Cancer Cell Biology
Restricted expression of membrane type 1-matrix metalloproteinase by myofibroblasts adjacent to human breast cancer cells
Article first published online: 21 FEB 2003
Copyright © 2003 Wiley-Liss, Inc.
International Journal of Cancer
Volume 105, Issue 1, pages 7–13, 20 May 2003
How to Cite
Bisson, C., Blacher, S., Polette, M., Blanc, J.-F., Kebers, F., Desreux, J., Tetu, B., Rosenbaum, J., Foidart, J.-M., Birembaut, P. and Noel, A. (2003), Restricted expression of membrane type 1-matrix metalloproteinase by myofibroblasts adjacent to human breast cancer cells. Int. J. Cancer, 105: 7–13. doi: 10.1002/ijc.11012
- Issue published online: 27 MAR 2003
- Article first published online: 21 FEB 2003
- Manuscript Accepted: 8 NOV 2002
- Manuscript Revised: 25 SEP 2002
- Manuscript Received: 27 MAY 2002
- Communauté Française de Belgique (Actions de Recherches Concertées)
- Commission of European Communities
- Fonds de la Recherche Scientifique Médicale
- Fonds National de la Recherche Scientifique (FNRS, Belgium)
- Fédération Belge Contre le Cancer
- Fonds spéciaux de la Recherche (University of Liège)
- Centre Anticancéreux près l'Université de Liège
- FB Assurances
- Fondation Léon Frédéricq (University of Liège)
- Région Wallonne
- Fonds d'Investissements de la Recherche Scientifique (CHU, Liège, Belgium)
- Federal Office for Scientific, Technical and Cultural Affairs (O.S.T.C., Brussels, Belgium)
- α smooth muscle actin;
The membrane type-1 matrix metalloproteinase (MT1-MMP), a protease originally identified in breast carcinoma, is characterized by its capacity to activate other MMPs (MMP-2 and MMP-13) and to degrade extracellular matrix. Our study was undertaken to localize and identify the MT1-MMP expressing cells in human breast adenocarcinomas. A textural analysis of images obtained by immunohistochemistry and in situ hybridization showed precisely the co-expression of alpha smooth muscle actin (αSM actin) and MT1-MMP in myofibroblasts. MT1-MMP expression is confined to myofibroblasts in close contact with tumor cells. In sharp contrast, the expression of MMP-2 was more widely distributed in both αSM actin positive and negative cells close to and at distance from cancer cell clusters. Our in vitro observations are consistent with the higher level of MT1-MMP expression and of MMP-2 activation observed in αSM actin positive fibroblasts derived from breast tumors, as compared to normal breast fibroblasts. Collectively, these results implicate myofibroblasts as major producer of MT1-MMP in breast cancer and emphasize the importance of stromal-epithelial cell interactions in their progression. © 2003 Wiley-Liss, Inc.
Breast tumor development is dependent upon the formation of a supporting tumor stroma that comprises extracellular matrix proteins (ECM), newly formed blood vessels, inflammatory cells and activated fibroblasts. The complex interplay between malignant cells themselves, endothelial cells and stromal fibroblasts remains poorly understood. Such a cross-talk involves at least angiogenic factors, growth factors, cytokines, chemokines and proteases.
ECM proteolysis contributes to tumor growth, angiogenesis and metastatic dissemination1, 2 and involves the action of matrix metalloproteinases (MMPs), a group of at least 23 zinc dependent and matrix-degrading neutral enzymes with different but overlapping substrate specificities. MMPs are synthesized as inactive zymogens (Pro-MMPs) and their activation by proteolytic cleavage is a rate-limiting step for their catalytic function. Membrane type-1 MMP (MT1-MMPs) is a membrane anchored MMP and has a pivotal function in connective tissue metabolism3 and activation of pro-MMP-2 (progelatinase A).4, 5 MT1-MMP binds its physiological inhibitor TIMP-2 that acts as an adapter molecule mediating pro-MMP-2 binding to cell surface MT1-MMP.6, 7 MT1-MMP not only activates pro-MMP-2 and pro-MMP-38 but also displays a broad spectrum of activity against ECM components such as fibrillar collagens, fibronectin, vitronectin, laminin, fibrin, and proteoglycans.9, 10, 11 The key role played by MT1-MMP during tumor growth and angiogenesis is supported by (i) the correlation existing between MT1-MMP expression and the malignancy of breast cancers,12, 13(ii) the capacity of MT1-MMP to promote angiogenesis in fibrin matrix,14 as well as tumoral angiogenesis15; and (iii) the upregulation of vascular endothelial growth factor (VEGF) expression associated with MT1-MMP overproduction,16, 17 Whether MT1-MMP participates in tumor progression through its capacity to activate pro-MMP-2 or through other functions remains to be determined.
An important issue during the last 10 years was to determine the cellular source of MMPs involved during cancer progression. Studies of breast adenocarcinoma have shown that stromal cells are strongly involved in the MMP production. Thus, mRNAs for interstitial collagenase (MMP-1), MMP-13,18, 19 stromelysin-3 (MMP-11),20 gelatinase A (MMP-2),21 MT1-MMP12, 13 and their inhibitor TIMP-2 have been all found to be expressed by fibroblast-like cells.22 Although the fibroblastic cells expressing uPA,23 MMP-1124 or MMP-1313 have been identified as myofibroblasts, the subset of cells expressing MT1-MMP or MMP-2 in breast carcinomas has not been elucidated. The precise respective cellular origin of several MMPs in tumors has even been precluded by methodological limitations.
Our study was undertaken to gain further insight into the comparative pattern of MT1-MMP and MMP-2 expression in relation to the subtypes of fibroblastic cells at the border of tumoral cells. A textural analysis of images obtained by immuno-histochemistry and in situ hybridization was carried out to compare the spatial distribution of MT1-MMP, MMP-2 and αSM actin in human breast carcinomas. Grey level image transformation, binary image processing and measurements have been carried out on serial stained sections. The distribution of MMPs and αSM actin in relation to tumor cells was determined by an original morphological method. MT1-MMP transcripts were confined to αSM actin-positive myofibroblasts in close contact to invasive tumor cells. In sharp contrast, MMP-2 mRNA was expressed by both αSM actin positive and negative fibroblasts broadly distributed in the stromal compartment. Consistent with the in vivo observations, MT1-MMP expression and MMP-2 activation were higher in fibroblasts derived from breast carcinomas than from normal breast tissues. Altogether, these differential patterns of MT1-MMP and MMP-2 expression, as well as the close association of MT1-MMP to tumor cells emphasize the importance of MT1-MMP during cancer progression.
MATERIAL AND METHODS
Source and preparation of tissues
The breast human tissues were derived from biopsies of 5 normal breast tissues, 25 ductal carcinomas (5 in situ and 20 invasive carcinomas). The samples were fixed in formalin and embedded for immunohistochemistry and in situ hybridization studies.
Tissue paraffin sections (7 μm thick) were rehydrated and treated with 0.3% hydrogen peroxide for 5 min to quench endogenous peroxidase activity. Non-specific binding was blocked with serum (Dako, Carpinteria, CA) for 20 min. Slides were incubated for 1 hr with monoclonal antibody raised against αSM actin (1/2000) (Sigma, St. Louis, MO). After 3 washes for 10 min in PBS, subsequent steps were carried out with the peroxidase LSAB kit (labeled-streptavidin biotin method; Dako). Peroxidase activity was shown with AEC (3-amino-9-ethylcarbazole) chromogen, which generated a red-brown product. All slides were counterstained with Mayers haematoxylin, mounted and observed under a Zeiss fluorescence microscope.
In situ hybridization
Tissue paraffin sections (7 μm thick) were rehydrated and treated with 0.2 N HCl for 20 min at room temperature, followed by 15 min at 37°C in proteinase K (1mg/ml in Tris-EDTA-NaCl; Sigma). Sections were then washed in 2× SSC, acylated in 0.25% acetic anhydride in 0.1 M triethanolamine for 10 min and hybridized overnight with35S-labeled antisense RNA transcripts prepared from MT1-MMP cDNA (1,760 bp),25 MMP2 cDNA (1,500 bp) (kindly provided by G. Murphy) and TIMP2 cDNA (1,035 bp) (kindly provided by Dr. Y. DeClerck). The samples were then treated with RNase (20 mg/ml) for 1 hr at 37°C to remove the unhybridized probes, washed under stringent conditions and detected autoradiographically by exposure to D19 emulsion (Kodak, Rochester, NY) for 21 days. The control sections were treated under the same conditions, but were hybridized with35S-labeled sense probes.
Image analysis was carried out on a workstation Sun SPARC30, using the software Visilog 5.0 from Noesis. Images were digitized in 760 × 540 pixels with 256 grey levels. Grey level image filtering and image binary processing have been carried out according to the following steps: (i) enhancement of local discontinuities or edges that delimit the darkest regions by gradient techniques, (ii) thresholding of the resulting gray level image with the purpose to extract only the darkest spots, (iii) filtration based on a size criterion to exclude the dark points lower than 10 pixels that have remained on the image after thresholding and which correspond to unspecific staining.
To determine the distribution of MMP mRNAs or αSM actin immuno-staining in relation with the tumor cells, preinvasive areas of ductal carcinomas were analyzed. The boundaries of each tumor cell islet has been drawn manually. The density of staining was then calculated in successive rings obtained by the dilatation 1,2 … n times of these boundaries. The density of labeling is represented in a function of the distance to tumor cell islets.
Cells were cultured in DMEM, supplemented with 10% (v/v) heat inactivated FCS, penicillin–streptomycin (100 IU/ml–100 μg/ml), 10 mM HEPES buffer, 2 mM glutamine at 37°C in a humid atmosphere. All culture reagents were purchased from Gibco-Life Technologies (Merelbeke, Belgium). Normal human breast fibroblasts (NBF1 and NBF2) were derived from explants of human breast stroma.26 Breast tumor-derived fibroblasts (BTF1 and BTF2) were isolated from primary infiltrating mammary carcinomas according to Wang et al.27 and used between passages 4–10. The fibroblastic nature of these cells was confirmed by their morphology, the expression of vimentin, the absence of cytokeratin using immunohistochemistry. Experiments carried out in triplicate for each condition and each type of fibroblasts were repeated 3 times.
Subconfluent cells cultured on glass were fixed with cold methanol. Nonspecific antibody binding was blocked by incubation in 3% BSA-PBS for 30 min. Cells were incubated 1 hr, at room temperature, with the monoclonal αSM actin antibody (1/400) (Sigma) in 1% BSA-PBS. After washings with BSA-PBS, a secondary FITC-conjugated antibody (rabbit anti-mouse antibody, Sigma) was applied for 1 hr in the dark. Simultaneously, cell nuclei were stained with bisbenzimide (25 mg/ml). After extensive washings, cells were mounted in buffered glycerin and the percentage of positive cells was evaluated under an Olympus microscope.
Preparation of conditioned medium and cell extracts
Cells (105 cells per well) were seeded in DMEM supplemented with 10% FCS. After cell adhesion, medium was removed and cells were extensively washed with serum-free medium. Cells were then maintained in fresh serum-free DMEM during 48 hr. The conditioned medium (CM) was harvested, centrifuged to remove cell debris and stored at −20°C. To prepare cell extracts, cells were incubated at 4°C with 50 mM Tris (PH 7.4), 150 mM NaCl, 1% (v/v) Igepal, 1% (v/w) (Sigma), Triton X-100, 1% (w/v) sodium deoxycholate (Sigma), 0.1% (w/v) SDS, 5 mM iodoacetamide (Sigma), 2 mM phenylmethylsulfonylfluoride (Sigma). Proteins extracts were centrifuged and stored at −20°C. Protein concentrations in cell extracts were determined with DC protein assay (BioRad, Hercules, CA).
Cell extracts (20 μg of total proteins) were mixed with sample buffer and 2.5% (v/v) β-mercaptoethanol, boiled 5 min and separated on 10% SDS-PAGE gel. Proteins were electrically transferred to a nitrocellulose membrane (polyscreen PVDF hybridization transfer membrane, NEN Life Science Products). After saturation in 3% BSA, 25 mM Tris, 150 mM NaCl, 0.1% Tween 20 (v/v) during 1 hr, the filter was incubated overnight, at 4°C, with primary monoclonal antibody (mAb 2D7) raised against the hemopexin-like domain of MT1-MMP (1/200; kindly provided by MC. Rio, IGBMC, Illkirch, France). After washings, the membrane was exposed to a secondary goat anti-mouse peroxidase-conjugated antibody (1/1000; DAKO, Copenhagen, Denmark). Signals were identified by chemoluminescence reagent kit (NEN Life Science products).
Gelatinolytic analysis was carried out by gelatin zymography as described previously.28 The amount of conditioned medium analyzed by zymography for each assay was standardized according to the protein concentration (2 μg total cell proteins were loaded). Medium conditioned by subconfluent fibrosarcoma HT080 cells treated with 12-O-tetradecanoylphorbol-13-acetate (TPA) (10 ng/ml) for 48 hr, was used as gelatinase standards. The relative intensities of the gelatinolytic bands corresponding to the different MMP-2 forms were evaluated by scanning densitometry using a GS-700 Imaging Densitometer (Bio-Rad) equipped with Molecular Analyst Software. Results were expressed as relative percentages of total MMP-2 activity.
Differences between the experimental conditions were evaluated using the Student's t-test (p-values <0.05 were considered as significant).
Identification of MT1-MMP mRNA expressing stromal cells in breast carcinomas
According to previous studies,21, 29, 30 MMP-2 mRNAs were found in both normal tissues and malignant lesions, whereas MT1-MMP mRNAs were not detected in benign lesions or normal tissues (data not shown). Nevertheless, MMP-2 was strongly expressed in numerous fibroblasts present in invasive carcinomas, whereas their distribution was sparse and weaker in fibroblasts of normal tissues. In invasive carcinomas, MT1-MMP transcripts were confined to fibroblasts in close contact to invasive tumor cells (Fig. 1a). The use of serial sections showed that this particular localization of MT1-MMP mRNAs paralleled that observed for a subset of fibroblastic cells immunoreactive for αSM actin indicating that they are myofibroblasts (Fig. 1c). To more precisely study the spatial distribution of MT1-MMP, MMP-2 and αSM actin, image analysis has been carried out. Figure 1 shows a representative example of invasive breast carcinoma. For each case analyzed, 3 images of the same region, at the same magnification, were acquired and filtered with in situ hybridization for MT1-MMP (Fig. 1d), MMP-2 (Fig. 1e) or αSM actin immunostaining (Fig. 1f). To quantify the spatial distribution of these markers, image binary processing was carried out and MT1-MMP or MMP-2 is represented in green and αSM actin in blue (Fig. 1d–f). The superposition of binary images obtained from MT1-MMP and αSM actin stainings resulted in red spots located at the vicinity of tumor cells (Fig. 1g). This demonstrates that all MT1-MMP expressing cells were positive for αSM actin and were myofibroblasts. On the opposite, the spatial distribution of MMP-2 was sparse and only partly superposable to that of the αSM actin (Fig. 1h). Although all MT1-MMP expressing cells were αSM actin positive cells, both αSM actin positive and negative cells expressed MMP-2. Only few fibroblastic cells were both positive for MMP-2 and its activator (Fig. 1i).
We developed an original approach to determine the distribution of MMP mRNAs or αSM actin immuno-staining in relation with the tumor cells. The density of labeling is represented in a function of the distance from tumor cell cluster (Fig. 2c,d). For MT1-MMP and αSM actin, staining density decreased abruptly with the distance to tumor islets, whereas MMP-2 labeling remained quite constant whatever their location relative to the tumor front (Fig. 2). These observations demonstrate that only myofibroblasts in close association with tumor cells produced MT1-MMP.
In vitro comparison of fibroblasts extracted from normal or tumoral breast tissue samples
To further characterize the fibroblastic subset expressing MT1-MMP in breast tissue samples, we have isolated fibroblastic cells from normal breast and mammary carcinomas. More than 60% of tumor-derived fibroblasts (BTF1 and BTF2) were intensively positive for αSM actin staining. In sharp contrast, a very weak immunostaining was detected in <40% of normal breast fibroblastic cells.
Cell extracts were analyzed by Western blotting with an antibody raised against the hemopexin domain of MT1-MMP (Fig. 3). Three different immunoreactive proteins were detected in BTF1 and BTF2 extracts: the 60 kDa activated MT1-MMP, as well as the 43 kDa and 37 kDa inactive species33 (Fig. 3). The activated MT1-MMP represented almost 60–70% of the enzyme detected. The 43 kDa form corresponding to 25–35% of the MT1-MMP species produced has been described as an inactive MT1-MMP that results from the processing of the 60 kDa form during MMP-2 activation.31, 32, 33 An additional processed form of MT1-MMP of 37 kDa33 represents 5–10% of MT1-MMP species. Normal fibroblasts also produced these 3 MT1-MMP species in similar proportion, but at lower concentrations. Indeed, normal fibroblasts produced 5–10 times less MT1-MMP than their tumoral counterpart (Fig. 3).
Zymography analysis of conditioned media by each cell types showed that BTF1 and BTF2 secreted important amounts of MMP-2 in its pro-form (72 kDa) and fully mature form (59 kDa) (Fig. 4). On sharp contrast, only a faint band of pro-MMP-2 was detected in the conditioned medium by normal breast fibroblasts (both NBF1 and NBF2). The MMP-2 production was 10–15 times less important in normal fibroblasts than in their tumoral counterpart. Interestingly, the active MMP-2 form represented 5–10% of total MMP-2 secreted by BTF vs. 0.1–1% in NBF supernatant.
Previous investigations on different types of cancer have shown that host stromal cells are the principal source of MMPs and that stromal MMP overexpression in vivo is closely related to tumour invasion.13, 19, 20, 21, 22, 34, 35 In breast cancer, the presence of MT1-MMP could be related to malignancy because its expression level is lower in normal tissues and benign proliferation than in both preinvasive and invasive areas.4, 12, 36 Although MT1-MMP has been reported by several groups as a characteristic feature of stromal fibroblastic cells associated with breast adenocarcinoma,12, 13, 35 the identity of these stromal cells was unknown. Over the last few years, it has become increasingly clear that there is a considerable heterogeneity in fibroblastic cell populations. On the basis of immunohistochemistry and in situ hybridization carried out on serial sections, we observe that MT1-MMP is confined to αSM actin positive cells in close contact to tumoral cells. Myofibroblasts typically express specific actin isoforms normally present only in smooth muscle cells. Because the MT1-MMP and αSM active positive cells were not directly associated with vascular structures, they were considered as myofibroblasts. Myofibroblastic differentiation of fibroblasts has been described in various processes such as tumor-stroma development, wound healing and fibrosis.37, 38 Myofibroblasts have been shown to originate primarily from fibroblasts and vascular smooth muscle cells.39 The possibility, however, that cancer cells also convert into myofibroblasts can not be excluded.40
Although the precise role of myofibroblasts is not yet well understood, their capacity to produce different proteases such as MMP-11,24 MMP-13,19 or urokinase plasminogen activator23 have been reported. By developing a computer-assisted image analysis, we were able to more precisely localize the MT1-MMP transcripts and to demonstrate, for the first time, that the subset of fibroblastic cells expressing MT1-MMP is exclusively myofibroblasts in contact to tumor cells. As shown by an original quantitative morphological method, the abrupt decrease of labeling density of MT1-MMP transcripts and αSM actin with the distance from tumor cells demonstrated the close association of MT1-MMP expressing myofibroblasts to tumor cells (Fig. 3). The density of the MMP-2-related staining remained more constant and decreased slowly demonstrating the large distribution of MMP-2 expression at distance from tumor cells. Taken together, these results demonstrate that the subset of fibroblastic cells expressing MT1-MMP is exclusively myofibroblasts in contact to tumor cells, whereas MMP-2 is produced by different fibroblastic subpopulations.
These in situ observations are consistent with our in vitro study demonstrating that tumor-derived fibroblasts expressed MT1-MMP and αSM actin at higher levels than cells derived from normal tissues. They also displayed a higher capacity to activate MMP-2. Interestingly, several factors known to regulate MMP expression have also been implicated in the myofibroblastic differentiation. For instance, cytokines released by cancer cells are likely to be involved in the activation of fibroblasts into myofibroblasts. Transforming growth factor β has been shown to induce αSM actin expression in cultured fibroblasts.41, 42 It is particularly noteworthy that cocultivation of fibroblasts with tumor cells induced the differentiation of fibroblasts into myofibroblasts43 and the upregulation of MMP-2 and MT1-MMP expression by fibroblasts.27, 30, 44 These findings are consistent with the observation that fibroblasts promoted tumor progression in animal models through their production of MMPs.30, 45, 46 In addition, our results confirm that stromal cells derived from malignant tumors are phenotypically and may be different genetically from fibroblasts of normal tissue origin.47, 48, 49, 50 They also indicate that tumor-derived fibroblasts keep at least some of their features in culture.
It is clear, therefore, that cancer cells cooperate with stromal cells to produce MT1-MMP that may participate in the invasion process through its capacity to activate pro-MMP-2. Previous studies both in animal models and on human tissue47, 48, 49, 50 have shown an association between MMP-2 expression and invasive or metastatic behavior, but it is an increase in the level of active MMP-2 that correlates with the invasive phenotype to a greater degree than total MMP-2 expression. Knowing the key role played by MT1-MMP in pro-MMP-2 activation, MT1-MMP distribution is thus likely to be a stronger indicator of MMP-2 activity and thereafter of tumor progression. This is consistent with the close association of MT1-MMP expression and invasive tumor cells. The larger distribution of MMP-2 than MT1-MMP mRNA suggests that MMP-2 produced by different subpopulation of fibroblastic cells is activated focally by MT1-MMP expressing myofibroblasts surrounding invasive cancer cells. Gilles et al.29 have also observed a similar focal coexpression of MT1-MMP and collagen Type I in a subpopulation of stromal cells close to tumor clusters. In correlation with in vitro data demonstrating that MT1-MMP is implicated in collagen-stimulated MMP-2 activation, the authors have suggested that this focal localization of MT1-MMP and collagen Type I represents hot spots for local degradation and invasive progression. In addition, it is worth noting that MMP-13 that is activated by MT1-MMP is produced by myofibroblasts and could contribute to the transition of ductal carcinoma in situ lesions to invasive carcinomas.19 These observations raise the possibility that MT1-MMP might contribute in tumor progression by activating pro-MMP-13.
Alternatively, despite the role of MT1-MMP in MMP-2 activation, the MT1-MMP could also participate in tumor progression through other mechanisms. The contribution of MT1-MMP in different physiological and pathological processes by mechanisms independent to its capacity to activate pro-MMP-2 is suggested by the more aggressive phenotype associated with MT1-MMP deficiency than that of MMP-2.3, 4, 5, 6, 7 In addition to the direct degradation of extracellular matrix components, MT1-MMP degrades important cell surface receptors such as tissue transglutaminase, integrins and CD44.51, 52 MT1-MMP could also be involved in signal transduction53 as suggested by the association between VEGF overproduction and MT1-MMP overexpression in glioma cells17 or breast adenocarcinoma cells.16
In conclusion, it appears evident that several stromal extracellular matrix degrading proteases such as MT1-MMP, MMP-11, MMP-13 and urokinase participate in a cooperative manner in cancer progression. Our in vivo and in vitro data, in accordance with other studies,19, 23, 24 strongly support the concept that the main source of these enzymes is myofibroblasts. We provide evidence that myofibroblasts in close association with tumoral cells produced MT1-MMP, emphasizing the importance of stromal-epithelial interface during cancer progression.
This work was supported by grants from the Communauté française de Belgique (Actions de Recherches Concertées), the European Commission (FP4 and FP5 programs), the Fonds de la Recherche Scientifique Médicale, the Fonds National de la Recherche Scientifique (FNRS, Belgium), the Fédération Belge Contre le Cancer, the C.G.R.I.-F.N.R.S.-INSERM Coopération, the Fonds spéciaux de la Recherche (University of Liège), the Centre Anticancéreux près l'Université de Liège, the FB Assurances, the Fondation Léon Frédéricq (University of Liège), the D.G.T.R.E. from the “Région Wallonne”, the Fonds d'Investissements de la Recherche Scientifique (CHU, Liège, Belgium), the Interuniversity Attraction Poles (I.U.A.P.) from the Federal Office for Scientific, Technical and Cultural Affairs (O.S.T.C., Brussels, Belgium), Rhône-Poulenc Rorer Pharmaceuticals (Collegeville, USA) and Roche Diagnostics GmbH (Penzberg, Germany).
- 33Mutation analysis of membrane type-1 matrix metalloproteinase (MT1- MMP). The role of the cytoplasmic tail Cys(574), the active site Glu(240), and furin cleavage motifs in oligomerization, processing, and self-proteolysis of MT1-MMP expressed in breast carcinoma cells. J Biol Chem 2001; 276: 25705–14., , , , , , .
- 52Processing of integrin αV subunit by MT1-MMP stimulates migration of breast carcinoma cells on vitronectin and enhances tyrosine phosphorylation of FAK. J Biol Chem 2001; 27: 27., , , , .