Membrane Type 4 Matrix Metalloproteinase (MT4-MMP, MMP-17) Is a Glycosylphosphatidylinositol-anchored Proteinase*

Among the five membrane-type matrix metalloproteinases (MT-MMPs), MT1-, MT2-, MT3-, and MT5-MMPs have about a 20-amino acid cytoplasmic tail following the transmembrane domain. In contrast, a putative transmembrane domain of MT4-MMP locates at the very C-terminal end, and the expected cytoplasmic tail is very short or nonexistent. Such sequences often act as a glycosylphosphatidylinositol (GPI) anchoring signal rather than as a transmembrane domain. We thus examined the possibility that MT4-MMP is a GPI-anchored proteinase. Our results showed that [3H]ethanolamine, which can be incorporated into the GPI unit, specifically labeled the MT4-MMP C-terminal end in a sequence-dependent manner. In addition, phosphatidylinositol-specific phospholipase C treatment released the MT4-MMP from the surface of transfected cells. These results indicate that MT4-MMP is the first GPI-anchored proteinase in the MMP family. During cultivation of the transfected cells, MT4-MMP appeared to be shed from the cell surface by the action of an endogenous metalloproteinase. GPI anchoring of MT4-MMP on the cell surface indicates a unique biological function and character for this proteinase.

Matrix metalloproteinases (MMPs), 1 also called matrixins, are a family of Zn 2ϩ -dependent metalloendopeptidases that is involved in the degradation of extracellular matrix (ECM) in physiological and pathological conditions (1,2). To date, 20 mammalian MMPs have been identified by cDNA cloning (3)(4)(5)(6), and they can be subgrouped into 15 soluble-type and 5 membrane-type MMPs (MT-MMPs). With the exception of stromelysin 3 (MMP-11), soluble-type MMPs are secreted as zymogens and can act at distant sites following activation (3). Thus, they can participate in the degradation of a broad area of ECM, for example during tissue resorption. On the other hand, MT-MMPs have a hydrophobic amino acid stretch that is embedded in the plasma membrane at the C terminus and expressed on the cell surface (6 -10). Thus, MT-MMPs are thought to be responsible for the degradation of ECM at the cell periphery, and this may modulate various cellular functions in tissue such as proliferation, apoptosis, differentiation, and migration, etc. (11).
Cell surface proteins attach to the plasma membrane in at least two different ways, via transmembrane domains that are comprised of a stretch of hydrophobic amino acids or through a glycosylphosphatidylinositol (GPI) anchor (12). The former includes cytokine receptors, G-proteins, and integrins. Because they have signaling domains on both sides (11), these types of proteins act as interfaces connecting the outside and inside of the cells. In the case of GPI-anchored proteins, attachment to the plasma membrane is via a GPI molecule that contains 2-3 fatty acids. The protein is synthesized as a precursor having a 15-20 hydrophobic amino acid stretch at the C terminus (GPI signal). The hydrophobic amino acid stretch is cleaved off in the endoplasmic reticulum lumen, and the ectodomain is transferred to the GPI moiety (12). Thus, the mature GPI-anchored protein has no transmembrane or intracellular domains. Examples of such proteins include the urokinase receptor (uPAR) (13) and neural cell adhesion molecule (12,14).
The MT4-MMP (MMP-17) gene was originally cloned from a cDNA library derived from breast carcinoma cells (15). Among the MT-MMPs, MT4-MMP is unique in the following criteria. First, it is distantly related in the amino acid sequence to the other members (less than 40% sequence identity in the catalytic domain); the sequence identity among other members is more than 65%. Second, a putative transmembrane domain locates at the end of the C terminus. Thus, MT4-MMP lacks a cytoplasmic tail, whereas all the other MT-MMPs have a short cytoplasmic tail (20 -22 amino acids) (6 -9). Its C-terminal sequence suggests that MT4-MMP is a GPI-anchored enzyme. However, its characterization has been hampered to date because the reported cDNA (15) does not direct expression of the enzyme in transfected cells (16). A comparison of the reported cDNA sequences for MT-MMPs revealed that the cDNA for MT4-MMP does not have the region encoding a signal peptide common to all MMPs. On the other hand, we recently isolated cDNAs derived from new transcripts for human and mouse MT4-MMP that have a signal peptide coding sequence (16,17). Both human and mouse cDNAs for MT4-MMP can direct expression of the gene products when the cDNAs are expressed in the cells. Thus, newly obtained MT4-MMP cDNAs enabled us to examine whether MT4-MMP is a GPI-anchored protein.
In this report, we demonstrated that both the human and mouse MT4-MMP proteins are expressed on the cell surface of transfected cells as a GPI-anchored protein.
Construction of FLAG-tagged Human and Mouse MT4-MMP, hMT4/ MT1 TMCP , and Catalytically Inactive mMT4-MMP (E248A)-Fulllength mouse (GenBank TM accession number AB021224) and human MT4-MMP cDNAs (GenBank TM accession number AB021225) were subcloned into pSG5 expression vector (Stratagene, La Jolla, CA). To detect MT4-MMP protein by anti-FLAG M1 and M2 antibody, FLAG epitope (Asp-Tyr-Lys-Asp-Asp-Asp-Asp-Lys)-tagged MT4-MMP expression constructs were generated. The FLAG epitope was inserted either immediately downstream of the furin cleavage motif (following Arg 124 for mouse and Arg 128 for human) (MT4-F1) or one amino acid downstream of the motif (following Gln 125 for mouse and Gln 129 for human) (MT4-F2) (see Fig. 1). Anti-FLAG M1 antibody recognizes the FLAG epitope only when it is located at the N-terminal end of the protein in the presence of 1 mM Ca 2ϩ . Thus, M1 antibody is able to recognize the epitope only when MT4-F1 is processed at the furin cleavage motif. On the other hand, anti-FLAG M2 antibody recognizes the FLAG epitope at the N terminus or anywhere else in the molecule. Thus, it recognizes both MT4-F1 and MT4-F2 even without cleavage by a furin-like enzyme.
Construction of the Expression Vector for Human uPAR-The fulllength human uPAR cDNA clone was obtained from ATCC (ATCC number 65768). The insert was excised with EcoRI and XbaI and subcloned into the pSG5 expression vector.
Generation of Polyclonal Antibody against Mouse MT4-MMP Hemopexin-like Domain-Polyclonal anti-mouse MT4-MMP hemopexinlike domain (anti-MT4PEX) antibody was raised in a rabbit by injecting Escherichia coli-expressed mouse MT4-MMP hemopexin-like domain (Met-Asn 321 -Gly 550 ) using pET3a vector. The protein was purified and folded according to the method of Huang et al. (19). The antibody recognizes human and mouse MT4-MMP equally in Western blotting analyses. It is also specific for MT4-MMP and does not recognizes MT1-, -2, -3, or -5-MMPs (data not shown).
Cell Culture and Transfection-COS1 or Chinese hamster ovary cells (CHO-K1) were cultured in Dulbecco's modified Eagle's medium or Ham's F-12 medium supplemented with 10% fetal bovine serum and kanamycin, respectively, in a humidified incubator at 37°C. 16 h before the transfection was performed, cells were reseeded in 6-well plates at 2 ϫ 10 5 /well. Expression vectors for each protein were transfected using FuGENE6 TM according to the manufacturer's instructions. After 48 h, the cells were harvested and analyzed.
Western Blotting-To detect the protein in the culture supernatant, the protein in the medium was concentrated by treatment with 10% trichloroacetic acid. Cell lysate or trichloroacetic acid-concentrated protein samples were separated by SDS-polyacrylamide gel electrophoresis, and the proteins in the gel were transferred to a nitrocellulose membrane (Hybond-ECL, Amersham Pharmacia Biotech). After blocking with 10% fat-free dry milk in Tris-buffered saline (20 mM Tris-HCl, pH 7.5, 150 mM NaCl), the membrane was probed with anti-FLAG M2 monoclonal antibody (3 g/ml) to detect FLAG-tagged MT-MMPs or anti-MT4PEX serum (1:500) to detect MT4-MMP.
[ 3 H]Ethanolamine Incorporation, Immunoprecipitation, and Autoradiography-COS1 cells in 6-well plates were transfected with expression plasmids for FLAG-tagged MT4-MMPs and MT1-MMP. After 24 h, [ 3 H]ethan-1-ol-2-amine (Amersham Pharmacia Biotech) was added to the culture medium (100 Ci/ml), and the cells were cultured for another 24 h. Cells were washed once with phosphate-buffered saline and lysed in the RIPA buffer (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 10 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 M E-64, 0.02% NaN 3 ) at room temperature. Undissolved materials were spun down at 15,000 rpm in the Eppendorf tubes, and the supernatant was reacted with anti-FLAG M2-conjugated beads at room temperature for 1 h. The beads were washed with RIPA buffer three times, and materials bound to the beads were eluted with SDS-polyacrylamide gel electrophoresis loading buffer that contains 2-mercaptoethanol. The samples were subjected to SDSpolyacrylamide gel electrophoresis, and 3 H-labeled materials were detected on x-ray film after reaction of the gel with EN 3 HANCE (NEN Life Science Products). The FLAG-tagged MT4-MMP, chimera protein, and MT1-MMP in the samples were visualized by Western blotting using anti-FLAG M2 antibody.
Indirect Immunofluorescence Staining-CHO-K1 cells transfected with expression plasmids for the FLAG-tagged MT4-MMPs were reseeded on coverslips and cultured for 24 h. Cells were then fixed with 3% paraformaldehyde in phosphate-buffered saline. After blocking with 5% goat serum and 3% bovine serum albumin in Tris-buffered saline for 1 h at room temperature, cells were reacted with anti-FLAG M1/M2 antibodies (10 or 3 g/ml, respectively), polyclonal rabbit anti-MT4PEX serum (1:200), and polyclonal rabbit anti-human uPAR IgG (5 g/ml) at room temperature for 2 h. CaCl 2 (1 mM) was included throughout the procedure of washing and, in the case of the M1 antibody, incubation . Cy3-conjugated goat anti-mouse IgG and Alexa488-conjugated goat anti-rabbit IgG were used to visualize the antigen signal. The signals were analyzed by confocal microscopy (Bio-Rad).

Structural Characteristics of the C-terminal Region of MT4-MMP-
The amino acid alignment of mouse and human MT4-MMP with other MT-MMPs shows distinct differences in hydrophobicity at the C terminus (Fig. 1A), and this hydrophobicity pattern at the C-terminal end is rather similar to that of GPI-anchored proteins such as human Thy-1, human NCAM120, and human uPAR (Fig. 1A). Therefore, we have addressed the possibility that MT4-MMP is expressed on the cell surface as a GPI-anchored protein. The synthesis of the GPI membrane anchor unit includes the incorporation of ethanolamine for bridging the C-terminal amino acid and the glycan structure (see Fig. 1B) (12). Thus, GPI-anchored proteins can be labeled metabolically with [ 3 H]ethanolamine (12).  Fig. 2A, WB). As shown in Fig. 2A, [ 3 H]ethanolamine was specifically incorporated into both mMT4-F1 and hMT4-F1 but not into hMT4-F1/MT1 TMCP or hMT1-F1. The labeled band corresponds to one of the two bands (67-kDa species) for both mouse and human MT4-F1 detected by anti-FLAG M2 antibody ( Fig. 2A, WB, lanes 1 and 2). The incorporation of [ 3 H]ethanolamine into hMT4-MMP is C-terminal sequencedependent, because the chimeric hMT4-F1/MT1 TMCP was not labeled at all (lane 3). Thus, MT4-MMP was labeled with the [ 3 H]ethanolamine in association with a sequencespecific modification in the C-terminal region most likely by the GPI moiety. Therefore, the upper band detected on the Western blot in MT4-MMPs (68 kDa in mMT4-F1 and 71 kDa in hMT4-F1) that is not labeled with [ 3 H]ethanolamine is likely to correspond to the precursors for GPI anchoring, and the 67-kDa species is the GPI-anchored protein.

Incorporation of [ 3 H]Ethanolamine into
PI-PLC Treatment Releases MT4-MMP from the Cell Surface-Another indication of a GPI-anchored protein is the sensitivity for bacterial PI-PLC that cleaves the GPI moiety and releases the protein from the cell surface (see Fig. 1B) (12). Thus, the transfected COS-1 cells were treated with PI-PLC. As shown in Fig. 3A, although some levels of mMT4-F1 and hMT4-F1 were spontaneously released into the supernatant during incubation of the cells (Sup, lanes 1 and 3), PI-PLC  2 and 4). Upon treatment of the cells with PI-PLC, only slight decreases of MT4-MMPs were observed in the cell fraction (Cell, lanes 2 and 4). This is most likely because the intracellular MT4-MMPs remained to be transported to the plasma membrane surface because hMT4-F1 located at the cell surface was completely removed by PI-PLC treatment (Fig. 3B). In contrast, the type I transmembrane proteinase MT1-MMP was insensitive to PI-PLC treatment (Fig. 3A, lanes 5 and 6, and Fig. 3B). Taken together, we conclude that both mouse and human MT4-MMPs are GPI-anchored proteinases, and the GPI anchor signal is located in the C-terminal portion of the molecule, similar to other GPI-anchored proteins.
Shedding of MT4-MMP from the Cell Surface-During cultivation of the transfected COS-1, MT4-MMP appeared to be shed into the culture supernatant (Fig. 4A, lane 1). To test the possibility that proteinases release MT4-MMP from the cell surface, various proteinase inhibitors were tested for the inhibition of the shedding. The shedding of mMT4-MMP was not affected by mutation of the active site (Glu 248 to Ala) (Fig. 4A,  lane 7), co-transfection of TIMP-1 (lane 2) or TIMP-2 (lane 3), the presence of serine protease inhibitor (Pefabloc, lane 4), or the cysteine protease inhibitor (E-64, lane 6). The suppression was observed only in the presence of the hydroxamate MMP inhibitor, BB94, at 1 M (lane 5). The inhibition of the shedding by BB94 was also concentration-dependent (Fig. 4B), but the highest concentration (50 M) did not inhibit it completely. Thus, there may be another mechanism to shed the enzyme from the cell surface. Nonetheless, the data suggest that the shedding is at least in part due to a metalloproteinase.
Cell Surface Localization of the Active form of MT4-MMP-We next examined the localization of MT4-MMP in a Chinese hamster ovary cell line (CHO-K1) that is more suitable than COS-1 cells for analyzing subcellular localization. CHO-K1 cells transfected with the hMT4-F1 expression plasmid were co-immunostained with anti-FLAG M2 antibody and polyclonal rabbit anti-MT4PEX antibody without permeabilization. As shown in Fig. 5, MT4-F1 expressed on the cell surface was detected by M2 antibody (F1/M2). A similar result was obtained with anti-MT4PEX (F1/PEX). To confirm that the cell surface of MT4-MMP includes the mature active form that is processed at the furin motif, the cells were also stained with anti-FLAG M1 antibody that specifically recognizes the FLAG epitope located only at the N terminus of the protein. A similar staining pattern to that obtained with M2 or anti-MT4PEX antibody was obtained with the M1 antibody (F1/M1). To confirm the specificity of the M1 antibody, expression vector for another FLAG-tagged hMT4-MMP (hMT4-F2), which has the FLAG insertion one amino acid downstream of hMT4-F1, was transfected. The cells expressing hMT4-F2 showed a signal with M2 (F2/M2) or anti-MT4PEX antibody (F2/PEX) but not with M1 (F2/M1). This suggests that the signal detected in hMT4-F1-expressing cells with M1 antibody is the specific signal-detecting the furin cleavage site processed form. Unfortunately, the M1 antibody was ineffective for Western blotting to confirm the size of the processed species (data not shown).
Co-localization of MT4-MMP with uPAR on the Cell Surface-GPI-anchored proteins are thought to exist in microdomains formed by lipid rafts in the cell membrane that are rich in cholesterol and glycosphingolipids (20 -22). Therefore, we tested whether MT4-MMP can exist in the same microdomains with another GPI-anchored protein, uPAR (Fig. 6). CHO-K1 cells were co-transfected with hMT4-F1 or hMT1-F1 and uPAR expression vector and stained with anti-FLAG M1 (M1) and anti-human uPAR (huPAR) antibody (uPAR). As shown in Fig.  6, the hMT4-F1 (hMT4) and uPAR signals both showed a mixture of small and large dotlike staining patterns. When these images were merged, it became evident that MT4-F1 and uPAR signals co-localize in the bigger dots, showing a yellow color (Merge). Small dotlike signals on the cell body did not co-localize, as red (hMT4-F1) and green (uPAR) signals can be seen independently. On the other hand, the signal of MT1-MMP does not co-localize well with the major signal of huPAR, shown by numerous independent localizations of green and red signals (Fig. 6). The data suggest that these two nonrelated GPI-anchored proteins are likely to be present in the same microdomain on the cell surface. DISCUSSION In this paper, we have presented evidence that MT4-MMP is a GPI-anchored proteinase. The other four MT-MMPs are type I transmembrane proteins with a short cytoplasmic tail comprised of about 20 -22 amino acids (Fig. 1). Thus, MT4-MMP is unique in the MT-MMP subfamily not only in amino acid sequence homology but also in the method used for anchoring to the plasma membrane. Recent studies indicate that pericellular ECM degradation is regulated not only by the bulk levels of proteinases on the cell surface but also by localization of proteinases to the specific site on the cell surface to increase local concentration of the enzyme. In the case of MT1-MMP, swapping the transmembrane/cytoplasmic domain with that in the interleukin-2 receptor ␣ chain abolished the localization of the enzyme to the invadopodium structure of melanoma cells (23), suggesting that the transmembrane/cytoplasmic domain contains the signal to locate MT1-MMP to the specific site on the cell surface. On the other hand, MT4-MMP is tethered to the plasma membrane through the lipid and was expected to scatter uniformly on the cell surface because it lacks a cytosolic apparatus that may regulate localization. However, MT4-MMP-expressing CHO-K1 cells showed a dotlike staining pattern on their surface. Recent studies showed that GPI-anchored proteins are concentrated into the microdomains, or lipid rafts, on the living cell surface that are enriched for cholesterol and glycosphingolipids (21,22). Because microdomain-enriched GPI-anchored proteins can form clusters comprised of at least 15 molecules (21), some of the immunostained dotlike clusters may correspond to such microdomains. The enrichment of proteins in microdomains was shown not to be ectodomain-dependent but rather to be GPI moiety-dependent (21). Supporting this notion, our data also showed that hMT4-MMP co-localized with a nonrelated GPI-anchored protein, uPAR, on the cell surface of CHO-K1 cells. This also indicates that the cellular localization of MT4-MMP may be regulated in its GPI moiety-dependent manner. The GPI-anchored uPAR has previously been shown to associate with ␤ 2 integrin (CD18), and this interaction was shown to be important for leukocyte adhesion to the endothelial cells (24), local fibrin degradation, (25) and cell migration (26). Because MT4-MMP co-localizes with uPAR in the same microdomain clusters, these different types of proteinase systems (uPA/uPAR and MT4-MMP) may co-operate in conjunction with cell adhesion molecules such as CD18.
All the MT-MMPs, including MT4-MMP and many ADAM family members, have a furin cleavage motif at the end of the propeptide (7-9, 15, 27). Thus, they are thought to be activated intracellulary by furin or related proteinases and expressed on the cell surface as an active form. This feature is particularly important and effective for membrane-bound proteinases because membrane proteinase activity is likely to be associated with cell behavior and function (11). The active enzyme appearing on the cell surface can be used immediately or removed from the site by shedding when it is not needed. The MT4-F1 on the cell surface was detected as an active form using M1 antibody (Fig. 5). Because we could not quantify the pro and active forms of MT4-MMP on the cell surface, it is still possible that some fraction of pro MT4-MMP is transported to the cell surface without processing. In relation to this, it is of interest to know whether the 67-kDa GPI-anchored MT4-F1 represented in Fig. 2 retained the propeptide. However, the poor reactivity of M1 antibody in Western blotting prevented us from answering this question.
Several membrane proteins are known to be cleared from the cell surface by shedding. Shedding of the surface proteins, such as tumor necrosis factor ␣ receptor (28 -30), Fas ligand (31), and CD44 (32), is mediated by metalloproteinases as it is inhibited by broad spectrum hydroxamate metalloproteinase inhibitors such as BB94. Similarly, the shedding of MT4-MMP is also at least in part due to metalloproteinases that are insensitive to TIMPs. The shedding of MT4-MMP is not specific to COS1 cells because transfected CHO-K1 cells also shed the enzyme (data not shown). The shedding was not due to autocleavage of the MT4-MMP because an active site mutation (Glu 248 to Ala) of the enzyme (33) did not affect it at all (Fig.  4B). Therefore, it is likely that some other metalloendopeptidases, possibly ADAM family members, might be responsible for this shedding.
One GPI-anchored proteinase, membrane dipeptidase, was shown to be released from the cell surface by phospholipase C (34), and the shedding was induced by insulin stimulation in 3T3-L1 adipocytes. This shedding is not likely to be ectodomaindependent but rather GPI moiety-dependent, because cleavage occurred in the GPI moiety (34). Therefore, it is interesting to speculate that other GPI-anchored proteins including MT4-MMP may also be released from the cell surface in response to certain stimuli that activate the responsive phospholipase for shedding in vivo.
In conclusion, we have shown that MT4-MMP is a GPIanchored protein. This anchoring method is unique and thus indicates a unique biological function and character of MT4-MMP as a GPI-anchored proteinase.