Intracellular processing and activation of membrane type 1 matrix metalloprotease depends on its partitioning into lipid domains

Degradation of the extracellular matrix (ECM) is a critical process during cell invasion in both physiological and pathological processes such as morphogenesis, differentiation, cell migration, apoptosis, tumor invasion and neo-angiogenesis (reviewed in Basbaum and Werb, 1996). These events require the direct participation of released and exposed proteases including matrix metalloproteases (MMPs); MMPs are a family of related zinc-dependent proteases whose principal function appears to be the breakdown of extracellular matrix (ECM) proteins during tissue remodeling, growth, development, repair and cell invasion, physiological and pathological alike.

A considerable body of data now indicates that the expression and regulation of MMP activity is abnormal in many common forms of human malignancy; in fact, several MMPs are expressed in cancers at higher than normal levels (Edwards and Murphy, 1998; Egeblad and Werb, 2002) and, based on their characteristics, are believed to mediate the events leading to tumor cell metastasis. More than 25 members of this family have been identified that, as a group, degrade virtually all components of the ECM (Egeblad and Werb, 2002; Seiki, 2003). The MMP family can be broadly divided into soluble or membrane types. The latter are the minority and are membrane-bound proteins featuring a single type I transmembrane domain and a short cytoplasmic stretch (MMP14, MMP15, MMP16 and MMP24, or MT1-MMP, MT2-MMP, MT3-MMP and MT5-MMP), a glycosphingolipid anchor (MMP17 and MMP25, or MT4-MMP and MT6-MMP) or, as an exception, a type II transmembrane domain (MMP23).

MT1-MMP plays a pivotal role in a number of physiological and pathological processes via mechanisms that go beyond the degradation of ECM components (Seiki, 2003; Sounni et al., 2003). MT1-MMP displays proteolytic activity not only against a number of ECM components (d'Ortho et al., 1997) but also against other cell surface proteins (Belkin et al., 2001; Deryugina et al., 2002; Kajita et al., 2001) and has critical functions in skeletal embryogenesis (Holmbeck et al., 1999; Zhou et al., 2000) and cell migration (Hotary et al., 2000; Kajita et al., 2001). In some cases MT1-MMP might function irrespective of its proteolytic activity and might be involved in signal transduction events (Mu et al., 2002) triggered by cell interaction or other factors such as ligand binding.

A recent exciting development has been the finding that MT1-MMP confers tumor cells with an evident three-dimensional growth advantage in vitro and in vivo (Hotary et al., 2003). This requires pericellular proteolysis of the ECM. These observations emphasize the unique role of MT1-MMP in metastatis and reveal its function as a tumor-derived growth factor that regulates proliferation by controlling cell geometry within the confines of the three-dimensional ECM.

Similarly to most MMPs, MT1-MMP is synthesized as a pro-enzyme. It is generally thought that the propeptide needs to be removed for the protein to function properly and that this occurs through cleavage at a short amino-acid stretch at the C-terminus of the propeptide by the action of a prohormone convertase [furin or similar (Sato et al., 1996; Yana and Weiss, 2000)], although it has been proposed that MT1-MMP does not require the cleavage of its propeptide to activate MMP2 in certain cell types (Cao et al., 1996). In addition, activation of MT1-MMP could also occur by plasmin at the cell surface (Okumura et al., 1997). The subcellular locations where MT1-MMP is activated by furin are still uncertain; however, it could occur intracellularly before exit from the Golgi apparatus in the trans-Golgi network (TGN), similarly to MT3-MMP (Kang et al., 2002), or at the cell surface (Mayer et al., 2003). A further layer of complexity derives from the fact that mechanisms as diverse as autocatalytic processing, ectodomain shedding, homodimerization and internalization might all contribute to the modulation of MT1-MMP activity on the cell surface (reviewed by Osenkowski et al., 2004). Most recently it was shown that O-glycosylation might also contribute to the regulation of MT1-MMP substrate targeting (Wu et al., 2004). MT1-MMP has also been found to be partially located in detergent-resistant membrane fractions (DRM) (Annabi et al., 2001). Recently, new results have been published defining the role of DRM in MT1-MMP/CD44/caveolin interaction and hyaluronan cell surface binding (Annabi et al., 2004). Hence, MT1-MMP membrane partitioning could represent a mechanism for the regulation of its activities. At variance with these findings, it has been recently proposed that only a deletion mutant of MT1-MMP lacking the cytosolic C-terminal tail is partitioned into DRM (Rozanov et al., 2004).

Here, we investigate MT1-MMP intracellular processing and activation. We find that the association of MT1-MMP with different membrane subdomains establishes divergent metabolic fates and might be crucial in the control of its different activities. In particular, we show that, although the majority (80%) of MT1-MMP is sorted to detergent-resistant membrane fractions, it is only the minor (20%) detergent-soluble fraction of MT1-MMP, that undergoes intracellular processing to the mature form. Also, this processed MT1-MMP is exclusively responsible for ECM degradation in vitro. Finally, we show that furin-dependent processing of MT1-MMP occurs intracellularly after exit from the Golgi apparatus prior to arrival at the plasma membrane and provide an immunoelectron microscopy analysis of the potential MT1-MMP processing compartment.

All chemical reagents, unless otherwise stated, were purchased from Sigma Chemical Co. (St Louis, MO).

A375MM cells (hereafter referred to as A375 cells), an established human melanoma cell line (Kozlowski et al., 1984), were maintained in DMEM/F12 (1:1) nutrient mixture (Life Technologies, Rockville, MD). Medium was supplemented with 10% fetal calf serum, 4 mM L-glutamine, 100 U/ml penicillin and 100 μg/ml streptomycin. For transfection, cells were plated at 50% confluence, and the next day were incubated for 1 hour at 37°C with 0.45 μg/cm² DNA of interest and 2 μl/μg DNA TransFast (Promega, Madison, WI). Complete medium was then added. Experiments involving overexpression of exogenous proteins were usually performed the day after transfection.

MT1-MMP cDNA encoding an open reading frame from amino acid residues Met¹-Val⁵⁸³ and GFP-tagged-MT1-MMP were cloned in a pcDNA3.0 expression vector (Life Technologies) (Cao et al., 1996). FLAG-tagged α1-PDX cDNA (see Jean et al., 1998) was kindly provided by G. Thomas (Oregon Health Sciences University, Portland, OR) and subcloned in pcDNA3.0. GFP-tagged furin cDNA was kindly provided by J. Bonifacino (NIH, Bethesda, MD). The dominant-negative dynamin 2 mutant Dyn2^K44A and the ts045 temperature-sensitive Vesicular Stomatitis Virus G glycoprotein mutant (VSV-G)-GFP chimera were generous gifts of Drs M. McNiven (Mayo Clinic, Rochester, MN) and J. Lippincott-Schwartz (NIH, Bethesda, MD), respectively.

In all experiments and unless indicated otherwise, MT1-MMP was immunodetected or immunoprecipitated with a rabbit polyclonal antiserum (MMR2) raised against a GST-fused polypeptide corresponding to amino acids 113-538 of MT1-MMP and produced with the help of G. Di Tullio (Consorzio Mario Negri Sud, S. Maria Imbaro, Italy); where indicated, the commercial rabbit polyclonal immunopurified antibody M3927 (Sigma) raised against a synthetic peptide corresponding to the hinge domain of MT1-MMP was used. α1-PDX was immunodetected with a monoclonal anti-FLAG M2 antibody (Sigma). GFP-tagged proteins (in immunoelectron microscopy experiments) were detected with a rabbit polyclonal antibody to GFP (Abcam, Cambridge, UK). Caveolin-1 was detected with an anti-caveolin-1 antibody (Santa Cruz Biotechnology, Santa Cruz, CA). In immunofluorescence colocalizations studies anti-mannose 6-phosphate receptor (Affinity Bioreagents, Golden, CO) and anti-transferrin receptor (Zymed, San Francisco, CA) antibodies were used. Polyclonal sheep anti-TGN46 antibodies were a generous gift from S. Ponnambalan (University of Dundee, Dundee, UK). A rabbit polyclonal against the luminal domain of VSV-G was kindly provided by M. A. De Matteis (Consorzio Mario Negri Sud, S. Maria Imbaro, Italy). In immunoelectron microscopy studies, furin was detected with a polyclonal antibody (Affinity Bioreagents).

Optiprep gradient analysis of Triton X-100-insoluble material was performed using previously published protocols with some modifications (Arreaza et al., 1994). Briefly, cells were grown to confluence in 100 mm dishes and labeled for 2 hours with 0.1 mCi/ml [³⁵S]methionine (Amersham, Cologno Monzese, Italy). Cells were then washed in PBS and lysed with 1% Triton X-100 in TNE buffer (50 mM Tris-HCl, 150 mM NaCl, 2 mM EDTA, pH 8.0) for 10 minutes on ice. Lysates were scraped from dishes, adjusted with Optiprep to a final 35%, and then placed at the bottom of a centrifuge tube. A discontinuous Optiprep gradient (5-30% in TNE) was layered on the top of the lysates and the samples were ultracentrifuged at 120,000 g overnight in a swinging-bucket rotor. Fractions of 1 ml were harvested from the top of the gradient. MT1-MMP immunoprecipitation was performed with MMR2 antiserum after adding 1% Triton X-100 to the samples. Immunoprecipitates were recovered with protein A-Sepharose (Pharmacia, Milan, Italy), washed twice in TNE, resuspended with sample buffer and separated by SDS-PAGE; proteins were then electro-transferred to nitrocellulose filters and detected by autoradiography with the Fujifilm bioimaging analyzer system (BAS) 1800 II and relative software.

MT1-MMP-transfected A375 melanoma cells were labeled for different times, depending on the experiment, with 0.1 mCi/ml [³⁵S]methionine (Amersham) and chased for different times. Cells were cooled on ice and washed twice with ice-cold PBS (pH 8.0). After extraction with 1% Triton X-100 in TNE, soluble (S) and insoluble (P) fractions were separated by ultracentrifugation at 4°C (120,000 g, 1 hour). The pellet, containing Triton X-100-insoluble material, was further incubated with 0.2% SDS, 1% Triton X-100 in TNE at 37°C for 1 hour. MT1-MMP immunoprecipitation was performed with MMR2 antiserum. Immunoprecipitates were recovered with protein A-Sepharose (Pharmacia), washed twice in TNE, resuspended with sample buffer and separated by SDS-PAGE; proteins were then transferred to nitrocellulose filters and detected by autoradiography. Quantitative analysis was performed with the Fuji BAS software.

MT1-MMP-transfected A375 cells were cooled on ice, washed twice with ice-cold PBS and incubated twice for 20 minutes each with 1 mg/ml NHS-biotin (Pierce, Rockford, IL) in PBS at 4°C. Next, cells were washed with PBS and excess NHS-biotin was quenched for 20 minutes with 50 mM NH₄Cl in PBS. After extraction with 1% Triton X-100 in TNE (pH 7.4) at 4°C, soluble and insoluble material was separated by ultracentrifugation at 4°C (120,000 g, 1 hour). The pellet, containing Triton X-100-insoluble material, was further incubated with 2% SDS, 1% Triton X-100 in TNE (pH 7.4), containing a standard antiprotease mix at 37°C for 1 hour. Biotinylated proteins were captured by an overnight incubation in the cold room with streptavidine-agarose beads (Sigma). Supernatants, containing non-biotinylated proteins, were recovered, precipitated in 10% trichloro-acetic acid (TCA) for 30 minutes on ice, washed twice with ice-cold acetone and resuspended in sample buffer. The pellets, which contained biotinylated proteins, were washed twice in TNE (pH 7.4), resuspended in sample buffer and subjected to SDS-PAGE. Proteins were then electro-transferred to nitrocellulose filters and probed with the immunopurified polyclonal anti-MT1-MMP MMR2 and anti-caveolin-1 polyclonal antibodies to reveal the MT1-MMP and caveolin distribution profiles.

Fluorescent gelatin coated coverslips were prepared and the assay carried out as described (Baldassarre et al., 2003; Bowden et al., 2001). Cells were cultured on gelatin-coated coverslips for 16 hours. Most experiments were analyzed with a Zeiss LSM 510 laser scanning confocal microscope (Esslingen, Germany). Immunofluorescence images were acquired at high confocality (pinhole=1 Airy unit) to achieve the thinnest possible optical slices at the substrate-cell interface. To determine the number of degrading cells for each experiment we considered 100 random fields (containing at least 50 transfected cells) at a 63× magnification as previously described (Baldassarre et al., 2003). In some experiments involving multi-color labeling, fluorescence was acquired by wide-field fluorescence microscopy in each channel along the z-axis. To remove the blurring caused by the objective lens and reduce the background caused by fluorescence from out-of-focus regions, images were subjected to mathematical deconvolution with an acquired point spread function (Delta Vision, Issaquah, WA). Images were given a threshold to remove noise and enhance the contrast of structures of interest, and were superimposed and saved.

After treatments, cells were fixed in 4% paraformaldehyde for 15 minutes, permeabilized in PBS containing 0.02% saponin, 0.2% BSA and 50 mM NH₄Cl, incubated with the primary antibodies of interest for 1 hour and then incubated with fluorophore-conjugated secondary antibodies (Molecular Probes Europe BV, Leiden, The Netherlands) for 45 minutes. Finally, coverslips were mounted in the SlowFade (Molecular Probes) antifade reagent.

A375 melanoma cells were plated on 12 mm glass coverslips, transfected with VSV-G-GFP and incubated overnight at 40°C to allow for expression while blocking exit from the endoplasmic reticulum. The following day, cells were washed three times with complete medium containing 20 mM Hepes and incubated at 20°C for 30 minutes in the same medium and chased at 32°C in the presence or absence of 0.5% tannic acid. Cells were then fixed and processed for immunofluorescence as described above, with the difference that cells were not permeabilized prior to incubation with antibodies. A combination of an antibody directed against the lumenal/extracellular domain of VSV-G and a fluorophore-conjugated secondary antibody was used to detect plasma membrane staining and compared with total VSV-G measured as GFP fluorescence. For each treatment at least 20 cells were quantified. The experiment was repeated three times.

MT1-MMP-transfected cells were fixed with 2% formaldehyde and 0.2% glutaraldehyde in PHEM buffer (60 mM PIPES, 25 mM HEPES, 2 mM MgCl₂, 10 mM EGTA, pH 7.4) washed in the same buffer and collected by centrifugation. Pellets were embedded in 10% gelatin, cooled in ice, and cut into 1 mm³ blocks in the cold room. The blocks were infused with 2.3 M sucrose at 4°C for at least 2 hours, and frozen in liquid nitrogen. 50-60-nm-thick sections were cut at –120°C using an Ultracut R/FCS (Leica, Milan, Italy) equipped with an antistatic device (Diatome, Fort Washington, PA) and a diamond knife. Ultrathin sections were picked up in a mix of 1.8% methylcellulose and 2.3 M sucrose (1:1) as described (Liou et al., 1996). Cryosections were collected on formvar/carbon-coated slot copper grids and then incubated with different combinations of rabbit polyclonal antibodies followed by protein-A-conjugated gold particles of different sizes (Slot and Geuze, 1985). Double immunolabeling was performed as described before (Slot and Geuze, 1985; Webster, 1999) with optimal combinations of gold particle sizes. To reduce specificity problems with double labeling (when two rabbit secondary antibodies were used), following treatment with the first secondary antibody and protein A, sections were additionally fixed with 1% glutaraldehyde for 5 minutes, washed with 0.12% glycine and incubated in blocking solution containing 0.01% BSA-c (Aurion, Wageningen, The Netherlands), as previously recommended (Webster, 1999). Additionally, all sequences of primary antibody addition and combinations of gold particle size were tested. After labeling, sections were treated with 1% glutaraldehyde, counterstained with uranyl acetate, and embedded in methyl cellulose uranyl acetate as described (Slot et al., 1991). Automated data acquisition was carried out on a Tecnai 20 electron microscope at 200 kV (FEI/Philips Electron Optics) equipped with a slow-scan CCD camera.

To determine the distribution of MT1-MMP throughout the secretory pathway in transfected A375 cells, quantification was performed on over 30 cross-sections of cells expressing average levels of MT1-MMP. In detail, labeling density of MT1-MMP protein in different compartments was calculated as previously described (Mayhew et al., 2003), micrographs were taken at a 46,000× magnification for each marker. Fields were taken randomly with the only criterion being a well-preserved morphology. Labeling densities in Table 1 were expressed as the number of gold particles per intersections of the morphometrical grid. Estimation of colocalization on EM cryosections was performed according to (Martinez-Menarguez et al., 2001).

To analyze MT1-MMP, we produced two new rabbit polyclonal antisera (MMR1, MMR2) raised against the same GST-fused polypeptide corresponding to amino acids 113-538 of MT1-MMP. These were tested and compared with the commercially available affinity-purified polyclonal antibody M3927 (Sigma) by immunoprecipitation from untransfected and MT1-MMP-transfected A375 melanoma cells (Fig. 1A). The three different antibodies displayed uneven efficiency in the detection of a lower molecular weight (60 kDa) MT1-MMP band, presumably the fully processed mature form (see below). This is of great practical importance because (1) it suggests that detection of mature MT1-MMP is generally critical, as exemplified by comparing the antisera generated in different animals but raised against the same polypeptide (MMR1 and MMR2); and (2) the interpretation of experimental data involving the maturation of MT1-MMP obtained with different antibodies becomes a critical issue. In our experiments, since MMR2 appeared to be by far the most effective, it was used in all subsequent experiments

MT1-MMP-transfected A375 cells were pulse-labeled with [³⁵S]methionine for 5 minutes and, after different chase times, processed for immunoprecipitation with the polyclonal anti-MT1-MMP antibody MMR2. Three molecular forms of MT1-MMP could be detected of 65, 63 and 60 kDa (Fig. 1B). The higher molecular mass 65 kDa form has been previously observed (Maquoi et al., 1998; Yana and Weiss, 2000) and proposed to be a post-translationally modified unprocessed form (Maquoi et al., 1998); we observed that this form is glycopeptidase F-resistant and is also present if cells are treated with inhibitors of N- and O-glycosylation (not shown). The 63 kDa form (proMT1-MMP) had a slow turnover rate over a time frame of 2 hours; only a minor fraction was converted to the mature lower molecular weight 60 kDa form which instead appeared to turn over much more rapidly (Fig. 1B). When BB94, a broad-range metalloprotease inhibitor (Davies et al., 1993), was included during preincubation and chase, MT1-MMP maturation occurred normally which, at least in our conditions, contradicts previous results showing that MT1-MMP was functioning as a self-convertase (Rozanov and Strongin, 2003).

Both endogenous and overexpressed MT1-MMP have been partially located in detergent-resistant membranes (DRM) (Annabi et al., 2001). Insolubility in detergent is typical of glycolipid-rich domains, also referred to as `rafts' (Simons and Ehehalt, 2002). It was thus possible that two populations of MT1-MMP could coexist in transfected A375 cells and explain its peculiar processing pattern. We verified the Triton X-100 solubility of MT1-MMP in total membrane preparations from MT1-MMP-transfected A375 melanoma cells and found that the majority of MT1-MMP (at least 70-80%) was Triton X-100-insoluble and unprocessed (63 kDa pro-form; Fig. 1C). To exclude the possibility that insolubility was due to association with cytoskeletal elements or to the formation of non-specific protein aggregates, we further verified the distribution of MT1-MMP on an Optiprep-density gradient as described in Materials and Methods. We found that the majority of the 63 kDa proMT1-MMP floated towards the top of the gradient (Fig. 1D); the assay was validated by testing the distribution of caveolin-1, a bona-fide marker for raft-type membrane domains, and found that it co-fractionated with proMT1-MMP (Fig. 1D). Again, only the unprocessed 63 kDa proMT1-MMP was detected in the DRM-enriched fraction (lane 5 in Fig. 1D).

We thus addressed the possibility that differential MT1-MMP processing was occurring in association with its partitioning in different membrane subdomains. To this end, MT1-MMP-transfected cells were pulse-labeled with [³⁵S]methionine for 5 minutes and, after different chase times, lysed with 1% Triton X-100. Soluble (S) and insoluble (P) fractions were separated by ultracentrifugation and immunoprecipitated with polyclonal antibody MMR2 (Fig. 2). The detergent-soluble 63 kDa proMT1-MMP is rapidly processed to the mature form within 20 minutes and, after 60 minutes, is severely reduced (Fig. 2). As a whole, the soluble MT1-MMP pool undergoes a rapid turnover (Fig. 2) and is degraded starting from 2 hours (Fig. 1B); DRM-associated proMT1-MMP, instead, is stable and remains unprocessed even after a 2 hour-long chase (Fig. 2). This indicates that only the detergent-soluble pool of MT1-MMP undergoes maturation and rapid degradation; DRM-associated proMT1-MMP is, conversely, completely excluded from processing and is quite stable over time.

Overexpression of MT1-MMP in A375 cells led to, as expected, a massive degradation of cross-linked fluorescent gelatin (Fig. 3A). When monitored in vivo, with a GFP-tagged MT1-MMP chimera (which produces the same phenotype as wild-type MT1-MMP), this appeared to be an extremely rapid process (confocal time-lapse series; see Movie 1 in supplementary material) (ecm.mov). The question thus arises as to which molecular forms and populations of MT1-MMP are responsible for this biological activity. Clearly, in our experimental setting, proMT1-MMP is activated by furin. In fact, the serpin-like protein α1-PDX, a bioengineered α1-antitrypsin characterized as an extremely specific endocellular inhibitor of the pro-protein convertases furin and PC6 (Jean et al., 1998) completely abated MT1-MMP-induced (Fig. 3A) gelatin degradation by A375 cells.

A number of experiments were thus performed to understand better the processing of MT1-MMP and define the molecular form(s) responsible for the biological activity, a controversial issue in the literature. Since gelatin degradation relies on cell surface presentation of enzymatically active MT1-MMP, we verified the presence and membrane partitioning of MT1-MMP at the plasma membrane. To this end, surface proteins were biotinylated on ice (to block all membrane trafficking events) and then probed by western blotting analysis of the different membrane fractions (detergent-resistant vs detergent-soluble) and cellular locations (intracellular vs surface). We observed that, at steady state, the vast majority of MT1-MMP was intracellular, immature and associated with DRM (Fig. 3B). Although total plasma membrane MT1-MMP levels were relatively low, also considering that surface biotinylation is not very efficient, DRM-associated proMT1-MMP remained the dominant fraction over non-DRM proMT1-MMP (Fig. 3B). This evidence also implies that mature 60 kDa MT1-MMP, shown here to be associated with the detergent-soluble membranes, is rapidly removed from the cell surface after presentation. This is in good agreement with the rapid turnover of MT1-MMP as verified in our pulse-chase analysis (Fig. 1A). We then proceeded to analyze the effect of furin inhibition on MT1-MMP partitioning at the plasma membrane: although co-expression with α1-PDX totally inhibited the proteolytic processing of detergent-soluble MT1-MMP, proMT1-MMP was still present at the plasma membrane (Fig. 3B). Interestingly, in these conditions an increase of the 65 kDa unprocessed form of MT1-MMP was observed. As noted above, this has been previously reported (Maquoi et al., 1998; Yana and Weiss, 2000) and also occurs when treatments affecting transport or processing are applied such as Brefeldin A (Fig. S2 in supplementary material).

In conclusion, these results clearly indicate that: (1) only non-detergent-soluble proMT1-MMP is processed to the mature 60 kDa form; (2) this fully processed MT1-MMP alone is responsible for gelatinolytic activity; (3) the processing enzyme is most likely furin or a furin-like convertase as previously suggested; and (4) removal of the pro-domain is not a prerequisite for arrival of MT1-MMP at the plasma membrane.

The MT1-MMP processing compartment still needs to be defined. To address this issue, we first analyzed the subcellular distribution of MT1-MMP at steady state in A375 melanoma cells. MT1-MMP presented a perinuclear patchy distribution with scattered cytoplasmic elements and plasma membrane staining (Fig. 3A and Fig. 4A,D,G,J). Next we compared the localization of MT1-MMP to markers of the secretory pathway. Endoplasmic reticulum and Golgi apparatus markers showed no significant colocalization with MT1-MMP (not shown). When we tested an anti-TGN46 antibody in MT1-MMP-transfected A375 cells, a patchy asymmetrical perinuclear distribution was observed, with an apparently high, but not complete, colocalization with MT1-MMP (Fig. 4A-C); TGN46 is a protein known to cycle between the trans-Golgi network, the plasma membrane and the endosomal compartment and, although generally considered to be a trans-Golgi network marker, its actual distribution depends on the cell type. By contrast, the distribution of transferrin receptor and mannose 6-phosphate receptor (bona fide early and late endosome markers, respectively) only partially overlapped with MT1-MMP (Fig. 4D-F and G-I, respectively). In which of these compartments, therefore, does furin-dependent activation actually occur? Our pulse-chase analysis showed that fully processed MT1-MMP appeared only after at least 20 minutes, which suggests that it might occur in a late station along the biosynthetic-secretory pathway. In addition, a furin-dependent proteolytic activation is not expected to occur before the trans-Golgi network (Molloy et al., 1999). Nevertheless, to formally rule out the possibility that MT1-MMP activation occurs in an early station of the secretory pathway, we took advantage of the fungal toxin Brefeldin A. This treatment leads to intermixing between the endoplasmic reticulum and the Golgi apparatus and blocks exit from the Golgi apparatus itself (Fujiwara et al., 1988; Lippincott-Schwartz et al., 1989). Proteins normally present in the endoplasmic reticulum and the Golgi apparatus would still be modified by processing enzymes contained in those compartments. Despite this, there is no proteolytic processing of proMT1-MMP; moreover, its cell surface presentation is strongly reduced as expected (see Fig. S2 in supplementary material).

Next, we verified the extent of furin/MT1-MMP colocalization. Furin is, depending on the cell type, typically distributed throughout the various proprotein processing compartments including the trans-Golgi network, cell surface and endosomes (Mayer et al., 2004; Molloy et al., 1999). In A375 cells cotransfected with MT1-MMP and a furin-GFP chimera, there is an excellent correlation between the two proteins (Fig. 4J-L). Considering this evidence, we further searched for the MT1-MMP activation compartment. The first possible station considered for MT1-MMP processing was thus the trans-Golgi network, a complex structure located at the distal-most region of the Golgi apparatus that plays a central role in the sorting, processing and targeting of proteins of the secretory pathway en route to their final destination. When MT1-MMP was accumulated in the trans-Golgi network through the application of a 20°C temperature block (Griffiths et al., 1985), it remained unprocessed (Fig. 5A). Notably, previous reports have shown that, in these conditions, the enzyme activity of furin is not blocked (Song and Fricker, 1995). Within 30 minutes after temperature shift to 37°C (release from the block), intracellular trafficking as well as MT1-MMP processing resumed (Fig. 5A). Hence, our results are consistent with a MT1-MMP processing event occurring after exit from the trans-Golgi network. Considering all the above and that proteins blocked in the trans-Golgi network at 20°C are known to exit rapidly (Polishchuk et al., 2000), we suggest that the processing of MT1-MMP takes place beyond the trans-Golgi network during transport to the plasma membrane.

We thus determined whether processing of MT1-MMP takes place prior to or after delivery to the plasma membrane. Tannic acid has been used as an inactivator of fusion events at the plasma membrane (Newman et al., 1996; Polishchuk et al., 2004). We thus proceeded to use this experimental approach to prevent delivery of MT1-MMP to the cell surface after a 20°C temperature block. In the presence of tannic acid, proMT1-MMP processing still occurred, albeit not completely (Fig. 5A), even though some toxic effects of the treatment can be inferred from the increased degradation of mature MT1-MMP. This is taken to indicate that MT1-MMP processing can occur intracellularly in a post-trans-Golgi network compartment and prior to arrival at the plasma membrane. To confirm that tannic acid indeed functioned to block fusion at the plasma membrane in our experimental system, we verified the delivery of a ts045 temperature sensitive mutant of vesicular stomatitus virus G protein (VSV-G)-GFP chimera to the plasma membrane. This protein is blocked in the endoplasmic reticulum while cells are maintained at 40°C, is regularly transported through the Golgi apparatus up to the TGN (where it remains) at 20°C and is finally transported to plasma membrane at the permissive temperature of 32°C. As shown in Fig. 5B, control cells feature a clearly defined plasma membrane staining 60 minutes after release from the 20°C block; by contrast, cells treated with 0.5% tannic acid are unable to deliver VSV-G-GFP to the cell surface and present an accumulation of transport intermediates just below the plasma membrane. Thus, in A375 cells treated with tannic acid, fusion of transport intermediates with the plasma membrane is, in fact, completely impaired (Fig. 5B, graph). Complementary evidence was obtained by coexpression of MT1-MMP with Dyn2^K44A, a dominant negative mutant of the GTPase dynamin 2 and a potent inhibitor of both caveolar and clathrin-dependent endocytosis (Cao et al., 2000; Pelkmans and Helenius, 2002); in such conditions maturation of proMT1-MMP was not affected (not shown), which makes it unlikely that MT1-MMP is processed following cell surface presentation, for example after an endocytic step.

The extracellular face of the plasma membrane is thought to be the location where MT1-MMP processing by furin might occur (Mayer et al., 2003). Since our results are at variance with this notion, we further tested this possibility. Furin is a calcium-dependent enzyme (Molloy et al., 1992) and as such it would be completely inhibited in the absence of free (i.e. unchelated) Ca²⁺. Fig. 5C shows that MT1-MMP processing can still occur when extracellular Ca²⁺ is chelated with a supramaximal concentration of EGTA (5 mM). Maturation does not appear to be complete (Fig. 5C, graph), however, possibly due to a plasma membrane component of processing (Mayer et al., 2003) or leakage of EGTA into the processing compartment via fluid-phase endocytosis.

To gain a better understanding of the processing compartment, MT1-MMP-transfected A375 cells were processed for cryoimmuno-electron microscopy and immunolabeled with an immunopurified anti-MT1-MMP antibody (MMR2). We thus determined the intracellular distribution of labeling densities for MT1-MMP throughout the various compartments of the secretory pathway identified on the basis of their morphology. The distribution of MT1-MMP was not homogeneous and presented a clear, high peak of 76.6% (Table 1) in tubulo-saccular structures located between the trans-Golgi network and the plasma membrane (Fig. 6A-C). Notably, MT1-MMP expression levels had no effect on the morphology of the secretory structures (not shown). This compartment was thus defined as a whole as a post-trans-Golgi network compartment on the basis of our biochemical experiments and on the morphological observation that, at the ultrastructural level, it was not specifically labeled by the TGN marker TGN46 at steady state (see Fig. S1A,B in supplementary material). This compartment might also contain endosomal elements as suggested by our immunofluorescence studies, consistent with recent reports suggesting a recycling loop for MT1-MMP (Remacle et al., 2003; Wang et al., 2004).

When the labeling density of MT1-MMP in different compartments was compared with that of endogenous furin, it was established that 64.6% of the MT1-MMP-positive structures contained furin. Furthermore, 84.8% of the furin-positive structures contained MT1-MMP. This localization coincided with the post-trans-Golgi network compartment, where MT1-MMP was enriched, and which represented the foremost reservoir for endogenous furin in A375 cells (Fig. 6A-C). Interestingly, the post-TGN compartment where furin and MT1-MMP colocalize was not significantly affected by our 20°C block protocol (see Fig. S1C in supplementary material).

A previous report hypothesized that MT1-MMP is localized to caveolar DRM (Annabi et al., 2001); it has also been suggested that MT1-MMP is internalized via caveolae in addition to the clathrin-mediated pathway (Remacle et al., 2003). We thus examined whether the protein caveolin-1, a typical component of caveolae, colocalized with MT1-MMP at the plasma membrane. Fig. 6D clearly shows that MT1-MMP at the plasma membrane is not preferentially localized in caveolin-1-enriched domains. This is not in contrast with the biochemical data in Fig. 1D; in fact, although both MT1-MMP and caveolin-1 partition into DRM, caveolae represent only a subset of the total DRM fraction (Schnitzer et al., 1995).

Our data indicate that furin, the MT1-MMP-activating enzyme, is unable to process DRM-associated MT1-MMP which, as a result, reaches the plasma membrane in its 63 kDa immature form. A possible explanation for the inability of furin to activate the large fraction of DRM-associated MT1-MMP could be that they are physically segregated during transport and thus unable to interact. To test this, we examined the membrane partitioning of a furin-GFP chimera by transient co-transfection with MT1-MMP. As shown (Fig. 7A), despite the increased cellular burden due to overexpression, furin-GFP, in contrast to MT1-MMP, was exclusively associated with the detergent-soluble membrane fraction; notably, this partitioning was not affected by incubation at 20°C (see Fig. S1D in supplementary material). GFP chimeras have been extensively used to study membrane partitioning of diverse proteins with no reported alterations in normal profiles (Rodgers, 2002). Furthermore, in our experimental system, the furin-GFP chimera was enzymatically active, as its expression clearly increased the extent of MT1-MMP maturation (Fig. 7B). Taken together, these results further confirm that partitioning of MT1-MMP into different membrane sub-domains regulates its processing and biological activities. This also supports the notion that differential partitioning into DRM, which is also seen in the Alzheimer precursor protein (Ehehalt et al., 2003), reflects an intrinsic property of MT1-MMP.

In the present study we show that in a melanoma cell line, although the majority (80%) of MT1-MMP is sorted to DRM fractions, only the minor (20%) detergent-soluble fraction undergoes intracellular processing to the mature form. Also, only this processed MT1-MMP is responsible for ECM degradation in vitro. Furthermore, different experimental approaches were employed to: (1) provide direct evidence that MT1-MMP processing occurs intracellularly in a post-Golgi apparatus compartment, prior to arrival at the plasma membrane; and (2) define, at the ultrastructural level, the subcellular compartment where the majority of MT1-MMP colocalizes with furin and where processing presumably occurs.

MT1-MMP is clearly emerging as a multifunctional protein playing different physiological and pathological roles (for a review, see Seiki, 2003; Sounni et al., 2003). As a consequence, definition of MT1-MMP processing pathways, transport and the subcellular site of activation is of central importance.

In MT1-MMP-transfected A375 melanoma cells, the majority of MT1-MMP is associated with DRM in accordance with previous reports (Annabi et al., 2001; Annabi et al., 2004). By contrast, it has recently been reported that a MT1-MMP deletion mutant lacking the cytosolic C-terminal tail, but not the full-length protein, is stably partitioned into DRM (Rozanov et al., 2004). We extended these findings by analyzing the relationship between proteolytic processing and membrane subdomain association. We found that MT1-MMP undergoes proteolytic processing only if associated with detergent-soluble membranes. The immature 63 kDa form remains stably associated with DRM-enriched membranes and persists following the processing of the detergent-soluble proMT1-MMP and its degradation. In other words, the metabolic fate and thus the activity of MT1-MMP is directly dependent on its differential association with the two membrane subdomains. A first consequence is that two cellular and possibly functionally distinct pools of MT1-MMP can thus be defined: one that is normally and completely processed and thus available for ECM degradation and/or other proteolytic activities, and a second, whose function remains to be understood, that is potentially available at the plasma membrane for interactions with other molecules. A second consequence is that this phenomenon, if undetected (depending on the experimental procedures used), could have generated some of the discrepancies in the interpretation of MT1-MMP processing.

Most recently MT1-MMP was found in DRM and thus suggested to be associated with caveolin-1 (Annabi et al., 2001); also, it was recently hypothesized that MT1-MMP is internalized via caveolae as well as via the clathrin-mediated pathway (Remacle et al., 2003). In our experiments, we did not find morphological evidence of MT1-MMP association to caveolae-like domains. This is not incongruous since caveolae represent only a subpopulation of the total DRM-fraction and can actually be separately isolated (Schnitzer et al., 1995).

Previous findings (Yana and Weiss, 2000) suggested that MT1-MMP processing is dependent on a furin-like activity, although it has been proposed that MT1-MMP does not require the cleavage of its propeptide to activate MMP2 in certain cell types (Cao et al., 1996). In addition, activation of MT1-MMP could also occur by plasmin at the cell surface (Okumura et al., 1997). To verify the activation mechanism in our experimental system, we co-transfected MT1-MMP with α1-PDX in A375 cells. In these conditions, MT1-MMP-dependent gelatin degradation was completely inhibited, indicating that MT1-MMP needs to be processed to express direct or indirect gelatinolytic activity. In addition, since in such conditions the mature 60 kDa form is undetectable, alternative pathways of activation, such as by extracellular plasmin (Okumura et al., 1997), can be excluded. Finally, BB94, a broad-range metalloprotease inhibitor (Davies et al., 1993), did not affect MT1-MMP processing which, at least in our conditions, contradicts previous results showing that MT1-MMP was functioning as a self-convertase (Rozanov and Strongin, 2003). These data enhance the essential role of furin in the regulation of MT1-MMP processing and activity.

Despite the central role played by furin in many pathophysiological processes, the exact subcellular sites of processing and activation of its substrates remain elusive. This is also the case for MT1-MMP. Hence, a major concern is to identify the cellular location(s) where endoproteolytic cleavage of MT1-MMP occurs. A number of studies have established that furin is mainly localized in the trans-Golgi network but undergoes extensive cycling between the Golgi complex, endosome compartments and the plasma membrane (reviewed by Molloy et al., 1999). Notably, a high-resolution immunogold electron microscopy-based study has recently shown that, in addition to the Golgi apparatus, furin was clearly localized to endosomes and plasma membranes of a number of cell types (Mayer et al., 2004). Our results that MT1-MMP processing takes place in an intracellular, post-Golgi apparatus, pre-plasma membrane compartment are consistent with this widespread distribution of furin.

The extracellular face of the plasma membrane was recently proposed to be the MT1-MMP activation compartment based on experiments showing that, after chemical cross-linking, furin co-immunoprecipitates with proMT1-MMP at the plasma membrane (Mayer et al., 2003). The same study also showed a colocalization of the two proteins on the plasma membrane. We show that, at least in A375 melanoma cells, furin-dependent processing of MT1-MMP can occur in a post-TGN, pre-plasma membrane compartment.

At steady state, we observed transfected proMT1-MMP to be mainly associated with DRM albeit with a significant fraction over the plasma membrane. This suggests that (1) intracellular removal of the propeptide domain is not a prerequisite for MT1-MMP cell surface presentation; (2) DRM-associated proMT1-MMP is not subject to intracellular furin-dependent activation; and (3) proMT1-MMP is not functioning as a gelatinase. A possible mechanicistical explanation is provided by our observation that furin is excluded from detergent-resistant membranes. In fact, physical interaction between proteins associated with different membrane subdomains is extremely unlikely, as recently suggested for other proteins (Ehehalt et al., 2003). In our context, colocalization of furin and MT1-MMP in common structures supports the idea that detergent-soluble and -resistant MT1-MMP pools might coexist but not necessarily interact. This suggests that partitioning of MT1-MMP into different lipid subdomains, which is also seen in the Alzheimer precursor protein (Ehehalt et al., 2003), could be a mechanism for the regulation of its processing and activity. Interestingly, it was recently proposed that the cholesterol content of tumor cells is critical for the regulation of MT1-MMP cell surface presentation and MMP2 activation (Atkinson et al., 2004).

It has clearly emerged that MT1-MMP (and perhaps other membrane-type MMPs) is essential for the growth of human cancers and acts by disrupting the three-dimensional matrix that would otherwise impede cell proliferation (Hotary et al., 2003). Hence, full understanding of MT1-MMP physiology is crucial to develop strategies aimed at preventing escape of cells from a primary tumor (Yamada, 2003). Here we show that only a subpopulation of MT1-MMP, associated with detergent soluble membranes, are activated and rapidly degraded, perhaps by targeting to lysosomes (Takino et al., 2003). By contrast, DRM-associated MT1-MMP retains its propeptide and is stably associated with the plasma membrane. The functional role of DRM-associated MT1-MMP is not yet known. A potential indication stems from the findings that (1) hyaluronic acid cell-surface binding to its receptor CD44 is inhibited by the overexpression of wild-type MT1-MMP; this inhibiton can be overcome by shifting part of the MT1-MMP pool out of DRM by cholesterol extraction (Annabi et al., 2004); and (2) wild-type MT1-MMP overexpression triggers ERK activation (Annabi et al., 2004). A possibility that remains to be tested is whether these events are specifically triggered by the DRM-associated proMT1-MMP pool; in fact, DRM domains represent a preferred platform for a number of signaling receptors and cytoskeleton interactors (Simons and Toomre, 2000).

In conclusion, our experiments clearly indicate that the metabolic fate of proMT1-MMP is dependent on its partitioning into different lipid domains. In particular, furin-dependent activation, shown here to occur in a post-Golgi, pre-plasma membrane compartment, occurs exclusively at the expense of the detergent-soluble fraction. These findings establish a novel mode of regulation of MT1-MMP activity and suggest a potential molecular basis for its diverse functions.

We thank Arsenio Pompeo and Maria Inmaculada Ayala Grande for insightful discussions. R.B. and A.L. were supported by grants from the Italian Association for Cancer Research (AIRC, Milano, Italy) and the Italian Foundation for Cancer Research (FIRC, Milano, Italy). R.B. was also supported by the European Commission (contract LSHC-CT-2004-503049). J.C. was supported by a Scientist Development Grant from the American Heart Association and a New Investigator grant from the US Army Medical Research and Materiel Command. S.Z. was supported by a Merit Review grant and a Research Enhancement Award Program grant from the Department of Veterans Affairs. M.M. was supported by `Emilio Mussini' (2003) and FIRC (2004) Fellowships.