The Human Integrin α8β1 Functions as a Receptor for Tenascin, Fibronectin, and Vitronectin

The integrin family of adhesion receptors consists of at least 21 heterodimeric transmembrane proteins that differ in their tissue distribution and ligand specificity. The recently identified α8 integrin subunit associates with β1 and is predominantly expressed in smooth muscle and other contractile cells in adult tissues, and in mesenchymal and neural cells during development. We now show that α8β1 specifically localizes to focal contacts in cells plated on the extracellular matrix proteins fibronectin or vitronectin. In addition we show that human embryonic kidney cells (293), transfected with α8 cDNA, express α8β1 on their surface and use this receptor for adhesion to fibronectin and vitronectin. Furthermore, α8β1 binds to both fibronectin- and vitronectin-Sepharose and can be specifically eluted from either matrix protein by the arginine-glycine-aspartic acid (RGD)-containing peptide, GRGDSP. Because fibronectin and vitronectin adhesion appeared to be mediated by RGD, we examined additional RGD-containing proteins, including tenascin, fibrinogen, thrombospondin, osteopontin, and denatured collagen type I. We found that only tenascin was able to mediate adhesion of α8-transfected 293 cells. By using recombinant fragments of tenascin in adhesion assays, we were able to localize the α8β1 binding domain of tenascin to the RGD-containing, third fibronectin type III repeat. These data strongly suggest that tenascin, fibronectin, and vitronectin are ligands for α8β1 and that this integrin binds to the RGD site in each of these ligands through mechanisms that are distinct and separate from α5-and αv-containing integrins.

Integrins are a class of cell adhesion glycoproteins composed of two noncovalently associated subunits, ␣ and ␤. Each subunit contains a large extracellular domain, a transmembrane domain, and a short cytoplasmic domain. Integrins are known to bind to a wide variety of extracellular matrix proteins, including fibronectin, vitronectin, collagens, and laminins. The specificity of protein binding is determined by particular combinations of ␣ and ␤ subunit pairing. The ligand binding site is formed by the extracellular domain of both subunits and requires the presence of divalent cations. Many integrins interact with ligands through the tripeptide arginine-glycine-aspartic acid (RGD).
The ␣8 integrin subunit was originally identified by Bossy et al. (1) in the chick embryo nervous system and was shown to be a partner for ␤1. We have identified human ␣8, cloned and sequenced the cDNA, raised antibodies to the predicted cytoplasmic domain sequence, and determined its distribution in adult mammalian tissues (2). We found that ␣8 is predominantly expressed in a variety of visceral and vascular smooth muscle cells, kidney mesangial cells, and lung myofibroblasts (2).
To gain insight into potential functions of ␣8␤1 in vivo, we sought to determine potential ligands. We tested various ligands for their ability to direct ␣8␤1 to focal contacts, to bind to ␣8␤1 by affinity chromatography, and to support adhesion of ␣8-transfected cells. All of the results provide strong evidence that ␣8␤1 can function as an RGD-dependent receptor for tenascin, fibronectin, and vitronectin.

EXPERIMENTAL PROCEDURES
Materials-Fibronectin was prepared from human plasma as described by Engvall and Ruoslahti (3), and vitronectin was prepared from human plasma according to Yatohgo and co-workers (4). Type I collagen from rat tail, type IV collagen from human placenta, and fibrinogen were purchased from Sigma. Type I collagen was heatdenatured by incubating at 100°C for 15 min (5). Tenascin was purchased from Life Technologies, Inc. Recombinant tenascin fragments were a gift from Dr. Kathryn Crossin, Scripps Research Institute, La Jolla, CA (6). Osteopontin was a gift from Dr. Cecilia Giachelli, University of Washington, Seattle, WA. Laminin was a gift from Dr. Randy Kramer, University of California, San Francisco.
Immunofluorescence-Coverslips were coated overnight at 4°C with 10 -20 g/ml of the following extracellular matrix proteins: fibronectin, vitronectin, collagen types I and IV, denatured collagen type I, laminin, and fibrinogen. Coverslips were blocked with 3% bovine serum albumin (BSA), 1 phosphate-buffered saline (PBS) at 37°C for 30 min. Cells grown to 80% confluence were removed from tissue culture plates with 2 mM EDTA, PBS, washed with PBS, resuspended in serum-free Dulbecco's modified Eagle's medium, and plated onto the coverslips. Three hours after plating, cells were fixed and permeabilized with 2% paraformaldehyde, 0.1% Triton X-100 for 10 min. Coverslips were blocked with 3% BSA and then incubated with anti-␣8 antibody (10 g/ml) and anti-vinculin antibody (2 g/ml) for 1 h at room temperature. Unbound primary antibody was removed by washing with PBS. Coverslips were incubated with biotin-conjugated donkey anti-rabbit IgG (1:50) (Amersham Corp.) and Texas Red-conjugated goat antimouse IgG (1:100) (Caltag, South San Francisco, CA) for 1 h at room temperature, followed by PBS wash. Coverslips were then incubated with fluorescein conjugated streptavidin (1:100) (Amersham) for 15 min at room temperature, washed with PBS, and mounted with Vectashield (Vector Laboratories, Burlingame, CA).
To block endogenous production of extracellular matrix proteins, cells were incubated with the protein synthesis inhibitor cycloheximide (30 g/ml) for 3 h in serum-free medium. Cells were then detached with 2 mM EDTA and seeded onto fibronectin-and vitronectin-coated coverslips in the presence of cycloheximide.
Preparation of cDNA Expression Constructs-A cDNA containing the entire coding region of human ␣8 was initially constructed in pBluescript (Stratagene, La Jolla, CA). The full-length ␣8 cDNA was amplified by PCR using the previously reported cDNA clone HA33A (2) as a template. This clone lacks the signal peptide and contains a deletion of nucleotides 688 -732. Because the signal sequence of human ␣8 has not been determined, we joined the N terminus of mature ␣8 with a synthetic signal sequence derived from the ␣M (Mac-1) integrin subunit. We chose ␣M because the N-terminal four amino acid residues of ␣M and ␣8 are identical (FNLD). We designed two overlapping forward PCR primers that encoded for the signal peptide of the human ␣M integrin subunit (16) and the amino terminus of mature ␣8 (Mac 2F, 5Ј-CCTCCTCGAAAGCTTCTCCTTCCAGCCATGGCTCTCAGAGTCCTT-CTCTTAACAGC-3Ј; Mac 1F, 5Ј-AGAGTCCTTCTCTTAACAGCCTTAT-GTCATGGGTTCAACCTGGACGTGGAAAA-3Ј). The Mac 2F primer contains a HindIII recognition sequence (underlined), 12 nucleotides of the ␣M 5Ј untranslated region, followed by the ATG initiation codon (double underlined) and the sequence encoding the first nine amino acids of the ␣M signal peptide (ALRVLLLTA). The Mac 1F primer has a 20 nucleotide overlap (underlined) with the 3Ј end of the Mac 2F primer and encodes the remainder of the ␣M signal peptide (ALCHG) followed by the N terminus of mature ␣8 (FNLDV). The first PCR reaction used forward primer Mac 1F and a ␣8-specific 3Ј reverse primer and HA33A plasmid DNA as a template. The second PCR reaction used forward primer Mac 2F and the same ␣8-specific reverse primer and the first PCR reaction cDNA as a template. In order to correct the deletion, we amplified the region from clone HA15-A that does not contain the deletion (2) and used unique restriction sites (BclI and HpaI) in the ␣8 clone in order to ligate the fragment. All polymerase chain reactions were performed with Vent DNA polymerase (New England Biolabs) which has been reported to provide significantly higher fidelity than Taq polymerase. The plasmid was then completely sequenced using Sequenase 2.0 (U. S. Biochemical Corp., Cleveland, OH) and found to be free of mutations that would alter the encoded amino acid sequence. The plasmid was cut at the unique HindIII and XbaI sites in the pBluescript polylinker, and ligated into the mammalian expression vector pCDNAIneo between the unique HindIII and XbaI sites to generate pCDNAIneo␣8.
Transfection of Mammalian Cells-The human embryonic kidney cell line 293 and human colon carcinoma cell line SW 480 were transfected using the Lipofectin reagent (Life Technologies, Inc.) according to the manufacturer's instructions. Stably transfected cell lines were selected in medium containing the neomycin analog G418 (0.4 mg/ml). Cells were transfected with either pCDNAIneo␣8 (␣8-transfected cells) or pCDNAIneo alone (mock-transfected cells).
Cell Adhesion Assays-Non-tissue culture-treated plates were coated with increasing concentrations (0.3, 1, 3, 10, 20 g/ml) of fibronectin, vitronectin, thrombospondin, osteopontin, denatured collagen type I, fibrinogen, and intact tenascin. Plates were also coated with increasing concentrations of the recombinant tenascin fragments containing the third fibronectin type III repeat (TNfn3), the third fibronectin type III repeats in which the RGD site had been mutated to RAA (TNfn3RAA) or the fourth to sixth fibronectin type III repeats (TNfn4 -6). As a negative control, wells were coated with 1% BSA, PBS. Wells coated at 37°C for 1 h were washed with PBS and blocked with 1% BSA for 30 min at 37°C. Cells were detached with 2 mM EDTA, washed with PBS, and resuspended in serum-free Dulbecco's modified Eagle's medium (containing 200 g/ml CaCl 2 , 200 g/ml MgCl 2 ). 50,000 cells were added to each well, centrifuged at 10 ϫ g for 3 min to ensure uniform settling of cells, and incubated for 1 h at 37°C. Nonadherent cells were then removed by centrifugation (top-side down) at 10 ϫ g for 5 min. The attached cells were fixed and stained with 1% formaldehyde, 0.5% crystal violet. After washing with PBS, adherence was determined by absorption at 595 nm in a Microplate Reader (Bio-Rad). The data were reported as the mean absorbance of triplicate wells Ϯ S.E., minus the mean absorbance of BSA-coated wells.
Peptide blocking experiments were carried out in an analogous manner. Peptides were used at the following final concentrations: 125 M RGDGW, 125 M CRRETAWAC, 1 M 4C, and 100 g/ml GRGDSP.
The data reported represent the results from one clone of mocktransfected and ␣8-transfected 293 cells. The adhesion assay results were confirmed with four independent clones of ␣8-transfected 293 cells and with wild type (untransfected) 293 cells (data not shown).
Affinity Chromatography-Cells grown to 80% confluence were detached with 2 mM EDTA in PBS. Cells were centrifuged and washed three times in labeling buffer (50 mM Tris-HCl, 150 mM NaCl, 1 mM MnCl 2 , 20 mM glucose). Cells were surface labeled with 125 I-sodium iodide using the lactoperoxidase method (17).
Affinity columns were prepared by coupling the ligand (fibronectin or vitronectin) to cyanogen-bromide activated Sepharose as described previously (17). 125 I-Labeled cell lysates were applied to a 1-ml affinity column. The column was washed with 10 volumes column buffer (50 mM octylglucoside, 50 mM Tris-HCl, 1 mM divalent cation (MnCl 2 , CaCl 2 , or MgCl 2 ). Elutions were carried out using 4 volumes of GRGDSP peptide (1 mg/ml), followed by 2 volumes of column buffer, and 4 volumes of 10 mM EDTA in column buffer without divalent cations. Finally the columns were washed with 4 volumes of 1 M NaCl in column buffer. Fractions (1 ml) were collected and either analyzed directly by SDS-PAGE or subjected to immunoprecipitation and then analyzed by SDS-PAGE as described previously (17).

RESULTS AND DISCUSSION
Focal Contact Formation-Many cells in culture, including smooth muscle cells and fibroblasts, attach to extracellular matrix proteins at discrete sites called focal contacts or adhesion plaques. Focal contact localization of integrins is liganddependent; i.e. a particular integrin will accumulate at focal contacts only when its ligand is present in the substrate (18,19). We therefore examined focal contact localization of ␣8␤1 on cells plated on various extracellular matrix proteins as a first step toward identifying ligands for ␣8␤1. In our experiments we used a human smooth muscle cell line (HISM) and a rat embryo fibroblast (REF) cell line, which we found to express ␣8␤1 by immunoprecipitation (Fig. 1, lanes 1 and 2).
HISM cells plated on extracellular matrix proteins were analyzed by double-labeling immunofluorescence microscopy using antibodies to vinculin (to detect focal contacts) and ␣8 (Fig. 2). Three hours after plating on fibronectin, vitronectin, or collagen, HISM cells were well spread and formed vinculincontaining focal contacts. In cells plated on fibronectin or vitronectin, ␣8 co-localized to vinculin-containing focal contacts (Fig. 2, A-D). In contrast, in cells plated on collagen type I or denatured collagen type I, ␣8 did not localize to focal contacts, despite the abundance of focal contacts identified by vinculin staining (Fig. 2, E and F). Cells plated on tenascin adhered, but did not form vinculin-containing focal contacts (data not shown). Cells plated on fibrinogen or laminin attached poorly and did not spread or form focal contacts. No localized ␣8 staining was detected in these cells (data not shown). Identical results were obtained using rat embryo fibroblast cells (data not shown). It is unlikely that ␣8 localization was due to matrix protein secretion by the cells, because control experiments using cells pretreated with the protein synthesis inhibitor, cycloheximide, yielded similar results (Fig. 3). We were not able to demonstrate focal contact formation by ␣8-transfected or untransfected 293 cells. However, ␣8-transfected SW 480 colon carcinoma cells showed new localization of ␣8 to focal contacts when plated on fibronectin and vitronectin (data not shown). Thus, vitronectin and fibronectin, but not collagen, specifically promote localization of ␣8 to focal contacts. These data suggest vitronectin and fibronectin are ligands for ␣8␤1.
Heterologous Expression of ␣8 cDNA-The human embryonic kidney cell line 293 is highly permissive for transfection of heterologous cDNAs and does not express ␣8 as determined by immunoprecipitation (Fig. 1, lane 3). After transfection of 293 cells with ␣8 cDNA we obtained four independent clones that expressed ␣8 as detected by Western blot analysis (data not shown). Immunoprecipitation of surface-labeled cells demonstrated that ␣8␤1 was expressed on the cell surface (Fig. 1,  lane 4).
To determine whether the ␣8 transfectants have altered levels of other ␤1-associated ␣ subunits, we performed fluorescein-activated cell sorting analysis using a panel of anti-integrin antibodies (Fig. 4). Mock-transfected and ␣8-transfected 293 cells contained similar amounts of cell surface ␣2-, ␣3-, ␣5-, and ␣v-containing integrins. Thus, changes in adhesive properties of ␣8-transfected cells are not due to changes in the surface expression of other integrins.
Adhesion of ␣8-transfected 293 Cells-To determine whether ␣8␤1 can mediate cell adhesion to fibronectin and vitronectin, we compared the adhesive properties of mock-transfected and ␣8-transfected 293 cells. Fibronectin and vitronectin adhesion were not significantly affected by ␣8 expression (Fig. 5), although a slight increase in adhesion to 3 g/ml vitronectin was noted. When endogenous receptors for fibronectin and vitronectin were blocked with monoclonal antibodies, the contribution of ␣8␤1 was more apparent (Figs. 6 and 7). The adhesion of wild type or mock-transfected cells to fibronectin was almost completely inhibited (Ͼ90%) by either anti-␤1 antibody (P5D2) or anti-␣5 antibody (P3D10) (Fig. 6A). In contrast, ␣8-transfected cell adhesion was only partially blocked by anti-␣5 antibody (33%), but was completely blocked by anti-␤1 antibody. These data suggest that the ␣8-transfected cells are using ␣8␤1 in addition to ␣5␤1 to adhere to fibronectin.
Vitronectin adhesion of mock-transfected and wild type 293 cells was almost completely inhibited by the blocking anti-␣v antibody, L230 (Fig. 7A), consistent with previous reports (15,20). In contrast, vitronectin adhesion in ␣8-transfected cells was only partially inhibited (19%) by the anti-␣v antibody, L230, and inhibited by 41% using the anti-␤1 antibody, P5D2. Vitronectin adhesion was completely abolished using both antibodies in combination. These data suggest that ␣8-transfected cells are using ␣8␤1 in addition to ␣v-containing integrins, to adhere to vitronectin.
To further elucidate the binding characteristics of ␣8␤1 to fibronectin and vitronectin, adhesion assays were performed in the presence of three different synthetic peptides (Figs. 6B and 7B). Although integrins can interact through a common RGD site in the ligand, conformationally constrained peptides can discriminate between various RGD binding integrins. The cyclic peptide, CRRETAWAC, has recently shown to be highly selective for ␣5␤1 (13). At concentrations sufficient to block adhesion of ␣5␤1 to fibronectin, CRRETAWAC does not block ␣v␤1 fibronectin adhesion or ␣v-mediated vitronectin adhesion (13). In an analogous fashion, the cyclic peptide 4C selectively inhibits ␣v-mediated adhesion (21). In contrast, the peptides GRGDSP and RGDGW are able to block both ␣5and ␣v-  HISM cells (lane 1), REF cells (lane 2), 293 cells (lane 3), and  ␣8-transfected 293 cells (lane 4). Aliquots of 125 I-surface-labeled lysates were immunoprecipitated with anti-␣8 antibody. Proteins were analyzed by SDS-PAGE under nonreducing conditions. Positions of molecular size markers in kilodaltons are shown to the right.  (12). We took advantage of these selective peptides to further define the binding characteristics of ␣8␤1. We found that adhesion of ␣8-transfected cells to fibronectin was inhibited by the peptide RGDGW, but not by the CRRE-TAWAC peptide (Fig. 6B). The addition of the ␣5 blocking antibody, P3D10, to CRRETAWAC, did not significantly decrease fibronectin adhesion (data not shown). In contrast, adhesion of mock-transfected 293 cells to fibronectin was inhibited by either RGDGW or CRRETAWAC (Fig. 6B). Adhesion of mock-transfected and ␣8-transfected 293 cells to fibronectin was not affected by the ␣v-selective 4C peptide (data not shown). These results suggest that ␣8␤1 interacts with fibronectin by mechanisms that are similar to, but distinguishable from, those used by ␣5␤1.
Adhesion of ␣8-transfected cells to vitronectin was inhibited by the peptide RGDGW, but not by the 4C peptide (Fig. 7B). The addition of the ␣v blocking antibody, L230, to 4C did not further inhibit vitronectin adhesion (data not shown). In contrast, adhesion of mock-transfected 293 cells to vitronectin was inhibited by either RGDGW or 4C (Fig. 7B). Adhesion of mocktransfected and ␣8-transfected 293 cells to vitronectin were not affected by the ␣5-selective peptide, CRRETAWAC (data not shown). Thus, the 4C peptide, at the concentration used, blocks ␣vbut not ␣8␤1-mediated adhesion to vitronectin, suggesting that ␣8␤1 interacts with vitronectin through mechanisms distinct from those used by ␣v integrins.
Adhesion of ␣8␤1 to Additional RGD-containing Proteins-Because the above experiments suggested that ␣8␤1 was binding to the RGD sites in fibronectin and vitronectin, we examined additional RGD-containing proteins for the ability to bind to ␣8␤1. We tested the ability of mock-and ␣8-transfected 293 cells to adhere to tenascin, fibrinogen, thrombospondin, osteopontin, and denatured collagen type I (Fig. 10). We found that ␣8-transfected 293 cells adhered and spread well on tenascin, whereas mock-transfected 293 cells did not adhere (Fig.  10). Neither mock-transfected nor ␣8-transfected 293 cells adhered to fibrinogen, thrombospondin, osteopontin, or denatured collagen type I (Fig. 10). In contrast, ␤3-transfected 293 cells adhered well to fibrinogen, thrombospondin, osteopontin, and denatured collagen type I (Fig. 10).
Tenascin is a modular protein that contains several fibronectin type III repeats. An RGD site located in the third fibronectin type III repeat of tenascin (TNfn3) has been shown to mediate adhesion of several integrins, including ␣v␤3 and probably ␣v␤6 (22). To determine whether ␣8␤1 was mediating adhesion through this RGD site, we tested the adhesion of ␣8- and mock-transfected 293 cells to various recombinant fragments of tenascin (Fig. 11). The ␣8-transfected cells adhered to the RGD-containing fragment, TNfn3, in a concentration-dependent manner. As expected, mock-transfected 293 cells did not adhere to TNfn3. In contrast, ␣8-transfected 293 cells did not adhere to a fragment containing the fourth through sixth fibronectin type III repeats, TNfn4 -6 ( Fig. 11). To confirm that ␣8-transfected cells were binding through the RGD site of TNfn3, we tested adhesion of ␣8-transfected cells to a mutant TNfn3 in which the RGD site had been altered to RAA (TNfn3RAA). We found that this abolished adhesion of ␣8transfected 293 cells (Fig. 11). In a control experiment, ␣9transfected 293 cells, which have been shown to adhere to TNfn3 at a non-RGD site (14), adhered well to TNfn3RAA (data not shown).
Adhesion to TNfn3 by ␣8-transfected cells was abolished by the anti-␤1 antibody, P5D2 (Fig. 12). We also tested the ability of the selective, synthetic peptides to block adhesion to TNfn3. The peptides, GRGDSP and RGDGW blocked adhesion of ␣8transfected cells to tenascin (Fig. 12). In contrast, the cyclic CRRETAWAC peptide did not inhibit adhesion and the cyclic 4C peptide only partially inhibited adhesion to TNfn3 (Fig. 12). Adhesion of ␤3-transfected cells to tenascin was abolished by the 4C peptide (data not shown). Taken in concert, these results suggest that ␣8␤1 is binding to the RGD site in tenascin. However, this interaction must be somewhat distinct from the ␣v-tenascin interactions because it is not completely inhibited by the peptide 4C.
In summary, we have demonstrated that ␣8␤1 can bind to tenascin, fibronectin, and vitronectin by interacting with the RGD sites on these ligands. We also show that ␣8␤1 is capable of localizing to focal contacts on fibronectin and vitronectin on fibroblasts and smooth muscle cells. ␣8␤1 is eluted from both fibronectin and vitronectin affinity columns by an RGD-containing peptide. Our fibronectin adhesion data are in agreement with the recently published observations that chicken ␣8␤1 is able to support attachment, spreading, and neurite outgrowth by transfected cells (23). Two other ␤1-containing integrins are known to interact with RGD sites: ␣5␤1 and ␣v␤1. These ␣ subunits, along with ␣IIb, are the most closely  related to ␣8 (42-43% amino acid identity). In addition, these subunits share several other structural features with each other including the presence of post-translational cleavage and the absence of I domains. Thus, ␣8, ␣5, ␣v, and ␣IIb define a subfamily of ␣ subunits that have close sequence homology, bind to RGD-containing peptides, are post-translationally cleaved, and do not contain I domains. Despite the similarities to ␣v and ␣5, the binding specificity of ␣8␤1 is unique. In contrast to ␣5␤1, whose only known ligand is fibronectin, ␣8␤1 is more promiscuous and can bind to vitronectin and tenascin as well as fibronectin. In addition, binding by ␣8␤1 is not affected by the peptide CRRETAWAC, which efficiently blocks ␣5␤1. When compared to ␣v␤3, the binding repertoire of ␣8␤1 is more limited. Although both ␣8␤1 and ␣v␤3 bind to tenascin, fibronectin, and vitronectin, ␣8␤1 does not bind to several other ␣v␤3 ligands, including fibrinogen, thrombospondin and denatured collagen. Additionally, ␣8-mediated adhesion is not affected by the peptide 4C. Thus, the binding characteristics of ␣8␤1 are unique and distinguishable from both ␣5␤1 and ␣v␤3.
In adult mammalian tissues, ␣8␤1 is prominently expressed in vascular and visceral smooth muscle cells, kidney mesangial cells, and lung myofibroblasts (2). Tenascin, fibronectin, and vitronectin are thought to play a role in the response to injury and inflammation (24). Thus, ␣8␤1 may contribute to the functional changes that occur in smooth muscle cells during tissue repair. Since smooth muscle cells also express other fibronectin, vitronectin, and tenascin receptors, such as ␣5␤1, ␣v␤5, and ␣v␤3 (25), it will be important to determine the specific functional contribution of ␣8␤1 to smooth muscle cell behavior.