Inactivation of the retinoblastoma gene yields a mouse model of malignant colorectal cancer.

The retinoblastoma gene (Rb) is mutated at significant frequency in various human epithelial tumors, including colorectal cancer, and is strongly associated with metastatic disease. However, sole inactivation of Rb in the mouse has so far failed to yield epithelial cancers. Here, we specifically inactivate Rb and/or p53 in the urogenital epithelium and the intestine. We find that the loss of both tumor suppressors is unable to yield tumors in the transitional epithelium lining the bladder, kidneys and ureters. Instead, these mice develop highly metastatic tumors of neuroendocrine, not epithelial, origin within the urogenital tract to give prostate cancer in the males and vaginal tumors in the females. Additionally, we discovered that the sole inactivation of Rb in the intestine was sufficient to induce formation of metastatic colorectal adenocarcinomas. These tumors closely mirror the human disease in regard to the age of onset, histological appearance, invasiveness and metastatic potential. Like most human colorectal carcinomas, our murine Rb-deficient tumors demonstrate genomic instability and they show activation of β-catenin. Deregulation of the Wnt/β-catenin pathway is specific to the intestinal tumors, as genomic instability but not activation of β-catenin was observed in the neuroendocrine tumors. To date, attempts to generate genetically engineered mouse models of colorectal cancer tumors have yielded mostly cancer of the small intestine, which rarely occurs in humans. Our system provides the opportunity to accurately model and study colorectal cancer in the mouse via a single gene mutation.

INTRODUCTION

Inactivation of the retinoblastoma gene occurs in a variety of human cancers including retinoblastoma, small cell lung carcinomas, osteosarcomas, bladder cancer, prostate cancer and a subset of colorectal cancers[1]. However, while mouse models have established a causal correlation between Rb mutation and multiple cancers, the role of this tumor suppressor in epithelial cancer development is unclear. In the mouse, germline Rb loss causes embryonic lethality, while inactivation of one Rb allele leads exclusively to neuroendocrine tumors of the pituitary and thyroid[2]. Similarly, chimeras bearing Rb null cells develop only neuroendocrine tumors[3, 4]. This narrow tumor spectrum suggests that transformation of other cell types requires additional mutations, or that the short lifespan of the neuroendocrine tumor-bearing Rb mutant mice (one year maximum) precludes the possibility of slower developing tumors.

The tumor suppressive activity of the retinoblastoma protein, pRB, is at least partially dependent upon its ability to inhibit the E2F transcription factors and thereby prevent cell cycle entry[5]. By generating Rb−/−;E2f4−/− chimeric animals, we have previously shown that E2f4 loss inhibits early onset pituitary tumors, and increases the lifespan of Rb mutant animals[6]. Moreover, these older animals developed additional tumor types dependent on Rb loss. Specifically, 15% of Rb−/−;E2f4−/− chimeras sixteen months or older had cancer of the urothelium, which is the transitional epithelium lining the bladder, ureters and kidney calyces. Notably, we never observed urothelial tumors in Rb−;E2f4−/− or E2f4−/− germline mice[7, 8]. Thus, we hypothesized that E2f4 loss is not a direct driver of urothelial tumors but acts instead to extend the lifespan of Rb−/− chimeras and enable formation of slower developing Rb mutant bladder tumors. Importantly, over 50% of human urothelial cancers display Rb mutations[9], and our analysis of Rb−/−;E2f4−/− chimeras provided the first evidence that Rb loss is a driver of this tumor type.

Urothelial (or bladder) cancer does not spontaneously occur in mice and is still very understudied due to the difficulties in creating valid models that do not involve surgery[10, 11]. Given the complexity of the Rb−/−;E2f4−/− chimeras, we wished to generate a more tractable model of Rb mutant urothelial cancer. Previously, He et al. used Uroplakin-Cre to delete Rb and also p53 (mutated in more than 50% of human urothelial cancers), and found no bladder cancer[12]. Since Uroplakin-Cre is expressed at low penetrance in the bladder, we reasoned that a stronger Cre might boost the frequency of Rb and p53 mutation and thus the possibility of transformation. Thus, in this study, we crossed Rb and p53 conditional mutant mice with the Fabpl-Cre transgene. This is expressed very efficiently in urothelial cells, and is also active in the prostate and the intestinal epithelium[13]. Notably, human data indicate that a small but significant fraction of colorectal cancers carry Rb mutation (0.5–4%) but the significance of this is unclear[14, 15]. Our analysis of this new mouse model shows that inactivation of Rb and p53 is insufficient to initiate tumorigenesis in the urogenital epithelium even when nearly every cell is mutated. Instead, double mutant animals develop highly metastatic neuroendocrine tumors in the prostate and the female reproductive tract. Finally, we discovered that Rb mutation is sufficient to initiate intestinal tumors that closely resemble human colorectal cancer. This reveals a fundamental role for this tumor suppressor in the intestine, and provides a novel model system to study colon cancer.

RESULTS

Combined loss of Rb and p53 leads to urogenital neuroendocrine metastatic cancer

To assess the contribution of Rb to the tumorigenesis of multiple epithelia, we used the Fabpl-Cre transgenic in which Cre recombinase is driven by the liver fatty acid binding protein promoter. Previous reports indicate that this transgene is expressed in the urothelium, intestine and anterior prostate[13, 16]. In agreement, by crossing Fabpl-Cre to mice carrying a conditional LacZ reporter allele, we observed transgene expression at nearly 100% penetrance in the transitional epithelium cells lining the kidneys and the bladder (supplementary Fig. 1). Additionally, we observed LacZ expression in most of the colon (large intestine), in limited areas of the small intestine (mostly the distal part), and in the dorsal prostate (supplementary Fig. 1). We crossed Fabpl-Cre with Rb and p53 conditional mice (respectively Rb and p53) to generate cohorts of animals lacking Rb, p53 or both genes, which were aged alongside appropriate controls (Rb, and Rb without Fabpl-Cre). Between six and eight months, 7/8 FabplCre males had to be euthanized (Fig. 1a, grey lines). Gross and microscopic analysis revealed that these animals all displayed highly aggressive prostate tumors that had colonized lungs, liver and kidneys (Fig. 1b). Rb mutations exist in over 30% of human prostate cancers, and they associate strongly with androgen therapy resistance[17]. Human prostate tumors are typically epithelial, but often contain a neuroendocrine component that correlates with androgen therapy resistance, and a subfraction of human tumors are exclusively neuroendocrine[18]. Probasin-Cre induced deletion of Rb and p53 in the mouse prostate was found to yield malignant tumors co-expressing neuroendocrine and epithelial markers[19]. In contrast, the histological appearance (Fig. 1b) and staining for neuroendocrine (chromogranin A) and prostate epithelial (keratin 17/19) cell markers (Fig. 1c), showed that all but one of the tumors in our model were neuroendocrine. Likewise, morphological and immunochemical analyses detected exclusively neuroendocrine cells in the metastases (not shown). Since the presence of neuroendocrine cells in human prostate tumors associates with androgen therapy resistance, we examined the levels of androgen receptor (AR) in our Rb;p53 deficient tumors and metastases. We found that 3/8 expressed physiological levels of this receptor and 5/8 tumors showed low or no AR expression (Fig. 1d). Furthermore, all the metastases, even ones originating from AR positive tumors, had downregulated AR expression (Fig. 1d). Therefore, these tumors model neuroendocrine prostate tumors, the subset of human cancers that is both metastatic and therapy resistant.

Figure 1

Rb and p53 deficiency causes metastatic neuroendocrine tumors in the urogenital tract

(a) Kaplan-Meyer curves show reduced survival of Rb−/−;p53−/− double mutant animals, with males shown in grey (N=8), females in black (N=8), and p<0.0001 for both genders. (b) Histological analysis reveals the presence of multiple neuroendocrine-like tumors in the prostate (top) and their metastases in the lungs and livers (bottom). (c) Immunohistochemistry with neuroendocrine and epithelial-specific antibodies shows that the prostatic primary tumors are immunoreactive only to neuroendocrine (chromogranin A), not epithelial markers (CK17/19). (d) Immunological staining for androgen receptor of prostate tumors (left) indicates that most of them have under-physiological levels of the protein. The right panels show that the expression of the receptor dramatically decreases in the metastases compared to the primary tumors. (e) H&E staining of the vaginal tumors (top panels) and their metastases (bottom panels) shows that there are neuroendocrine.

Neuroendocrine tumors also appeared to be the primary cause of death of FabplCre females, between eight and fourteen months (Fig. 1a, black lines). Histological and immunochemical analysis showed these tumors originated in the vagina and spread to lung, liver and kidneys (Fig. 1e). There are no previous reports of Cre expression from this transgene in neuroendocrine vaginal cells. Thus, there must be rare and scattered neuroendocrine cells in the female genital tract, underscoring the extreme sensitivity of this cell type to transformation upon concomitant Rb and p53 loss.

Rb and p53 deficiency is insufficient for urothelial carcinogenesis

Histological examination of the FabplCre mutant males showed complete absence of tumors in the transitional epithelium that lines bladder, ureters and kidneys. This finding was unsurprising since the vast majority of these animals succumbed before eight months, and we have previously shown that Rb−/−;E2f4−/− chimeras do not display detectable tumors until sixteen months[6]. However, given the high penetrance of the FabplCre-driven recombination, we were disappointed by the complete absence of either malignant or benign bladder tumors in the mutant females, since these lived up to fourteen months. Instead, we found that the males and females both developed atypia of the umbrella cells along the entire urothelium with 94% penetrance (Fig. 2a). These cells had much bigger nuclei and higher DNA content than their corresponding controls. This phenotype had been previously described in the Uroplakin-Cre transgene model although, consistent with the lower expression of Cre in the urothelium, the authors detected umbrella cell atypia in only 20% of the Rb−/−p53−/− mutant animals[12]. Histological analysis of bladders harvested from p53 and Rb single deficient animals showed that this neoplasm arose from mutation of p53 alone (11 out of 13 animals), but not Rb alone (0 out of 26 animals), although combined mutations of Rb and p53 clearly increased its severity (Fig. 2a). The physiological implication of the umbrella cells atypia remains unclear as p53 single mutants lived 22 months on average and showed no signs of urogenital disease (Fig. 2b). These results suggest that in our previous Rb−/−;E2f4−/−chimeric model, loss of E2F4 was playing a contributing role in the development of Rb-deficient urothelial tumors.

Figure 2

Urothelial cells are not susceptible to tumorigenesis upon inactivation of Rb and/or p53

(a) H&E staining of bladders isolated from double Rb−/−;p53−/− mutant as well as Rb−/− and p53−/− single mutant reveals absence of tumors. Instead, anomalies in the umbrella cells, with macronuclei containing higher levels of chromatin (indicated by the arrows), are seen in Rb−/−;p53−/− double and p53−/− single mutant animals. (b) The absence of differences in the lifespan among p53−/− mutants and their controls shows that umbrella cell atypia does not overall affect survival (p=0.91).

Inactivation of Rb causes development of intestinal adenocarcinomas

Despite the lack of urothelial tumors, FabplCre mice died on average six months earlier than their corresponding controls (Fig. 3a). The primary cause of death was masses that occluded the intestinal tract in 11 out of 26 mice. These malignancies were found typically in twelve to eighteen months old mice, independent of gender, and were located in the small intestine (10/26; 38% of animals) and colon (6/26; 23% of animals). With the exception of a few relatively benign adenomas, these tumors were aggressive adenocarcinomas (Table 1). Indeed, most of the FabplCre;Rb mice bore multiple tumors with pathological stages varying from well-circumscribed polyps (stage II) to moderately or poorly differentiated tumors invading through the submucosa and the muscle wall (stage III and IV, Fig. 3b,c, Table 1). Importantly, we detected tumors that had traveled through mesenteric lymphatics and blood vessels and entered the hepatic portal system to colonize the liver, indicating that the mutant cells were metastastic (Fig. 3d). Prior studies, where Rb was deleted in the intestine using VillinCre and FabplCre transgenes, reported that Rb mutation alone is insufficient to cause cancer in animals younger than one year (Ref [20, 21] and Haigis, personal communication). We only detect intestinal tumors after this time point, explaining this apparent discrepancy. Interestingly, this time of onset essentially mirrors that at which humans sporadically develop the disease, around or after middle age. Most importantly, cancer of the colon has been very difficult to model genetically in mice. With few exceptions, existing models develop tumors in the small intestine, not the large intestine. Furthermore, those lesions are usually benign and unable to metastasize[22]. Our studies show for the first time that Rb ablation alone can cause aggressive, metastatic tumors in the intestine that closely mirror the human disease thus providing a model for colorectal cancer.

Figure 3

Rb-deficiency leads to invasive, poorly differentiated intestinal adenocarcinomas

(a) Rb mutant mice have a significantly reduced lifespan relative to control animals (p=0.0047). (b) H&E staining of different grades of small intestinal tumors in Rb mutant mice. The top panels show a low grade tumor while the bottom panels represent a poorly differentiated high grade (stage IV) and invasive tumor that traveled through the muscle wall to invade the mesentery. (c) The tumors found in the colon are highly invasive (top panels), and undifferentiated (bottom panels, note in both cases cells breaking through the muscle). (d) Rb-deficient intestinal tumors have metastatic potential. The top panels show a mesenteric vessel containing intestinal tumor cells; the bottom panel shows the portal vein in the liver clogged by intestinal cells, indicating that these cells, like their human counterpart can metastasize to the liver.

Table 1

Phenotypes of Fabpl-Cretg/+Rbc/c tumor-bearing mice.

AGE (months)	TUMORS IN THE SMALL INTESTINE	TUMORS IN THE LARGE INTESTINE	TUMORS IN OTHER ORGANS
MALES
10	none	4 Tumors. PT I
13	None	2 Tumors. 1 PT I, 1 PT II, invasive carcinoma
16	4 Tumors. 1 PTI, 1 PT II, 1 PT III, 1 PT IV blood and lymphovasculature invasion	1 PT III moderately differentiated non mucinus, invasion into mesenteric small vessel	intestinal metastases in the portal vein
17	9 Tumors. PT II	6 Tumors. 3 PTI, 2 PT II, 1 PT III, poorly differentiated, potentially invasive
17	2 Tumors. PT III and PT IV, lymphovasculature invasion	benign
17	1 PT III, carcinoma, vasculature invasion	none
22	none	none	liver tumor
23	none	none	lungs
FEMALES
8	2 Tumors. PT 1	none
12	5 Tumors. 4 PT II and 1 PT III, carcinoma with lymphovasculature invasion	none
13	none	none	thyroid
16	2 Tumors. PT II	4 Tumors. 3 PT II, 1 PT III-IV poorly diff. with neuroendocrine features
17	4 Tumors. PT I	none
18	6 Tumors. PT IV, poorly diff (neuroend and epithelial), metastatic in lymph nodes and mesentery	benign
18	none	none	pituitary
22	none	none	thyroid
23	7 Tumors. 5 PTI, 2 PTII	1 PT III

PT= pathological stage. PT II: Tumor invades muscularis propria. PT III: Tumor invades through the muscularis propria into pericolorectal tissues. PT IV: Tumor penetrates to the surface of the visceral peritoneum and/or directly invades or is adherent to other organs or structures.

p53 inactivation is not a prerequisite for the development of Rb-deficient intestinal tumors

The late onset of intestinal tumors in Rb-deficient mice suggests that, similar to human colon cancers, multiple mutations are required for tumor development. As p53 is mutated in about 50% of human colon cancers, we analyzed Rb−/−;p53−/− double and p53−/− single mutants for evidence of intestinal tumors. In agreement with a previous study[23], we observed no intestinal tumors in the p53−/− single mutants, even though they lived 22 months on average, indicating that p53 loss is insufficient to initiate intestinal tumorigenesis. We did observe a single tumor in three of the sixteen Rb−/−;p53−/− animals: a high-grade adenocarcinoma of the small intestine (PT IV), a low-grade adenocarcinoma of the large intestine (PT II), and a benign polyp (data not shown). This low incidence suggested that p53 inactivation does not obviously accelerate intestinal tumors arising from Rb loss. To test this hypothesis, we screened the Rb single mutant intestinal tumors for the presence of p53. Remarkably, every one (n=53) retained p53, and those examined showed concomitant expression of its downstream target p21Cip1 (Fig. 4). These findings strongly suggest that the Rb mutant tumors express functional p53 and therefore lack any selective pressure to inactivate this tumor suppressor. Altogether, our data argue that inactivation of Rb, and not p53, is promoting intestinal tumors in our model.

Figure 4

Rb-deficient intestinal tumors show activation of p53

(a) Analysis of p53 expression by immunostaining shows high levels of the protein in Rb-deficient small and large intestinal tumors. (b) Concomitant upregulation in the tumors of p21Cip1, a p53 direct target, shows that p53 is present in its active form. This indicates that p53 inactivating mutations are unlikely to play a role in the development of Rb−/−;p53−/− or Rb−/− intestinal tumors.

Genomic instability in Rb-deficient tumors

Human colorectal cancers are very heterogeneous and they accumulate mutations mostly through two types of genomic instability: microsatellite instability (MIN) or chromosomal instability (CIN). Microsatellite instability leads to amplification of short repetitive DNA sequences caused by loss of function mutations in mismatch repair (MMR) genes. Inactivation of MMR genes can occur sporadically in 15% of colorectal cancers resulting from somatic mutations of the MLH1 gene or germline mutations of the genes MSH2, MSH6 or MLH1 genes[24, 25]. In contrast, chromosomal instability occurs in the majority of human colon cancers, and involves whole chromosome gain or loss that is typically associated with large chromosomal rearrangements[24, 25]. Notably, Rb-deficiency has been shown to cause chromosomal instability by altering chromosome condensation and cohesion[26-28]. Thus, we hypothesized that Rb loss might contribute to the development of intestinal tumors in our FabplCre;Rb mice by promoting CIN and thus genomic instability. Since MIN and CIN pathways both lead to double stranded DNA breaks that can be identified via phosphorylated H2AX (γ-H2AX), an early marker of DNA damage, we performed immunochemical analysis of γ-H2AX on our tumor sections. This revealed DNA breaks in the tumorigenic intestinal epithelium, which supports the presence of genomic instability and explains p53 activation in the tumors (Fig. 4, left panels). To address CIN versus MIN, we examined the expression of MLH1 (that is in sporadic MIN) by immunohistochemistry. We readily detected MLH1 protein in the tumors as well as in Rb-deficient normal epithelium indicating that the mutant cells are unlikely to undergo genomic instability through the mechanism of microsatellite instability (Fig. 4, middle panels). On the contrary, the presence of high levels of MLH1 protein, together with upregulation of another component of the MMR system, MSH2, in our Rb-deficient intestinal tumors, indicates that the function of mismatch repair genes is activated rather than lost (Fig. 4, right panels). This is consistent with reports of other human tumor types, where the mismatch repair genes are upregulated upon DNA damage caused by genomic rearrangements. Therefore, altogether our results strongly suggest that Rb-deficient intestinal tumors undergo genomic rearrangements through chromosomal instability.

Activation of the Wnt pathway in Rb-deficient intestinal tumors

A nearly universal feature of human colorectal cancers is deregulation of the Wnt signaling pathway with consequent activation of β-catenin. This event typically occurs through direct mutation of the β-catenin gene or through truncating mutations of APC. These mutations allow β-catenin to translocate to the nucleus and cooperate with the TCF/LEF transcriptional factors to activate genes responsible for cell transformation[29]. To assess whether Rb loss impinges on the Wnt pathway, we analyzed the localization of β-catenin in the intestinal tumors by immunohistochemistry. We found that indeed Rb inactivation influences β-catenin status. As expected, β-catenin was localized to the cell membrane in the non-tumorigenic Rb-deficient intestinal epithelium, but it showed a highly heterogeneous pattern in the tumors. Analysis of both small and large intestinal tumors (n=35) identified regions covering 5–80% of the cross-sectional area where areas of membranous β-catenin co-exist with areas of nuclear and/or cytoplasmic accumulation (Fig. 6a). This pattern has been described in human colorectal cancer where the presence of diffuse nuclear+cytoplasmic or nuclear β-catenin is associated with invasive or potentially invasive regions of the tumor and is predictive of poor prognosis[30]. In agreement, immunochemical analysis of cells colonizing mesenteric lymph nodes shows diffuse nuclear+cytoplasmic β-catenin staining (Fig. 6a). Notably, all the tumors showed upregulation of Sox9, a known β-catenin-TCF target[31], confirming that the Wnt pathway is activated in Rb-deficient intestinal tumors (Fig. 6b). To determine if activation of the Wnt pathway resulted from loss of the β-catenin upstream regulator APC, we examined the levels of APC in the intestinal tumors. Immunochemical analysis revealed that the presence or absence of APC was a poor predictor of β-catenin status in Rb-deficient tumors (Fig. 6c). Specifically, in a subset of tumors (n=19/35) absence of the APC protein coincided with activated β-catenin, but other tumors and metastases (n=16/35) retained, or partially retained, APC expression despite the presence of nuclear/cytoplasmic β-catenin (Fig. 6c). Moreover, irrespective of the levels of APC expression, we found that nuclear β-catenin consistently co-localized with areas of DNA damage and DNA repair (Fig. 6c). This indicates the Wnt pathway is typically activated within the Rb-deficient intestinal tumors, but the underlying mutations vary, presumably as a result of the ensuing genomic instability. KRas is also frequently activated in early stages human colorectal cancer[25]. To determine whether this occurs in our Rb deficient intestinal tumors, we screened for the presence of phosphorylated MEK1 (P-MEK1 T286), and showed that this marker of kRas signaling was upregulated in 27/40 tumors analyzed (Fig. 6d). Taken together, these data show that our Rb deficient intestinal tumors have upregulated signaling pathways, including Wnt/β-catenin and kRas, similar to human colorectal tumors.

Figure 6

Nuclear accumulation of β-catenin in intestinal tumors indicates deregulation in the Wnt pathway

(a) IHC for β-catenin shows diffuse cytoplasmic staining as well as areas of nuclear accumulation in small and large intestinal tumors that generally associate with the invasive front of the tumors. The bottom panels show diffuse cytoplasmic+nuclear staining in a mesenteric metastasis. (b) The areas of β-catenin accumulation are coincident with Sox9 upregulation indicating that the Wnt pathway is activated in the intestinal tumors. (c) Analysis of Rb-deficient intestinal tumors shows variable APC expression. The top two panels represent small intestinal tumors, one with uniform APC expression and one with no expression despite the presence of nuclear β-catenin. The mid two panels show colon tumors expressing nuclear β-catenin, one with patchy localization of APC and the other lacking expression of the protein. The first, fourth and last panels are from the same tumors shown in Figure 4, allowing comparison of APC and β-catenin staining respectively. Note that the expression of APC is heterogenous, while the pattern of nuclear β-catenin mirrors the ones of DNA damage and DNA repair. (d) Immunostaining for phosphorylated MEK1 (P-MEK1 T286) shows expression of the protein in a proportion of Rb-deficient intestinal tumors (right panels) indicating activation of kRas signaling pathway.

Nuclear accumulation of β-catenin is specific to Rb-deficient intestinal, not neuroendocrine tumors

We wanted to establish whether deregulation of β-catenin is strictly associated with the onset of intestinal tumors or represents a more general phenomenon related to Rb inactivation. Thus, we analyzed the Rb−/−p53−/− mutant neuroendocrine tumors (n=8) and found that they all lacked detectable nuclear β-catenin, and generally retained both membranous β-catenin and detectable APC protein (Fig. 7a, b). This indicates that the Wnt/β-catenin pathway remains intact in the neuroendocrine tumors, while being abrogated in the intestinal tumors. By extension of this logic, it appears that distinct oncogenic signaling events are driving the Rb-intestinal versus the Rb;p53-deficient neuroendocrine tumors. Importantly, our analyses showed that the neuroendocrine tumors display high levels of MLH1 and MSH2 expression in a comparable manner to the intestinal tumors (Fig. 7b), arguing that the MMR pathway is activated in both settings. Accordingly, we found that the neuroendocrine tumors also contained areas positive for γ-H2AX, revealing the presence of DNA damage (Fig. 7b). These results strongly suggest that genomic instability represents a common event in the Rb-deficient tumors, but loss of Rb in neuroendocrine versus intestinal epithelial cells must activate distinct molecular pathways to induce tumorigenesis.

Figure 7

Neuroendocrine tumors undergo genomic instability without deregulating the Wnt pathway

(a) β-catenin staining of neuroendocrine tumors shows membranous localization of the protein. (b) IHC for APC indicates that this protein is not lost in the neuroendocrine tumors. However, the presence of DNA damage together with upregulation of MLH1 and MSH2 indicate that these tumors undergo genomic instability.

DISCUSSION

This study provides key insights into the consequences of Rb loss into a subset of epithelial tissues. We show that ablation of Rb, with or without p53 mutation, is unable to initiate cancer of the murine urothelium. Conversely, simultaneous inactivation of these proteins causes rapid onset of metastatic prostate and vaginal cancer. Most importantly, our model system yields the first evidence that Rb inactivation is sufficient to promote intestinal tumorigenesis. Below, we discuss the significance of each of these findings.

We found that despite nearly complete inactivation of Rb and/or p53 in urothelial cells, no tumors develop in aged animals indicating that these mutations are insufficient for tumor initiation. Given that about 50% of human tumors have mutation in Rb and p53, our results and the study from He et al[12] imply that inactivation of Rb and p53 contributes to the progression, but not the initiation, of bladder cancer, or that additional mutations are required for tumor onset. In support of the requirement for additional mutations is the presence of tumors in chimeric animals lacking both Rb and E2f4 (Ref. 6) and a study reporting that large T antigen, which targets p53 and the entire Rb family, causes development of invasive tumors in the mouse bladder[32].

Human neuroendocrine prostate tumors occur at low frequency[18]. Generally these cancers are epithelial but as they progress they show a substantial fraction of neuroendocrine cells. These cells express low levels of androgen receptor and they associate with acquisition of resistance to hormone therapy[33]. The formation of aggressive, AR low neuroendocrine tumors in the Rb−/−p53−/− mutants, clearly establishes that the combined loss of Rb and p53 differs in its ability to transform urothelial versus neurendocrine cell types. Moreover, as Rb mutations are generally associated with androgen therapy resistance[34], it also suggests that our model may offer a useful tool to study prostate neuroendocrine tumors. We note that our system differs from the TRAMP model, which typically yields epithelial prostate cancers that give rise to AR therapy-resistant metastases containing neuroendocrine cells[35]. We believe that our model could be used together with TRAMP mice and other models of epithelial prostate cancer to dissect the neuroendocrine component present in many advanced stage human cancers[19].

Our discovery that a significant number of Rb mutant mice develop high-grade intestinal tumors residing in the colon, provides novel insights into the role of Rb in colon cancer. Importantly, these Rb-deficient intestinal tumors have similar features to human colorectal cancer; the masses range from well to poorly differentiated, can be highly invasive and even metastasize to the liver, the first target organ in human colorectal cancer. Furthermore, the Rb-deficient colon tumors, similar to human colorectal cancer, have a mosaic distribution of β-catenin with cytoplasmic/nuclear staining interspersed with membranous localization. Finally, these tumors bear marks of genomic instability and upregulate the MMR system suggesting that they result from CIN, not MIN, mechanisms.

Notably, unlike humans, most genetically defined mouse models of intestinal tumorigenesis develop lesions in the small intestine rather than the colon[36]. The few existing systems of colon carcinogenesis develop tumors incapable of metastasizing, and thus cannot reproduce the primary cause of mortality in human colorectal cancer[37]. Only recently, combinations of surgical procedures in kRas;apc double mutant mice have yielded model systems in which tumors develop in the colon and can be metastatic[38]. However, these models require technical expertise, specific instrumentation and aseptic settings. Our model involves a single gene mutation and recapitulates the key features of human colon cancer.

Mutations in the Rb locus have been found in a significant fraction of colorectal carcinomas[14, 15]. Therefore, the function of Rb in the mouse intestine has already been investigated in different systems but due to the nature of the system utilized and/or the age ranges employed, its role in colorectal carcinoma had not been revealed previously. The only prior aging study was performed in Villin-CreRb−/− mice, which die by one year of age of pituitary and thyroid tumors and therefore preclude evaluation of Rb’s role in intestinal tumorigenesis[21]. Notably, studies in embryos and younger animals hinted at tumor enabling events from Rb loss in the intestine. In embryos, Rb ablation using a CollagenIA-Cre transgene causes an increase in number of enterocytes, paneth, goblet and endocrine cells which indicates a general effect on proliferation[39]. Using the Fabpl-Cre transgene, Haigis et al. showed that this ectopic proliferation becomes restricted to enterocytes in young adults[20]. Moreover, they found that simultaneous deletion of Rb and either one of the pRB-related proteins, p107 or p130, disrupted the intestinal architecture although no tumors were detected. While we cannot explain the lack of tumors in the Haigis study versus the presence in our own, we speculate that this may result from changes in the representation of the cells and/or their relative niches that are the targets of transformation in the context of normal tissue architecture.

In the present study we definitively show a requirement for Rb in suppression of colorectal carcinoma in mice. Moreover, our model suggests that tumorigenesis occurs with, and thus likely requires, synergistic activation of β-catenin, as we find nuclear accumulation in our tumors and upregulation of Sox9. We hypothesize that this Wnt pathway deregulation occurs through genomic rearrangements induced by Rb inactivation. In fact, the variation in staining of APC among tumors is suggestive of random mutations in the apc gene itself and/or in other members of the pathway. Thus, our data indicate that Rb loss is a driver of tumorigenesis in the colon that leads to Wnt/β-catenin activation. We note that Kucherlapati et al. found that Rb loss in mice accelerates intestinal tumors initiated by apc mutation[40]. In that study, additional deletion of Rb shifts the spectrum of apc-deficient tumors from small to large intestine and augments the overall tumor burden. This finding argues that Rb has targets beyond the Wnt/β-catenin pathway that could reflect the known effect of Rb inactivation in promoting cell cycle entry and/or additional cooperating events resulting from Rb loss. In line with the cell cycle hypothesis, while direct mutation of Rb is relatively rare in colorectal carcinoma (0.5–4%), inactivation of the Rb pathway is a seemingly universal feature of colorectal cancer[14, 15]. Several studies have shown that in addition to showing mutation in apc or the Wnt/β-catenin pathway, most intestinal tumors typically carry mutations in growth signaling pathways expected to inactivate Rb function and promote proliferation. For example, a comprehensive analysis of mutations in human cancer showed a significant frequency of inactivating mutations (17%) in the ubiquitin ligase FBXW7, which positively regulates Rb by targeting its inhibitor cyclin E for degradation[14]. One potential exception to this role is a report, conducted on a very small sample size and prior to the availability of deep sequencing, of super-physiological levels of Rb correlating with chromosome 13 amplification[41]. This phenomenon has been explained through the inhibitory effect of E2F1 on β-catenin. E2F1 has been shown to promote β-catenin degradation and inhibit its transcriptional activity in an APC-independent manner in Drosophila and in cell lines[42]. Therefore, the authors speculate that maintenance of Rb expression would be required to prevent E2F1 inhibition of β-catenin function. While this mechanism may operate in the small subset of tumors that retain Rb, data from this current study and others lead us to conclude that loss or inactivation of Rb can certainly enable the development of intestinal tumors.

Another feature that highlights the significance and broad applicability of our model is that chromosomal instability is the most likely mechanism operating in the tumors. Our results show that Rb-deficient tumors do not undergo microsatellite instability. On the contrary, they contain double stranded DNA breaks and up-regulate the mismatch repair system suggesting the presence of foci of DNA damage that recruit DNA repair components in line with the presence of genomic instability. These observations dovetail nicely with prior reports showing that Rb-deficiency alters chromosome condensation and promotes chromosome missegregation by altering centromeric structure through changes in chromosome cohesion[26-28]. Taken together, these studies and ours, argue for a causal effect between Rb deficiency and chromosomal rearrangements through different mechanisms and in different settings, suggesting that chromosomal instability is the driving force of the Rb-deficient tumors.

In conclusion, here we show that Rb deficiency in the intestine causes the onset of aggressive, invasive intestinal tumors that share numerous histological and biological mechanisms with human colorectal cancer, indicating that this mouse system will prove useful for the modeling of this deadly cancer type.

MATERIAL AND METHODS

Mouse strains and histological analysis

Mouse colonies were maintained in compliance with IACUC guidelines. Fabpl−4×@132-Cre[13], Rb2[43], p53[44], and Rosa26-LSL-lacZ (Jackson Labs) conditional mutant mice (all defined as c/c in the text) were maintained on a mixed 129/Sv;C57Bl/6 genetic background. Animals were dissected and tissues were fixed in formalin except when they were stained for X-gal as described previously[45]. 5μm paraffin sections were used for all analyses.

Immunohistochemistry

Immunostaining was performed with antigen retrieval (20–30′ in boiling 10mM sodium citrate pH 6.0, 0.05% Tween 20) and by incubating primary antibodies overnight at 4°C. Detection and staining was performed with Vectastain ABC kit and DAB kits (Vector laboratories). Slides were counterstained with Hematoxylin. Antibodies used: anti-APC #sc-896 and anti-AR #sc-816, from Santa Cruz Technology; anti-β-catenin #610181 (BD); anti-chromogranin A #ab1560-1, anti-MLH1 #ab92312, anti-Sox9 # ab5535, and anti-P-MEK1 ab59329 from Abcam; anti-ck17/19 #3984, anti-MSH2 #D24B5, and anti-γ-H2AX #9718, from Cell Signaling; anti-p21 #LS-C88585 (Lifespan Biosciences), and anti-p53 #CM5 (Vector Laboratories).

Supplementary Figure 1. Analyses of . (a) β-galactosidase staining of Fabpl-CreLacZ animals and their relative LacZ controls shows that Cre recombinase is expressed in all of the cells of the urothelium lining the bladder, kidney calyx and ureters. The bottom panels show LacZ expression in the dorsal prostate (b) Whole mount β-galactosidase staining of small and large intestines shows that expression of Fabpl-Cre is very pronounced in the colon but restricted to few spots in the distal section of the small intestine (top panels). Nuclear fast red counterstained sections of whole mounts indicate that Cre expression occurs throughout the intestinal epithelium in the villi and crypts.