Tumor cell traffic through the extracellular matrix is controlled by the membrane-anchored collagenase MT1-MMP

The metastatic spread of tumor cells to distant organs via hematogeneous routes represents the most important cause of morbidity and mortality in cancer (Hanahan and Weinberg, 2000; Chambers et al., 2002). In order for malignant cells to intravasate into the host vasculature, neoplastic cells express a tissue-invasive phenotype that allows them to infiltrate an ECM largely composed of a dense, cross-linked meshwork of type I collagen fibrils (Hay, 1991).

To date, multiple proteinases have been identified that impact on tumor cell behavior in vivo as a consequence of their ability to cleave growth factors, growth factor receptors, cell-associated molecules, or chemokines/cytokines (Egeblad and Werb, 2002; Turk et al., 2004). However, with specific regard to tissue-invasive activity the required proteolytic machinery remains the subject of speculation. Current evidence suggests that cancer cells negotiate collagenous barriers by mobilizing redundant proteolytic systems that each serve to dissolve the interceding matrix, thus precluding a necessary role for any single proteinase (Brinckerhoff and Matrisian, 2002; Matrisian et al., 2003). Further, it has been proposed that even if ECM hydrolysis is inhibited, tumor cells can adopt an amoeboid phenotype that accommodates tissue trafficking by mechanical means alone (Friedl and Wolf, 2003).

Although the genomic instability that characterizes the neoplastic process may allow each distinct cancer cell to adopt its own unique strategy for negotiating the ECM, recent gene-profiling studies support a generic program for initiating the metastatic process (Ramaswamy et al., 2003). Indeed, tissue-infiltrating cancer cells of multiple origins have been noted to assume a mesenchymal cell-like phenotype similar to that displayed by fibroblasts (Thiery, 2002; Friedl and Wolf, 2003). Like tumor cells, fibroblasts are also endowed with the ability to remodel and infiltrate collagen-rich tissues as a function of their roles in growth, development, wound healing, and inflammation (Grinnell, 2003; Dong et al., 2004). Interestingly, the proteolytic or nonproteolytic systems engaged by this “professional” collagen-remodeling cell type also remain undefined, as does the possibility that normal and neoplastic cells share a conserved stratagem for invading the ECM.

Herein, we report that fibroblasts and tumor cells are able to tunnel through dense barriers of cross-linked type I collagen in vitro or in vivo via a virtually indistinguishable proteolytic process that requires matrix metalloproteinase (MMP) activity. Further, we demonstrate that fibroblasts deficient for the membrane-anchored metalloproteinase MT1-MMP are unable to display pericellular collagenolytic activity and are incapable of expressing a collagen-invasive phenotype in vitro or in vivo. Likewise, gene silencing of MT1-MMP in human carcinoma or fibrosarcoma cells similarly results in a complete loss of membrane-associated collagenolytic activity and renders tumor cells unable to infiltrate host tissues. These data identify MT1-MMP as a required pericellular collagenolysin that confers both normal and neoplastic cells with tissue-invasive activity.

After in vitro fibrillogenesis, type I collagen forms a gel-like network of interlocking fibrils stabilized by intermolecular, aldolimine cross-links similar to those generated in vivo (Hay, 1991). When cultured atop this three-dimensional substratum and exposed to a chemotactic gradient of PDGF-BB in the presence of serum, fibroblasts invade the collagen gel (Fig. 1 A). Tumor cells of mesenchymal or epithelial origin (e.g., HT-1080 and SCC-1 cells, respectively) also infiltrate the three-dimensional collagen barriers over the course of the 4-d culture period (Fig. 1 D). Coincident with the invasive process, both fibroblasts and tumor cells remodel the collagen substratum, leaving a network of tunnels that crisscross the traversed matrix (Fig. 1, B and E). Significantly, the channels formed are not the products of mechanical remodeling alone (Sawhney and Howard, 2002), as tunnel walls are lined by immuno-detectable collagen denaturation products (Fig. 1 C). Likewise, expression of the collagen-invasive phenotype by fibroblasts or tumor cells correlates with the accumulation of type I collagen degradation products (Fig. 1, F and G).

To identify the class of proteolytic enzymes responsible for driving this invasive/degradative process, fibroblasts or tumor cells were stimulated in the absence or presence of cysteine (E-64d), aspartyl (pepstatin A), serine (SBTI), or MMP (BB-94) inhibitors (Hiraoka et al., 1998; Hotary et al., 2000). Under these conditions, the synthetic MMP inhibitor BB-94 alone is able to block invasion (Fig. 1, A and D), tunnel formation (Fig. 1, B and E), and collagenolysis (Fig. 1, C, F, and G). Similar results are obtained with breast, pancreatic, or prostate carcinoma cell lines where BB-94 inhibits invasion and collagen degradation by more than 90% (Fig. S1). However, requirement for proteolytic activity during invasion is confined to physiologically relevant, cross-linked collagen gels. When matrices are prepared from pepsin-extracted collagen, which, unlike collagen networks assembled in vivo, cannot form intermolecular cross-links and are stabilized by noncovalent forces alone (Snowden and Swann, 1979; Capaldi and Chapman, 1982), tunnels are no longer observed in the wake of the migrating cells (Fig. 1 E) and BB-94 does not block fibroblast or tumor cell invasion (Fig. S2).

As fibroblasts and tumor cells both express collagenolytic activity during invasion, efforts were initiated to identify the proteolytic systems that mediate pericellular degradation. During matrix remodeling events in vivo, fibroblasts express MMP zymogens that can undergo processing to active forms (Brinckerhoff and Matrisian, 2002). Similar to the repertoire of MMPs expressed by fibroblasts activated in vivo (Okada et al., 1997; Bullard et al., 1999; Wu et al., 2002), fibroblasts stimulated with PDGF atop a three-dimensional collagen substratum in vitro express MMP-13, MMP-8, MMP-2, MT1-MMP, MT3-MMP, MMP-3, and MMP-9 as well as the MMP inhibitor TIMP-2 (Fig. 2 A). Further, wild-type cells process proMMP-2 to an active form as observed in vivo (Wang et al., 2000; Fig. 2 A).

When fibroblasts are stimulated atop a bed of rhodamine-labeled collagen fibrils in the presence of serum, the field of collagenolytic activity is confined to the cell–matrix interface where circulating antiproteinases may have limited access (Fig. 1 B). As observed during invasion (Fig. 1), pericellular collagen degradation is likewise sensitive to inhibition by BB-94 or TIMP-2 (Fig. 2, B and C) and unaffected by serine, cysteine, or aspartyl proteinase inhibitors (not depicted).

Murine fibroblasts potentially degrade subjacent collagen by mobilizing the direct-acting collagenolysins, MMP-13, MMP-8, MMP-2, or MT1-MMP (Brinckerhoff and Matrisian, 2002). In addition, MT1-MMP may also promote collagenolysis by catalyzing the processing of either proMMP-2, proMMP-8, or proMMP-13 to active enzymes (Holopainen et al., 2003; Seiki et al., 2003). To identify individual MMPs or MMP cascades that drive the collagenolytic phenotype, fibroblasts were isolated from MMP-13–, MMP-8–, MMP-2–, or MT1-MMP–deleted mice and the ability of resting or PDGF-stimulated cells to degrade type I collagen assessed. As shown in Fig. 2 A, each of the knockout cells express the full repertoire of MMPs detected in wild-type littermate cells, save for the targeted gene product. Though MMP-13 or MMP-8 are potent collagenolysins (Brinckerhoff and Matrisian, 2002) and can be detected in fibroblast releasates (not depicted), fibroblasts null for either of these MMPs degrade collagen comparably to control cells in either the absence or presence of PDGF (Fig. 2, B and C). Similarly, although recent reports posit a preeminent role for MMP-2 in degradative events after its TIMP-2–dependent activation by MT1-MMP (Brinckerhoff and Matrisian, 2002; Seiki et al., 2003), MMP-2^−/− as well as TIMP-2^−/− fibroblasts also mount normal collagenolytic responses, effectively eliminating a required role for the MT1-MMP/TIMP-2/MMP-2 axis in this proteolytic process (Fig. 2, B and C). MMP-3 has been implicated in the activation of both procollagenases and proMMP-9 (Benbow et al., 1999; Ramos-DeSimone et al., 1999), but neither MMP-3^−/− nor MMP-9^−/− fibroblasts display defects in the collagenolytic phenotype (Fig. 2, B and C).

As fibroblasts null for each of the major secreted collagenases degrade collagen in an area restricted to the pericellular space, we next considered a direct attack by the membrane-anchored metalloprotease MT1-MMP. Indeed, MT1-MMP–deleted fibroblasts display an almost complete defect in collagenolytic activity relative to wild-type cells though the cells express MMP-13, MMP-8, MMP-2, and TIMP-2 comparably to control fibroblasts and retain a residual ability to process proMMP-2 to its active form (Fig. 2, A–C), presumably via MT3-MMP (Hotary et al., 2002). Consistent with a direct role for MT1-MMP in the collagenolytic phenotype, collagen degradation by wild-type fibroblasts was blocked completely by either BB-94 or exogenous TIMP-2, MMP inhibitors whose substrate profile includes the membrane-anchored MMPs (Fig. 2, B and C), but not by high concentrations of TIMP-1, an MMP inhibitor that preferentially targets secreted MMPs (not depicted; Hotary et al., 2002, 2003). The collagenolytic-null phenotype of MT1-MMP^−/− fibroblasts is reversed by reexpressing full-length MT1-MMP, but not by an MT1-MMP construct harboring an inactivating point mutation in the catalytic domain (i.e., MT1-MMP E²⁴⁰→A; Fig. 2 B). Consistent with the restricted ability of secreted proteinases to operate effectively in the presence of serum antiproteinases, the collagenolysis-null phenotype of MT1-MMP^−/− fibroblasts cannot be rescued when cells are transfected with a fully active, but soluble, form of MT1-MMP (i.e., ΔMT1-MMP; Fig. 2 B). Though mouse adult fibroblasts can express low levels of a species-specific orthologue of mammalian MMP-1, termed mColA (Balbin et al., 2001), subjacent proteolysis is not affected by transfecting wild-type or MT1-MMP^−/− cells with a mColA expression vector (Fig. 2 B).

Importantly, the inability of MT1-MMP^−/− fibroblasts to degrade collagen in a serum-containing environment is not due to a generalized defect in proliferation or the mobilization or activation of secreted collagenases. Although MT1-MMP can affect cell proliferation (Hotary et al., 2003), this effect is limited to three-dimensional growth and, as such, wild-type and null fibroblasts replicate at indistinguishable rates atop the collagen substratum (at day 5, 2.1 ± 0.3 × 10⁵ and 1.9 ± 0.2 × 10⁵ cells for wild-type and MT1-MMP^−/− fibroblasts, respectively). Further, when MT1-MMP^+/+ or MT1-MMP^−/− cells are triggered under serum-free conditions and the medium supplemented with exogenous plasminogen, both wild-type and null cells extend collagenolytic activity beyond their cell borders in a manner consistent with the activation of secreted collagenases (Fig. 2 B; Netzel-Arnett et al., 2002) and degrade collagen comparably (i.e., wild-type and MT1-MMP^−/− fibroblasts solubilize 8.7 ± 1.1 and 7.9 ± 1.0 μg of hydroxyproline, respectively). Nonetheless, in an antiproteinase-rich milieu that recapitulates that encountered in vivo, membrane-anchored MT1-MMP serves as the dominant collagenase.

Tumor cells, like fibroblasts, also express secreted and membrane-anchored MMPs (Seiki et al., 2003) and efficiently degrade subjacent collagen via a BB-94–sensitive process (Fig. 3 A). Consistent with a role for MT1-MMP in pericellular collagenolysis, high dose TIMP-1 is unable to block degradation, whereas proteolysis is inhibited completely by TIMP-2 (Fig. 3 A). To determine directly the role of MT1-MMP in the tumor cell collagenolytic phenotype, the expression of the membrane-anchored enzyme was targeted with a siRNA construct in HT-1080 or SCC-1 cells. As shown in Fig. 3 B, MT1-MMP is silenced effectively without affecting MMP-2 or MMP-9 expression, though, as expected, the MT1-MMP–dependent processing of proMMP-2 to its active form is suppressed. Further, like MT1-MMP^−/− fibroblasts, MT1-MMP siRNA-treated cancer cells display no defects in two-dimensional proliferation or migration across collagen-coated surfaces (Fig. S3). However, in the absence of endogenous MT1-MMP, the pericellular collagenolytic activity of HT-1080 cells or SCC-1 cells was lost completely (Fig. 3 C). Further, the inability of MT1-MMP siRNA-treated tumor cells to mediate subjacent collagenolysis is not reversed when the cells are transfected with chimeric constructs of human MMP-1, MMP-13, or MMP-2 that undergo constitutive processing to active forms (i.e., MMP-1_RXKR, MMP-13_RXKR, or MMP-2_RXKR expression vectors; Hotary et al., 2003; Fig. 3 C). Although siRNAs can potentially induce nonspecific effects by mediating off-target gene regulation (Scacheri et al., 2004), collagenolytic activity is restored completely when the MT1-MMP siRNA-treated tumor cells are transfected with a mouse MT1-MMP construct that displays only limited homology through the human MT1-MMP siRNA-targeted sequence (Fig. 3 C). Thus, cancer cells, like fibroblasts, rely on MT1-MMP to express subjacent collagenolytic activity.

Although fibroblasts and tumor cells both use MT1-MMP as a pericellular collagenolysin, the invasive phenotype displayed by cells negotiating three-dimensional ECM barriers is a more complex form of cellular behavior and conceivably requires complementation by other proteolytic systems (Murphy and Gavrilovic, 1999). Hence, monolayers of wild-type and null fibroblasts were established atop three-dimensional collagen gels and the cells exposed to a chemotactic gradient of PDGF. Under these conditions, MMP-13^−/−, MMP-8^−/−, MMP-2^−/−, TIMP-2^−/−, MMP-3^−/−, and MMP-9^−/− fibroblasts each invade the collagen gels comparably to wild-type cells while degrading similar amounts of collagen (Fig. 4, A–C). Likewise, though plasmin can activate secreted MMPs, fibroblasts isolated from plasminogen-null mice and suspended in plasminogen-null sera display an invasive phenotype indistinguishable from controls (i.e., 98 ± 10% of control; n = 3). Recent papers have promoted a potential role for the transmembrane serine prolyl peptidase, fibroblast activation protein/seprase, in collagen remodeling events (Ghersi et al., 2002), but the null fibroblasts also retained full invasive activity (i.e., and 95 ± 6% of control; n = 3). In contrast, MT1-MMP^−/− fibroblasts are unable to express an invasive phenotype over the 6-d culture period (Fig. 4, A–C). Though cell behavior can be altered when cells are embedded within, as opposed to cultured atop, a three-dimensional matrix (Cukierman et al., 2002, Hotary et al., 2003), MT1-MMP^−/− cells sustain a similar, and complete, deficiency in three-dimensional invasive behavior relative to wild-type fibroblasts (Fig. 4 A). However, under either two- or three-dimensional conditions, the MT1-MMP^−/− invasion-null phenotype is rescued when cells are transfected with an expression vector encoding wild-type (Fig. 4 A), but not either soluble ΔMT1-MMP, inactive MT1-MMP E²⁴⁰→A, mColA, human MMP-1_RXKR, human MMP-13_RXKR, or human MMP-2_RXKR (not depicted). However, defects in MT1-MMP^−/− fibroblasts are confined to the type I collagen substrate as the null cells migrate across collagen-coated surfaces comparably to controls, exert a normal collagen-contractile response (Fig. S4), and, as reported previously, invade three-dimensional matrices of cross-linked fibrin via an MT3-MMP–dependent process (Hotary et al., 2002).

Although fibroblasts use MT1-MMP to cross barriers of type I collagen assembled in vitro, the extracellular matrix in vivo is a more complex mix of collagens, glycosaminoglycans, glycoproteins, and peptidoglycans (Hay, 1991). To determine whether or not MT1-MMP likewise regulates tissue-invasive activity in vivo, wild-type or null fibroblasts labeled with fluorescent nanobeads were applied to the surface of the chicken chorioallantoic membrane (CAM), a type I collagen-rich ECM barrier commonly used to study invasive processes (Kim et al., 1998). As reported previously (Armstrong et al., 1982), wild-type fibroblasts readily invade the CAM in the course of a 3-d culture period (Fig. 5 A). Though the pattern of distribution of the invading fibroblasts is variable between tissue sections and individual chick CAMs (unpublished data), both the percentage of fibroblasts that cross the CAM surface and the depth of the leading front of invasion is highly reproducible. By either of these criteria, neither MMP-13^−/−, MMP-8^−/−, MMP-2^−/−, MMP-3^−/−, MMP-9^−/−, plasminogen^−/−, nor fibroblast activation protein^−/− fibroblasts display discernible defects in invasive activity (Fig. 5, A and B). In contrast, MT1-MMP-null fibroblasts are completely unable to penetrate the CAM surface unless the cells are transfected with wild-type, but not a MT1-MMP E²⁴⁰→A, expression vector (Fig. 5, B and C).

To determine if a required role for MT1-MMP in fibroblast invasion extends to tumor cells, collagen-invasive activity of the MT1-MMP siRNA-treated cancer cells was assessed in vitro and in vivo. As shown in Fig. 6, type I collagen invasion and the associated degradation of the surrounding matrix by HT-1080 or SCC-1, although unaffected by the control siRNA, are suppressed almost completely by the MT1-MMP siRNA (Fig. 6, A and B). Validated siRNA-mediated targeting of either tumor cell MMP-1, MMP-2, or MMP-9 did not affect collagen-invasive activity (unpublished data). A role for MT1-MMP in tumor cell invasion was not limited to HT-1080 or SCC-1 cells, as similar results are obtained when MT1-MMP expression is silenced in MDA-MB-231, PANC-1, or DU-145 cells where the number of invasive foci is reduced by more than 75% (Fig. S5). MT1-MMP siRNA-treated tumor cells retain full ability to invade either pepsin-extracted collagen gels devoid of intermolecular cross-links or fibrin matrices (Fig. S3). Similarly, (though in contrast with other recent reports; Ueda et al., 2003; Iida et al., 2004) invasion through Matrigel, an extract of noncross-linked ECM macromolecules (Kleinman et al., 1982), is unaffected in MT1-MMP–targeted tumor cells (Fig. S3). However, invasion through cross-linked collagen matrices and the associated collagenolytic activity are restored fully by transfecting the MT1-MMP siRNA-treated tumor cells with the mouse MT1-MMP construct (Fig. 6, A and B).

Unlike fibroblasts, tumor cells transplanted atop the CAM not only invade the surrounding tissues but gain access to the host vascular and/or lymphatic beds where human cell DNA can be detected as Alu sequences via PCR-based assay on extracts prepared from the lower CAM (Kim et al., 1998). To determine if MT1-MMP plays a necessary role in driving cancer cell invasion and intravasation, tumor cells were treated with either siRNA controls or MT1-MMP siRNA, and the in vivo phenotype was monitored. After a 3-d incubation period, HT-1080 and SCC-1 cells treated with siRNA controls readily invade the CAM surface (Fig. 6 C). The invasive phenotype was not limited to the chick extracellular matrix, as tumor cells cultured atop acellular human dermis and explanted onto the CAM likewise penetrated this type I/III collagen-rich barrier (Fig. 6 D). After penetrating the CAM interstitium, the tumor cells further breach the vascular bed and gain entry to the lower CAM as detected by Alu sequence (Fig. 6 E). In contrast, MT1-MMP siRNA-treated cells are unable to express invasive activity and fail to access the lower CAM unless transfected with mouse MT1-MMP (Fig. 6, C and E). Neither siRNAs directed against MMP-2 nor MMP-1 inhibit significantly tumor cell invasion (i.e., percent invasion for control siRNA-, MMP-2 siRNA-, and MMP-1 siRNA-treated HT-1080 cells is 23.9 ± 6.7%, 22.2 ± 6.3%, and 21.2 ± 2.6%, respectively). These data support a necessary, if not critical, role for MT1-MMP in the metastatic process by arming cancer cells with the ability to invade interstitial tissues and gain access to the host vasculature.

In vivo, cancer cells initiate metastatic behavior by penetrating the type I collagen-rich ECM, infiltrating vascular or lymphatic compartments, and disseminating to distant sites (Hanahan and Weinberg, 2000; Chambers et al., 2002). Multiple proteolytic systems have been shown to impact on tumor cell behavior in vivo (Matrisian et al., 2003; Turk et al., 2004), but the identity of the enzymes that confer tissue-invasive activity remain controversial. As tumor cells engage the proteolytic machinery necessary to invade surrounding tissues, they adopt a fibroblast-like morphology—a phenotypic switch that suggests that tumor cells access purposefully the same gene programs engaged by normal cells that traffic through the ECM (Thiery, 2002; Friedl and Wolf, 2003). Herein, we demonstrate that the membrane-anchored metalloproteinase MT1-MMP serves as the dominant purveyor of the tissue-invasive activities necessary to support the trafficking of normal and neoplastic cells through the ECM. Although fibroblasts as well as tumor cells also mobilize secreted collagenases, which may associate with the cell surface (Dumin et al., 2001; Netzel-Arnett et al., 2002), such binding interactions were unable to confer these cells with invasive activity. Presumably, only MT1-MMP is directed specifically to sites targeted for degradation by a combination of motifs embedded not only in the transmembrane domain and intracellular cytosolic tail but also the collagen-binding hemopexin domain on the extracellular face of the enzyme (Tam et al., 2002; Seiki et al., 2003). This conclusion should not be construed to suggest that secreted collagenases are inoperative in vivo. Rather, the ability of secreted collagenases to resorb bulk tissue would be restricted to those sites where the antiproteinase shield is overwhelmed by a relative proteinase excess or inactivated by either reactive oxidants or proteolytic attack (Weiss, 1989).

The ability of fibroblasts and tumor cells to mobilize a pericellular collagenase would appear to provide a mechanistic basis for the tunnel tracks we detected in cell-trafficked collagen matrices. Indeed, the unique ability of MT1-MMP, rather than secreted MMPs, to drive invasion correlates precisely with the singular role that the membrane-anchored proteinase plays in pericellular collagenolysis. Although invasion may require MT1-MMP to cleave collagen as well as one or more cell surface targets in tandem (Seiki et al., 2003), it should be noted that MT1-MMP-null cells display no defects in collagen adhesion, collagen-contractile activity, or migration, and invade three-dimensional gels constructed of cross-linked fibrin or noncross-linked collagen comparably to wild-type cells. Furthermore, invasion is inhibited when tumor cells confront barriers of a collagenase-resistant mutant form of type I collagen (Ellerbroek et al., 2001; Hotary et al., 2003). Finally, we have found that invasive activity can be conferred by a chimeric MT1-MMP construct wherein the catalytic domain of the wild-type enzyme has been replaced with that of the “classic” collagenase MMP- 1 (unpublished data). Together, these results support a model where MT1-MMP confers fibroblasts and tumor cells with invasive activity by acting as a membrane-tethered collagenase.

In vivo, tissue-invasive fibroblasts and cancer cells not only negotiate collagenous barriers, but the more complex mix of ECM-associated macromolecules that constitute the interstitial matrix (Hay, 1991). When applied to the CAM surface, wild-type fibroblasts and cancer cells both cross the CAM surface and invade the underlying mesenchyme via a process dependent completely on MT1-MMP—a finding consistent with the ability of MT1-MMP to cleave multiple interstitial collagen types, glycoproteins and proteoglycans (Seiki et al., 2003). However, carcinoma as well as fibrosarcoma cells not only invade locally but also intravasate into the chick vasculature and/or lymphatic bed (Kim et al., 1998). Despite the more aggressive phenotype of the transformed cells, the invasive activity of cancer cells is likewise paralyzed by silencing MT1-MMP expression. Although the specificity of siRNA-mediated silencing must be interpreted cautiously (Scacheri et al., 2004), we note that the inhibitory effects exerted on tumor cells are reversed by reexpressing an MT1-MMP construct designed to escape siRNA targeting and that the observed tumor cell phenotype recapitulates that of MT1-MMP^−/− fibroblasts. Hence, contrary to expectations that multiple proteolytic systems can confer invasive activity (Egeblad and Werb, 2002; Matrisian et al., 2003; Turk et al., 2004), in the absence of MT1-MMP, none of the five cancer cell lines studied herein are able to mobilize any of their remaining repertoire of serine-, cysteine-, aspartyl-, or metalloproteinases alone, or in combination with chick proteinases, to drive invasion or intravasation.

Although our findings highlight the importance of proteolytic remodeling of the ECM during invasion, recent attention has focused on the ability of tumor cells to potentially negotiate collagenous barriers by switching to an amoeboid-like form of movement that is permissive for the proteinase-independent patterns of migration more commonly associated with myeloid cells (Huber and Weiss, 1989; Friedl and Wolf, 2003; Wolf et al., 2003). However, we note that tumor cell amoeboid movement was observed in vitro within the confines of three-dimensional gels constructed from pepsin-extracted collagen that no longer contains the telopeptide region critical for the structural integrity of type I collagen gels (Snowden and Swann, 1979; Capaldi and Chapman, 1982). Similarly, nonproteolytic systems have been reported to confer tumor cells with the ability to traverse Matrigel-coated surfaces (Hotary et al., 2000; Sahai and Marshall, 2003), a noncross-linked composite of basement membrane-associated macromolecules (Kleinman et al., 1982). As collagens found in the interstitial matrix are normally cross-linked in vivo (Hay, 1991), we posit that MT1-MMP plays a dominant role in most tissue-invasive processes as MT1-MMP–suppressed tumor cells were unable to negotiate type I collagen gels, the CAM interstitial matrix, or human dermis. This conclusion remains at odds with reports that MMP inhibition fails to block tumor cell migration in vivo (Wolf et al., 2003) and that MT1-MMP^−/− mice develop normally (Holmbeck et al., 1999). First, with respect to the amoeboid behavior of tumor cells observed in vivo, HT-1080 cells were treated with reversible MMP inhibitors in vitro and antiproteinases were washed away before in vivo analysis (Wolf et al., 2003). Hence, HT-1080 cells would be free to mobilize MT1-MMP and initiate invasion. Second, we find that the concentrations of MMP inhibitors required to block tumor cell migration in our long term in vitro studies are far higher than those reached in either animal studies or clinical trials (Matrisian et al., 2003). Third, although MT1-MMP^−/− mice display no discernible defects at birth, the synthesis and maturation of type I collagen fibrils is confined largely to the postnatal period (Caulfield and Borg, 1979; Mays et al., 1988). Indeed, we have found that MT1-MMP^−/− fibroblasts can navigate through the type I collagen-poor neonatal dermis, but remain entrapped in the type I collagen-rich dermis of 4-wk-old mice (unpublished data). Although collagen content and structure can differ dramatically from tissue to tissue even during postnatal states (Hay, 1991), gene-expression profiles demonstrate that type I collagen is up-regulated consistently at both primary and metastatic sites in a wide variety of human carcinomas (Ramaswamy et al., 2003). Hence, MT1-MMP may play a required role for invasion even at sites where the collagen content of the normal, pre-affected tissue is low. Further, while our work emphasizes the importance of MT1-MMP, preliminary studies indicate that MT2-MMP may also bestow recipient cells with tissue-invasive activity (Hotary et al., 2000; unpublished data). However, the more restricted pattern of MT2-MMP expression in vivo suggests that this alternate collagenolysin likely plays a more limited role than MT1-MMP in regulating the invasive phenotype.

For more than 30 yr, various proteinases have been linked to tumor invasion, intravasation, or metastasis (Brinckerhoff and Matrisian, 2002; Matrisian et al., 2003; Turk et al., 2004). However, with increasing frequency, proteolytic enzymes appear to regulate tumor cell behavior in an indirect fashion, e.g., by solubilizing matrix-bound growth factors, or by processing cadherins, integrins, and chemokines (Egeblad and Werb, 2002; Seiki et al., 2003; Turk et al., 2004). In contrast, our data, coupled with more recent findings (Hotary et al., 2003), support a model wherein MT1-MMP functions as a direct-acting, pro-invasive factor that confers tumor cells with the ability to penetrate connective tissue barriers, proliferate within the confines of the three-dimensional ECM, and initiate the metastatic process. Given the widespread distribution of MT1-MMP both at the invading front of cancer cells and in the surrounding stroma (Seiki et al., 2003; Hlubek et al., 2004) as well as the role of tissue-infiltrating fibroblasts in supporting tumor growth (Dong et al., 2004), we posit that MT1-MMP serves as the key protease mobilized by normal and neoplastic cells as they traverse and remodel ECM barriers in the in vivo setting.

Fibroblasts were isolated from dorsal dermal explants of 2–4-wk-old mice with targeted deletions in MT1-MMP (Holmbeck et al., 1999), MMP-2 (Itoh et al., 1998; provided by S. Itohara, RIKEN Brain Science Institute, Saitama, Japan), MMP-9 (Liu et al., 1998; provided by R.M. Senior and J.M. Shipley, Washington University, St. Louis, MO), MMP-3 (Mudgett et al., 1998), MMP-8 (Balbin et al., 2003), seprase/fibroblast activation protein (Niedermeyer et al., 2000; provided by E. Ostermann and E. Patzelt, Boehringer Ingelheim, Vienna, Austria), plasminogen (Ploplis et al., 1995; provided by T. Bugge, National Institutes of Health, Bethesda, MD), or TIMP-2 (Wang et al., 2000) as well as their wild-type littermates as described previously (Havemose-Poulsen et al., 1998). MMP-13^−/− fibroblasts were recovered from knockout mice whose characterization will be described elsewhere. Fibroblasts (passage 2–6) were cultured in DME supplemented with either 10% heat-inactivated FCS (Atlanta Biologicals) or heat-inactivated mouse sera, 100 U/ml penicillin, 100 μg/ml streptomycin, 0.25 μg/ml fungizone, and 2 mM l-glutamine. Human HT-1080 fibrosarcoma cells (American Type Culture Collection), SCC-1 squamous carcinoma cells (provided by T. Carey, University of Michigan, Ann Arbor, MI), MDA-MB-231 breast carcinoma cells (American Type Culture Collection), PANC-1 pancreatic carcinoma cells (American Type Culture Collection), and DU-145 carcinoma cells (American Type Culture Collection) were cultured as described previously (Hotary et al., 2003).

Fibroblasts were cultured atop type I collagen in the presence and absence of 10 ng/ml PDGF in media supplemented with 10% serum, and total RNA was isolated using TRIzol reagent (GIBCO BRL). RT-PCR was performed with 1 μg of total RNA and 10 μM of specific primers (Kim et al., 1998; Hotary et al., 2002) using One-Step RT-PCR System reagent (Life Technologies). The identity of each PCR product was confirmed by sequence analysis.

Fibroblasts or tumor cells were transiently transfected with control vector (PCR3.1 Uni; Invitrogen), mColA, mouse MT1-MMP (gift of M. Seiki, University of Tokyo, Tokyo, Japan), human MT1-MMP, catalytically inactive human MT1-MMP (EA mutant, Glu 240→Ala), transmembrane-deleted human MT1-MMP (ΔMT1-MMP), or expression vectors encoding proprotein convertase-activatable forms of human MMP-1, MMP-2, or MMP-13 (MMP-1_RXKR, MMP-2_RXKR, and MMP-13_RXKR) using Fugene6 (Roche) as described previously (Hotary et al., 2003).

Collagen gel films (∼100 μg distributed uniformly over an area of 2 cm²) were incubated with 0.05 M of sodium borate buffer, pH 9.0, for 1 h and labeled with tetramethylrhodamine isocyanate (Molecular Probes) for 45 min. Fluorescence images of rhodamine-labeled collagen were captured by confocal microscopy. Fibroblasts or tumor cells (5 × 10⁴) were added to the collagen monolayer and incubated for either 5 or 3 d, respectively. Where indicated, cells were labeled with either phalloidin, calcein-AM, or DAPI (Molecular Probes). Collagen degradation products were quantified by hydroxyproline release after an ethanol precipitation step (70% vol/vol; Creemers et al., 1997). Degradation assays were performed in the presence of 10% FCS, in autologous mouse sera (i.e., from wild-type, MMP-2^−/−, MMP-9^−/−, or TIMP-2^−/− mice) or where indicated, under serum-free conditions in the presence of 20 μg/ml of plasminogen (Enzyme Research). Results are presented as the mean ± 1 SEM of at least three experiments.

Type I collagen was acid or pepsin extracted from rat or mice tail tendons (Birkedal-Hansen, 1987; Hotary et al., 2000). Alternatively, pepsin-extracted bovine dermal collagen (Vitrogen; Nutacon) was commercially obtained. Collagen gels were prepared in 24-mm Transwell dishes (3-μm pore size; Corning, Inc.) with 1 ml of type I collagen used at a final concentration of 2.2 mg/ml (Hotary et al., 2000). After gelling (45 min at 37°C), 1.5–2 × 10⁵ cells in media supplemented with either 10% heat-inactivated FCS or 10% mouse serum (from either wild-type, TIMP-2^−/−, plasminogen^−/−, MMP-2^−/−, or MMP-9^−/− mice) were added to the upper chamber. PDGF-BB (Upstate Biotechnology) was added to the lower compartment of the Transwell chambers to initiate fibroblast invasion or 50 ng/ml hepatocyte growth factor (HGF; Genentech, Inc.) to trigger tumor cell invasion. Where indicated, gels and media were supplemented with recombinant tissue inhibitor of metalloproteinases (7.5 μg/ml TIMP-1 or 2.5 μg/ml TIMP-2; Fuji Chemical Industries), the synthetic MMP inhibitor BB-94 (5 μM final concentration in 0.1% DMSO; British Biotechnology), the cysteine proteinase inhibitor, E-64d (100 μM; Sigma-Aldrich), the aspartate proteinase inhibitor pepstatin A (50 μM; Sigma-Aldrich), or the serine proteinase inhibitors, soybean trypsin inhibitor, SBTI (100 μg/ml) or aprotinin (200 μg/ml; Roche). Invasive activity was visualized by either phase-contrast microscopy or confocal laser microscopy of rhodamine-labeled type I collagen gels. The number of invading cells per high powered field and leading front depth of invasion (three or more cells) were quantified as the mean ± 1 SEM of at least three experiments as described previously (Hotary et al., 2002). Collagen degradation was monitored by hydroxyproline release as described in the section Collagen degradation assays.

To analyze the three-dimensional invasive activity of fibroblasts or tumor cells, 10⁵ cells were embedded in 100 μl of collagen in a 96-well plate and cultured for 24 h. The collagen-cell plugs were then transferred to 24-well plates and further embedded in 1 ml of collagen gels and cultured for 7 d. Cell migration from the central plug into the surrounding collagen was monitored by phase-contrast microscopy.

Gelatin zymography for secreted MMP-2 and MMP-9 and Western blot analysis were performed as described previously (Hotary et al., 2003).

The antisense strand of siRNA was targeted against a 21-nt MT1-MMP sequence (5′-AACAGGCAAAGCTGATGCAGA-3′; nt 228–248). The nucleotide sequence was scrambled to generate a siRNA control sequence (5′-AAGTGATCAAGCACCGAAGAG-3′). hMMP-1, hMMP-2, and hMMP-9 were targeted using 21-nt sequences (5′-AAGATGTGGACTTAGTCCAGA-3′, nt 86–106 for hMMP-1; 5′-AATACCATCGAGACCATGCGG-3′, nt 274–294 for hMMP-2; and 5′-AAGGAGTACTCGACCTGTACC-3′, nt 1066–1086 for hMMP-9). siRNA oligonucleotides (QIAGEN) were introduced into tumor cells (50–100 nM) using a nucleofector kit and electroporation (Amaxa Biosystems). Using fluorescent oligonucleotides, ∼99% of the tumor cells were transfected. 20 h after electroporation, siRNA knockdown was complete for up to 72 h.

In vivo fibroblast or cancer cell invasion and intravasation assays were conducted using 11-d-old chick embryos wherein wild-type or null fibroblasts, HT-1080 cells, or SCC-1 cells (10⁵ cells labeled with fluoresbrite 45-nm-diam carboxylate microspheres; Polysciences) were seeded atop the CAM for 3 d as described previously (Kim et al., 1998; Chambers et al., 2002). The percentage of invading cells was quantified as a function of the percentage of the total cell-associated fluorescence localized beneath the CAM surface in five or more randomly selected fields (ImageQuant [version 5.2]; Molecular Dynamics Inc.). Depth of invasion from the CAM surface was defined as the leading front of three or more invading cells in 10 randomly selected fields.

In selected experiments, HT-1080 cells were cultured atop an acellular fragment of human dermis (Alloderm; Life Cell), and then explanted atop the CAM for 3 d to initiate invasion.

Frozen sections of fibroblast-collagen gel cocultures were incubated with a polyclonal antibody against the denatured fragment of type-I collagen (gift of R. Poole, McGill University, Montreal, Canada) as described previously (Hollander et al., 1995). Collagen gels were prepared for transmission EM as described previously (Hotary et al., 2002).

Confocal imaging was performed on a laser scanning fluorescent microscope (model LSM 510; Carl Zeiss MicroImaging, Inc.) equipped with acquisition software (Version 3.2 Service Pack 2) using either 40× or 63× C-Apochromat (1.2 NA) water immersion objectives at 25°C. Samples were mounted in fluorescent mounting media (DakoCytomation). Triple-color imaging of DAPI, Alexa 488-phalloidin or calcein-AM, and rhodamine-red was obtained at 364 nm, 488 nm, and 543 nm, respectively, for selective excitation.

All other bright-field, fluorescent, or phase-contrast images were captured using either an upright fluorescent microscope (Leica; 40× objective/0.7 NA) or inverted microscopy (Leica; 10× objective/0.3 NA), and digitally acquired using a Spot digital camera (Software v.3.3; Diagnostic Instruments) at RT. Image-processing software (Photoshop 7; Adobe) was used to overlay images and to enhance equally image color and clarity.

Figs. S1–S5 are available in the online supplemental material. Fig. S1 shows the effects of BB-94 on collagen invasion and degradation by carcinoma cell lines. Fig. S2 shows the effect of BB-94 on invasion through three-dimensional gels of pepsinized collagen. Fig. S3 shows the effects of MT1-MMP siRNA on tumor cell migration, growth, morphology, and invasion. Fig. S4 shows migration and contractile activity of MMP-null fibroblasts. Fig. S5 shows type I collagen invasion by MT1-MMP siRNA-treated carcinoma cells.

We thank Robert M. Senior and J. Michael Shipley for providing access to the MMP-9-null mice, S. Itohara for the MMP-2-null mice, E. Ostermann and E. Patzelt for the fibroblast activation protein-null mice, and Thomas Carey for SCC-1 cells.

This work was supported by National Institutes of Health grants R01 CA071699 (S.J. Weiss), R01 CA088308 (S.J. Weiss), and PO1 HL29594 (S. Shapiro).