TIMP-3 Binds to Sulfated Glycosaminoglycans of the Extracellular Matrix*

Of the four known tissue inhibitors of metalloproteinases (TIMPs), TIMP-3 is distinguished by its tighter binding to the extracellular matrix. The present results show that glycosaminoglycans such as heparin, heparan sulfate, chondroitin sulfates A, B, and C, and sulfated compounds such as suramin and pentosan efficiently extract TIMP-3 from the postpartum rat uterus. Enzymatic treatment by heparinase III or chondroitinase ABC also releases TIMP-3, but neither one alone gives complete release. Confocal microscopy shows colocalization of heparan sulfate and TIMP-3 in the endometrium subjacent to the lumen of the uterus. Immunostaining of TIMP-3 is lost upon digestion of tissue sections with heparinase III and chondroitinase ABC. The N-terminal domain of human TIMP-3 was expressed and found to bind to heparin with affinity similar to that of full-length mouse TIMP-3. The A and B β-strands of the N-terminal domain of TIMP-3 contain two potential heparin-binding sequences rich in lysine and arginine; these strands should form a double track on the outer surface of TIMP-3. Synthetic peptides corresponding to segments of these two strands compete for heparin in the DNase II binding assay. TIMP-3 binding may be important for the cellular regulation of activity of the matrix metalloproteinases.

there are at least 20 MMPs, all able to digest various ECM components (3,4).
The MMPs, in turn, are regulated by tissue inhibitors of metalloproteinases or TIMPs. The major function of the TIMPs is to inhibit MMPs; any imbalance in which the activities of MMPs outweigh the TIMP levels will favor tissue destruction and pathological processes (5,6). The TIMPs also possess growth stimulatory and regulatory activities (7,8). The four members of the TIMP family all have similar secondary structures of six loops stabilized by six highly conserved disulfide bonds. The TIMPs all bind tightly, albeit with widely varying affinity, to the various MMPs. The x-ray structure (9) shows that the N-terminal cysteine chelates the active site zinc. TIMPs have N-and C-terminal domains, each with three loops. The N-terminal domain of TIMP-1 folds readily and displays full inhibitory activity (10).
TIMP-3 has several features that distinguish it from the other TIMPs. First, it is the only TIMP to bind tightly to the ECM: it was first observed as a transformation-sensitive protein bound to the ECM of chick embryo fibroblasts (11) and extractable with SDS or guanidine. This protein was subsequently shown to be TIMP-3 (12). Second, TIMP-3 is the only TIMP to inhibit members of the ADAM (a disintegrin and metalloprotease domain) family such as tumor necrosis factor-␣-converting enzyme (13); this may account for its ability to induce apoptosis (14). It is the only TIMP to inhibit shedding of L-selectin (15) and interleukin-6 receptors (16). Third, TIMP-3 is the only TIMP directly implicated in a disease process: Ser-Cys mutants of TIMP-3 accumulate in Bruch's membrane of the eye and cause Sorsby's fundus dystrophy (17). TIMP-3 also promotes the detachment of transformed cells from the ECM (18) and is involved in the formation, branching, and expansion of epithelial tubes and in regulating trophoblast invasion of the uterus (19).
The present study is concerned with the mechanism of binding of TIMP-3 to the extracellular matrix. We recently reported that matrilysin could bind to heparan sulfate in rat uterine tissues (2). In this work, it was noted that heparin not only extracted matrilysin from the tissue, but also solubilized TIMP-3. The present results indicate that heparan sulfate and other sulfated glycosaminoglycans may be responsible for the binding of TIMP-3 to the ECM.

EXPERIMENTAL PROCEDURES
Extraction of TIMP-3-Uteri were collected 1 day postpartum from Harlan Sprague-Dawley rats (Harlan), weighed (ϳ2 g), washed 3ϫ with cold 50 mM Tris, pH 7.5, 0.03% sodium azide, and homogenized in 20 ml of this buffer containing 0.1% Triton X-100 with a Polytron for 6 min at 4°C. The mixture was centrifuged at 11,000 rpm for 20 min. The pellet was washed 2ϫ and resuspended in the same volume of cold 50 mM Tris, pH 7.5, 0.03% sodium azide, 50 M Z-Phe-chloromethyl ketone, 50 M aminoethyl-benzenesulfonyl fluoride. The suspension was divided at 0.5 ml per tube. Extractants (50 l, Sigma) were added to a final * This research was supported by United States Public Health Services Grants AR-40994 (to K. B.) and AR-16940 (to J. F. W.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18  concentration of 0.2 mg/ml of the various GAGs and 2 mg/ml of pentosan and suramin. Extraction at 4°C for 30 min was followed by centrifugation at 14,000 rpm for 10 min. For the heat extraction procedure, see Ref. 20. To destroy TIMPs, control extracts were reduced with 5 mM dithiothreitol for 25 min, 24°C, and thiol groups were blocked with 10 mM iodoacetic acid. For the heparinase III (Flavobacterium heparinum, heparitinase I, Sigma) and chondroitinase ABC (Sigma) digestion, pellets were resuspended in 50 mM Tris, pH 7.5, 0.03% sodium azide, 5 mM CaCl 2 , 10 M ZnCl 2 , 50 M Z-Phe-chloromethyl ketone, and 50 M aminoethyl-benzenesulfonyl fluoride and incubated with 0.2 unit of enzyme/ml at 37°C for 18 h. Controls were incubated without added enzyme to check for endogenous activity.
Reverse Zymography-Components for 12.5% SDS-polyacrylamide gel were mixed with gelatin (final concentration 1 mg/ml) and a proprietary mixture of gelatinases (University Technology International, Calgary) and cast as gels. Extracts containing TIMPs were electrophoresed, and the gels were then washed 3ϫ with 2.5% Triton X-100/50 mM Tris/5 mM CaCl 2 /0.03% azide and 3ϫ with 50 mM Tris, pH 7.5, 5 mM CaCl 2 , 50 M Z-Phe-chloromethyl ketone, and 10 mM phenylmethylsulfonyl fluoride. The gels were incubated in this latter mixture at 37°C for 18 h and stained with Coomassie Blue. The blue gelatin staining was cleared by gelatinase action except where TIMP bands blocked this activity. Marker TIMPs (mouse) were also obtained from the University Technology International, Calgary.
Confocal Microscopy-Frozen tissue sections (5 m) from 26-to 32-h postpartum rat uterus were air-dried and soaked in 95% ethanol for 10 min. Sections were prefixed with 100% methanol for 20 min and rinsed 3ϫ with PBS. For the pretreated group, some sections were washed with 100 l of heparin (20 mg/ml) for 1 h at 24°C or digested for 18 h in a moist chamber at 24°C with heparinase III, or chondroitinase ABC, or a combination of the two. These enzymes (0.2 unit/ml) were used in 50 mM acetate buffer, pH 6.5, containing 0.1 M ZnCl 2 , 5 mM CaCl 2 , 5 mM MgCl 2 , 10 mM phenylmethylsulfonyl fluoride, 0.05% sodium azide, and the following inhibitors at 0.1 mM: Z-Phe-chloromethyl ketone, tosyl-Phe-chloromethyl ketone, tosyl-Leu-chloromethyl ketone, antipain, and BB-94 (British BioTech Pharmaceuticals Ltd.). Control sections were treated with the same mixture without enzyme. The sections were rinsed twice with PBS.
Sections that were not pretreated were postfixed for 5-10 min with 3.7% paraformaldehyde, rinsed with PBS, and blocked with 10% heatinactivated normal rabbit serum at 24°C in PBS for 40 min. They were then exposed to goat polyclonal antibody against human TIMP-3 (Santa Cruz Biotechnology, Santa Cruz, CA) and mouse monoclonal antibody (IgM) against heparan sulfate (Sagaku) for 1 h. The TIMP-3 antibody was raised against the C-terminal 20 amino acids and recognized both rat and human TIMP-3. Sections were washed 3ϫ in PBS. Primary antibodies were reacted with a mixture in blocking reagent of rabbit anti-goat IgG polyclonal antibody conjugated to fluorescein isothiocyanate and rabbit anti-mouse IgM polyclonal antibody conjugated to Texas Red (both from Jackson ImmunoResearch Laboratories). Sections were washed 3ϫ in PBS for a total of 45 min. Controls included sections in which the first antibody was omitted, sections treated only with goat anti-human TIMP-3 followed by rabbit anti-mouse IgM antibody, and sections treated with mouse anti-heparan sulfate followed by rabbit anti-rabbit IgG antibody (to show absence of cross-species reaction).
Washed sections were covered with SlowFade mounting solution (Molecular Probes, Eugene, OR), and a coverslip was applied. Sections were analyzed with a Zeiss confocal laser scanning microscope (LSM 510) equipped with a 25-milliwatt krypton-argon laser and a 10-milliwatt helium-neon laser. Excitation was at 488 nm and emission at 530 and 590 nm for fluorescein isothiocyanate and Texas Red, respectively. Images were captured at 0.5-m increments along the Z axis and converted to composite images by LSM 510 software.
Recombinant TIMP-3-BHK cells transfected with mouse TIMP-3 (mT3PNUT.BHK, University Technology International, Calgary) were grown at 37°C in Dubecco's modified Eagle's medium/nutrient mixture F12 (Life Technologies, Inc.) containing 5% fetal calf serum and 1% penicillin and streptomycin. Conditioned medium (50 ml) was collected from 75-cm 2 cell cultures confluent for 3-4 days. Medium (1 ml) from BHK-TIMP-3 cell cultures was mixed with 4 ml of buffer suspension of 0.5 mg of heparin-agarose beads (Sigma, H-0402) and stirred at room temperature for 4 h. The mixture was poured in a small column then washed with 50 mM Tris, pH 7.5, plus 0.15 M NaCl. Stepwise elution was then carried out using increasing amounts of NaCl or heparin in Tris buffer Expression of N-TIMP-3 in E. coli-Human TIMP-3 cDNA from a placental cDNA library was kindly provided by Dr. H. Nagase, Univer-sity of Kansas Medical Center. A set of primers, 5Ј-AGTCATATGTG-CACATGCTCG3-Ј (forward) and 5Ј-GCGGCCGCGTTACAACCCAG-GTG-3Ј (reverse), was used in a one-step polymerase chain reaction to amplify the cDNA insert encoding N-terminal TIMP-3 (residues Cys 1 -Asn 121 ). The amplified insert was digested by NdeI and NotI and ligated into pET21b vectors (Invitrogen Inc.). The ligation reaction mixture was used to transform the Escherichia coli DH5␣ competent cells. Plasmid DNA from positive clones was purified using the Qiagen kit and digested by NotI and NdeI at 37°C for 3 h. The correct clone was confirmed by DNA sequencing and used to transform into the expression host E. coli BL21(DE3). Cells containing pET3a-N-TIMP-3 were grown in 3 ml of LB/ampicillin medium at 37°C overnight then inoculated into 1-to 6-liter batches. When A 600 reached 0.6 -0.8, the culture was induced with isopropyl-␤-D-thiogalactopyranoside (0.4 mM) for protein expression. Cells were grown for another 3 h and harvested by centrifugation.
Purification of N-TIMP-3-N-TIMP-3 was expressed as a fusion protein with a C-terminal His tail. Inclusion bodies were dissolved in 50 ml of loading buffer (5 mM imidazole, 0.5 M NaCl, 20 mM Tris-HCl, pH 7.9, 6 M guanidine) and centrifuged at 10,000 rpm for 40 min. The supernatant was loaded onto a Ni-NTA column (7 ϫ 80 mm, Qiagen) equilibrated with loading buffer. The column was washed at room temperature with 60 mM imidazole, 0.5 M NaCl, 8 M urea, 20 mM Tris-HCl, pH 7.9, then eluted with 500 mM imidazole, 0.25 M NaCl, 8 M urea, 10 mM Tris-HCl, pH 7.9. Fractions from the Ni-NTA column were analyzed by SDS-polyacrylamide gel electrophoresis.
DNase II-based Homogeneous Heparin Binding Assay-In this assay, binding of heparin to DNase II was assessed (21). Competitive test compounds were added at 4°C for 20 min in pH 4.8 acetate buffer ϩ 5 mM dithiothreitol; then substrate was added for digestion at 37°C. Heparin concentration was adjusted to inhibit DNase II activity by 90%; compounds binding heparin reversed this inhibition. The percentage inhibition observed in the presence of heparin and added test species was plotted versus log concentration to yield the dose-response curve for the given species. Several peptides from the A and B strands of TIMP-3 and RHAMM 401-411 (a heparin-binding peptide from the Receptor for Hyaluronic Acid-Mediated Mobility) were synthesized (Genemed).

Sulfated Glycosaminoglycans Extract TIMP-3 from Postpartum Rat Uterus
Tissue-Various sulfated compounds were tested as extractants; Fig. 1 illustrates that heparin/heparan sulfates and chondroitin sulfates were effective extractants at 0.2 mg/ml, but keratan sulfate showed only a weak effect. Sulfated compounds such as pentosan polysulfate and suramin also liberated TIMP-3, but higher concentrations (2 mg/ml) were required (Fig. 1). Heat extraction, effective for matrilysin (2), was much less so for TIMP-3. Two bands of TIMP-3 (27 and 22 kDa, corresponding to glycosylated and nonglycosylated forms (22), appeared in the reverse zymogram. Most of the TIMP-1 and TIMP-2 appeared in the initial Triton extract (not shown). A band corresponding in position to TIMP-2 was also extracted by GAGs, suramin, and pentosan (Fig. 1); this may be TIMP-2 that was bound to gelatinase A, but positive identification of the inhibitor has not been made. Heparin and suramin also extracted a small inhibitory protein of 16 kDa that was not sensitive to reduction/alkylation and has not been identified. Optimal extraction of TIMP-3 was achieved at 2-4 mg of heparin/ml or 1-2 mg of suramin/ml (not shown). A parallel gel was run on SDS-polyacrylamide gel electrophoresis and stained with Coomassie Blue (right portion of Fig. 1) to show that the dark bands interpreted as TIMPs in the reverse zymograms were not due to nonspecific protein bands. The intensity of the glycosylated TIMP-3 (T3*) band appears to be less in the heparan sulfate/heparin lanes than in the CS lanes. This is attributed to interaction with a comigrating band of matrilysin extracted by heparin but not by chondroitin sulfate; this protease band can be directly visualized in the heat extract lane.
TIMP-3 was identified by its position on the reverse zymogram and by its sensitivity to destruction by reduction and alkylation (Fig. 2). A polyclonal antibody to human TIMP-3 (Santa Cruz) was used to show that heparin removed the factor from the uterus (see Fig. 4 below), but the antibody proved unsuitable for Western blotting of rat TIMP-3. Therefore, the identification of TIMP-3 depended on size, sensitivity to reduction and alkylation, and binding to matrix as shown by loss from tissue sections following treatment that increases TIMP-3 in reverse zymograms. TIMP-3 is the only TIMP that binds to the extracellular matrix (5). Fig. 3 illustrates further extraction studies; reverse zymography is not quantitative and permits only rough qualitative comparisons. It was also affected by MMPs that migrated to the same region and reacted with the TIMPs. The initial tissue extract in Triton X-100 (normally discarded) contained some TIMP-3. This might reflect binding of TIMP-3 to proteoglycans of the cell membranes, which were disrupted by Triton, or TIMP-3 in complex with gelatinase A and B (23). SDS should have released the bulk of the inhibitor (based on its ability to release the more tightly binding MMP-7 (2)), although some might remain bound to MMP-2 through its hemopexin domain (23). The highly positively charged heparin antagonist, poly-Dlysine, extracted TIMP-3; it is suggested that it competed with TIMP-3 for binding sites on the GAG chains. Digestion with chondroitinase ABC released less TIMP-3 than did heparinase III digestion.
Confocal Microscopy-The postpartum rat uterus contained both heparan sulfate (Fig. 4A) and TIMP-3 (Fig. 4B), which were largely localized near the uterine lumen, in the epithelial cells, and in their underlying basement membrane. There was little of either molecule in the deep stroma (right side of Fig.  4C). Superimposition of images indicates colocalization of the two proteins with some small patches of green remaining, perhaps on chondroitin sulfate. Washing with heparin (Fig. 4E) completely eliminated the TIMP-3 staining. Digestion with chondroitinase ABC gave some reduction in TIMP staining (Fig. 4G); but a better estimate of the chondroitinase-sensitive binding is probably provided by the residual staining in Fig. 4I. Digestion with heparinase III gave extensive losses of both heparan sulfate and TIMP-3 (Fig. 4, H and I). Both components were completely removed by digestion with the two enzymes together (Fig. 4, J and K). Incubation of sections without added enzyme did not lead to significant losses of either component (not shown).
Properties of Full-length and C-terminally Truncated TIMP-3-In reverse zymography (Fig. 5A), mouse TIMP-3 activity could be detected in BHK-TIMP-3 cells but not in the mocktransfected cell line. Two forms of TIMP-3 were found in the medium: a 27-kDa form, corresponding to the glycosylated form (24) and a 22-kDa nonglycosylated form. To see TIMP-3 in the medium, it was necessary to culture for several days, presumably until binding sites in the ECM are first filled. The denatured truncated human N-TIMP-3 (ϳ14 kDa) was prepared from E. coli and folded ("Experimental Procedures"). Similar amounts of protein were found before and after folding (Fig.  5B) but only the folded form was inhibitory (Fig. 5C).
Culture medium containing recombinant mouse TIMP-3 was mixed with heparin-agarose beads and packed in a small column (see "Experimental Procedures"). The initial concentration of NaCl was 0.15 M and the wash was with 0.2 M. Stepwise increases in salt concentration eluted TIMP-3 between 0.3 and 0.8 M NaCl, with the peak at about 0.5 M (Fig. 6A). Both the glycosylated and unglycosylated forms emerged at the same position. Further washing with 2% SDS removed a small amount of residual TIMP (about 5%). The purified and folded human N-TIMP-3 bound in similar fashion (Fig. 6B); both monomeric and dimeric forms were eluted with a peak also around 0.5 M NaCl but with somewhat longer tailing. This tailing, with a distinct band at 0.9 M NaCl was attributed to the propensity of the truncated TIMP to aggregate. Such aggregates may be eluted at higher salt, but the aggregates did not appear on the gel because of dissociation by SDS in the gel. The elution of full-length and truncated TIMP-3 around 0.5 M NaCl indicated that the major heparin binding site was located in the N-terminal part of the protein.
Three peptides were synthesized based on the A and B strand sequences of rat TIMP-3 (25): Pep1 ϭ residues 19 -32 (IRAKVVGKKLVKEG); Pep2 ϭ residues 41-52 (IKQMKMY-RGFSKM); and the spanning peptide Pep3 ϭ residues 19 -52 (IRAKVVGKKLVKEGPFGTLVYTIKQMKMYRFHSKM). This last peptide contained 9 basic residues potentially involved in heparin binding. It can be seen from Fig. 7 that the two shorter peptides have similar affinity for heparin and that this is about 10-fold less than the binding affinity of the long peptide (IC 50 ϭ 3.5 M). All three bound more firmly than the RHAMM 401-410 peptide (KQKIKHVVKLK). N-TIMP-3 was unstable under the reducing conditions of this assay and could not be measured. DISCUSSION There has been no systematic study of the extraction of TIMP-3 from the ECM. It was earlier noted that guanidine and SDS are effective extractants (26). Here we show that nega-tively charged molecules (heparin, various GAGs, and polysulfated compounds) gave extensive extraction comparable to SDS, but positively charged compounds such as polylysine were also extractants. However, enzymatic treatment with heparinase III and chondroitinase ABC gave extensive extraction, pointing to the negatively charged GAG molecules as binding sites, so positively charged compounds such as polylysine probably competed with TIMP-3 for binding to heparin. A preponderance of basic over acidic residues (26:13) in TIMP-3 also supports this interpretation. It is interesting to note that cultured mouse mesangial cells produce TIMP-3 in the medium only after pentosan polysulfate or heparin treatment (27); we suggest this might be due to the release of TIMP-3 bound to the ECM of the cultures.

FIG. 5. Expression of recombinant mouse TIMP-3 in mammalian BHK cells and C-terminally truncated human N-TIMP-3 in
The N-terminal domain of TIMP-3 contains 17 positive and 8 negative charges (25) and exhibits the OB (oligosaccharide/ oligonucleotide binding) fold (9). Because GAGs are similar in charge and linearity to oligosaccharides/nucleotide polymers, the N-terminal domain of TIMP-3 could be a binding site for GAGs. The binding of TIMP-3 is strong, but not nearly as strong as the binding of matrilysin: TIMP-3 is eluted from a heparin affinity column with 0.5 M NaCl, whereas matrilysin is not eluted at 2 M NaCl (2). Matrilysin also has a great many more positive residues that might participate in binding. The fact that chondroitinase ABC appeared to release TIMP-3 from tissue but not as efficiently as heparinase III (Figs. 3 and 4) suggests that the binding is not highly specific. Several types of sulfated GAGs may serve as binding sites, but heparan sulfate chains are probably the major sites. Full-length TIMP-3 eluted from a heparin affinity column at 0.5 M NaCl. This matches exactly the results of Butler et al. (23); they also showed that TIMP-1 and TIMP-2 did not bind to heparin at 0.15 M NaCl, supporting our hypothesis that TIMP-3 is distinctive in binding to heparin. They further showed that glycosylation at the site Asn 184 near the C terminus has no effect on binding and elution, comparable to the results in Fig. 6. We found that the N-domain of TIMP-3 bound to heparin with affinity similar to that of full-length TIMP-3, indicating that most of the free energy of binding is contributed by the N terminus. It must be mentioned that our results are not in agreement with those of Langton et al. (22) who expressed the N-domain of human TIMP-3 in COS-7 cells. In that case, full-length TIMP-3 bound to the ECM produced by the COS cells, whereas the N-domain did not and adding the N-domain of TIMP-2 to the C-domain of TIMP-3 gave partial but not full binding. We cannot explain this contradictory finding except to suggest that COS cells may make a different ECM than other cells.
The three-dimensional structure of TIMP-3 is not known but is assumed to be similar to that of TIMP-1 and TIMP-2; for both cases we know the complete x-ray crystallographic structure (9,28) and the NMR structure of the N-domain (29,30). In these two structures there are two ␤-strands, A and B, within the first disulfide-bonded loop. These strands lie across the protein surface and remain exposed when TIMP binds to MMPs. Fig. 8 shows models of the A and B strands (plus 5 further residues) of human TIMP-3 based on human TIMP-1 and -2. We have focused on this region, because its sequence is quite distinct from that of the A and B strands of the more readily soluble TIMP-1 and TIMP-2. This region contains 9 basic residues and only 1 acidic residue (Glu 30 ), which is directed downward and interacts with Lys 89 (not shown). TIMP-1 has only 6 basic residues and 3 acidic, and TIMP-1, although it also has 9 basic residues, has 7 acidic residues. So, only TIMP-3 has a large excess of positive charge in this region, which might explain its unique matrix-binding property. In both models (Fig. 8) 6 basic residues are directed upward from the surface and 3 are lying on the surface where they are H-bonded to nearby residues. However, these 3 might also contact a heparin chain if their H-bonds were disrupted. There is a tendency for the basic residues to form a double track, which might interact with sulfate groups attached to alternate sides of the heparin backbone. Either model would predict that a single chain of heparin/ heparan sulfate would be stretched along the TIMP surface and contacting, at a minimum, Lys 30 , Lys 26 , Lys 22 , Lys 42 , Arg 20 , and Lys 45 , with additional basic residues possibly joining through an induced fit effect.
This study has included human, rat, and mouse TIMP-3. It is reasonable to expect them all to bind to heparin in an almost identical fashion. In the entire N-domain, there are only 3 amino acid differences between human and rat/mouse and none involving the basic and acidic residues. The data from rat uterus indicate that TIMP-3 is produced in large amount in the postpartum uterus and that it is colocalized to a great extent with heparan sulfate. Both compounds occur in the epithelium and upper stroma and not in the deeper stroma of the endometrium. At the same time in involution, the uterus is also producing a high level of matrilysin in the epithelial cells around the lumen. One role of TIMP-3 might be to regulate the matrilysin activity so that it does not attack the immediately underlying stroma. In this respect, it is interesting that matrilysin plays a role in protection against bacterial infection. In the intestine, this takes the form of activating precursors of ␣-defensin peptides with bactericidal activity (31). It is possible that a similar defense role is played in the postpartum uterus, which would be quite susceptible to infection, and that TIMP-3 would be an important regulator of this process.
Why would TIMP-3 be firmly bound to the ECM, whereas the other TIMPs are relatively soluble? Anand-Apte et al. (32) have suggested several reasons, based on the ability of TIMP-3 to suppress growth of tumors when melanoma cells are transfected with TIMP-3. First, the deposition of TIMP-3 in the surrounding matrix may prevent local expansion of the tumor by blocking MMP activity. Second, TIMP-3 may retard the release of sequestered growth factors, needed for tumor growth, from the ECM. Third, the protein may inhibit angiogenesis, preventing adequate blood supply to the tumor.
The anti-adhesive property of TIMP-3 has been observed in fibroblast cells (33), and this might sensitize the anoikis processes. Interestingly, whereas TIMP-1 and -2 are known for their growth effects (5), overexpression of TIMP-3 can induce apoptosis in melanoma cells (34) and smooth muscle cells (35). In certain cell types such as melanoma, smooth muscle, MCF-7, and HT109 cells overexpression of TIMP-3, but not of TIMP-1 or TIMP-2, promotes entry into the cell cycle, which could lead to induction of apoptosis (36), and stabilizes pro-apoptotic surface molecules such as tumor necrosis factor-␣ receptor (37). It has been shown that TIMP-3 added exogenously to cell culture can induce apoptosis by extracellular action and that this effect is not due to inhibition of matrix metalloproteinases as shown by the lack of effect of the inhibitor BB-94 (36). Finally, many GAG chains are part of proteoglycan molecules attached to the cell surface; TIMP-3 bound in this location would be wellpositioned to inhibit the shedding of tumor necrosis factor-␣ (10) and the syndecans 1 and 4 (38).