Overexpression and β-1,6-N-Acetylglucosaminylation-initiated Aberrant Glycosylation of TIMP-1

Background: Early TIMP-1 accumulation may impede cancer progression, and cancer cells need to reduce the antiproteolytic burden. Results: Early overexpression and later aberrant glycosylation of TIMP-1 support tumor progression. Conclusion: Concomitant overexpression of TIMP-1 and GnT-V directs accelerated tumor growth and cancer progression in vivo and in vitro. Significance: An answer to the debate on whether TIMP-1 is pro- or anti-oncogenic is given. There has been ongoing debate over whether tissue inhibitor of metalloproteinase-1 (TIMP-1) is pro- or anti-oncogenic. We confirmed that TIMP-1 reinforced cell proliferation in an αvβ3 integrin-dependent manner and conferred resistance against cytotoxicity triggered by TNF-α and IL-2 in WiDr colon cancer cells. The cell-proliferative effects of TIMP-1 contributed to clonogenicity and tumor growth during the onset and early phase of tumor formation in vivo and in vitro. However, mass-produced TIMP-1 impeded further tumor growth by tightly inhibiting the activities of collagenases, which are critical for tumor growth and malignant transformation. Tumor cells could overcome this impasse by overexpression of N-acetylglucosaminyltransferase V, which deteriorates TIMP-1 into an aberrant glycoform. The aberrant glycoform of TIMP-1 was responsible for the mitigated inhibition of collagenases. The outbalanced activities of collagenases can degrade the basement membrane and the interstitial matrix, which act as a physical barrier for tumor growth and progression more efficiently. The concomitant overexpression of TIMP-1 and N-acetylglucosaminyltransferase V enabled WiDr cells to show a higher tumor growth rate as well as more malignant behaviors in a three-dimensional culture system.

regards, the roles of TIMP-1 in tumorigenesis and cancer progression are quite contradictory. Moreover, it is puzzling to find that TIMP-1 is accumulated in tumor tissues (19,20), and the accumulated TIMP-1 is associated with poor prognosis (21).
We here provide evidence that the normally glycosylated form of TIMP-1 promotes tumor cell proliferation and confers resistance against cytotoxicity triggered by cytokines via an MMP-independent mechanism during the early stage of cancer. The antiproteolytic burden created by TIMP-1 accumulation is reduced by aberrant glycosylation initiated by N-acetylglucosaminyltransferase V (GnT-V). In other words, colon cancer cells utilize the growth-potentiating activity of TIMP-1 at the early stage of cancer and employ an aberrant glycosylation mechanism for TIMP-1 to nullify the MMP-dependent, cancer-suppressive effects to support cancer progression. We provide in vivo and in vitro evidence as well as results from a clinical investigation that support this understanding.

Modulation of GnT-V and TIMP-1 Expression in WiDr
Cells-GnT-V-overexpressing cells and mock cells were established following transfection of mgat5/pCXN (neo) and the empty vector, respectively, into the parental WiDr cells, a derivative of the human colonic adenocarcinoma cell line HT-29 (22), using Lipofectamine Plus TM reagent (Invitrogen) according to the manufacturer's instructions. A stable clone that overexpresses GnT-V was selected after dozens of subcultures. TIMP-1 mutant genes were generated using the standard megaprimer methods, where Thr 32 and Ser 80 were mutated to valine and alanine, respectively. Wild type TIMP-1 and the mutant genes were cloned into pcDNA 3.1 hygro(ϩ) plasmid vector (Invitrogen). The cloned vectors were transfected into WiDr:mock or WiDr:GnT-V cells. The stable TIMP-1 transfectants were confirmed by an immunoblot analysis. Cells were maintained as a monolayer in RPMI 1640 medium containing 10% fetal bovine serum at 37°C, supplied with 5% CO 2 .
Western and Lectin Blot Analyses-Protein preparations were resolved on 10 -15% SDS-polyacrylamide gels and transferred electrically onto PVDF membranes (Immobilon-P, Millipore). The membranes were blocked in 0.05% Tween 20-TBS containing 5% skim milk (immunoblot) or 3% BSA (lectin blot) and then incubated with primary antibodies or biotin-labeled L 4 -phytohemagglutinin (L 4 -PHA) or concanavalin A. After hybridizing with HRP-labeled secondary antibody (Cell Signaling) or HRP-avidin conjugates (Vector Laboratories, Inc.), the membranes were reacted with ECL TM Western blotting detection reagents (Amersham Biosciences) and exposed to an x-ray film for 1-2 min. The band intensity was calculated from the digitized, scanned files using ImageJ software (National Institutes of Health, Bethesda, MD).
In Vivo Animal Studies-Six-week-old C.By.Cg-Foxn1 nu (nude) mice were housed and maintained in an animal facility under specific pathogen-free conditions with continuous microbiological monitoring. A sterilized commercial diet (Harlan) and water were given ad libitum. Cells (2 ϫ 10 6 cells in 100 ml of PBS) were injected subcutaneously into the thighs of nude mice. The nude mice (n ϭ 10/group) were inspected daily, and tumor size was measured three times daily in two dimensions with calipers. The tumor volumes were calculated using the formula, length ϫ (width) 2 /2. For histopathological examination, subcutaneous tumor masses were sectioned from the sacrificed mice and fixed in 10% neutral buffered formalin. After fixation, the specimens were routinely processed, embedded in paraffin, and stained with hematoxylin and eosin.
Wound Healing Assay-Cells were grown to near confluence in 6-well plates and scraped using a 20-l pipette tip. Floating cells were removed after washing with PBS buffer and grown for 5 days. The scratched regions were photographed daily at a ϫ200 magnification. Distance filled with grown cells was calculated at different times, and the relative healing velocity was obtained from the regression curves.
BrdU Incorporation Assay-A total of 1 ϫ 10 6 cells were incubated with 50 ng/ml tumor necrosis factor (TNF)-␣ and 0.2 g/ml IL-2 for 2 days and then pulsed with 20 M BrdU (Sigma) for 1 h at 37°C. The cells were then fixed with cold 95% ethanol, washed with PBS, and resuspended in 0.4 mg/ml pepsin in 0.1 N HCl for 30 min at room temperature to release nuclei. The nuclei were pelleted, incubated with 2 N HCl for 30 min, and neutralized with 0.1 M Na 2 B 4 O 7 . Nuclei were washed with PBS containing 0.5% Tween 20 and 0.1% BSA. Nuclei were then incubated with anti-BrdU antibody (BD Biosciences) for 90 min in the dark at room temperature. The nuclei were then stained with FITC-labeled goat anti-mouse antibody (1:50) for 30 min at room temperature in the dark. The nuclei were washed with PBS, incubated with propidium iodide (0.1 mg/ml) and RNase A (10 g/ml) overnight at 4°C, and then analyzed using a FAC-SCalibur flow cytometer (BD Biosciences).
FACS Analysis-Cells were cultured in the presence of 50 ng/ml TNF-␣ and 0.2 g/ml IL-2 for 2 days. The treated cells were double-stained with Annexin-V-FITC and propidium iodide (PI) as described elsewhere (23). Fluorescence was detected using a fluorescence-activated cell sorter to analyze necrotic (PIϩ), non-apoptotic (negative for both dyes), early apoptotic (Annexinϩ/PIϪ), and late apoptotic cells (Annexinϩ/PIϩ). Fluorescence parameters were gated using unstained control cells, and 10,000 cells were counted for each sample.
In Vitro Collagenolysis Inhibition Assays-Cells were cultured in serum-free RPMI1640 medium in the absence of the pH-indicating dye. The conditioned media were retrieved and concentrated using an Amicon ultracentrifugal filter with a molecular mass cut-off of 10 kDa (Millipore). Equal amounts of proteins were mixed with 8 M fluorogenic substrate, DAB-CYL-GABA-PQGL-(EDANS)-AK-NH 2 (Calbiochem), at 37°C in 50 mM Tris-HCl buffer (pH 7.5) containing 150 mM NaCl, 5 mM CaCl 2 , 0.1 mM ZnCl 2 , and 0.02% Brij-35. The hydrolysis activity was kinetically measured in an LS 45 luminescence spectrometer (PerkinElmer Life Sciences) at excitation and emission wavelengths of 338 and 495 nm, respectively.
Immunoprecipitation-Cells were suspended in PBS buffer containing 0.02% Nonidet P-40 and sonicated on ice. The cell lysates were cleared by centrifugation at 10,000 ϫ g for 10 min. The protein extracts were subject to immunoprecipitation using an integrin ␤3 monoclonal antibody (Santa Cruz Biotechnology, Inc., Santa Cruz, CA). The precipitated preparations were used for the subsequent Western blot analysis.
Matrigel-based Three-dimensional Culture-Three-dimensional culture was performed as described elsewhere with minor modifications (24). Briefly, 24-well plates were precoated with 50 l of Matrigel. Approximately 1,500 cells suspended in 150 l of Matrigel were seeded onto the precoated plate. RPMI 1640 medium containing 1% (v/v) FBS was covered and replaced every 2 days. If necessary, TNF-␣ and IL-2 were added at the concentrations aforementioned. The cells were incubated at 37°C for up to 15 days to allow aggregated spheroids to form. The cell aggregates were periodically taken to calculate the formed tumor size. Tumor size was calculated by comparing the lengths of tumors with a mesh of known lengths on photographs. Cell volume was calculated from the formula as used for in vivo experiments. All assays were performed at least three times, with each assay being tested in triplicate.
Statistical Analysis-Statistical differences between groups were determined by Student's t test. Values of p Ͻ 0.05 were taken as significant.

Manipulation of the Expression Level and Glycan Structure of TIMP-1 in Colon Cancer
Cells-GnT-V is a Golgi-located enzyme that is widely known to be overexpressed in various malignant solid tumors (25), producing ␤1,6-N-acetylglucosamine (GlcNAc)-branched N-glycan. WiDr is a colon cancer cell line with a minimal expression of GnT-V (26) into which GnT-V (or MGAT5) gene was forced to be stably overexpressed. L 4 -PHA is a plant-derived lectin that specifically recognizes a ␤1,6-GlcNAc branch attached to the core mannose residue of N-glycan. Overexpression of GnT-V was confirmed by immunofluoresce using L 4 -PHA as a probe. The surfaces of cells with GnT-V overexpression were more reactive to L 4 -PHA binding, compared with the mock transfectant cells (Fig. 1A). The secreted proteins from cells with GnT-V overexpression were also more reactive, as assessed by a lectin blot analysis using L 4 -PHA (data not shown).
TIMP-1 is a glycoprotein that has two N-linked glycosylation sites, at Asn 30 and Asn 78 , and is known to accumulate in various solid tumors, including colon cancer (21). To identify the role of glycans of TIMP-1 and the alterations in cancer progression, the aglycosylation mutant TIMP-1 gene was generated by changing the codon sequences for Thr 32 and Ser 80 to those for Val 32 and Ala 80 , respectively (Fig. 1B). The wild type or mutant TIMP-1 gene was transfected into the mock and GnT-V-overexpressing cells to establish cancer cell lines in which the glycan structure and expression level of TIMP-1 are manipulated. TIMP-1 shows a molecular mass of ϳ28.5 kDa, of which N-glycans at both asparagines account for approximately 8 kDa. The stable transfectant cells were confirmed after more than 30 rounds of subculture by immunoblot and lectin blot analyses (Fig. 1C). The aglycosylated TIMP-1 mutant protein was detected with an estimated molecular mass of 20 kDa in the transfectant cells of both mock and GnT-V-overexpressing cells. L 4 -PHA-reactive aberrant glycoforms of TIMP-1 were observed predominantly in GnT-V-overexpressing cells. The mutant protein showed a 2.5-2.9 fold expression level, compared with the endogenous TIMP-1. Expression level of the wild type TIMP-1 was increased by ϳ3-fold for TIMP-1 transfectant cells. Cathepsin X was used as a loading control because expression of cathepsin X does not fluctuate by GnT-V expression (26).
To clarify the role of the expression level of TIMP-1 in colon cancer, small hairpin RNA was forced to express to interfere with the endogenous TIMP-1 expression. The stable clones that show the most efficient suppression of TIMP-1 expression were selected, and TIMP-1 expression was also tested after more than 30 rounds of subculture (Fig. 1D). The relative expression levels of TIMP-1 and the mutant form are shown in Biphasic Effects of TIMP-1 in Tumor Formation and Proliferation in Vivo-To investigate the effects of TIMP-1 in tumor formation and proliferation in WiDr cells, each transfectant cell line was injected subcutaneously into the thighs of nude mice, and the tumor volumes were measured daily up to 60 days after injection ( Fig. 2A). Tumor masses began to be detectable from 10 days after injection in the case of tumors with the highest growth rate. Mice were sacrificed 60 days after injection, and the formed tumors were resected from mice for further analysis. Analysis of the tumor growth curves revealed three distinct features. 1) Overall, GnT-V transfectant cells showed an earlier solid tumor formation and a higher tumor growth rate, compared with mock cells. 2) Tumor growth showed a biphasic pattern that can be arbitrarily divided into an early phase (days 0 -35) and a later phase (days 35-60), and tumors had a higher growth rate in the late phase than in the early phase. 3) The expression level and the glycosylation status of TIMP-1 affected the tumor proliferation, and they worked differently between the early and late phase (Fig. 2B).
Analysis of the tumor growth rate in the early phase revealed that the growth rate was proportional to the secretion level of glycosylated TIMP-1 (Fig. 2C) with a squared correlation coefficient (r 2 ) of Ն0.90 regardless of GnT-V expression. Although GnT-V expression contributed to the higher tumor proliferation rate, the extent of dependence on TIMP-1 level was nearly identical between the mock and GnT-V-overexpressing cells. It is of note that overexpression of aglycosylated TIMP-1 did not affect the tumor growth rate in vivo, indicating that glycosylation is critical for TIMP-1 to exert effects on tumor formation and growth at this time point.
In the late phase of tumor growth, the overall rates increased compared with the early phase, as mentioned above. The most interesting feature in this phase is that overexpression of aglycosylated TIMP-1 resulted in a lower tumor growth rate, compared with the control cells (Fig. 2D). TIMP-1 acts as a biological modulator through an MMP-independent mechanism or SEPTEMBER 21, 2012 • VOLUME 287 • NUMBER 39

JOURNAL OF BIOLOGICAL CHEMISTRY 32469
an MMP-dependent mechanism (27). Aglycosylated TIMP-1 maintains MMP-inhibitory activity (28), and it is thought that the aglycosylated TIMP-1 blocks the proteolytic activities that are critical for tumor growth in the tumor microenvironment. Notably, the down-regulation of MMP-9 resulted in a slower tumor growth in the late phase than in the early phase (supplemental Fig. 1), indicating the pivotal role of MMP-9 in the late phase of tumor growth in vivo. We previously reported that an aberrant glycoform of TIMP-1 was produced in WiDr cells with GnT-V overexpression, and the aberrant form showed mitigated inhibitory activity for MMP-2 and MMP-9 (26).
Consistent with the previous report, a higher tumor growth rate was observed in cells with GnT-V overexpression. Taken together, these results suggest that tumor formation and growth are dependent on the secretion level of glycosylated TIMP-1, and aberrant glycosylation of TIMP-1 by GnT-V could enable MMPs to outbalance the inhibitory action of TIMP-1, thereby accelerating tumor expansion in the late phase.

TIMP-1 Has Tumor Cell Proliferation Activity-
The cell growth-potentiating activity of TIMP-1 was tested in vitro using the XTT assay, which shows better performance than the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assay with respect to experimental variation and dynamic range (29). The secretion level of glycosylated TIMP-1 was associated with the proliferation capacity of WiDr and the transfectant cells (Fig. 3A). Overexpression of TIMP-1 potentiated proliferation of both parental and GnT-V-overexpressing cells, whereas down-regulation of TIMP-1 decreased proliferative capacity (p Ͻ 0.05). Interestingly, aglycosylated TIMP-1 did not show growth-potentiating activity, which was consistent with the results obtained from the tumor growth curve in vivo (Fig.  2B). These results were observed both in the parental and GnT-V-overexpressing cells. The effect of glycosylated TIMP-1 on the cell proliferation was confirmed by the wound-healing assay. The wound-filling activity was increased by overexpression of glycosylated TIMP-1, whereas aglycosylated TIMP-1 showed no such activity (Fig. 3B). The overall wound-filling rate A, mannosyl ␤1,6-GlcNAc linkages on the cell surfaces were traced by immunofluorescence using biotin-labeled L 4 -PHA as a probe. Nuclei were stained with DAPI. B, aglycosylated mutant TIMP-1 gene was generated by sitedirected mutagenesis, by which gene sequences coding for threonine (T) and serine (S) were changed into those for valine and alanine. C, vector or vector constructs containing aglycosylated human TIMP-1 gene (hT1 aglyco ) or wild type (hT1 wt ) were stably transfected, and the TIMP-1 expression and glycan structure were checked by immunoblot (IB) and lectin blot (LB) analyses. An aglycosylated form of TIMP-1 was observed with an estimated molecular mass of around 20 kDa. An aberrant glycan structure of TIMP-1 comprising the mannosyl ␤1,6-GlcNAc linkages was detected in GnT-V-overexpressing cells. D, down-regulation of TIMP-1 was achieved through interference by small hairpin RNA. The secretion level of TIMP-1 was compared by immunoblot analysis. E, relative TIMP-1 level was quantitatively measured by two different methods. Total TIMP-1 level was measured by the enzyme-linked immunosorbent assay, and glycosylated TIMP-1 was quantified by an immunoblot assay followed by digitized measurement of intensity. Values are means of five independent experiments with S.E. (error bars).
was higher in GnT-V-overexpressing cells, compared with mock cells. However, dependence of wound-filling activity on glycosylated TIMP-1 level was also observed in GnT-V-overexpressing cells.
The cell proliferative activity of TIMP-1 was validated in a specially devised culture dish. The culture flask was separated into three compartments (A, B, and C) by silicon-based barriers, as illustrated in Fig. 3C. The height of the barrier between A and B compartments was lowered for conditioned culture media to freely diffuse between the compartments. The dishsilicon interfaces were tightly sealed, and the C compartment was isolated from the other compartments. To facilitate the diffusion, the dishes were constantly agitated on an orbital shaker. Cells with knockdown expression of TIMP-1 were plated in the B and C compartments, and mock or transfectant cells were cultured in the A compartment. After cells were stabilized, cells in B and C were subjected to scratching, and their wound healing rate was monitored. As the results indicate, the low wound healing activity of cells with down-regulation of TIMP-1 in B was marginally rescued by mock cells placed in A, whereas a significant recovery of the activity in B was observed by the placement of cells with overexpression of glycosylated TIMP-1 in A, compared with the activity of cells in C. However, overexpression of aglycosylated TIMP-1 in A did not result in such a dramatic rescue of wound healing activity (Fig. 3D). Taken together, these results indicate that the proliferation of cancer cells is dependent on the secretion of glycosylated TIMP-1, and the dependence is also seen for the aberrant glycoform of TIMP-1. All of these effects were identically observed for GnT-V transfectant cells.
TIMP-1 Controls Apoptosis Triggered by TNF-␣ and IL-2-TNF-␣ is a pleiotropic inflammatory cytokine and causes necrosis of some types of tumors either alone (30) or in combination with chemotherapeutic agents and other cytokines (31). TNF-␣ and IL-2 are typical cytokines that tumor cells frequently encounter in a tumor microenvironment, where various stromal cells secrete the cytokines during cancer progression (32,33), and combinatorial treatment of these cytokines two-dimensionally measured by calipers, and the tumor volumes were derived from the formula provided under "Experimental Procedures." Ten mice were used for each cell line, and the tumor size was measured three times per day. C, tumor volumes that were measured until the 35th day of injection were regressed, and the slope was taken as the tumor growth rate. The tumor growth rate was again expressed as a function of the glycosylated TIMP-1 level. The squared correlation coefficients (r 2 ) were 0.93 and 0.90 for mock and GnT-V transfectant cells, respectively. D, tumor volumes that were measured from the 35th to 60th day of injection were regressed, and the slope was taken as the tumor growth rate. Of note, overexpression of aglycosylated TIMP-1 negatively affected the tumor growth independent of GnT-V expression (n ϭ 10). * and #, p Ͻ 0.05. Error bars, S.E.

TIMP-1 Overexpression and Aberrant Glycosylation in Cancer
showed an efficacy in several clinical trials (34). However, resistance against the cytotoxic effects of TNF-␣ creates environments for tumor cells to co-evolve and metastasize in various ways (35).
The effects of TIMP-1 expression on the resistance against the cytotoxic effects by TNF-␣ and IL-2 were investigated by monitoring apoptotic and necrotic features of WiDr and the transfectant cells upon treatment with the cytokines. WiDr cells express TNF-␣ receptor 1 (TNF-R1), and transfection of the MGAT5 gene did not alter the receptor expression (Fig.  4A). TIMP-1 expression did not affect the expression levels of the receptors for TNF-␣ and IL-2 (data not shown). Moreover, their receptors were not attacked by GnT-V enzyme. Although treatment with either TNF-␣ or IL-2 did not show any indication of cytotoxicity as assessed by flow cytometry, the concomitant treatment with 50 ng/ml TNF-␣ and 200 ng/ml IL-2 triggered both apoptotic and necrotic responses in WiDr and the transfectant cells. However, mock cells were more vulnerable to the cytotoxicity by the cytokines, compared with WiDr/GnT-V cells (Fig. 4B), indicating the cytokine-induced resistance by GnT-V via a yet to be fully defined mechanism independent of TIMP-1. However, the resistance by GnT-V was augmented by the high TIMP-1 secretion (Fig. 4C). Downregulation of TIMP-1 by RNA interference reversed the apoptotic resistance, confirming that TIMP-1 secretion contributes to avoidance of the cytotoxic effect induced by a combination of cytokines that is encountered by cancer cells in the tumor microenvironment and act as one of the barriers to tumor cells fulfilling metastasis.
TIMP-1 Binding to ␣v␤3 Integrin Potentiates Proliferation and Survival Signaling-The molecular mechanism underlying the anti-apoptotic and proliferative effects of TIMP-1 was  3). B, after cell monolayers were subjected to scratching, the wound-filling activity was observed at ϫ400 magnification. The relative wound-filling activity was calculated by measuring the filled distance divided by hours on the photographs (n ϭ 3). A and B, * and #, p Ͻ 0.05. C, a culture dish was divided into three compartments with silicon, in which diffusion of culture media between A and B compartments was allowed to occur. The interfaces at the bottoms of dishes were tightly sealed to prevent leakages of culture media through them. D, the wound-filling activity of shhT1a/mock cells in B and C compartments was compared. In compartment A, various TIMP-1 transfectant cells were cultured (n ϭ 3). Error bars, S.E. investigated. Integrin ␤3 was immunoprecipitated using an anti-integrin ␤3 antibody, and the binding counterpart integrin ␣v was found to be co-immunoprecipitated (Fig. 5A). It is of note that TIMP-1 was also co-immunoprecipitated in a TIMP-1 level-dependent manner. Interestingly, TIMP-1 from GnT-V-overexpressing cells appeared to readily bind to the integrin dimer, and the aglycosylated TIMP-1 did not show any interaction with ␣v␤3 integrin, which may, at least in part, explain why GnT-V confers higher cell proliferation and survival effects and the aglycosylated TIMP-1 has no cell-proliferative activity, as is seen Figs. 2-4. When treated with TNF-␣ and IL-2, cells showed the caspase-8-dependent apoptotic death, and only the glycosylated TIMP-1 showed a preventive effect on the caspase-8 activation (Fig. 5B). Likewise, the glycosylated TIMP-1 served to maintain ERK activation even under the apoptotic atmosphere.
Down-regulation of TIMP-1 was responsible for the sensitization of ERK inactivation induced by TNF-␣ and IL-2 without affecting to the expression of ␣v␤3 integrin (Fig. 5C). The sensitized ERK inactivation resulted in the down-regulation of cyclin D1 and cyclin A, whose activities are required for cell cycle G 1 /S transition. In line with this observation, down-regulation of TIMP-1 showed preventive effects on G 1 /S transition in GnT-V overexpression cells as accessed by a BrdU incorporation assay (Fig. 5D), which was also observed in mock cells. The involvement of TIMP-1 in regulation of cancer cell survival was confirmed by treatment with a neutralizing antibody of TIMP-1 in the culture media. The concomitant treatment of TNF-␣ and IL-2 resulted in the decreased fraction of cells in S-phase. However, treatment with a neutralizing antibody for TIMP-1 decreased the fraction of cells in S-phase both in the presence and absence of TNF-␣ and IL-2 (Fig. 5E). Overexpression of TIMP-1 rescued the treated cells from the resting states, inducing transition into the synthetic phase. Taken together, these results indicate that the glycosylated form of TIMP-1 binds to ␣v␤3 integrin and augments the survival and proliferation signals by the integrin. The heavily glycosylated TIMP-1 from GnT-V-overexpressing cells was found to more readily bind to the integrin molecules. However, determination of whether TIMP-1 binds directly to the integrin molecules or their interaction is mediated by other molecules, such as galectins, awaits further scrutiny.
Effect of Aberrant Glycosylation of TIMP-1 on the Collagenolytic Activity-Cancer cell proliferation was reported to decrease in MMP-9-deficient mice compared with wild type mice (36), and a similar result was obtained in our observations in vivo (supplemental Fig. 1). Moreover, reduced metastasis and angiogenesis of melanoma have been observed in mice that were genetically modified to lack MMP-2 expression (37), indicating the critical role of collagenolytic activity in tumor progression. A histopathological analysis of a tumor formed 60 days after subcutaneous injection indicates that the tumor cells, bordered with dashed lines in Fig. 6A, exhibited an invasive property whereby they proliferate by infiltrating skeletal muscle fibers (filled arrow) and the surrounding endomysium (open arrow), the major component of which is wisps of collagen. The smaller tumors showed lower invasiveness with less vasculature and lower penetration (data not shown). Taken together, these results are suggestive that the net collagenolytic activities are a critical factor for cancer invasion in the cancer-stromal niche.
One of the factors governing the invasive potential is the balance between collagenolytic potentials and the inhibitory action by TIMPs. MMP-2 and MMP-9 are major matrix metalloproteinases responsible for deregulation of the ECM. Because the proteases and TIMP-1 are secreted by active tumor cells, the conditioned media, at least partly, reflect the balance between the collagenolytic activity and the inhibitory potential and thus the invasive potential. As an indicator of this, the balanced collagenolytic activity of the conditioned media was measured in vitro. Each cell line was cultured in serum-free media in the absence of dye, and the conditioned media were retrieved. Equal amounts of proteins were mixed with a fluorogenic substrate, and the hydrolysis activity was kinetically measured in a fluorospectrophotometer. The slope of formation of fluorescent products was used to determine the relative collagenolytic activity (Fig. 6B). The kinetic results indicate that cells with high GnT-V expression showed higher collagenolytic activity, compared with the mock cells. Normally glycosylated TIMP-1 substantially inhibited the collagenolytic activity, whereas TIMP-1 down-regulation resulted in elevated activity (Fig. 6C). Of note, aglycosylated TIMP-1 retained inhibition for collagenolysis, which is consistent with a previous report (28). FIGURE 5. Reinforcement of ␣v␤3 integrin survival and proliferation signaling by TIMP-1. A, TIMP-1 was co-immunoprecipitated (IP) with ␣v␤3 integrin in a TIMP-1 level-dependent manner. The heavily glycosylated TIMP-1 interacted more readily with the integrin, and the aglycosylated form of TIMP-1 did not bind to the integrin dimer. WB, Western blot. B, cells were treated with TNF-␣ and IL-2 as for Fig. 4, and the activation of caspase-8 and ERK was monitored. C, cells were treated with TNF-␣ and IL-2, and the proliferation signaling that is mediated by ␣v␤3 integrin and the downstream ERK activation was investigated, including cyclin D1 and cyclin A expression. D, incubated with 50 ng/ml TNF-␣ and 200 ng/ml IL-2 for 48 h, cells were then pulsed with BrdU for 1 h prior to analysis by FACS. Due to the aneuploidy of the cells, a gate in the region below 500 on the x axis was chosen, and thus the percentages do not total to 100%. E, cells were incubated in the presence or absence of the neutralizing TIMP-1 antibody at a concentration of 10 g/ml. Effects of the TIMP-1 antibody on G 1 /S transition were tested by a BrdU incorporation assay. Change in cell proliferation was also investigated following treatment with the cytokines. Results were obtained from three independent experiments. * and #, p Ͻ 0.05. Error bars, S.E.
Interestingly, the aberrant TIMP-1 was responsible for the mitigated inhibition of collagenolysis, thereby reinforcing the destructive and invasive potential of cancer cells. Although MMP-2 and MMP-9 were attacked by GnT-V, as assessed by lectin blot analysis using L 4 -PHA, the difference in the glycoforms did not affect the collagenolytic activity of the two enzymes (supplemental Fig. 2). Taken together, these results suggest that the aberrant glycosylation of TIMP-1 plays a critical role in tumor expansion and regulation of the invasive potential of colon cancer cells. Additionally, other factors that make GnT-V-overexpressing cells more aggressive include the observation that MT1-MMP expression is induced by GnT-V overexpression. As a result, MMP-2 was more readily activated by the membrane-bound MMP, and the free form of active MMP-2 was overproduced (supplemental Fig. 3).
Effects of Aberrant Glycosylation of TIMP-1 on Proliferation in Three-dimensional Culture-It is thought that a monolayer culture system does not mirror the behavior of cancer cells in situ, where cell-ECM interactions are critical for cancer cells to evolve. This prompted us to confirm the results obtained from the monolayered cells in the three-dimensional culture system. We also sought to test if the results obtained in vivo can be reproduced in vitro. For this, cells were embedded in Engelbreth-Holm-Swarm (EHS)-derived matrices as described under "Experimental Procedures." Type I collagen is the most abundant collagen in the human body, found in the endomysium and interstitial matrix, and acts as a physical barrier for cell migration and interferes with the proliferation of both normal and cancer cells (38). However, EHS-derived matrices do not contain a sufficient amount of type I collagen (38,39), which we supplemented to EHS-derived matrices at a final concentration of 0.02% (w/v) prior to plating. TNF-␣ and IL-2 were added beginning at 4 days after seeding to recreate the cytotoxic atmosphere occurring in the tumor environments. Embedded into the matrices, each cell was cultured in low serum conditions (1% FBS, v/v) to enhance the effect of TIMP-1 on tumor cell proliferation. The tumor spheres formed were photographed daily.
As seen in Fig. 7A, a high expression of TIMP-1 contributed to elevating clonogenicity and forming spheres with high tumor volumes. Aberrant glycosylation of TIMP-1 by GnT-V showed a synergistic effect on the tumor expansion, which would otherwise be marginal. When cultured in the presence of excess MMP-2 and MMP-9 inhibitors, cells experienced poor clonogenicity, proliferation, and even an apoptotic death for a portion of cells, indicating the critical role of the MMP activity in tumor formation and progression.
Images of tumor spheres that were formed 12 days after seeding clearly illustrate the dependence of clonogenicity and cancer progression on TIMP-1 expression level and aberrant glycosylation of TIMP-1 by GnT-V (Fig. 7B). High expression of TIMP-1 accounted for the higher clone number and voluminous tumor spheres, whereas a portion of cancer cells either showed a lower growth rate or even failed to form a colony in Most interestingly, tumors that concomitantly express TIMP-1 and GnT-V at a high level showed an aggressive morphology with an invasive front (as indicated by the arrows in Fig. 7B), which was difficult to observe in a monolayered cell culture system as well as even in vivo. Cancer cells with marginal expression of GnT-V were symmetrically spherical in shape and appeared to undergo more physical pressure from the ECM as tumors grew. This result is believed to firmly support the claims that the time-dependent interplay of TIMP-1 and GnT-V contribute to tumorigenesis and cancer progression in colon cancer.

DISCUSSION
TIMP-1 is implicated in various biochemical processes by controlling the activities of MMPs and is also known to enhance cell proliferation via an MMP-independent mechanism. These dual activities of TIMP-1 are often contradictory, and the net outcome of TIMP-1 action is complicated and even paradoxical. Especially in cancer, it has not been clearly defined whether TIMP-1 is pro-oncogenic or anti-oncogenic. Many lines of evidence indicate that TIMP-1 promotes cell proliferation and confers resistance against apoptotic cues (40). However, overexpression of TIMP-1, in general, suppresses cancer progression and metastasis in vivo (17,18), which is also contradictory to the reports that accumulation of TIMP-1 is frequently observed in several types of cancer (19,20) and correlated with poor prognosis (21).
Here we suggest a plausible model in which TIMP-1 accumulation and subsequent aberrant glycosylation by GnT-V collaborate to direct clonogenicity and cancer progression (supplemental Fig. 4). During the onset of tumor development, it is an urgent matter for a transformed single cell to secure clonogenicity and survive against the cytotoxic and apoptotic signals, including cytokines and immune attacks. The effect of the base-ment membrane and the interstitial matrix as a physical barrier is negligible at this early stage. It appears that cancer cells utilize the cell-proliferative effect of TIMP-1 to achieve this goal. Overexpression of TIMP-1 positively affected colon cancer cell proliferation (Fig. 3) and resulted in resistance against the cytotoxic effects by concomitant treatments with TNF-␣ and IL-2 by binding to ␣v␤3 integrin and reinforcing the survival and proliferation signal (Figs. 4 and 5). Actually, overexpression of TIMP-1 culminated in a higher rate of tumor formation and growth at the early stage in vivo (Fig. 2). As tumor growth continues, tumor foci encounter physical barriers by the basement membrane and the interstitial matrix, one of the main components of which is collagens, substrates for MMP-2 and MMP-9. The persistent overexpression of TIMP-1, however, in turn acts as a negative regulator for the collagenolytic enzymes and contributes to pause or retardation of subsequent cancer progression. Notably, the tumor growth was nearly halted in the presence of an excessive concentration of inhibitors for MMP-2 and MMP-9 in the three-dimensional culture system (Fig. 7A). To break this stalemate, cancer cells appear to operate "aberrant glycosylation machinery" by which TIMP-1 deteriorates into a less functional glycoform following the addition of ␤1,6-GlcNAc to the core N-glycan. ␤1,6-GlcNAc attached to the core mannose of N-glycan is a preferred substrate for galactosyltransferases and N-acetyltransferases, generating polylactosamine stretch and sialic acid at the terminus (26). This aberrant glycan structure was previously analyzed by a glycan analysis using mass spectrometry, and the aberrant TIMP-1 was found to significantly lose its binding affinity for active MMP-2 and MMP-9 (26). An enhanced collagenolytic activity was observed in WiDr cells with GnT-V overexpression (Fig. 6), and the outbalanced collagenolytic activity was responsible for the subsequent tumor growth in vivo (Fig. 2) and the malignant phenotype in vitro (Fig. 7). Although aglycosylated TIMP-1 is not a form that is found under normal states in nature, it served to support the idea that aberrant glycosylation of TIMP-1 is necessary for accelerated tumor growth. Aglycolsylated TIMP-1 is not attacked by overexpressed GnT-V because of a lack of N-glycan, and the aglycoform retains the collagenaseinhibitory activity (28) (Fig. 6C). Accordingly, cells overexpressing aglycosylated TIMP-1 and GnT-V showed retarded tumor growth in the later phase (Fig. 2).
Many tumors are of epithelium origin and characterized by tight cell-cell adhesion and polarity (41,42). For tumor cells to obtain invasive and metastatic properties, it is suggested that the epithelial cells need to transform into mesenchymal-like cells, a process called epithelial-mesenchymal transition (EMT). The canonical pathway for EMT is triggered by TGF-␤ (43), and other pathways have also been reported (44 -47). Evidence from a number of research groups indicates that several MMPs are involved in EMT. MMP-3 induces EMT in mouse mammary epithelial cells through increased expression of Rac1b and induction of reactive oxygen species (48). MMP-28 (49) and MMP-7 (50) were reported to induce EMT in lung cancer. Activation of MMP-2 and MT-MMP has also been shown to be essential in the EMT of the endocardial cushion cells (51). Tumor cells with an EMT-like phenotype were, although rare, observed in the three-dimensional culture only when GnT-V and TIMP-1 were concomitantly overexpressed (data not shown). Actually, an EMT-like phenotype was observed in GnT-V transgenic mice (52). Because TIMP-1 is a pan-MMP inhibitor (9), overexpression and aberrant glycosylation of TIMP-1 may lead to perturbations in balances with various MMPs that are possibly involved in EMT. Therefore, there is strong motivation to explore which MMP is involved in EMT in colon cancer, whether fluctuations of TIMP-1 affect the MMP-involved EMT, and how collapse in TIMP-1/MMP balance is implicated in EMT.
Cancer cells that form cancer foci dynamically interact with the surrounding stroma that contains fibroblasts, endothelial cells, inflammatory cells, and ECM; participate in the formation of the tumor microenvironment; and evolve into more malignant cells (53). The tumor microenvironment surrounding cancer and stromal cells is best realized in an in vivo model system despite the immunocompromised characteristics in several model mice. Therefore, it is thought that in vivo models provide the native three-dimensional microenvironment and best reflect the behaviors and responses of cancer cells in the tumor environment in situ. However, observations of time course changes in terms of cell morphology are limited, and the final results are attainable only when animals are sacrificed. To overcome this limitation, a three-dimensional in vitro culture system has been employed. Because it is a simplified system and dynamic cellular and chemical interactions are limited, it does not mirror bona fide dynamic behaviors of tumor cells in situ. Here, we attempted to mimic the complex nature of the human tumor microenvironment in a three-dimensional culture system.
We used a laminin-rich matrix derived from EHS as a supporting material. It contains a limited amount of type I collagen, the most abundant collagen in the human body, and acts as a physical barrier for cell migration and interferes with the proliferation of both normal and cancer cells (38). We used an EHS-based matrix supplemented with Type I collagen. Although non-epithelial cells, including immune cells, endothelial cells, and fibroblasts, were not co-cultured, we attempted to establish apoptotic environments by treating TNF-␣ and IL-2, which occur during interaction of cancer cells with those non-epithelial cells. The concentration of FBS was lowered to 1% (v/v) during three-dimensional culture, with the aim of reflecting the conditions of limited nutrient and molecular factors in the tumor microenvironment. These manipulations were found to suffice to reproduce the in vivo results in this three-dimensional culture system. Deletion of any of these manipulations did not produce in vitro results similar to those obtained from the in vivo xenograft experiments. More comprehensive information on the molecular dynamics, the chemical nature of matrix, and the co-culturing system would help to enhance the three-dimensional culture system such that it could serve as an effective supplemental system for in vivo models and provide information that would otherwise be unobtainable.