Cholesterol-independent Interactions with CD47 Enhance α_vβ₃ Avidity^*

Integrins form complexes with other transmembrane proteins that can modulate their functions. Lateral integrin partners include growth factor receptors, tetraspanins, and CD47, an Ig family protein also known as integrin-associated protein (IAP) (1). CD47 is characterized by a single extracellular IgV domain, a 5-TM¹ region known as the multiple membrane-spanning (MMS) domain, and a short cytoplasmic tail that is alternatively spliced (2, 3). CD47 is both physically and functionally associated with β₃ and β₁ integrins (4, 5), and mediates a range of adhesion-related processes. Ligation of CD47 with thrombospondin (TSP), or a peptide, 4N1K, derived from its C-terminal domain, augments cell spreading and migration on α_vβ₃ substrates and α_IIbβ₃- and α₂β₁-mediated platelet aggregation (6, 7). These actions are inhibited by pertussis toxin treatment, indicating a role for heterotrimeric G_i in CD47 signaling (8). In fact, detergent-stable complexes of CD47 and G_iα with both α_vβ₃ and α_IIbβ₃ have been isolated, suggesting that CD47 and the integrin may combine to form a functional 7-TM complex that functions like a G protein coupled receptor (4, 6, 8).

The α_vβ₃-CD47-G_i complex has also been found in Triton X-100-resistant membrane domains, or rafts, which are thought to act as organizing platforms for diverse signaling pathways, including those involving growth factors, G protein-coupled receptors, and T cell activation (9, 10). Depletion of membrane cholesterol was shown to destabilize the α_vβ₃-CD47-G_i complex in C32 melanoma and ovarian carcinoma cells, suggesting that intramembrane contacts controlling its localization to rafts may promote or even be required for complex assembly and G_i-coupled signaling (9). CD47-mediated signaling also occurs in Jurkat T cells, where treatment with certain anti-CD47 antibodies promotes apoptosis via a G_i-mediated pathway involving cyclic AMP regulation of cAMP-dependent kinase (11). Ligation of CD47 with other antibodies has been shown to have a synergistic effect on T cell activation that required the MMS domain of CD47 for both the T cell response and for raft localization (10). Thus raft localization may play a role in organizing CD47 signaling complexes in a number of systems.

Recently, effects of CD47 on β₃ integrin functions have been reported that appear to be independent of downstream signaling. α_IIbβ₃-mediated platelet aggregation was stimulated by TSP peptide in energy-depleted cells shown to be deficient in signaling (12). Furthermore, 4N1K peptide-induced activation of α_IIbβ₃, as measured by binding of PAC-1 antibody to transfected cells, required only the presence of the extracellular IgV domain of CD47, added either as a soluble protein or expressed with a generic membrane anchor (12). In addition, a previous study had shown that expression of the CD47 extracellular IgV domain in ovarian carcinoma cells was required for α_vβ₃-mediated binding of vitronectin-coated beads (13). These data indicate that the physical association of only the IgV domain of CD47 can affect β₃ integrin function. Thus CD47 may influence β₃ functions through interactions that involve different regions of the molecule. While it seems that signaling-dependent actions of CD47 may require cholesterol-rich domains, the possible role of membrane domains in more direct effects on the conformation of its associated integrin is unknown.

In this study, we have examined the ability of CD47 association to promote activated states of α_vβ₃, and addressed the role of detergent resistant membrane domains in CD47 modulation of integrin function. CD47 interactions with α_vβ₃ were found to promote both integrin activation induced by RGD peptide, and α_vβ₃ avidity in binding immobilized substrates by enhancing integrin clustering. These effects resulted from CD47-α_vβ₃ complexes that were resistant to cholesterol depletion, although they still resided in Triton-insoluble membrane domains. These CD47-α_vβ₃ complexes were distinct from those localized in cholesterol-rich domains, which were disrupted by cyclodextrin treatment, and required the MMS segment of CD47 for integrin and G protein association. Thus, both the physical nature of CD47-α_vβ₃ complexes and their effect on integrin functions can be influenced by their location in different membrane microdomains.

MATERIALS AND METHODS

Reagents and Cell Lines—The human ovarian carcinoma cells, known as OV10, were maintained as previously described (4). The stable transfections of α_vβ₃, CD47, and IgV-GPI have been previously reported by Lindberg et al. (13), and the cells used in this study were derived from the same lines. For all experiments, OV10 cells were previously treated with serum-free media for 16-18 h. Cell manipulations were carried out in Hanks Balanced Salt Solution containing 20 mm HEPES pH 7.5 (HBSS) with additional reagents for each application. Anti-β₃ mAbs AP3 (14) and 7G2 (15), and anti-CD47 2D3 and B6H12 (16) have been previously described. The anti-α_vβ₃ mAb LM609 and anti-α_vβ₅ mAbs P1F6 were generous gifts of Dr. David Cheresh (Scripps Research Institute), and the two ligand-induced binding sites (LIBS) antibodies, LIBS1 and LIBS6, were generously provided by Dr. Mark Ginsberg (Scripps Research Institute). Polyclonal antibodies for Western blots of integrin β₅ subunit, G_iα, Gβ, Src, Lyn, and FAK were obtained from Santa Cruz Biotechnology, and the mAb for SHP-2 was purchased from Transduction Laboratories. FITC-anti-mouse IgG (Sigma Chemical Co.) was employed as the secondary antibody for flow cytometry experiments, and horseradish peroxidase conjugates of antirabbit IgG and anti-mouse IgG (Jackson Laboratories) were used for Western blots. The peptides 4N1K (KRFYVVMWKK), 4NGG (KRFYGGMWKK), and GRGDSP were synthesized as previously described (17). BCECF-AM was purchased from Molecular Probes, and BS³ was from Pierce Chemical Co. Optiprep™ was obtained from Axis-Shield (Oslo, Norway). All other reagents were purchased from Sigma Chemical Company unless otherwise stated.

Cholesterol Depletion of OV10 Cells—5 × 10⁶ cells were brought up at 1.25 × 10⁶/ml in serum-free Iscoves media containing 0.1% fatty acid-free BSA with or without 10 mm methyl-β-cyclodextrin (Aldrich Chemicals) and incubated on a rocker at 37 °C for 20 min. This procedure has been shown to reduce total cellular cholesterol by 40% in OV10 cells (9). After washing, cells were resuspended in appropriate media or buffers for cell adhesion, lysis, or FACS staining.

Flow Cytometry—Expression of CD47, β₃, α_vβ₃, and LIBS epitopes in OV10 cells was measured by flow cytometry. 2 × 10⁴ cells were resuspended in 50 μl of FACS buffer (1% BSA, 2 mm MgCl₂, 1 mm CaCl₂ in HBSS) with 10 μg/ml of primary antibody and incubated on ice for 30 min. For detection of LIBS binding, GRGDSP was included at 0.1-2mm, and the incubation was carried out at 37 °C for 15 min followed by 15 min on ice. Cells were washed and incubated in 50 ml of the same buffer containing goat anti-mouse IgG-FITC (Fc-specific) at a 1:50 dilution for an additional 30 min. After washing, the cells were diluted to 0.4 ml and analyzed on a flow cytometer using the Cell Quest software program.

Cell Adhesion—A rapid, non-washing, fluorescence-based assay (18) was employed to measure cell adhesion at an early time point before cell spreading occurred. OV10 cells were resuspended at 1.25 × 10⁶/ml in 5 ml of HBSS containing BCECF, and incubated on a rocker at 37 °C for 20 min. Cells were washed, resuspended in 4 ml of binding buffer (1.5% BSA, 2 mm MgCl₂,1mm CaCl₂ in HBSS) at 0.2 × 10⁶/ml, and dispensed into V-well plate at 100 μl/well. The plate was incubated at 37 °C for 10 min, and immediately centrifuged at 1000 × g for 10 min at 4 C. Unbound cells were detected from triplicate fluorescence measurements (fluorescein parameters) of cells at the bottom of wells as measured with an FLX800 plate reader (Bio-Tek Instruments, Inc.). Total cell fluorescence was determined from measurements of cells in uncoated wells. Percent cell binding was calculated based on the total cell fluorescence for each cell type and the unbound cell fluorescence for individual samples.

Isolation of Detergent-stable Complexes—Monoclonal antibodies AP3 and B6H12 were coupled to sheep anti-mouse 4.5-μm magnetic beads, and vitronectin was coupled to 4.5-μm tosyl-activated magnetic beads (Dynal) according to the manufacturer's protocols. 5 × 10⁶ cells were resuspended in 1 ml of HBSS with 0.1 mm MnCl₂, and bound to substrate-coated magnetic beads (2 × 10⁷) by mixing at room temperature for 20 min. Unbound cells were removed in the supernatant after placing the sample in a magnet, and the beads were washed 1× in the same way. Bead-bound cells were taken up in 1 ml of lysis buffer (HBSS containing 10 mm CHAPS, 250 mm sucrose, 0.1 mm MnCl₂, and protease inhibitor mixture), and mixed for 10 min at room temperature. Beads were resuspended in 20 μl of HBSS with 2.5 units of DNase I and incubated for 5 min at room temperature. Complexes were extracted with 2× sample buffer at 65 °C for 15 min and analyzed by SDS-PAGE and Western blotting.

Cross-linking of β₃ Aggregates—Cells were bound to AP3 or vitronectin-coated magnetic beads as described above. After washing, bead-bound cells were resuspended in 1 ml of HBSS containing 0.5-5 mm BS³, and mixed at 4 °C for 90 min. The beads were resuspended in 1 ml of phosphate-buffered saline containing 0.1 m glycine and 1% BSA, and mixed at room temperature for 15 min. The isolated beads were then treated with lysis buffer and DNase I as described above, followed by analysis of β₃ integrin by SDS-PAGE and Western blotting.

Detergent-free Isolation of Rafts—The preparation was carried out according to the method of Smart et al. (19). Approximately 100 × 10⁶ cells were resuspended in 1 ml of ice cold buffer A (20 mm Tris, pH 7.8, 0.25 m sucrose, 1 mm EDTA) and lysed with 20 strokes of a Dounce homogenizer, followed by extrusion through a 20-gauge needle 20×. The lysate was centrifuged at 1000 × g for 10 min, and the supernatant was collected. The pellet was subjected to the same homogenization procedure just described, and the supernatants were combined and brought up to 2 ml with buffer A. The sample was layered over 8 ml of 30% Percoll in buffer A, and centrifuged in a 75 Ti rotor on a Beckman L8-80 M Ultracentrifuge at 84,000 × g. The plasma membrane fractions normally migrated near the middle of the gradient as detected by Western blotting for transferrin receptor (not shown), and could be visually distinguished as a discrete band. This region was collected by pipette and brought up to 2 ml with buffer A. To this sample was added 1.84 ml of buffer C (50% Optiprep in 120 mm Tris, pH 7.8, 0.25 m sucrose, 6 mm EDTA) + 0.164 ml buffer A, over which was layered an 8-ml Optiprep gradient of 10-20% (made from dilutions of buffer C in buffer A). The sample was centrifuged in an SW41 rotor at 52,000 × g for 90 min. The top 5 ml containing the light membranes were collected and mixed with 4 ml of buffer C. The lower 4 ml were pooled and taken as the raft-depleted membranes. The light membrane mixture was placed in an ultracentrifuge tube, over, which was layered 3 ml of a 4% Optiprep solution (made by diluting buffer C with buffer A). The sample was centrifuged in the SW41 rotor as before to produce a membrane band clearly visualized at the gradient interface. This material was collected by pipette and taken as the purified raft fraction.

Separation of Triton-soluble from Triton-insoluble Cell Fractions— This procedure was similar to the method described by Maile et al. (20), and is based on the association of detergent-resistant domains with the cytoskeletal pellet (21). 5 × 10⁶ cells were lysed in 1 ml of ice-cold lysis buffer (0.2% Triton X-100 in HBSS containing 2 mm MgCl₂,1mm CaCl₂, and protease inhibitor mixture) and clarified by centrifugation at 14,000 × g for 10 min at 4 °C. The supernatant was collected and taken as the Triton-soluble fraction. The pellet was resuspended at room temperature in 1 ml of HBSS containing 1% Nonidet P-40, 0.25% sodium deoxycholate, 2 mm MgCl₂,1mm CaCl₂, and protease inhibitors, followed by centrifugation as before. This supernatant was collected and taken to be the Triton-insoluble fraction, with the remaining pellet containing the insoluble cytoskeleton. SDS-PAGE and Western blotting were carried out on 50-μl aliquots of the Triton-soluble and Triton-insoluble fractions for each condition.

RESULTS

Binding of LIBS Antibodies to α_vβ₃ Is Enhanced by CD47 Expression—To determine if CD47 expression can have a direct influence on α_vβ₃ activation, binding of antibodies recognizing LIBS was measured for OV10 cells co-transfected with CD47 and β₃ integrin. The transfection and selection of clones for the cell lines employed in this study have been previously reported (13). Transfection of these cells with only β₃ chain results in expression of α_vβ₃ as a result of assembly with endogenous α_v. This does not result in overexpression because the level of α_vβ₃ expression is limited by the level of α_v and is near the level of α_vβ₅ expressed in wild-type OV10 cells (9). CD47 was expressed in these cells in its wild-type form and as a fusion protein containing the IgV domain linked to a glycophosphatidylinositol anchor (IgV-GPI) (13). As compared with wild-type CD47, the IgV-GPI species has been shown to be deficient in downstream T cell signaling induced by CD47 ligation (22) (10). On the other hand, as discussed above, IgV-GPI supported vitronectin-coated bead binding when co-expressed with αvβ3, and so may serve as an indicator of signaling-independent effects of CD47 (13). As measured by flow cytometry (Fig. 1, A and B), expression levels of wild-type CD47 and IgV-GPI were within 5% of each other, and levels of α_vβ₃ were also comparable among the cell lines employed, with a general order of IgV-GPI > CD47-> CD47+ (Fig. 1B). To correct for the small differences in integrin expression, anti-LIBS mAb binding was normalized to the binding of non-LIBS antibody LM609. Based on the LIBS1/LM609 binding ratio, neither wild-type CD47 nor IgV-GPI activated α_vβ₃ in the absence of RGD or Mn²⁺ (not shown). However, RGD-induced binding of LIBS-1 was enhanced with expression of either CD47 species (Fig. 1, A and C), with IgV-GPI producing the largest effect. Binding of LIBS6 was also increased with CD47 expression, and again IgV-GPI produced a larger increase in the maximal level of RGD-induced binding compared with wild-type CD47 (Fig. 1D). The enhanced binding appeared to result from an increase in the maximal binding at saturating RGD concentrations (Fig. 1E). Thus surface expression of the IgV domain of CD47 resulted in an increase in the number of “activable” integrin molecules, or the fraction of α_vβ₃ susceptible to stimulation by RGD peptide. The thrombospondin C-terminal peptide 4N1K was also tested for an ability to induce LIBS binding, but at concentrations up to 100 μm, no effect was observed either in the presence or absence of RGD for any cell line (not shown). Because cholesterol has been shown to be a component of the CD47-integrin complex (9), the effects of cholesterol depletion on activation of LIBS epitopes was examined. Rather than reducing anti-LIBS binding to CD47-expressing cells, cholesterol treatment had only small effects on binding of LIBS1 or LIBS6 for any of the cell lines, and differences in relative levels of anti-LIBS binding with CD47 or IgV-GPI expression was unchanged by cyclodextrin treatment (Fig. 1, C and D). Thus membrane cholesterol depletion did not disrupt the CD47-mediated changes in α_vβ₃ that are responsible for anti-LIBS binding. That IgV-GPI induced a larger change in binding of LIBS antibodies than wild-type CD47 may indicate that transmembrane contacts might even constrain the ability of the extracellular domain of CD47 to induce conformational changes in α_vβ₃. In any case, the stimulation of RGD-induced binding of LIBS1 and LIBS6 indicated that IgV-integrin interactions promoted a conformational change in α_vβ₃ that is associated with its activation.

CD47 Enhances α_vβ₃-mediated Adhesion through Cyclodextrin-resistant Effects—To evaluate the effects of CD47 on α_vβ₃- mediated cell adhesion, a rapid binding assay was used to measure OV10 cell attachment to V-well plates coated with vitronectin at high (5 μg/ml) and low (1 μg/ml) densities. At the high density, a high percentage of cells (65-80%) were bound irrespective of CD47 expression, and for all cell lines, treatment with RGD peptide reduced the binding to low levels (3-7%). Thus the assay was specific for RGD-containing substrates. Because the OV10 cells can bind to vitronectin through both α_vβ₃- and α_vβ₅-mediated adhesion, the α_vβ₅-blocking antibody P1F6 (4) was employed to determine α_vβ₃-specific effects. Cell attachment at the high ligand density showed no effect of either wild-type CD47 or IgV-GPI expression, and no major reduction was observed in cells treated with P1F6 (Fig. 2A). Thus, at this level of vitronectin, α_vβ₃-mediated adhesion supported nearly maximal cell attachment that was not reliant on CD47 expression. However, with cyclodextrin treatment, binding of CD47-deficient cells was reduced to about 20%, but cells expressing wild-type CD47 or IgV-GPI bound at levels of about 40 and 60%, respectively (Fig. 2A). While it is not known precisely how cholesterol depletion disrupted cell adhesion, alteration of the membrane environment could in theory influence either the ligand binding conformation of both α_vβ₃ and α_vβ₅, or the clustering of integrin molecules, as has been reported for α₅β₁ and LFA-1 (23). However, expression of either CD47 or IgV-GPI partially protected against this effect (Fig. 2A). Co-treatment with P1F6 and cyclodextrin did not further reduce the binding of the cells compared with cyclodextrin alone (Fig. 2A), indicating that α_vβ₃-mediated adhesion predominated in the cholesterol-depleted cells, and the influence of CD47 was exerted on that integrin. Treatment with LM609 to block α_vβ₃-mediated adhesion did not reduce cell binding (Fig. 2A), indicating that α_vβ₅ alone also supported adhesion at a high vitronectin density. However, blocking α_vβ₃ in cyclodextrin-treated cells reduced binding to a low level irrespective of CD47 expression (Fig. 2A), indicating that α_vβ₅-mediated adhesion was not rescued from the disruptive effects of cholesterol depletion. Thus expression of either wild-type CD47 or IgV-GPI protected the cells against the disruption of cell binding caused by cholesterol depletion by enhancing α_vβ₃-mediated adhesion, but not adhesion conferred by α_vβ₅.

At the low ligand density, cell binding was reduced to about one-fourth of the high vitronectin density level for all three cell lines, and was blocked in all cases with RGD peptide (Fig. 2B). Whereas no effect of wild-type CD47 or IgV-GPI expression was detected in untreated cells, with P1F6 treatment, cell binding to vitronectin was reduced to a low level in CD47-deficient cells (Fig. 2B). Thus, at low substrate conditions, α_vβ₃-mediated binding was not supported, and α_vβ₅-mediated adhesion predominated. However, for cells expressing wild-type CD47 or IgV-GPI, only moderate or no reduction of binding was observed with P1F6 treatment (Fig. 2B), indicating that when α_vβ₅-mediated binding was blocked at low substrate conditions, the remaining α_vβ₃-mediated adhesion was CD47-dependent. For cyclodextrin-treated cells at the low vitronectin density, binding of CD47-deficient cells was reduced to a low level, and wild-type CD47 conferred only minor protection (Fig. 2B). However, binding of IgV-GPI-expressing cells was resistant to cholesterol depletion, and co-treatment with P1F6 and cyclodextrin did not result in a further significant disruption of cell adhesion (Fig. 2B), indicating the residual adhesion was because ofα_vβ₃. Thus IgV-GPI exhibited an even greater protective effect against the cyclodextrin-induced disruption of cell binding than wild-type CD47. Blocking α_vβ₃-mediated binding with LM609 treatment reduced binding to a lower level in all the cell lines. However, this amount of α_vβ₅-mediated binding was eliminated with cholesterol depletion with no effect of CD47 expression (Fig. 2B), indicating that a CD47-independent effect of cholesterol depletion on α_vβ₅-mediated binding was displayed at the low ligand density.

In summary, expression of either wild-type CD47 or IgV-GPI was shown to enhance α_vβ₃-mediated adhesion, as detected both by an increase in binding of cells at low ligand density (Fig. 2B) and by protection against cyclodextrin-induced disruption of cell binding conferred at both high and low substrate conditions (Fig. 2A).

CD47 Enhances α_vβ₃ Avidity for Binding Immobilized Substrate—The effect of CD47 or IgV-GPI expression to increase the fraction of integrins activated upon treatment with saturating RGD peptide (Fig. 1E) suggests that their interactions with α_vβ₃ can shift the integrin conformational equilibrium toward the activated state. However, enhanced cell adhesion could also be derived from an increase in avidity because of changes in the aggregation state of α_vβ₃ that result in cell-matrix contacts with a higher valency. To determine if expression of CD47 enhances α_vβ₃ avidity, cell attachment to plates coated at high and low densities of anti-β₃ antibody AP3 was measured. AP3 ligation of integrin molecules should occur at a constant antibody-antigen affinity, thus any difference in cell binding must be derived from an increase in α_vβ₃ avidity. At the higher AP3 density, no significant differences in cell adhesion were observed among the three cell types, with each displaying the same maximal level of binding. At the lower antibody density, binding of CD47-deficient cells was reduced by 80%, but binding of cells expressing wild-type CD47 expression was reduced by only about 20%, and IgV-GPI-expressing cells bound to the low density of AP3 to the same extent as on high density AP3 (Fig. 3A). Thus, expression of either wild-type CD47 or IgV-GPI increased the binding of α_vβ₃ to immobilized AP3 antibody, presumably by increasing the valency of cell surface contacts by enhancing integrin clusters. As with vitronectin binding, the effects of CD47 became superfluous at higher ligand densities (Fig. 2A).

We next examined the effects of cholesterol depletion on α_vβ₃ avidity. Membrane cholesterol levels have been reported to affect the clustering of α₅β₁ in rat fibroblasts (23), and in T-cells, via the GPI-anchored protein CD24 (24). Because we have shown that anti-LIBS binding is not significantly affected by cyclodextrin treatment (Fig. 1, C and D), its ability to disrupt cell adhesion to vitronectin (Fig. 2, A and B) may have been based in part on an inhibition of integrin clustering. At the high AP3 density, we observed a dramatic decrease in binding of CD47-deficient cells, consistent with a disruption of integrin clustering (Fig. 3A). However, nearly twice the level of binding was observed for wild-type CD47-expressing cells as for deficient ones, and IgV-GPI expression provided nearly complete protection against the cyclodextrin-induced disruption (Fig. 3A). Thus, as was also observed for vitronectin binding, expression of IgV-GPI gave an even greater enhancement of AP3 binding than wild-type CD47, and was more resistant to cholesterol depletion (Fig. 3A). For cell binding to AP3 at a low coating concentration, cholesterol depletion was completely disruptive in all the cell types, suggesting that CD47 expression could not compensate for the decreased valency resulting from both the reduced clustering induced by cyclodextrin and the low substrate density. We conclude from these experiments that interactions of the extracellular domain of CD47 with α_vβ₃ enhance binding of the integrin to immobilized substrates by promoting a greater degree of clustering to increase integrin avidity.

4N1K Treatment Promotes IgV-α_vβ₃ Interactions That Enhance Integrin Avidity—Recently, the thrombospondin C-terminal peptide 4N1K was reported to induce α_IIbβ₃-mediated activation in signaling-deficient cells via CD47 (12), suggesting that the peptide may interact with CD47 to produce signaling-dependent and -independent effects on β₃ integrins. Furthermore, a recent report has shown that treatment with a similar peptide can increase α_vβ₃-mediated adhesion (25). However, we saw no effect of 4N1K on the binding of LIBS antibodies to OV-10 cells expressing either wild-type or IgV-GPI, either alone or with GRGDSP treatment (not shown), indicating that the peptide did not influence RGD-induced activation. We therefore tested the ability of 4N1K to enhance α_vβ₃ avidity by measuring OV-10 cell binding to AP3-coated plates, at coating concentrations that supported only nonspecific attachment by untreated cells. Although 4N1K has been reported to have CD47-independent effects on integrin functions in some situations (25, 26), no increased binding with peptide treatment was observed for CD47-deficient cells (Fig. 3B). However, enhanced binding was indeed observed with 50 μm 4N1K treatment of OV-10 cells expressing wild-type CD47 and IgV-GPI (Fig. 3B). Treatment with the control peptide 4NGG did not promote increased binding, and 4N1K treatment did not affect cell binding to uncoated, blocked wells (Fig. 3B). Thus, 4N1K peptide association with CD47 must stabilize a conformation of the IgV domain that interacts with α_vβ₃ to promote integrin clustering. Cyclodextrin treatment resulted in reduced background binding to less than 5%, as well as blocking the 4N1K effect (not shown), suggesting that its disruptive effects on avidity could not be overcome with peptide treatment.

CD47 Forms Cyclodextrin-resistant Complexes with α_vβ₃ through Its Extracellular Domain—Previous studies have reported a CD47-α_vβ₃ complex that is sensitive to cholesterol depletion (9), but the effects on integrin avidity that we report are not. We used immunoadsorption assays in detergent lysates with and without cyclodextrin treatment to distinguish direct association from co-localization in cholesterol-rich rafts. Cells expressing wild-type CD47 were allowed to bind magnetic beads coated with anti-β₃ mAb AP3 and lysed in CHAPS. Complexes were rapidly isolated and analyzed for signaling molecules associated with rafts (27). Indeed, Src, Lyn, SHP-2, and G_i were detected in association with the αvβ3 complexes (Fig. 4A). Prior cyclodextrin treatment virtually eliminated all of these associations, an indication of the loss of integrity of CHAPS-resistant rafts (Fig. 4A). The level of wild-type CD47 was also reduced, indicating that a significant fraction of the CD47 was released with loss of raft integrity, and was thus likely to be indirectly associated with the integrin. Adsorption of complexes with anti-CD47-coated beads more clearly demonstrated β₃ association that was cyclodextrin-resistant in either CHAPS (Fig. 4B, left panel) or Triton X-100 (Fig. 4B, right panel). Thus α_vβ₃ appeared to reside in several membrane compartments, with some associated with CD47 in cyclodextrin-sensitive rafts, some associated in cyclodextrin-resistant complexes, and some unassociated with CD47. Adsorption of IgV-GPI displayed cyclodextrin-resistant β₃ association in CHAPS lysates (Fig. 4B), indicating that the extracellular domain of CD47 was sufficient to bind α_vβ₃. In contrast, although integrin β₅ subunit and both forms of CD47 were clearly detected in cell lysates, β₅ was barely detected in CHAPS immunoprecipitates adsorbed with either wild-type CD47 or IgV-GPI, and no trace at all was observed with cholesterol depletion (Fig. 4C). These observations are consistent with the previous report that α_vβ₅ is present but not enriched in cholesterol-rich rafts, and the lack of association of this integrin with CD47 (9) (13). IgV-GPI adsorption from Triton X-100 lysates was greatly reduced upon cholesterol depletion, and β₃ was not detected (Fig. 4B, right panel), indicating that the stability of the complex in this detergent displayed a cholesterol dependence that was not observed in CHAPS solutions. However, detection of G_iα with wild-type CD47 was largely resistant to cyclodextrin treatment in complexes isolated with either detergent (Fig. 4B, left panel), indicating its association was not strictly based on raft integrity and was probably direct. On the other hand, G_iα was only found with IgV-GPI in CHAPS lysates, and its detection was completely cyclodextrin-sensitive (Fig. 4B, left panel), indicating that the MMS region is required for G protein association. In summary, analysis of immunoadsorbed complexes showed CD47 association with α_vβ₃ that was cyclodextrin-resistant and required only the IgV domain, in contrast to raft-localized signaling molecules and G proteins.

CD47 Promotes Formation of High Molecular Weight β₃ Species That Can Be Chemically Cross-linked—CD47-dependent effects on avidity are presumed to influence clustering of α_vβ₃ and should result in closer integrin-integrin contacts upon oligomerization. To detect oligomeric α_vβ₃, chemical cross-linking of cells with adsorbed anti-β₃ coated magnetic beads was employed, using the membrane-impermeable reagent BS³. Detection of cross-linked species by this method requires close proximity of surface-exposed lysines on associated proteins, which also must be in a favorable orientation. Treatment with BS³ before bead binding resulted in detection of only monomeric β₃ (Fig. 5A, lanes 1 and 4), indicating that cross-linked species were not formed in the absence of antibody engagement. Furthermore, this particular reagent did not cross-link the α_v and β₃ subunits under these conditions. However, upon BS³ treatment following bead binding to the cells, but before cell lysis, higher order β₃-containing species were detected by Western immunoblotting of the bead-bound material (Fig. 5A). At least three types of complexes with a molecular weight greater than monomeric β₃ were detected. For CD47-deficient cells, an ∼300 kDa species was detected at a significant level, and was further enriched with a higher concentration of BS³ (Fig. 5A, lanes 2 and 3). Trace amounts of an approximately ∼200 kDa species and a very high molecular weight band that barely migrated were also detected, and the larger species were not increased at the higher concentration of cross-linker (Fig. 5A, lanes 2 and 3). In contrast, for CD47-expressing cells, the ∼300 kDa species was absent, but instead, very high molecular weight species were detected at significant levels even at the lower concentration of cross-linker (Fig. 5A, lane 5). These species were further enriched at the higher concentration, and only then did the ∼300 kDa species appear (Fig. 5A, lane 6). Thus CD47 expression promoted the formation of the highest molecular weight β₃-containing aggregates. We also used beads coated with vitronectin, a natural ligand of α_vβ₃. Once again, the very high molecular weight species were present only in CD47-containing complexes (Fig. 5A, lanes 9 and 10). However, the ∼300 kDa species was absent in both cell types (Fig. 5A, lanes 7-10) suggesting that it might be a complex of β₃ with the AP3 antibody. Treatment of cells with cross-linker before binding of cells to vitronectin-coated beads also did not produce higher molecular weight species (not shown). Thus, formation of the very large integrin aggregates required the presence of CD47. CD47 itself was only detected in monomeric form when treated with the same conditions (not shown), indicating it may not be accessible to direct covalent linkage with integrin. While the detection of cross-linked species indicates a protein-protein interaction, lack of detection does not rule it out, and could be because of poor orientation of lysines for cross-linking, destruction of the antibody epitope by BS³, or lack of reagent accessibility to the protein-protein contacts. Cross-linked complexes from IgV-GPI-expressing cells adsorbed to anti-β₃ antibody AP3 were similar to those of wild-type CD47, with formation of the very high molecular weight β₃ species and no detection of the ∼300 kDa species (Fig. 5B). Thus, formation of the large β₃ aggregates was mediated by the extracellular IgV domain of CD47 alone. Cyclodextrin treatment inhibited formation of cross-linked complexes in CD47-expressing cells, however, with IgV-GPI expression, cholesterol depletion resulted in complexes that were not significantly different compared with untreated cells (Fig. 5B). Thus, as suggested in the cell adhesion assays, promotion of β₃ aggregates by IgV-GPI was more resistant to cholesterol depletion than were aggregates formed by CD47.

MMS Region of CD47 Is Required for Targeting α_vβ₃ to Cholesterol-rich Rafts—To determine the dependence on CD47 of αvβ3 localization to cholesterol-rich membrane microdomains, we utilized detergent-free membrane preparations (19). This method has been used to isolate ordered lipid domains that display the greatest cholesterol enrichment, and are free of detergent-related artifacts (28). Purified plasma membranes were fractionated on Optiprep™ gradients, resulting in a small protein peak floating near the top (rafts) that separated from the major protein peak near the bottom (raft-depleted membranes), with a difference in the cholesterol/protein ratio of about 10-fold (Fig. 6A). These rafts are cholesterol-rich, and are distinguished from the more general “detergent-resistant domains” (DRMs). The large majority of the total plasma membrane protein resided in the lower half of the gradient, with the light fractions representing about 2% of the protein loaded (Fig. 6A). Cyclodextrin treatment resulted in elimination of the rafts, while not altering recovery of the bulk plasma membrane fractions (not shown). Western blots revealed that, for equal amounts of protein, the raft fraction was enriched in G_iα and Gβ (Fig. 6B), consistent with the known targeting of these molecules to ordered domains by lipid modifications. Wild-type CD47 and IgV-GPI were found in both raft and raft-depleted membrane fractions (Fig. 6C), and levels of β₃ were the same in raft-depleted membranes from all three cell lines (Fig. 6C, right). However, upon analysis of equal amounts of protein from isolated rafts, β₃ was present in significant amounts only when the rafts were derived from cells expressing wild-type CD47 (Fig. 6C, left). Expression of IgV-GPI did not result in a β₃/CD47 ratio in rafts that was significantly different from CD47-deficient cells (Fig. 6C). Thus IgV-integrin interactions alone were not sufficient for targeting integrin to cholesterol-rich domains, indicating that transmembrane or cytoplasmic contacts in the wild-type CD47-α_vβ₃ complex are required for the integrin to partition into the light membranes. Because localization of CD47-α_vβ₃ complexes in these domains has been found to be required for the G protein-coupled cell spreading stimulated by thrombospondin-based agonists (9), this finding supports the model in which integrin association is required for downstream signaling (8). In T cells, IgV-GPI was not able to support downstream effects that included increases in [Ca²⁺]_i, actin polymerization and PKCθ association with the cytoskeleton, that were correlated with synergistic activation with the TCR induced by CD47 ligation (22) (10). Thus, the MMS domain appears to be required for CD47 signaling in both of these systems. However, it is clear that IgV-GPI does in fact associate with α_vβ₃ (Fig. 4B) and alters its avidity (Fig. 3, A and B), although this must occur in heavy membrane fractions, which are not rich in membrane cholesterol (Fig. 6A). This conclusion is consistent with the observation that, as noted above, the heavy membrane region of the gradient contained both wildtype CD47 and IgV-GPI in the same relative enrichment as the raft fractions, as well as the majority of the α_vβ₃ (Fig. 6C). Thus, interactions of the IgV domain of CD47 with α_vβ₃ that influence integrin activation and avidity do not require localization to cholesterol-rich rafts, in contrast to downstream signaling which also requires the MMS region of CD47.

CD47 and IgV-GPI Associate with α_vβ₃ in Cyclodextrin-resistant Membrane Domains—While IgV-GPI interactions with αvβ3 could not take place in cholesterol-rich rafts (Fig. 6C), their localization in other types of DRMs may mediate the effects of CD47 on integrin function. It is becoming increasingly clear that membrane microdomains are heterogeneous (29). Recent studies have found that raft isolation based on detergent insolubility produces a more heterogeneous mixture of membrane domains than detergent-free preparations (30). In fact, detection of CD47 in light fractions from gradients of detergent lysates has been shown to be unaltered by cyclodextrin treatment (9), and other studies have reported the existence of similar cyclodextrin-resistant domains (31) (32). Recently, it was demonstrated that CD47 can move out of rafts by following its partitioning between Triton-soluble cell extracts and Triton-insoluble fractions that are associated with low speed cytoskeletal pellets (20). Using this method, we found both CD47 and IgV-GPI exclusively in the Triton-insoluble pellets, indicating they resided in DRMs associated with the cytoskeleton (Fig. 7A, lanes 9 and 11). Thus, this pellet fraction must include the cholesterol-rich rafts that were detected in detergent-free preparations (Fig. 6A). Cyclodextrin treatment resulted in the release of more than half of the wild-type CD47 in a Triton-soluble form (Fig. 7A, lane 4 versus lane 10), presumably as a result of disruption of cholesterol-rich rafts. IgV-GPI, on the other hand, persisted in the pellet and was not detected in the soluble extract even after cyclodextrin treatment (Fig. 7A, lanes 6 and 12). Complexes of α_vβ₃ with IgV-GPI were therefore localized in cyclodextrin-resistant domains (CRDs), while wild-type CD47-α_vβ₃ complexes were distributed in both CRDs and cholesterol-rich rafts. In all three cell lines, β₃ integrin was mostly Triton-soluble (Fig. 7A, upper panel), and this soluble fraction was increased with cyclodextrin treatment (lanes 2, 4, and 6), indicating that Triton extraction of α_vβ₃ from ordered lipid domains was increased with cholesterol depletion. However, the release of integrin with cyclodextrin treatment was increased when either CD47 or IgV-GPI was present (Fig. 7A, lanes 4, 6, and 9-12). Thus, CD47 enhanced partitioning of α_vβ₃ into detergent-resistant domains though IgV interactions that did not require the MMS region.

To demonstrate that CD47 associates with α_vβ₃ in the cyclodextrin-resistant domains, protein complexes were adsorbed to anti-CD47 coated beads from octylglucoside extracts of the Triton-insoluble pellets. These immunoprecipitation conditions were similar to those used in the previous study reporting an association of the IgV domain of CD47 with α_vβ₃ (13). Western analysis revealed the presence of β₃ upon adsorption of either wild-type CD47 or IgV-GPI from the Triton-insoluble, but octylglucoside-soluble fractions (Fig. 7B). Furthermore, detection of β₃ was not blocked by cyclodextrin treatment for either species (Fig. 7B), indicating that the CD47-α_vβ₃ interaction was stable upon cholesterol depletion. Thus, CD47 associates with α_vβ₃ in cyclodextrin-resistant domains via IgV interactions, which led to enrichment of the integrin in those membrane compartments.

DISCUSSION

CD47 can modulate integrin function in two ways. The first of these is analogous to the well described activation of integrins such as α_IIbβ₃ on platelets by G protein-coupled receptors like the thrombin receptor (33). This so called inside-out signaling leads to activation of kinase pathways that modify integrins or their associated proteins, which in turn promotes a more active, perhaps higher affinity, state of the integrin (33). In addition to this classical signaling pathway, accumulating evidence indicates that direct interactions between CD47 and the integrins, which it modulates play a role. Importantly, ligation of CD47 by TSP-derived agonist peptides such as 4N1K and 7N3 (VVM motif peptides) seems to promote both pathways of integrin activation. We and others (4, 25, 34) have reported that these agonist peptides act via a G_i-dependent signaling pathway to augment integrin functions. In addition, there have been reports that the agonist peptides can augment integrin function even when G_i is inactivated by pertussis toxin, suggesting that there may be G protein-independent routes of action (3). In fact a recent report showed that the soluble, extracellular IgV domain of CD47, when added to cells, could activate α_IIbβ₃ (12). This occurred only when the agonist peptide 4N1K was present, suggesting that 4N1K induces a conformation of the CD47 IgV domain that interacts favorably with the extracellular domain of α_IIbβ₃.

In this study we used the panel of OV-10 cells expressing the integrin α_vβ₃ alone, along with wild-type CD47 or with the extracellular IgV domain of CD47 engineered to be expressed anchored to the cell surface via GPI (13). This CD47 variant allowed us to study the signaling (G_i) independent actions of CD47 while maintaining the IgV domain in its normal location on the cell surface. The levels of CD47 and IgV-GPI expression in the OV10 cells in this study are in line with levels seen for natively expressed CD47 in a number of cell types, and does not represent overexpression. Likewise, the α_vβ₃ utilizes the endogenously expressed pool of α_v and is equivalent to the level of α_vβ₅ in wild-type cells. Given the earlier report of a role for cholesterol in stabilizing the CD47-integrin-G_i complex (9), we investigated the effect of cholesterol chelation on the ability of these two forms of CD47 to impact α_vβ₃ function. To our surprise, the GPI anchored CD47 IgV domain was as active as wild-type CD47 in promoting α_vβ₃-mediated adhesion and conformational changes associated with LIBS mAb binding, and was even more effective than native CD47 in protecting against the disruption of cell adhesion associated with the removal of cholesterol and dissolution of cholesterol-rich raft domains. The mechanism of these effects appears to be promotion of integrin clustering by CD47 and this clustering is not solely dependent on cholesterol-rich domains. Nonetheless, the inclusion of the integrin in cholesterol-rich light membranes along with associated Gi and other signaling proteins occurs only when native CD47 containing the MMS domain is expressed. Wild-type CD47 clearly exists in cholesterol-rich rafts as well as in other types of membrane domains. Thus it appears that CD47 residence in different membrane domains may alter its functions in terms of how integrin modulation is accomplished, and in the nature of that modulation

The mechanism by which CD47 promotes α_vβ₃ clustering is not known, but its interactions presumably stabilize an integrin conformation that is able to form integrin-integrin contacts. Recent studies of α_IIbβ₃ have shown that homo-oligomerization of β₃ can drive integrin clustering via transmembrane contacts, and that mutations which promote oligomerization also increase binding to fibrinogen (35). Similarly, we show that CD47 interactions promote both clustering of α_vβ₃ and increased activation as measured by binding of LIBS antibodies. Furthermore, models of integrin activation indicate that it is accompanied by separation of the α- and β-chains (36), suggesting that homo-oligomerization involves activated conformations, and drives a conformational equilibrium toward the activated state. That CD47 expression enhances the binding of anti-LIBS antibodies (Fig. 1, A-E) also indicates that it influences an activated integrin state. These events may be mediated by recruitment of α_vβ₃ to detergent-resistant domains (DRMs). A comparison of the effects of cholesterol depletion on the ability of wild-type CD47 versus IgV-GPI to alter α_vβ₃ avidity and raft localization supports such a mechanism. While wild-type CD47 expression conferred partial resistance to cyclodextrin-induced disruption of cell adhesion to α_vβ₃ ligands, IgV-GPI-expressing cells exhibited almost complete protection (Figs. 2 and 3). This was correlated with the partial conversion of wild-type CD47 to a Triton-soluble state with cholesterol depletion, while IgV-GPI localization in Triton-insoluble domains was completely resistant to cyclodextrin treatment (Fig. 7). Thus, IgV-GPI appeared to be superior to wild-type CD47 in its ability to confer protection against cyclodextrin-induced disruption of cell adhesion as a result of its increased enrichment in cyclodextrin-resistant membrane domains. A correlation between raft localization and integrin clustering has been previously reported in fibroblasts, where membrane cholesterol levels affected the clustering of α₅β₁ (23), and in T-cells, where LFA-1-mediated cell binding was shown to be enhanced by clustering of membrane rafts via the GPI-anchored protein CD24 (24). In raft membranes, integrin-integrin contacts may be promoted by a more extreme hydrophobic environment, or by other physical properties such as membrane thickness (37). Raft localization could also regulate integrin clustering by its association with the actin cytoskeleton. In T cells, both raft localization and increased PKCθ association with the cytoskeleton was correlated with stimulation of IL-2 synthesis induced by CD47 ligation in synergy with TCR (10), suggesting that CD47 may link T cell signaling to the cytoskeleton through detergent-resistant domains. Although clustering of certain integrins can be enhanced with release of cytoskeletal constraints in lymphocytes (38), these linkages may also provide adhesive strength. Regulated distribution of transmembrane proteins to cytoskeletal-linked DRMs has been shown in mouse mammary epithelial cells, where CD44 partitioning was correlated with its redistribution to the leading edge of polarized cells (21). In a similar manner, CD47 may serve as a link between integrin surface clusters and the cytoskeleton. Interestingly, the cytoplasmic tail of CD47 has been found to associate with cytoplasmic proteins termed PLICs (proteins linking intermediate filaments to CD47), which may provide a link between CD47 containing complexes and the cytoskeleton (39). The ability of IgV-GPI to increase tethering of α_vβ₃ to the cytoskeleton (Fig. 7) indicates that the linkage need not be through CD47 itself, but that the IgV domain of CD47 confers upon the integrin an enhanced ability to engage cytoskeletal elements.

Recent structural studies of α_vβ₃ have provided new insights into how lateral or cis integrin ligands might inhibit or promote integrin activation. The crystal structure of α_vβ₃ reported by Xiong et al. (40, 41) showed the integrin in a bent conformation, and ultrastructural studies have depicted closed and open configurations that correlate with low and high affinity states, respectively (42). Thus regulation of integrin activation can be seen in terms of a conformational equilibrium that may be influenced in either direction by the action of ligands, antibodies or associated proteins. We have demonstrated in this study that CD47, though its extracellular IgV domain, can increase the appearance of LIBS epitopes that are exposed with RGD binding which leads to a more open conformation of the integrin (33). Thus in addition to effects on integrin clustering by recruitment to ordered lipid domains, CD47 may have direct conformational influences by stabilizing the open conformation of integrin heterodimers. However, IgV interaction alone was not sufficient for targeting integrin to cholesterol-rich rafts (Fig. 5B), indicating that CD47-αvβ3 complexes in these domains may have different properties. Because the MMS region is strictly required for G protein association (Fig. 4B), localization to cholesterol-rich rafts may be an activating mechanism, with a requirement of integrin transmembrane contacts for G protein coupling (2). The finding that a CD47 extracellular epitope, detected by the mAb 10G2, is exposed with cholesterol depletion (9), indicates that membrane-proximal contacts influence IgV structure, and may therefore affect extracellular integrin contacts. Translocation of the CD47-α_vβ₃ complex between cholesterol-poor and cholesterol-rich domains could therefore alter the CD47 influence on integrin conformation, as well as activate G_i-mediated signaling. In support of such a pathway, it has been reported that IGF-1 treatment of human SMC increased the affinity of α_vβ₃ for soluble vitronectin, and was accompanied by translocation of CD47 from Triton-insoluble to Triton-soluble fractions (20). Binding of 4N1K by the IgV domain could also produce signaling-dependent versus direct conformational outcomes. A recombinant cell-binding domain of thrombospondin has been shown to produce G_i-mediated effects requiring both VVM motifs from which the 4N1K sequence was derived, suggesting that G protein activation occurs via CD47 oligomerization (43). 4N1K-induced oligomerization of CD47 could lead to formation of higher order clusters through its association with integrin aggregates. While the process of G protein coupling by CD47 has not been fully elucidated, our findings that CD47-integrin complexes are altered by the lipid environment of the membrane may lead to new ways to approach the mechanism of signaling.

Our finding that CD47 can modulate α_vβ₃ avidity may resolve the apparently conflicting observations that CD47 exerts both signaling-dependent and signaling-independent effects on β₃ integrin function. Furthermore, we have shown that integrin recruitment to different membrane microdomains by CD47 can affect integrin activation by direct association, or by signaling through recruitment into cholesterol-rich rafts containing G_i. Thus different pools of α_vβ₃ can be regulated differently by CD47, perhaps leading to different biological readouts such as motility, protection from apoptosis or gene expression. The striking ability of different membrane environments to alter the interaction of CD47 with its integrin partner represents a unique example of how raft localization can modulate integrin functions, and may serve as a more general model for the interaction of other classes of integrin-associating proteins.