Cysteine Array Matrix Metalloproteinase (CA-MMP)/MMP-23 Is a Type II Transmembrane Matrix Metalloproteinase Regulated by a Single Cleavage for Both Secretion and Activation*

Matrix metalloproteinases characterized so far are either secreted or membrane anchored via a type I transmembrane domain or a glycosylphosphatidylinositol linkage. Lacking either membrane-anchoring mechanism, the newly discovered CA-MMP/MMP-23 was reported to be expressed as a cell-associated protein. In this report, we present evidence that CA-MMP is expressed as an integral membrane zymogen with an N-terminal signal anchor, and secreted as a fully processed mature enzyme. We further demonstrate that L20GAALSGLCLLSALALL36 is required for this unique membrane localization as a signal anchor and its secretion is regulated by a proprotein convertase motif RRRR79sandwiched between its pro- and catalytic domains. Thus, CA-MMP is a type II transmembrane MMP that can be regulated by a single proteolytic cleavage for both activation and secretion, establishing a novel paradigm for protein trafficking and processing within the secretory pathway.

Proteolysis mediated by metalloproteinases has been implicated in a diverse range of biological processes such as signal transduction (1), precursor processings (2,3), angiogenesis (4,5), and the turnover of extracellular matrix (ECM) 1 (5)(6)(7)(8). The matrix metalloproteinases (MMP) are a distinct family of related genes specializing in ECM remodeling by virtue of their abilities to degrade all ECM components under physiological conditions (8 -10). To achieve precision in ECM remodelings, the activities of MMPs are controlled at multiple steps from transcription, translation, zymogen activation, to inhibitions by tissue inhibitor of matrix metalloproteinases (TIMPs) (9 -12). Recently, localization and trafficking of MMPs have been recognized as important mechanisms of regulation, but remain poorly defined.
MMPs discovered so far can be classified into two categories based on cellular localization: secreted or membrane-bound. The secreted MMPs are generally synthesized and discharged to the extracellular milieu where they are activated, bind to, and degrade their ECM substrates (9,11). On the other hand, members of the MT-MMP subgroup are limited to the cellular membranes, mostly the plasma membrane via a type I transmembrane domain or glycosylphosphatidylinositol anchor (13)(14)(15)(16)(17)(18)(19)(20). Compared with their secreted counterparts, these membrane-bound MMPs may be able to express their activity with both precision and intensity against subjacent substrates, thus, exerting profound proteolytic activities in a wide range of biological and pathological processes such as cell migration, invasion, angiogenesis, and development (5,6,13,17,(21)(22)(23)(24)(25). The transmembrane domain of MT1-MMP has been shown to be critical for its ability to activate pro-MMP-2 and mediate cell invasion and migration (13,(22)(23)(24). More dramatically, MT1-MMP-deficient mice are the first to display developmental abnormalities, in contrast to the apparent normal phenotypes reported for those from secreted MMPs, underscoring the significance of the membrane-bound MMPs (6,25,26).
CA-MMP, a novel MMP identified recently (27), is apparently the mouse orthologue of human MMP-23 (28). Tissue distributions of this enzyme suggest that it may play an important role in the functions of lung, heart, as well as the reproductive organs such as ovary and prostate (27,28). Enzymatically, purified CA-MMP behaves like a classic MMP: detectable on zymography and inhibited by tissue inhibitor of matrix metalloproteinases (TIMPs) (27). Structurally, CA-MMP shares only two features with the rest of the MMP family, a conserved catalytic domain and a basic motif RRRR 79 (27,28). Otherwise, CA-MMP resembles little of a MMP. First, while all MMPs discovered so far are regulated by a latency mechanism based on a cysteine switch (29), CA-MMP may employ an entirely different mechanism for latency because there is no cysteine in its putative prodomain (27). Second, we suggested that CA-MMP differ from the other MMPs by having two novel domains downstream of its catalytic domain: a cysteine-array (CA) and an Ig-like domain (27), replacing the hinge and hemopexin-like domains implicated in interactions with substrates and intergrins (30,31). Experimentally, CA-MMP has been shown as a cell-associated protein with no detectable secretion in COS cells (27). Given its lack of a C-terminal transmembrane domain or glycosylphosphatidylinositol anchor found in the MT-MMPs, we hypothesize that CA-MMP may possess a novel mechanism for cellular localization. In this report, we present evidence that CA-MMP is localized as a type II transmembrane proteinase via a signal anchor. Furthermore, a specific cleavage at Arg 79 releases active CA-MMP into the extracellular milieu. The apparent coupling of secretion and activation for a membrane-anchored zymogen defines an unprecedented strategy to regulate proteolytic activity.

MATERIALS AND METHODS
Chemicals, Cell Lines, Cell Culture, and Immunological Reagents-All common laboratory chemicals and proteinase inhibitors were from Sigma. Restriction enzyme products were from Promega (Madison, WI) or Roche Molecular Biochemicals (Indianapolis, IN). COS, MDCK, and its derivatives were maintained as described (19). Cell culture reagents and fetal bovine sera were from Life Technologies (Rockville, MD). CA-MMP stable clone 30-11 was described previously (27). Anti-CA-MMP antisera were generated against a glutathione S-transferase-CA-MMP fusion protein in rabbit as described (32). Anti-FLAG M2 mono-clonal antibody and its agarose conjugates were purchased from Sigma.
Expression Constructs and Transfections-pCR3.1uni-CA-MMP was described previously (27). The GFP fusion construct was made by inserting the CA-MMP coding region as a polymerase chain reaction fragment in front of a modified GFP that has a A to T mutation in the first codon ATG (preventing potential internal initiation) as shown in Fig. 5A. Serial deletions of CA-MMP were generated by high fidelity polymerase chain reaction with pfu polymerase (Stratagene, CA) and cloned upstream of GFP as described above (see Fig. 6A). Liposomemediated DNA transfections into COS and MDCK cells were performed as described (19). Stable lines were selected in the presence of G418 (400 g/ml) and screened by fluorescent microscopy for GFP and/or Western blotting.
Cell Fractionations and Differential Extractions-These procedures were carried out essentially as described (33). Briefly, cells grown to 100% confluence were scrapped in PBS and washed two times with PBS, then treated with buffer A (10 mM Tris-HCl, pH 7.4, 10 mM KCl, 10 mM MgCl 2 , 1 M dithiothreitol, 1 mM phenylmethylsulfonyl fluoride, 1 g/ml leupeptin, 10 g/ml aproptin) on ice for 15 min. The cells were then homogenized and centrifuged at 1,600 ϫ g for 8 min at 4°C. The pellet were collected as crude nuclei (Nuc), and the supernatant was further centrifuged at 100,000 ϫ g for 30 min at 4°C to separate the cytosol (Cyt) from the membranes including ER, Golgi, and plasma membranes. The pellet fractions (either nuclei or membranes) were divided equally, and extracted with high salt (Buffer A, 350 mM NaCl), alkali/EDTA (pH 11.0/10 mM), or Triton X-100 (1% in TBS) for 1 h on ice (34,35). The resulting extraction mixtures were centrifuged at 100,000 ϫ g for 30 min at 4°C to pellet the remaining materials from the soluble mixtures. The pellets were dissolved with SDS-PAGE sample buffer.
Saponin Extraction of Cells-COS cells transiently transfected with  CA-MMP 1-20 GFP, CA-MMP 1-39 GFP, or CA-MMP 1-75 GFP were washed  with PBS (3 times), then extracted with saponin (500 l/well, 0.1% in 75 mM potassium acetate, 25 mM Hepes buffer, pH 7.2) for 30 min at 25°C (35). The extraction mixture were collected and centrifuged to separate the extracts from the cell pellets. The pellet were lysed with RIPA buffer for 15 min on ice.
Protein Purification, SDS-PAGE, Western Blot, and N-terminal Sequencing-All proteins were purified on anti-FLAG-M2 affinity columns as described (36). Briefly, serum-free conditioned media were collected after incubating with cells for 48 h. For cell lysates, cells were washed in ice-cold PBS (3 times) and extracted in TBS (50 mM Tris-HCl, pH 7.5, 150 mM NaCl) containing 1% Triton X-100 and proteinase inhibitors (1 mM phenylmethylsulfonyl fluoride, 1 g/ml leupeptin, 10 g/ml aproptin). The conditioned media or cell lysates were cleared of any debris by centrifugations, loaded onto an M2 column (1 ml of resuspended gel) pre-washed with TBS. The bound materials were washed with TBS and eluted competitively with FLAG peptide (2 molar excess) and collected in fractions. The fractions were analyzed by SDS-PAGE, Western blotting, and/or zymography as described (32,36,37). For N-terminal sequencing, protein samples were transblotted on PVDF membranes and submitted to the Michigan State University Macrochemical Facility for sequencing.
Immunoprecipitations and Glycosylation Analysis-N-Glycosylation in vivo was investigated by inhibition with tunicamycin (10 g/ml) during metabolic labeling. Cells from stable line 30-11 were washed in PBS (3 times) and allowed to incubate with Met/Cys-free RPMI (Life Technologies, Inc.) for 30 min. Then, fresh Met/Cys-free RPMI supplemented with [ 35 S]Met and Cys (100 Ci/ml, PerkinElmer Life Science) was added and allowed to incubate for 2 h in the presence or absence of tunicamycin (10 g/ml). The cells were lysed and analyzed by immunoprecipitation with anti-CA-MMP antisera as described (38). For enzymatic deglycosylation, purified CA-MMP (ϳ100 ng, see Fig. 3B) were reduced and denatured, and then treated with N-glycanase (5 units, Roche Molecular Biochemicals) for 20 h.
Confocal Microscopy-Confocal scans were performed at the University of Minnesota Biomedical Image Processing Laboratory on a MRC Bio-Rad system. The images were analyzed by Confocal Assistant and Adobe photoshop 5.0.

Structural
Features for CA-MMP-In addition to CA-MMP isolated from mouse (27), its orthologues have been isolated from human (28) and rat (accession number AB 010960). As shown in Fig. 1A, CA-MMP and its homologues are highly conserved throughout the molecule, especially the catalytic domain from Phe 80 -Gly 255 . In addition to the novel domains such as the CA and Ig-fold identified earlier (27), analysis of hydrophobic/hydrophilic profiles for all three homologues revealed a nearly identical N-terminal hydrophobic segment of 17 residues (Fig. 1B). Curiously, no cleavage sites for signal peptidases were identified. Instead, this hydrophobic segment may serve as a potential transmembrane (TM) anchor for CA-MMP. Thus, CA-MMP may be expressed as a novel type II or III transmembrane proteinase with a TM domain at its N terminus, in sharp contrast to a TM located at the C termini for the type I transmembrane MT-MMPs (Fig. 1C) (14,15,17,19).
CA-MMP as an Integral Membrane Protein-Consistent with the idea that CA-MMP contains an N-terminal transmembrane anchor, its product appears to be expressed as a cell-associated protein in all cell lines examined including COS, MDCK (27), T47D, and MCF-7 cells (data not shown). To further analyze the nature of its cell association, a stable line expressing CA-MMP (27) was analyzed further by cell fractionations coupled with differential extractions (Fig. 2A). The cell homogenates were separated into three distinct fractions: cytosol, membrane, and nuclei as illustrated in Fig. 2A. Most of the CA-MMP can be partitioned into the nuclei and membrane frac-  5), and Triton X-100 (lanes 6 and 7) as illustrated in A (see "Materials and Methods"). The extracts (i.e. supernatants in lanes 2, 4, and 6) as well as the remaining pellets (lanes 3, 5, and 7) were analyzed by Western blotting. Note that only Triton X-100 is effective in extracting CA-MMP from the membrane fractions (lane 6 versus lanes 2 and 4). D, the nuclei-associated CA-MMP is an integral membrane protein as well. Nuclear fractions from 30-11 (C, lanes 1, 3, 5, 7, 9, and 11)  tions, but not the cytosol (Fig. 2B, lanes 3 and 5 versus 1). Fractions from control transfected MDCK cells were all negative (Fig. 2B, lanes 2, 4, and 6). Next, we wished to determine whether CA-MMP is an integral membrane protein by performing extractions with high salt, alkali, or Triton X-100. As shown in Fig. 2C, CA-MMP can only be extracted by the nonionic detergent Triton X-100, but resistant to extractions with either high salt or alkali conditions (lanes 6 versus 2 and 4). These results demonstrate that CA-MMP is integrated into the membrane and strongly argue against it being a typical secretory protein (35,39). When similar extractions were performed for the nuclear fractions, an identical profile of extraction was observed, indicating that the CA-MMP in the nuclear fractions is also integrated into the membranes (Fig. 2D, lanes 9 versus 1 and 5). Thus, CA-MMP is not a typical secretory protein, rather, an integral membrane one with a unique intracellular distribution.

The 56-kDa CA-MMP Is Modified by Met Aminopeptidases but Retains the N-terminal Hydrophobic Segment as a Signal
Anchor-The cell fractionation and differential extraction experiments in Fig. 2 demonstrated that CA-MMP is a transmembrane protein, consistent with the prediction that it contains a potential transmembrane domain at its N terminus ( Fig. 1). In eukaryotic cells, proteins are targeted to the ER through the recognition of an N-terminal hydrophobic sequence in the nascent polypeptide chain by signal recognition peptides, which in turn interacts with the signal recognition peptides receptors on the rear endoplasmic reticulum (40,41). Simulta- FIG. 3. The N-terminal hydrophobic segment remains with CA-MMP as a signal anchor. A, signal peptide or signal anchor? An N terminus between residues 26 and 80 would argue that the hydrophobic segment is removed as a signal peptide (41). On the other hand, an N terminus within the first 19 residues would indicate that CA-MMP is a type II or III transmembrane protein with an N-terminal signal anchor. B, purification and N-terminal sequencing of cell-associated CA-MMP. Cell-associated CA-MMP was solublized in Tris-buffered saline with Triton X-100 (1%) and purified through an anti-FLAG affinity column. Materials throughout the purification were analyzed by SDS-PAGE (lanes 1-11) and Western blotting (lanes 12-22) as described (36). neously, the ribosome interacts with the Sec61p complex to form a channel through which the nascent polypeptide traverses the ER membrane (40,41). During the translocation process, a classic signal peptide would be removed by signal peptidases when the nascent polypeptide chain is 75-150 residues in length (39,41). Consequently, no classic signal pep-tides have been shown to remain with the fully synthesized proteins in vivo (41). In contrast, a signal anchor remains with the mature protein after completion of translation (35,41). To differentiate whether the N-terminal hydrophobic sequence in CA-MMP is a classic signal peptide or a signal anchor, we wished to determine whether it is cleaved or retained (Fig. 3A), by purifying the cell-associated CA-MMP and sequencing its N terminus. Shown in Fig. 3B is a representative purification from approximately 10 8 cells. The input, output, washes, and eluted fractions were analyzed by SDS-PAGE and detected by either R-250 staining or immunoblotting with anti-FLAG-M2 antibody (lanes 1-11 or 12-22, Fig. 3B). A 56-kDa species along with another ϳ120-kDa species was detected by R-250 staining (Fig. 3B, fractions 5-7). Both species reacted with M2 antibody confirming their identity as the fusion proteins and an intact C terminus with a FLAG tag (Fig. 3B, lanes 17-22). The higher molecular mass species may be either a multimeric form of the 56-kDa monomer or complex with other proteins. Since the blot was generated under denaturing and reducing conditions, this complex is apparently resistant to both denaturation and reduction, suggesting strong intermolecular interactions. To determine its N terminus, the 56-kDa species was sequenced for 10 cycles as described (37,38). As shown in Fig. 3B (bottom panel), an N terminus of G-RA-LRPEA was obtained with cycles 2 and 4 non-callable. This sequence matches perfectly with the predicted sequence 2 GCRACLRPEA of mouse CA-MMP/MMP-23 with positions 2 and 4 as Cys, a residue known to be non-callable by N-terminal sequencing. An N terminus starting at Gly 2 suggests that CA-MMP be processed by Met aminopeptidases to remove the translational initiation Met (42). Since the purified 56-kDa species can be detected by the Cells from a stable transfectant of CA-MMP-GFP were fixed, mounted, and scanned on a confocal system from Bio-Rad. One focal plane is presented to show signals at the ER-Golgi complex (arrow) and the nuclear membrane. One vertical scan is also presented at the lower panel with the marked positions for ER and nucleus (n, nuclei). Bar, 20 m. anti-FLAG-M2 antibody, it is a finished translational product, not an intermediate. Thus, the retention of 20 LGAALSGLCLL-SALALL 36 in mature CA-MMP excludes the possibility that it is a cleavable signal peptide (Fig. 3A).
CA-MMP Is Glycosylated at Multiple Asn Residues in the Secretory Pathway-For proteins with uncleaved signal anchors, they can assume either an N cyt /C exo (type II) or N exo /C cyt (type III) orientation on the membrane (35). Based on the charge rules, CA-MMP should have a type II orientation (43). Should CA-MMP be a type II transmembrane protein, we would predict that any or all of its 4 potential N-glycosylation sites be post-translationally modified in the lumen of the secretory pathway (Figs. 1 and 4A). On the other hand, there should be no N-glycosylation for CA-MMP if it assumes a type III orientation with the bulk of its coding region in the cytosol (Fig. 4A). Thus, N-glycosylation could provide evidence whether CA-MMP is a type II or III protein. We have taken two independent approaches to probe if CA-MMP is N-glycosylated. First, we performed metabolic labeling of CA-MMP expressing cells with [ 35 S]Met in the presence or absence of tunicamycin, a compound that inhibits N-type glycosylation (44). As shown in Fig. 4B, two CA-MMP species around 50 and 46 kDa were detected by immunoprecipitation from cultures treated with tunicamycin (lane 3), as compared with a 56-kDa one from untreated cells (lane 2), suggesting that N-glycosylation contribute 6 -10 kDa to the molecular mass of cell-associated CA-MMP. Second, we treated purified CA-MMP (ϳ100 ng) with N-glycanase (5 units). The 56-kDa CA-MMP species was converted into two species at 50 and 46 kDa as a result of deglycosylation (Fig. 4B, lanes 4 versus 5), confirming that CA-MMP is indeed N-glycosylated. The apparent molecular mass of 46 kDa is close to the estimated mass of 44.5 kDa for the CA-MMP open reading frame, with the remaining 1.5 kDa being contributed by the C-terminal FLAG tag. Therefore, the 46-kDa species may represent the fully deglycosylated form of CA-MMP. On the other hand, the 50-kDa species may contain additional post-translational modifications such as O-glycosylation, especially considering the fact that it was observed after both tunicamycin treatment and N-glycanase digestion (Fig. 4B). Further investigation is needed to ascertain the nature of the apparent increase of 4 kDa in mass for this minor species. Nonetheless, these data demonstrate that the 56-kDa CA-MMP is N-glycosylated, perhaps at multiple sites. Thus, CA-MMP should be considered as a type II protein (35). Fig. 2 suggest that a substantial amount of CA-MMP remains in the nuclear fraction. To further map the localization of CA-MMP, we have tagged a reporter gene, GFP, to the carboxyl end of CA-MMP and analyzed its cellular localization in both COS and MDCK cells (Fig. 5). As expected, a ϳ78-kDa fusion protein was detected in COS cells transiently transfected with pCR3.1uni-CA-MMP-GFP (Fig. 5B, lane 4). Confocal microscopic analysis of the same transfected COS cells revealed that CA-MMP-GFP is localized primarily in the ER/Golgi and the nuclear membrane (Data not shown). One of the concerns is that overexpression may alter the trafficking of a GFP-tagged protein. Therefore, we generated a panel of stable transfectants in MDCK cells with a range of expression levels. The localizations of CA-MMP in these transfectants are quite consistent over a wide range of expression levels. Shown as a representative in Fig. 5C, a moderately expressed CA-MMP-GFP localizes to the ER-Golgi complex (upper panel). Consistent with the cell fractionation experiments (Fig. 2B), a portion of CA-MMP is also associated with the nuclear mem-brane. A vertical scan of the same cell confirmed its presence on the nuclear membrane (Fig. 5C, lower panel). Since the nuclear membrane is practically a prominent component of the ER (45), it is reasonable to conclude that CA-MMP is primarily localization in the ER/Golgi apparatus.

CA-MMP Is Localized in the Nuclear-ER-Golgi Membrane Complex-Previous attempts by immunohistochemistry have localized CA-MMP intracellularly in 30-11 cells (27). Cell fractionations shown in
The N-terminal Signal Anchor Peptide Is Responsible for Targeting CA-MMP to the Membrane-Theoretically, a protein anchored by the signal anchor can be localized anywhere within the secretory pathway from the ER to the plasma membrane. The highly concentrated localization of CA-MMP in the ER/Golgi apparatus may be controlled by additional targeting signals throughout its coding region. To define the minimal sequence required for its localization, we have constructed a series of C-terminal deletions fused to GFP as illustrated in Fig. 6A (46). When expressed in COS cells, these mutants produced the expected proteins as judged by their apparent molecular masses (Fig. 6B). Consistent with the observed localization of the full-length fusion, all deletions behave similarly by localizing to the nuclear/ER/Golgi membrane network (data not shown). Stable clones were established and shown as representatives in Fig. 6, C and D. Deletions removing the catalytic/CA/Ig-like domains are capable of preserving an almost identical localization pattern (Fig. 6C). Results from these mutants argue that the signal responsible for the observed localization of CA-MMP in ER/Golgi network resides in the first 75 amino acids including the signal anchor peptide.
To directly test the role of the signal anchor peptide in CA-MMP localization, we constructed two additional deletion mutants within the first 75 residues (Fig. 6A, bottom). CA-MMP 1-39 GFP further removes the remaining portion of the prodomain, whereas CA-MMP 1-20 GFP deletes the signal an-  8 and 11, 12). B, the signal anchor is required for converting the reporter gene GFP into an integral membrane protein. COS cells transfected with CA-MMP 1-20 GFP (lanes 1 and 4), CA-MMP 1-39 GFP (lanes 2 and 5), and CA-MMP 1-75 GFP (lanes 3 and 6) were extracted with saponin (35). The extracts (lanes 1-3) and residual pellets (lanes 4 -6) were analyzed by Western blot as described above. Note that CA-MMP 1-20 GFP is extracted efficiently by saponin while the other two fusion proteins are resistant.
chor peptide. As shown in Fig. 6B, both constructs drive efficient expression of the fusion proteins in COS cells (Fig. 6B,  lanes 9 and 10). As expected, both CA-MMP 1-75 GFP and CA-MMP 1-39 GFP are localized to the nuclear/ER/Golgi network as the full-length fusion (Fig. 6D, panels B and C). Interestingly, CA-MMP 1-20 GFP is distributed throughout the cytosol as GFP does (Fig. 6D, panels A and F), suggesting that the signal anchor is responsible for targeting the reporter GFP to the nuclear/ER/Golgi network. To show that GFP does not have a dominant effect on the localization of its fusion partners, we have constructed and analyzed GFP fusions to MT3-MMP, a known plasma membrane protein; and MT6-MMP 1-280 , a secreted protein (18). MT3-MMP is a type I transmembrane MMP with a classic TM domain at its C terminus (15). When GFP is fused to its C terminus, the resulting protein is expressed stably (data not shown) and localized to the lateral membrane between two adjacent cells (Fig. 6D, panel D). 2 The fusion protein between GFP and TM-truncated MT6-MMP 1-280 is distributed in secretory vesicles throughout the cell body as reported (Fig. 6D, panel E) (18). Expressed alone, GFP is a cytosolic protein as expected (Fig. 6D, panel F) (46). The distinct localizations of these control proteins argue against the idea that GFP alters the distribution of its fusion partners in our experimental systems.
The subcellular localizations of these CA-MMP-GFP fusions were further confirmed by cell fractionations and extractions. As shown in Fig. 7A, CA-MMP 1-20 GFP is detected primarily in the cytosolic fraction while CA-MMP 1-39 GFP and CA-MMP 1-75GFP are membrane-associated, in agreement with the confocal analysis shown in Fig. 6D. The residual CA-MMP 1-20 GFP in the membrane and nuclear fractions (Fig. 7A, lanes 6 and 10) can be extracted with high salt and alkali while both CA-MMP 1-39 GFP and CA-MMP 1-75 GFP are resistant as expected (data not shown). To prove their distributions as either soluble or membrane proteins, we performed extraction with saponin, an agent known to extract soluble proteins from cells and cellular vesicles (35). As shown in Fig. 7B, CA-MMP 1-20 GFP was extracted efficiently from cells while CA-MMP 1-39 GFP and CA-MMP 1-75 GFP remain associated with the cells (lanes 1 and  4 versus lanes 2, 5 and 3, 6). Taken together, it is reasonable to conclude that 1) the first 39 residues are sufficient to target a reporter protein, GFP, to the ER/Golgi membrane network and 2) the signal anchor peptide is responsible for the observed localization of CA-MMP as a type II transmembrane protein.
CA-MMP Is Secreted as an N-terminal Processed Enzyme-So far, we demonstrated that the full-length CA-MMP is expressed as a type II transmembrane protein primarily localized in the nuclear/ER/Golgi networks via a signal anchor. Given the fact that degradation of ECM takes place extracellularly, CA-MMP should be at least displayed on cell surface or secreted into the intercellular spaces for physical contacts with the substrates. Although its signal anchor peptide is not cleaved off during translation, CA-MMP does possess a conserved motif (RRRR 79 ) for PCs (27,28,47). A specific cleavage at the carboxyl end of this motif by members of the PC family would not only activate CA-MMP by removing its prodomain, but also release the processed enzyme into the lumen of the secretory pathway for secretion. To test this hypothesis, we searched for secreted CA-MMP in the conditioned media from the stable cell line described in Fig. 2. Although little CA-MMP is detected from the culture media directly, further concentration (ϳ20-fold) allowed us to detect CA-MMP-specific products by Western blotting (Fig. 8A, lane 2). Since the secreted species (ϳ66 kDa) is significantly larger than the intracellular one at 56 kDa (Fig. 2, lane 2), we performed N-glycanase treatment to demonstrate that it is N-glycosylated (Fig. 8B, lanes 2 versus 1) as the intracellular CA-MMP (Fig. 4B). Interestingly, the deglycosylated forms of secreted CA-MMP remain larger than their intracellular counterparts (Fig. 4B), suggesting that additional post-translational modifications other than N-glycosy-2 T. Kang and D. Pei, unpublished data.  , lane 3). The upper band (ϳ400 ng) were excised from PVDF membrane and submitted for sequencing (see "Materials and Methods"). The resulting N terminus is YTLTPARLRX as shown on the right. D, schematic diagram of CA-MMP processing and activation. An N terminus starting at Tyr 80 suggests that CA-MMP is processed at the conserved PC motif RRRR 79 (downward arrow) to generate a secreted and active species with an intact C terminus (FLAG tagged). lations must have occurred intracellularly immediately prior to secretion or extracellularly after secretion. To define its molecular identity, we collected 1.5 liter of the conditioned media and performed affinity chromatography to purify the secreted CA-MMP. As shown in Fig. 8C, the eluted CA-MMP reacts with the M2 antibody by Western blotting (lane 1), exhibits gelatinolytic activity on zymography as described for CA-MMP (lane 2) (27), and can be detected by R-250 staining (lane 3). The larger species identified from the SDS-PAGE staining corresponds to the enzymatically active CA-MMP, whereas the smaller one does not, perhaps, as a truncated form (Fig. 8C, lane 3). Thus, we determined the N terminus of the upper band by sequencing. As shown in Fig. 8C, 10 cycles of protein sequencing of the upper species revealed an N terminus of YTLTPARLRX, which matches the CA-MMP sequence from Tyr 80 to Arg 88 , the predicted N terminus of its catalytic domain. Consistently, the secreted CA-MMP can cleave gelatin as substrates in solution, validating its status as an active species (data not shown). Thus, the secreted CA-MMP is a processed product of its intracellular zymogen by proteolytic cleavage at the carboxyl end of Arg 79 , perhaps via a PC-dependent mechanism as defined for MMP-11 (38). As illustrated in Fig. 8D, the proteolytic cleavage at Arg 79 not only releases CA-MMP into the media, but also converts its zymogen into an active species.
Regulation of Secretion by the RRRR Motif-Among proteolytic enzymes capable of cleaving at Arg residues, members of the PC family can specifically recognize a RX(K/R)R motif (47). Furin, the prototypic member of this family, has been demonstrated to process the zymogen of MMP-11 at a conserved RXKR motif (38). To prove that this motif is important for CA-MMP secretion and activation, we have constructed a substitution mutant converting the RRRR 79 motif into AAAA 79 . As expected, the 4R-4A mutant is retained in cells without any secretion (data not shown), suggesting that the RRRR 79 motif is required for CA-MMP secretion. To further define its role in regulating CA-MMP secretion, we analyzed two GFP fusion proteins described in Fig. 6A, CA-MMP 79 GFP and CA-MMP 75 GFP. These two constructs were designed to assess the role of the RRRR motif in regulating the secretion of its downstream fusion partner GFP. These two fusion proteins behave similarly both at the expression levels (Fig. 6B, lanes 2 and 3) and intracellular localizations (Fig. 6C, panels A and B). When the conditioned media from these two cells were analyzed for GFP secretion, only CA-MMP 79 GFP expresses a secreted species whereas CA-MMP 75 GFP does not (Fig. 9A, lanes 2 versus  1). Thus, the RRRR motif is responsible for releasing GFP into the media, apparently in agreement with the mutational data for full-length CA-MMP (data not shown). To ascertain the specificity of processing, we purified the secreted GFP species from the conditioned media of CA-MMP 79 GFP cells (Fig. 9B) and determined its N terminus as GLVSKGEELF, which matches perfectly with the deduced sequence from the fusion junction between CA-MMP 1-79 and GFP (Fig. 9D). The first residue glycine (G) comes from the right half of the SmaI site (CCCGGG). The second residue leucine (L) is a result of mutating the initiation codon, ATG, to TTG that codes for leucine. The remaining portion, VSKGEELF, corresponds to residues 2 to 9 of GFP. Thus, CA-MMP 79 recapitulates the full-length CA-MMP by preserving the cleavage specificity at Arg 79 .
Finally, we also determined the N terminus of intracellular CA-MMP 75 GFP by sequencing to see if it retains the hydrophobic sequence as a signal anchor similar to the full-length CA-MMP (Fig. 3). Shown in Fig. 9, we purified the cell-associated CA-MMP 75 GFP (Fig. 9C) and sequenced it N terminally. The N terminus is G-RA-LRPEA, matching residues 2 to 11 of CA-MMP as the full-length protein does (Figs. 9D and 3). There-fore, CA-MMP 75 also preserves the post-translational processing by Met aminopeptidases and retains the hydrophobic signal anchor just like the full-length CA-MMP (Fig. 3). Together, the signals regulating the trafficking, secretion, and activation of CA-MMP appear to reside in the first 79 residues including the signal anchor, L 20 GAALSGLCLLSALALL 36 , and the PC motif RRRR 79 . DISCUSSION Proteolysis is a basic mechanism of regulation in biology as exemplified by the caspase-mediated execution of cell death (48), proteosome-regulated progression of cell cycle (49), and antigen processings (50). The diversity and complexity of proteolysis may rival some of the better known biochemical processes such as phosphorylation/dephosphorylation. Irreversible and destructive in nature, proteolysis must be controlled tightly. With ECM components as substrates, members of the MMP family are encoded and synthesized as zymogens to be either secreted or displayed as surface proteinases via a type I transmembrane domain or glycosylphosphatidylinositol link, providing multiple steps for cells to regulate their activities  lanes 1 and 4) was passed through M2 column (lanes 2 and 5), eluted with FLAG peptides (lanes 3 and 6) as described (36). The materials were analyzed by either SDS-PAGE/Brilliant Blue R-250 staining (lanes 1-3) or Western blotting with M2 antibody (lanes 4 -6). The eluted GFP (ϳ500 ng) was transblotted onto PVDF membrane and sequenced N-terminally as described in the legend to Fig. 3 to yield an N terminus of GLVSK-GEELF (see D). C, purification and sequencing of cell-associated CA-MMP 1-75 GFP. Confluent culture of cells (10 ϫ 150-mm dishes) expressing CA-MMP 1-75 GFP were washed and lysed as described in the legend to Fig. 3. Lysates (lane 1) were purified through a M2 column to yield the eluted materials (lane 2) as detected by Western blotting using M2 antibody (lanes 1 and 2). The eluted protein (ϳ500 ng) was excised from a PVDF membrane and sequenced as described in the legend to Fig. 3. An N terminus of G-RA-LRPEA was obtained (see D). D, summary of the sequencing data from B and C. The CA-MMP 1-79 GFP mini-gene is presented schematically (lower portion) to illustrate the two regions sequenced N-terminally in B and C. Note that the fusion junction between CA-MMP 1-79 and GFP is cleaved (upward arrow) to yield the secreted GFP in A (lane 2).
under both physiological as well as pathological conditions (for reviews, see Refs. 9 and 10). Here we report the characterization of CA-MMP as a type II membrane-anchored zymogen that is proteolytically processed at a conserved motif and converted into a secretory and mature enzyme. This unique mechanism of localization and processing may allow cells to control the proteolytic activity of CA-MMP more efficiently in ECM remodeling events.
CA-MMP is the first MMP to be demonstrated as a type II transmembrane protein. Among ϳ25 MMPs discovered so far, they can be classified into two basic groups based on their cellular localizations. The first group is relatively large and made of secretory MMPs including collagenases, gelatinases, and stromeolysins (10). The second group is smaller and made of the 6 MT-MMPs (14)(15)(16)(17)(18)(19). Based on the membrane anchoring mechanisms, these MT-MMPs may be further divided into type I transmembrane MMPs for MT1-3, 5-MMPs, and the glycosylphosphatidylinositolanchored MMPs (MT4, 6-MMPs) (20). Thus, CA-MMP would establish the third subclass of membrane-bound MMPs as a novel type II transmembrane MMP. Without any precedence, we initially questioned the validity of the sequence-based prediction of CA-MMP being a type II transmembrane protein (Fig. 1). Although consistent with the finding that the majority of CA-MMP is cellassociated (27), CA-MMP could be associated with the cells by means other than being transmembrane-anchored. Thus, we devoted our effort to (i) verify that CA-MMP is an integral membrane protein by cell fractionation and differential extractions with high salt, alkali, and Trition X-100 (Fig. 2) (35,39), (ii) demonstrate by N-terminal sequencing that L 20 GAALSGLCLLSALALL 36 is retained with CA-MMP as a signal-anchor rather than being a cleavable signal peptide (41), and (iii) establish that the first 39 residues including the signal anchor is sufficient to localize a soluble reporter protein, GFP, to membrane and the signal anchor is required for the observed localizations. Taken together, these three independent lines of evidence argue strongly that CA-MMP is an integral membrane protein with L 20 GAALSGLCLLSALALL 36 as the signal anchor. As for its topology, CA-MMP was predicted to be a type II transmembrane protein based on the charge rules (35,43). The 19 residues that precede the signal anchor contain a net charge of ϩ2 for CA-MMP and its orthologues whereas the downstream sequences contain a Ϫ2 for rat and mouse and 0 for human. This charge distribution would favor the transfer of the carboxyl side into the lumen (the positive inside rule) (43). We proved this pre-diction by demonstrating that CA-MMP is N-glycosylated, a process unique for protein residing in the lumen of the secretory pathway (Fig. 4). Since the 4 likely N-glycosylation sites (Asn 93 , Asn 149 , Asn 233 , and Asn 317 , see Fig. 1A) are located downstream of the signal anchor, the carboxyl side must reside in the lumen in order to being N-glycosylated (Fig. 4). Together, this evidence argues that CA-MMP is a type II transmembrane protein.
The coupled secretion and activation of CA-MMP is regulated by a conserved PC motif RRRR 79 . Being a type II transmembrane zymogen, CA-MMP assumes a reverse topology in comparison to the MT-MMPs (14,15,17,19). The obligatory removal of its N-terminal prodomain would release the mature enzyme into the lumen and result in secretion. Sandwiched between its pro-and catalytic domains, there is a motif, RRRR 79 , which could be recognized by PCs (27,28). A similar RXKR motif has been demonstrated to be important for zymogen activation of MMP-11 through a process mediated by furin, a prototype PC (38). Thus, a single cleavage at Arg 79 could not only release CA-MMP from the membrane, but also remove its N-terminal prodomain. Indeed, we mapped the cleavage site for secreted CA-MMP at Arg 79 (Figs. 3 and 8), thus, confirming a coupled process for both secretion and activation. This is an interesting strategy because cells secrete only active CA-MMP, not latent ones like other secretory MMPs (9, 10). Accordingly, we speculate that CA-MMP functions in short distances from its origin of expression since its long distant efficacy would certainly be compromised by free tissue inhibitor of matrix metalloproteinases (TIMPs) in the extracellular milieu. Alternatively, the demand for CA-MMP activity is so acute that only an efficient and timely delivery of active species may meet the requirement. Both features would fit well with rapidity of the ovulation process where CA-MMP is expressed at relatively high levels (28). 3 The identification of its target substrates during the ovulation process will shed light on its proposed role in breaking the wall of ovarian capsules to release oocytes (28).
FIG. 10. A proposed model for the biosynthesis, processing, and secretion of CA-MMP. CA-MMP is synthesized in both the ER and its related outer nuclear membrane as a type II transmembrane protein. The majority of the protein is stored inside the lumen of ER and nuclear envelopes where it is modified post-translationally with N-and/or O-glycosylation. Trafficking to the Golgi networks allow it to be further modified as well as recognized by a putative PC at the conserved RRRR 79 motif (see illustrations in the two boxes). Subsequent cleavage at Arg 79 may take place in the trans-Golgi vesicles to release the soluble and active CA-MMP into secretory vesicles (S.V.) which mediate secretion by fusion with the plasma membrane (PM). the RX(K/R)R motif in these proproteins converts latent protein into active species, executing an irreversible step in biological regulations. In an evolutionary sense, the recruitment of a similar motif for the coupled secretion and activation of CA-MMP represents the first precedence with an unusual degree of degeneracy by converting a two-step process with implied flexibility to a single rigid one. The tradeoff may be efficiency and ease of control for dispensing the "right" amount of CA-MMP activity. It is not clear whether this strategy is advantageous over the standard processes of secretion mediated by signal peptidases and subsequent activation in the extracellular milieu for most of the secretory MMPs (10,12). Considering its divergent structure from the other MMPs, CA-MMP is certainly evolved with a dramatically different design (27,28). We can only speculate that such a unique design of domain structure warrants a coupled strategy for secretion and activation. Through a signal anchor, the cells retain latent CA-MMP within the cells, most likely in the ER network (Figs. 6 and 10), effectively eliminating any possibility of unwanted activation in the extracellular milieu. Since members of the PCs are known to be localized in the trans-Golgi network (47), trafficking of CA-MMP to the trans-Golgi complex would trigger the cleavage at Arg 79 and result in the release of active CA-MMP (Fig. 10). Thus, we predict that a key regulatory and rate-limiting step in controlling CA-MMP activity be the trafficking from ER to the trans-Golgi network (Fig. 10). Consequently, the observed low level of secretion in CA-MMP transfected MDCK cells may represent only the basal level of secretion in an overexpression system. It is possible that endocrine factors responsible for orchestrating the well executed turnover of ECM during ovulation may actually regulate the trafficking of CA-MMP from ER to the trans-Golgi network where the presumed cleavage by PCs takes place (Fig. 10). This attractive hypothesis is testable if cells from ovulating ovaries become available. Furthermore, the significance of the coupled secretion and activation could be tested in vivo by a "knock-in" approach to replace the signal anchor with a classic signal peptide. It would be interesting to see if such as a modified CA-MMP can substitute the wild type gene in vivo. Nonetheless, the discovery that CA-MMP/MMP-23 is a type II transmembrane proteinase with a coupled secretion and activation mechanism not only introduces a novel mechanism of regulation for MMP-mediated proteolysis but also establish an interesting paradigm for protein trafficking and processing within the secretory pathway.