Transforming Growth Factor β1 Suppresses Nonmetastatic Colon Cancer at an Early Stage of Tumorigenesis¹

Many lines of evidence demonstrate that the TGF⁴-β signaling pathway is important in suppressing the development of colon cancer. Cell lines derived from human colon tumors are frequently resistant to the growth-inhibitory effects of TGF-β1, and this resistance is often associated with increased invasive characteristics (1, 2). Mutations in the human TGFBR2 have been found in both sporadic and inherited colon cancers exhibiting increased rates of genetic instability (3, 4, 5). In addition, inactivating mutations in SMAD2 and SMAD4, two members of the family of intracellular proteins responsible for transducing signals from the activated TGF-β receptors, have been found in human colon cancer (6, 7, 8). Deficiency of SMAD4 in the Apc^Δ716 mouse model of small intestinal cancer has been shown to increase adenoma size and promote progression to invasive carcinoma (9). Finally, SMAD3-deficient mice bred onto a 129 background have been shown to develop metastatic colon cancer (10).

The multifunctional nature of the TGF-β family suggests several potential mechanisms by which defects in the TGF-β pathway may contribute to tumorigenesis. A commonly held view is that TGF-β1 prohibits tumor cell proliferation because TGF-β1 inhibits epithelial cell growth in vitro. TGF-β1 promotes remodeling of the extracellular matrix, which may mediate tumor cell-matrix interactions and epithelial cell differentiation (11). In addition, TGF-β1 is a potent regulator of immune and inflammatory cells (12, 13, 14, 15). By regulating immune cell function, it is thought that TGF-β1 reduces local production of growth factors and tissue damage induced by free radicals (16, 17). Consequently, growth control, regulation of epithelial cell differentiation and cell matrix interaction, and protection from genetic damage caused by inflammatory cells could all be significant in the initiation, promotion, or progression of colon cancer.

A mixed strain (129S6 × CF-1) of Rag2^−/− mice, which lack both B and T cells, develops an inflammation-associated hyperplasia specific to the cecum and colon shortly after weaning. We examined the role of TGF-β1 in suppressing intestinal cancer by introducing a Tgfb1 null mutation into this strain of mouse. Because immunocompetent Tgfb1^−/− mice die at ∼3 weeks of age due to organ failure resulting from a multifocal inflammatory disease (12, 13), placement of the Tgfb1 null allele on an immunodeficient background permits the Tgfb1^−/− mice to live to adulthood (18) so that tumorigenesis can be followed. Tgfb1^−/− Rag2^−/− mice rapidly develop carcinoma. By 5 months of age, all Tgfb1^−/− Rag2^−/− mice exhibit multiple carcinomas in the cecum and colon. In contrast, ceca and cola from nearly all Tgfb1^+/+ and Tgfb1^+/− Rag2^{− /−} mice remain hyperplastic, suggesting that inflammation-associated hyperplasia in the absence of TGF-β1 predisposes to cancer. No differences in inflammation, cell proliferation, or apoptosis were found among hyperplastic tissues from Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb^−/− Rag2^{− /−} mice. This suggests a critical role for TGF-β1 in nonproliferative aspects of crypt epithelial tissue integrity early in tumorigenesis, and it implies that other TGF-β ligands mediate suppressive activities later in tumorigenesis.

One breeding pair of 129S6 × CF-1 Tgfb1^+/− Rag2^{− /−} mice was generously provided by Dr. Robert L. Coffman (DNX Transgenics, Princeton, NJ). These mice were generated by crossing CF-1 Tgfb1^+/− mice with 129S6 Rag2^−/− (Taconic, Germantown, NY) mice. F₁ offspring were backcrossed to 129S6 Rag2^−/− mice to generate 129S6 (75%) × CF-1 (25%) Tgfb1^+/− Rag2^{− /−} offspring, which in turn were backcrossed to 129S6 Rag2^−/− mice for three additional generations. Subsequent breedings involved intercrossing 129S6 (97%) × CF-1 (3%) Tgfb1^+/− Rag2^{− /−} mice. Further backcrossing of Tgfb1^+/− Rag2^{− /−} mice onto a 129S6 background was not attempted because inbreeding Tgfb1^−/− mice beyond N3 generations into many backgrounds results in >50% embryo loss due to an unknown lethality (19). Mice were genotyped for Tgfb1 alleles by PCR as described (18). All mice were maintained in a specific pathogen-free facility in accordance with standard animal use protocols, and sentinel mice were routinely screened for pathogens.

The cecum and colon were dissected free from mesenchyme, opened longitudinally, and flushed of contents. After examining the luminal surface and scoring for tumors, half of the cecum and colon was frozen in liquid nitrogen and stored at −80°C. The remaining tissue was immersion fixed in 4% phosphate-buffered formalin and embedded in paraffin. A minimum of two H&E-stained sections from the cecum and colon of each mouse was evaluated by a single observer (G. P. B.), who was unaware of the genotypes of the mice from which the samples were taken. Disease stage was based upon the most severe lesion (hyperplasia, adenoma, or carcinoma) present within each sample. In addition, characterization and severity of lesions were corroborated in a blind fashion by comparative pathologists at two other institutions.

For APC and β-catenin immunostaining, deparaffinized, 5-μm sections were submerged in 0.1 m sodium citrate (pH 6.0), heated in a microwave at full power for 5 min, and cooled to room temperature. The sections were washed in PBS and incubated with a rabbit polyclonal antibody directed against either the COOH terminus of the APC protein (sc-896 diluted 1:200; Santa Cruz Biotechnology, Santa Cruz, CA) or the β-catenin protein (C-2206 diluted 1:500; Sigma Chemical Co., St. Louis, MO). The primary antibody was detected with the Vectastain Rabbit ABC kit (Vector Laboratories, Burlingame, CA). Final staining was developed with 3,3-diaminobenzidine, and sections were counterstained with aqueous hematoxylin.

For detection of proliferating cells, mice received i.p. injections of 120 μg of BrdUrd/g of body weight 1 h before sacrifice. Mice were prepared for histology as described above. Deparaffinized sections were pretreated with 2 n HCl (for 20 min at 37° C), neutralized with 1% boric acid buffer (for 1 min at 37°C), and treated with trypsin (1 mg/ml each of trypsin and CaCl₂ in 50 mm Tris, pH 7.5, for 3 min at 37°C). After trypsinization, the sections were washed in PBS and incubated with anti-BrdUrd (BU-33 diluted 1:1000; Sigma). The primary antibody was detected using the Vectastain Mouse ABC kit. Final immunostaining was developed with 3,3-diaminobenzidine, and sections were counterstained with aqueous hematoxylin. Three ×40 microscope fields from each section were photographed, and epithelial cells were scored for positive BrdUrd staining.

For identification of apoptotic cells, TUNEL was performed on tissue sections using the fluorescein-based In Situ Cell Death Detection kit (Boehringer Mannheim, Indianapolis, IN) according to the manufacturer’s instructions. Nuclei were counterstained with 5 μg/ml of bisbenzimide in PBS for 5 min at room temperature. Three ×40 microscope fields from each section were photographed, and cells were scored for positive TUNEL fluorescence.

For measuring MPO activity, frozen cecum was homogenized in 10 volumes of 0.2 m phosphate buffer (pH 7.4) and microcentrifuged at 14,000 rpm for 15 min at 4°C. The supernatant was discarded, and the pellet was homogenized in 1 ml of 0.05 m potassium phosphate buffer (pH 6.0) containing 0.5% (w/v) hexadecyltrimethylammonium bromide. The resuspended pellet was microcentrifuged at 14,000 rpm for 15 min at 4°C, and the supernatant was collected. MPO activity was determined using a 3,3′,5,5′-tetramethylbenzidine liquid substrate system (Sigma). The enzymatic reaction was stopped with 0.5 m H₂SO₄ after a 30-min incubation at room temperature. The yellow color was read at 450 nm. MPO activity in the samples was calculated from a standard curve generated with purified MPO. For measurement of NOS activity, cecal homogenates were assayed for the ability to convert [³H]arginine to [³H]citrulline using the NOS Detect Assay kit (Stratagene, La Jolla, CA) according to the manufacturer’s instructions. Protein concentrations of homogenates in both assays were determined using the Micro BCA Protein Assay kit (Pierce, Rockford, IL).

DNA was isolated from excised, individual polyps by digestion with 300 μg/ml of proteinase K in lysis buffer [50 mm KCl, 10m m Tris (pH 8.3), 0.45% NP40, 0.45% Tween 20, and 1 mm EDTA]. Primer sets representing eight different chromosomes (DXMit99, D3Mit22, D4Mit37, D5Mit48, D6Mit14, D11Mit4, D16Mit4, and D17Mit93) were purchased from Research Genetics (Huntsville, AL). Primers used to amplify across the (GT)₃ repeat of the Tgfbr2 gene were: forward, 5′ -AAATTCCCAGCTTCTGGCTC-3′; and reverse, 5′ -TTCTGGAATCTTCTCCTGGG-3′. All PCR amplifications were performed using one ³²P end-labeled primer and one unlabeled primer. Cycling conditions for PCR were: 95°C for 90 s, followed by 35 cycles of 57°C for 50 s, 72°C for 90 s, and 95°C for 30 s. Amplified products were separated on a urea-formamide 8% polyacrylamide gel. Gels were exposed to film at −80°C for 1–2 h.

Rag2^−/− mice (20) on a mixed strain background (∼ 97% 129S6 and ∼3% CF-1) spontaneously develop mucosal hyperplasia of the cecum and colon. This hyperplasia can develop as early as 1 week after weaning and is independent of the Tgfb1 genotype of the mouse. Hyperplasia first appears in the body and base of the cecum and eventually disseminates to the distal colon. Hyperplastic ceca and cola contain thickened folds of intestinal wall, which appear as a collection of broad, sessile polyps. As many as 32 polyps may be present within affected tissues. Hyperplastic regions are characterized by elongated crypts with reduced numbers of goblet cells. Crypt organization is maintained (Fig. 1, A–C). These areas are associated with significant granulocytic infiltration into the stroma (Fig. 1, B–C). The inflammation appears to be a response to the normal bacterial flora, a condition often observed in other immunocompromised mice such as interleukin 2, interleukin 10, and T-cell receptor α-deficient mice (21, 22, 23). The inflammation is also strain specific because it was not detected in Rag2^−/− mice on non-129 backgrounds (data not shown; Refs. 20 and 23).

Tgfb1^−/− Rag2^{− /−} mice develop adenomas and carcinomas in association with the hyperplasia, whereas Tgfb1^+/+ or Tgfb1^+/− Rag2^−/− mice rarely develop such lesions. Adenoma is characterized by profound expansion of the mucosal layer, a further reduction in the number of goblet cells, moderate loss of typical mucosal architecture, dilated cysts, and the presence of well-differentiated crypts within the submucosa (Fig. 1,D). Carcinoma is characterized by a significant loss of mucosal organization, stratification of the crypt epithelium (Fig. 1, E and F), and nuclear atypia. Multiple tumors are typically present within the cecum and colon of each Tgfb1^−/− Rag2^{− /−} mouse. No evidence of inflammation, hyperplasia, primary tumors, or metastases has been detected in any other tissue from Tgfb1^−/− Rag2^−/− mice, including the small intestine, mesenteric lymph nodes, and liver. Mice with hyperplasia or adenomas rarely show signs of illness other than mild diarrhea. Mice with carcinomas frequently display a prolapsed rectum, loss of appetite, and bloody feces. Postmortem examination generally reveals an obstructive polyp within the proximal or distal colon.

The development of adenoma and carcinoma in Tgfb1^−/− Rag2^{− /−} mice occurs at a significantly earlier age and a higher frequency than in Tgfb1^+/+or Tgfb1^+/− Rag2^−/− mice (Fig. 2). At 3 months of age, 27% (4 of 15) of Tgfb1^−/− Rag2^{− /−} mice presented with adenoma or carcinoma, whereas none of the Tgfb1^+/+ Rag2^−/− (0 of 15) mice exhibited adenoma or carcinoma (Fig. 2 A). At 6 months of age, 100% (9 of 9) of Tgfb1^−/− Rag2^{− /−} mice displayed cancer in the cecum and colon, whereas only 20% (1 of 5) of Tgfb1^+/+ Rag2^{− /−} and 43% (3 of 7) of Tgfb1^+/− Rag2^−/− presented with carcinoma. Consequently, decreasing expression of Tgfb1 increases the frequency with which hyperplasia progresses to adenoma and carcinoma. The rapid loss of normal mucosal architecture in the absence of TGF-β1 suggests a role for TGF-β1 in maintaining or restoring normal mucosal integrity after insult. This model provides a unique opportunity to investigate the specific role of TGF-β1 at this critical transition in the development of colon cancer.

TGF-β1 is a potent inhibitor of epithelial cell proliferation in vitro (24, 25). In the absence of TGF-β signaling, it is generally assumed that control of cell cycle arrest is relaxed and cell proliferation is increased. Alternatively, a reduced rate of cell death with an unchanged rate of proliferation in the absence of TGF-β signaling would result in a net increase in cell number. Either event would allow for the selection of preneoplastic cells within hyperplastic ceca and cola of Tgfb1^−/− Rag2^{− /−} mice. To test for this, cell proliferation and apoptosis measurements were performed. There was no significant TGF-β1-dependent difference in the fraction of proliferating epithelial cells among Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb1^−/− Rag2^{− /−} mice (Table 1). Although the proliferative zone is expanded in hyperplastic Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb1^−/− Rag2^{− /−} ceca relative to Rag2^+/+ ceca, the location of proliferating cells along the crypt is similar for all mice and limited to the lower two-thirds of the crypt (Fig. 3, A–C). As expected for adenoma and carcinoma, cell proliferation in tumors of Tgfb1^+/− Rag2^{− /−} (Fig. 3,D) and Tgfb1^−/− Rag2^{− /−} mice (Fig. 3,F) extended to the luminal edge of the crypt. In the 27 Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb1^−/− Rag2^{− /−} mice analyzed (19 with hyperplasia and 8 with adenoma or carcinoma), we have not detected proliferating cells at the luminal edge of either normal or hyperplastic crypts, even when those crypts are adjacent to adenomas or carcinomas. Cell death, as detected by TUNEL analysis, occurs in stromal cells and crypt epithelium at similar rates among Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb1^−/− Rag2^{− /−} mice (Fig. 3, F and G; Table 1). Epithelial cell densities are also independent of the Tgfb1 genotype of the mouse (Table 1). From these results, we conclude that the absence of TGF-β1 in hyperplastic cecum does not affect the rate of cell proliferation or cell death, thus suggesting that the event initiating dysplasia in Tgfb1^−/− Rag2^{− /−} mice occurs through a mechanism other than excessive cell proliferation.

The first sign of disease in Tgfb1 Rag2^{− /−} mice is an inflammation-associated hyperplasia. Because TGF-β1 has been shown to suppress inflammation, the rapid development of adenoma and carcinoma in the Tgfb1^−/− Rag2^{− /−} mice may be due to greater mucosal damage and disruption as a result of increased inflammatory activity. To quantitate the level of inflammation, we measured the activity of two inflammatory enzymes, MPO and NOS, in hyperplastic ceca from Tgfb1^+/+, Tgfb1^{+ /−}, or Tgfb1^−/− Rag2^{− /−} mice. MPO is produced by neutrophils and has been shown in multiple rodent models of intestinal ischemia to be directly proportional to tissue neutrophil number (26). NOS activity, which is produced constitutively at low levels, is elevated in inflammatory conditions due to increased NOS production by resident macrophage and tissue-infiltrating monocytes (27, 28). Hyperplastic ceca of Tgfb1^−/− Rag2^{− /−} mice exhibit statistically similar MPO activities, as do ceca from Tgfb1^+/+ or Tgfb1^+/− Rag2^{− /−} mice (Fig. 4,A). NOS activities from inflamed ceca of all genotypes were significantly elevated relative to noninflamed ceca; however, total NOS enzyme activities within hyperplastic ceca of Tgfb1^+/+ and Tgfb1^−/− Rag2^{− /−} mice are all similar (Fig. 4 B). Therefore, the absence of TGF-β1 does not noticeably affect the extent of inflammation associated with cecal hyperplasia in these mice.

During inflammation, the activity of destructive oxygen radical and peroxynitrite species can cause damage to DNA, leading to a greater risk of mutations (29). It is possible that in the absence of TGF-β1, cells of the inflamed cecum and colon accumulate mutations, thereby promoting more severe disease. To test this, we examined genomic DNA from hyperplastic, adenomatous, and neoplastic tissues from Tgfb1^{+ /+}, Tgfb1^+/−, or Tgfb1^−/− Rag2^{− /−} mice for evidence of microsatellite instability. DNA from polyps from 20 different mice was examined with eight dinucleotide repeat microsatellite markers representing eight different chromosomes. Fig. 5 shows representative results from screening polyp DNA with three of these markers. None of the 320 loci examined in hyperplastic, dysplastic, or neoplastic ceca exhibited microsatellite instability.

In some human colon cancers, insensitivity to TGF-β1 is coincident with increased microsatellite instability due to faulty DNA repair mechanisms (3, 4). Present in these cancers are mutations in TGFBR2 at two sites. The most commonly mutated region is a 10-base adenine mononucleotide repeat in exon 3. The second site is located in exon 7 of TGFBR2 and consists of two (GT)₃ dinucleotide repeats separated by 53 bases (3, 5). The murine Tgfbr2 gene lacks the exon 3 A₁₀ repeat found in humans. However, it does contain the homologous paired (GT)₃ repeat in exon 7. PCR primers designed to span this repeat were used to detect possible mutations within the (GT)₃ sequences. No alterations in the (GT)₃ repeats were observed (Fig. 5, D and E). Consequently, the absence of TGF-β1 does not lead to microsatellite instability or to Tgfbr2 mutations similar to those found in human colon cancer.

Mutations in the APC gene have been found in most (> 95%) human sporadic and inherited colon cancers (30, 31). The majority of these mutations result in a truncated mRNA and immunohistochemically undetectable protein (32, 33). Similarly, tumors in the Apc^Min mouse also exhibit loss of immunohistochemically detectable APC (34, 35, 36). Loss of functional APC within intestinal epithelium is thought to lead to dysregulation of cell differentiation and proliferation through a mechanism involving β -catenin (37). β-Catenin influences cell behavior directly by binding to E-cadherin at the cell surface (38) and indirectly by binding members of the TCF/LEF family of transcription factors prior to translocating to the nucleus (39). In the absence of functional APC, β-catenin is lost from its normal location at the cell membrane and is distributed throughout the cytoplasm and nucleus (40, 41).

We examined the possibility that APC levels might be reduced or absent in the epithelium of these tissues or that β-catenin may be inappropriately distributed in epithelial cells. Immunostaining with an antibody directed against the COOH terminus of APC detects protein in epithelial cells at the luminal edge of hyperplastic crypts in Tgfb1^−/− Rag2^{− /−} (Fig. 6,A) and throughout epithelial cells of neoplastic cecal tumors from Tgfb1^−/− Rag2^{− /−} mice (Fig. 6,B). In hyperplastic ceca from Tgfb1^−/− Rag2^−/− mice, positive immunostaining for β-catenin is located in the peripheral cytoplasm and at intercellular junctions but not in the nucleus of mucosal epithelial cells (Fig. 6,C). In epithelial cells of the neoplastic tumors, cytoplasmic staining is slightly more diffuse, whereas nuclear staining remains absent (Fig. 6 D). Therefore, a deficiency in TGF-β1 does not result in a redistribution of β -catenin to the nucleus as occurs in the absence of APC, suggesting that the tumor-suppressive effects of TGF-β1 do not act through the antiproliferative mechanisms attributed to APC.

Adult Tgfb1^−/− Rag2^−/− mice bred onto a 129S6 background develop cancerous tumors in the cecum and colon at an early age. Here we demonstrate that TGF-β1 suppresses early events in colon cancer development. Although all three TGF-βs signal through a common receptor element (42), TGF-β2 and TGF-β3 are not involved in early events because they continue to be expressed in the cecum and colon of Tgfb1^−/− Rag2^{− /−} mice (data not shown). Observations from both human colon cancers and cell culture have suggested that resistance to TGF-β is associated with late events in tumorigenesis. However, this conclusion is based on results from mutations that inactivate the receptor or downstream signaling molecules that eliminate signaling from all three TGF-βs. Because SMAD3-deficient mice develop metastatic colon cancer (10) and Tgfb1^−/− Rag2^{− /−} mice develop only nonmetastatic colon cancer, it is reasonable to hypothesize that TGF-β2 and/or TGF-β3 inhibits metastasis at late stages of tumorigenesis. To demonstrate a role for TGF-β2 and/or TGFβ3 in colon tumor metastasis in the mouse, the perinatal lethality of the Tgfb2 and Tgfb3 knockout mice will first need to be circumvented (43, 44, 45).

Systematic evaluation of tumorigenesis in these mice indicates that the tumor-suppressor activity of TGF-β1 is not directed at cell proliferation, suppression of inflammation, or maintenance of genetic stability. Comparison of hyperplastic mucosa among Tgfb1^+/+, Tgfb1^{+ /−}, and Tgfb1^−/− Rag2^{− /−} mice reveals similar increases in cell turnover rates and cell densities, as well as a similar distribution of proliferative cells along the crypt. This is in contrast to previous in vitro studies in which many epithelial cells are strongly growth inhibited by the addition of TGF-β1 but it is consistent with the phenotype of the Tgf-b1 knockout mouse, which lacks any generalized epithelia hyperplasia in the absence of inflammation. The extent of inflammation in the ceca and cola of Tgfb1^−/− Rag2^{− /−} mice, as measured by levels of NOS and MPO, is not greater than in Tgfb1^+/+ Rag2^−/− mice. This implies that TGF-β1 is not participating in suppressing granulocytic inflammation in the mouse intestine, despite ample in vitro and in vivo evidence showing that TGF-β1 is a potent regulator of immune and inflammatory cells. Additionally, microsatellite instability is not detected in hyperplastic or tumor tissue from the cecum and colon of Tgfb1^−/− Rag2^{− /−} mice, thus arguing against a direct genome protective role for TGF-β1.

The single difference detected in cecum and colon tumorigenesis in the absence of TGF-β1 is the rapid loss of crypt and mucosal architecture in the presence of hyperplasia. Loss of crypt architecture is thought to be one of the earliest and most significant events in tumorigenesis (46). Normal crypt division, which is infrequent in the adult cecum and colon, begins with epithelial budding at the base of a single crypt and proceeds by growth of an epithelial septum to yield two crypts (47). Crypt division initiated by hyperplasia increases the number of crypts, which can distort the normal mucosal organization. In addition, the absolute number of proliferating cells in the cecum and colon is increased, despite the normal distribution and proportion of proliferating cells. This increases the population of cells at risk for acquiring changes that circumvent proliferation and differentiation controls. Further loss of mucosal organization can occur when crypt epithelial cells invaginate or evaginate in response to the cellular overcrowding of hyperplasia. Cells within these crypt branches are no longer subject to the same directional migration and extrusion as cells from the original crypt. Because most hyperplasia does not progress to adenoma, this implies that mechanisms must be in place to control the nature of crypt growth in response to insult and to maintain mucosal organization. Because the Tgfb1^−/− Rag2^{− /−} mice develop adenomas rapidly from hyperplastic tissue, these results indicate that TGF-β1 is involved in maintaining mucosal organization after epithelial cell proliferation has been initiated. In support of this conclusion, Thorup (48) showed that chemically induced aberrant crypt foci in rat colon, which are thought to be the earliest detectable preneoplastic lesions, express reduced amounts of TGF-β1, and Mikailowski et al. (49) showed that controlled release of TGF-β1 into the peritoneum reduced the formation of these dysplastic foci.

One possible target of the TGF-β1 tumor suppressor activity involves differentiation. TGF-β1 is expressed as a gradient along the normal colon crypt. The differentiated epithelium of the luminal tips are associated with the highest levels of TGF-β1, whereas the less differentiated, proliferating stem cells of the crypt base are associated with the lowest levels of TGF-β1 (50, 51). In part because of this observation, a role for TGF-β1 in maintaining crypt epithelial cells in a differentiated state has been proposed. However, unaffected crypts within hyperplastic ceca and cola in Tgfb1^−/− Rag2^{− /−} mice possess a normal epithelial cell profile consisting of well-differentiated columnar epithelium and mucin-producing goblet cells. Thus, any effect of TGF-β1 on epithelial cell differentiation is not directed at establishing terminal differentiation.

Similar to Tgfb1^−/− Rag2^−/− mice, the earliest sign of disease described in the APC mouse model of intestinal cancer is mucosal dysplasia (35, 52). Given the frequent mutations found in APC in human colon cancers and the large body of data implicating APC in regulation of epithelial interactions, we investigated the possibility that the absence of TGF-β1 may affect epithelial cell function through APC. TGF-β1 does not suppress intestinal cancer via regulation of APC levels because full-length APC was detected in epithelial cells of tumors in the Tgfb1^−/− Rag2^{− /−} mouse at levels comparable with controls. Although it cannot be ruled out that Apc has acquired point mutations in the tumors from Tgfb1^−/− Rag2^{− /−} mice, it seems unlikely because the majority of APC mutations in both humans and mice result in truncation of the protein product or loss of the wild-type allele. Additionally, we do not detect microsatellite instability in tumors from our mice, which suggests that the absence of TGF-β1 does not invoke a mechanism that allows the rapid accumulation of gene mutations. The phenotype of the Apc^Δ716/Smad4 compound heterozygous mouse also suggests that TGF-β1 does not function directly through APC (9). If the two molecules acted in a hierarchical fashion, the compound heterozygotes would be expected to have the same phenotype as the Apc heterozygotes. The presence of larger, invasive tumors in the compound heterozygotes suggests that APC and TGF-β1 act synergistically to protect against intestinal cancer. The detection of full-length APC in the intestinal tumors of SMAD3-deficient mice is consistent with these observations (10).

In conclusion, we present a mouse model of large intestinal cancer in which carcinoma develops from an inflammation-associated hyperplasia. We have demonstrated that TGF-β1 suppresses intestinal cancer by preventing the early transition from an organized hyperplasia to dysplasia rather than by inhibiting epithelial cell proliferation, granulocyte-mediated inflammation, or genetic instability.

We thank Colleen York and Christy Sloan for excellent mouse husbandry and W. Y. Sun for assistance with genotyping. We thank Dr. Robert Coffman for providing a starting colony of 129S6 × CF-1 Tgfb1^+/− Rag2^{− /−} mice and Dr. Kathy Heppner-Goss at the University of Cincinnati for providing paraffin sections of intestinal tumors from Apc^Min mice and valuable suggestions. We thank Drs. Ann Kier and Mark McArthur at Texas A&M University and Drs. Cindy Besch-Williford and Craig Franklin at the University of Missouri for histopathological consultation.