Aggressive pancreatic ductal adenocarcinoma in mice caused by pancreas-specific blockade of transforming growth factor-β signaling in cooperation with active Kras expression

Normal pancreas development in the pancreas epithelium-specific Tgfbr2 knockout mice

To investigate the effect of TGF-β signaling in pancreas development and disease, we first generated pancreas epithelium-specific Tgfbr2 knockout mice by crossing the Ptf1a^cre/+ mice (Kawaguchi et al. 2002) with the Tgfbr2^flox/flox mice (Chytil et al. 2002), resulting in the genotype Ptf1a^cre/+;Tgfbr2^flox/flox. By using the Ptf1a promoter, the Cre recombination and consequent Tgfbr2 knockout were expected to take place in all three pancreas epithelia lineages: duct, acinar, and islet cells. While the Tgfbr2^−/− mice were embryonic lethal (Oshima et al. 1996), the Ptf1a^cre/+;Tgfbr2^flox/flox mice were born at the expected frequency. PCR analysis demonstrated the recombined allele of the Tgfbr2 gene in the pancreas, and Cre-recombinase expression was uniformly detected in the acinar, ductal, and islet cells by X-gal staining of the mice further crossed with Rosa26 reporter mice (genotype: Ptf1a^cre/+;Tgfbr2^flox/flox;Rosa26r) (Soriano 1999), which confirmed the expected Tgfbr2 knockout in a pancreas epithelium-specific manner (Fig. 1A–C).

The pancreas of Ptf1a^cre/+;Tgfbr2^flox/flox mice appeared very normal (Fig. 1D). We analyzed the expression of islet cell markers, glucagon, insulin, somatostatin, pancreatic polypeptide, and the acinar cell marker amylase, which revealed that somatostatin-positive cells were more frequently observed in the Tgfbr2 knockout mice at the 1-d and 7-wk stages with statistical significance (p < 0.01). However, this difference disappeared by 12 wk of age, and other markers showed no difference between the knockout and control pancreas (Fig. 1E). Glucose tolerance tests and body weight monitoring also showed no significant changes compared with the controls (data not shown).

Thus, the knockout of Tgfbr2 had only subtle effects on pancreas development and the mice have shown no phenotype up to 18 mo. The apparently normal pancreas in the Ptf1a^cre/+;Tgfbr2^flox/flox mice indicated that the blockade of TGF-β signaling in the pancreatic epithelium was insufficient to initiate pancreas tumor formation. Other proto-oncogene activation might be required for spontaneous tumor initiation.

Knockout of Tgfbr2 rapidly promotes mutant Kras-induced PanIN progression to ductal adenocarcinoma

The Pdx1-Cre;LSL-Kras^G12D/+ or Ptf1a^cre/+;LSL-Kras^G12D/+ mice have been reported to have mPanIN lesions as early as at 2 wk of age, but most of the lesions did not progress into advanced cancer within 1 yr (Hingorani et al. 2003). We crossed the LSL-Kras^G12D/+ mice with our Tgfbr2 knockout mice and generated mice of the genotype Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox. These mice had active Kras expression plus Tgfbr2 knockout both in a pancreas epithelium-specific manner.

These mice, which were obtained at the expected Mendelian frequency at birth, began to demonstrate abdominal distension at ∼6–7 wk of age, which allowed us to “diagnose” their pancreatic tumor. The distension was due to ascites and the tumor itself, which was firm and palpable through the skin. The mice frequently demonstrated weight loss, bloody ascites, and sometimes jaundice, and were sacrificed when they appeared to be in distress; some died suddenly. The median survival was 59 d (8.4 wk) (Fig. 2A–C; Supplementary Table 1). These symptoms recapitulate well those frequently found in human PDAC. No animals with other genotypes showed any symptoms at this stage, except for one case (Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+) that had a cystic dilatation of the obstructed main pancreatic duct. A Kaplan-Meyer curve showed a dramatically shortened survival of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox with statistical significance (p < 0.001) (Fig. 2A).

The mice were dissected when they became distressed, mostly at 7–10 wk of age. The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumor uniformly occupied the whole pancreas and was irregularly shaped with a homogeneous white and firm appearance (Fig. 2B,C). No normal pancreas tissue was seen macroscopically. The duodenum loop was compressed by the tumor, resulting in obvious duodenal and gallbladder dilatation. Splenomegaly was also frequently observed. At this stage, the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ and Ptf1a^cre/+;LSL-Kras^G12D/+ mice showed only single or a couple of 1-mm-sized nodules, if any, which were also white and firm, while the vast majority of the pancreas still appeared normal. The other genotype pancreases (LSL-Kras^G12D/+;Tgfbr2^flox/flox, Ptf1a^cre/+;Tgfbr2^flox/flox, and Ptf1a^cre/+) showed no abnormal appearance. In those pancreatic tissues, the recombined alleles of Kras and Tgfbr2 genes were detected as expected according to the genotype, while the nonpancreas tissue (liver), a negative control, showed no recombination (Fig. 2D).

Upon microscopic examination, the Ptf1a^cre/+;LSL-Kras^G12D/+ mice demonstrated mPanIN lesions as reported previously (Hingorani et al. 2003) within a large amount of normal pancreas that contained well-organized acinar, ductal, and islet tissue (Fig. 3A). The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice showed similar histology. However, both genotypes showed no PDAC lesions that had grown beyond the basement membrane at the 7–10-wk stage. The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors exhibited uniformly well-differentiated glandular architecture (n = 25), which occupied the entire pancreas, resulting in almost complete loss of normal pancreatic tissue. The carcinoma cells frequently showed nuclear atypia and invasion beyond the basement membrane (Fig. 3B–D). The nascent invasion of tumor cells into the stroma was also revealed by pan-cytokeratin immunostaining (Fig. 3E). Carcinoma cells showed positive ductal markers, CK19 and mucin (Fig. 3F,G), and were negative for the acinar and islet marker, amylase, and insulin (data not shown), indicating ductal adenocarcinoma. Notably, the tumors were also richly distributed with a stromal component, showing positive for vimentin and smooth muscle actin staining (Fig. 3B,I,K), which was also found in the area surrounding mPanIN lesions in the Ptf1a^cre/+;LSL-Kras^G12D/+ pancreas (Fig. 3H,J). This indicated that stromal expansion had started in the early tumorigenesis stages. Moreover, a distinct and important feature of this mouse model is that the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox tumors did not show sarcomatoid architecture, which was seen in one-third of the Kras^G12D activation plus Ink4a/Arf knockout model (Aguirre et al. 2003). This feature is seen infrequently in human PDAC, providing another line of evidence that our mouse model is a close approximation of aggressive human PDAC. The penetrance of tumor appearance across all mice of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox genotype examined was 100%, and tumors were always accompanied by a whole panel of mPanINs and acinar–ductal metaplasia lesions (Fig. 3L).

The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDACs and Ptf1a^cre/+;LSL-Kras^G12D/+ mPanINs were strongly positive for EGFR, erbB2/Her2, and phospho-ERK (Fig. 4), while normal pancreatic ducts showed faint or negative staining (data not shown). This indicates that these signals were activated from the early stage of tumors, another feature closely paralleling the human disease (Hezel et al. 2006). E-cadherin was well maintained in the mPanIN lesions of Ptf1a^cre/+;LSL-Kras^G12D/+ mice, demonstrating typical cell–cell junctional staining, whereas the PDAC lesions of Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox mice revealed heterogeneity, with some areas completely lacking the junctional staining, while in many other areas it was still retained (Fig. 4). These results were consistent with another PDAC model (Hingorani et al. 2005) and suggested that the tumor was developing epithelial discontinuity and more invasive characteristics.

The pancreas epithelium-specific knockout of Tgfbr2 in the context of Kras^G12D expression revealed a dramatic enhancement of pancreatic tumor progression, and the tumor recapitulated human PDAC very closely.

Age progression of lesions in Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox model

We next examined mice at earlier ages to assess the severity of development of these lesions over the postnatal period. At 3.5 wk of age, the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox pancreas showed less frequent, but multifocal, isolated mPanINs and acinar–ductal metaplasia lesions surrounded by abundant normal pancreas area (Supplementary Fig. 1A–C), as was also observed in similar-age Ptf1a^cre/+;LSL-Kras^G12D/+ mice. At this stage, mPanIN-type lesions could be clearly distinguished from those representing acinar–ductal metaplasia; the distribution of mPanIN-1A, mPanIN-1B, mPanIN-2, and mPanIN-3 was 22%, 39%, 28%, and 11%, respectively. No invasive cancer lesion was found at this stage. At 5 wk of age, the distribution had shifted toward 8.3%, 20%, 17%, and 29%, respectively, in the same order, and invasive cancer lesions, which had invaded beyond the basement membrane, already accounted for 26% in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice (Supplementary Fig. 1D,E). Mice with other genotypes had no PDAC lesions even at 7–10 wk of age, when the whole pancreas in Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice had already become substituted by the tumor, resulting in loss of almost all normal acinar and islet tissue (Fig. 3B–D). We suggest that the order of appearance of the progressively more severe lesions is consistent with the hypothesis that there is a progression through a preceding stage of mPanIN lesion.

Invasive potential of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC

Lungs and livers of the tumor-bearing mice were subjected to histological analysis for metastasis (procedure in Supplemental Material). Although the tumors were very large, no distant metastasis was observed at 7–10 wk. The tumors showed minimal invasion into the duodenum, intrapancreatic lymph nodes, lymphatic channels, and/or nerve of peripheral pancreas, all of which are commonly found in human PDAC. Under this late-stage tumor burden, portal vein thrombosis was frequently detected (8/25 = 32%). The main trunk of the portal vein was completely obstructed with thrombus frequently accompanied by fibroblast involvement, which resulted in multiple necrotic areas in the distal liver parenchyma (Supplementary Fig. 2). The mice also demonstrated paraplegia in three cases. This was likely because of circulation failure due to direct oppression of large vessels by the rapidly growing tumor.

Although most Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice had to be sacrificed at 7–10 wk of age, three of this genotype survived up to 24–27 wk. All of them demonstrated PDAC with prominent desmoplasia and liver metastasis, and two showed lung metastasis, diaphragmatic and duodenal invasion, and peritoneal dissemination (Supplementary Fig. 2). These findings revealed that the Kras^G12D expression plus Tgfbr2 knockout tumor had substantial invasive and metastatic potential, again recapitulating aggressive human PDAC. The age of ∼25 wk was similar to the median survival of the Kras^G12D expression plus p53 inactivation model described previously (Hingorani et al. 2005). It is possible that the development of invasion and metastasis in all of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice might require a period longer than most of them in this study could survive.

Molecular analysis of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumor cells

Next, we established pancreatic cell lines from the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC, as well as from the mPanIN-type lesions in the Ptf1a^cre/+;LSL-Kras^G12D/+ pancreas. To address whether the mutant Kras expression plus Tgfbr2 knockout is an independent tumorigenesis pathway from those of previous mouse models, the status of other tumor suppressor genes highly related to human PDAC, p16^INK4a, p53, and Smad4 was examined using these cell lines. The coding sequences of those genes in several primary PDAC cell lines revealed no significant mutations, although one silent point mutation in Smad4 in one cell line was found (data not shown).

In Western blotting, Smad4 protein expression was detected in all cell lines examined (Fig. 5A). p53 was also expressed in most cell lines, and γ-irradiation induced p53 protein expression in most of the cell lines examined, which indicated normal p53 function in these cells (Fig. 5A,B). In contrast, the p16^INK4a protein level varied among the cell lines, and p19^ARF protein, which is derived from the alternative reading frame of the DNA locus shared with p16^INK4a, was not detected in any cell lines including the Ptf1a^cre/+;LSL-Kras^G12D/+ cells (Fig. 5A; data not shown). In RT–PCR analysis, p16^INK4a mRNA expression correlated well with that of protein, while p19^ARF mRNA expression was detected in most of the cell lines (Fig. 5C). The discrepancy of p19^ARF mRNA and protein expression remains to be elucidated.

The type II TGF-β receptor protein (TβRII) expression and TGF-β–Smad signaling were also analyzed. As expected by the Cre-mediated deletion, TβRII expression was lost in all PDAC cell lines examined, and upon TGF-β1 stimulation, downstream Smad3 phosphorylation was abrogated in those PDAC cell lines, while it was clearly observed in the Tgfbr2-intact cells (Fig. 5A,D).

PI3K–Akt signaling was also examined in those cell lines. It was expected that both mPanIN and PDAC cells have Akt activation, since the PI3K–Akt pathway is one of those downstream from the active Kras signals. On the other hand, the TGF-β signal activates the PI3K–Akt pathway, sometimes synergistically with the Kras signal (Janda et al. 2002). Our results showed Akt phosphorylation in both PDAC and mPanIN cells (Supplementary Fig. 3), which indicated that blockade of the TGF-β signal does not inhibit PI3K–Akt signaling in the Ptf1a^cre/+; LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC cells, and it is consistent with the cumulative data suggesting activation of this signal in human PDAC (for review, see Hezel et al. 2006). Cyclin D1, cyclin-dependent kinase 2 (CDK2), and CDK4 protein expression was also examined. Both the mPanIN and PDAC cell lines examined expressed abundant amounts of these proteins (Supplementary Fig. 3).

Taken together, the mutant Kras expression plus Tgfbr2 knockout was a sufficient combination for aggressive PDAC formation and progression through to large tumors with metastatic potential. Although some PDAC cells showed loss of both p16^INK4a and TβRII proteins, others retained intact p16^INK4a protein expression. The precise relation of the status of Tgfbr2 and p16^INK4a, as well as p19^ARF, remains to be investigated.

Heterozygous loss of Tgfbr2 promotes PDAC progression along with mutant Kras expression

Reduced TGF-β signaling (haploinsufficiency at the Tgfbr2 locus) also seems to interact synergistically with expression of the activated Kras protein in terms of inducing tumors. Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice also have a reduced survival compared with control animals and a much longer latency than the Ptf1a^cre/+; LSL-Kras^G12D/+;Tgfbr2^flox/flox (Fig. 2A). Different from the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice, most of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice did not show obvious symptoms (abdominal distension, weight loss, palpable hard tumor, ascites, or jaundice). When dissected at ∼200 d or later stage, firm, white-colored nodule(s) were frequently palpable in the pancreas head and/or body area; however, the tail area appeared normal.

The tumors in Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice had a similar histology of well-differentiated ductal adenocarcinoma to that of the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox tumors (Fig. 6A). Notably, the area of hard, white-colored mass in the pancreas head dominantly showed ductal adenocarcinoma histology, while the area of normally appearing pancreas tail dominantly showed mucin-rich glandular architecture corresponding to mPanIN-1B or mPanIN-2 (Fig. 6B). Normal pancreas structure had almost disappeared already.

These tumors frequently demonstrated gross metastasis to the liver and lung, direct invasion to the duodenum, and also peritoneal dissemination (Supplementary Table 2). The frequency of metastasis or invasion was much higher in the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/+ tumors dissected (6/12 = 50%) than in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors, although a certain number of the former cohort are still alive and must be examined later. It is suggested again that the latter cases could not survive long enough for metastasis or invasion due to the enormous tumor burden.

To determine whether the tumor had lost the wild-type allele of the Tgfbr2 gene through loss of heterzygosity (LOH), we performed laser-capture microdissection to obtain tumor epithelial DNA for subsequent PCR analysis. Both the primary and metastatic tumors still retained the wild-type allele of the Tgfbr2 gene (Fig. 6C). The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ PDAC cell lines retained the wild-type Tgfbr2 allele and showed TβRII protein expression and Smad3 phosphorylation upon TGF-β stimulation, indicating that TβRII protein function was also retained in those tumor cells (Supplementary Fig. 4), although some cells showed LOH and lost TβRII expression, which might be secondary changes arising during cell culture. Smad4 protein was also expressed in all the cells examined (Supplementary Fig. 4). Taken together, it is still possible that point mutation of the Smad3 or 4 gene caused biallelic inactivation of the TGF-β–Smad pathway; the data indicate that complete disruption of TβRII signaling is not necessarily required for PDAC formation, and haploinsufficiency of TGF-β signaling might be associated with PDAC progression, in the context of the mutant Kras expression.

Survival observation of Ptf1a^cre/+;LSL-Kras^G12D/+ mice has still been ongoing. Among eight mice observed for >200 d, one mouse developed metastatic pancreatic tumor at 249 d of age. For precise statistical analysis between the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ and Ptf1a^cre/+;LSL-Kras^G12D/+ mice, longer observation of the Ptf1a^cre/+;LSL-Kras^G12D/+ group is required. However, there appears to be a clear difference in terms of tumor frequency, latency, and histology between these two lines. By the time the first invasive tumor developed in the Ptf1a^cre/+;LSL-Kras^G12D/+ line, seven invasive tumors had occurred out of 25 Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/+ mice (7/25 = 28%). Thereafter, 18 of 25 mice developed invasive pancreatic tumor (18/25 = 72%) with median survival 347 d. According to the longer observations in the published studies, the median survival of the Kras-activated mice was described as ∼15 mo (Hingorani et al. 2005) or 57 wk (Bardeesy et al. 2006), which are longer than that of the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/+ group in our study, although the latency might be affected by the difference of background strains.

Notably, three of eight Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/+ tumors examined demonstrated a dominant sarcomatoid histology (50% or more) as well as the ductal adenocarcinoma (Fig. 6D–F; Supplementary Table 2). The sarcomatoid feature appeared similar to those in the previously described models (Aguirre et al. 2003; Bardeesy et al. 2006). All the three tumors demonstrated aggressive metastasis and/or direct invasion, in which the spindle-shaped cells were spreading along with the ductal tumor cells (Fig. 6E,F). Again, these sarcomatoid tumors were not observed in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice, indicating a distinct aspect of homozygous and heterozygous knockout of the Tgfbr2 gene. Moreover, the Pdx1-Cre;LSL-Kras^G12D/+ and Ptf1a^cre/+;LSL-Kras^G12D/+ pancreatic tumors, in all the cases published or that we observed, showed a dominant sarcomatoid or a significant poorly differentiated histology, specifically, 100% sarcomatoid in one recent study (Hingorani et al. 2003; Bardeesy et al. 2006). Taken together, heterozygous loss of Tgfbr2, shown in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice, might promote PDAC progression differently from both, the homozygous knockout in the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox mice and intact Tgfbr2 in the Ptf1a^cre/+; LSL-Kras^G12D/+ mice.

Connective tissue growth factor is strongly expressed in Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC stroma

What is the underlying mechanism in the rapid tumor progression in the Tgfbr2 knockout in the context of the mutant Kras expression? Strongly up- or down-regulated genes in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC compared with the Ptf1a^cre/+;LSL-Kras^G12D/+ mPanIN might provide important clues to understand the mechanism, and lead to the further identification of novel therapeutic targets for human PDAC. In this study we found that connective tissue growth factor (CTGF) was much more strongly expressed in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC tissues than in the Ptf1a^cre/+;LSL-Kras^G12D/+ mPanIN lesions (Fig. 7A–D). Immunohistochemistry revealed that the mPanIN stroma showed negative staining in contrast to the mPanIN epithelia (Fig. 7A,B). In clear contrast to this, the PDAC demonstrated a strong CTGF staining mainly in the stromal area (Fig. 7C,D), which is consistent with the previous report that the main source of CTGF in normal tissue is the stroma (Grotendorst 1997). The expression was especially strong in the stroma adjacent to the tumor epithelia (Fig. 7C,D), which suggested an active nonautonomous tumor–stromal inductive interaction. Since CTGF is known as a downstream target of TGF-β signaling (Grotendorst 1997) and PDAC tissue is also reported to express high levels of TGF-β (Friess et al. 1993), we next examined active TGF-β1 expression in the pancreas tissues by ELISA. It revealed higher TGF-β1 expression in both the Ptf1a^cre/+;LSL-Kras^G12D/+ mPanIN and the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC tissues, in a clear contrast to the control LSL-Kras^G12D/+; Tgfbr2^flox/flox and the Ptf1a^cre/+;Tgfbr2^flox/flox pancreas (Supplementary Fig. 5). This suggested that high TGF-β1 expression might have begun at the mPanIN stage and be sustained through PDAC progression.

Pancreatic tumors arising in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox recapitulate human PDAC

The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice rapidly developed PDAC, which resembled human disease closely in both symptoms and histopathology. Recently, the consensus regarding the criteria for determining mouse PanIN and PDAC was published, which documented that mouse PDAC, which developed through mPanIN lesions originally from the duct cells, should be considered as a close parallel of human PDAC, with clear distinctions between mPanIN and acinar–ductal metaplasia (Hruban et al. 2006). The tumors in this study exhibited this tumorigenesis pathway as with the previous LSL-Kras^G12D-based models (Aguirre et al. 2003; Hingorani et al. 2003, 2005). According to molecular analyses of the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox PDAC cell lines, the PDAC could develop without any alteration of other human PDAC-related tumor suppressor genes, p53, Smad4, and p16^INK4a, which indicated that the combination of Tgfbr2 knockout and active Kras expression was an independent tumorigenesis pathway of mouse PDAC from those of the previous mouse models. On the other hand, some of the PDAC cell lines showed both p16^INK4a and Tgfbr2 inactivation. It also suggested that there might be a certain underlying association in the p16^INK4a and TGF-β signaling pathways, which is similarly found as p16^INK4a and Smad4 inactivation in human PDAC (Rozenblum et al. 1997). Taken together, the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mouse represents a novel mouse model of human PDAC.

The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors uniformly demonstrated well-differentiated ductal adenocarcinoma, but not sarcomatoid tumor, in contrast to the Pdx1-Cre;LSL-Kras^G12D/+;Ink4a/Arf^lox/lox model (Aguirre et al. 2003), one-third of which showed sarcomatoid tumor, an infrequent type in human PDAC. Both models similarly showed a very short survival, which was much shorter than that of the Pdx1-Cre;LSL-Kras^G12D/+;LSL-Trp53^R172H/+ model (median survival: 8.4 wk vs. 5 mo) (Hingorani et al. 2005). The Ptf1a^cre/+; LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors also showed distant metastasis, direct invasion, and peritoneal dissemination, only when the mice survived until a similar age to those of the Pdx1-Cre;LSL-Kras^G12D/+;LSL-Trp53^R172H/+ model, which showed highly metastatic tumors. In the recent study (Bardeesy et al. 2006), Pdx1Cre;LSL-Kras^G12D/+;p53^lox/lox, Pdx1-Cre;LSL-Kras^G12D/+; p53^lox/+, and Pdx1-Cre;LSL-Kras^G12D/+;p53^lox/+;p16^+/− show a similar phenotype to human PDAC. Although occasional undifferentiated carcinoma was seen, these three show basically differentiated PDAC with distant metastasis, unless they die early due to the enormous tumor burden. This relationship of metastasis frequency and survival period was also seen in our model. Thus, in terms of short survival and differentiated histology, the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors might be one of the closest mouse models of human PDAC, suggesting the impact of blockade of TGF-β signaling.

These previously described LSL-Kras^G12D-based models used the Pdx1 promoter to drive Cre recombinase. In this study, the Ptf1a^cre/+ system showed pancreas-specific, homogeneous Cre recombination, not only in acinar cells but also ductal cells (Fig. 1B,C). In contrast, the Pdx1-Cre system shows more heterogeneity in Cre expression (Hingorani et al. 2003). Moreover, the Ptf1a^cre/+ system was more specific to pancreas than the Pdx1-Cre model. The Pdx1-Cre system developed extra-pancreas tumors (Aguirre et al. 2003; Hingorani et al. 2005), which might affect several aspects of the disease conditions. In contrast, the Ptf1a^cre/+ system did not show any extra-pancreatic phenotype in this study, even in the cerebellum (data not shown), where Ptf1a is also highly expressed and important for normal organ development (Hoshino et al. 2005). This suggests that there might be a certain advantageous circumstance in the pancreas, and not in the cerebellum, for tissue-specific context-dependent response of cells in the Ptf1a^cre/+ system; that is, the impact of Kras^G12D activation and blockade of TGF-β signaling becomes prominent only in the pancreas, not in the cerebellum. This may be because of a different proliferative mechanism, or prolonged plasticity and proliferative potential in the pancreas compared with the cerebellum, or stromal interaction that is deeper in pancreas than in the brain region. Accordingly, the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice may have several advantages compared with other existing mouse models.

Underlying mechanism of PDAC progression induced by Tgfbr2 knockout in the context of active Kras expression

As described above, the blockade of TGF-β signaling has a great impact on Kras-induced PDAC progression. What is the mechanism underlying the dramatic tumor progression in the Tgfbr2 knockout in the context of active Kras expression? Since TGF-β is a growth inhibitor of epithelial cells, it is possible that deprivation of this signal by knockout of Tgfbr2 might confer privileges to those cells for autonomous proliferation, resulting in rapid tumor formation. However, Tgfbr2 knockout alone did not cause any abnormalities up to 1.5 yr of age. It was suggested that Kras activation might be required first as a tumor-initiating event, and then the inactivation of either tumor suppressor gene p16^INK4a, p53, Tgfbr2, or Smad4 should occur for promoting tumor progression, which seems similar to the clinical findings for human PDAC. It seems also the same in other organs, that Tgfbr2 knockout alone does not initiate tumor formation, unless certain oncogene activation or chemical initiation exists (Forrester et al. 2005; Lu et al. 2006). To clarify the direct effect of TGF-β signal blockade on the mutant Kras-initiated PDAC progression, Tgfbr2 knockdown in the Ptf1a^cre/+;LSL-Kras^G12D/+ cells using siRNA might provide a better understanding of this tumorigenic process.

TGF-β signaling appears to be important especially in epithelial–stromal interaction (Bierie and Moses 2006). The vast stroma uniformly found in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox tumors also suggested there might be active interaction between the tumor and stroma. In this study, we showed strong CTGF expression in the PDAC stroma, especially the area adjacent to the tumor epithelium, which might reflect an active tumor–stromal interaction. Although CTGF is known as one of the target genes downstream from the TGF-β signal, high levels of TGF-β1 found in both the PDAC and mPanIN might not be able to explain the clear contrast of CTGF expression between the two, high in PDAC and low in mPanIN stroma. There might be another mechanism in the strong CTGF expression in the PDAC stroma. It was recently reported that CTGF promoted growth and invasion of pancreatic cancer cell lines in vitro, and its specific antibody attenuated orthotopic pancreatic tumor progression in vivo (Aikawa et al. 2006). Therefore, testing the activity of CTGF-blocking antibodies in this new mouse PDAC model might be effective. A positive result in this respect would lend strong support for developing a novel approach for target therapy of PDAC; that is, to target molecules expressed not just in the tumor epithelium itself but in the stromal cells that provide nonautonomous maintenance and progression signals, as was proposed in a recent study (Lu et al. 2006). In this regard, other stromal components (matrix metalloproteinases, tissue inhibitors of metalloproteinases, etc.) might be further investigated.

The level of TGF-β signal might be a significant determinant of pancreatic tumor formation and histology

The Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice also developed PDAC while retaining one wild-type Tgfbr2 allele, which indicated that complete knockout of Tgfbr2 is not always required, but that decreased TGF-β signaling can promote quite severe PDAC progression. The tumors progressed much more slowly than the complete Tgfbr2 knockout, presenting with less symptoms, although resulting in a similar final histology and a more frequent metastatic spread. In terms of decreased level of signaling, reduced Tgfbr2 expression by epigenetic silencing is already reported in human lung and prostate cancer (Zhang et al. 2004; Zhao et al. 2005), although there are controversial studies in pancreatic cancer that showed decreased or increased Tgfbr2 mRNA expression in the tumor tissues (Wagner et al. 1999; Venkatasubbarao et al. 2000). We found a gradation of tumors in the Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice, in which the PDAC lesions were frequently located in the pancreas head and body with low-grade mPanIN lesions in the tail area. The same pattern likely exists in earlier-age Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox mice, according to the 5-wk-old mice study, in which there was a minor invading tumor area in the pancreas head, contained within an abundant normal pancreatic tissue. What determines the difference in propensity to PDAC progression between the pancreas head and tail, which is also seen in human PDAC, remains to be investigated. It is known that the body and tail of the pancreas are derived from the dorsal pancreatic bud, while the head of the pancreas is mostly derived from the ventral pancreatic bud (Werlin 2004). This suggests that the difference of phenotypic response to carcinogenic genetic changes between the pancreas head and tail might be connected to genetic difference of their initial developmental anlagen, the ventral and dorsal buds.

In this study, three among eight Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+ mice examined demonstrated sarcomatoid tumor along with well-differentiated PDAC. In contrast, all 25 Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox mice developed well-differentiated PDAC, and none of them showed sarcomatoid histology. Since the sarcomatoid tumors always accompanied differentiated PDAC, it was likely derived through the mechanism of epithelial–mesenchymal transdifferentiation (EMT). Because the TGF-β signal is known to induce EMT as one of its major functions (Zavadil and Bottinger 2005), the amount of intact TGF-β signaling might be one of the important determinants of pancreatic tumor histology; that is, the more the TGF-β signal is intact, the more sarcomatoid tumor is obtained through EMT. Among the previously described PDAC models, the tumors that had both p16^INK4a alleles inactivated demonstrated a higher frequency of sarcomatoid histology (Aguirre et al. 2003; Bardeesy et al. 2006), all of which seemed to have intact TGF-β signaling. The relation of p16^INK4a inactivation and EMT remains to be investigated. As described above, some areas of the Ptf1a^cre/+; LSL-Kras^G12D/+;Tgfbr2^flox/flox tumor, however, showed E-cadherin mislocalization (Fig. 4), which might be associated with EMT. The Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox PDAC cells have more E-cadherin dysfunction than the mPanIN cells, but might still retain more functional E-cadherin than other PDAC cells that showed a sarcomatoid pattern.

It has been reported that either the wild-type p16^INK4a, Ink4a/Arf, or p53 allele was lost by LOH or inactivated by promoter methylation in the tumors of compound heterozygous genotype mice in the context of the Pdx1-Cre;LSL-Kras^G12D/+ background (Bardeesy et al. 2006). Therefore, haploinsufficiency of TGF-β signal might be a unique mechanism in PDAC progression.

Different phenotypes of Tgfbr2 knockout and Smad4 knockout in the context of active Kras expression

This study has demonstrated that blockade of TGF-β signaling might have significance in PDAC formation; however, mutation or deletion of the Tgfbr2 gene is found in <5% of human pancreatic cancer, among them especially in medullary carcinoma, a rare variant of pancreatic cancer (Goggins et al. 1998), although epigenetic silencing of Tgfbr2 may occur more frequently in human PDAC.

In contrast, the incidence of Smad4 mutation or deletion is 55%. Therefore, one might suggest that the Smad4 knockout (plus Kras^G12D expression) could provide a model more closely recapitulating human PDAC than that of the Tgfbr2 knockout plus Kras^G12D expression model (Kras + Tgfbr2KO) presented here. Pancreas-specific Smad4 knockout mice (Pdx1-Cre;Smad4^lox/lox and Ptf1a^cre/+;Smad4^lox/lox) also showed normal pancreas development similar to the Tgfbr2 knockout (N. Bardeesy and R.A. DePinho, pers. comm.). In contrast, pancreas-specific Kras^G12D expression plus Smad4 knockout (Kras + Smad4KO) mice (Pdx1-Cre;LSL-Kras^G12D/+;Smad4^lox/lox and Ptf1a^cre/+;LSL-Kras^G12D/+; Smad4^lox/lox) developed large pancreatic tumors by ∼10 wk with 100% penetrance. However, the tumor phenotype of Kras + Smad4KO differed from the Kras + Tgfbr2KO in that they were less aggressive. The Kras + Smad4KO mice developed intraductal papillary mucinous neoplasm (IPMN) that progressed to frank PDAC at low penetrance (N. Bardeesy and R.A. DePinho, pers. comm.); on the other hand, Smad4 deficiency cooperated with Kras^G12D and Ink4a/Arf heterozygosity to promote rapid development of PDAC. The more aggressive character of the tumor in the Kras + Tgfbr2KO compared with the Kras + Smad4KO (with intact Ink4a/Arf) might be explained by the idea that the knockout of Tgfbr2 abrogates all signaling downstream from the TGF-β receptors, having more impact than preventing only Smad4-dependent signal transduction. Since Smad4 is also a central signal mediator downstream from bone morphogenetic protein (BMP) and activin (Miyazono et al. 2000), one may suggest that blockade of BMP or activin signaling might be contributing to human PDAC progression. However, our study demonstrates that blockade of TGF-β signaling (with Kras^G12D expression) is sufficient to promote aggressive PDAC progression.

In conclusion, the Ptf1a^cre/+;LSL-Kras^G12D/+; Tgfbr2^flox/flox mice present a promising PDAC model with a very close approximation, by several key criteria, to the severe aggressive form of the human disease. Further intensive investigation of this model will provide a better understanding of the lethal PDAC and assist in developing new therapies.

Generating mouse lines

Tgfbr2^flox/flox mice (C57BL/6) (Chytil et al. 2002) were mated with Ptf1a^cre/+ mice (C57BL/6/DBA/2) (Kawaguchi et al. 2002) to generate Ptf1a^cre/+;Tgfbr2^flox/flox mice (C57BL/6/DBA/2). Then, the Ptf1a^cre/+, Tgfbr2^flox/flox, and LSL-Kras^G12D/+ mice (129/SvJae/C57BL/6) (Jackson et al. 2001) were interbred to obtain Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/flox, Ptf1a^cre/+;LSL-Kras^G12D/+;Tgfbr2^flox/+, and littermate control mice on a mixed C57BL/6/DBA/2/129/SvJae background. Some mice were also bred with Rosa26 lacZ reporter mice (C57BL/6) (Soriano 1999) for the purpose of tracing Cre-recombinase expression driven by the Ptf1a promoter. All mice were housed in the Animal Care Facility at Vanderbilt University following the Association for the Assessment and Accreditation of Laboratory Animal Care guidelines.

Mouse genotyping and recombined allele detection

Genotyping of Ptf1a^cre, LSL-Kras^G12D, floxed Tgfbr2 alleles, and LSL-Kras^G12D and Tgfbr2 recombined allele detection were performed by using oligonucleotide primers as described previously (Jackson et al. 2001; Chytil et al. 2002; Kawaguchi et al. 2002; Hingorani et al. 2003; Cheng et al. 2005). Recombined alleles were detected by using genomic DNA extracted from paraffin-embedded pancreas and liver tissue.

Histopathology

Mouse tissue was harvested and processed as described before (Forrester et al. 2005). Immunohistochemistry for vimentin and smooth muscle actin was performed by the Vanderbilt Histology Core Facility. Other antibodies and procedures used are in the Supplemental Material.

Establishing primary pancreatic cell lines

Mouse pancreas cells were separated and cultured in Vitrogen/Fibronectin-coated dishes in the RPMI media supplemented with 20% fetal bovine serum and antibiotics to obtain pure pancreas epithelial cell populations (Aguirre et al. 2003; Forrester et al. 2005). After the cells reached pure population, all experiments were performed within eight passages, to avoid possible gross genomic changes during long-term in vitro culture.

Molecular analysis of the primary pancreatic cell lines

RNA and protein of the primary mouse pancreatic cell lines were extracted for subsequent RT–PCR, sequencing, and Western blotting. Specific reagents and procedures are in the Supplemental Material.

Laser-capture microdissection

Epithelial tissue of primary pancreatic tumor, metastatic liver tumor, normal liver, and normal duodenum of the Ptf1a^cre/+; LSL-Kras^G12D/+;Tgfbr2^flox/+ mice was microdissected by laser, and DNA was extracted as described before (Cheng et al. 2005). Subsequently, PCR was performed for detection of wild-type, floxed, and recombined alleles of Tgfbr2.

Other experimental procedures are described in the Supplemental Material.