Induction of tissue inhibitor of metalloproteinases-3 is a delayed early cellular response to hepatocyte growth factor

Hepatocyte growth factor (HGF) stimulates mitogenic, motogenic, and morphogenic responses in various cell types. We analysed HGF-responsive cells by differential display PCR to identify HGF-induced genes that mediate these biological events. One of the genes identified encoded a member of the tissue inhibitor of metalloproteinases (TIMP) family, TIMP-3. HGF transiently induced TIMP-3 mRNA in keratinocytes as well as kidney and mammary epithelial cells maximally between 4 and 6 h post-stimulation. Increased TIMP-3 protein secretion returned to basal levels within 18 h, while the expression of gelatinases A and B remained unchanged, suggesting that temporary suppression of matrix degradation is a delayed early response to HGF. Ectopic overexpression of TIMP-3 in cultured leiomyosarcoma cells conferred an epithelial morphology, reduced cell growth rate, anchorage-independent growth, and matrix invasion in vitro. Antisense suppression of TIMP-3 was associated with a scattered, fibroblastic cell morphology, as well as enhanced proliferation, anchorage-independent growth, and matrix invasion. A survey of tumor cell lines revealed an inverse relationship between metastatic potential and TIMP-3 expression level. These data suggest that early, transient TIMP-3 expression mediates specific HGF-induced phenotypic changes, and that loss of TIMP-3 expression may enhance the invasion potential of certain tumors.


Introduction
Hepatocyte growth factor (HGF) was originally identi®ed as a potent mitogen for hepatocytes (reviewed in Michalopoulos and DeFrances, 1997) but also stimulates the proliferation of endothelial cells and a variety of epithelial cell types such as keratinocytes, melanocytes, and kidney epithelium (reviewed in Rubin et al., 1993;Zarnegar and Michalopoulos, 1995). HGF also stimulates cell scattering (Stoker and Perryman, 1985), branching tubulogenesis (Montesano et al., 1991), cellular invasion through extracellular matrix substrates (Giordano et al., 1993;Rong et al., 1994) and promotes tumor metastasis (Weidner et al., 1990;Jeers et al., 1996). Multiple mRNA species tran-scribed from a single HGF gene encode three distinct proteins: full-length HGF and two truncated HGF isoforms known as NK1 and NK2, which consist of the N-terminal domain linked in tandem with the ®rst one or two kringle domains, respectively. Both truncated isoforms retain scatter activity, but their mitogenic activities dier substantially: while NK1 is mitogenically active with nearly the potency of HGF, NK2 displays little or no activity, and can potently antagonize HGF-stimulated DNA synthesis Cioce et al., 1996).
We have surveyed HGF-responsive cell lines by dierential display PCR (DD ± PCR) to identify delayed early gene induction events that correlated with speci®c HGF biological activities. DD ± PCR analysis of HGF-treated MDCK cells revealed the transient induction of tissue inhibitor of metalloproteinases-3 (TIMP-3). The speci®c biological impact of TIMP-3 expression was assessed by altering TIMP-3 levels in the absence of HGF treatment. Striking changes were observed in cell morphology, anchorage-independent cell growth, and cellular invasiveness in vitro that suggest a mechanism by which sustained HGF stimulation may promote cellular invasion and metastasis in certain human cancers.

HGF-regulation of TIMP-3 transcription
Reverse transcribed mRNA from MDCK cells that had been stimulated with either HGF or HGF/NK2 for 6 h was subjected to DD ± PCR. Dierentially displayed bands were isolated from the gel, ream-pli®ed, and cloned. One of the clones encoded a member of the tissue inhibitor of metalloproteinases family (TIMP), TIMP-3. TIMP-3 mRNA level showed a dose-dependent increase in response to HGF treatment of MDCK cells (Figure 1a). HGF induction of TIMP-3 mRNA was observed at concentrations as low as 10 ng/ml of HGF, with maximal induction (tenfold) observed at 300 ng/ml. HGF/NK2 stimulated a lower level of TIMP-3 mRNA induction (approximately threefold) at much higher ligand concentrations (1000 ng/ml; Figure 1a). It is noteworthy that HGF/NK2 activates c-Met but does not stimulate DNA synthesis at any dose in any HGF target cell line tested , suggesting that the subset of c-Met signaling pathways activated by this HGF isoform are sucient for TIMP-3 induction.
The temporal pro®le of TIMP-3 induction was examined by Northern analysis of RNA samples prepared from MDCK cells that had been treated with serum, HGF or HGF/NK2 for 1 ± 20 h. TIMP-3 showed a dramatic increase in mRNA levels in response to HGF, with high levels of TIMP-3 mRNA being observed from 2 ± 10 h ( Figure 1b). Peak induction of TIMP-3 was observed after 4 ± 6 h of growth factor exposure. Interestingly, a signi®cant decrease in TIMP-3 message was observed after 10 h; at 20 h, HGF-stimulated cells displayed lower levels than untreated controls, suggesting that TIMP-3 production is tightly regulated by HGF. This is consistent with previous reports documenting the transient nature of serum-stimulated TIMP-3 induction in ®broblasts, where a striking correlation was found between TIMP-3 induction and the G1 phase of the cell cycle (Wick et al., 1994). TIMP-3 induction by HGF/NK2 was similar in temporal pro®le to that of HGF, but decreased below control levels after 8 h of HGF treatment (Figure 1b).
The growth factor speci®city of TIMP-3 induction was assessed by comparing the eects of several other growth factors, including epidermal growth factor (EGF), insulin-like growth factor-1 (IGF-1), plateletderived growth factor (PDGF), keratinocyte growth factor (KGF), basic ®broblast growth factor (bFGF), transforming growth factor-b1 (TGF-b) and serum on TIMP-3 expression in MDCK cells. Northern analysis revealed that HGF and TGF-b stimulated TIMP-3 mRNA induction fourfold and tenfold, respectively ( Figure 2a). It is noteworthy that while TGF-b can potently suppress HGF-stimulated DNA synthesis in several epithelial cell types, it does not inhibit HGFstimulated MDCK cell scattering (unpublished observations), and a recent report indicates that TGF-b selectively inhibits HGF-stimulated branching morphogenesis in MDCK without blocking tubulogenesis (Sakurai and Nigam, 1997). Together these results Figure 1 (a) Northern analysis of the induction of TIMP-3 mRNA by HGF and HGF/NK2 in MDCK cells. Total RNA (15 mg/ sample) was isolated from MDCK cells that had been stimulated with the indicated concentrations (ng/ml) of HGF or HGF/NK2 for 6 h. The ®lter was ®rst hybridized with a TIMP-3 cDNA probe (upper panel) and subsequently hybridized to a cDNA encoding glyceraldehyde 3-phosphate dehydrogenase (GAPDH, lower panel) to detect variations in sample loading. The location of the 28S ribosomal RNA on the TIMP-3 blot is shown on the left. (b) Northern analysis of the time course of TIMP-3 mRNA induction by HGF and HGF/NK2 in MDCK cells. Total RNA (15 mg/sample) was prepared from MDCK cells treated with 5% fetal calf serum, HGF (50 ng/ml) or HGF/NK2 (300 ng/ml) for the indicated times. Filters were ®rst hybridized with TIMP-3 cDNA (upper panel) and then GAPDH (lower panel). The position of the 28S and 18S ribosomal RNAs are marked on the left. The ®rst and last lanes (7) of the blot contain total RNA prepared from serum-starved MDCK cells suggest the convergence of HGF and TGF-b intracellular signaling pathways at one or more levels.
TIMP-3 protein level was examined by immunoblot analysis of cell lysates prepared from serum-, TGF-b, HGF/NK2-and HGF-stimulated MDCK cells. As shown in Figure 2b, TGF-b, HGF/NK2 or HGF treatment stimulated signi®cant increases in TIMP-3 protein relative to serum-treated controls. When MDCK cells were stimulated with TGF-b and HGF together, no additive eects on TIMP-3 secretion were observed (data not shown). Interestingly, although TGF-b, HGF/NK2 and HGF showed dierent levels of induction of TIMP-3 mRNA, each stimulated TIMP-3 protein secretion to the same extent, suggesting that TIMP-3 expression may be regulated at both transcriptional and post-transcriptional levels. Together these data demonstrate that HGF and TGF-b induce TIMP-3 gene transcription and stimulate increased TIMP-3 protein secretion in MDCK cells within the same time period.
To determine if other HGF-target cells also displayed the TIMP-3 response, two other epithelial cell lines, Balb/MK keratinocytes and B5/589 mammary epithelial cells  were analysed for TIMP-3 mRNA. As shown in Figure 2c, both cell lines were positive in this regard, suggesting that TIMP-3 induction may be part of a general response to HGF stimulation. HGF did not induce TIMP-3 in NIH3T3 ®broblasts, which do not express c-Met, but serum-stimulated induction of TIMP-3 was observed in these cells, and in B5/589 cells (Figure 2c).

Changes in the balance of TIMP-3 and gelatinase activities
The HGF-stimulated secretion of TIMP-3 is one means by which this growth factor may initiate extracellular matrix remodeling as observed in tissue culture model systems. The composition and rate of turnover of extracellular matrix is determined by a balance between the rates of synthesis and degradation of its components. Degradation is determined by a balance between the opposing activities of proteases and protease inhibitors. HGF-stimulated production of u-PA, MMP-1, and MMP-3 led to their gradual accumulation over 24 ± 48 h (Pepper et al., 1992;Dunsmore et al., 1996), suggesting that during the period over which HGF transiently stimulates TIMP-3 production, the net protease/inhibitor balance may favor decreased matrix degradation. To test this hypothesis, we examined MMP levels over 18 h following HGF stimulation. TIMP-3 shows highest speci®c activity toward MMP-2 and -9 in vitro (Apte et al., 1995), thus we focused on the production of these enzymes by two HGF-responsive cell lines. By SDS ± PAGE/gelatin zymography, the majority of gelatinase activity present in MDCK cells appeared to be the 92 kDa MMP-9; very little 72 kDa MMP-2 was observed ( Figure 3a). In contrast, the human leiomyosarcoma cell line SK-LMS-1 displayed prominent MMP-2 activity, and little detectable MMP-9 activity (Figure 3a). Although a detectable, transient increase in MMP-2 activity was observed in both cell lines at 4 ± 6 h, no substantial changes were observed in the MMP-2 or MMP-9 activities in either cell line over 18 h (Figure 3a).
The MMP-2 and MMP-9 protein content of control or HGF-treated MDCK and SK-LMS-1 cells was examined speci®cally by immunoblotting of gelatin-Sepharose-fractionated conditioned medium ( Figure  3b). Both MMP-2 and -9 bind tightly to this anity matrix (Mazzieri et al., 1997). The 72 kDa MMP-2 was clearly detected in SK-LMS-1 cells, and although a subtle increase was detected at 4 ± 6 h, no signi®cant increase in the amount of MMP-2 was observed over this 18 h period. Little, if any 72 kDa MMP-2 was detected in medium conditioned by MDCK cells, consistent with the results observed by gelatin zymography (Figure 3a and b). Proteins with both Western blot analysis of TIMP-3 protein secretion by MDCK cells stimulated by HGF, HGF/NK2 or TGF-b. Subcon¯uent MDCK cells were treated with serum (7), TGF-b (10 ng/ml), HGF/NK2 (300 ng/ml) or HGF (100 ng/ml) for 6 h. Cell lysates (50 mg/sample) were resolved by 14% SDS ± PAGE, transferred to polyvinylidine di¯uoride (PVDF) membrane, and immunodetected using anti-TIMP-3 antisera and chemiluminescence. (c) Northern analysis of TIMP-3 in B5/589 human mammary epithelial cells, Balb/MK keratinocytes, and NIH3T3 ®broblasts. Serum starved (7) cells were treated for 6 h with 50 ng/ml HGF (`H') or 10% serum (`S'), as indicated above each lane, and RNA samples were prepared as described in Figure 1. Ethidium bromide staining of agarose gels is shown in the lower panels to verify equal sample loading lower and higher molecular masses increased over the period, but these bands lacked gelatinase activity (Figure 3a and b). Although their identities are unknown, they were not observed when the same samples were probed with anti-MMP-9 and visualized using the same secondary antibody and detection system ( Figure 3b, left and centre panels). Immunoblotting with anti-MMP-9 con®rmed the identity of the 92 kDa gelatinase observed in MDCK cell conditioned medium, and showed that no signi®cant increase in the production of this protein was evident after 6 h of HGF treatment (Figure 3b). No MMP-9 was detected in SK-LMS-1 cell conditioned medium (Figure 3b). Immunoblotting of MDCK cell extracts with anti-TIMP-3 under the same conditions clearly demonstrates the transient HGF-stimulated increase in TIMP-3 protein, peaking at 6 h ( Figure 3b). Thus, production of these TIMP-3 sensitive proteases remained constant over the period of HGF-stimulated TIMP-3 induction, suggesting a transient shift in the protease/inhibitor balance toward protease inhibition.

HGF-independent modulation of TIMP-3 expression
We anticipated that the speci®c impact of TIMP-3 induction would be dicult to evaluate at the cellular level in light of the pleiotropic eects of HGF. To create a system in which TIMP-3 levels were modulated in the absence of other HGF-stimulated changes, we sought to identify a cell line with relatively low levels of TIMP-3 expression that would allow assessment of ectopic overexpression of TIMP-3, as well as eective antisense suppression. A comparison of TIMP-3 expression by MDCK, SK-LMS-1, and NIH3T3 cells is shown in Figure 4. MDCK cells express 50-fold more TIMP-3 protein than any of the other cell lines under normal culture conditions. Thus, although several HGF activities have been well characterized using MDCK cells, they were not an optimal system for independently manipulating TIMP-3 expression. SK-LMS-1 cells displayed relatively low basal expression of TIMP-3, while NIH3T3 cells displayed an intermediate level ( Figure 4). SK-LMS-1 cells were stably transfected with full-length sense and antisense TIMP-3 cDNA constructs, and antibioticresistant cultures were analysed for steady-state TIMP-3 protein expression level by immunoblotting. As shown in the right panel of Figure 4, the sense TIMP-3 transfectant displayed substantially increased TIMP-3 expression relative to control cells transfected with vector alone, while TIMP-3 expression was undetectable in the TIMP-3 antisense transfectants. The SK-LMS-1/TIMP-3 transfectants were thus ideally suited for evaluating the impact of TIMP-3 expression in the absence of other HGF-stimulated phenotypic changes.

TIMP-3-mediated changes in cell shape, growth and invasiveness
Striking morphological dierences among the three SK-LMS-1 transfectants were immediately obvious ( Figure 5). While the control vector transfectant resembled untransfected SK-LMS-1 cells ( Figure 5, top panel), TIMP-3 overexpressing cells clustered into  Whole cell lysates were prepared from cells under normal growth conditions, and analysed by immunoblotting with anti-TIMP-3 antibody as described in Figure 2b. Note that in the left-hand panel, 80 mg of protein was loaded for each cell line except MDCK, where only 20 mg was loaded. Samples from stable SK-LMS-1 transfectants (right panel; 40 mg/sample) expressing vector (V), full-length TIMP-3 sense (S), or antisense (AS) constructs, were prepared similarly. The lane marked`C' shows the migration of a puri®ed TIMP-3 standard, the position of which is also indicated at the left close-knit colonies, and generally had a more epithelial than ®broblastic appearance ( Figure 5, centre panel). In contrast, TIMP-3 antisense transfectants grew as a uniform dispersion of randomly oriented single cells, each with elongated bipolar shape or occasionally with three or more thin cytoplasmic projections (Figure 5,bottom panel).
In addition to the gross morphological changes associated with the modulation of TIMP-3 expression, dierences in the rate of growth of the transfectants were readily apparent. A comparison of the growth of sparsely-plated TIMP-3 transfectants over a period of 15 days is shown in Figure 6a. The growth of TIMP-3 overexpressors was obviously retarded relative to the control vector transfectant (Figure 6a, left and centre panels). We also observed accelerated growth of cells transfected with antisense to TIMP-3 (Figure 6a, right panel). The apparently dense staining of TIMP-3 overexpressors was due to dye precipitated onto the surface of these cells, and not to more dense packing of equal numbers of cells. To document these dierences in more detail, the growth of the three SK-LMS-1/ TIMP-3 transfectants was compared over a 20 day period (Figure 6b). The results con®rmed the signi®cantly faster growth of TIMP-3 antisense transfectants relative to TIMP-3 overexpressors (P50.001 after 10 days; Figure 6b), and indicate that  Additional evidence supporting this hypothesis was found by evaluating the anchorage-independent growth of the three SK-LMS-1 transfectants in soft agar. Control vector transfectants exhibited a high level of colony formation in soft agar, consistent with their ability to form tumors in nude mice (Jeers et al., 1996). This level of colony formation was signi®cantly decreased in TIMP-3 overexpressors (Figure 7, top and centre panels). In contrast, antisense suppression of TIMP-3 expression was associated with very aggressive colony formation in this assay, reaching a level eightfold greater than TIMP-3 overexpression ( Figure  7, bottom panel). These results are consistent with those of Bian et al (1996), which show that TIMP-3 can act as a potent growth suppressor and that TIMP-3 can suppress the invasive phenotype that correlates with aggressive growth in soft agar and tumorigenesis in nude mice.
We have examined several tumor cell lines to determine whether a correlation could be found between metastatic phenotype and TIMP-3 expression level. CaSki and HA1780 cells, derived from metastatic cervical and ovarian carcinomas, respectively, showed very low levels of TIMP-3, as did A204, a highly invasive rhabdomyosarcoma-derived cell line ( Figure  8). HT1080 cells, derived from a ®brosarcoma, have been characterized as fourfold more invasive than normal ®broblasts, and threefold less invasive than A204 cells (Albini et al., 1987). Interestingly, HT1080 cells express TIMP-3 at levels intermediate between M426 normal human ®broblasts, and A204 cells (Figure 8). RD, a rhadbomyosarcoma-derived cell line, grow as spindle shaped cells similar to the SK-LMS-1/TIMP-3 antisense transfectants, grow similarly in soft agar, and express similar levels of TIMP-3 protein (Figure 8 and data not shown). SAOS-2 cells are derived from an osteogenic sarcoma, display an epithelial-like morphology, but express much lower levels of TIMP-3 protein than the normal epithelial cells (B5/589 and MDCK) and ®broblasts (M426) included for comparison ( Figure 8). Though preliminary, these data suggest that TIMP-3 expression correlates inversely with the invasive properties of certain human cancers.
The invasive properties of the SK-LMS-1/TIMP-3 transfectants were examined directly using three dimensional matrices of Matrigel or collagen I, both of which have been used extensively in the past to characterize the invasive properties of cultured cells. Cells were seeded on top of Matrigel matrices in growth medium alone or supplemented with HGF, and  prepared from M426 embryonic lung ®broblasts, B5/589 mammary epithelial cells, CaSki epidermoid cervical carcinoma cells, A204 rhabdomyosarcoma cells, RD rhabdomyosarcoma cells, SAOS-2 osteogenic sarcoma cells, HA1780 ovarian carcinoma cells, HT1080 ®brosarcoma cells, and MDCK kidney epithelial cells (10 mg) were resolved by 12% SDS ± PAGE, transferred to PVDF, and immunodetected using anti-TIMP-3 antisera and chemiluminescence. The position of the 24 kDa TIMP-3 protein band is indicated at the right invasion was monitored daily by microscopy. After 3 days, control (vector-transfected) SK-LMS-1 cells displayed modest invasion into the underlying matrix ( Figure 9, upper left panel), while the TIMP-3 antisense transfectants showed much more aggressive matrix invasion (Figure 9, upper right panel). In contrast, TIMP-3 overexpressors formed smoothsurfaced spheroid aggregates, and matrix invasion was completely abolished (Figure 9, upper centre panel). Stimulation with HGF moderately increased Matrigel invasion by control (vector) and antisense TIMP-3 transfectants, while in the TIMP-3 overexpressors an increase in the size of the spherical aggregates was observed, again in the total absence of Matrigel invasion (Figure 9, lower panels). Consistent with these ®ndings were the invasive properties of the SK-LMS-1 transfectants observed using three-dimensional collagen gels. By focusing below the surface monolayer 48 h after initial seeding, control SK-LMS-1 cells were seen to invade the underlying collagen matrix mostly as single elongated cells ( Figure 10). Invasion was almost totally inhibited by addition of the synthetic metalloproteinase inhibitor BB94 (1 mg/ml), suggesting a requirement for metalloproteinase activity (data not shown). Qualitative observations revealed that 24 h after seeding, control vector and antisense TIMP-3 transfectants had already penetrated below the surface of the gel, whereas TIMP-3 overexpressors were still mostly con®ned to the gel surface. Fortyeight hours after seeding, the three cell lines had invaded the underlying gel to dierent extents; TIMP-3 overexpressors were the least invasive (Figure 10b), TIMP-3 antisense transfectants were most invasive (Figure 10c), while the vector-transfected control cells were between these two extremes (Figure 10a). A quantitative analysis of the number of cells which had migrated to depths of 100 mm and 200 mm below the surface of the gel 48 h after seeding demonstrated that TIMP-3 overexpressors were less invasive than control cells, whereas TIMP-3 antisense transfectants were signi®cantly more invasive (P50.001; Figure 10d).

Discussion
The TIMP family of metalloproteinase inhibitors regulate the activity of proteinases such as collagenases, gelatinases and stromelysins, and several lines of evidence suggest that TIMPs are required for extracellular matrix (ECM) homeostasis (reviewed in Stetler-Stevenson et al., 1996;Anand-Apte et al., 1996). TIMP-3 was originally puri®ed from Rous sarcoma virus-infected chicken embryo ®broblasts (Blenis and Hawkes, 1984;Staskus et al., 1991). Isolation of the TIMP-3 cDNA and recombinant expression revealed that the encoded protein promoted detachment of transformed cells from the ECM, accelerated morphological changes characteristic of transformation, and enhanced the growth of normal cells in low serum conditions (Pavlo et al., 1992;Yang and Hawkes, 1992). TIMP-3 was also identi®ed as a serum-inducible delayed early-response gene in human diploid fibroblasts (Wick et al., 1994). At least two speci®c pathological ECM-related defects have been linked to TIMP-3: speci®c inherited mutations in the TIMP-3 gene cause night blindness in Sorsby's fundus dystrophy (Weber et al., 1994), and elevated retinal TIMP-3 mRNA expression has been correlated with simplex retinitis pigmentosa (Jones et al., 1994).
TIMP-3 is expressed in adult kidney, placenta, lung, ovary, uterus, brain, muscle, and bone (Leco et al., 1994;Hurskainen et al., 1996). TIMP-3 is also Figure 9 Eects of TIMP-3 overexpression and antisense suppression on SK-LMS-1 cell invasion into Matrigel matrices. Cells seeded on three-dimensional Matrigel matrices in growth medium alone (upper panels) or growth medium supplemented with HGF (lower panels) were photographed 96 h later by focusing below the surface of the gel using phase contrast optics and a 206 objective as described in Materials and methods. SK-LMS-1 cells transfected with vector alone (V, left panels) exhibited some short processes that extend into the matrix, which are not visible in cells transfected with full-length TIMP-3 sense cDNA (S, center panels), while cells transfected with antisense to TIMP-3 were highly branched and extended multiple processes into the underlying matrix (AS, right panels) expressed developmentally in the latter half of gestation primarily in cartilage, placental trophoblasts, various epithelia, and muscle (Apte et al., 1994). Consistent with its high expression in kidney, we identi®ed TIMP-3 by DD ± PCR as an HGFinduced gene in MDCK cells. HGF-induction of TIMP-3 in keratinocytes and mammary epithelial cells suggests that it may be a general HGF response. The high level of TIMP-3 expression in placental trophoblasts is noteworthy in light of the role of HGF in directing placental development. Two independent HGF gene deletion studies reported severe placental defects, particularly in the number and organization of trophoblast cells (Schmidt et al., 1995;Uehara et al., 1995). The proliferation and dierentiation of trophoblast cells, which are presumably derived from embryonic ectoderm, is normally stimulated by HGF produced by allantoic mesenchyme (Uehara et al., 1995). The role of HGF as a paracrine mediator of epithelial morphogenesis in this system exempli®es what has become a well-accepted paradigm of HGF action: HGF secreted by mesenchymal cells stimulates its receptor expressed in adjacent epithelial cells in various tissues throughout normal development Sonnenberg et al., 1993). Although the role of TIMP-3 in trophoblast dierentiation remains to be determined, HGF is likely to regulate its expression in this setting.
The extensive use of tissue culture model systems has helped uncover the mechanisms by which HGF exerts its pleiotropic eects. HGF stimulates scattering of MDCK cells (Stoker and Perryman, 1985), promotes their invasion through ECM substrates (Giordano et al., 1993;Rong et al., 1994), and induces tubular morphogenesis in collagen or ®brin gels (Montesano et al., 1991). These processes are thought to depend, in part, upon ECM remodeling via controlled proteolytic degradation, associated with an HGF-induced increase in u-PA and its receptor (Montesano et al., 1991;Pepper et al., 1992). Two other members of the metalloproteinase family, stromelysin-1 (MMP-3) and collagenase-1 (MMP-1), are induced by HGF in primary keratinocytes, suggesting a general role for these enzymes in HGF-induced ECM remodeling (Dunsmore et al., 1996). While HGF-stimulated production of u-PA, MMP-1, and MMP-3 led to their gradual accumulation over 24 ± 48 h (Pepper et al., 1992;Dunsmore et al., 1996), we observed that HGF-stimulated TIMP-3 secretion was maximal within 6 ± 8 h, and decreased thereafter. This is consistent with the cell-cycle speci®city of TIMP-3 induction reported by Wick and coworkers (1994), where it was noted that no other known gene exhibited such G1-speci®c induction. These data, together with the absence of signi®cant changes in the activities of MMP-2 and MMP-9 during this period reported here, suggest that ECM turnover is temporarily suspended, then gradually increases, following HGF stimulation. During the ®rst 6 h of HGF treatment, MDCK cells spread (increasing projected surface area at least twofold), intercellular contacts remain intact, and the number and size of F-actin stress ®bres increase, events that are thought to represent increased cell anchorage (Stoker and Perryman, 1985;Dowrick et al., 1991;Ridley et al., 1995). Cells at the colony periphery also show increased lamellipodia and membrane ruing, and all of these early events appear to require Ras and Rac signaling (Ridley et al., 1995). From 6 ± 18 h, intercellular contacts diminish, stress ®bers disappear, and cell motility increase (Stoker and Perryman, 1985;Dowrick et al., 1991). Although the initial cell spreading coincides temporally with rising TIMP-3 secretion, two observations suggest that it may be TIMP-3 independent: (1) cycloheximide inhibited cell spreading only slightly, but potentially inhibited subsequent scattering (Ridley et al., 1995), suggesting that spreading occurs in the absence of increased TIMP-3 protein synthesis; and (2) SK-LMS-1 cells overexpressing TIMP-3 actually show less projected cell surface area than vector controls ( Figure 5). Interestingly, however, the morphology of SK-LMS-1/TIMP-3 antisense transfectants, and preliminary observations (not shown), suggest that TIMP-3 may somehow contribute to the increased cell adhesion that is believed to occur in this initial phase of MDCK cell response.
Several dierences in morphology between SK-LMS-1/TIMP-3 sense and antisense transfectants correlate with HGF-stimulated changes in MDCK cells. The TIMP-3 overexpressing cells form more tightly grouped colonies than control or antisense transfectants, and intercellular contacts are maintained in the initial phase of HGF-stimulated MDCK cell scatter. The later disruption in MDCK intercellular contacts temporally overlaps peak and declining TIMP-3 levels, and coincides with the gradual shift toward ECM degradation that would accompany the long term HGFinduction of several MMPs. SK-LMS-1/TIMP-3 transfectants more closely resemble MDCK cells than controls, particularly the lack of isolated single cells, polygonal cell shape, and strict contact-inhibition of growth. Coincidentally, steady-state TIMP-3 expression by MDCK was found to be much greater than the normal ®broblasts or other epithelial cell types examined. At con¯uence, MDCK cells form a monolayer and do not overgrow (or migrate over) each other, even in the presence of HGF (Dowrick et al., 1991). SK-LMS-1/TIMP-3 transfectants also exhibit more strict monolayer, contact-inhibited growth than control transfectants. This is in striking contrast to SK-LMS-1/TIMP-3 antisense transfectants: cells are spindleshaped, lack tight intercellular contacts, lack contact inhibition, and grow at a faster rate than controls.
Although the mechanism by which TIMP-3 exerts such profound changes on cell morphology and growth is not known, evidence from studies of all TIMPs supports two distinct pathways. Intracellular signals generated in response to ECM turnover, such as those transduced by integrin receptors, may be modulated by TIMP-mediated suppression of MMP activity (Huhtala et al., 1995). For example, MMP activity expressed by melanoma cells seeded on a collagen gel substrate changed the basis of the cell-ECM interaction from ligation of the a 2 b 1 -integrin to an a 1 -mediated interaction (Montgomery et al., 1994). The second mechanism ± the existence of speci®c cell surface TIMP receptors ± has been proposed to explain the growth promoting activities of TIMP-1 and -2 in erythroid cells, ®broblasts, and keratinocytes (Docherty et al., 1985;Golde et al., 1980;Westbrook et al., 1984;Avalos et al., 1988;Stetler-Stevenson et al., 1992;Bertaux et al., 1990). The existence of two pathways may hold an explanation for the apparent disparity between the data presented here ± decreased proliferation by TIMP-3 overexpressors and enhanced proliferation by antisense TIMP-3 transfectants ± and reports of growth promotion by TIMP-1 and -2.
We observed that TIMP-3 antisense suppression enhanced colony formation by SK-LMS-1 in soft agar, as well as their ability to invade three-dimensional Matrigel or collagen-I matrices, while TIMP-3 overexpression strongly inhibited both anchorage-independent growth and matrix invasion. These properties of TIMP-3 are consistent with those reported previously for TIMP-1 and -2. TIMP-1 lowers the metastatic potential of melanoma cells (Khokha, 1994), TIMP-1 suppression confers oncogenicity to normal ®broblasts (Khokha et al., 1989), and TIMP-1 gene deletion results in enhanced invasion of normal dierentiated cells and altered metastasis of transformed cells (Alexander and Werb, 1992;Soloway et al., 1996). Moreover, down-regulation of TIMP-1 and -2 may contribute signi®cantly to the invasive potential of human glioblastoma (Mohanam et al., 1995), and we have observed that several tumor cell lines, particularly those derived from metastatic tumors, express little or no TIMP-3 protein. Many other studies document the positive role of MMPs in promoting tumor invasion and metastasis (reviewed in Stetler-Stevenson et al., 1996). Thus TIMP/MMP imbalance, caused by TIMP suppression or MMP overexpression, can result in increased cellular invasiveness. The modulation of this balance appears to be an important and highly orchestrated cellular response to HGF: an early phase of TIMP-3 induction leading to MMP suppression, followed by a later phase of MMP induction and increased cell invasiveness. HGF does not appear to modulate TIMP-1 expression in the cell lines used in our study (data not shown), and extensive analysis of TIMP-2 expression suggests that it is largely constitutive (Leco et al., 1994). Together with the results reported here, these ®ndings support a pivotal role for TIMP-3 as an early mediator of HGF-regulated TIMP/ MMP balance.
HGF/c-Met signaling plays an essential role in normal development and adult homeostasis at least in part by mediating mesenchymal-to-epithelial conversion and promoting epithelial organization and dierentiation. The inappropriate expression of c-Met in certain mesenchymal cells can lead to a carcinogenic transformation in which the tumor cells express both mesenchymal and epithelial markers (Tsarfaty et al., 1994). We have found that HGF induces the transient expression of TIMP-3, and that this protein can suppress anchorage-independent cell growth. Since HGF also induces the sustained production of MMPs that promote cellular invasion, the observed sequence of TIMP and MMP induction events in the appropriate milieu may be essential for normal HGFdirected development. Inappropriate sustained HGF signaling, such as that observed in a variety of tumors that coexpress c-Met and HGF, would be expected to result in a long term reduction in TIMP-3 expression and sustained MMP accumulation. The resulting TIMP/MMP imbalance may contribute to a highly invasive and possibly metastatic phenotype.

Materials
Tissue culture medium, growth supplements, and RT-H 7 were obtained from GIBCO ± BRL. Plasticware was obtained from Costar and Nunc. Human recombinant HGF and recombinant TGF-b were gifts from Drs George Vande Woude and Anita Roberts, respectively. Puri®ed chicken TIMP-3 protein standard was generously provided by Dr Susan Hawkes. Two polyclonal TIMP-3 antibodies raised against its C-terminal amino acid sequence, and cross-reactive with human, mouse, and chicken TIMP-3, were used; one was obtained from Triple Point Biologics, and the other was a gift from Dr Susan Hawkes. Anity-puri®ed anti-gelatinase-A and -B antibodies were gifts from Dr William Stetler-Stevenson. Puri®ed MMP-2 (72 kDa) was from Boehringer Mannheim. Recombinant human EGF was obtained from Collaborative Research. Recombinant human PDGF, IGF-1 and bFGF were from Upstate Biotechnology, Inc. Recombinant human KGF was prepared as described previously (Ron et al., 1993). Recombinant NK2 was prepared as described previously (Stahl et al., 1997). RNazolB was obtained from Tel-Test. Nytran membrane was obtained from S&S. p-Iodotetrazolium violet and gelatin-agarose were from Sigma. T7 polymerase was obtained from Pharmacia. [ 32 P]dCTP was from Amersham. Reagents for DD ± PCR were obtained from Gen Hunter, Inc. The TA cloning vector was obtained from Invitrogen. Matrigel was purchased from Collaborative Research/Becton Dickinson. Synthetic metalloproteinase inhibitor BB94 was the generous gift of Drs J Gordon and P Brown, British Biotechnology, Cowley, UK.
Tissue culture and cell transfection MDCK cells were grown at 378C in DMEM supplement with 10% fetal bovine serum (FBS). Cells were grown in dishes that had been precoated with human ®bronectin at 1 mg/cm 2 . When the cells reached 60% con¯uency, the medium was replaced with DMEM+5% FBS alone or with the indicated growth factors. The leiomyosarcoma cell line SK-LMS-1 was grown in DMEM+10% FBS. NIH3T3 mouse ®broblasts were grown in DMEM+10% calf serum. Full-length TIMP-3 cDNA in pCEV29 (Lorenzi et al., 1996) in the sense or antisense orientation was transfected into SK-LMS-1 cells by calcium phosphate precipitation and stable transfectants were obtained by selection in media containing G418 (500 mg/ml).

RNA isolation and Northern blot analysis
MDCK cells were washed three times with ice-cold PBS and total RNA was isolated using RNazolB according to the manufacturer's protocol. RNA concentration was estimated by measuring absorbance at 260 nm. Samples of total RNA (15 mg) were run on a 1% agarose/ formaldehyde gel and transferred by capillary blotting onto Nytran membrane. Radioactive DNA probes were prepared using T7 polymerase and [ 32 P]dCTP according to the manufacturer's instructions. Hybridization was performed in 50% formamide (40% for MDCK) followed by two high stringency washes at 588C in 0.16SSC (0.26SSC for MDCK) and 0.1% SDS.

Dierential display
Total RNA (200 ng) isolated from MDCK cells grown in 5% serum alone or supplemented with HGF (50 ng/ml), or NK2 (300 ng/ml) was reverse transcribed with 200 U RT ± H 7 with a one base anchored oligo dT primer. PCR reactions and subsequent product puri®cations were performed using the Gen Hunter Kit according to manufacturer's instructions. Final products were then subcloned using the TA cloning vector and subjected to restriction analysis and dideoxy sequencing.

Immunoblotting
Subcon¯uent cell cultures were treated with HGF, or left untreated, for the time intervals indicated, and cell lysates were prepared by extraction on ice with 50 mM Tris buer (pH 7.4) containing 1% Triton X-100, 0.1% SDS, and protease and phosphatase inhibitors. After clearing by high speed centrifugation, protein amount was determined using the BCA protein assay (Pierce). Equal amounts of protein were prepared in SDS sample buer and samples were resolved by SDS ± PAGE, transferred to PVDF, probed with anti-TIMP-3 and detected by chemiluminescence. A puri®ed TIMP-3 standard was included in addition to conventional molecular mass standards.
Cell growth rate, anchorage-independent growth, and photomicroscopy The eects of TIMP-3 on SK-LMS-1 cell growth were quantitated by seeding multiwell tissue culture plates with SK-LMS-1 cells transfected with vector alone, full-length TIMP-3 cDNA, or full-length antisense TIMP-3 cDNA at a density of 100 cells/well. At various times, cells were ®xed, stained with Giemsa, and multiple wells were scanned using a UMAX PowerLook II high resolution scanner. Quantitative densitometry and descriptive statistical analysis was performed on the stored images using NIH Image software. Mean values of optical density, in arbitrary units, were plotted on the Y-axis against days after seeding on the X-axis. The results of a representative experiment are shown.
Anchorage-independent growth in soft agar was performed and quantitated as described previously (DiFiore et al., 1987). Brie¯y, SK-LMS-1 transfectants were suspended at tenfold serial dilutions (1610 5 to 1610 3 cells) in 0.5% Seaplaque agarose in growth medium in duplicate. Cells were fed once a week with growth medium containing 10% FCS. Parallel sets of SK-LMS-1 transfectants were plated in duplicate into 60 mm dishes to determine the number of colonies formed per plating density. Cell colonies in the soft agar assays were visualized after staining with p-iodotetrazolium violet by bright®eld microscopy using a Zeiss Axiovert 135TV microscope with a 2.56 objective; general cell morphology was visualized by phase contrast microscopy with a 106 objective. Percentage of soft agar growth was determined by counting the number of cells forming colonies of diameter greater than 50 mm per total colony number. The percentage presented is the average of three measurements. Note that in antisense transfectants many cell colonies of diameter greater than 100 mm were visualized. Digital images were recorded using a Hamamatsu XC77 CCD video camera, Hamamatsu Camera Controller Model C2400, IPLab Spectrum software (Signal Analytics, Inc.) and a Power Macintosh 7200 computer.

Zymography
MDCK or SK-LMS-1 cells were washed in serum-free medium and incubated in the presence or absence of HGF for time intervals up to 18 h, as indicated in the ®gure legends. Media were then collected and incubated at 48C for 1 h with 100 ml gelatin-agarose equilibrated in 50 mM Tris-HCl, 150 mM NaCl, 5 mM CaCl 2 , and 0.02% Tween 20, pH 7.6 (Mazzieri et al., 1997). The beads were washed three times with the equilibration buer containing 200 mM NaCl, and proteins were extracted with 70 ml of nonreducing Laemmli buer for SDS ± PAGE. Equal volume of samples were loaded on Novex precast 10% gels for gelatin zymography according to the manufacturer's protocol to analyse gelatinase activity. Equal volumes of samples were analysed under reducing conditions by immunoblotting to determine the amount of each gelatinase, and equal loading of lanes was con®rmed by staining parallel blots with Auro Dye Forte (Amersham). MDCK cell lysates were prepared by extraction on ice with 50 mM Tris buer (pH 7.4) containing 0.5% Triton X-100, 150 mM NaCl, and protease and phosphatase inhibitors. Protein amount was determined by BCA protein assay (Pierce) and equal amounts of samples were analysed by immunoblotting to verify TIMP-3 induction.

In vitro invasion assays
A 50% solution of Matrigel in DMEM (300 ml) was added to each well of 48-well plates and kept at 378C for 30 min to gel. SK-LMS-1/TIMP-3 transfectants, at a density of 25 000 cells/ml of DMEM+5% FCS alone or supplemented with HGF, were seeded at 300 ml per well into the Matrigel-containing plates. Media were replaced and HGF was added every other day at 200 ng/ml. After 4 days cell invasion was photographed by focusing through the gel using phase contrast optics in a Zeiss Axiovert 135TV microscope with a 206 objective.
The invasion of cells into three-dimensional collagen gels was analysed essentially as described (Montesano and Orci, 1985). Brie¯y, eight volumes of a solution of type I collagen extracted from rat tail tendons (approximately 1.5 mg/ml) were quickly mixed with one volume of 106MEM and one volume of sodium bicarbonate (11.76 mg/ml) on ice, and 400 ml aliquots of this mixture were dispensed into 16 mm tissue culture wells and allowed to gel at 378C for 10 min. SK-LMS-1/TIMP-3 transfectants were seeded onto the gels at 2610 4 cells/well in 400 ml of complete medium. After 48 h, the cultures were ®xed in situ with 2.5% glutaraldehyde in 100 mM sodium cacodylate buer (pH 7.4). To quantitate invasion, in each of three separate experiments, ®ve randomly selected ®elds measuring 161.4 mm were photographed at two dierent levels (100 mm and 200 mm) beneath the surface monolayer using the calibrated ®ne focusing micrometer and the 106 phase contrast objective of a Nikon Diaphot TMD inverted photomicroscope. Invasion was quantitated on positive prints by counting the number of cells identi®able in each photographic ®eld. Values are shown as mean number of invading cells per photographic ®eld, and statistical signi®cance was determined using the ANOVA single factor test.