The Integrins α3β1 and α6β1 Physically and Functionally Associate with CD36 in Human Melanoma Cells

Lateral association between different transmembrane glycoproteins can serve to modulate integrin function. Here we characterize a physical association between the integrins α3β1 and α6β1 and CD36 on the surface of melanoma cells and show that ectopic expression of CD36 by CD36-negative MV3 melanoma cells increases their haptotactic migration on extracellular matrix components. The association was demonstrated by co-immunoprecipitation, reimmunoprecipitation, and immunoblotting of surface-labeled cells lysed in Brij 96 detergent. Confocal microscopy illustrated the co-association of α3 and CD36 in cell membrane projections and ruffles. A requirement for the extracellular domain of CD36 in this association was shown by co-immunoprecipitation experiments using surface-labeled MV3 melanoma or COS-7 cells that had been transiently transfected with chimeric constructs between CD36 and intercellular adhesion molecule 1 (ICAM-1) or with a truncation mutant of CD36. CD36 is known to engage in signal transduction and to localize to membrane microdomains or rafts in several cell types. Toward a mechanistic explanation for the functional effects of CD36 expression, we demonstrate that in fractionated Triton X-100 lysates of the MV3 cells stably transfected with CD36, CD36 was greatly enriched with the detergent-insoluble fractions that represent plasma membrane rafts. Significantly, when these fractionated lysates were reprobed for endogenous β1 integrin, it was found that a 4-fold increase in the proportion of the mature protein was contained within the detergent-insoluble fractions when extracted from the CD36-transfected cells compared with MV3 cells transfected with vector only. These results suggest that in melanoma cells CD36 expression may induce the sequestration of certain integrins into membrane microdomains and promote cell migration.

Integrins comprise a large family of ␣␤ heterodimeric transmembrane proteins that function as key receptors for cellular attachment to the extracellular matrix and to cellular ligands (1). Integrin function extends beyond cell adhesion per se, and integrin-mediated signaling influences a range of biological processes including cellular division and migration, differentiation, and apoptosis (reviewed in Refs. [1][2][3][4][5]. Integrin function is complex and is regulated at many levels. The binding of ligand to integrins can initiate signaling (1,2), with bidirectional "cross-talk" occurring between integrins and other pathways (e.g. Refs. 6 and 7), including the functional modulation of other integrin molecules (8 -10). Much emphasis has also focused on the process of "inside-out" signaling, where intracellular events enacted through the relatively short cytoplasmic tail sequences of integrins have a profound influence on the extracellular function of the integrin (1). In addition to the molecular interactions occurring between cytoplasmic proteins and the cytoplasmic domains of integrins (reviewed in Refs. 11 and 12), emerging evidence indicates that a number of surface transmembrane glycoproteins can associate with integrins and modulate their cellular functions (12).
Several classes of cell surface glycoproteins have been documented to play a role in integrin-mediated events, the most widely studied being the integrin-associated protein (IAP 1 / CD47) and the unrelated transmembrane 4 superfamily (TM4SF or tetraspannins). IAP associates with the ␣ v ␤ 3 integrin (13) and influences integrin-mediated signal transduction, phagocytosis, cellular migration, and chemotaxis (14 -17). In vascular smooth muscle cells IAP also associates with the ␣ 2 ␤ 1 integrin and modulates the chemotactic response of these cells to soluble collagen (18). Numerous TM4SF members have been shown to associate with ␤ 1 integrins and with other TM4SF members (19 -27). For example, the TM4SF molecule CD63 has been shown to specifically associate with ␣ 3 ␤ 1 and ␣ 6 ␤ 1 integrins (21), and the CD63-␣ 3 ␤ 1 complex was found to be associated with phosphatidylinositol 4-kinase activity, suggesting a novel integrin-signaling pathway (28). Furthermore, transfection studies with CD63 in a melanoma cell line resulted in marked inhibition of cell motility, an effect probably mediated by its influence on ␤ 1 integrin function (29). Similarly, another TM4SF member, CD9, also has been shown to associate with ␤ 1 integrins and to be implicated in cellular signaling, adhesion, motility, and differentiation (19, 20, 22, 30 -32). It is believed that both IAP and the TM4SF molecules complex with integrins via extracellular domains, since the Ig domain of IAP is essential for effects on ␣ v ␤ 3 integrin function (33) and the ␤ 1 integrin association with CD63 occurs independently of the integrin ␣-subunit cytoplasmic tail (20). More recently, Yaunch et al. (34) demonstrated using chimeras that the CD151 (TM4SF)/␣ 3 ␤ 1 association required extracellular domains found in CD151 and the ␣-chain of the integrin. Furthermore, in K562 cells, in which the ␣ 4 ␤ 7 integrin initiates cell adhesion under flow conditions, the extracellular domains of the integrin determine its localization to microvilli, where both IAP and TM4SF proteins were also found to be strongly concentrated (35). Given the ubiquitous expression of both IAP and TM4SF, this suggests that regulation of integrin localization and function by these accessory molecules may play a widespread role in cellular processes.
Other surface molecules have also been implicated in the modulation of integrin-mediated functions. The widely expressed CD98 molecule (initially described as an early T-cell activation antigen) was found to regulate ␤ 1 integrin activation in a genetic complementation assay (36), and a physical association between CD98 and certain ␤ 1 integrins has recently been described (cited in Ref. 36); in the case of CD98, however, the association with ␤ 1 integrins appears to occur between cytosolic domains of the two glycoproteins. The CD98 interaction with ␤ 1 integrins appears to modulate the activation of T-cells (38) as well as the adhesion of breast carcinoma cells (39). Other examples of integrin-associated molecules that may modify or regulate integrin function include the ErbB-2 oncogene, a receptor tyrosine kinase associated with breast cancer progression in vivo. This interaction demonstrated to facilitate phosphatidylinositol 3-kinase-dependent in vitro cell invasion by erbB-2-transformed 3T3 cells (39). The cancer metastasispromoting molecule, urokinase plasminogen activator receptor (uPAR) also associates with and regulates ␤ 1 and ␤ 2 integrin function (40).
Another candidate co-associated regulator of integrin-function is the transmembrane glycoprotein CD36. The archetype member of a small gene family, CD36 is unrelated to integrins or to currently known integrin-accessory molecules. CD36 expression is prominent on the surface of platelets, capillary endothelial cells, macrophages, and cultured cell lines from a variety of sources including melanoma (reviewed in Refs. 41 and 42). A study using chemical cross-linking demonstrated that CD36 is spatially associated with the ␣ IIb ␤ 3 integrin on the surface of platelets (43), and although the functional consequence of this association is not known, it has been shown that platelet activation achieved by the addition of anti-CD36 antibodies involves signaling through this integrin (44,45). It is not known whether CD36 directly associates with integrins on other cell types (12), but there are several lines of evidence to suggest that certain functions ascribed to CD36 directly involve or implicate involvement of integrins.
The phagocytic uptake of apoptotic cells by macrophage requires the coordinate function of CD36 and the ␣ v ␤ 3 integrin, possibly with involvement of the extracellular matrix protein thrombospondin functioning as a molecular bridge (46). Additionally, both CD36 and the ␣ v ␤ 5 integrin have been shown to be involved in the phagocytosis of shed photoreceptor rod outer segments by retinal pigment epithelial cells (47,48). Interestingly, CD36 functions as an adhesive cellular receptor for both thrombospondin (49) and collagen (50), specificities that also overlap with integrins of both the ␣ v and ␤ 1 integrin subfamilies (1,11). Fluorescence-activated cell sorting of C32 mela-noma cells into populations expressing either high or low levels of CD36 showed that high CD36 expression defined a more fibroblastoid cell phenotype, with this population demonstrating enhanced tumor growth and motility in in vivo studies using nude mice (51). Additionally, transfection of CD36 into human umbilical vein endothelial cells resulted in aberrant morphological changes affecting in vitro tube formation (52). Taken together, these data suggest that certain functions attributed to CD36 directly involve or implicate involvement of integrins, particularly phagocytosis, cell adhesion, and migration. How this accessory function is enacted is not known.
Possibly significant in this regard is the reported association of CD36 with plasma membrane microdomains or "rafts" enriched in cholesterol, sphingomyelin, and glycolipids. Rafts can be isolated biochemically as low density Triton X-100-insoluble membranes, and such isolates have been termed detergentinsoluble glycosphingolipid-enriched complexes (DIGs), also described as DRMs, GEMs, TIM, TIFF, and LDTI (reviewed in (53)(54)(55)). CD36 was found to be highly enriched in DIGs from a variety of cellular sources including lung endothelium (56), platelets (57), and transfected COS-7 cells (58). Raft domains have been implicated in a variety of cellular processes including signal transduction, membrane trafficking, cholesterol homeostasis, the regulation of apoptosis, and cell motility (53-55, 59, 60). The plasma membrane invaginations known as caveolae may represent a specialized structural form incorporating raft components contained within the structural coat protein caveolin (61). Early studies suggested that integrins were not compartmentalized to rafts/caveolae (56,62), but subsequent reports have indicated that a proportion of ␤ 1 integrins may physically and functionally associate with caveolin. For example, caveolin-1 can function to link Fyn kinase with the ␣ 1 integrin subunit, this resulting in downstream signaling promoting cell cycling (63). Furthermore the GPI-linked uPAR, ␤ 1 integrins and caveolin-1 co-immunoprecipitate in a single complex, and this complex appeared necessary for integrin-mediated adhesion and signaling (40,64). Interestingly some Srcfamily kinases, including Fyn, Lyn, and Yes known to associate with CD36 (65,66) are also associated with DIGs (56,57,67), suggesting that CD36 signaling and perhaps other CD36-mediated functions may be enacted via raft domains.
The present study was initiated to examine further the reported effect of CD36 on melanoma cell morphology and growth (51). A melanoma cell line that did not express endogenous CD36 was identified, and permanent CD36 transfectants were isolated. No differences were found in the morphology or growth rates of the transfectants compared with the parent cell line, but the CD36-transfected cells displayed higher haptotactic migration on extracellular matrix substrates. The possibility that integrins might be implicated in this functional effect was investigated by seeking a co-association between CD36 and endogenous integrins on the melanoma cell surface. Coimmunoprecipitation studies revealed a physical association between CD36 and the integrins ␣ 3 ␤ 1 and ␣ 6 ␤ 1 , and this association was also shown on C32 melanoma cells expressing endogenous CD36. We then used transient transfection of chimeric constructs of CD36 to demonstrate that the association between CD36 and ␤ 1 integrins requires the extracellular domain of the CD36 molecule. Finally, this association may occur within raft domains/DIGs, since ectopic expression of CD36 increased the proportion of ␤ 1 integrin(s) found within this fraction.

EXPERIMENTAL PROCEDURES
Cell Lines and Antibodies-The MV3 melanoma cell line (a gift of Dr. G. N. P. van Muijen) was isolated from a lymph node metastasis as described previously (68). C32 and WM115 melanoma and COS-7 mon-key kidney cell lines were obtained from the ATCC (Manassas, VA). All cells were routinely cultured in Dulbecco's modified Eagle's medium (CSL, Parkville, Australia) supplemented with 10% fetal bovine serum (Cytosystems, Castle Hill, Australia).

Construction of Expression Vectors and Cell Transfection-
The human cDNAs for CD36 (a gift of Dr. A. W. Boyd) and ICAM-1 (gift of Dr. D. Simmons, University of Oxford, Oxford, UK) in the pCDM8 vector were subcloned to the high level eukaryotic expression vector pEF.BOS (71) using XbaI restriction endonuclease digestion. To substitute the carboxyl-terminal cytoplasmic tail of CD36 with the ICAM-1 cytoplasmic tail (CD36/ICAM tail chimera) the pEF.BOS.ICAM-1 plasmid was used as template for polymerase chain reaction amplification using oligonucleotides 5Ј-cGcAtGcctctataaccgccagcggaag-3Ј and 5Ј-ccccagggccatgcctcccagc-3Ј, which introduced an SphI restriction site to the 5Ј-end of the product (mismatched oligonucleotide bases are represented in uppercase type). Following digestion with SphI/MscI, the resulting fragment was then subcloned to an MscI/SphI partially digested "vector" fragment of pEF.BOS.CD36, thereby inserting the 31-amino acid ICAM-1 cytoplasmic tail following CD36 cysteine residue 466. A chimera consisting of the amino terminus of ICAM-1 and the CD36 carboxyl-terminal transmembrane region and cytoplasmic tail (ICAM/CD36) was prepared by digestion of pEF.BOS.CD36 and pEF-.BOS.ICAM-1 plasmid DNA with HincII/MscI and MscI restriction endonucleases, respectively. The MscI fragment encoding the 5Ј-sequence of ICAM-1 and the HincII/MscI fragment encoding the 3Ј sequence of CD36 were ligated together, resulting in the ICAM-1 extracellular region amino acids 1-409 fused to residues 390 -472 of CD36. To manufacture an amino-terminal truncation of CD36 with the FLAG epitope tag placed at the N terminus (FN⌬TM.CD36), the pEF-.BOS.CD36 vector was used as template DNA for polymerase chain reaction amplification using the oligonucleotides 5Ј-ggaggtattctaTCTA-GAAtGggagacctgc-3Ј and 5Ј-ccccagggccatgcctcccagc-3Ј. The amplification mutated valine 29 in CD36 to methionine (just outside the predicted amino-terminal transmembrane) and introduced an XbaI restriction site. The polymerase chain reaction product was then restricted with XbaI and subcloned to the XbaI site of the sIL3.FLAG.pEF.BOS vector, a derivative of pEF.BOS containing the IL-3 signal sequence peptide upstream of a FLAG epitope sequence (a gift of Dr. A. W. Boyd). The resulting construct encodes the IL-3 signal peptide, the FLAG epitope, and the CD36 peptide beginning at glycine residue 30 of the wild-type CD36 sequence. All custom synthesis oligonucleotides were purchased from Bresatec (Adelaide, Australia). As required, polymerase chain reaction products were T/A-cloned (pGEM-T, Promega Corp., Madison, WI) before restriction digestion. All vector constructs were validated by automated sequencing analysis.
Stable transfections of MV3 cells were performed using the Lipo-fectAMINE reagent (Life Technologies) according to the manufacturer's instructions. A 20:1 excess of the pEF.BOS vector (which lacks a eukaryotic selection marker) to pREP9 (encoding neomycin phosphotransferase; Invitrogen, San Diego, CA) was used, followed by selection with 500 g/ml G418 (Life Technologies). This procedure facilitated a high level of expression of CD36, negating the need for dilution cloning. For transient expression, MV3 and COS-7 cells were electroporated (300 V/250 microfarads using a Bio-Rad Gene Pulser) with 150 g/ml plasmid DNA in Hebs buffer (20 mM Hepes, pH 7.05, 137 mM NaCl, 5 mM KCl, 0.7 mM Na 2 HPO 4 , 6 mM dextrose), and the cells were analyzed 2-4 days post-transfection.
Flow Cytometry-Cells were harvested using trypsin-versene solution (CSL) and washed in phosphate-buffered saline (PBS). The respective primary mAbs were then incubated at 4°C for 30 min, after which the cells were washed twice with PBS and the bound mAb was detected by incubation with fluorescein isothiocyanate-conjugated sheep antimouse immunoglobulin (F(ab) 2 fragment; Silenus, Dandenong, Australia). Cells were washed again; fixed using 2% (w/v) glucose, 1% (w/v) formaldehyde, 0.02% (w/v) azide in PBS; and analyzed on a Becton Dickinson (Mountain View, CA) FACScan.
Cell Adhesion and Haptotactic Cell Migration Assays-Cell adhesion to extracellular matrix proteins (collagen type I, fibronectin, or laminin; Sigma) was measured as described previously (72). For migration assays, the underside of replicate Transwell migration inserts (8-m pore size, 6.5-mm diameter; Costar, Corning, NY) were coated for 60 -90 min at 37°C with the indicated matrix proteins (20 g/ml) or with bovine serum albumin (BSA; 0.5% (w/v)) in 200 l of PBS. The wells were washed once with migration buffer (serum-free medium supplemented with 0.5% BSA), and 500 l of migration buffer was added to the bottom chamber. Cells were detached using trypsin-EDTA (0.25% (w/v) and 5 mM, respectively) and washed three times in migration buffer before seeding to the top of the filter (10 5 cells in 100 l). After 24 h at 37°C, adherent cells were detached using trypsin/EDTA from the upper and lower membrane surfaces and separately pooled with the supernatants collected from each respective chamber. Cells were counted on a CASY-1 Cell Analyser (Scharfe Systems GmbH, Reutlingen), and cell migration was calculated as the percentage of cells harvested from the underside of the filter divided by the total number of cells counted. All assays were performed on triplicate wells and repeated several times.
Cell Surface Labeling and Immunoprecipitation-A modified lactoperoxidase method was used for cell surface radiolabeling with Na 125 I (Australian Radioisotopes, Lucas Heights, Australia) as described previously (23). For biotin labeling of cell surface antigens, cells were resuspended in 10 mM sodium tetraborate, pH 8.8, 150 mM NaCl containing 50 g/ml biotinamidocaproate N-hydroxysuccinimide ester (Sigma) and incubated for 15 min at room temperature before quenching of the reaction by the addition of NH 4 Cl to a 10 mM final concentration. After washing with PBS, 125 I/biotin-labeled cells were lysed for 60 min at 4°C with Brij 96 lysis buffer (1% polyoxyethylene 10 oleyl ether (Sigma) in 25 mM Hepes, pH 7.5, 150 mM NaCl, 5 mM MgCl 2 supplemented with protease inhibitors (2 mM phenylmethylsulfonyl fluoride, 10 g/ml soybean trypsin inhibitor, and 2.5 mM iodoacetamide (all from Sigma)). Insoluble material was then removed by centrifugation for 10 min at 14,000 ϫ g. The lysates were precleared twice using protein A beads (Amersham Pharmacia Biotech) and/or rabbit anti-mouse immunoglobulins (DAKO, Carpentaria, CA) coupled to CNBr-activated Sepharose beads (Amersham Pharmacia Biotech) in combination with human transferrin (Calbiochem) coupled beads for 1-2 h at 4°C. Lysates were then incubated for 2 h (biotin) or overnight ( 125 I) with either mAb combined with rabbit anti-mouse immunoglobulin-coupled beads, with pAb with protein A beads (indirect capture), or with mAb directly coupled to beads (as described above). Because CD36 was poorly represented in cell surface immunoprecipitates relative to ␤ 1 integrins, in some experiments immunoprecipitations were weighted for CD36 detection by using 8 -9 times the input cell lysate relative to the integrin immunoprecipitations. The beads were then washed four times in lysis buffer, eluted with Laemmli sample buffer, and resolved by SDS-PAGE. Gels containing radioiodinated samples were fixed, stained, and dried, and the labeled protein bands were visualized by autoradiography. Gels containing biotinylated samples were transferred to nitrocellulose, and the bands were visualized by blotting with streptavidin-biotin horseradish peroxidase complexes (Amersham Pharmacia Biotech) as described below for immunoblotting.
Where samples were subjected to secondary immunoprecipitation, the beads were first washed with lysis buffer, and the antigen(s) were eluted from the beads using acidic buffer (50 mM glycine-HCl, pH 2.5, 0.15 M NaCl, 0.1% Triton X-100). The eluate was immediately neutralized with 2 M Tris, pH 8.0, passed through a PD10 column (Sephadex G25M; Amersham Pharmacia Biotech), and eluted in 1% Triton X-100 containing lysis buffer. The eluate was then subjected to a second round of immunoprecipitation and processed as described above.
Immunoblotting-Cells were lysed and immunoprecipitated as de-scribed above. Samples were resolved by SDS-PAGE and transferred to nitrocellulose using 25 mM Tris, 192 mM glycine, 20% methanol transfer buffer. Nitrocellulose membranes were blocked in 5% skim milk in TTBS (20 mM Tris, pH 7.5, 500 mM NaCl, 0.05% Tween 20) before incubation with specific antibodies in 1% skim milk TTBS. Membranes were then incubated with horseradish peroxidase-conjugated goat antimouse or rabbit immunoglobulins (Bio-Rad), and horseradish peroxidase complexes were detected by the Renaissance chemiluminescence system (PerkinElmer Life Sciences). Immunofluorescent Cell Staining and Confocal Microscopy-MV3 cells were harvested as described above and allowed to attach to glass coverslips overnight. Following fixation with 4% formaldehyde (w/v) in PBS for 20 min at room temperature, the coverslips were washed in 0.1% BSA, PBS (PBSA) and permeabilized with Triton X-100 (0.1% (v/v) in PBSA) for 10 min at room temperature. Antibodies (diluted in PBSA) were then added overnight at 4°C, and the coverslips were washed four or five times in PBS before mounting in Mowiol 4 -88 (0.1 g/ml; Calbiochem) dissolved in a mixture of 12 ml of Tris-HCl (pH 8.5), 6 ml of Citifluor AF1 (Citifluor Ltd., London, UK), and 6 ml of distilled water. Fluorescein isothiocyanate-conjugated anti-integrin mAbs together with an anti-CD36 (11H5) TAMRA conjugate (prepared according to the manufacturer's recommended protocol; Molecular Probes, Inc., Eugene, OR) were used for the two-color analysis. Cells were examined by laser-scanning confocal microscopy (Zeiss LSM510, Zeiss Microscopes, Welwyn, UK).
Detergent-insoluble Glycosphingolipid-enriched Membrane Raft Domains-DIGs were prepared by floatation in sucrose density gradients as described previously (58). Briefly, whole cell lysates solubilized in 1% Triton X-100 in MES-buffered saline (pH 6.5) were adjusted to 40% sucrose and applied under a discontinuous 5-30% sucrose gradient. Following ultracentrifugation, DIGs were found to separate as a low density light-refractive band. Twelve fractions were collected from the top of the gradient, and equal protein amounts of each were analyzed by SDS-PAGE and immunoblotting (protein concentrations were determined using the BCA assay; Pierce). Densitometric analysis was performed on nonsaturating exposures of immunoblot films. Images were captured using a UMAX powerlook II scanner, and densitometric quantitation was performed using the NIH Image software package, version 1.6.1 (written by W. Rasband, National Institutes of Health). Enrichment in DIGs was calculated as the total antigen density in DIGcontaining fractions divided by the total antigen density detected across the gradient (percentage of antigen in DIGs).

RESULTS
The Integrins ␣ 3 ␤ 1 and ␣ 6 ␤ 1 Associate with CD36 on the Surface of Melanoma Cells-In order to determine whether CD36 influenced the adhesion and migration of melanoma cells in a system amenable to detailed analysis, we identified a well characterized metastatic melanoma cell line (MV3; Ref. 68) that was deficient in CD36 expression and transfected this with CD36. Several bulk populations of cells stably transfected with CD36 were derived by G418 selection, and one (uncloned) population expressing levels of CD36 comparable with that endogenously expressed by C32 and WM115 melanoma cells was selected for further study (MV3.CD36, Fig. 1A). In contrast to the report of Wong et al. (51), who found that C32 melanoma cells expressing high levels of CD36-antigen displayed a more fibroblastoid appearance relative to low CD36-antigen-expressing cells, both control (pEF.BOS vector) and CD36-transfected MV3-melanoma cells appeared morphologically indistinguishable (Fig. 1B), and we could detect no difference in the growth rates of the two populations (data not shown). Adhesion assays measuring attachment to fibronectin, collagen, or laminin substrates showed no significant differences between the MV3.BOS and MV3.CD36 cells either measured under time-or matrix protein concentration-dependent conditions (data not shown). In contrast, in cell migration assays carried out in Transwell chambers, the CD36-expressing cells were found to display significantly increased migration on fibronectin, laminin, and collagen substrates but not on the BSA substrate control (Fig. 1C). Wong et al. (51) did not measure cell migration in vitro, but the high CD36-expressing C32 cells were more metastatic in vivo, and this, together with a more fibroblastoid morphology, suggests that these cells would be more migratory than the low CD36-antigen expressing cells. We did not test this directly by cell sorting, but for the C32 mixed population the corresponding migration rates were 32.4 Ϯ 3.3, 13.7 Ϯ 11.9, 28.7 Ϯ 6.0, and 1.5 Ϯ 0.7% on fibronectin, collagen, laminin, and BSA (control) substrates, respectively. The MV3 cells are highly metastatic in vivo (68) even in the absence of CD36 expression, and in vitro the nontransfected parental cells already display a somewhat motile appearance; therefore, it is perhaps not surprising that CD36 expression did not alter the morphological appearance of these cells. However, the significant increase in the already high motility of these cells on extracellular matrix components indicates that CD36 expression functionally interacts with the motility-promoting machinery of these cells.
Next we examined the transfected cells for a physical association between CD36 and integrins, since these molecules provide the mechanical elements of cell migration. The transfected MV3 cells were surface-labeled with radioiodine and lysed with Brij 96 detergent, and the soluble cell lysates were subjected to immunoprecipitation for CD36 or ␤ 1 integrins. As shown in Fig. 2A, the CD36 antibody immunoprecipitated a band at the position of CD36 (ϳ85 kDa) together with two bands coincident with the immunoprecipitated ␤ 1 integrins (Fig. 2A). The associated protein bands were absent from control immunoprecipitations ( Fig. 2A) and from CD36 immunoprecipitates of nontransfected MV3 cells (data not shown). A weak band at ϳ85 kDa corresponding to the mobility of CD36 was also observed in the ␤ 1 integrin precipitate. Due to the relative abundance and labeling intensity of ␤ 1 integrins and also because CD36 is relatively poorly immunoprecipitated from the cell surface, an 8:1 lysate input weighting for the CD36 immunoprecipitation was used. This result appeared to suggest that CD36 was physically associating with ␤ 1 integrins in the transfected MV3 cells.
To determine whether this was the case in melanoma cells expressing endogenous CD36, the same immunoprecipitation experiments were carried out with lysates of surface radioiodinated C32 and WM115 melanoma cells. As before, CD36 antibodies co-precipitated labeled bands in the region of ␤ 1 integrins, although in these assays no CD36 was observed in the ␤ 1 integrin immunoprecipitate under these conditions (Fig. 2B). These results suggest that the putative ␤ 1 integrin(s) association with CD36 is not restricted to transfected cells but is representative of melanoma cell lines.
To identify specific integrin(s) associating with CD36, reprecipitation assays were employed. Phenotyping of the MV3 cells by flow cytometry demonstrated that they expressed each of the ␣ 1 to ␣ 6 subunits known to associate with the ␤ 1 integrin subunit (1,11). By this analysis, the ␣ 2 , ␣ 4 , and ␣ 6 integrin subunits appeared to be most abundant with relatively lesser amounts of ␣ 1 , ␣ 3 , and ␣ 5 antigen (Fig. 3A). These cells have been reported not to express ␣ v ␤ 3 (73), and we confirmed these results, although the cells were found to express ␣v␤5 (data not shown). In addition, MV3 cells are known not to express ␣ 6 ␤ 4 integrins (74). Immunoprecipitation analysis of Brij 96 lysates from surface radioiodinated MV3-CD36 cells substantiated these findings, although in this case the ␣ 5 integrin in particular appeared to label or precipitate less well or to be less abundant. In this experiment, it was also noted that ␣ 3 and ␣ 6 integrin subunit antibodies precipitated a labeled band with the same migration as immunoprecipitated CD36 (Fig. 3B). This result suggested that ␣ 3 and ␣ 6 are the predominant ␤ 1 integrins associating with CD36 in the transfected MV3 cells. To substantiate this notion, surface-labeled MV3-CD36 cells were lysed in Brij lysis buffer and CD36 immunoisolated with anti-CD36 antibodies directly coupled to Sepharose 4B beads. Bound proteins were eluted from the beads with brief acid treatment and neutralized, and aliquots of the eluate were reprecipitated with antibodies to the ␤ 1 and ␣ 1 -␣ 6 integrin subunits. The results (Fig. 3C) indicated that the major integrin to co-precipitate with CD36 was ␣ 3 ␤ 1 with trace amounts of ␣ 6 ␤ 1 also associating with CD36. In repeat experiments confirming these results, we found in addition by reprecipitation with antibodies to ␣ v , ␤ 3 , ␤ 5 , and ␤ 6 that none of these ␣ v integrins were associated with CD36 on the transfected MV3 cells (data not shown).
The co-association was then confirmed by immunoblotting (Fig. 3D). MV3.CD36 cells were lysed in Brij lysis buffer and immunoprecipitated for CD36 and for ␣ 2 , ␣ 3 , and ␣ 6 integrins. When the resulting immunoprecipitates were immunoblotted for CD36, both ␣ 3 and ␣ 6 lanes showed co-precipitated CD36, whereas ␣ 2 did not. Taken together, these results suggested that ␣ 3 ␤ 1 and ␣ 6 ␤ 1 form a physical association with CD36 on the surface of transfected MV3 cells. To ensure that the association also applied to endogenous CD36 on the surface of melanoma cells, the elution and reprecipitation experiment described above was repeated with C32 melanoma cells. These cells also express each of the integrins ␣ 1 -␣ 6 , although ␣ 1 , ␣ 2 , ␣ 4 , and ␣ 5 are relatively less abundant by both flow cytometry and immunoprecipitation analyses (data not shown). With the C32 melanoma cells, reprecipitation of the integrins from the acid eluate obtained from the CD36 immunoprecipitate revealed associated ␣ 3 ␤ 1 , but in this case not ␣ 6 ␤ 1 (Fig. 3E), but it is likely that the relatively weaker association with ␣ 6 ␤ 1 seen with the MV3 cells (Fig. 3C) would not be seen in precipitates from C32 cells in which the precipitated signal is much weaker.
To confirm that the association between the ␤ 1 integrins and CD36 occurred in vivo and to identify the cellular location(s) of

FIG. 2. ␤ 1 integrins co-precipitate with CD36 in both transfected and endogenously expressing melanoma cells. A,
MV3.CD36 cells (ϳ6 ϫ 10 7 cells) were cell surface-labeled with radioiodine and lysed in 1% Brij 96 lysis buffer, and the lysate was subjected to immunoprecipitation analysis (i.p.) as described under "Experimental Procedures." The lysate was divided such that 80% of the material was used for specific immunoprecipitation for CD36 (VM58, IA7, and IE8 mAb mixture) and 10% each for the control immunoprecipitation (Cntrl; protein A beads alone) and immunoprecipitation with the ␤ 1 pAb. Immunoprecipitates were resolved by 7.5% SDS-PAGE under reducing conditions and visualized by autoradiography after 2 days. B, WM115 and C32 melanoma cells (ϳ6 ϫ 10 7 cells) were analyzed as described for A. Autoradiography was performed for 8 days.
this association, dual color immunofluorescence and confocal microscopic analysis were performed on the transfected MV3 cells after staining at the cell surface for ␣ 2 or ␣ 3 integrins and CD36. No specific association between ␣ 2 integrin and CD36 integrin was identified in these cells (data not shown). In contrast, there was strong co-localization between CD36 and ␣ 3 integrin in a high proportion of the cells examined, and the sites of association were identified as being predominantly on cell surface projections and/or membrane ruffles (Fig. 4).
Association with Integrins Requires the Extracellular Domain of CD36 -To elucidate the nature of the CD36/␤ 1 integrin association, chimeras between CD36 and ICAM-1 were utilized (Fig. 5A). These constructs were transiently transfected into the MV3 melanoma cells, and 3 days following transfection, the cells were surface-labeled with biotin, lysed in Brij 96 lysis buffer, and immunoprecipitated with antibodies directed against CD36, ICAM-1, and ␤ 1 integrins. The results demonstrate that, as was shown for the permanently transfected MV3.CD36 cells, immunoprecipitates of CD36 from MV3 cells transiently transfected with wild-type CD36 co-precipitated ␤ 1 integrins (Fig. 5B); note that using the biotinylation protocol, the ␣-subunit(s) of ␤ 1 integrins are less strongly labeled in comparison with radioiodination (compare with Fig. 3B). The substitution of the CD36 carboxyl-terminal cytoplasmic tail with that of ICAM-1 did not affect the ␤ 1 integrin association in the transfected cells ( Fig. 5B; CD36/ICAM tail). Additionally, as demonstrated for wild-type CD36 above (Fig. 3D), immunoblotting of the immunoprecipitates of specific ␤ 1 integrins with antibodies against CD36 revealed that the CD36/ICAM tail chimera co-precipitated with ␣ 3 and ␣ 6 integrins but not with the ␣ 2 subunit (Fig. 3C), corroborating the results presented in Fig. 5B and further confirming the specificity of the CD36-␤ 1 integrin interaction.
The MV3 cells were then transfected with an ICAM/CD36 chimera, encoding the extracellular domain of ICAM-1 and the CD36 carboxyl-terminal transmembrane/cytoplasmic sequences (Fig. 5A), and the cell lysates were precipitated with an anti-ICAM-1 mAb. This antibody was found to precipitate endogenous ICAM-1 from the control transfected MV3 cells (Fig. 5D, BOS), and no integrin bands were seen to co-precipitate with the ICAM-1 molecule. In cells transfected with the ICAM/CD36 construct, the chimeric molecule was found to co-precipitate with endogenous ICAM-1, but again there were no integrin bands identified in the precipitate (Fig. 5D). There-FIG. 3. Integrins ␣ 3 ␤ 1 and ␣ 6 ␤ 1 specifically associate with CD36 in transfected MV3 and C32 melanoma cells. A, the expression of the integrin ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , ␣ 5 , and ␣ 6 subunits on MV3 parental cells was examined by flow cytometry. Open profiles represent reactivity with the control mAb (HB57) in comparison with the black profile (the indicated anti-integrin ␣-subunit mAbs). B, MV3.CD36 cells (ϳ12 ϫ 10 7 cells) were cell surface-labeled with radioiodine and lysed in 1% Brij 96 lysis buffer, and the lysate was subjected to immunoprecipitation analysis as described under "Experimental Procedures." The lysate was divided equally, and specific immunoprecipitations (i.p.) were performed using the anti-CD36 specific mAbs (IA7 and IE8) directly coupled to Sepharose 4B (lane 1) or with mAbs to the integrin subunits ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , ␣ 5 , and ␣ 6 (lanes 2-7, respectively). Immunoprecipitates were resolved on 7.5% SDS-polyacrylamide gels under nonreducing conditions and visualized by autoradiography after 6 days. C, MV3.CD36 cells were radiolabeled as described for B, and the cell lysate was immunoprecipitated using 500 l of IA7/IE8-Sepharose beads, before acid elution of the bound antigen(s) as described under "Experimental Procedures." The eluate was divided equally, and secondary immunoprecipitations were performed for CD36 (lane 1), the ␤ 1 integrin subunit (mAb 13; lane 2), or the integrin subunits ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , ␣ 5 , and ␣ 6 (lanes 3-8, respectively). Immunoprecipitates were resolved on 7.5% SDS-PAGE gels under nonreducing conditions and visualized by autoradiography after 6 weeks. D, MV3.CD36 cells were lysed and immunoprecipitated as for B. The cell lysate (ϳ25 g of protein; lane 1) and specific immunoprecipitations for either CD36 (IA7/IE8; lane 2), and the integrin subunits ␣ 2 , ␣ 3 , and ␣ 6 (lanes 3-5, respectively) were resolved by SDS-PAGE under reducing conditions. The gel was transferred to nitrocellulose, and immunoblotting was performed with the CD36 pAb as described under "Experimental Procedures." E, C32 melanoma cells (ϳ12 ϫ 10 7 cells) were radiolabeled and lysed as described for B. The lysate was first immunoprecipitated for CD36, the bound antigen(s) was eluted, and then secondary immunoprecipitation was performed as described in C for either CD36 (lane 1) or the ␤ 1 integrin subunits ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , ␣ 5 , and ␣ 6 (lanes 2-7, respectively). Immunoprecipitates were resolved on 7.5% SDS-PAGE gels under nonreducing conditions and visualized by autoradiography after 12 weeks. For all experiments shown, similar results were obtained on at least two separate occasions. fore, since no association was seen when the extracellular region of CD36 was replaced with ICAM-1, although the transmembrane domain and cytoplasmic tail of CD36 were left intact, these data suggest that the association between CD36 and integrin molecules occurs extracellularly.
These findings established that the association between the ␤ 1 integrins and CD36 occurs in the absence of the carboxylterminal cytoplasmic tail of CD36. However, the predicted CD36 polypeptide contains 471 amino acids with hydrophobic regions resembling transmembrane domains present adjacent to both the amino-and carboxyl-terminal regions. The short amino-terminal hydrophobic region resembles an uncleaved signal peptide (75) with only the initiator methionine being absent in the mature protein (76). From these data, it has been proposed that CD36 is ditopic in an "inverted horseshoe" membrane configuration (41), although it has been controversial as to whether the amino-terminal transmembrane domain actually serves to anchor the protein (77)(78)(79). If the amino-terminal regions does form a transmembrane domain, therefore, it is possible that an association between the ␤ 1 integrins and CD36 could occur in this region, either in the amino-terminal cytoplasmic domain or in the transmembrane domain.
To investigate this, a molecule was constructed lacking the amino-terminal and cytoplasmic domains; this construct was tagged with a FLAG epitope and incorporates a cleavable signal sequence (FN⌬TM.CD36; Fig. 6A). Upon transfection, however, this construct was very poorly expressed in the MV3 cells. Preliminary analyses revealed that reasonable levels of expression could be obtained with this construct in transiently transfected COS-7 cells (data not shown). Therefore, COS-7 cells were transfected with the control BOS vector, wild-type CD36, or the FN⌬TM.CD36 construct, and the lysates of biotin-labeled cells were subjected to immunoprecipitation analysis. Immunoprecipitates of wild-type CD36 showed a specific coprecipitated band with identical mobility to the ␤ 1 integrin subunit (Fig. 6B), indicating that the transfected human CD36 can associate with endogenous monkey ␤ 1 integrin(s). Similarly, both the amino-terminal truncate of CD36 (FN⌬TM.CD36; Fig. 6B) and the CD36/ICAM tail mutant ( Fig. 5A and data not shown) also co-precipitated endogenous ␤ 1 integrins from these cells. Taken together, these results establish that the association between ␤ 1 integrins and CD36 requires the extracellular domain of CD36.
Expressed CD36 Sequesters a Proportion of ␤ 1 Integrin to DIGs in Transfected MV3 Melanoma Cells-Membrane microdomains, or rafts, are specialized compartments of the plasma membrane that are highly enriched in cholesterol and glycosphingolipids (53)(54)(55)80). They are also greatly enriched in many signal-transducing molecules and are therefore thought to represent signaling microdomains within the membrane (56,57,67). These rafts are relatively insoluble in detergents such as Triton X-100 by virtue of their lipid constituents, and this feature has enabled the well validated biochemical approach of DIG isolation to identify the proteins located within the microdomains (53,56,67,80). Relatively few transmembrane glycoproteins have been identified as raft-associated in this way. However, CD36 has been characterized as one such molecule that has been shown to partition into DIGs isolated from murine lung endothelial cells (56), human platelets (57), and transfected COS-7 cells (58), where it may engage in signal transduction (44,65,81). We therefore determined whether CD36 also partitioned into DIGs isolated from the transfected MV3 cells, and, because this may provide a possible mechanism for the influence of CD36 on integrin function, we examined whether CD36 expression altered the proportion of ␤ 1 integrin contained in DIGs.
MV3.BOS and MV3.CD36 transfected cells were lysed with 1% Triton X-100 in MES-buffered saline buffer and subjected to floatation on sucrose gradients as described under "Experimental Procedures." Following fractionation of the gradient, equal protein amounts from each fraction were separated on SDS-PAGE, and the distribution of various antigens was analyzed FIG. 4. Co-localization of the ␣ 3 ␤ 1 integrin with CD36 in transfected MV3 melanoma cells by confocal microscopy. MV3.CD36 cells adhered to glass coverslips were immunostained for both the ␣ 3 integrin subunit and CD36 as described under "Experimental Procedures." Two captured images are shown: ␣3 (green) and CD36 (red) channel images are merged to show areas of co-localization (highlighted by arrowheads). Bar, 10 m. by immunoblotting. As expected, CD36 was not detected in the BOS-transfected cells, but, following CD36 transfection, it was found to be greatly enriched in the fractions representing the light-refractile DIG band (Fig. 7, compare A and B). Control and CD36-transfected lines displayed abundant caveolin, also concentrated in the DIG fractions. The same fractions were also probed for an ER resident protein, calreticulin (82), and, as expected, this did not co-migrate with the DIGs but was primarily confined to the fractions of the gradient corresponding to soluble proteins (Fig. 7, A and B). The ␤ 1 integrin subunit was observed predominantly in the soluble region of the gradient from both CD36 and control transfectants, the upper band representing the mature molecule and the lower band corresponding to the precursor (69). However, whereas in the control (MV3.BOS) transfectants a small amount of ␤ 1 integrin was found in DIGs, being entirely the mature form, the proportion of this mature protein found in the DIG fraction was considerably relatively enriched in fractions from the CD36 transfectants. CD9, a member of the TM4SF family and known to form a physical association with ␤ 1 integrins (20,22,23,25,32), was also examined for its distribution in these gradients. This molecule was well represented in most fractions but was most enriched in fractions 5 and 6 just below the DIGs and above the "soluble" region of the gradient. These fractions do not contain much biotinylated material from lysates of cells surface-labeled with biotin, and they are also enriched in the Golgi marker GS28 2 ; therefore, they may represent intracellular material. Notably, however, there appeared to be no difference in the relative amounts of CD9 in the DIG fractions from the controlor CD36-transfected cells. Densitometric analyses of the data confirmed these observations. The wild-type CD36, wild-type ICAM-1, CD36/ICAM tail, and ICAM/CD36 constructs encode 472, 532, 497, and 492 amino acids, respectively. B, MV3 cells (ϳ2 ϫ 10 7 cells) were transiently transfected with the empty pEF.BOS vector (BOS), wild-type CD36, or the CD36/ ICAM tail chimera (CITail). Following transfection, the cells were biotin-labeled on the cell surface and lysed in 1% Brij lysis buffer, and immunoprecipitations (i.p.) were performed for CD36 (VM58, IA7, and IE8 mAb mixture) using 90% of the total cell lysate and for ␤ 1 integrins (␤ 1 pAb) using 10% of cell lysate. The samples were resolved by 7.5% SDS-PAGE under reducing conditions and transfered to nitrocellulose, and biotinylated proteins were visualized with streptavidin-horseradish peroxidase as described under "Experimental Procedures." C, MV3 cells (ϳ10 ϫ 10 7 cells) were transiently transfected with the CD36/ ICAM tail construct before lysis with 1% Brij lysis buffer. Immunoprecipitations were performed for CD36 (11H5 mAb) and for the indicated integrin subunits ␣ 2 , ␣ 3 , and ␣ 6 before resolving the samples by SDS-PAGE under reducing conditions. The gel was transferred to nitrocellulose, and immunoblotting was performed for CD36 as described under "Experimental Procedures." D, MV3 cells (ϳ2 ϫ 10 7 cells) were transiently transfected with either CD36, BOS, or the ICAM/CD36 (I/ CD36) chimera and analyzed as described for B. The lysates were subjected to immunoprecipitation with either the anti-CD36 mAb mixture (CD36; IA7/IE8), or with anti-ICAM-1 (BOS and I/CD36; IH4 mAb). Note that the increased migration seen with the expressed ICAM/CD36 construct results from differences in glycosylation. All results were confirmed in three separate experiments. In these analyses (Fig. 7C), the proportion of the label for each marker in the DIG fractions was measured as a fraction of the total antigen density across the gradient (see "Experimental Procedures"). Both CD36 and caveolin were contained primarily in the DIG fractions, and the proportion of caveolin in these fractions was the same in both transfected cell populations. Similarly, the ER resident protein, calreticulin, was found across the soluble fractions with little protein identified in DIGs; this distribution also did not alter after CD36 transfection. In contrast, the proportion of mature ␤ 1 integrin found in the DIG fraction, although small, was found to be increased ϳ4-fold in the CD36 transfectants compared with the vector-only controltransfected cells (17.8 versus 4.5%, respectively). CD9 was distributed across the gradient, with a substantial proportion found in the insoluble fractions; however, this distribution was unaltered between the two transfected cell populations.
These results suggest that CD36 can sequester a proportion of ␤ 1 integrin into membrane microdomains, where it is more likely to encounter signaling molecules that could modulate the function of the integrins. It is possible that substantially higher proportions of ␣ 3 ␤ 1 and ␣ 6 ␤ 1 were specifically sequestered, but blotting antibodies to the ␣-subunits were not available for this study. DISCUSSION It is increasingly being recognized that integrin function can be modulated by association with other transmembrane glycoproteins, as well as by cytosolic proteins, and several such integrin-associated glycoproteins have been identified. To this list can now be added CD36. As demonstrated here in melanoma cells, CD36 forms a physical association with the ␤ 1 integrins ␣ 3 ␤ 1 and ␣ 6 ␤ 1 , and this association requires the extracellular domain of CD36; co-expression of CD36 in MV3 melanoma cells results in increased migration on extracellular matrix components, indicating that the association has functional consequences. This requirement for the extracellular domain of CD36 for integrin association is shared with other molecules laterally associated with integrins, including IAP, the GPI-linked uPAR, and, notably, members of the TM4SF family (20,24,(33)(34)(35)40).
The specific integrins identified as associating with CD36 were ␣ 3 ␤ 1 and, to a lesser extent, ␣ 6 ␤ 1 . It is striking that these particular integrins are also those most frequently identified as associating with members of the TM4SF family (21,22,25,27,28,32). In order to demonstrate the association of CD36 and integrins by co-immunoprecipitation, we lysed the cells in Brij 96 detergent. Some co-precipitation (although less) was also demonstrated after cell lysis in 1% octyl glucoside, but the association was disrupted after lysis of the cells in 1% Nonidet P-40 or 1% Triton X-100. 2 These features are shared with several reported TM4SF-integrin associations, and Hemler and associates (21,22,24,25) have rigorously demonstrated that such associations are not the result of detergent artifact but rather depend upon the strength of the association and its capacity to withstand "stringent" detergents. Some suggested TM4SF-integrin associations have been based upon co-immunoprecipitation results after lysis in CHAPS, Brij 99, or Brij 58 but do not withstand lysis in Brij 96 (as used in the present study), which is more hydrophobic; therefore, the CD36-integrin association is reasonably robust. Recently, however, Yaunch et al. (26,34) reported a direct association between an extracellular domain of the TM4SF protein CD151 and integrin ␣ 3 ␤ 1 that occurred at high stoichiometry even after cell lysis in 1% Triton X-100 and was relatively resistant to denaturing detergents. Further, an association between CD151 and ␣ 3 ␤ 1 integrin was found on every cell and tissue type examined. CD151 can also associate with other TM4SF proteins under milder conditions of lysis (26), leading these authors to postulate that some accounts of TM4SF-integrin association may be due to co-precipitation with CD151-␣ 3 ␤ 1 complexes. These findings suggest the possibility that CD36 may form part of a ternary complex with the integrin and CD151. Perhaps in support of this, in a chemical cross-linking study of platelets, Dorahy et al. (43) demonstrated that the platelet integrin ␣ IIb ␤ 3 formed a physical association with both CD36 and the TM4SF protein, CD9. Studies are in progress to determine whether CD36 associates with CD151 in melanoma cells, but it should be noted that whereas CD151 associates with the immature ␣ 3 precursor (34), CD36 was found to associate only with the mature ␤ 1 molecule.
An important finding in the present study was that ␤ 1 integrin contained within the DIG fraction isolated from MV3 were lysed with 1% Triton X-100 in MES-buffered saline (pH 6.5) and subjected to sucrose density gradient fractionation, SDS-PAGE, and immunoblotting with the indicated antibodies. The immunoblots represent equal protein loading (5 g of protein/fraction) of the 12 ϫ 1-ml fractions collected from the top of the sucrose gradient. The boxed fractions indicate the location of the light-refractive DIGs. Immunoblots with polyclonal antibodies against CD36, caveolin, and calreticulin were assayed under reducing conditions, whereas polyclonal anti-␤ 1 integrin and anti-CD9 (mAb FMC56) were analyzed under nonreducing conditions of electrophoresis (7.5-12% gradient SDS-PAGE). C, enrichment of each antigen in DIGs (determined from the immunoblots shown in A and B) was determined by densitometric analysis as described under "Experimental Procedures." Similar distributions were found for each antigen in at least two separate experiments. melanoma cells transfected with CD36 was relatively enriched compared with control cells. Several studies have validated the concept that such biochemically isolated DIGs are representative of specialized plasma membrane microdomains or rafts (53,56,67,80). Caveolin, the marker coat protein of the distinct membrane structures known as caveolae was also found in this fraction, raising the possibility that association with CD36 facilitates the entry of ␤ 1 integrins into caveolae. Caveolae are thought to bud from membrane microdomains in a process controlled by caveolin oligomers, and, in addition to their characteristic enrichment in cholesterol and sphingolipid, they are greatly enriched in cellular signaling proteins (56,57,67). Integrins are not generally considered to locate to caveolae (56,62), but recent work has revealed that a fraction of ␤ 1 integrin in Fisher rat thyroid cells associates with both caveolin and the protein-tyrosine kinase Fyn, and Fyn-dependent Shc-mediated signaling in response to integrin ligation was found to be dependent upon the co-expression of caveolin-1 (63). Also, by use of antisense methodology in kidney 293 cells, Wei et al. (64) found that caveolin was important to ␤ 1 integrin-dependent adhesion to fibronectin and the activation of focal adhesion kinase. The authors propose that ␤ 1 integrins form a complex with caveolin that is stabilized by the nonintegrin, GPI-linked urokinase receptor, uPAR; this complex, in turn, regulates the ability of caveolin to regulate members of the Src family of protein-tyrosine kinases surrounding integrins.
Integrins co-precipitate with endogenous and transfected uPAR in a complex that also contains caveolin (40,83). Caveolin itself can also bind cholesterol, Src family kinases, heterotrimeric G proteins, and Ha-Ras (84,85), thereby contributing to the formation of caveolae rich in signaling molecules. It might be considered therefore that by associating with ␤ 1 integrins through its external domain, CD36 can substitute for or mimic uPAR in providing a bridge for caveolin association. Three observations militate against this possibility. First, in our co-precipitation studies, we were unable to detect any caveolin in the CD36/␤1 precipitates by immunoblotting. 2 Second, in an immunoultrastructural study of the CD36-transfected MV3 cells, it was found that the CD36 appeared to compete with caveolin such that cells expressing high levels of CD36 displayed considerably fewer of the flask-like invaginations representative of morphological caveolae. 3 The reason for this is not known at present, but this result does suggest that CD36 is not simply a passive facilitator of caveolae association. Third, CD36 locates to biochemically isolated plasma membrane rafts even in the absence of caveolin; in platelets that lack caveolin expression, CD36 is found highly enriched in the DIG fraction together with active Lyn protein-tyrosine kinase (57,81). Further, in these enucleate cells, CD36 physically associates with the platelet integrin ␣ IIb ␤ 3 (43) and with the tyrosine-protein kinases, Fyn, Lyn, and Yes (65). CD36 itself may bind cholesterol (86) and fatty acids (87) and co-precipitates with Src family protein-tyrosine kinases from a number of other cell types, including melanoma cells (65,66). Together, these data suggest that CD36 may play a role in guiding integrins to signaling rafts. Many hemic cells lack caveolin expression and do not form caveolae (88), and the down-regulation of caveolin expression is a feature of cell transformation (89). In contrast, CD36 expression is highly prominent in hemic cells such as platelets and macrophages (41,42) as it is on melanoma cells in situ (90) and breast cancer cells (91). We tentatively suggest, therefore, that in those cells expressing CD36 but lacking caveolin expression, CD36 may play an important role in the regulation of integrin function.
Precisely how CD36 might regulate integrin function remains to be determined, but several scenarios can be envisioned. By the recruitment of adjacent integrins into membrane microdomains together with Src-related kinases, CD36 could serve to promote integrin-mediated signaling. Further, membrane microdomains are highly enriched in gangliosides (80), which themselves can regulate protein kinase function (92) and, more particularly, have been shown physically to associate with Arg-Gly-Asp-directed integrins and to be involved in their adhesive function (72,93,94). Not necessarily unrelated to such a role in sequestration is the possibility that CD36 can function as an accessory molecule, complementing integrin ligation with certain extracellular matrix ligands. Precedents for integrin accessory molecules regulating or stabilizing the adhesive function of associated integrins include uPAR, which contains a vitronectin binding site and can regulate the relative binding of ␣ 5 ␤ 1 integrin to fibronectin or vitronectin (40), and the 67-kDa monomeric laminin receptor that regulates the laminin-binding function of the ␣ 6 ␤ 4 integrin (95). CD36 has been described as a cell surface receptor for both collagen and thrombospondin (49,50). Tandon and colleagues (96,97) have recorded that CD36 expression influences the rate of platelet attachment and spreading on type I collagen mediated by the ␣ 2 ␤ 1 integrin. ␣ 3 ␤ 1 integrin also can bind collagen; however, in our study with transfected MV3 melanoma cells we could detect no influence of CD36 expression on the degree or rate of spreading of these cells on collagen or on the ␣ 6 ␤ 1 substrate, laminin (data not shown). CD36 binding to the pericellular matrix glycoprotein, thrombospondin, may be a more attractive candidate for the modulatory role of CD36 in cell migration. Thrombospondin is abundantly secreted by many migratory cell types, including melanoma cells; it can bind to several extracellular matrix proteins and, functioning as an antiadhesive molecule, can promote cell migration (reviewed in Ref. 98). Thrombospondin contains several distinct domains able to bind to discrete cell surface receptors, including CD36 and ␣ v and ␤ 1 integrins (98), and fibroblast migration on thrombospondin has been shown to include a role for both integrins and CD36 (99). Of interest, the integrin-associated protein (IAP or CD47), which has been demonstrated to physically associate with integrins and to modulate their function (13)(14)(15)(16)33), also functions as a receptor for the carboxyl-terminal region of thrombospondin (17,100). In a recent study, Wang and Frazier (18) demonstrated that IAP physically associated with ␣ 2 ␤ 1 integrin on aortic smooth muscle cells and that thrombospondin peptides able to bind IAP stimulated the chemotaxis of those cells on gelatin-coated filters.
Finally, at the initiation of this study, we expected to find an association between CD36 and ␣ v ␤ 3 integrin because of reports of the coordination of these receptors in the phagocytic uptake of apoptotic cells (46). However, we were unable to demonstrate co-precipitation of these molecules from WM115 melanoma cells that express both in abundance, or from MV3 cells transfected with both CD36 and ␤ 3 . 2 The ready demonstration of a CD36-␣ 3 ␤ 1 association under these conditions suggests that, if CD36 and ␣ v ␤ 3 act coordinately in phagocytosis, they do so without being in physical association. However, while a role for CD36 in this function has been well validated in transfection studies (101) and a CD36 family member, croquemort, has been found in genetic studies to be essential for the removal of apoptotic cells during Drosophila development (102), similar studies have not been carried out to determine the role of ␣ v ␤ 3 . Studies documenting a role for ␣ v ␤ 3 in this function have depended upon antibody-induced or peptide agonist-induced inhibition (46, 103-106). More recent work has established that the engagement of ␤ 3 integrins with antibody or ligand can 3 R. G. Parton, R. F. Thorne, and G. F. Burns, unpublished data. result in the down-regulation of the function of ␤ 1 integrins in a signal-mediated process known as transdominant inhibition (8,9). Further, it has been established that ␣ 3 ␤ 1 participates in phagocytosis mediated by human breast cancer cells (107), most of which express CD36 (91). Therefore, the findings reported here of an association between CD36 and ␣ 3 ␤ 1 integrin may indicate that the conclusions of studies implicating ␣ v ␤ 3 in the phagocytic process may warrant further evaluation to assess the role of ␤ 1 integrins and to consider the possibility that ␣ v ␤ 3 engages in integrin receptor cross-talk.