Tumor-derived Osteopontin Is Soluble, Not Matrix Associated*

The secreted phosphoprotein osteopontin (OPN), when immobilized on a surface, supports cell adhesion, prevents apoptosis of endothelial cells, and is a ligand for the (cid:1) v (cid:2) 3 integrin, which is important in endothelial cell biology and neovascularization. OPN synthesized by tumor cells stimulates tumor growth, but the mecha-nism by which the protein acts remains unclear. One possibility, therefore, is that OPN may exert its effects on tumor growth by enhancing angiogenesis. While OPN is found at high levels in bone, where it is a component of the mineralized matrix, we have asked here whether OPN present in tumors is similarly extracellular matrix associated. We have shown that OPN is detectable in tumor extracts and in serum of tumor-bearing mice, and that the protein in tumors and in serum can be synthesized by both tumor and the host cells. Biochemical fractionation of tumor tissue confirmed that there is little if any association of OPN with the insoluble fraction. Im-munochemical analysis of murine mammary tumors shows no co-localization of OPN with the extracellular matrix, identified by laminin staining. Ras -transformed cells in culture produce abundant OPN, however, the protein was found to be associated with the cell fraction but not with the matrix fraction. An enzyme-linked immunosorbent assay was used to demonstrate that OPN in conditioned medium from these cells fails to associate with extracellular matrix components, including laminin and fibronectin, in vitro . Recombinant OPN (GST-OPN) when coated

The secreted phosphoprotein osteopontin (OPN), when immobilized on a surface, supports cell adhesion, prevents apoptosis of endothelial cells, and is a ligand for the ␣ v ␤ 3 integrin, which is important in endothelial cell biology and neovascularization. OPN synthesized by tumor cells stimulates tumor growth, but the mechanism by which the protein acts remains unclear. One possibility, therefore, is that OPN may exert its effects on tumor growth by enhancing angiogenesis. While OPN is found at high levels in bone, where it is a component of the mineralized matrix, we have asked here whether OPN present in tumors is similarly extracellular matrix associated. We have shown that OPN is detectable in tumor extracts and in serum of tumor-bearing mice, and that the protein in tumors and in serum can be synthesized by both tumor and the host cells. Biochemical fractionation of tumor tissue confirmed that there is little if any association of OPN with the insoluble fraction. Immunochemical analysis of murine mammary tumors shows no co-localization of OPN with the extracellular matrix, identified by laminin staining. Ras-transformed cells in culture produce abundant OPN, however, the protein was found to be associated with the cell fraction but not with the matrix fraction. An enzyme-linked immunosorbent assay was used to demonstrate that OPN in conditioned medium from these cells fails to associate with extracellular matrix components, including laminin and fibronectin, in vitro. Recombinant OPN (GST-OPN) when coated onto a plastic surface can support human umbilical vein endothelial cell adhesion, suppressing apoptosis and allowing cell cycle progression, at concentrations from 1 to 50 g/ml. Soluble GST-OPN in the same concentration range has no effect on HU-VECs held in suspension. Thus, we conclude that OPN associated with tumors is primarily soluble, and that soluble OPN can neither support endothelial cell proliferation nor prevent apoptosis of these cells in the absence of adhesion.
The secreted phosphoprotein osteopontin is synthesized by both osteoblasts and osteoclasts, and accumulates in bone (1), where it is required for optimal osteoclast function (2). OPN 1 is also expressed in a variety of other tissues, notably those with a major epithelial component, where it is excreted into a variety of fluids: the function of the protein in these tissues is less well defined. The protein is expressed at high levels in a variety of tumors and transformed cells (3)(4)(5)(6), and expression can be either in the tumor cells themselves (3,5,7) or in tumorassociated macrophages (6). In a syngeneic mouse model of tumorigenesis, OPN production by tumor cells themselves stimulates tumor growth (8). The function of OPN in tumorigenesis is still not clear. In particular, the cell type(s) with which OPN interacts to stimulate tumor growth is not known.
One well described function of OPN in vitro is as a cell adhesion molecule when it is coated onto a plastic surface (9,10). The protein contains an RGD sequence that can mediate cell adhesion (11), although other non-RGD sequences in the molecule can also mediate adhesion: these include the SV-VYGLR sequence in human OPN (12,13), as well as other sequences that may have adhesion functionality (14 -17). Adhesion of cells to OPN is in most cases through a variety of integrins (reviewed in Ref. 18, see also Refs. 13 and 19); the non-integrin receptor CD44 may also mediate cell adhesion to OPN (15,20). Most notably, OPN is a well characterized ligand for the ␣ v ␤ 3 integrin, which is important for angiogenesis and endothelial cell survival (21,22). Thus, one possibility for the function of OPN in tumorigenesis is to promote survival of endothelial cells: indeed the protein has been demonstrated to have this function in vitro, as an insoluble adhesive ligand (23).
Like endothelial cells, osteoclasts express high levels of the ␣ v ␤ 3 integrin: this integrin is critical for osteoclast function (24). OPN is present at high levels in bone (25,26), and is an important ligand for the osteoclast ␣ v ␤ 3 integrin as evidenced by reduced bone resorption in osteopontin-deficient mice (2). Thus, in bone, OPN is a well established component of the extracellular matrix, and can function in vivo in this tissue as a cell attachment molecule. However, osteopontin has a high affinity for hydroxyapatite by virtue of its general acidic nature, its high degree of phosphorylation, as well as due to a run of 9 -10 aspartates in the NH 2 -terminal half of the molecule (27). Thus, its accumulation in bone may be due to its affinity for the mineral, rather than the protein component of bone tissue.
It is still not clear, therefore, if OPN functions as a cell attachment molecule in soft tissues. While numerous cell types have been demonstrated to adhere to OPN coated onto a plastic surface, including endothelial cells, fibroblasts, transformed cells, and smooth muscle cells (10, 11, 23, 28 -30), this activity is only physiologically relevant if these cell types encounter immobilized OPN in vivo. An important question to be addressed therefore is whether OPN is associated with the extracellular matrix in non-calcified matrices. Of special interest is the situation in tumors: if OPN, which is expressed at quite high levels in many tumor types, is functioning as a cell adhesion molecule it should be immobilized in the matrix. In the work presented here, we have used a variety of techniques to localize OPN in tumor tissues and cells, and conclude that OPN made by tumors is primarily, if not exclusively, soluble.

MATERIALS AND METHODS
Reagents-Laminin (Sigma), fibronectin (Invitrogen), collagen I and matrigel (Collaborative Research/BD) were obtained from commercial sources, while recombinant GST-OPN was isolated as described (31). mOPN was isolated from ras-transformed 3T3 cell conditioned medium by using DEAE-Sepharose chromatography. Antibodies to laminin and fibronectin were obtained from Sigma. Antibodies to OPN were: goat anti-rat OPN antibody OP199 (28) or rabbit anti-human OPN antibody LF 123 (32): both these antibodies were used at a dilution of 1:1000; they both react with mouse OPN and were used interchangeably. LF123 has a slightly higher affinity for mouse OPN, but cross-reacts nonspecifically with a few other bands in tumor and cell lysates: the specificity of these antibodies has been described in detail elsewhere (33). For immunohistochemistry, pooled, affinity purified mouse antimouse OPN 2 was used. For immunofluorescence, monoclonal antimouse 2A1 2 was used; this antibody reacts with an epitope in the C-terminal half of OPN. These antibodies were raised in the OPN Ϫ/Ϫ mice.
Tumors and Tumor Extracts-The tumors used here were derived from two different experimental systems. Mammary tumors arising spontaneously in MMTV-c-myc/MMTV-v-Ha-ras transgenic mice, either wild type or OPN-deficient (34), were used for analysis of extracellular matrix. For the analysis of OPN accumulation in sera and tumors in hosts of different genotypes, tumors were induced by injection of ras-transformed 3T3 cells as described (8) into either wild type or OPN-deficient mice. In all cases, mice were sacrificed by exsangination, and tumors were removed and immediately flash frozen on liquid nitrogen. Powdered frozen tissue was stored at Ϫ70°C until use. Tumor extracts were made by resuspending the powdered tissue in RIPA buffer containing 0.1 mM phenylmethylsulfonyl fluoride, 1 mg/ml leupeptin, and 1 mg/ml aprotinin, and subjecting the extract to 3 cycles of freeze/thaw. For preparation of extracellular matrix from mammary tumors, 50 mg of powdered tissue were resuspended in 50 mM Tris, pH 7.5, with 0.2% Triton X-100 containing protease inhibitors as above (Tris/Triton), and sonicated. The suspension was centrifuged (15,000 ϫ g; 5 min) and the supernatant retained. The pellet was resuspended in Tris/Triton, and subjected to two more cycles of sonication and centrifugation. The final pellet was resuspended in 50 mM Tris, pH 7.5, 6 M urea. OPN in serum was collected by addition of 1/10 volume each of 15% barium chloride and 3.8% sodium citrate on ice. The resulting precipitate was collected, and washed at 0°C sequentially with 0.5 volume of 15% barium chloride and H 2 O. Proteins were eluted by boiling the washed pellets in 20 l of SDS loading buffer containing 0.2 M sodium citrate. Western blotting was performed as described (33).
Cultured Cell Extracts-ras-transformed wild type and OPNdeficient 3T3 cell lines (8) were plated in Dulbecco's modified Eagle's medium containing 3% fetal bovine serum on 35-mm wells previously coated with 0.3 mg/ml collagen I in 20 mM acetic acid or with 0.7 mg/ml matrigel (Collaborative Research). The cells were allowed to become confluent and the extracellular matrix and soluble fractions obtained using a modification of the procedure of Gospodarowicz and co-workers (35). Briefly, the cells were washed and lysed in 10 mM Tris, pH 7.5, 0.5% Triton X-100, 10 mM phenylmethylsulfonyl fluoride, and the supernatant retained. The wells were then washed with 20 mM NH 4 OH in 10 mM Tris, pH 7.5. Material remaining in the well was scraped into 10 mM Tris, pH 7.5, containing 2% SDS. Control wells (whole cell extracts) were solubilized directly in 2% SDS. Alternatively, the cells were removed with 3 mM EDTA in PBS, after which the matrix was washed and solubilized as above. These methods are commonly used for the preparation of extracellular matrix (36,37). An alternative method, removal of cells by incubation with cytochalasin B, was not effective for these transformed cells (data not shown). For OPN and fibronectin, 1/5 of the cell or ECM extracts were separated by PAGE; for laminin, 1/60 of the extracts were used. Equal proportions rather than equal amounts of protein were used because the cell lysate contains proportionately much more protein than the ECM.
ELISA-Wells of 96-well plates were coated with matrix components at 200 g/well in PBS overnight at 4°C. These wells were blocked the next day with 2% BSA. Conditioned medium from either wild type or OPN-deficient ras-transformed cells was then added to the coated well and incubated at room temperature for 2 h. The same conditioned medium, and dilutions as indicated were applied directly to uncoated wells. All wells were then washed with water, and the matrix containing wells, and some of the wells containing directly coated conditioned medium were fixed with 4% paraformaldehyde for 30 min at room temperature. Following additional washes, OPN was detected with affinity purified polyclonal antimouse OPN 2 at 1:3000 dilution, followed by horseradish peroxidase-labeled goat anti-mouse IgG (Bio-Rad) at a 1:1000 dilution. After washing the horseradish peroxidase substrate, ABTS (0.3 mg/ml in 55 mM citrate, pH 4.0, with .03% H 2 O 2 ) was added and incubated for 10 -30 min in the dark.
Immunohistochemistry-Serial sections from myc/ras mammary tumors (38) (fixed in methacarn and embedded in paraffin) were rehydrated and endogenous peroxidases blocked in 3% H 2 O 2 in methanol. Tumors arising in both wild type and OPN Ϫ/Ϫ mice were used (34). Sections were blocked with goat serum (1:20 in PBS) for 30 min. Primary antibodies were diluted in 1% BSA, 1% goat serum, and incubated with tissue sections for 1 h or overnight. Affinity purified polyclonal anti-mouse OPN (1:1000) was biotinylated with the DAKO ARK system before application to the tissue. Rabbit anti-laminin was from Sigma, was used at 1:300 and detected with biotinylated goat anti-rabbit IgG. All sections were reacted with the ABC reagent (Vector ABC Elite) prior to visualization of the antibody reactivity by staining with diaminobenzidine (Sigma Fast DAB). Sections were lightly counterstained with hematoxylin and mounted in permount. Control sections were incubated without primary antibody.
Confocal Microscopy-Paraformaldehyde sections from myc/ras mammary tumors (38) were rehydrated and subjected to antigen retrieval (0.01 M citrate, pH 6.0, 100°C, 10 min). Sections were blocked with goat serum (1:20 in PBS) for 1-2 h. Primary antibodies were diluted in 1% BSA, 1% goat serum, and incubated with tissue sections overnight. Affinity purified monoclonal antibody 2A1 at 0.18 g/ml was biotinylated with the DAKO ARK system before application to the tissue, and was detected with rhodamine-avidin D (Vector Labs, 16 g/ml). Anti-laminin was detected with fluorescein isothiocyanate-labeled goat anti-rabbit IgG (Jackson). Rhodamine and fluorescein isothiocyanate signals were collected and a single optical section is shown. Controls with no primary antibody showed fluorescence that was indistinguishable from untreated tissue (data not shown).
Endothelial Cell Growth-Human umbilical vein endothelial cells (HUVECs; Clonetics from pooled donors) were cultured in M199 with 10% fetal bovine serum (Hyclone), 4 ng/ml aFGF, 75 g/ml endothelial cell growth supplement (Sigma), and 5 units/ml heparin, and used before PDL 15. Prior to experiments, confluent cells were incubated overnight in M199 with 1 ϫ ITS supplement (Invitrogen), in the absence of serum and growth factors. Quiescent cells were detached with trypsin, washed with M199, and plated at 5 ϫ 10 4 cells/well in M199 containing 1 ϫ ITS and 2% BSA in 48-well plates coated with different substrates. Wells were coated as described below. For apoptosis, cells were incubated overnight in the absence of growth factors, and then labeled for 15 min with 8 g/ml Hoescht 33342. Floating and adherent cells were collected, fixed, and counted under fluorescence illumination. For [ 3 H]thymidine incorporation, cells were incubated overnight with growth factors (10 ng/ml FGF plus 5 units/ml heparin), then 2 Ci/ml [ 3 H]thymidine were added and the cells were incubated for an additional 5-6 h. Cells were washed, fixed in 7% trichloroacetic acid, then solubilized in 0.5 N NaOH, 0.5% SDS and counted. Control experiments showed that this protocol recovered all the trichloroacetic acid-precipitable counts in the wells.
For adhesion assays, exponentially growing cells in complete medium were harvested with trypsin, resuspended in complete medium, and washed in Dulbecco's modified Eagle's medium. Cells were resuspended in 2% BSA, and plated in 96-well plates (Immulon II) at 2 ϫ 10 4 cells/well. The wells had been previously coated with different substrates as indicated. After 2 h, the wells were washed and the cells stained with crystal violet. Following elution of the dye with acetic acid, absorbance of the wells at 630 nm was determined. Substrates were diluted in PBS and coated on the wells overnight at 4°C; the wells were subsequently blocked with 2% BSA for 30 min at 37°C. Substrates used were mOPN (OPN purified from ras-transformed 3T3 cells), GST-OPN (recombinant mouse OPN as a GST fusion protein), fibronectin (Invitrogen), laminin (Sigma), and recombinant GST protein. Unless otherwise indicated, proteins were used at 10 g/ml.

RESULTS
ras-transformed 3T3 lines were generated from both wild type and OPN-deficient mice: these cell lines have been previously described (8), and shown to give rise to tumors following subcutaneous injection into syngeneic wild type hosts. These tumors form much more slowly when the injected cells are deficient for OPN production, illustrating the important role that OPN plays in the process of tumorigenesis (8). In other experiments (data not shown), these same cell lines were injected into control (wild type) and OPN-deficient host mice, and tumors allowed to develop. When the tumors reached a volume of 2000 -3000 mm 3 , serum was prepared from the tumor-bearing animals and the tumors were excised. OPN levels in these sera and in tumor extracts were analyzed by Western blotting as shown in Fig. 1.
When wild type, OPN-expressing tumors developed in wild type or in OPN-deficient hosts (Fig. 1A, lanes 2-4 and 9 -11), OPN protein accumulated to significant levels in the serum. This is consistent with the high level of expression of OPN in the wild type cell lines. Interestingly, in wild type animals bearing OPN-deficient tumors, serum levels of OPN were again elevated over control (non-tumor bearing, wild type mice) levels ( Fig. 1A, lanes 5-7). OPN was not detected in the serum of OPN-deficient mice bearing OPN-deficient tumors (data not shown).
Extracts of OPN-expressing tumors similarly contained OPN, as shown previously for OPN-expressing tumors devel-oped in wild type hosts (3,34), and the amount of protein was similar whether or not the host animal expressed OPN (Fig.  1B, lanes 1-2 and 5-6). OPN-deficient tumors developed in wild type mice expressed easily detectable OPN, and in some cases the level of OPN in these tumors was as high as that in wild type tumors (Fig. 1B, lanes 3 and 4). Again, no OPN was detected when extracts from OPN-deficient tumors developed in OPN-deficient hosts were analyzed (Fig. 1B, lane 7). We conclude from these results that much of the OPN made by tumors is secreted into the circulation, and that sufficient OPN is produced by the host cells in response to tumor growth to result in substantial levels of the protein in the tumor itself (as detected by Western blotting), and also in elevated serum levels.
Biochemical fractionation of tumor extracts was performed to determine whether the OPN present in tumors was associated with the ECM. Tumors arising spontaneously in transgenic mice expressing myc and v-Ha-ras specifically in the mammary gland (34,38) were flash-frozen and resuspended in Triton X-100 containing buffer, sonicated to disrupt the cell structure, and insoluble material collected by centrifugation. Following three cycles of resuspension and sonication in the same buffer, equal amounts of protein from the first supernatant and the final pellet were analyzed by SDS-PAGE followed by Western blotting (Fig. 2). The extracellular matrix proteins fibronectin and laminin were retained in the pellet fraction, confirming that these pellets indeed contained the extracellular matrix material. OPN, on the other hand, was localized exclusively in the soluble fraction: no OPN immunoreactivity could be detected in the pellet, at least in the amount of protein loaded in the well of the PAGE gel (30 g). Cross-reacting proteins identified in both the soluble and pellet fractions are present in both WT and OPN Ϫ/Ϫ tumors, indicating that they are not OPN. Laminin and fibronectin reactivity appear also in the soluble fraction: this reactivity may be due to soluble protein present in the tumor, or to partial solubilization of the ECM by the Triton/sonication protocol used. Even if the latter is the case, there is a clear differentiation between the pattern of association of OPN and the former proteins with the insoluble pellet.
The localization of osteopontin in tumor tissue was further examined in these spontaneously arising tumors by immunohistochemistry. For these experiments we chose to use the spontaneously arising tumors rather than those resulting from subcutaneous injection of transformed cells, reasoning that the spontaneous tumors would have a more extensive and well developed extracellular matrix. Accumulation of OPN was variable in these tumors, with areas of high expression and regions where OPN was undetectable (Fig. 3: panel B, OPN ϩ/ϩ). Some tumor samples expressed very low levels of OPN (data not shown). In regions where OPN expression was high, however, immunoreactivity appeared over tumor cell cytoplasm. The extracellular matrix in these tumors was identified by staining of adjacent sections for laminin, which was associated particularly with connective tissues and with the basement membrane of blood vessels (Fig. 3: panels A and D, LAM), with some reactivity appearing associated with the tumor cells themselves. Double immunofluorescence for OPN and laminin, ex-amined at higher magnification in a single optical section by confocal microscopy (Fig. 3, panel C), indicated that this laminin reactivity was localized over the cytoplasm of the cells. Thus OPN and laminin co-localize in the cytoplasm of the tumor cells, as these cells are presumably synthesizing both proteins. Laminin was not detected surrounding individual tumor cells: thus, we could not identify the ECM, if any, associated with the tumor cells themselves. However, laminin was clearly localized underlying the tumor vasculature and in regions of connective tissue, and OPN was conspicuously absent from these regions. The arrow in the upper left of Fig. 3C, indicates a cell staining strongly positive for OPN. This cell is directly adjacent to the basement membrane of a large blood vessel, and OPN is clearly not present in this basement membrane. Thus the OPN secreted from this cell does not associate with the nearby basement membrane. OPN expression was not detected in macrophages, identified by F4-80 staining (data not shown). This is in accordance with previous results from in situ hybridization, indicating that OPN in these tumors is produced primarily by the tumor cells themselves (3), and contrasts with results obtained with the subcutaneous tumors (Fig. 1), highlighting the variability of the host anti-tumor response.
ras-transformed 3T3 cells in culture generate considerable amounts of OPN in the conditioned medium: we have estimated that the level of OPN in medium conditioned overnight by such cells at confluence can reach 10 g/ml (data not shown and Fig. 4). These transformed cells grow rapidly and are not growth arrested at confluence, so they do not generate substantial extracellular matrix of their own. To determine whether the OPN made by these cells in culture associates with the ECM, wild type and OPN Ϫ/Ϫ transformed cells were plated onto matrigel and collagen I as surrogate extracellular matri- In the upper panels, a tumor from a wild type (ϩ/ϩ) mouse was used. Controls (lower panels) included tumors from OPN Ϫ/Ϫ mice (Ϫ/Ϫ) reacted with laminin and OPN antisera, and an adjacent section from the wild type mouse mice incubated as above but in the absence of primary antibody (ϩ/ϩcon). Detection was as described under "Materials and Methods," and immunoreactivity was visualized by DAB (brown) staining. The sections were lightly counterstained with hematoxylin. The arrows in the first two panels indicate capillary ECM reactive with laminin that does not contain OPN. Immunofluorescence (panel IF) was performed using a section from a different tumor: OPN was detected with monoclonal antibody 2A1, visualized with rhodamine fluorescence (red); laminin was visualized with fluorescein isothiocyanate (green). The image shown is a single optical section from confocal microscopy. The arrow, upper left, shows a cell expressing high levels of OPN immediately adjacent to the basement membrane of a large blood vessel: OPN is clearly not associated with this basement membrane. Scale bars, panel C, 20 m; panel E, 100 m. All immunohistochemistry sections were photographed at the same magnification as panel E.
ces. Confluent cells on the different matrices were then fractionated into soluble and ECM components, and the presence of OPN in the different fractions determined by Western blotting. This fractionation was performed using either Triton X-100 (Fig. 4) or EDTA (Fig. 5) to remove the cells. When the cells were plated on either matrigel or collagen and then removed with Triton X-100, only a very slight OPN reactivity could be detected in the ECM fraction: this reactivity is seen in both the WT and OPN Ϫ/Ϫ samples, and is probably a cross-reacting protein (Fig. 4, A and B). The presence of ECM components exclusively in the ECM fraction was confirmed by reacting parallel blots with antibody to laminin (Fig. 4C). When the cells were plated on matrigel, and then removed with EDTA, again, OPN was undetectable in the ECM fraction, but was observed in the cell lysate, and was abundant in the conditioned medium ( Fig. 5A and data not shown). Both laminin and fibronectin, on the other hand, were retained in the ECM fraction (Fig. 5, B  and C).
The ability of OPN made by tumor cells to interact with extracellular matrix components was further tested by ELISA. Different ECM components were coated onto wells of 96-well plates, and then incubated at room temperature with conditioned medium from OPN-expressing tumor cells, or from analogous OPN Ϫ/Ϫ cells. The wells were then washed, fixed with paraformaldehyde, and reacted with anti-OPN antibodies. Fixation was performed to ensure that any bound OPN would be retained through the multiple washings included in the ELISA procedure. No detectable OPN was retained on wells coated with fibronectin, matrigel, collagen I, or gelatin (Fig. 6A): while slight immunoreactivity was observed in some wells (e.g. matrigel) it was the same whether the wells were incubated with WT or OPN Ϫ/Ϫ conditioned medium, or with BSA only (Fig.  6A, compare matrigel WT and OPN Ϫ/Ϫ with matrigel only). OPN in the conditioned medium diluted 1:90 could be readily detected when coated directly on the plastic, either with or without fixation (Fig. 6B). Thus, if any of the OPN from this conditioned medium associates with these matrix components, it is less than 1% of protein present in the conditioned medium.
HUVECs express high levels of the ␣ v ␤ 3 integrin (39): function of this integrin is critical for endothelial cell growth and survival in vivo (22). These cells adhere to immobilized OPN with a similar dose response as to other adhesive ligands such as fibronectin and laminin (Fig. 7). Post-translational modification of OPN (phosphorylation and glycosylation) is not required for the protein to support adhesion of these cells (compare mOPN and GST-OPN). HUVECs require adhesion to the extracellular matrix to proliferate, and rapidly undergo apoptosis when held in suspension in the absence of growth factors (40,41). As tumor OPN appears to be primarily soluble, we have asked whether soluble OPN can have any of the same effects on HUVECs as immobilized OPN. HUVECs were starved for growth factors overnight, and plated on wells coated with fibronectin, or with different concentrations of OPN. In other wells, the cells were prevented from adhering to the dish by plating on BSA. Soluble OPN at different concentrations was added to the cells kept in suspension. These cells were then incubated overnight in the presence of 10 ng/ml aFGF, and their ability to transit the cell cycle assessed by incubation with [ 3 H]thymidine. Thymidine incorporation was similar when the cells were plated on FN, or on all concentrations of OPN tested (Fig. 8A). However, cells held in suspension by plating on BSA were unable to incorporate thymidine, and soluble OPN (sOPN) had no effect on these cells ability to enter the cell cycle.
When HUVECs were plated as described above and incubated overnight in the absence of growth factor, cells held in FIG. 4. Western blot of OPN and laminin in matrix and cells following Triton X-100 extraction. Wild type or OPN Ϫ/Ϫ rastransformed 3T3 cells were plated on either matrigel (1:20 dilution) or collagen (0.1 mg/ml) and allowed to become confluent. The cells were removed from the dish with 0.5% Triton X-100 (cells lanes), leaving the extracellular matrix behind. The wells were then washed with 20 mM NH 4 OH, and then the ECM was solubilized with 2% SDS (ECM lanes). For comparison, parallel wells were lysed directly in 2% SDS (total lanes). Each sample was originally 500 l, and equal volumes were loaded in each lane as indicated below. For the blots reacted with anti-OPN antibody (panels A and B), extracts were precipitated with acetone, and the pellet resuspended and loaded; for the blot reacted with the anti-laminin antibody, the total extract was loaded directly. suspension (plated on BSA-coated wells) underwent apoptosis. While immobilized OPN at different concentrations or immobilized fibronectin could suppress this apoptotic response, soluble OPN had no effect on the apoptotic response of cells held in suspension (Fig. 8B). Control experiments indicated that GST-OPN coated onto the wells remained immobilized throughout the course of the experiment. Conversely, we have not formally eliminated the possibility that some small fraction of the soluble OPN may become immobilized during the experiment. However, the lack of any effect of soluble OPN above that of BSA alone or GST (Figs. 8, A and B, compare sOPN, BSA, and GST samples) supports the idea that the soluble OPN added to the wells remains soluble throughout the experiment. If there is some immobilization of the soluble OPN, then it is an insufficient amount to have an effect in these assays. Thus, we conclude that while immobilized OPN can support endothelial cell proliferation, and suppress the apoptosis occurring in the absence of adhesion, soluble OPN, such as that associated with tumors, has no such effects. Thus, endothelial cell proliferation and survival is probably not regulated by OPN in tumorigenesis, and the protein is likely acting as a cytokine in either an autocrine or paracrine manner. The target cells of this soluble OPN produced by tumors is still unknown: it could be the tumor cells themselves or a variety of different host cell types.

DISCUSSION
Osteopontins association with bone matrix as well as its homology (chiefly the RGD sequence) with extracellular matrix proteins has earned it a reputation as an extracellular matrix protein. Structurally, however, the protein has significant differences from well characterized matrix proteins such as fibronectin, laminin, and collagen. OPN is a small protein (ϳ300 amino acids) with no recognized domain or repeat structure (42). Unlike these other matrix proteins, OPN is not known to form organized multimeric structures. Thus, if it were to associate with the insoluble extracellular matrix, it must do so through interaction with other matrix components. In bone, OPN associates with the matrix largely through its high affinity for hydroxyapatite: OPN is not released from bone by chaotropic reagents alone, but requires high concentrations of a calcium chelator such as EDTA for release from the bone matrix (26).
There are numerous reports, however, that OPN binds specifically to components of the extracellular matrix, and that it is a substrate for transglutaminase cross-linking (43,44). In bone extracts, for example, OPN immunoreactive protein migrates at high molecular weights (44), and is thought to be cross-linked by transglutaminase through conserved glutamine residues. OPN can also be cross-linked in vitro to fibronectin (45). This cross-linking of OPN could serve to immobilize the protein in the extracellular matrix. In addition, OPN in vitro can bind specifically to collagen (46,47). These observations are taken as evidence for OPNs association with the extracellular matrix in non-mineralized tissues. However, it seems that this latter interaction may not occur readily in soft tissues such as tumors in vivo. In our experiments, binding of OPN to collagen was undetectable in cells in culture: OPN produced by tumor cells failed to associate with collagen coated on the culture dish in either the presence or absence of the cells. Thus, while this association may occur under defined conditions, with purified components, it does not appear to occur at relevant levels in the more complex conditions of the cell culture system used or in tumor tissue. For instance, if OPN were cross-linked in tumor ECM to fibronectin, we would expect to see retention of OPN in the tumor insoluble material, which retains abundant fibronectin (Fig. 2), but this is not the case. No OPN could be detected in the pellet fraction in this experiment.
We have utilized a variety of techniques to isolate extracellular matrix, and see no association of OPN with these different FIG. 6. ELISA of OPN association with matrix components. Fibronectin, matrigel, collagen I, gelatin, and BSA as a negative control, all at 200 g/ml, were coated on wells of a 96-well plate. Undiluted serum-free conditioned medium freshly collected from ras-transformed cells was added to each well and incubated for 2 h. After washing, the matrix containing wells were fixed with 4% paraformaldehyde, as were some of the control wells as indicated. ELISA for OPN was then performed as described under "Materials and Methods." A, fibronectin (FN), matrigel (MTG), collagen (COL), gelatin (GEL), and BSA-coated wells were incubated with either WT (striped bars) or OPN-deficient conditioned medium (dotted bars). Control wells contained matrigel (MTG ONLY) and BSA (BSA ONLY) incubated in the absence of conditioned medium. Conditioned medium (COND MED) coated directly on the plate is included as a positive control. B, in the same experiment WT or OPN Ϫ/Ϫ conditioned media was coated directly on the plate undiluted (UNDIL) or after serial dilutions as indicated. Some wells were fixed (FIX) to assess the effects of fixation on antibody reactivity. HUVECs were trypsinized, resuspended in serum containing medium, and washed 3 times in serum-free medium before plating at 2 ϫ 10 4 cells/ well. The cells were allowed to adhere for 2 h. Adhesion was quantified by staining the washed wells with crystal violet: the dye bound to the cells was eluted following washing, and the amount of eluted dye measured at a wavelength of 630 in an ELISA reader. matrix preparations. Triton X-100 extraction of cellular components yields an ECM preparation that retains diverse proteins such as tenascin (48), apolipoprotein E (36), and plasminogen activator inhibitor (37), but not OPN (Fig. 4). In culture, removal of the cells with EDTA removes some of the matrix, as shown by the presence of laminin in the cell lysate fraction (Fig.  5). This is probably due to the extensive mixing required to remove the cells, and likely represents fragments of ECM that become detached during this process. However, considerable laminin and fibronectin remain in the ECM fraction, and there is no evidence of OPN in this preparation.
The experiments described here indicate that at least in tumor tissue, OPN does not associate significantly with the extracellular matrix. This is especially clear in regard to the basement membrane elaborated in vivo by endothelial cells: these structures are devoid of OPN immunoreactivity by immunohistochemistry and immunofluorescence. Thus, it is unlikely that OPN facilitates tumorigenesis through a proliferative or protective effect on endothelial cells, since we also showed that soluble OPN does not mediate these effects. An important aspect of the experiments presented here is the inclusion of the OPN Ϫ/Ϫ tumors as controls for the immunohistochemistry, demonstrating the specificity of the antibody reaction with mouse OPN.
It appears that most, if not all, of the OPN produced by the tumors in vivo is exported to the circulation. This is consistent with the appearance of OPN in a variety of extracellular fluids (49), and with the observation of elevated OPN levels in the plasma of patients with metastatic disease (50). This is the first report, however, in which serum OPN has been shown to be elevated as a result of host cell expression. It has been known that OPN is expressed by macrophages infiltrating tumors, and that often these cells show higher levels of expression than the tumor cells themselves (6). Immunohistochemical results indicate that in wild type hosts injected with ras-transformed 3T3 cells, macrophages infiltrate both wild type and OPN-deficient tumors to a similar extent, and that numerous cells showing reactivity with OPN antibodies are present in OPN-deficient tumors (data not shown). The Western blot analysis of Fig. 1 indicates that the production of OPN exclusively by these host cells can result in levels of OPN in the tumor that approaches that seen when the tumor expresses OPN. This observation highlights the extent of involvement of OPN in the host response to tumorigenesis.
OPN has been characterized as a cell attachment molecule in a variety of models, and this attachment can be both RGD-dependent and -independent (10,15,51). Cell attachment is critical for growth and survival of a wide variety of cell types, in many cases due to a requirement for signaling through integrin-stimulated pathways (52,53). Thus it is not surprising that OPN, which binds to both integrin and non-integrin adhesive receptors, can support cell proliferation and survival when immobilized on a plastic surface (23). However, our results indicate that in vivo, OPN is unlikely to serve these functions in the absence of its immobilization in the extracellular matrix. Thus, at least in the case of OPN produced by tumors, the protein is functioning as a soluble molecule. To understand the role of OPN in promoting tumorigenesis, therefore, the focus must be on functions supported by soluble OPN. Recently, several systems have been described in which soluble OPN can enhance cell proliferation (54 -56) or affect cell functioning (57)(58)(59). In addition, we have shown that soluble OPN can affect macrophage function. 3 These new insights into the role of soluble OPN will help to advance our understanding of the role of OPN in tumorigenesis.

FIG. 8. Effect of soluble (sOPN) and immobilized OPN on HU-VEC proliferation and apoptosis.
HUVECs were grown to 80% confluence and made quiescent by overnight incubation in serum-free medium without growth factors. The cells were then plated on wells previously coated with GST-OPN at 1-50 g/ml (1-50, OPN), fibronectin (10 g/ml, FN), with GST protein (10 g/ml, GST) or held in suspension by plating on wells coated with 2% BSA. Cells held in suspension were treated with soluble GST-OPN at different concentrations (sOPN 1-50). Panel A, [ 3 H]thymidine incorporation. Cells were incubated for 24 h in the presence of aFGF and heparin. For the last 5 h, [ 3 H]thymidine was added to 2 Ci/ml. The cells were then washed, fixed with trichloroacetic acid, and solubilized prior to counting. Total counts/ well are shown for each substrate or addition. Panel B, apoptosis. Cells were incubated overnight in the absence of growth factors, plated on different substrates, and treated with soluble GST-OPN as indicated above. After 22 h, the cells were labeled with Hoescht 33342 for 15 min. Floating and adherent cells were collected, fixed with 4% paraformaldehyde, and the number of fragmented and intact nuclei determined by fluorescence microscopy. Percent apoptotic cells is equal to the percentage of cells with fragmented nuclei.