Tissue Inhibitor of Metalloproteinases-2 (TIMP-2) mRNA Expression in Tumor Cell Lines and Human Tumor Tissues*

Human tissue inhibitor of metalloproteinase-2 (TIMP-2) was cloned and sequenced from an A2056 human melanoma cell cDNA library. When the se- quence was compared with that of human TIMP-1 at both the nucleotide and deduced amino acid levels, the homology appeared closer at the protein level than at the nucleotide level, suggesting that these inhibitors diverged early in the evolution of this gene family. Comparison of the deduced amino acid sequence for TIMP-2 with that of human TIMP-1 shows that there are two regions in which the similarity is below the overall average of 66%. It. is postulated that these regions are responsible for the unique ability of TIMP-2 to bind to the latent form of the 72-kDa type IV collagenase. Polyclonal anti-TIMP-2 antisera recog- nized TIMP-2 but not TIMP-1 on immunoblotting. Northern blot analysis of RNA from A2058

Human tissue inhibitor of metalloproteinase-2 (TIMP-2) was cloned and sequenced from an A2056 human melanoma cell cDNA library. When the sequence was compared with that of human TIMP-1 at both the nucleotide and deduced amino acid levels, the homology appeared closer at the protein level than at the nucleotide level, suggesting that these inhibitors diverged early in the evolution of this gene family. Comparison of the deduced amino acid sequence for TIMP-2 with that of human TIMP-1 shows that there are two regions in which the similarity is below the overall average of 66%. It. is postulated that these regions are responsible for the unique ability of TIMP-2 to bind to the latent form of the 72-kDa type IV collagenase.
Polyclonal anti-TIMP-2 antisera recognized TIMP-2 but not TIMP-1 on immunoblotting. Northern blot analysis of RNA from A2058 human melanoma, HT-144 human melanoma, HT-1080 human fibrosarcoma, and WI-38 fetal lung fibroblast cell lines demonstrated two distinct transcripts of 1.0 and 3.5 kilobases (kb) for timp-2 mRNA. Both transcripts are down-regulated in response to transforming growth factor-p but are unchanged in response to phorbol ester treatment. This is in contrast to the up-regulation of timp-1 transcripts by these agents and indicates that timp-2 and timp-1 are independently regulated in cell culture.
Northern blot analyses of matched normal and tumor tissue samples from five cases of human colorectal carcinoma were performed. Normal and tumor tissues contain both the l.O-and 3.5-kb transcripts.
However, in the tissue samples the ratio of the 3.5-kb transcript to the l.O-kb transcript was markedly elevated.
No evidence of down-regulation of timp-2 transcript levels was noted in the tumor tissues. This is in contrast to the elevated timp-1 transcript levels seen in these tumor samples. Thus, timp-2 mRNA transcript levels are differentially regulated from timp-2 levels in uiuo as well as in cell culture.
The collagenase family enzymes are a group of metalloproteinases which are secreted in the zymogen form and degrade both the collagenous and noncollagenous components of the extracellular matrix. The overproduction and unrestrained activity of these enzymes have been linked to a variety of pathologic conditions such as rheumatoid arthritis and malignant conversion of tumor cells (Okada et al., 1986;Harris et al., 1984;Werb et al., 1977;Liotta et al., 1980;Kalebic et al., 1983). The down-regulation of metalloproteinase collagenolysis and proteolysis may occur through naturally occurring inhibitor proteins, such as the tissue inhibitors of metalloproteinases (TIMPs).' TIMP-1 is a glycoprotein with an apparent molecular size of 28.5 kDa which forms a complex of 1:l stoichiometry with activated interstitial collagenase, activated stromelysin, and the 92-kDa type IV collagenase Stricklin, 1983: Welgus et al., 1985a;. The gene coding for TIMP-1 has been cloned, sequenced, and mapped to the X chromosome (Carmichael et al., 1986;Docherty et al., 1985;Mullins et al., 1988;Mahtani and Willard, 1988). The secreted protein has 184 amino acids and six intramolecular disulfide bonds. The same cells which produce interstitial collagenase are capable of synthesizing and secreting TIMP-1 (Welgus et al., 1985b;Herron et al., 1986). Thus, the net collagenolytic activity for these cell types is the result of the balance between activated enzyme levels and TIMP-1 levels. Studies have shown an inverse correlation between TIMP-1 levels and the invasive potential of murine and human tumor cells (Khokha et al., 1989).
Recently we have isolated, purified, and determined the complete primary structure of a second member of the TIMP family, TIMP-2 (Stetler-Stevenson et al., 1989). TIMP-2 is a 21-kDa protein which selectively forms a complex with the latent proenzyme form of the 72-kDa type IV collagenase (Stetler-Stevenson et al., 1989;Goldberg et al., 1989). The secreted protein has 192 amino acid residues and is not glycosylated. TIMP-2 shows an overall 71% similarity to TIMP-1 at the amino acid sequence level. The positions of the 12 cysteine residues are conserved with respect to those present in TIMP-1, as are three of the four tryptophan residues. TIMP-2 inhibits the type IV collagenolytic activity and the gelatinolytic activity associated with the 72-kDa enzyme (Stetler-Stevenson et al., 1989). Inhibition studies demonstrated that complete enzyme inhibition occurred at a 1:l molar ratio of TIMP-2 to activated 72-kDa type IV collagenase (Stetler-Stevenson et al., 1989). Thus, unlike TIMP-1, TIMP-2 is capable of binding to both the latent and activated forms of type IV collagenase. Cell culture studies using cell lines that produce a variety of collagenase family enzymes, as well as both TIMP-1 and TIMP-2 suggest that TIMP-2 preferentially interacts with the 72-kDa type IV collagenase (Stetler-Stevenson et al., 1989;Goldberg et al.,   . Thus, like interstitial collagenase activity which is the balance of activated enzyme and TIMP-1, the net 72-kDa type IV collagenase activity may depend upon the balance between the levels of activated enzyme and TIMP-2. Augmented type IV collagenolytic activity has been associated with the metastatic phenotype in a number of experimental systems. This could possibly be due to the increased production and activation of the 72-kDa type IV collagenase enzyme. However, decreased production of TIMP-2 could also result in greater effective enzyme activity. To examine the regulation of TIMP-2 we have isolated and sequenced a cDNA clone for human TIMP-2. Comparison of the cDNA sequence of timp-2 with that of timp-1 suggests that these inhibitors diverged early in the evolution of the TIMP family. We have used this probe to measure the levels of timp-2 mRNA in human tumor cell lines and the effects of phorbol ester and transforming growth factor p (TGF-@l) treatment on timp-2 mRNA levels. These effects were contrasted with those noted for timp-1 mRNA levels. Finally, we have examined timp-2 and timp-1 mRNA levels in a series of human colon adenocarcinomas and adjacent normal colonic mucosa.

Human Melanoma
Cell cDNA Library Preparation, Screening, and DNA Sequencing-Oligo(dT)-select poly(dA) mRNA was prepared from human A2058 melanoma cells using standard methods. 1 pg of purified mRNA was used to prepare double-stranded cDNA using a commercially available cDNA synthesis kit (Amersham Corp.). This cDNA was methylated using EcoRI methylase (Promega), linked to EcoRI linkers (Promega), restricted with EcoRI, and ligated to EcoRIdigested h-GEM-4 (Promega). The ligations were packaged (Gigapack Gold, Strategene) and the optimal reactions were pooled to give 1.5 x lo6 recombinants. 7.5 X lo5 recombinants were screened using oligonucleotide 27-40. Oligonucleotide 27-40, a 45-mer, with the sequence, 5'GAGAAGGAGGTGGACTCTGGCAATGACATCTAT-GGCAACAACATC3', corresponds to the reverse translation of residues 27 through 40 of the previously sequenced TIMP-2 protein.
Oligonucleotide 27-40 was synthesized on a Biosearch 8700 DNA synthesizer by means of @-cyanoethyl phosphoramidite chemistry and was labeled using y-[32P]ATP (Amersham Corp.) and T4 kinase (Bethesda Research Labaoratories). Library screening was performed using standard techniques. DNA sequencing was performed using dideoxy methodology and [?S]dATP (Du Pont-New England Nuclear).
Cell Culture, RNA Isolation, and Northern Blot Analysis-All cell lines except A2058 melanoma cells were obtained from the American Type Culture Collection, Rockville, MD. HT-1080 human fibrosarcoma cells, Wi-38 human embryonic lung fibroblasts, HT-144 human melanoma, and A2058 human melanoma cells were grown to 80% confluence in Dulbecco's modified Eagle's medium (GIBCO). The medium was then replaced with Dulbecco's modified Eagle's medium supplemented with 0.5% ITS+ (Collaborative Research) and 25 eg/ ml gentamycin. The medium was changed after 4 h and culture continued for 20 h prior to the addition of 10 rig/ml 12-O-tetra- The cDNA insert of clone pTP-MO1 was sequenced in both directions using dideoxy methodology. The predicted amino acid sequence is shown under the DNA sequence. The putative polyadenylation signal is underlined. decanoylphorbol-13.acetate (TPA, Sigma) or 5 rig/ml TGF-Pl (R & D Systems).
Total cytoplasmic RNA was isolated from cell lines as described by Gough (1988). mRNA was isolated using the FAST-TRACK mRNA isolation kit (Invitrogen). Tissue mRNA was isolated from frozen tissue fragments. Tissue fragments were obtained from partial colectomy specimens at the time of surgery. The pathologic diagnosis of all five cases was invasive adenocarcinoma. Tissue samples were also obtained from adjacent uninvolved mucosa. Frozen tissue was pulverized in liquid Nz using a mortar and pestle. The tissue powder was then dissolved in 4 M guanidine isothiocyanate, 3 M sodium acetate, 0.84% P-mercaptoethanol, pH 6.0. Total cytoplasmic RNA was isolated by pelleting through 5.7 M cesium chloride, 3 M sodium acetate, pH 6.0. Aliquots of RNA were applied to formaldehyde, 1% w/v agarose gels and electrophoresed before transfer onto Nytran filters (Schleicher & Schuell). The RNA was UV cross-linked to the filter and hybridized using standard conditions. The cDNA probes were labelled with ol-["'P]dCTP using a random primer labeling kit (Bethesda Research Laboratories). Filters were autoradiographed at -80 "C!, and quantitation on developed film was performed by scanning densitometry using an LKB 2202 laser densitometer.
A TIMP-1 cDNA probe was prepared by polymerase chain amplification reaction using cDNA prepared from oligo(dT)-selected mRNA from HT-1080 TPA-treated cells. The cDNA was prepared using a cDNA synthesis kit (Bethesda Research Laboratories) according to the manufacturer's directions. The primers for this amplification were also synthesized on the Biosearch 8700 DNA synthesizer and had the following sequences corresponding to sequences reported in the human TIMP-1 cDNA clone (Carmichael et al., 1986): primer 82 (bases 118-146), 5'TGCACCTGTGTCCCACCCCACCCACAGA-CG3'; primer 83 (complement to bases 640-669), 5'GGCTATC-TGGGACCGCAGGGACTGCCAGGT3'.
The polymerase chain amplification reaction was performed for 40 cycles using 55 "C annealing temperature and standard conditions for denaturation and extension. This reaction consistently yielded a single 551-base pair product (P551) encoding for the mature TIMP-1 protein sequence.

Z'ZMP-2
Purification and Antibody Preparation-TIMP-2 was purified from human melanoma A2058 cell-conditioned media by gelatin affinity chromatography and reverse-phase HPLC as previously described (Stetler-Stevenson et al., 1989). The purified human TIMP-2 was used as the antigen for preparation of polyclonal rabbit antihuman TIMP-2 antibodies using standard immunization protocols previously described (Hoyhtya et al., 1988). RESULTS We have previously determined the primary structure of the TIMP-2 protein by direct amino acid sequencing (Stetler-Stevenson et al., 1989). This information was used to prepare a synthetic oligonucleotide probe, oligo 27-40, which was used to screen a cDNA library constructed from mRNA isolated from human A2058 melanoma cells. Ten clones were isolated, and the nucleotide sequence of the cDNA insert in the longest clone, pTP-MOl, is presented in Fig. 1. The insert contains 1062 base Dairs excluding the polv(A) tail and encodes for the -_-.
pro-TIMP-2 protein of 220 amino acids. This protein se-quence includes a 26-residue signal peptide sequence and a mature TIMP-2 protein of 194 amino acids. The 130-nucleotide-long 3'-untranslated region contains a polyadenylation signal 30 bases upstream from the 3' end of the RNA.
Comparison of the amino acid sequence of TIMP-2 deduced from the cDNA clone with that determined by direct amino acid sequencing of overlapping endoproteinase-derived peptide fragments shows excellent agreement. The original sequence contained only 192 amino acids. The previously unidentified residues correspond to the glycyl residue at position 92 and the prolyl residue at the carboxyl terminus. Other changes are noted in Fig. 2. The homology of mature TIMP-2 with TIMP-1 at the predicted protein level is 37.6% identity and 65.6% overall similarity.
Rabbit anti-human TIMP-2 polyclonal antibodies were used in an immunoblot comparison of bovine TIMP-1 and human TIMP-2. As shown in Fig. 3, bovine TIMP-1 migrates at approximately 28 kDa, as previously reported for this glycosylated protein, compared with TIMP-2 which migrates at 21 kDa and is unglycosylated.
Northern blot analysis of oilgo(selected mRNA isolated from the A2058 human melanoma cell line revealed two specific tinp-2 mRNA species with approximate sizes of 3.5 and 1.0 kb (Fig. 4)  Lanes A and C contained 1 pg of TIMP-1. Lanes B and D contained 1 pg of TIMP-2. Lanes A and B were silver-stained following electrophoresis. Lanes C and D were stained with rabbit anti-human TIMP-2 antisera (1:200) followed by goat anti-rabbit horseradish peroxidase complex after transfer to Immobilon membrane (Millipore). Oligo(dT)-selected RNA was isolated from cells as described under "Experimental Procedures." After transfer to Nytran filters RNA was hybridized with "'P-labeled probe pT2-MOl, specific for tinp-2, or alternatively, probe P551, specific for timp-1.
Relative positions of the 28 and 18 S ribosomal RNA bands are noted as are the approximate transcript sizes.
fold induction of the timp-1 transcript levels. These findings are consistent with the induction of timp-1 by this agent as previously reported (Edwards et al., 1985;Welgus et al., 1985b).  of differential repression by TGF-fil. Treatment of A2058 melanoma cells with TGF-@l reduced steady-state timp-2 transcript levels to 46 and 59% of control values for the 1.0 kb and 3.5kb transcripts, respectively. Treatment of HT-144 cells with TGF-/31 resulted in reduction of timp-2 transcripts to 42 and 47% of control levels for the 1.0 and 3.5-kb message, respectively. Both timp-2 transcripts showed a more moderate TGF-fll-induced reduction in the HT-1080 fibrosarcoma cell line of 25% from control levels. These effects are in contradistinction to the results observed for timp-1 message levels.
In the tumor cell lines (HT-144, HT-1080, and A2058) TGF-@l induced an increase in timp-1 steady-state transcript levels to 150% of control values. Treatment of Wi-38 fibroblasts with TGF-Pl resulted in a g-fold increase of message levels. These effects for timp-1 are consistent with the report by Overall et al., (1989) which demonstrated that TGF-/3 induces timp-1 mRNA levels. These observations clearly demonstrate that timp-2 and timp-1 are differentially regulated in all four cell lines tested. Northern blot analyses of tissue from five primary human colorectal tumors and patient-matched adjacent normal mucosa were performed using both the timp-1 and timp-2 probes (Fig. 6). Ethidium bromide staining of the formaldehyde gels prior to transfer demonstrated equal loading of RNA in all lanes (data not shown). The results demonstrate that both the LO-and 3.5-kb timp-2 transcripts are present but that the l.O-kb message level is markedly reduced compared with that seen in the RNA isolated from cultured human tumor cells. Transcript levels for timp-2 show little correlation with tissue origin (normal uerws tumor). In most tumor samples the timp-2 transcript levels show no difference from the adjacent normal tissue, with the exception of tumor T2 in which there is a marked decrease in timp-2 message levels compared with that seen in the adjacent normal colonic mucosa N2. These data suggest that primary colon adenocarcinomas may be heterogeneous with respect to the levels of timp-2. However, timp-1 transcript levels are obviously elevated in all tumor tissue samples, including sample T2, compared with normal adjacent mucosa. Again, these observations suggest that timp-2 and timp-1 are independently regulated.

DISCUSSION
Isolation and sequencing of a cDNA clone for human TIMP-2 was performed. Characterization of the TIMP-2 cDNA clone pTP-MO1 confirmed that TIMP-2 is a unique gene product independent of TIMP-1. The protein sequence predicted from the TIMP-2 cDNA is in excellent agreement (95%) with the amino acid sequence obtained for this protein by direct protein sequencing (Stetler-Stevenson et al., 1989). Pustell Matrix analysis of the homology distribution between these two predicted protein sequences using a cut-off value of 66% and an 8-amino acid overlap demonstrates that there are two areas in which the homology falls below this average value (Fig. 7A). TIMP-2 shows a distinct preference for binding to the latent form of the 72-kDa type IV collagenase in the presence of both other latent metalloproteinases and TIMP-1 (Stetler-Stevenson et al., 1989;Goldberg et al., 1989). However, both forms of TIMP will inhibit activated type IV collagenase. Thus regions of amino acid sequence that are highly conserved between these proteins, such as those that exceed the overall homology value of 66%, may be responsible for the known shared functions of these proteins, inhibition of the activated collagenase family of enzymes. Areas of low homology are likely to be responsible for those functions which are unique for individual TIMP molecules. Thus, the regions of low homology between residues 45 and 70 and the carboxyl terminus of TIMP-2 may be responsible for the binding of TIMP-2 to the latent form of the 72-kDa type IV collagenase. Such regions must exist and account for the failure of TIMP-1 antibodies to detect TIMP-2, as previously reported , as well as the failure of TIMP-2 antibodies to detect TIMP-1 as demonstrated in the current report (Fig. 3).
Comparison of the cDNA sequences of human timp-2 with human timp-1 shows little homology considering that seen at the amino acid level (Fig. 7B). This result implies that these genes diverged early in the evolution of this gene family. The lack of homology at the cDNA level may also explain why timp-2 mRNA transcripts are not detected in Northern blot analyses using timp-1 probes and also why screening cDNA libraries with timp-1 probes fails to yield timp-2 clones.
Northern blot analyses of oligo(dT)-selected poly(A) RNA as well as total cytoplasmic RNA from human tumor cell lines detect the presence of two transcripts when probed with the full-length timp-2 cDNA clone pT2-Mol. The origin of these two specific transcripts remains to be determined. The size difference is too large to be easily accounted for by differences in 3'-polyadenylation, although large differences in 3'-portions of the pro-cYP(1) collagen transcripts, attributable to alternative polyadenylation signals, have been observed (Myers et al., 1983). It is possible that alternative 5'-untranslated regions could account for the different transcript sizes, as has been demonstrated for insulin-like growth factor II mRNAs (Irminger et al., 1987). TGF-Pl has been demonstrated to increase timp-1 mRNA levels in human gingival fibroblasts (Overall et al., 1989). In the presence of other growth factors, TGF-8 also has a selective reciprocal effect on interstitial collagenase and timp-1 expression (Edwards et al., 1987). TGF-P selectively represses the induction of interstitial collagenase but interacts synergistically to superinduce TIMP-1. The results of the present report are consistent with these previous observations. Both TPA and TGF-/31 induced increases in timp-1 transcript levels over basal levels. This suggests that timp-1 up-regulation occurs in a similar fashion in the melanoma and fibrosarcoma cells studied in the present report as in the fibroblast cells studied by other investigators. We have previously demonstrated that TGFB-1 induces the 72-kDa type IV collagenase mRNA and protein levels in the same tumor cell lines studied in the present report. ' We now demonstrate that TGF-81 decreases timp-2 mRNA transcript levels. Thus, TGF-/31 treatment has an opposite effect on timp-2 compared with the 72-kDa type IV collagenase transcript levels in human tumor cells. This suggests that TGF-81 treatment may result in augmented type IV collagenolytic activity due to up-regulation of the enzyme coupled with down-regulation of an associated inhibitor, TIMP-2. This could result in an enhanced invasive phenotype of tumor cells treated with TGF-/31, although TGF-@l does induce an increase in timp-1 transcript levels. These observations demonstrate the complex multilevel regulation of type IV collagenolytic activity. However, it is clearly evident that the transcriptional regulation of timp-2 is independent of timp-1.
Northern blot analyses of human colorectal tumor and adjacent normal tissues again demonstrated two mRNA transcripts when probed with the timp-2 cDNA clone. However, in the RNA from these tissue samples the 3.5-kb transcript clearly predominates, with only trace amounts of the l.O-kb message detectable in the normal tissues. There was no correlation of timp-2 transcript levels and adenocarcinoma tissues. However, transcript levels for timp-1 show a distinct correlation with the malignant tumor samples. All adenocarcinema tissue samples showed elevated timp-1 levels compared with adjacent normal colonic mucosa. This result was highly unexpected in light of the malignant nature of the tumor tissues examined and the report of Khokha et al. (1989) which demonstrated an inverse correlation between timp-1 mRNA levels and the invasive phenotype. Immunohistochemical studies of the distribution of TIMP-1 protein in these tumor samples will be helpful in developing an understanding of these observations.
In summary, we have cloned and sequenced a full-length cDNA which encodes the pro-TIMP-2 protein. At the protein level there is a marked homology between members of the TIMP family. However, at the nucleotide level this homology is notably less, suggesting early divergence of these genes. Finally, the data presented clearly demonstrate that timp-2 is regulated independently of timp-1, both in cell culture as evidenced by studies using TPA and TGF-@l and in vivo as evidenced by comparison of transcript levels for these inhibitors in human colon adenocarcinoma tissue and adjacent normal colonic mucosa.