Inactivation of TIF1γ Cooperates with KrasG12D to Induce Cystic Tumors of the Pancreas

Inactivation of the Transforming Growth Factor Beta (TGFβ) tumor suppressor pathway contributes to the progression of Pancreatic Ductal AdenoCarcinoma (PDAC) since it is inactivated in virtually all cases of this malignancy. Genetic lesions inactivating this pathway contribute to pancreatic tumor progression in mouse models. Transcriptional Intermediary Factor 1 gamma (TIF1γ) has recently been proposed to be involved in TGFβ signaling, functioning as either a positive or negative regulator of the pathway. Here, we addressed the role of TIF1γ in pancreatic carcinogenesis. Using conditional Tif1γ knockout mice (Tif1γlox/lox), we selectively abrogated Tif1γ expression in the pancreas of Pdx1-Cre;Tif1γlox/lox mice. We also generated Pdx1-Cre;LSL-KrasG12D;Tif1γlox/lox mice to address the effect of Tif1γ loss-of-function in precancerous lesions induced by oncogenic KrasG12D. Finally, we analyzed TIF1γ expression in human pancreatic tumors. In our mouse model, we showed that Tif1γ was dispensable for normal pancreatic development but cooperated with Kras activation to induce pancreatic tumors reminiscent of human Intraductal Papillary Mucinous Neoplasms (IPMNs). Interestingly, these cystic lesions resemble those observed in Pdx1-Cre;LSL-KrasG12D;Smad4lox/lox mice described by others. However, distinctive characteristics, such as the systematic presence of endocrine pseudo-islets within the papillary projections, suggest that SMAD4 and TIF1γ don't have strictly redundant functions. Finally, we report that TIF1γ expression is markedly down-regulated in human pancreatic tumors by quantitative RT–PCR and immunohistochemistry supporting the relevance of these findings to human malignancy. This study suggests that TIF1γ is critical for tumor suppression in the pancreas, brings new insight into the genetics of pancreatic cancer, and constitutes a promising model to decipher the respective roles of SMAD4 and TIF1γ in the multifaceted functions of TGFβ in carcinogenesis and development.


Introduction
Pancreatic Ductal AdenoCarcinoma (PDAC), characterized by a ductal cell-type differentiation pattern, is the most common type of pancreatic cancer, accounting for more than 85% of pancreatic neoplasms. PDAC is the fourth leading cause of cancer-related mortality and carries an overall 5-year-survival rate of less than 5% [1]. The poor outcome of these patients is due to late diagnosis and resistance to current therapies. PDAC appears to arise from precursor lesions known as Pancreatic Intraepithelial Neoplasia (PanINs) or from two types of cystic tumors: Mucinous Cystic Neoplasms (MCNs) and Intraductal Papillary Mucinous Neoplasms (IPMNs) [2]. Mucinous cystic neoplasms are cysts lined by mucin-producing epithelial cells usually associated with an ovarian-type of stroma. These cysts do not communicate with the larger pancreatic ducts. IPMNs form intraductal papillary projections replacing the normal duct epithelium, secrete mucin, and communicate with ducts. IPMNs are currently classified according to their pattern of apparent histological differentiation into three main subtypes: intestinal (with the neoplastic epithelium resembling the intestinal epithelium), the most frequent, pancreatobiliary and gastric [3].
Recurrent genetic alterations have been identified in human PDAC [4,5]. Sporadic cases, which represent the vast majority of PDAC, are associated with activation of the KRAS oncogene (.90% of cases) and inactivation of the INK4A/ARF (.80% of cases), TP53 (.50% of cases) and SMAD4/DPC4 (.50% of cases) tumor suppressors. Inherited pancreatic cancers represent approx-imately 5-10% of all pancreatic cancers. In a high proportion of familial pancreatic cancers, the genetic alterations causing the disease are still unknown. However, several germinal mutations associated with complex familial syndromes have been shown to significantly increase the risk of developing pancreatic cancer (BRCA2, INK4A, STK11/LKB1, PRSS1, hMLH1 and hMSH2) [1,6].
In the last five years, a series of genetically engineered mouse models of PDAC have been developed based on these signature gene mutations [7]. For instance, expression of a constitutively active Kras mutant protein (Kras G12D or Kras G12V ) induces PanINs that eventually progress towards PDAC [8,9,10]. Kras activated mutants act in concert with inactivation of the p53 [11], Ink4A/Arf [12,13], and TbRII [14] tumor suppressors to accelerate development of PDAC. These models and others [15][16][17][18][19] support the concept that progression towards invasive PDAC involves emergence from different precancerous lesions (PanINs, MCNs and IPMNs) depending on the associated genetic alterations.
The TGFb pathway appears to be of particular importance to PDAC tumor suppression, since it is inactivated in virtually all cases of this malignancy [20], and since genetic lesions inactivating the pathway-inactivation of Smad4 or TbRII and over-expression of inhibitory Smad7contribute to pancreatic tumor progression in mouse models [14,[21][22][23][24]. Transforming growth factor beta (TGFb) is a secreted polypeptide belonging to a wide family of cytokines and growth factors including TGFbs, Bone Morphogenetic Proteins (BMPs) and activins [25,26]. Upon binding to its receptors, TGFb triggers phosphorylation of the SMAD2 and SMAD3 transcription factors. Phosphorylated SMAD2 and SMAD3 then interact with SMAD4. The SMAD2/3/4 complex accumulates within the nucleus, binds to DNA and activates the transcription of target genes leading to proliferative arrest or apoptosis of epithelial cells.
Transcriptional Intermediary Factor 1 gamma (also named TIF1c/TRIM33/RFG7/PTC7/Ectodermin) [27,28] appears to contribute to TGFb signaling, although its precise functional role is not clear. Some data point toward TIF1c as a negative regulator of the pathway through its capacity to mono-ubiquinate SMAD4 and limit SMAD4 nuclear accumulation [29,30,31]. In contrast, other studies have suggested that TIF1c plays an important positive role in transducing TGFb signaling through its interaction with SMAD2 and SMAD3 [32].
Here we wished to determine whether TIF1c contributes to tumorigenesis consistent with a function within the TGFb signaling pathway. We have focused on pancreatic exocrine tumors based on the prominent role played by TGFb signaling in these malignancies. Using a conditional mouse strain, we show for the first time that Tif1c is an important gene whose loss of function cooperates with Kras G12D activation to induce cystic pancreatic tumors resembling human IPMNs. We also report that TIF1c expression is down-regulated in human PDAC and some types of precursor lesions, supporting the relevance of our mouse model to human malignancy.

Results/Discussion
To selectively abrogate Tif1c expression in the pancreas, we crossed conditional Tif1c knockout mice [33] with Pdx1-Cre mice [34]. Pdx1 is a gene expressed in the common progenitor to all pancreatic lineages during early embryogenesis, hence Pdx1-Cre transgenic mice exhibit recombination of floxed alleles in pancreatic cells from all lineages (endocrine, acinar, centroacinar and ductal cells) [35]. Pdx1-Cre;Tif1c lox/lox animals were born at expected ratios and showed normal lifespan without obvious developmental or physiological alterations. Live imaging techniques (Positron Emission Tomography, PET and Magnetic Resonance Imaging, MRI), histological techniques (immunodetection of insulin, glucagon, PPY, chymotrypsine, F4/80, CD3, MPO), metabolic tests (glucose tolerance) did not reveal any significant differences between wild-type and Pdx1-Cre;Tif1c lox/lox littermates (n.20, between 3 weeks and 2 years of age) (data not shown). As expected, immunohistochemistry experiments showed that Tif1c was expressed in the nuclei of pancreatic cells in wildtype mice and that this staining was lost in the Pdx1-Cre;Tif1c lox/lox pancreas ( Figure S1). In all, these observations show that Tif1c is dispensable for normal pancreatic development and function in the mouse.
Activating KRAS mutations occur early in human PDAC pathogenesis and give rise to slowly progressing PanINs in mouse models [8,10,12]. We then asked whether Tif1c inactivation could modify the phenotype or latency of the pancreatic lesions induced by Kras G12D . To that end, we generated Pdx1-Cre;LSL-Kras G12D ; Tif1c lox/lox mice (n = 12, Table S1). All animals looked healthy at the time they were euthanized (the oldest animal was sacrificed at the age of 189 days). Since pancreatic lesions are often asymptomatic, we decided to explore in vivo the pancreas of these mutant mice (n = 4) by PET and MRI imaging techniques. Pdx1-Cre;LSL-Kras G12D ;Ink4A/Arf lox/lox mice, which exhibit rapid PDAC progression, were also employed in these studies. Strikingly, MRI imaging performed on Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox animals revealed an hypertrophic pancreas with multifocal cystic lesions exhibiting T2 hypersignals visible as early as 7 weeks after birth (T2 weighted scans allow detection of cysts as they are sensitive to water content) ( Figure 1B). Such lesions were absent in the pancreas of wild-type and Pdx1-Cre;LSL-Kras G12D animals ( Figure 1A) and were clearly different from those observed in Pdx1-Cre;LSL-Kras G12D ;Ink4A/Arf lox/lox mice, which harbor solid tumors exhibiting a T1 isosignals (T1 weighed scans allow detection of solid tumors) ( Figure 1C). PET imaging did not show significant increased metabolic activity in the abdomen of wildtype ( Figure 1D) and Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice ( Figure 1E) whereas Pdx1-Cre;LSL-Kras G12D ;Ink4A/Arf lox/lox mice had abdominal lesions with readily detectable metabolic activity

Author Summary
Inactivation of the TGFb tumor suppressor pathway contributes to the progression of Pancreatic Ductal AdenoCarcinoma (PDAC), a devastating malignancy. Transcriptional Intermediary Factor 1c (TIF1c) has recently been proposed to be involved in TGFb signaling, a pathway inactivated in virtually all cases of this malignancy. To address the role of TIF1c in pancreatic carcinogenesis, we used conditional Tif1c knockout mice. In a genetic background expressing a constitutively active mutation of KRAS oncogene (Kras G12D ) recurrently found in patients with PDAC, Tif1c inactivation induces pancreatic precancerous lesions resembling those observed in the absence of Smad4, a key player involved TGFb signal transduction. This observation strengthens the notion that TIF1c plays an active role in TGFb signaling. Interestingly, we also found that TIF1c expression was markedly down-regulated in human pancreatic tumors supporting the relevance of our findings to human malignancy. Characterization of new players involved in the outbreak of early pancreatic lesions that will eventually evolve into invasive pancreatic cancer is crucial to detect the disease earlier and eventually develop new therapeutic drugs.
( Figure 1F). Macroscopic analysis of the Pdx1-Cre;LSL-Kras G12D ; Tif1c lox/lox pancreas confirmed the presence of numerous cysts affecting the entire organ without macroscopic evidence of invasive carcinoma ( Figure 1H and Figure S2) whereas the pancreas of Pdx1-Cre;LSL-Kras G12D ;Ink4A/Arf lox/lox mice was invaded by a firm and homogeneous mass ( Figure 1I) and the pancreas from wild-type or Pdx1-Cre;LSL-Kras G12D animals had a normal macroscopic appearance ( Figure 1G). The size of these cysts observed in the pancreas from Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox was variable and most of them contained papillary projections ( Figure 1K). These lesions clearly contrast with the invasive tumors of ductal morphology identified as PDAC in the pancreas of Pdx1-Cre;LSL-Kras G12D ;Ink4A/Arf lox/lox ( Figure 1L). The histological analysis of the 12 Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox animals (Table  S1) revealed the presence of cystic lesions in 100% of these mice, such cysts being never observed in Pdx1-Cre;LSL-Kras G12D or wild type control animals. Quantitative analysis revealed that the area occupied by the abnormal pancreas exceeded 50% by the age 6 weeks in 6/7 Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice whereas it represented less than 20% in 6/6 Pdx1-Cre;LSL-Kras G12D mice (Table S1). Collectively these data demonstrate that inactivation of Tif1c actively cooperates with activated Kras G12D to induce cystic tumors of the pancreas.
To carefully compare the pancreatic lesions observed in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox with the Pdx1-Cre;LSL-Kras G12D controls, we performed a sequential histological analysis of pancreas from animals (n = 12) euthanized at different ages (Table S1). Contrary to wild-type mice (Figure 2A-2D), Pdx1-Cre;LSL-Kras G12D mice gradually developed focal PanINs by the age of about 10 weeks ( Figure 2E-2H). Strikingly, in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox pancreas, PanINs, signs of acute inflammation as well as enlarged and dilated ductal structure resembling budding cysts were observed as early as 3 weeks of age ( Figure 2I and Figure S3). At later time points, inflammatory tissue and PanINs were mainly replaced with cystic lesions becoming more numerous and of larger size ( Figure 2J-2L). Microscopic examination revealed that the lining of the cystic structures characteristically found in the pancreas from Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice was formed by epithelial cells with a cuboidal or cylindrical morphology. These cells formed numerous thick papillary projections in the cyst lumen. The axis of these projections usually contained masses of small monomorphic cells with an endocrine morphology (Insets in Figure 2I-2L).
We performed immunohistochemical studies to characterized the evolving pancreatic lesions in these mice. We observed staining for chymotrypsin and insulin, which decreased with age, indicating a replacement of exocrine and endocrine components, together with abnormal ductal structures in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice ( Figure 3). There was a notable disappearance of wellorganized endocrine islets with age coinciding with the accumulation of endocrine cells within the papillary projections bulging within the lumen of the cysts.
To more precisely identify the nature of these lesions, several lineage markers were explored by immunohistochemistry. We first verified that Tif1c expression was lost in pancreatic ducts from Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice ( Figure 4E) compared to normal observed ducts in wild-type mice ( Figure 4A). In the normal pancreas, cytokeratin 19 (CK19) is specifically expressed by ductal cells lining the secretory ducts ( Figure 4B). We verified that most of the epithelial cells lining the cystic lesions observed in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice were positive for CK19 ( Figure 4F); this is consistent with a ductal phenotype for these cells. However, in contrast to the cells lining the normal secretory ducts ( Figure 4C), many cells lining the cysts were mucus-secreting and stained for Alcian blue ( Figure 4G). The cells with an endocrine appearance present within the intra-cystic papillary projections were CK19 and Alcian blue negative. There was no evidence of ovarian-type stroma or of invasive or microinvasive carcinoma, even on serial sections, suggesting that these cystic tumors resemble human IPMNs. In all, the cystic lesions observed in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice show distinctive characteristics, including the presence of intraepithelial endocrine pseudo-islets, suggestive of a mixed, endocrine-exocrine, lesion [7].
TIF1c has recently been proposed to be involved in TGFb signaling [29,32]. The resemblance between the cystic lesions (either IPMNs [21,23] or MCNs [22]) observed in Pdx1-Cre;LSL-Kras G12D ;Smad4 lox/lox mice, and the cystic lesions we observed in the Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice, reinforces an active role of TIF1c in TGFb signaling. However, we cannot rule out the possibility that TIF1c could also be involved in other signaling pathways. Interestingly, we observed that Smad4 expression was almost undetectable in epithelial cells lining the papillary projections observed in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice whereas it was detectable in epithelial cells lining the cysts ( Figure 4H) or in normal ducts ( Figure 4D). IPMNs observed in Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice always contain a significant endocrine component, a rare event in IPMNs observed in Pdx1-Cre;LSL-Kras G12D ;Smad4 lox/lox mice [21,23]. This observation suggests that TIF1c and SMAD4 could differentially regulate endocrine versus exocrine differentiation in a context of an activated KRAS oncogenic protein.   Based on the prominent cooperation noted between Kras activation and Tif1c inactivation in promoting cystic pancreatic tumors in our mouse model, we speculated that TIF1c expression may be down-regulated in human pancreatic tumors. To test this hypothesis, we analyzed by quantitative RT-PCR the expression level of TIF1c and SMAD4 mRNA in 20 PDAC and 16 peritumoral tissues coming from surgical specimens removed for therapeutic purposes (peritumoral tissues were not available for 4 of these patients). The cellularity of the samples used for molecular analysis was verified histologically. Our results show that TIF1c expression is significantly decreased in the tumors as compared to peritumoral tissues (P = 0.0054) ( Figure 5A). We also compared TIF1c expression levels in each individual tumor along with the peritumoral tissue from the same patient (n = 16). Our results show that TIF1c expression is significantly down-regulated in most patients and is not up-regulated in any patient ( Figure 5B). We next examined TIF1c protein pattern of expression by immunohistochemistry in human pancreatic cancers and their precursors. In peritumoral tissues from PDAC, TIF1c was detected in the majority of the nuclei of acinar, ductal and endocrine cells ( Figure 5C). Centroacinar cells are more difficult to identify in routinely stained sections; however, since no epithelial cell population devoid of TIF1c expression has been detected in the normal pancreas, it can be assumed that they also express TIF1c. In PDAC (16 cases), TIF1c nuclear expression level was significantly decreased as compared to the peritumoral tissue. In 8 cases, TIF1c expression was heterogeneous, with large numbers of negative cells coexisting with scattered positive cells ( Figure 5D). In 2 cases, TIF1c was even undetectable ( Figure 5E). In IPMNs (samples from 10 patients, all with the intestinal subtype according to current classifications [3]), almost all neoplastic cells in areas of low grade dysplasia (present in the 10 cases) displayed a weak nuclear positivity whereas in areas of high grade dysplasia (present in the 10 cases), more than 50% of cells were negative for TIF1c; this was especially the case along the papillary projections ( Figure 5F). In PanINs (samples from 15 patients, with grade 1 in 12, grade 2 in 10 and grade 3 in 8), the expression of TIF1c was usually retained in grade 1 and 2 lesions (data not shown), but was undetectable in a variable proportion of cells in grade 3 lesions ( Figure 5G). In MCNs (8 cases), TIF1c protein was strongly expressed by all neoplastic cells, even in areas of high grade dysplasia and in foci of microinvasive carcinoma ( Figure 5H). We showed in the same set of tumors that SMAD4 expression was also down-regulated in high-grade PanINs, IPMNs and PDAC, while remained highly expressed in MCNs ( Figure S4).
None of the Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice (n = 4) sacrificed after the age of 13 weeks showed an aggressive cancer developed from IPMNs. This observation is consistent with epidemiological data in humans showing that IPMNs only rarely give rise to aggressive tumors [3]. Interestingly, IPMNs were reported in Pdx1-Cre;LSL-Kras G12D ;Smad4 lox/lox mice but no PDAC was found by the age of 13 weeks (n = 8) [21]. Another group observed in Pdx1-Cre;LSL-Kras G12D ;Smad4 lox/lox a significant proportion of PDAC between 23 and 33 weeks [23]. This suggests that the minimal latency period to see the onset of aggressive tumors may have not been reached in our study or the number of animals studied is too low. We have been in the process of ''aging'' a cohort of Pdx1-Cre;LSL-Kras G12D ;Tif1c lox/lox mice to address this specific point even if we cannot rule out the possibility that these animals die before developing aggressive tumors because of pancreatic failure due to growing cysts. The molecular mechanism supporting the cooperation between activated Kras G12D mutation and Tif1c inactivation to induce the formation of IPMNs is still unknown. TAK1 (TGFb Associated Kinase 1), which has recently been proposed to explain R-Ras and TGFb cooperation in breast tumors, is an interesting candidate [36].
The relationship between TIF1c and SMAD4 and their respective role in TGFb signaling have been a subject of extensive investigation and debate in the last few years [37]. Indeed, published data support distinct models whereby TIF1c could function as either a negative regulator of TGFb signaling [29,30] or a complementary agonist of TGFb signaling [32]. In the ''antagonist model'', TIF1c negative function relies on its ability to mono-ubiquitinate and relocate SMAD4 into the cytoplasm. In the ''agonist model'', TIF1c competes with SMAD4 for binding to SMAD2 and 3 and form TIF1c-SMAD2/3 complexes regulating SMAD4-independent TGFb responses. One can envision that these models, both supported by compelling biochemical and in vivo evidence, are not mutually exclusive and that one of them may be predominant depending on the cellular context. The experimental evidence we present here suggest that TIF1c works with SMAD4 as a complementary agonist molecule during pancreatic tumorigenesis. In the presence of activated Kras, Tif1c loss-offunction induces cystic lesions resembling those observed in the absence of Smad4 suggesting that both molecules act in concert to prevent tumor progression. Even if this hypothesis needs further demonstration, it is strengthened by the observation that TIF1c expression is decreased in human pancreatic tumors and our observation that loss of Tif1c does not significantly impair Smad4 expression level or Smad4 target genes expression (data not shown). The existence of a joint effort between TIF1c and SMAD4 to maintain TGFb-mediated tissue homeostasis has been proposed before. Indeed, during erythroid differentiation, TIF1c mediates the differentiation response while SMAD4 mediates the antiproliferative response [32]. In a recent work, we showed that Tif1c controlled iNKT (invariant Natural Killer T) cell expansion whereas Smad4 maintained their maturation state [33]. A recent comprehensive genetic analysis of .20,000 transcripts in 24 pancreatic cancers failed to identify point mutations, amplifications, deletion or translocations in the TIF1c gene [20]. The present study strongly spurs us toward looking for TIF1c genetic alterations in a larger set of pancreatic tumors. Besides, chromosomal breakpoints chromosome on 1p13.1 containing TIF1c gene have been reported in acute megakaryocytic leukemias [38], osteochondromas [39], bronchial large cell carcinomas [40] and childhood papillary thyroid carcinomas [41]. Interestingly, we recently demonstrated that abrogation of the closely related Tif1a gene in mice caused hepatocellular carcinoma [42]. These observations reinforce the idea according to which TIF1c loss of function could play an active protective role during tumorigenesis. TIF1c overexpression has been suggested by others to facilitate tumorigenesis in other organs by inhibiting SMAD4-mediated growth inhibition and motility in response to TGFb [29,30]. This observation may reflect an active role of TIF1c during tumor progression depending on the organ and involving a anti-SMAD4 mechanisms (''antagonist'' model).
In conclusion, we demonstrated in a mouse model that inactivation of Tif1c cooperates with activated Kras G12D to induce cystic pancreatic tumors. Characterization of new players involved in the outbreak of early pancreatic lesions that will eventually evolve into invasive pancreatic cancer is crucial to detect the disease earlier and eventually develop new therapeutic drugs. Further work to decipher the respective roles of SMAD4 and TIF1c in PDAC as well as the functional cooperation between KRAS and TIF1c could bring new insight into the etiology of pancreatic cancer, and generate a better understanding of the multifaceted role of TGFb in carcinogenesis and development.

Mice
Tif1c lox/lox [32] mice harboring floxed exons 2-4 were generated by K.Y. and R.L. will be described elsewhere. Briefly, using a genomic clone that contains a portion of the Tif1c gene, we generated a targeting vector in which a PGK Neo selection cassette flanked by two loxP sites was introduced into intron 1 and a third loxP site inserted into intron 4. This targeting vector was designed with the expectation that upon homologous recombination and subsequent Cre recombinase-mediated excision, exons 2, 3 and 4 along with the PGK-Neo cassette would be deleted, thereby causing a frameshift mutation with a premature termination codon in exon 5. The putative product of this deleted gene corresponds to a truncated protein lacking part of the RING finger-B box-coiled coil (RBCC) motif and the entire C-terminal region of the TIF1c protein, which contains the conserved PHD finger/bromodomain unit.

Human samples
Cryopreserved tumoral and peritumoral tissue samples were obtained from an institutional tissue bank, the Tumorothèque des Hospices Civils de Lyon (Centre de Ressources Biologiques, Hospices Civils de Lyon). In accordance with French ethical rules, samples were from patients having given their informed consent or from deceased patients. Prior to molecular analysis, the quality and cellularity of tissue samples was verified histologically; tumor tissues were selected in order to contain a significant amount of neoplastic cells; peritumoral tissues were constantly altered by reactive fibrotic changes associated with a loss in acinar tissue and a massive ductular proliferation.

RNA analysis
Liquid nitrogen frozen human tumors were blended using a Pro200 homogenizer (Pro Scientific Inc.) in a 5 M guanidine solution. Total RNA was further purified by RNeasy mini kit (Qiagen). The cDNA was used as template with RT Kit SuperScript II (Invitrogen). quantitative RT-PCR was performed as previously described [48].

Live imaging
For MRI and PET experiments, mice were anesthetized using 3% isoflurane inhalation (TEM Sega, Lormont, France) and maintained in 1.5% isoflurane atmosphere during experiments.
For PET experiments, the mice were catherized in the caudal vein (24 gauge), injected with 250 mCi of 300 mL of radioactive 18-Fluorodéoxyglucose (FDG). After 90 minutes to allow FDG fixation, images were acquired during 15 minutes (constant 2% isoflurane atmosphere) using the ''TEP clearPET'' (Raytest, Inc.). MRI acquisitions were made with a BioSpec-7T system (Bruker, Ettlingen, Germany) using a 32 mm inner-diameter emission/reception volume coil (Rapid Biomedical, Würzburg, Germany). T1/T2-weighted contrast sequences synchronized to respiration were acquired for each mouse. A RARE (Rapid Acquisition with Relaxation Enhancement) sequence (TR/TE 3500/38.1 ms) with fat saturation was used. Geometric parameters were: a series of 18, 750 mm thick sections, 33 mm field of view, and 2566256 pixel matrix. Voxel size was therefore 12961296750 mm 3 . Figure S1 Loss of nuclear Tif1c protein expression in the pancreas of Pdx1-Cre; Tif1c lox/lox mice. Immunohistochemistry showed that Tif1c was expressed in the nuclei of pancreatic cells in wild-type mice and that this staining was lost in a Pdx1-Cre;TIF1c lox/lox pancreas. Quantitative RT-PCR to detect SMAD4 expression from the 16 PDAC for which peritumoral tissue was available (A). For each patient represented by an individual bar, SMAD4 expression is represented as a percentage of variation relative to mRNA expression in the peritumoral tissue from the same patient (B). SMAD4 protein expression pattern was also assessed by immunohistochemistry. In the normal pancreas (C), SMAD4 is strongly detected in endocrine islets, a faint labeling is visible in acinar cells and ductal cells. In adenocarcinomas (n patients = 20), no labeling for SMAD4 was detected whereas adjacent residual endocrine cells are positive (D). In another example of adenocarcinoma, SMAD4 is faintly but readily detectable in neoplastic cells; the labeling is both cytoplasmic and nuclear (E). In IPMN grade 3, SMAD4 expression is heterogeneous; most cells are negative, while a few scattered cells retain a faint expression, usually nuclear (F). In PanIN grade 3, SMAD4 expression is either undetectable, as exemplified in the largest figure (note the persistent expression in adjacent residual endocrine cells (arrow)), or heterogeneous, as shown in the inset (G). In MCN, SMAD4 expression is usually strong, in low grade (large figure) as well as in high grade (inset) lesions (H).