Loss of Responsiveness to Transforming Growth Factor β Induces Malignant Transformation of Nontumorigenic Rat Prostate Epithelial Cells

TGF⁷-βs are multifunctional polypeptide growth factors with many biological activities that are relevant to the suppression of tumorigenesis. These include inhibition of epithelial cell proliferation, induction of apoptosis or terminal differentiation, and maintenance of genomic stability (1, 2, 3, 4, 5). The widespread distribution of both TGF-βs and their cognate receptors in essentially all epithelial tissues suggests that the TGF-β family may play key roles in homeostasis in multiple organ systems. Recent advances in the molecular basis of TGF-β signal transduction have yielded insights into this process (2, 6). Signaling by TGF-β is initiated by binding to the type II TGF-β receptor, prototype of a novel family of transmembrane serine-threonine kinases, followed by recruitment and phosphorylation of a second serine-threonine kinase, the type I receptor. The activated type I receptor in turn phosphorylates and activates downstream signaling components of the Smad family, which translocate to the nucleus and effect changes in gene expression. With the identification of the genes for these components, evidence has emerged that the entire TGF-β response system may function as a novel tumor suppressor pathway. Homozygous inactivation of TGF-β receptors or Smad signal transducers has now been demonstrated in a variety of primary human tumors or tumor cell lines, particularly those of the gastrointestinal tract (2, 7, 8, 9).

Although initial evidence supporting the tumor suppressor role of the TGF-β pathway was correlative, there is now substantial support for a causal connection. We and others have shown that transgenic inactivation of TGF-β response, using a dominant-negative mutant type II TGF-β receptor in vivo, enhances tumorigenesis in the mammary gland and skin (10, 11). Conversely, restoration of wild-type TGF-β receptors can reduce tumorigenicity in several tumor cell lines that have defective receptors (12, 13, 14). Homozygous loss of Smad3 or Smad4/DPC4 promotes tumorigenesis in the colon in some experimental mouse models (15, 16), and we have demonstrated recently that TGF-β1 heterozygous null mice show enhanced tumorigenesis in the liver and lung when challenged with carcinogens (17). Thus, the major issues are no longer whether the TGF-β system has tumor suppressor activity but rather when, where, and how it serves this function.

The prostate is the leading site for new cancer incidence in the American male, with 29% of all cancers newly diagnosed in males in 1998 occurring at this site (18). The TGF-β system is a candidate tumor suppressor pathway in the prostate based on a number of lines of evidence. Both TGF-βs and their cognate receptors are expressed in the developing and mature prostate (19, 20, 21), and TGF-β can inhibit growth and induce apoptosis in prostatic epithelial cells both in vitro and in vivo (22, 23, 24). These observations suggest a role for endogenous TGF-βs in the maintenance of prostatic epithelial homeostasis. Several recent studies have shown that immunohistochemical staining for the type I and type II TGF-β receptors is decreased in the malignant prostatic epithelium, with diminished expression correlating with increased Gleason grade of the tumor, and associating with poor prognosis (25, 26, 27). The LNCaP human prostate cancer cell line has been reported by different groups to have either a structural mutation in the type I receptor or an absent type II receptor, and restoration of the respective wild-type receptor in these cells restores sensitivity to TGF-β and reduces their colony-forming ability in vitro (14, 28). However, as for many other tumor types, advanced human prostate carcinomas show a paradoxically increased staining for TGF-β1 (25), and plasma levels of TGF-β1 are significantly elevated in patients with invasive prostate cancer (29). Furthermore, experimental overexpression of TGF-β in rat prostatic carcinoma cells increases their tumorigenicity (30). Thus, the role of TGF-β in prostatic carcinogenesis is likely to be complex and may differ at different stages of the tumorigenic process.

One of our goals has been to develop model systems that will allow us to clarify these complex roles for the TGF-β system in the prostate. We have previously developed and characterized a rat prostatic epithelial cell line, NRP-152, with a number of properties that make it particularly attractive for this type of analysis (23). This spontaneously immortalized epithelial cell line was derived from a histologically normal region of the dorsolateral prostate from a carcinogen-treated Lobund-Wistar rat. NRP-152 cells are highly sensitive to TGF-β, are nontumorigenic, and have stem cell-like properties in vitro (31). Furthermore, they can reconstitute normal prostatic ductal epithelium when coimplanted with embryonic urogenital sinus mesenchyme in vivo (32). In modeling the initiated prostatic epithelial cell, they provide a highly useful system in which to examine the potential role of TGF-β in the early stages of prostatic tumorigenesis. Here, we have introduced a dominant-negative mutant form of the TGF-β type II receptor (DNR) into the NRP-152 cells to test the hypothesis that loss of the type II TGF-β receptor function plays a causal role in prostatic tumorigenesis. Expression of the DNR abolished in vitro responses to TGF-β and caused the cells to become tumorigenic in nude mice. This is the first demonstration that loss of TGF-β responsiveness can induce the malignant transformation of a preneoplastic cell, and our results provide strong evidence that the type II TGF-β receptor has tumor suppressor activity in the prostate.

The human TGF-β type II receptor sequence between nucleotides 1 and 687, encoding the entire extracellular and transmembrane domains together with seven amino acids of intracellular domain followed by a stop codon at the 3′ end, was generated by PCR amplification from the pCMV5-HA-TβRIIK277R plasmid (33). The PCR product was ligated into the NcoI/BamHI sites of the proretroviral vector pMFG (kindly provided by Dr. S. Kim, Seoul National University, Seoul, South Korea) to generate the construct pMFG-DNR. Then a 1.5-kb BamHI fragment containing an IRES and the neomycin-resistance gene (neo) was inserted into the BamHI site of pMFG-DNR. In this way, pMFG-DNR-IRES-Neo was generated (Fig. 1). The sequence was verified by dye terminator sequencing using the ABI Big Dye sequencing kit and found to be correct. However, it should be noted that the HA tag present both in this construct and in the widely used parent pCMV5-HA-TβRIIK277R plasmid (YDVPDYASL) differs slightly from the recognized consensus sequence for the immunodominant HA epitope (YPYDVPDYA). In our experience, this modified tag has a decreased affinity for commercially available anti-HA antibodies, making it difficult to detect except at very high expression levels. Expression in pMFG-DNR-IRES-Neo is driven by the retroviral LTR and yields a bicistronic message in which translation of the downstream neomycin resistance cistron is initiated using the IRES. A similar construct (pMFG-CAT-IRES-Neo), which contains the bacterial CAT gene in place of the DNR was used in our study as a negative control.

NRP-152 cells were grown in GM1 medium (DMEM/F12, 10% fetal bovine serum, 0.1 μm dexamethasone, 5 μg/ml insulin, 20 ng/ml epidermal growth factor, and 10 ng/ml cholera toxin) and passaged as described (23, 34). All cells were used at passages less than P40 and were periodically checked for Mycoplasma contamination.

The proretroviral plasmid was transfected into the amphotropic packaging cell line BOSC 23 (35) to generate infectious retrovirus. Viral titers were typically ∼10⁷ infectious particles/ml. After infection, NRP152 cells were allowed to recover for 24 h in GM1 medium, and then G418 (200 μg/ml) was used to select stably infected cells. After 10 days of selection, surviving cells were pooled and cultured for further analysis, and clonal subpopulations were also derived by ring-cloning. Subsequently, cells were cultured in the presence of 80 μg/ml of G418 at all times.

RNA was isolated from cells in culture by guanidinium isothiocyanate extraction and column purification using the RNeasy kit (Qiagen, Santa Clarita, CA) as described (36). Tumor RNA was isolated from mouse tissues using the Trizol reagent according to the manufacturer’s instructions (Life Technologies, Gaithersburg, MD). For Northern blot analysis, blots were hybridized with the relevant ³²P-labeled probe using Church hybridization conditions (37). RT-PCR was performed using the SuperScript One-Step RT-PCR System (Life Technologies). The primer pair was designed to be specific for the bicistronic mRNA originating from the retroviral constructs and were located in the 5′-untranslated region derived from the pMFG vector (sequence, 5′-GCATCGCAGCTTGGATACAC-3′) and the IRES fragment (5′-TTATTCCAAGCGGCTTCGGC-3′).

Cell monolayers at 50–60% confluency were incubated with 200 pm ¹²⁵I-labeled TGF-β1 (93 μCi/μg; DuPont NEN, Boston, MA) at 4°C for 4 h with gentle shaking. Freshly prepared disuccinimidyl suberate (Pierce Chemical Co., Rockford, IL) at a final concentration of 0.6 mm was used to cross-link the bound ¹²⁵I-labeled TGF-β to the receptors at 4°C for 1 h. After solubilization in lysis buffer [1% Triton X-100, 10 mm Tris-HCl (pH 7.4), 1 mm EDTA, 0.1 mm phenylmethylsulfonyl fluoride, 1 μg/ml pepstatin, and 1 μg/ml leupeptin], protein samples were separated by SDS-PAGE on a 10% gel under reducing conditions, and cross-linked complexes of ¹²⁵I-labeled TGF-β1 bound to the TGF-β receptors were visualized by autoradiography.

Cell proliferation was measured by [³H]thymidine incorporation or by the colorimetric MTT incorporation assay (38). For the [³H]thymidine incorporation assay, 50,000 cells/well were plated in 24-well plates in DMEM/F12, 1% calf serum. After growing for 24 h, TGF-β or BMP2 were added at various concentrations for an additional 24 h. Cells were then pulsed with 0.25μCi/well of [³H]thymidine, and incorporated label was determined as described (39). For the MTT assay, cells were plated at 100 cells/well in 96-well plates in GM1 medium. After varying periods of time, cells were incubated with 25 μl/well of 5 mg/ml MTT for 2 h at 37°C and lysed by addition of 100 μl/well of 20% SDS in N,N-dimethylformamide for 24 h prior to reading the absorbance at 570 nm. All samples were done in triplicate for the [³H]thymidine incorporation assay and in replicates of nine for the MTT assay.

Cells (2 × 10⁶) were plated in 10-cm dishes in DMEM/F12 containing 1% calf serum. TGF-β1 (10 ng/ml) was added 24 h after plating, and cells were detached by trypsinization 48 h later. Internucleosomal DNA ladder was detected with a modification of TACS apoptotic DNA ladder kit (Trevigen, Gaithersburg, MD) as described (24) .

Cells were plated and treated with TGF-β as above, except that the medium included 0.1 μm dexamethasone. Induction of fibronectin was determined by harvesting cells after 24 h of TGF-β exposure for Northern blot analysis. Cell differentiation status was determined 72 h after TGF-β addition by examination of cell morphology and by Northern blot analysis probing for CK-18, a marker of luminal differentiation (31).

Cell suspensions trypsinized from subconfluent monolayers were suspended in a mix of DMEM/F12:Matrigel (Becton Dickinson Labware, Bedford, United Kingdom; 1:1, v/v) at a density of 1 × 10⁷ cells/ml. Matrigel was included to shorten tumor latency and enhance tumor growth (23). Cells were used within two passages (pooled cells) or five passages (ring clones) after infection, and the cumulative number of passages from the time of original derivation of the NRP152 parent cell line was <35. Five-week-old intact male athymic mice were inoculated s.c. on the hind flank with 0.2 ml of the cell suspension (2 × 10⁶ cells). Mice were palpated weekly. Pilot experiments showed that there was no increase in tumor size after 7–8 weeks after injection, at which point palpable lesions were 0.4–0.5 cm in diameter (data not shown); therefore, all mice in the definitive experiment were sacrificed at 8 weeks after inoculation. For this study, there were five to nine animals/group and two injection sites for each animal, for a total of 10–18 potential tumor sites/group. Tissue harvested from the injection site was bisected. Half of the tissue was fixed in neutral buffered formalin and used for routine histology, while the remaining half was snap-frozen in liquid nitrogen for molecular analysis. Histological slides were read by a board-certified pathologist (W. T. P.).

NRP-152 cells are very difficult to stably transfect with high efficiency.⁸ To overcome this problem, we generated replication-defective retroviruses containing either a truncated dominant-negative mutant form of the type II TGF-β receptor (DNR) or the CAT gene as a control (Fig. 1). After viral transduction and G418 selection, pools of transduced cells were harvested for further analysis, and a number of ring clones were also generated. Those ring clones with the highest expression (NRP-152 DNR #71 and NRP-152 CAT #74) were selected for analysis along with the cell pools. Northern blots showed that the 4.5-kb bicistronic DNR/neo mRNA was expressed in the DNR pool and DNR#71 clone in great excess over the endogenous type II receptor mRNA, which was undetectable at this exposure (Fig. 2,A). Ligand affinity cross-linking using ¹²⁵I-labeled TGF-β showed that the DNR protein was also expressed at a very high level, ∼50-fold higher than the endogenous type II receptor (Fig. 2,B). The DNR appeared as a specific band migrating with an apparent molecular weight of M_r 46,000–55,000 upon reduction, suggesting that the M_r 25,000 protein core is being heavily glycosylated, as is also seen for the native type II receptor (4). Consistent with our previous findings with the DNR construct (10), there was some up-regulation of endogenous type I, type II, and type III receptor in the cells expressing the DNR (Fig. 2 B). This may reflect changes in the formation or turnover of the endogenous receptor-ligand complex in the presence of the DNR. Overall, the results indicate that DNR protein is highly expressed in the transduced cells, in high stoichiometric excess over the endogenous receptor, and that it is processed correctly and can specifically bind TGF-βs.

We have shown previously that growth of the NRP-152 cells is potently inhibited by TGF-β (23). To test whether the DNR construct could block this effect, growth inhibition assays were performed on the transduced cells. Both the DNR pool and the clone (DNR#71) were resistant to growth inhibition with treatment up to 5 ng/ml of TGF-β1, TGF-β2, or TGF-β3, whereas NRP-152 parent cells and CAT pool cells retained sensitivity to all TGF-βs (Fig. 3). Like TGF-βs, BMPs and their cognate receptors are expressed in the prostatic epithelium and have growth-inhibitory effects (40). Given the structural similarities between the receptors for the TGF-β superfamily members, we tested whether the DNR construct could interfere with the effects of the BMPs. However, although BMP-2 significantly inhibited the proliferation of the parent NRP-152 cells, this effect was not blocked by the presence of the DNR (Fig. 3), suggesting that the dominant-negative effect of the DNR construct is specific for the TGF-βs.

TGF-β1 can induce apoptosis in NRP-152 cells under certain culture conditions (24). Treatment with TGF-β1 (10 ng/ml) for 48 h induced DNA laddering, a hallmark of apoptosis, in both NRP-152 parent and NRP-152 CAT pool cells when these were grown in medium containing low serum (Fig. 4, Lanes 1–4). The presence of the DNR completely blocked the ability of TGF-β to induce apoptosis (Fig. 4, Lanes 5–8). Similar results were obtained with the terminal deoxynucleotidyl transferase nick end labeling method of apoptosis detection on cell monolayers (data not shown).

NRP-152 cells have the phenotype of basal prostatic epithelial cells. When these cells are cultured in growth factor-deficient medium under conditions that suppress apoptosis, the cells undergo growth arrest and luminal differentiation, and treatment with TGF-β enhances this effect (31). For the parent and CAT control cells, culturing in low serum for 72 h caused a slight induction of CK-18, a marker of luminal differentiation (Fig. 5,A, Lanes 1 and 3). A more dramatic induction was seen upon treatment with TGF-β1 (Fig. 5,A, Lanes 2 and 5). However, the DNR completely blocked the induction of CK-18 (Fig. 5,A, Lanes 5–8). These results were confirmed by analysis of cell morphology, with parent and CAT pool cells showing a luminal phenotype, whereas the DNR pool and clone retained their basal phenotype after treatment with TGF-β (Fig. 5 B).

In addition to its effects on cell proliferation, apoptosis, and cell differentiation, TGF-β also regulates the expression of extracellular matrix genes, such as FN (41). Interestingly, we found that the presence of the DNR could decrease, but not totally block, the induction of FN mRNA expression by TGF-β (Fig. 6). Quantitation of the Northern blots showed that TGF-β induced a 4–6-fold increase in FN mRNA in parental and CAT control cells but only a 2.5-fold increase in DNR cells. This suggests either that the matrix gene induction response is less dependent on the activity of the type II receptor than are the other responses, as has been suggested by others (42, 43), or that the matrix gene induction response may include a component that is independent of the classical type II TGF-β receptor. Overall, our results showed that the DNR can completely abolish the ability of TGF-β to inhibit proliferation, induce apoptosis, and induce differentiation of the NRP-152 cells in vitro, while partially diminishing the ability to induce extracellular matrix gene expression.

We have shown previously that NRP-152 cells make TGF-βs at low levels, and that addition of anti-TGF-β antibodies stimulated the incorporation of [³H]thymidine, suggesting that these cells have a functional negative TGF-β autocrine loop (34). Consistent with this, we found that under conditions of low serum, the DNR cells incorporated more [³H]thymidine than did parental or CAT control cells, indicating that a greater fraction of the cells were in, or had passed through, S-phase (Fig. 7,A). Under the same culture conditions, the DNR cells showed little or no CK-18 or FN expression, whereas the parent and CAT control cells showed a higher basal level of expression of both genes, which could be attributed to induction by autocrine TGF-β (see Fig. 5,A, Lanes 1 and 3 versus Lanes 5 and 7, and Fig. 6, Lanes 1 and 3 versus Lanes 5 and 7). Thus, the DNR can block the ability of autocrine TGF-β to decrease the fraction of cells in S-phase or to induce luminal differentiation, and it diminishes basal levels of expression of FN. However, under the conditions of low serum that are required to see an effect of TGF-β on the fraction of cells in S-phase, as measured by [³H]thymidine incorporation, the cells do not show sustained proliferation but rather undergo apoptosis or differentiation. In the enriched medium that is required for sustained proliferation, the DNR had no effect on lag phase, exponential growth rate, or saturation density (Fig. 7 B). This is consistent with previous work that showed that high cell density, serum, growth factors, or extracellular matrix can block the growth-inhibitory response of the rat prostate cancer cell line, MATLyLu, to TGF-β (44). Overall, our results suggest that phenotypic expression of a TGF-β autocrine loop by the NRP-152 cells is critically dependent on the culture conditions, but that when expressed, it can be blocked by the DNR.

All of the results above indicate that the DNR can effectively inactivate signaling from the type II TGF-β receptor in the NRP-152 cells and block responses to TGF-β, whether this is derived from autocrine production or from exogenous sources. This allowed us to address the key question, i.e., whether loss of TGF-β responsiveness could induce malignant transformation in this preneoplastic cell line. To do this, we tested the tumorigenicity of the NRP-152 parent cells and their genetically modified derivatives by injection into nude mice. Mice injected with either DNR pool or clone started to form palpable lesions by 5 weeks after inoculation, and by 8 weeks, 21 of 34 injection sites showed palpable lesions, compared with only 1 of 30 for parental or CAT control cells. All of the palpable lesions were excised and analyzed histologically, and the majority (>60%) of the lesions formed by the DNR cells were found to be carcinomas, whereas the sole lesion in the control group was a squamous cyst. A summary of the results is shown in Table 1. Overall, the DNR pool and clone formed carcinomas at 13 of 34 possible sites compared with 0 of 30 for the parent or CAT control cells (P = 0.0001). Similar results for tumor take were obtained in a second independent experiment in which tumors formed at 12 of 20 sites for the DNR pool and clone, whereas no tumors (0 of 20) formed with parent or CAT cells (P = 0.0001; data not shown). This indicates definitively that inactivation of the type II TGF-β receptor can cause malignant transformation.

At the histological level, all of the carcinomas were characterized by the presence of cells having large pleiomorphic nuclei with prominent nucleoli and by frequent mitotic figures. For the experiment shown, 8 of the 13 carcinomas were classified as adenocarcinomas, showing identifiable glandular structures (Fig. 8,A), whereas 5 were identified as squamous carcinomas, containing sheets and whorls of keratinizing cells with sharply defined cells borders and foci of dyskeratotic cells (Fig. 8,B). Squamous cysts (Fig. 8 C) were also a frequent finding, arising at 11 of 34 sites in mice injected with NRP-152 DNR pool or clone and in 1 of 30 sites for parental or CAT control cells (P = 0.003). In three cases, cysts coexisted with carcinoma and comprised the bulk of the palpable lesion.

To determine whether expression of the DNR was maintained by the tumors, RNA was prepared from five representative tumors and analyzed by Northern blot. Tumor RNA that was negative on the Northern blot was reanalyzed by RT-PCR to increase the sensitivity of detection. Together, the results showed persistent expression of the DNR in four of five tumors analyzed (Fig. 9). The RNA from the one negative tumor was quite degraded, suggesting that it may have been a false negative. The presence of the DNR in tumors after 8 weeks of growth in the nude mouse suggests that continued expression of the DNR was probably important for tumor maintenance, not just initial establishment. However, because the tumors were no longer growing progressively at 8 weeks despite a relatively aggressive histology, possibly by this stage DNR expression had dropped below a threshold required to fully block TGF-β responsiveness. The relative level of DNR expression in the DNR tumors and the DNR cell lines could not be compared directly because of the variable contribution of DNR-negative host cells to the tumors. We are presently generating a tetracycline-regulatable form of the DNR retrovirus. This will allow us to rigorously address the questions of whether persistent expression of the DNR is necessary for tumor maintenance and whether relative DNR expression levels are important in determining whether the transduced cells form benign cysts or carcinomas in vivo.

We have provided the first evidence that loss of TGF-β responsiveness can induce malignant transformation of a nontumorigenic epithelial cell line. The NRP-152 prostatic epithelial cell line that we have used for this study has stem cell-like properties, as indicated by its ability to reconstitute a functional prostatic epithelium when recombined in vivo with embryonic urogenital sinus mesenchyme (32). Because work in other systems has suggested that the stem cell compartment of epithelial tissues may be the prime target of transforming events (45), the NRP-152 cells are a particularly attractive model system for elucidating key steps in prostatic tumorigenesis. Our data suggest that loss of TGF-β responsiveness may be a critical early event in this process, and, by extension, that TGF-β plays a key role as a tumor suppressor in the normal prostate.

Defects in TGF-β receptors or downstream signaling components are found in a significant fraction of primary human tumors (7, 8). In hereditary nonpolyposis colorectal cancer, inactivation of the type II TGF-β receptor has been shown to occur at the late adenoma-to-carcinoma transition (46). However, for other human tumors, it is not clear at which point in the tumorigenic process the tumor suppressor activity of the TGF-β pathway is critical. Immunohistochemical studies have shown that expression of the type II TGF-β receptor is reduced in human prostate tumors when compared with surrounding normal tissue, and that expression is further diminished as the histological grade of the tumor increases (25, 26, 27). Our data indicate that this loss of receptor expression is likely to be causally involved in the development of prostate tumors, rather than being an epiphenomenon. Furthermore, our data suggest that loss of receptor function at the early stages of prostatic tumorigenesis could have particularly profound effects by causing malignant transformation of the initiated cells.

The majority of the tumors that were formed by the DNR cells were adenocarcinomas, which is the predominant histotype of human prostatic tumors. This provides further support for the utility of the NRP-152 model system. However, it should be noted that the DNR adenocarcinomas had a more extensive stromal component than is normally seen in the human tumors. The remaining DNR tumors were squamous carcinomas, which are a relatively common finding in genetically engineered or chemically induced mouse models of tumorigenesis, and may simply reflect some fundamental difference in biology between mice and humans. Alternatively, although squamous metaplasia is relatively rare in the human prostate, it is seen in association with androgen deprivation (47), and it is possible that the s.c. xenograft site does not provide adequate androgenic stimulation to prevent squamous differentiation. In some cases, the carcinomas that formed were associated with large squamous cysts, and cysts also arose from the DNR cells independently of tumor formation. TGF-β has been shown to inhibit cystogenesis in MDCK canine kidney epithelial cells through a mechanism involving the induction of type V collagen (48). Thus, the absence of a TGF-β response in the DNR cells may cause microenvironmental changes that are permissive for cystic differentiation, and this differentiation program may compete with tumorigenesis in the xenograft setting. The local level of DNR expression may be important in determining the outcome.

The role of the TGF-β system in tumorigenesis is complex. Data from other systems suggest that loss of TGF-β response probably cooperates with initiating and other mutations to increase tumor promotion and progression, at least up to the stage of locally invasive carcinoma. For example, after treatment with the carcinogen DMBA, transgenic mice expressing a DNR in basal keratinocytes were observed both to form skin tumors earlier than the wild-type controls and to show a higher rate of conversion from papillomas to malignant carcinomas (11). This agrees with our present findings in the prostate and our previous demonstration that transgenic overexpression of a DNR in the mouse mammary gland enhances tumorigenesis in response to carcinogens (10). In contrast, loss of TGF-β responsiveness suppresses the subsequent development of the metastatic phenotype in skin and mammary gland (49, 50, 51). Thus, the tumor suppressor activity of TGF-β may be conditional on the specific stage of tumorigenesis.

The complexity of TGF-β action in tumorigenesis is likely to be attributable in part to the fact that, in addition to its tumor suppressor activities, TGF-β has activities that enhance tumorigenesis. The first oncogenic activities of TGF-β that were described, i.e., induction of angiogenesis and suppression of the immune surveillance system, had stromal targets (52, 53). However, more recently, it has emerged that TGF-β may have directly pro-oncogenic effects on the epithelium through induction of the epithelial to mesenchymal transition, alteration in extracellular matrix and adhesion proteins, and induction of parathyroid hormone-related peptide. The unifying hypothesis is that TGF-β has tumor suppressor activity up to the stages of invasion and metastasis, provided that the epithelial response is intact. However, at the late stages of tumorigenesis or if epithelial response is lost, the oncogenic activities dominate. Frequently, this is accompanied by an up-regulation of TGF-β expression, presumably reflecting the positive selective advantage of overexpressing this newly oncogenic molecule. Consistent with this theoretical framework, advanced human prostate tumors, like other highly aggressive metastatic tumors, show increased TGF-β expression (25), and the experimental overexpression of TGF-β in prostate cancer cells has been shown to enhance invasion and metastasis (30).

Interestingly, because TGF-β has now been shown to have direct oncogenic effects on the epithelium, we need to consider the possibility that complete loss of TGF-β responsiveness in the target epithelium may actually be less oncogenic than a partial loss (51). Work from a number of systems suggests that the ratio of type II to type I TGF-β receptors can qualitatively affect cellular responses (54, 55), and we and others have shown that a higher level of type II receptor is required for growth inhibition than for induction of matrix gene expression (Refs. 42 and 43 and the present work). Furthermore, the growth-inhibitory effects of TGF-β on prostate tumor cells can be overcome by contextual effects such as high cell density, changes in extracellular matrix or growth factors (44). Thus, at low TGF-β RII levels or in certain biological contexts, the growth-inhibitory response may be lost, whereas other, potentially pro-oncogenic, activities may persist. It may be significant that although most advanced human prostate tumors have reduced expression of TGF-β receptors at the immunohistochemical level (25, 26, 27), only one of the three commonly studied human prostatic tumor cell lines has a structural mutation in a TGF-β receptor (28). This is consistent with the idea that an epigenetic diminution in type II TGF-β receptor levels may have a selective advantage over genetic inactivation leading to total loss. Clearly, it will be important to alter both TGF-β ligand and receptor levels systematically in a given cell type and determine what effect different combinations of these have on the various stages of tumorigenesis. The NRP-152 cells should prove a highly useful model system for this type of analysis in the prostate.

We thank Dr. Anita Roberts for continual intellectual support, Dr. Robert Lechleider for critical reading of the manuscript, Melanie Barnes for excellent technical assistance, and Drs. John Letterio and Masa-Aki Shibata for much helpful advice. We are grateful to Dr. Seong-Jin Kim for provision of the pMFG-IRES-neo proretroviral vector, to Dr. Jeffrey Wrana for the pCMV5-HA-TβRIIK277R plasmid, and to Dr. Mary-Beth Eiden for expert advice in the use of retroviruses.