The Shedding of Betaglycan Is Regulated by Pervanadate and Mediated by Membrane Type Matrix Metalloprotease-1*

Betaglycan is a membrane-anchored proteoglycan that binds transforming growth factor-β (TGF-β) via its core protein. A soluble form of betaglycan can be released by proteolytic cleavage (also known as shedding) of the membrane-bound form, yielding soluble betaglycan. The mechanism leading to the generation of soluble betaglycan is poorly understood. Because the membrane and soluble forms of betaglycan have opposite effects regulating the availability of TGF-β, it is important to characterize the shedding of betaglycan further. Here we present evidence showing that in certain cell types, pervanadate, a general tyrosine phosphatase inhibitor, induces the release of the previously described fragment that encompasses almost the entire extracellular domain of betaglycan (sBG-120). In addition, treatment with pervanadate unveils the existence of a novel 90-kDa fragment (sBG-90). Noticeably, the cleavage that generates sBG-90 is mediated by a tissue inhibitor of metalloprotease-2-sensitive protease. Overexpression of all membrane type matrix metalloproteases (MT-MMPs) described to date indicates that MT1-MMP and MT3-MMP are endowed with ability to generate sBG-90. Furthermore, the patterns of expression of different MT-MMPs in the cell lines used in this study suggest that MT1-MMP is the protease involved in the shedding of sBG-90. Overexpression of MT1-MMP in COS-1 cells, which do not express detectable levels of this metalloprotease, confirms the feasibility of this hypothesis. Unexpectedly, during the course of these experiments, we observed that MT2-MMP decreases the levels of MT1-MMP and betaglycan. Finally, binding competition experiments indicate that, similar to the wild type membrane betaglycan, sBG-90 binds the TGF-β2 isoform with greater affinity than TGF-β1, suggesting that once released, it could affect the cellular availability of TGF-β.

Transforming growth factor-␤ (TGF-␤) 1 is the prototype of a superfamily of growth factors involved in the control of cell proliferation, differentiation, development, and extracellular matrix production (1,2). TGF-␤ controls many physiological processes, and disturbances in its regulation or signaling can lead to disease (3,4). TGF-␤ signals into the cell through two cell surface serine/threonine kinase receptors, the TGF-␤ type I and type II receptors. Signaling is initiated when TGF-␤ promotes the association of type I and type II receptors. Receptor I is then phosphorylated and activated by the constitutively active kinase of receptor II. The activated receptor I kinase then phosphorylates certain members of the Smads, a novel family of transcriptional regulatory proteins which transduce the TGF-␤ signal to the nucleus (5,6).
TGF-␤ has two coreceptors, betaglycan and endoglin, which regulate TGF-␤ access to its signaling receptors. Betaglycan and endoglin are transmembrane glycoproteins with large extracellular regions that bind TGF-␤ and very similar cytoplasmic regions without any identifiable signaling motif (7)(8)(9). Betaglycan, also known as type III receptor, is a membrane proteoglycan whose glycosaminoglycan chains consist of heparan and chondroitin sulfate. In general, membrane betaglycan is considered a positive regulator of TGF-␤ because it increases the affinity of the binding of TGF-␤ to receptor II, enhancing cell responsiveness to TGF-␤ (10,11). This effect is particularly significant for TGF-␤2, the isoform for which betaglycan has higher affinity, and appears to be mediated by a "presentation complex" formed by betaglycan, TGF-␤ and the type II receptor (10). Betaglycan also regulates the actions of activins, inhibins, and BMPs, other members of the TGF-␤ superfamily. Specifically, upon binding of inhibin A, betaglycan can associate with the type II receptors for activin or BMP. Because of their presence in such nonsignaling complexes, the availability of these type II receptors is reduced, becoming limited to associating with their corresponding type I receptors and thereby preventing activin or BMP actions (12,13). Thus, betaglycan can establish diverse ligand-dependent interactions with type II receptors for at least three members of the TGF-␤ superfamily. On the one hand, the complex formed with TGF-␤ and its type II receptor enhances TGF-␤ actions. Importantly, this latter function explains key features of the betaglycan-null mice phenotype (14). On the other hand, those formed with inhibin A and the corresponding type II receptors antagonize the effects of activin or BMP. As a matter of fact, up to date, betaglycan stands as the only identified inhibin A receptor capable of mediating its characteristic activin antagonism (15,16). Although the glycosaminoglycan chains are not required for all of these functions (10,17), they may modulate them. It has been reported that in some cell lines the nature of the glycosaminoglycan attached to membrane betaglycan core protein may sterically prevent its association with the TGF-␤ type II receptor, turning it into an inhibitor of TGF-␤ (18). Betaglycan is also required for the epithelial-mesenchymal transition of cardiac endothelial cells which leads to heart valve formation. For this reason, in addition of being a TGF-␤ coreceptor, a more direct function of betaglycan in TGF-␤ signaling has been proposed (19).
A natural soluble form of betaglycan can be found in serum, extracellular matrices, and the conditioned medium of several cell lines (20). In contrast to membrane betaglycan, soluble betaglycan inhibits TGF-␤ binding to cell surface receptors in vitro (21,22). Thus, betaglycan could be a dual modulator of TGF-␤ activities: as a membrane protein it is an enhancer of TGF-␤ actions, and as a soluble protein it is an inhibitor (21,22). This latter property has been exploited therapeutically to block TGF-␤ actions in a carcinogenesis animal model using a recombinant soluble betaglycan (23). Several studies have proposed that the natural soluble form of betaglycan is generated by a proteolytic cleavage of the membrane-bound form (10,20,21). However, despite its functional importance, the proteolytic cleavage of betaglycan has been poorly characterized.
The proteolytic cleavage of the extracellular domain is not particular for betaglycan; many transmembrane proteins undergo a similar process, frequently referred to as "ectodomain shedding." The shedding of a large number of membrane proteins is a regulated process that can be enhanced by PMA, a phorbol ester activator of protein kinase C and other signal transduction pathways, and is mediated by zinc metalloproteases (for reviews, see Refs. 24 -26). The disintegrin and metalloprotease ADAM 17, also known as TACE (Tumor necrosis factor ␣-Converting Enzyme), has been implicated in the vast majority of protein kinase C-activated shedding events analyzed to date (27)(28)(29). In a previous report we showed that, in contrast to the majority of proteins, the shedding of betaglycan is not stimulated by PMA and remains unaffected in a mutant cell line defective in the shedding of a wide variety of proteins (30). In this study, we analyzed the shedding of betaglycan in different cell lines, and unexpectedly, we found a novel form of the soluble receptor. The shedding of betaglycan can be stimulated by pervanadate and is mediated by a metalloprotease that is inhibited by TIMP-2 but not by TIMP-1. Transfection experiments indicate that MT1-MMP and/or MT3-MMP, or a similar metalloprotease, is responsible for the shedding of betaglycan.
Cell Culture-All cells used in this study were obtained from the American Type Culture Collection (Rockville, MD). L6E9 cells stably expressing Mycgag Ϫ were grown in Dulbecco's modified Eagle's medium (DMEM) containing 20% fetal bovine serum (Invitrogen) in the presence of 400 g/ml Geneticin (Invitrogen). Wild type CHO cells and M1 mutant cells stably expressing HABG were grown in DMEM containing 10% fetal bovine serum in presence of 400 g/ml Geneticin. CHO and COS-1 cells were grown in DMEM containing 10% fetal bovine serum.
cDNAs and Transfections-The Mycgag Ϫ betaglycan mutant, described previously (21), was subcloned in the pcDNA3 and stably transfected in L6E9 cells as described before (17). This gag Ϫ betaglycan construct is a double point betaglycan mutant of the serines that are the glycosaminoglycan chains acceptor amino acids (S535A/S546A) and expresses betaglycan only as core protein, greatly simplifying the analysis of the betaglycan shedding products. Wild type CHO cells and M1 mutant CHO cells defective in pro-HA/TGF-␣ shedding have been described elsewhere (32). Wild type CHO cells and M1 mutant cells were stably transfected with HABG construction subcloned in the pCEP4 vector using the calcium phosphate precipitate method (33). The HAgag Ϫ and HAgag Ϫ FLAG were constructed from the wild type rat betaglycan cDNA tagged at the amino terminus with the HA epitope. Both constructions were created replacing in the wild type betaglycan cDNA a fragment of a previously described gag Ϫ mutant in the XhoI and BglII sites (21). The FLAG epitope was inserted at the carboxyl terminus for the HAgag Ϫ FLAG and for the HABGFLAG constructions by PCR. All DNA manipulations were verified by nucleotide sequencing using standard techniques (34). For transient expression in COS-1 and CHO cells the HAgag Ϫ and the HAgag Ϫ FLAG constructions were subcloned in the pCMV5 vector (35). The cDNA encoding hamster MT1-MMP tagged at the amino terminus with the Myc epitope has been described before (36). The human MT2-MMP, human MT3-MMP, and HA-tagged mouse MT4-MMP, human MT5-MMP, and human MT6-MMP cDNAs were kindly provided by Dr. López-Otín (37)(38)(39). Transient L6E9 and CHO transfection was done with LipofectAMINE (Invitrogen) following the manufacturer's recommendations. Transient COS-1 transfection was done with the diethylaminoethyl-dextran method (40). Cells were assayed at 48 h post-transfection.
Northern Blot Analysis-Total RNA was isolated from cell cultures using TriZol (Invitrogen), and 10 g from each cell line was electrophoresed in 0.9% agarose/formaldehyde gels and transferred to nylon membranes Hybond-N (Amersham Biosciences). The cDNA probes used were hamster MT1-MMP, corresponding to a 2,669-bp XbaI/HindIII fragment of the plasmid pcD MycMT1 (36); human MT2-MMP, comprising a 1,996-bp EcoRI/PstI fragment from the plasmid pSG MT2-MMP; and human MT3-MMP, of a 1,330-bp EcoRI fragment from the plasmid pSG MT3-MMP. The cDNA probe for glyceraldehyde-3-phosphate dehydrogenase was from rat. Labeling of cDNA probes was done using the Random Primer Labeling Kit (Roche Applied Science) and [␣-32 P]dCTP (Amersham Biosciences) according to the manufacturers' instructions. Hybridization was performed overnight at 65°C in 0.5 M phosphate (pH 7.2), 7% SDS, 10 mM EDTA. Bands were visualized using a PhosphorImager (Amersham Biosciences).
Immunoprecipitation-Approximately 7 ϫ 10 5 L6E9 cells stably expressing Mycgag Ϫ were incubated for different periods of time in the absence or presence of 100 M pervanadate in medium supplemented with 0.2% fetal bovine serum. Cells were lysed in lysis buffer (PBS containing 1% Nonidet P-40, 5 mM EDTA, and protease inhibitors). Medium or cell lysates were immunoprecipitated with anti-betaglycan ectodomain antibody 822. Immune complexes were collected with protein A-Sepharose.
Western Blotting-For Western blotting, immune complexes or aliquots from medium or cell lysates were electrophoresed in polyacrylamide gels and transferred to polyvinylidene difluoride membranes (Millipore). Membranes were probed using 1/2000 of the anti-betaglycan ectodomain 822, anti-c-Myc, or anti-HA. Anti-MT1-MMP LEM 2/63 was used directly. Anti-MT2-MMP was used at 1/200 for transfected MT2-MMP and at 10 g/ml for endogenous MT2-MMP. Anti-MT3-MMP was used at 10 g/ml. Immunoblots were revealed using ECL (Amersham Biosciences) and autoradiography. Only where indicated was the ECL Plus kit (Amersham Biosciences) used. Metabolic Labeling and Immunoprecipitation-Approximately 7 ϫ 10 5 cells were labeled for 2 h with 250 Ci/ml 35 S-Pro Mix (Amersham Biosciences) in DMEM without L-methionine and L-cysteine (Sigma). The label was chased for 2 h in DMEM containing with 10% fetal bovine serum (complete medium). Then the cells were incubated in the presence of different compounds, as indicated, for 30 min in DMEM containing 0.2% fetal bovine serum. The medium was collected and centrifuged at 14,000 rpm for 2 min to remove debris. The cells were washed twice with cold PBS and lysed in PBS containing 1% Nonidet P-40, 5 mM EDTA and 1 mM phenylmethylsulfonyl fluoride. Cell lysates and medium were immunoprecipitated with different antibodies. Immune complexes were collected with protein A-Sepharose (for antibodies 822 and 821) or protein G-Sepharose (for anti-c-Myc and anti-HA antibodies), washed three times with PBS containing 0.1% Triton X-100 and 0.1% SDS, separated in SDS-PAGE, and bands visualized using a PhosphorImager. Densitometric analysis was carried out using Image-Quant software.
Affinity Labeling and Competition Experiments-TGF-␤2 was radiolabeled with 125 I using chloramine T (41). Baculoviral recombinant soluble betaglycan was purified as described before (22). Amounts of soluble betaglycan sBG-90 were determined by Western blot. Briefly, medium from pervanadate-treated L6E9 cells stably expressing Mycgag Ϫ was concentrated in an Amicon Ultrafiltration chamber (Millipore) and dialyzed against PBS. Aliquots from medium were separated in SDS-PAGE, subjected to Western blotting, and sBG-90 was quantified comparing with different amounts of recombinant soluble betaglycan as standards (data not shown). For competition experiments, recombinant soluble betaglycan and medium with sBG-90 (10 ng/assay) were affinity labeled with 100 pM 125 I-TGF-␤2 in the presence of different concentration of unlabeled TGF-␤1 or TGF-␤2 for 3 h in PBS supplemented with 0.05% Triton X-100. Cross-linking was done by the addition of 0.1 mg/ml disuccinimidyl suberate (Pierce). Labeling reactions were quenched after 15 min with Tris/HCl, pH 7.5, at a final concentration of 10 mM. The reaction mixture was immunoprecipitated with anti-c-Myc monoclonal antibody and protein G-Sepharose. Immunoprecipitates were separated by SDS-PAGE and visualized by a PhosphorImager.

RESULTS
Pervanadate Activates the Shedding of Betaglycan-Ectodomain shedding is frequently a regulated process. However, the shedding of betaglycan is insensitive to phorbol esters, calcium ionophores, or serum factors, three agents that activate the shedding of most proteins analyzed to date (30,32,42). Recently, we have shown that the shedding of the tyrosine kinase receptor HER-2, which is also insensitive to these treatments, can be activated by pervanadate, a tyrosine phosphatase inhibitor (43). Therefore, we analyzed the effect of pervanadate on the shedding of the betaglycan. Because of the presence of glycosaminoglycan chains, wild type betaglycan migrates in SDS-PAGE as a broad band of 180 -250 kDa, making it difficult to analyze the soluble products of the shedding reaction. Thus, to facilitate this assay, we stably transfected L6E9 myoblasts with a c-Myc-tagged rat membrane betaglycan devoid of glycosaminoglycan chains (L6E9/Mycgag Ϫ cells) (Fig. 1A). A short treatment of L6E9/Mycgag Ϫ cells with pervanadate induces the liberation of a 90-kDa fragment, which is immunoprecipitated by the polyclonal antibody 822 (directed against the betaglycan extracellular region) and by the monoclonal antibody directed against the c-Myc epitope engineered at the amino terminus (Fig. 1B). As expected, this soluble 90-kDa product (hereafter referred to as sBG-90) is not immunoprecipitated by the polyclonal antibody 821 directed against the cytoplasmic tail of betaglycan (Fig. 1B). The size of sBG-90 differs from the one reported previously (110 -120 kDa) for the soluble betaglycan species produced spontaneously by 3T3-L1 cells (20).
To determine whether L6E9/Myc gag Ϫ cells also secrete the 110 -120-kDa form, we analyzed the basal shedding of betaglycan in these cells. Fig. 1C shows that in the absence of any treatment, a 120-kDa betaglycan fragment (hereafter referred to as sBG-120) is slowly shed to the medium, being barely visible at 7 h but clearly detectable by 24 h. This slow process very much resembles the "constitutive" shedding that has been invoked to explain the existence of the naturally occurring forms of soluble betaglycan (30). Interestingly, Fig. 1C shows  that sBG-90 is also generated by this constitutive mechanism. Thus, these data suggest that, at least two cleavages at different sites of its ectodomain may generate two different soluble forms of the receptor, sBG-120 and the shorter sBG-90 form.
The effect of pervanadate on the shedding of sBG-90 is specific because vanadate or H 2 O 2 was without effect (Fig. 1D). Thus, these data indicate that the production of sBG-90 may be regulated by phosphorylation/dephosphorylation. On the other hand, as reported previously for the 110 -120-kDa fragment (30), the release of sBG-90 it is not regulated by PMA or fetal bovine serum (Fig. 1E).
Different Cell Lines Produce Different Soluble Betaglycan Fragments-To determine whether the generation of sBG-90 is a particular property of L6E9 myoblasts, we analyzed the shedding of betaglycan in different cell lines. Therefore, we transiently transfected the HAgag Ϫ FLAG construct (Fig. 1A) in CHO and COS-1 cells and compared the products of the betaglycan shedding reaction by these cells with those generated by L6E9/Mycgag Ϫ cells after pervanadate treatment. As shown in Fig. 2A, pervanadate also increased betaglycan ectodomain shedding in CHO and COS-1 cells. However, CHO cells released, in a pervanadate-activated manner, the two forms of soluble betaglycan ectodomain, sBG-90 and sBG-120 ( Fig. 2A). In contrast, COS-1 cells stimulated with pervanadate only released the sBG-120 product ( Fig. 2A). Thus, the generation of sBG-90 and sBG-120 is cell type-dependent, but the shedding of both fragments can be stimulated by pervanadate. Fig. 2A also shows that minor forms of soluble betaglycan are produced by CHO (ϳsBG-130) and COS-1 cells (ϳsBG-130 and ϳsBG-100). However, because their production is not stimulated by pervanadate, we did not characterize them any further.
Some authors have suggested that the cytoplasmic tail is involved in the maturation of membrane betaglycan (44). Because the experiments in Fig. 2A were performed with a carboxyl end FLAG-tagged betaglycan construct, it was important to determine whether or not this tag had any effect in the shedding of the receptor. For that purpose, we compared the pervanadate-stimulated shedding of betaglycan constructs with or without the FLAG tag (Fig. 1A). As show in Fig. 2B, the presence or absence of this 8-residue tag at the carboxyl end of the receptor did not affect the shedding process. In addition, this experiment showed that the shedding of the proteoglycan form of betaglycan is also stimulated by pervanadate, indicating that the presence of glycosaminoglycan chains does not interfere with this activation.
Pervanadate-inducible Betaglycan Shedding Is Not Mediated by TACE-We have reported previously that the constitutive shedding of betaglycan is not affected in a mutant cell line (M1 cells) defective in the shedding of a variety of other proteins (30). It has been shown recently that this mutant cell line has a specific defect that prevents the activation of TACE, but not other metalloproteases (45). Therefore, to determine whether active TACE participates in the shedding of betaglycan induced by pervanadate, we analyze the effect of pervanadate on the M1 mutant and on the wild type CHO cells that were stably transfected with HA-tagged betaglycan (HABG construct, see Fig.  1). The levels of the soluble form of betaglycan generated by M1 cells treated with pervanadate were indistinguishable from those generated by wild type cells under the same conditions (Fig. 2C), indicating that TACE does not mediate the pervanadate-stimulated shedding of betaglycan.
The Shedding of sBG-90 Is Sensitive to TIMP-2-The vast majority of shedding events analyzed to date is executed by zinc-dependent metalloproteases. To characterize the shedding activity stimulated by pervanadate, we tested the effect of different protease inhibitors. As expected, whereas the hydroxamates BB-94 and TAPI-2 prevented the production of sBG-90 (Table I and Fig. 3A), compounds that block serine, cysteine or aspartic proteases had little or no effect (Table I). This result indicates that a zinc-dependent metalloprotease is necessary for the shedding of sBG-90. Hydroxamates are known to inhibit the matrixins (MMPs) as well as reprolyisins (ADAMs, metalloprotease disintegrins). To define further the proteolytic activity responsible for the production of sBG-90, we analyzed the effect of two members of the family of the tissue inhibitors of metalloproteases, TIMP-1 and TIMP-2, which are specific for certain MMPs and metalloprotease disintegrins (46). Fig. 3B shows that low concentrations of TIMP-2 blocked the shedding of sBG-90, whereas TIMP-1 had little or no effect, indicating that a metalloprotease sensitive to TIMP-2 is necessary for the shedding of sBG-90 in L6E9/Mycgag Ϫ cells.
Similar experiments with COS-1 cells transfected with the HAgag Ϫ construct (Fig. 3C) indicated that the pervanadatestimulated shedding of sBG-120 is blocked by BB-94 and unaffected by TIMP-1. Interestingly, TIMP-2, at concentrations that completely inhibited the cleavage of sBG-90, did not have a significant effect on the cleavage of sBG-120 (Fig. 3C), indicating that different proteases are involved in the shedding of the different forms of soluble betaglycan.
Overexpression of MT1-or MT3-MMPs Augments the Shedding of sBG-90 -The activity of several metalloprotease-disintegrins putatively involved in ectodomain shedding, such as TACE, is not inhibited by TIMP-2 (46), further supporting that the generation of sBG-90 is not dependent on these metalloproteases. In contrast, other metalloproteases putatively involved in ectodomain shedding, such as certain MT-MMPs, are inhibited by TIMP-2 (37). To date, six MT-MMPs have been described (37)(38)(39)(47)(48)(49). To analyze the putative role of MT-MMPs on the shedding of sBG-90, we analyzed the effect of the overexpression of these metalloproteases on the production of sBG-90 by L6E9 cells. As shown in Fig. 4A, overexpression of MT1-or MT3-MMP in L6E9/Myc gag Ϫ cells clearly induced the production of sBG-90 by untreated cells, whereas MT2, MT4-, MT5-, and MT6-had no effect. In addition, MT1-had a modest but reproducible effect, increasing the production of sBG-90 in cells treated with pervanadate. These results indicate that MT1-, MT3-MMPs or a similar metalloprotease, could be responsible for the production of sBG-90 in L6E9 cells. On the other hand, overexpression of MT2-MMP partially inhibited the production of sBG-90 induced by pervanadate. A similar effect was observed in CHO cells transiently transfected with betaglycan and MT2-MMP (data not shown and see Fig. 7), opening the possibility that MT2-MMP influences the expression of betaglycan or, alternatively, that of the endogenous protease responsible for the shedding of sBG-90.
To confirm and extend these results we performed a similar Blots were revealed with ECL Plus kit and exposed to x-ray films for different times as indicated. experiment using COS-1 cells, which, in contrast to L6E9 cells, produce predominantly sBG-120 upon the treatment with pervanadate. Thus, we transiently coexpressed in COS-1 cells the HAgag Ϫ FLAG construct together with each one of the MT-MMPs or with the empty expression vector and analyzed the appearance of the sBG fragments in the presence or absence of pervanadate. As in the case of L6E9/Mycgag Ϫ cells, although overexpression of MT4-, MT5-, and MT6-MMP had little or no effect, overexpression of MT1-and MT3-MMP induced the production of sBG-90 in untreated and treated cells (Fig. 4B). This induction was concomitant with a profound decrease in the production of sBG-120, indicating a competition between MT1-, MT3-MMP and the protease responsible for the generation of sBG-120. Identical results were observed when COS-1 cells were transfected with a HAgag Ϫ construct devoid of the FLAG epitope at the carboxyl terminus (data not shown), suggesting that the cytosolic tail does not participate decisively in this process. As observed in the case of sBG-90 (Fig. 4A), the over-expression of MT2-MMP had an inhibitory effect on the pervanadate-induced production of sBG-120 in COS-1 cells (Fig.  4B), which suggests that this metalloprotease could affect the expression of betaglycan and/or of the protease responsible for the generation of sBG-120. Importantly, in the experiments shown in Fig. 4, the presence of each one of the transfected MT-MMPs was confirmed by Western blot analysis of cell lysates, indicating that the lack of effect of some metalloproteases was not the result of failed expression. Thus  (50), did not express this MT-MMP (Fig. 5A). Cell lysates subjected to Western blot analysis confirmed the results obtained by Northern blot (Fig.  5B). On the other hand, MT3-MMP mRNA was undetectable in CHO, COS-1, and L6E9 cells (Fig. 5C). Thus, there is an excellent correlation between the presence of MT1-MMP and the ability to shed sBG-90 in the cell lines studied. This fact, along with the lack of detectable levels of MT3-MMP mRNA, suggests that, at least in L6E9 and CHO cells, the metalloprotease responsible for the generation of sBG-90 is MT1-MMP.
MT2-MMP Decreases MT1-MMP and Betaglycan-One surprising result from the above studies was the observation that overexpression of MT2-MMP decreases the production of sBG-90 by L6E9 cells and abolishes the production of sBG-120 in cotransfection experiments in COS-1 cells (Fig. 4). A possible explanation for these findings is that MT2-MMP affects the expression of betaglycan and/or the protease responsible for the generation of sBG-90, likely MT1-MMP in L6E9 cells. To clarify these points, as a first approach, we determined the level of expression of MT2-MMP in COS-1, L6E9, and CHO cells with a well characterized antibody. The Western blot in Fig. 5D indicated that MT2-MMP is practically absent in these cell lines. Northern blot analysis (not shown) showed that MT2-MMP mRNA was undetectable in the same lines. This prompted us to determine whether or not increasing amounts of MT2-MMP had any effect on the levels of MT1-MMP and/or betaglycan. For that purpose, we carried out cotransfection experiments. Coexpression in COS-1 cells of increasing amounts of MT2-MMP with HAgag Ϫ or Myc/MT1-MMP, separately (Fig. 6, A and B, respectively), or together (Fig. 6C), had a detrimental effect on the levels of MT1-MMP and betaglycan. Western blot analysis of total cell lysates reproducibly demonstrated that HAgag Ϫ decreases as MT2 increases (Fig. 6A). Similarly, as revealed by an antibody against its epitope tag (c-Myc), as well as an antibody directed against its catalytic region (LEM 2/63), the levels of transfected Myc/MT1-MMP were decreased by MT2-MMP in a dose-dependent manner (Fig. 6B). Importantly, in CHO cells the same effects were observed on the levels of endogenous MT1-MMP (Fig. 6D) and transfected HAgag Ϫ (Fig. 6E) showing that the observed effects, at least in the case of MT1-MMP, are not exclusive of the transfected proteins. Consistent with the above results, the shedding of both forms of soluble betaglycan sBG-90 and sBG-120 was decreased in a concentration-dependent manner by MT2-MMP expressed in COS-1 (Fig. 7A) and CHO cells (Fig.  7B). Collectively, these results show that overexpression of MT2-MMP induces a down-modulation of ectopically expressed betaglycan, explaining the decrease in the production of sBG-120 and sBG-90 observed in different cell lines. In addition, MT1-MMP, likely responsible for the generation of sBG-90, is also down-modulated by MT2-MMP.  D). The locations of sBG-90 and recombinant soluble betaglycan (rsBG) are indicated. Because the total level of binding was lesser for the sBG-90 fragment, its gels (A) were exposed 4 times longer than the gels containing the rsBG (B).

sBG-90 Binds TGF-␤1 and TGF-␤2 with Relative Affinities
Similar to the Wild Type Membrane Receptor-To characterize the TGF-␤ binding properties of sBG-90, we performed ligand binding competition assays. For this purpose medium from L6E9/Mycgag Ϫ cells treated with pervanadate were affinity labeled in solution with a constant amount of 125 I-TGF-␤2 and an increasing concentration (from 0 to 4 nM) of competing unlabeled TGF-␤1 or TGF-␤2 (Fig. 8A). As a reference for comparison, we also tested the baculoviral recombinant soluble betaglycan (Fig. 8B), which binds TGF-␤ similar to the wild type membrane receptor (22). Binding of 125 I-TGF-␤2 to sBG-90 (Fig. 8C) was effectively competed by cold TGF-␤2 but not by the TGF-␤1 isoform. This pattern of competition, which was exhibited by recombinant soluble betaglycan (Fig. 8D), is characteristic of wild type betaglycan (17). Therefore, this result opens the possibility that sBG-90 is a specific modulator of TGF-␤ and or any of its other known ligands. DISCUSSION Here we show that betaglycan shedding can be stimulated by pervanadate, a general inhibitor of tyrosine phosphatases, and results in the production of two distinct fragments with the ability to bind TGF-␤. Pervanadate activates the shedding of several cell surface proteins, some of them, such as HER4 (51), proamphiregulin (51), syndecan-1 (52), and amyloid-␤ protein precursor (53), are also shed in response to PMA, a well known activator of TACE. In contrast, the shedding of betaglycan (this study), HER 2 receptor (43), and TNF-related activation-induced cytokine (TRANCE) in CHO and fibroblast cells (54) can be stimulated by pervanadate but not by PMA, indicating that TACE is not involved in these shedding events.
Upon treatment with pervanadate, different cell lines produce different forms of soluble betaglycan. L6E9 myoblasts secrete a ϳ90-kDa fragment (sBG-90), COS-1 cells a ϳ120-kDa fragment (sBG-120), and CHO cells produce both of them (Fig.  9). The fact that the production of both of these forms is inhibited by the hydroxamate BB-94 suggests that betaglycan is the substrate of one or more zinc-dependent metalloproteases. The effect of the TIMPs has been used previously to characterize the identity of the metalloprotease(s) involved in a given shedding event. For example, TACE is inhibited by TIMP-3, but not by TIMP-2 or TIMP-1 (55), and the majority of the MT-MMPs are inhibited by TIMP-2, but not by TIMP-1 or TIMP-3 (37,38,(47)(48)(49)56). Our data demonstrate that the pervanadate-activated proteolytic activity that results in the production of the sBG-90 fragment in L6E9 cells is inhibited by TIMP-2 but not by TIMP-1. This result, along with that obtained with mutant cells defective in TACE activity (M1 cells), confirms that TACE is not involved in the pervanadate-stimulated release of betaglycan and argue in favor of members of MT-MMPs. To date six members of the subfamily of MT-MMP have been described (37)(38)(39)(47)(48)(49). It has been shown that among the six MT-MMPs, only MT4-MMP is not inhibited by TIMP-2 (37), and MT6-MMP is inhibited by TIMP-2 as well as TIMP-1 (56), ruling out MT4-and MT6-MMP as the pervanadate-stimulated sBG-90 sheddase. Accordingly, we observed that overexpression of these MT-MMPs did not increase sBG-90 production. On the other hand, overexpression of MT1-MMP and MT3-MMP in L6E9 cells significantly increased the basal production of sBG-90, and MT1-MMP significantly enhanced its pervanadate-stimulated release, indicating that these metalloproteases participate in the shedding of betaglycan. The experiment performed with COS-1 cells overexpressing MT1-and MT3-MMP strongly supports this conclusion. Parental COS-1 cells do not express MT1-or MT3-MMP and therefore constitute functional knock-outs for these proteases (Fig. 5). In agreement with the proposal that these MT-MMPs are involved in the production of sBG-90, COS-1 cells produce sBG 120 preferentially. Furthermore, upon transfection with MT1-and MT3-MMP, COS-1 acquires the ability to generate sBG-90 (Fig. 4B). The loss of the regulated release of sBG-90 when MT1-and MT3-MMP are transfected in L6E9 and COS-1 cells may be caused by the overexpression of these proteases, which may overcome the regulatory mechanisms that normally operate upon the endogenous enzymes. Therefore, these data allow us to conclude that MT1-and/or MT3-MMP has the ability to produce sBG-90 upon stimulation with pervanadate. However, the fact that L6E9 and CHO cells, which are able to shed sBG-90, lack expression of MT3-MMP, strongly suggests that, at least in these cell lines, the metalloprotease responsible for the generation of this form of soluble betaglycan is MT1-MMP (Fig. 9). Nonetheless, our transfection experiments indicate that cell lineages expressing MT3-MMP are also able to generate sBG-90.
In addition to its role in the breakdown of components of the extracellular matrix and activation of other MMPs, MT1-MMP is linked directly to the shedding of several unrelated membrane proteins (57). Overexpression of MT1-MMP by glioma and fibrosarcoma cells led to proteolytic cleavage of cell surface tissue transglutaminase (58). Soluble recombinant MT1-MMP is able to cleave pro-TNF-␣ to its mature form in vitro (59). Interestingly, one of the shedding activities described for TRANCE is similar to the one reported here. The pervanadatestimulated shedding of TRANCE is inhibited by TIMP-2, but FIG. 9. Schematic representation of the proteases involved in the shedding of betaglycan ectodomain identified in the present work. Membrane-bound betaglycan (BG) may be the substrate of different zinc metalloproteases leading to cleavage of its ectodomain and to release of different forms of the soluble receptor (sBG-90 and sBG-120). L6E9 cells constitutively shed sBG-90 and sBG-120 and upon stimulation with pervanadate (PV) increase their shedding of sBG-90, apparently by stimulation of their endogenous MT1-MMP (MT1). In COS-1 cells, which lack MT1-and MT3-MMP, pervanadate stimulates an unknown metalloprotease (?) that leads to the augmented shedding of sBG-120. In CHO cells, pervanadate, acting via endogenous MT1-MMP and unknown metalloprotease, augments the shedding of both sBG-90 and sBG-120, respectively. By an unknown mechanism, the ectopic expression of MT2-MMP (MT2) decreases betaglycan and MT1-MMP in COS-1 and CHO cells. not by TIMP-1, and MT1-MMP has been proposed as the metalloprotease responsible (54). It has been reported that the cell adhesion molecule CD44 (60) and syndecan-1 (61) are directly shed by MT1-MMP and MT3-MMP. In these studies it was also demonstrated that coexpression of the others members of MT-MMP was without effect. Thus, it is not unexpected that MT1-MMP and MT3-MMP may have similar substrates, as we have observed here for betaglycan.
Regarding the identity of the protease that mediates the production of sBG-120 all we can conclude at this moment is that it is different from those generating sBG-90. Two findings support this conclusion: its lack of inhibition by TIMP-1 and TIMP-2, and the fact that its activity is overcome when cells are transfected with MT1-or MT3-MMPs. Further studies will be necessary to determine which protease mediates the pervanadate-stimulated production of sBG-120 (Fig. 9).
The fact that betaglycan ectodomain is shed by different proteases is not unprecedented; several reports have demonstrated the possibility of different shedding activities capable of cleaving the same substrate. For example, TRANCE can be released by two different metalloproteases: one in fibroblasts and CHO cells, and another in COS-7 cells (54). The metalloprotease in fibroblasts and CHO cells is stimulated with pervanadate but not by PMA and inhibited by TIMP-2 but not TIMP-1, whereas the metalloprotease in COS-7 cells is refractory to pervanadate and is not inhibited by TIMP-1 orϪ2 (54). Pro-TNF-␣ is cleaved by different proteases in different cells and conditions. TACE has been described as the principal pro-TNF-␣ sheddase in many cell types (29,62,63), but MMP-7 (matrilysin) (64), ADAM 10 (65), and MT4-MMP (37) can also cleave pro-TNF-␣. On the other hand, the existence of a metalloprotease activity alternative to TACE with the ability to shed pro-TGF-␣ has been demonstrated recently (66).
Deletion mutagenesis analyses have revealed two TGF-␤ binding regions in the betaglycan ectodomain (17,21,22,67). One region is at the amino-terminal half (related to endoglin, Fig. 1A) and the other at the carboxyl-terminal half (related to uromodulin, Fig. 1A). The recently described inhibin A binding region resides at the carboxyl-terminal half of betaglycan ectodomain (17). Based on their apparent sizes, we could have foretold that both sBG-90 and sBG-120 would have TGF-␤ binding activity (which they do; Fig. 8 and data not shown) because they retain at least one entire ligand binding domain. The data presented in Fig. 8 indicate that although the total binding of sBG-90 is lower than that of recombinant soluble betaglycan, its relative TGF-␤ isoform affinities are comparable with those shown by the wild type receptor. This is not surprising because sBG-90 is predicted to contain the complete ligand binding site related to endoglin which, as we have shown before, has a TGF-␤ isoform preference similar to that of the wild type receptor (17). Some potential roles for soluble receptors include increasing the half-life of the ligand or acting as an inhibitor for ligand binding to the cell membrane-anchored receptor (26). Whether or not sBG-90 and/or sBG-120 has such functional roles or behave as soluble recombinant betaglycan, inhibiting TGF-␤ binding to cell surface receptors (21,22), remains to be determined experimentally.
A remarkable observation from the present studies is the effect of MT2-MMP on the levels of betaglycan and MT1-MMP (Fig. 9). Although at this moment the mechanism involved is unknown, the data shown in Fig. 6 demonstrate that MT2-MMP decreases the cellular content of betaglycan and MT1-MMP in a dose-dependent manner. This effect is not likely the result of a decreased level of expression of the plasmids encoding MT1-MMP or betaglycan in cells cotransfected with differ-ent combinations of plasmids because it is not observed in cells overexpressing MT1-, MT3-, MT4-, MT5-, or MT6-MMP. This notion is supported further by the fact that endogenous MT1-MMP is also decreased by overexpression of MT2-MMP. In addition to being relevant for the shedding of betaglycan, this effect of MT2-MMP may have wider consequences because it implies that MT2-MMP is capable of down-regulating MT1-MMP. To our knowledge, there are four reports that show concomitant expression of these two metalloproteases, in glioblastoma tissues (68), in pancreatic tumor cell lines (60), in primary cultures of smooth muscle cells from aorta (69), and in placenta tissues (70). Nonetheless, in view of the paramount importance that MT1-and MT2-MMPs have in the invasiveness, morphogenesis, and metastatic potential of epithelial cells (71,72), this observation deserves further investigation.