP or hFII ⌬7 cCD47 after spreading on SIRP␣ for 20 min. Error bars, S.E. difference is very similar to that found above for binding soluble SIRP␣; thus, by at least two consistent measures, this humanto-cow chimera of 7 aa in fragment II contributes significantly to the locus of species-specific CD47-SIRP␣ interactions.
Locus Knock-out Abrogates Cell-Cell Adhesion-The physiological significance of the interactions identified above was evaluated in direct studies of cell-cell adhesion using a flow cytometric assay (43). Human monocytes express SIRP␣, so fresh human monocytes were fluorescently labeled with a membrane dye and then incubated with the various CHO cells to assess formation of cell-cell conjugates. Such conjugates are visible in optical microscopy (Fig. 7, upper inset) and were quantitatively assayed by flow cytometry. Only the CHO cells expressing human CD47 showed interactions with human monocytes above background levels (Fig. 7A, circled region). CHO cells expressing either the hFII ⌬7 cCD47 mutant or mouse-CD47 did not interact with human monocytes any more than control cells expressing just GFP. The number of CHO cells in clusters reduces to background levels (Fig. 7B) in the presence of soluble SIRP␣ or SE7C2, which is a SIRP␣-specific CD47 blocking antibody (6,23). These results indicate that reduced avidity limits cell-cell interactions and suggest that the 7 aa in CD47 fragment II are important to species-specific binding of human SIRP␣.

DISCUSSION
Broader Physiological Significance-CD47 is found on many cells and interacts with SIRP␣ on phagocytes, modulating many processes in mouse. However, the immunophysiological significance of this interaction in humans is less clear, especially since some severe deficiencies in CD47 on human RBCs show no evidence of enhanced macrophage or monocyte interactions (17)(18)(19). Our results are consistent with past reports that establish the human CD47-SIRP␣ interaction (6,20,21,23), but the data here show that human versus mouse interactions are biochemically distinct. Differences in affinity could influence downstream signaling, which is clear in other synapses, such as the immunological synapse between T cells and antigen-presenting cells (44) and also in the B cell synapse (45).
More generally, the relation between cell-cell adhesion and receptor-ligand affinities is not always clear (46). The results here reveal a nonlinear relationship for CD47-SIRP␣ (Fig. 7B). It is possible that a few mutations in one protein or the other would weaken the interaction and provide for species discrimination in this system.
Quantitative Determinations of Binding Affinities in a Cellular Context-As a general method here suited to flow cytometry, we have exploited GFP tagging of receptors for quantita-tive measurements of ligand affinities (or avidities) in transiently transfected cells. GFP allows monitoring of receptor export and membrane localization as well as determinations of lateral diffusion. Receptor density is readily calibrated so that in ligand binding assays one can easily normalize for the bound fraction and determine the K d for saturable and specific ligandreceptor interactions.
The monovalent interaction between human CD47-SIRP␣ Ig domains is reported to be K d(mon) Ϸ 2-8 M based on surface plasmon resonance measurements (21,47), so that our K d (ϭ 0.7 M) measured here probably reflects multivalent, physiologically relevant affinities in clusters (i.e. avidity). The simplest estimate for the average number n of molecules in each cluster on the membrane is n ϭ K d(mon) /K d Ϸ 7 and is consistent with estimates of the mean SIRP␣ per anti-GST cluster (see "Experimental Procedures"). High lateral mobility of CD47 (human, mouse, and hFII ⌬7 c mutant) in the CHO cell membrane (supplemental Fig. S2) would certainly allow such microclusters to form. No effect is seen in soluble SIRP␣ binding with agents that depolymerize the cytoskeleton (supplemental Fig. S1B), consistent with freely diffusible CD47 on the CHO cell. Importantly, these observations also hold for CD47 that is not tagged with GFP. Although the GFP tag has no influence on SIRP␣ binding (supplemental Fig. S1A), GFP-tagged CD47 has not been tested for signaling functions of native CD47.
Glycomodulation of Human CD47-SIRP␣ Interactions-CD47 is heavily glycosylated, with all five N-linked sequons probably occupied (34), but no definitive role of glycosylation had yet been determined. Core glycosylation would seem sufficient for the export of at least the soluble form of CD47, since it could be prepared in insect cells (48). Removal of all N-linked sugars does not affect the export of human protein, since expression levels are similar to those of wild type (Fig. 2), but mouse protein export is affected. Removal of sugars on human CD47 is found to lead to enhanced interactions with soluble SIRP␣ just as deglycosylated soluble SIRP␣ interacts more strongly with CD47 (Fig. 2), consistent with past reports (35).
These results are consistent with the notion that hyperglycosylation of either protein could suppress detectable association, as seen with B16 melanoma cells that express hyperglycosylated SIRP␣ (35). However, the role of sugar type cannot be ruled out, since B16 melanoma cells also overexpress ␤-1,6-N-acetylglucosaminyltransferase (GnT-V/Mgat5) that introduces ␤-1,6-GlcNAc-branched N-glycans strongly linked with increased cell invasiveness and metastatic potential (49 -51).
Beyond glycosylation, at least one other post-translational modification in CD47 has been reported to affect the affinity of CD47 for SIRP␣. A "long range disulfide" postulated between the CD47 Ig domain and transmembrane domain reportedly orients the Ig domain for optimal association so that knocking out the disulfide reduces the activity by about 50% (20). We made the same C15S mutant in our full-length hCD47, and this also showed reduced SIRP␣ binding in our system (not shown). Consistent with this, we found a similar reduction in binding to SIRP␣ when we expressed the Ig domain of hCD47 (not shown) as a fusion to the transmembrane domain of IL2R␣ (40). Posttranslational modifications thus modulate affinity but are not absolute requirements for the CD47-SIRP␣ interaction. Adhesion and Spreading Processes Involving CD47-CD47-SIRP␣ interactions can clearly mediate adhesion (4,6,25), but here we show that adhesion is followed by dynamic cell spreading (Fig. 3, A-C) that could contribute, for example, to cell-cell interactions between a T-cell and an antigenpresenting cell (52). SIRP␣ diffusion on cell surfaces but not on the coated surfaces here (e.g. Fig. 4A) would tend to foster clusters and favor cell-cell adhesion, but the trend is opposite here. In the context of cell-cell interactions, one intriguing aspect of monocytes that might explain the nonlinear trend with affinity ( Fig. 7) arises from the fact that phagocytes generally express both SIRP␣ and CD47 (6). Thus, any cis interactions between these two proteins would tend to inhibit trans interactions.
Phylogenetic Variation in CD47 Sequence Helps to Identify SIRP␣ Binding Locus-Protein binding partners involved in host defense, such as the classic KIR-MHC pair have co-evolved to maintain interaction specificity (53). The reported role of CD47 as a "marker of self " (14) prompted an evaluation of whether such specificity in interaction is maintained across species, especially given the considerable sequence variation observed in CD47 and SIRP␣. We recently demonstrated species specificity in SIRP␣ interactions with CD47 on red cells between humans and mice (25). The species specificity seen here with CHO-displayed CD47 (Fig. 4) seems not to be simple allor-none as previously reported (54,55); the interaction depends strongly on the density of molecules and their glycosylation status ( Figs.  1 and 2).
The primary locus of species specificity is in a 36-amino acid stretch between the cysteines forming the core disulfide in the Ig domain (fragment II). Comparison of pig-cow differences suggested 12 critical differences (five identical and one conservative substitution (Arg to Lys) between humans and pigs; two identical and one conservative substitution in fragment II) that seemed likely to explain the divergence in human SIRP␣ binding (Fig. 6). These 12 changes (humans to cows) reduced the apparent binding constant (K a ϭ K d Ϫ1 ) to human SIRP␣ by Ͼ98%. We tested the effects of seven changes (humans to cows) in fragment II and demonstrated a 70% reduction in the apparent binding constant for human SIRP␣. Only one of the identified seven amino acids (Asn 55 ) is conserved between mouse and human CD47, making results from both strategies consistent. The lack of any reduction in SIRP␣ binding when two glycosylation-related changes (N55R and D62N; hFII ⌬2 cCD47) were combined suggests that these two residues are probably less important for binding than the other five changes. Human to cow changes elsewhere (four in fragment I and one in fragment III) did not result in any measurable reduction in SIRP␣ binding. Single aa changes in fragment II also did not have any detectable effects, probably because the current assay is not capable of detecting Ͻ15% changes in soluble SIRP␣ binding (e.g. Fig. 2C) or because the seven key changes act cooperatively (synergistically) to bring about the reduction in soluble SIRP␣ binding. It is possible that select changes in fragment I or fragment III that are proximal to fragment II sites can reduce the SIRP␣ binding further to that seen with the h⌬12cCD47 construct.
Homology Models of the Binding Ig Domains of Human CD47 and SIRP␣-The recently solved crystal structure of the MOG Ig domain (29) (Protein Data Bank code 1PK0) provided an optimal starting template for a homology model of hCD47. MOG and hCD47 proteins are 17% identical and 29% similar, and the core cysteines in the two Ig domains bracket a very similar number of residues (71 for MOG versus 72 for CD47) with perfect alignment of a key, core tryptophan (Fig. 8A). Past models were based on rat CD2 (24) and show far bigger disparities. The initial homology model here for human CD47 was Cysteines involved in the disulfide bond and conserved tryptophan are marked with arrows. ␤-Strands in the template structure and per PSIPRED (57) for hCD47 are underlined. B, homology model of human CD47 showing ␤-sheets formed by strands ABED and GFCCЈCЉ in a schematic diagram embedded in a space-filling model. The 12 critical differences based on human, pig, and cow CD47 sequence comparison are highlighted as colored spheres, with changes in fragment 2 in red and other changes in green. The long range disulfide between Cys 15 and Cys 245 is schematically indicated as a yellow line. Models with identical topology of ␤-strands and similar overall structure were obtained for human Ig-CD47 using the Protein Data Bank structures of both coxsackievirus and adenovirus receptor (CAR; Protein Data Bank code 1EAJ chain A) and CD86/B7-2 (Protein Data Bank code 1I85 chain A) with root mean square deviation of C ␣ Ͻ 1.3 Å for modeled nonloop regions of the Ig domain. refined by molecular dynamics, and the resulting structure shows a typical immunoglobulin variable domain fold (Fig. 8B). The conserved tryptophan sits in a hydrophobic pocket between the two ␤-sheets (not shown), and Cys 15 orients toward the membrane (shown), as would be expected for it to participate in a disulfide bond with a cysteine on a transmembrane loop (20). Considerable solvent exposure (40 -80%) of the five asparagine residues that are experimentally known to be glycosylated in human CD47 (34) is also consistent with expectations.
Of the seven amino acids identified here to be critical for SIRP␣ binding to the CD47 Ig domain, a cluster of five charged amino acids (Arg 45 , Asp 46 , Lys 56 , Asp 62 , and Glu 69 ) are on the sheet formed by the ␤-strands GFCCЈCЉ. The other two are neutral and nearby (Ala 53 and Asn 55 ). Since two of the five charges (Lys 45 and Asp 46 ) are conserved in pig CD47, charge may contribute significantly to the speciesspecific interactions.
We also compared the sequences of the N-terminal Ig domains of human SIRP␣1 and SIRP␣2/BIT (brain immunoglobulin-like molecule with tyrosine-based activation motifs), both of which bind CD47, with that of human SIRP␤1, which does not bind CD47 (23) (Fig. S3A). Based on this comparison, seven residues are responsible for determining the CD47-binding phenotype. A homology model of the N-terminal Ig domain of human SIRP␣1 using a Fab as a template (Protein Data Bank code 1RHH; sequence alignment in Fig. S3B) was constructed ( Fig. S3C) with the seven residues highlighted, three of which are charged (Glu 2 , Asp 10 , and Gln 37 ). Five of the seven residues (Glu 2 , Thr 26 , Val 27 , Gln 37 , and Met 72 ) seem to cluster near the loop regions in the upper part of the structure, whereas the other two (Asp 10 and Pro 74 ) are separate and distal. The high sequence identity of the SIRP␣1 Ig domain to antibody V L regions points to the possibility of loop residues involved in binding CD47, with key differences identified here strongly modulating these loop interactions. Thus, SIRP␣-CD47 could interact like the CTLA4 and B7-2 (CD86) pair (56), with loops in SIRP␣ interacting with the strands in CD47.