Osteopontin N-terminal domain contains a cryptic adhesive sequence recognized by α9β1 integrin

Osteopontin is an adhesive glycoprotein implicated in numerous diseases associated with inflammation and remodeling. There are several structural domains in osteopontin that are of particular interest. The RGD motif is a cell attachment sequence shown to be critical for cell adhesion through αv-containing integrins. In close proximity to the RGD domain is the thrombin cleavage site. Previous observations suggest that thrombin cleavage of osteopontin occurs in vivo and may be physiologically important. To study the functional significance of osteopontin cleavage by thrombin, we made glutathione S-transferase-osteopontin fusion proteins. These proteins contain either the N- or C-terminal domains expected to be formed following thrombin cleavage at the Arg169-Ser170 peptide bond. We compared these osteopontin fragments with native osteopontin in their ability to support adhesion of several different cell lines and identified the receptors mediating these interactions. Our data show that the N-terminal osteopontin fragment, which contains the RGD domain, supports adhesion of a melanoma cell line that is unable to bind native osteopontin. This suggests that osteopontin adhesive interactions may be regulated by thrombin cleavage. We also demonstrate that osteopontin contains a cryptic binding activity, which can be recognized by a novel osteopontin receptor. This receptor has been identified as the α9β1 integrin.

Osteopontin is a multifunctional glycoprotein that promotes cell adhesion and migration. Previous studies have suggested that osteopontin plays a role in bone resorption, tumorigenesis, and metastasis (1). More recently, osteopontin has been implicated in a number of disease states associated with inflammation and tissue remodeling (2)(3)(4)(5)(6).
Many cellular interactions with osteopontin are mediated through integrin receptors. Integrins are capable of generating signals that control many aspects of cell behavior including differentiation, adhesion, migration, and apoptosis (7,8). The ␣ v ␤ 3 integrin allows a variety of cell types to adhere and migrate to osteopontin (9 -13). In addition to ␣ v ␤ 3 , the ␣ v ␤ 5 and ␣ v ␤ 1 integrins were recently found to mediate adhesion of human aortic smooth muscle cells (12) and human embryonic kidney cells to osteopontin (14). A weak interaction was also demonstrated between ␣ 4 ␤ 1 and osteopontin in the macrophage line, P388D1 (15). Interestingly, occupancy of osteopontin with different receptors has distinct functional consequences. For example, in smooth muscle cells, ␣ v ␤ 3 , ␣ v ␤ 1 , and ␣ v ␤ 5 mediate adhesion, but only ␣ v ␤ 3 can support migration (12).
Osteopontin contains several interesting structural domains. The RGD domain is an adhesive motif found in many matrix molecules (7) and is critical for ␣ v integrin-dependent cell adhesion and migration to osteopontin (12,16). Osteopontin is also susceptible to proteolytic fragmentation. There are two conserved thrombin cleavage sites in human osteopontin. The major thrombin cleavage site is at residues Arg 169 -Ser 170 , which is 6 amino acids C-terminal from the RGD domain. A second potential thrombin-cleavage site is within the RGD domain (Arg 160 -Gly 161 ) (17). Previous studies have shown that osteopontin proteolytic fragments are found in vivo and may have physiological importance (17,18). In addition, both osteopontin and thrombin are likely to be localized together at sites of injury, inflammation, angiogenesis, and in tumors. The functional activity of cleaved osteopontin, however, is unclear. One report demonstrated that thrombin cleavage destroyed RGD-mediated cell adhesion (19). In contrast, a second report showed that thrombin treatment enhanced osteopontin cell adhesive activity (20). One explanation for the discrepancy is that the interaction with osteopontin fragments may be mediated through distinct receptors in different cell types. A key to understanding the function of osteopontin, therefore, is identifying the receptors that interact with not only the full-length molecule, but with any functional proteolytic fragments.
To study the functional significance of osteopontin fragmentation, we performed adhesion experiments using glutathione S-transferase-osteopontin (GST-OPN) 1 fusion proteins. For these studies, we created osteopontin peptides that contain either the N-or C-terminal domains expected to be formed following thrombin cleavage at the Arg 169 -Ser 170 site, which is 6 amino acids C-terminal from the RGD adhesive motif. We compared the osteopontin fragments in their ability to support adhesion of several different cell lines with that of native osteopontin and identified the receptors mediating these interactions. These studies show that the N-terminal fragment of osteopontin contains a functional RGD domain recognized by ␣ v ␤ 3 , as well as a cryptic adhesive sequence recognized by the ␣ 9 ␤ 1 integrin.

EXPERIMENTAL PROCEDURES
Cell Lines-Bovine aortic endothelial cells were isolated from bovine aortas as described previously (21). Mo and Mo␣ v melanoma cells were maintained in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) containing 10% fetal calf serum. Mo and Mo␣ v were both derived from M21 melanoma cells. One subclone, Mo␣ v expresses high levels of ␣ v ␤ 3 . Mo is a subclone expressing no detectable levels of the ␣ v subunit. These cell lines were provided to us by Dr. Mark H. Ginsberg (The Scripps Research Institute) and have previously been described (22). Cell lines SW480 and 293 were obtained from ATCC.
Adhesive Proteins and Recombinant Osteopontin Fragments-Native osteopontin was purified from conditioned medium of rat pup smooth muscle cell cultures as described previously (9). Laminin and fibronectin were purchased from Life Technologies, Inc., and collagen I was purchased from Collaborative Biochemical Products (Bedford, MA).
Osteopontin N-and C-terminal proteins were generated by thrombin cleavage of bacterially expressed GST-OPN fusion proteins. Expression plasmids containing GST-OPN were generated by cloning polymerase chain reaction-amplified N-and C-terminal osteopontin fragments into BamHI/EcoRI sites of pGEX-2T (Pharmacia Biotech Inc.). The 5Јprimer CGCGGATCCATACCAGTTAAACAGGCT and the 3Ј-primer TCCCCCGGGTCACCTCAGTCCATAAAC were used to amplify the Nterminal osteopontin fragments, 10N and 30N, from the plasmids OP10 and OP30 respectively. The plasmid OP10 (37) was provided by Dr. Larry Fisher. OP30 (38) was obtained from ATCC. The C-terminal 10C fragment was amplified from OP10 using 5Ј-primer CGCGGATC-CAAATCTAAGAAGTTTCGC and 3Ј-primer TCCCCCGGGTTAATT-GACCTCAGAAGA. The GST-OPN fusion constructs were DNA sequence-verified. Escherichia coli JM109 cells transformed with these GST-OPN plasmids were grown in LB with 150 g/ml ampicillin and then induced with 0.1 mM isopropyl-1-thio-␤-D-galactopyranoside for 2 h at 37°C to express the fusion proteins. The GST-OPN fusion proteins were purified basically according to the manufacturer's instructions (GST gene fusion system, Pharmacia, Piscataway, NJ) with glutathione Sepharose beads. The OPN N-or C-terminal fragments were separated from GST-bound beads by treating with 0.1 units of biotinylated thrombin/g of GST-OPN (Novagen, Madison, WI) for 2 h (either at room temperature for the 10C fragment or at 37°C for the 10N and 30N OPN fragments). The cleavage reaction was stopped with biotinylated-PPACK (400 ng/unit of biotinylated thrombin). Supernatants were collected, and biotinylated thrombin and PPACK were removed by incubation with streptavidin-agarose beads (Pierce) and separation of beads from supernatant.
The full-length recombinant osteopontin was generated as a histidine-tagged protein. An expression plasmid containing histidine-tagged osteopontin (His-OPN) was generated by cloning a polymerase chain reaction fragment containing the full-length splice variant of human osteopontin (OP10, Ref. 37) into the BamHI site of vector pQE30 (Qiagen, Chatsworth, CA). E. coli transformed with the His-OPN plasmid was grown in LB with 100 g/ml ampicillin and induced with isopropyl-1-thio-␤-D-galactopyranoside at 37°C to express the histidine-tagged protein. The His-OPN was purified from bacterial cells according to manufacturer's instructions (QIAexpressionist kit, Qiagen), chromatographed on Ni-nitrilotriacetic acid resin, and eluted with 0.2 M imidazole. The purified His-OPN was analyzed by SDS-PAGE.
Cell Adhesion Assay-Adhesion of BAEC and melanoma cells to ligand-coated microtiter plates was performed as described (9). Briefly, matrix proteins or osteopontin fragments were coated onto 96-well Maxisorp microtiter plates (Nunc Inc., Naperville, IL) overnight at 4°C and then blocked 10 mg/ml bovine serum albumin (BSA) in PBS for 1 h at 37°C. Cells were suspended in Dulbecco's modified Eagle's medium (melanoma cells) or Waymouth's medium (BAEC) containing 1 mg/ml BSA and preincubated with and without antibodies or peptides for 15 min at 37°C. Melanoma cells (100,000) or BAEC (30,000) were added to wells and allowed to incubate at 37°C for 1 h. Absorbance (595 nM) of toluidine blue-stained adherent cells was measured. Under these conditions, absorbance was proportional to cell number (9). Cell adhesion assays for SW480 cells were performed as above with slight modifications (39). FACS Analysis-Integrin expression was analyzed by fluorescent flow cytometry. After dispersion, 0.5 ϫ 10 6 cells were treated with the primary antibody or an irrelevant antibody for 30 min at 4°C in binding buffer (PBS containing 2 mg/ml BSA and 0.02% NaN 3 ). They were then washed with the same buffer and incubated with secondary antibody conjugated to phycoerythrin (Biomeda, Foster City, CA) for 30 min at 4°C, washed twice with PBS, and resuspended in PBS containing 2% paraformaldeyde for FACScan analysis (Becton Dickinson, Rutherford, NJ).
Cell Surface Biotinylation and Immunoprecipitation-Cell surface proteins were labeled with biotin essentially as described (40). A 5.0 ϫ 10 6 suspension of cells was incubated with 1 mg/ml sulfosuccinimidyl-6-(biotinamido)hexanoate (Pierce) for 30 min on ice. Cells were washed three times with PBS and incubated in lysate buffer (PBS with 1% Triton X-100, 200 M phenylmethylsulfonyl fluoride, 0.5 g/ml leupeptin and 2 g/ml aprotinin) at 4°C for 30 min. To immunoprecipitate surface-biotinylated proteins, the lysate supernatant (2.0 ϫ 10 6 cell equivalence) was added to PBS containing 1% Triton, 0.5 mg/ml BSA, and fresh protease inhibitors and precleared with 40 l of 50% (v/v) protein A-Sepharose CL-4B (Pharmacia). The supernatants were immunoprecipitated with the anti-integrin antibody or a mouse IgG as a negative control at 4°C. Immune complexes were recovered by binding to protein A-Sepharose and washing five times with IP wash buffer (50 mmol/liter Tris, pH 7.4, 0.5 mol/liter NaCl, 2 mmol/liter phenylmethylsulfonyl fluoride, 0.1% Triton X-100, and 0.1% Tween 20). After samples were separated by electrophoresis on 8% polyacrylamide-SDS gels under nonreducing conditions, the proteins were transferred to polyvinylidene difluoride membrane (DuPont NEN). The membrane was blocked with 10% nonfat dry milk in TBST buffer (10 mmol/liter Tris base, pH 8, 150 mmol/liter NaCl, and 0.05% Tween 20) at room temperature for 1 h. After washing, blots were incubated for an additional hour with streptavidin-biotinylated horseradish peroxidase complex (Amersham Corp.), and proteins were visualized by the addition of a chemiluminescence reagent according to the manufacturer's instructions.

Expression of Recombinant Osteopontin Fragments in E. coli
Cells-The osteopontin proteins that were used in this study include native osteopontin, recombinant human full-length osteopontin, and recombinant N-and C-terminal human osteopontin fragments that would be formed following thrombin cleavage (Fig. 1). Two N-terminal fragments were used, 10N and 30N, which refer to two different splice variants of osteopontin. The 30N splice variant contains an additional 14 amino acids (NAVSSEETNDFKQE), which correspond to exon 5.
Human osteopontin fragments that contain either the Nterminal domain or the C-terminal domain were amplified by polymerase chain reaction and cloned into the BamHI/EcoRI sites of the expression vector pGEX-2T. The resulting plasmids contained the N-terminal domain of osteopontin including amino acids 17-169 (10N and 30N) or amino acids 170 -317 (10C) fused in frame to the 3Ј-end of the GST gene. The 30N fragment is identical to the 10N fragment except that it includes the alternate splice exon 5. Glutathione S-transferaseosteopontin fragment fusion proteins synthesized during a 2-h induction with isopropyl-1-thio-␤-D-galactopyranoside, were purified from bacterial lysates by affinity chromatography on glutathione-agarose beads. The pGEX-2T vector includes a thrombin cleavage site between GST and the inserted protein. Therefore, to cleave osteopontin fragments from GST, the beads were treated with biotinylated thrombin. The biotinylated thrombin was then separated from the osteopontin fragments by affinity chromatography on streptavidin-agarose beads. The resulting proteins were analyzed on SDS-PAGE and are shown in Fig. 2. The C-terminal osteopontin domain (10C) contains amino acids 170 -317 and has an apparent molecular mass of 25 kDa. The two alternatively spliced N-terminal fragments (30N and 10N), contain amino acids 17-169. The 30N fragment includes exon 5, while the 10N fragment lacks this exon. The apparent molecular masses are 30 and 26 kDa, respectively. The C-terminal osteopontin domain (10C), contains a lower molecular weight protein, which is most likely the result of truncated GST-10C translational products.
Cell Adhesion to Native and Recombinant Fragments of Osteopontin-To compare the adhesive function of full-length osteopontin with fragments that would be formed following thrombin cleavage at the Arg 169 -Ser 170 peptide bond, we performed cell attachment assays with native osteopontin, recombinant human full-length osteopontin, and recombinant N-and C-terminal human osteopontin fragments. Adhesion assays were carried out with both bovine aortic endothelial cells and two different subpopulations of human melanoma cells. Bovine endothelial cells adhered to full-length native OPN and both splice variants of the N-terminal osteopontin fragment (30N and 10N). There was no adhesion to the C-terminal osteopontin domain (10C) or the GST, which was used as a control (Fig. 3).
We next performed adhesion assays with two subpopulations of human melanoma cell lines. The melanoma cell lines Mo␣ v and Mo were derived from M21 (22). Mo␣ v was previously shown to express high levels of the ␣ v ␤ 3 integrin. Mo lacks expression of the ␣ v subunit; therefore, these cells fail to express many of the known osteopontin receptors (␣ v ␤ 3 , ␣ v ␤ 1 , and ␣ v ␤ 5 ). As expected, Mo␣ v , but not Mo cells adhered to native osteopontin (Fig. 4A). Similar results were seen with human full-length recombinant osteopontin (not shown). Surprisingly, both cell lines adhered to the 30N and 10N osteopontin fragments (Fig. 4B). There was no adhesion to the C-terminal domain or the GST control. Because Mo cells interacted with the N-terminal domain of osteopontin and not the native protein or full-length recombinant protein, these data suggest that osteopontin adhesive interactions may be regulated by proteolytic fragmentation. Furthermore, the N-terminal region apparently contains a cryptic adhesive activity that is not exposed in the full-length molecule.
Mo␣ v Adhesion to Native Osteopontin and the N-terminal Osteopontin Fragment Is ␣ v ␤ 3 -and RGD-dependent-Interaction with osteopontin has been shown in many different cell types to be mediated through the ␣ v ␤ 3 integrin and to be RGD-dependent. Two additional osteopontin receptors, ␣ v ␤ 1 and ␣ v ␤ 5 , were first identified in human smooth muscle cells (12) and later in human embryonic kidney cells (14). To determine the receptors mediating the interaction of Mo␣ v with osteopontin N-terminal domain, we performed adhesion assays in the presence of neutralizing integrin antibodies. The interaction between Mo␣ v and the N-terminal domain was completely blocked by both the anti-␣ v ␤ 3 mAb (LM609) and an anti-␣ v mAb (L230) (Fig. 5A). This indicated that the osteopontin adhesive function was dependent on the ␣ v ␤ 3 integrin. Adhesion of Mo␣ v to native osteopontin was also ␣ v ␤ 3 -dependent. The interaction between Mo␣ v and the native osteopontin was completely blocked by anti-␣ v ␤ 3 mAb LM609 but not by anti-␣ IIb ␤ 3 , an irrelevant antibody control (Fig. 5B). The adherence of untreated cells and anti-␣ IIb ␤ 3 -treated cells were similar. As expected, the interaction of Mo␣ v with the osteopontin N-terminal domain was mediated through the RGD sequence. adhesion to the osteopontin N-terminal domain in a dose-dependent manner with an IC 50 of 12 M (Fig. 5C).
Mo Cell Adhesion to the N-terminal Osteopontin Fragment Is ␤ 1 -dependent and Poorly Blocked by RGD Peptides-Mo melanoma cells lack all ␣ v -containing integrins (22). Since the known osteopontin receptors are ␣ v ␤ 3 , ␣ v ␤ 1 , ␣ v ␤ 5 , and ␣ 4 ␤ 1 , the interaction of Mo with the N-terminal fragment might be through ␣ 4 ␤ 1 or a novel osteopontin receptor. Binding of Mo to the N-terminal fragment was cation-dependent (Fig. 6A), suggesting that this receptor was an integrin. To identify the integrin, Mo adhesion to the osteopontin N-terminal domain was carried out in the presence of integrin-neutralizing antibodies. As shown in Fig. 6B, the interaction between Mo and the N-terminal osteopontin domain was entirely blocked by P4C10, a neutralizing ␤ 1 mAb. A nonblocking ␤ 1 antibody, LM534, failed to inhibit adhesion. The ␣ subunit mediating this interaction was not ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , or ␣ 5 , because blocking antibodies to these subunits had no effect on Mo adhesion (Fig. 6C).
The ␤ 1 interaction with the N-terminal osteopontin fragment was only partially dependent on RGD (Fig. 6D). The GRGDSP peptide inhibited adhesion with an IC 50 of 250 m, which was about 20 times higher than that observed for GRGDSP inhibition of Mo␣ v binding to the N-terminal fragment of osteopontin. These data suggest that the N-terminal osteopontin peptide may contain an additional adhesive domain, distinct from RGD, which is recognized by the ␤ 1 integrin on Mo cells.
Identification of Surface-expressed ␤ 1 Integrins from Mo Melanoma Cells-The antibody blocking studies suggested that the ␣ subunit responsible for the ␤ 1 -mediated adhesion of Mo to the N-terminal fragment was not ␣ 1 , ␣ 2 , ␣ 3 , ␣ 4 , or ␣ 5 . There are four additional ␣ subunits known to form heterodimers with the ␤ 1 integrin subunits: ␣ 6 , ␣ 7 , ␣ 8 , and ␣ 9 . It is unlikely that ␣ 6 mediates adhesion, because Mo cells express extremely low levels of this integrin by FACS analysis (Table I). To determine if ␣ 8 ␤ 1 was involved, we measured the adhesion of ␣ 8 -containing 293 cell transfectants to the N-terminal osteopontin fragment in the presence of neutralizing antibodies. The adhesion was completely blocked by an anti-␣ v antibody, suggesting that ␣ 8 ␤ 1 does not mediate adhesion by itself (not shown). To determine if the ␣ 9 subunit was associated with surface-expressed ␤ 1 integrin, we immunoprecipitated surface-biotinylated cells with P4C10, a ␤ 1 mAb. Immunoprecipitation revealed multiple ␣ chains with apparent molecular mass between 100 and 180 kDa associated with the ␤ 1 integrin (Fig. 7A). Two bands with molecular masses of approximately 140 and 115 kDa were particularly abundant. These sizes correspond to the molecular weight of the ␣ 9 and ␤ 1 subunits, respectively. To further identify these chains, an ␣ 9 antibody was used to immunoprecipitate surface-biotinylated proteins. Fig. 7B shows that the ␣ 9 antibody immunoprecipitated abundant levels of two proteins, corresponding to ␣ 9 and ␤ 1 subunits, respectively (34). ␣ 9 ␤ 1 Mediates Adhesion of ␣ 9 -transfected SW480 and Mo Cells to the N-terminal Domain of Osteopontin-Because a significant level of ␣ 9 ␤ 1 was expressed on Mo cells, we next determined if ␣ 9 ␤ 1 can mediate cell adhesion to the N-terminal fragment by using a cell line that was stably transfected with ␣ 9 . SW480 cells, human colon carcinoma cells that normally do not express ␣ 9 ␤ 1 integrin, were stably transfected with a plasmid encoding for the ␣ 9 integrin subunit, pcDNAIneo␣ 9 . These cells have been shown to adhere to a tenascin fragment containing the third fibronectin type III repeat, while mock transfectants do not adhere to this fragment (41). The ␣ 9 -transfectants also adhered to the N-terminal osteopontin fragment (30N), and the ␣ 9 antibody blocked this adhesion by about 60% (Fig. 8A). This suggests that ␣ 9 ␤ 1 mediates adhesion to the N-terminal osteopontin fragment but that other receptors are also involved.
The ␣ 9 blocking antibody was also used in adhesion assays with Mo melanoma cells. The antibody inhibited Mo adhesion to the N-terminal osteopontin fragment by 84%. The effects of this antibody were specific, because it did not interfere with non-␣ 9 -mediated adhesion to laminin (LN), collagen type I (COL I), or fibronectin (FN) (Fig. 8B). In addition, this antibody did not affect the ␣ v ␤ 3 -mediated adhesion of Mo␣ v to the Nterminal osteopontin fragment (not shown).

Mo and Mo␣ v Express Similar Levels of Integrin Receptor Subunits Except
␣ v ␤ 3 -There are several possible explanations for why Mo and not Mo␣ v melanoma cells interact with the N-terminal osteopontin fragment through the ␣ 9 ␤ 1 integrin. The most obvious explanation is that surface expression of ␣ 9 ␤ 1 is greater on Mo cells compared with Mo␣ v . FACS analysis demonstrated that this was not the case. Both melanoma cell lines contained similar levels of the ␣ 9 integrin on their surfaces (Table I). In addition, all the integrins tested had similar expression patterns in both cell lines except for the ␣ v -containing integrins. Each line expressed relatively similar levels of ␣ 2 , ␣ 3 , ␣ 9 , and ␤ 1 integrin subunits and low or undetectable levels of ␣ 1 , ␣ 4 , ␣ 5 , and ␣ 6 . The most significant difference between these two cells was the ␣ v , ␤ 3 , and ␣ v ␤ 3 expression. As expected, ␣ v , ␤ 3 , and ␣ v ␤ 3 were expressed at high levels on Mo␣ v melanoma cells and were undetectable on Mo cells. Thus, the difference in ability of Mo and Mo␣ v to adhere to N-terminal osteopontin fragment is not accounted for by ␣ 9 receptor density.

DISCUSSION
This study examined the potential role of proteolytic fragmentation on osteopontin function. We compared the N-and C-terminal GST-recombinant osteopontin fragments expected to be formed following thrombin cleavage at the Arg 169 -Ser 170 site with native osteopontin in their ability to mediate adhesion of several cell lines. The results demonstrated that: 1) osteopontin adhesive interactions may be regulated by proteolytic fragmentation such as seen with thrombin, 2) two different splice variants of the N-terminal osteopontin fragments have identical adhesive properties, and 3) N-terminal osteopontin fragment contains two distinct integrin-binding activities. One is the RGD-dependent ␣ v ␤ 3 -binding activity. The second is a cryptic binding activity for the ␣ 9 ␤ 1 integrin. Thus, proteolytic fragmentation may be a way of controlling or altering osteopontin's receptor specificity and thus its function.
Using N-and C-terminal recombinant fragments, we have shown that two splice variants of the N-terminal, but not the C-terminal, domain can support adhesion of bovine aortic endothelial cells and two subpopulations of human melanoma cell lines. The two subpopulations of melanoma cells differ in ␣ v integrin expression. Mo␣ v contains high levels of the ␣ v ␤ 3 in-  7. Immunoprecipitation of surface-labeled proteins with ␤ 1 and ␣ 9 antibody. Mo cells were surface-biotinylated, and the cell lysate was immunoprecipitated with ␤ 1 mAb (P4C10) (A) or ␣ 9 antibody (1057) (B). Proteins were separated by SDS-PAGE and transferred to polyvinylidene difluoride membrane. The membrane was then incubated with streptavidin-biotinylated horseradish peroxidase, which was visualized using chemiluminescence. The two bands at 140 and 115 kDa correspond to the molecular mass of ␣ 9 and ␤ 1 , respectively. Mouse IgG was used as a control antibody (CTL).

TABLE I Flow cytometry analysis of integrin subunit expression on Mo and
Mo␣ v melanoma cells Mo and Mo␣ v cells were stained with mAbs specific to ␣ v (mAb 1980), ␣ 1 (5E8D9), ␣ 2 (P1E6), ␣ 3 (P1B5), ␣ 4 (P4G9), ␣ 5 (P1D6), ␣ 6 (J1B5), ␣ 9 (Y9A2), ␤ 1 (P4C10), ␤ 3 (SZ.21), ␤ 6 (E7P6), ␣ v ␤ 3 (LM609), ␣ v ␤ 5 (P1F6), or mouse IgG followed by phycocrythrin-labeled anti-mouse IgG and analyzed by FACS scan. Peak fluorescence intensity ϭ linear fluorescence intensity (test antibody)/linear fluorescence intensity (control antibody). tegrin. Mo lacks expression of the ␣ v subunit. Because these cells lack ␣ v , they fail to express any of the known osteopontin receptors and do not adhere to native osteopontin. It was therefore surprising to find that the Mo cells could attach to the N-terminal osteopontin fragment. Mo cells also failed to adhere to the human recombinant full-length osteopontin, indicating that the difference in adhesion of Mo cells to native osteopontin and N-terminal recombinant fragment was not simply due to glycosylation or phosphorylation. These results suggested that the adhesion of Mo cells was through a non-␣ v osteopontin receptor and that a cryptic adhesive activity was exposed in the N-terminal osteopontin peptide fragment. Further analysis demonstrated that the two subpopulations of melanoma cells adhered to the N-terminal fragment with different receptors and bound distinct adhesive domains. Mo␣ v cells adhered to the N-terminal fragment through the ␣ v ␤ 3 integrin. This interaction was RGD-dependent. Mo cells, which lack ␣ v and fail to bind the native protein, adhered to the N-terminal fragment through the ␣ 9 ␤ 1 integrin. In addition, human urinary osteopontin that was cleaved with thrombin in situ also supported ␣ 9 ␤ 1 -mediated adhesion of Mo cells. 2 The interaction of ␣ 9 ␤ 1 with the N-terminal fragment was less effectively blocked by RGD peptides, suggesting there may be an additional adhesive domain distinct from RGD. A non-RGD adhesion function for osteopontin has previously been reported. A fragment from endoproteinase Arg-C-digested rat osteopontin, which lacked the RGD domain, supported adhesion of human fibroblasts (42). However, in this study, the activity was found in the C-terminal half of the molecule; therefore, this potential adhesive domain must be distinct from the ␣ 9 ␤ 1 site found in the N-terminal half of osteopontin.
The only other known ligand for ␣ 9 ␤ 1 is tenascin (41). Like osteopontin, the ␣ 9 ␤ 1 binding domain in tenascin appears to be distinct from the RGD adhesion motif (41). However, Mo adhesion to the osteopontin fragment is at least partially inhibited by RGD, suggesting that either ␣ 9 ␤ 1 recognizes osteopontin and tenascin by somewhat different mechanisms or that other RGD-dependent receceptors are involved. In normal adult tissue, osteopontin and ␣ 9 are expressed on most epithelia and could potentially colocalize (34,(43)(44)(45). In diseased tissue, osteopontin is highly up-regulated. Abundant osteopontin is found at the interface between malignant and normal tissue and at sites of inflammation and tissue remodeling (5,46). These are also sites where thrombin and thrombin-cleaved fragments of osteopontin are likely to be found. It would be interesting if ␣ 9 is coordinately up-regulated at these sites. If so, the ability of osteopontin to promote adhesion, migration, or other cellular functions may be regulated in the presence of thrombin by exposing the cryptic domain.
Cryptic integrin-mediated binding activities have also been identified in other adhesive proteins. For example, laminin contains a cryptic peptide site that becomes functional after proteolysis and supports ␣ v ␤ 3 -mediated adhesion of rat osteoclasts (47). Collagen also contains a cyptic site that is exposed following denaturation. Native type I collagen in its helical conformation supports ␣ 1 ␤ 1 -, ␣ 2 ␤ 1 -, and ␣ 3 ␤ 1 -meditated adhesion. These interactions are disrupted by heating or proteolysis of collagen, revealing a cryptic ␣ v ␤ 3 binding activity (48). The exposure of novel binding activities following proteolytic fragmentation is particularly relevant in remodeling tissues where proteases are active.
Both Mo and Mo␣ v express equal amounts of surface ␣ 9 integrin, but only the Mo cells use this receptor for N-terminal osteopontin interactions. There are several possible explanations for this phenomenon. First, the activation state of ␣ 9 ␤ 1 on the two cell lines may differ. It is known that ␤1 integrins can exist in different activation states, which can affect the affinity of ligand binding (49 -52). The second possibility is that ␣ 9 ␤ 1 function could be regulated by the ligation of another integrin. For example, the interaction of ␣ v ␤ 3 with ligand could transmit a signal that inhibits the affinity modulation and/or function of ␣ 9 ␤ 1 . This type of "cross-talk" between integrins has previously been reported in several cell types (53)(54)(55).
Two previous reports have examined the functional consequences of osteopontin cleavage with thrombin. One report demonstrated that thrombin cleavage destroyed RGD-mediated adhesion (19). In contrast, a second report showed that thrombin treatment enhanced osteopontin-mediated adhesion (20). In agreement with the second report, our studies demonstrate that the N-terminal fragment that is expected to be formed following thrombin cleavage can support adhesion. It is clear from these studies as well as others that the interaction of cells with osteopontin and osteopontin fragments is mediated through distinct receptors in different cell types. This could explain the discrepancy between different studies. Another possible explanation is that the conditions used to cleave osteopontin with thrombin may result in partial or complete proteolysis at an additional thrombin cleavage site. A second potential cleavage site is at residues Arg 160 -Gly 161 , which is within the RGD domain. If the RGD were destroyed by proteolytic cleavage, it is likely that ␣ v ␤ 3 -mediated interactions would not take place.
There are several different splice variants of osteopontin. Two of the variants are due to variable usage of exon 5, which contains 14 amino acids. The functional significance of alternative splicing in this gene is not known. In this study we compared the adhesive function of recombinant osteopontin N-terminal fragments of each splice variant. Our results dem- FIG. 8. A, adhesion of ␣ 9 -transfected SW480 cells to N-terminal osteopontin fragment 30N. SW480 cells, stably transfected with a plasmid encoding for ␣ 9 integrin subunit pcDNAIneo␣ 9 , were plated in triplicate wells (50,000 cells/well) coated with 40 nM recombinant osteopontin fragment (30N) in the absence of antibody (no Ab) or in the presence of the ␣ 9 ␤ 1 blocking antibody (Y9A2). Cells were allowed to attach for 1 h at 37°C. Attached cells were quantified as absorbance at 595 nm of crystal violet-stained wells. Each bar represents the mean Ϯ S.E. of triplicate wells. B, adhesion of the ␣ v -null melanoma cells (Mo), to extracellular matrix molecules in the presence and absence of ␣ 9 and ␤ 1 neutralizing antibodies. Mo cells were preincubated with and without the neutralizing ␣ 9 (Y9A2) and ␤ 1 (P4C10) antibodies for 15 min at 37°C. Cells were then plated in wells coated with laminin (LN) (10 g/ml), fibronectin (FN) (10 g/ml), collagen I (COL I) (10 g/ml), 30N osteopontin fragment (40 nM), or 10N osteopontin fragment (40 nM) and allowed to attach for 1 h. The attached cells were quantitated as described in the legend to Fig. 3. Each data point represents the percentage of adhesion compared with cells not treated with antibody. onstrate that both splice variants had identical adhesive functions. Exon 5 contains the sequence NAVSSEETNDFKQE. The two serine residues are potential sites for phosphorylation or O-linked glycosylation; therefore, our data do not rule out the possibility that there is a functional difference following posttranslational modification.
In conclusion, we have demonstrated that the N-terminal fragment of osteopontin contains two distinct activities. We predict from these results that osteopontin fragmentation by proteases, such as thrombin, are important in the regulation of receptor specificity and, thus, the function of osteopontin.