In vivo reprogramming drives Kras-induced cancer development

The faithful shutdown of the somatic program occurs in the early stage of reprogramming. Here, we examined the effect of in vivo reprogramming on Kras-induced cancer development. We show that the transient expression of reprogramming factors (1–3 days) in pancreatic acinar cells results in the transient repression of acinar cell enhancers, which are similarly observed in pancreatitis. We next demonstrate that Kras and p53 mutations are insufficient to induce ERK signaling in the pancreas. Notably, the transient expression of reprogramming factors in Kras mutant mice is sufficient to induce the robust and persistent activation of ERK signaling in acinar cells and rapid formation of pancreatic ductal adenocarcinoma. In contrast, the forced expression of acinar cell-related transcription factors inhibits the pancreatitis-induced activation of ERK signaling and development of precancerous lesions in Kras-mutated acinar cells. These results underscore a crucial role of dedifferentiation-associated epigenetic regulations in the initiation of pancreatic cancers.

I nduced pluripotent stem cells (iPSCs) can be established from somatic cells by the transient expression of pluripotencyrelated transcription factors (TFs) 1 . Given that epigenetic regulation plays a central role in cell fate determination and maintenance, this technology allows for the reorganization of epigenetic regulation from the differentiated somatic cell state into the pluripotent state without affecting genetic information [2][3][4] . Recent studies revealed the stepwise epigenetic changes during iPSC derivation 5,6 . In early-stage reprogramming, the faithful shutdown of the somatic program occurs through the silencing of somatic cell-specific enhancers, which is associated with a loss of original cell identity. Upon prolonged expression of the reprogramming factors, gradual activation of the pluripotency-related transcriptional network is observed in late-stage reprogramming.
Notably, such events in cellular reprogramming are similarly observed in cancer development. Cancer cells often lack terminal differentiation ability, suggesting that repression of the somatic cell program is involved in cancer development. In contrast, poorly differentiated cancers that lose most differentiation features sometimes exhibit an activation of pluripotency-related genes such as targets of Nanog, Oct3/4, and Sox2 (ref. 7 ). Moreover, poorly differentiated cancers with pluripotency-related gene expressions are associated with high malignancy, especially in the mammary glands 7 . Collectively, these findings suggest that some aspects of cellular reprogramming are associated with cancer development.
Indeed, our previous study demonstrated that premature termination of in vivo reprogramming leads to kidney cancer development through altered epigenetic regulation 8 . Consistent with the partial reprogramming state, these cancer cells lose kidney cell-specific molecular signatures while they partially acquire the trait of embryonic stem cells (ESCs) including selfrenewing capacity. Notably, these cancers resemble Wilms' tumor, which is the most common childhood kidney cancer. Furthermore, these cancer cells were readily reprogrammable into iPSCs that are capable of differentiating into non-cancerous kidney cells 8 . These results raised the possibility that reprogramming-associated epigenetic regulation has a significant impact on childhood cancer development, which is also in agreement with recent observations that childhood cancers harbor relatively few genetic mutations. However, the functional significance of epigenetic regulation related to cellular reprogramming remains largely unclear in adult cancer development.
Pancreatic cancer is one of the most common causes of cancer mortalities in adults in developed countries. The median survival period is less than 6 months and the 5-year survival rate is 3-5% 9,10 . The most common type of pancreatic cancer is pancreatic ductal adenocarcinoma (PDAC). PDAC is one of the wellcharacterized cancers for multistep cancer progression models that have "Big 4" mutations (mutations in KRAS, CDKN2A, TP53, and SMAD4) 11,12 . Histologically, PDAC is thought to arise from a spectrum of preneoplastic lesions with ductal morphology that are designated as pancreatic intraepithelial neoplasia (PanIN) 13 . Since early PanIN lesions harbor mutations or amplification of the KRAS at high frequency (over 90%), it has been proposed that KRAS mutation is responsible for PanIN formation and thus is an initial event during pancreatic cancer development 14,15 . In contrast to premalignant lesions, PDAC often harbors additional mutations such as a loss of CDKN2A and inactivating mutations at TP53 and SMAD4 11 . Together, current consensus holds that pancreatic cancer arises through a stepwise progression with characteristic genetic mutations for each histologically distinguishable stage.
In the present study, based on the shared properties between cellular reprogramming and cancer development, we examined the effect of in vivo reprogramming in the development of Kras-induced cancer. We show that reprogramming-mediated repression of somatic cell enhancers, which is associated with early stage of reprogramming, in conjunction with Kras mutation results in rapid development of PDAC. We also show that partial acquisition of the ESC signature, which occurs later stage of reprogramming, causes the development of cancers that resemble human α-fetoprotein (AFP)-producing cancer. These results highlight the crucial role of reprogramming-related epigenetic regulation in Kras-induced cancer in vivo.

Results
Kras and p53 mutations are insufficient for ERK activation. We first generated Pdx1-ires-Cre knocked-in ESCs by homologous recombination (Fig. 1a and Supplementary Fig. 1A), and then Pdx1-ires-Cre knocked-in mice to induce pancreas-specific Cre-LoxP recombination. Lineage tracing analysis of Pdx1-ires-Cre mouse using Rosa26 LacZ reporter allele 16 confirmed that the Cre-LoxP recombination occurs in almost all pancreatic cells ( Supplementary Fig. 1B, C), which was consistent with a previous report using transgenic mice expressing Pdx1-Cre 17 . Similarly, we knocked-in LoxP-Stop-loxP (LSL)-HA tag-Kras G12D allele into endogenous Kras locus in ESCs ( Supplementary Fig. 1D) and generated knocked-in mice. We then generated LSL-HA tag-Kras G12D and Pdx1-ires-Cre (hereafter HA-KC) compound mice to examine the expression of Kras G12D oncoprotein (Fig. 1a). HA immunostaining exhibited membranous expression of Kras G12D oncoprotein in the pancreatic cells of HA-KC mice at 6 weeks of age (Fig. 1b). However, despite the expression of Kras G12D oncoprotein, most pancreatic cells were histologically normal ( Fig. 1b) except for the focal formation of early PanIN, indicating that Kras mutation alone cannot transform pancreatic acinar cells.
To further investigate the effect of oncogenic mutations on the ERK signaling pathway and aberrant proliferation, we next generated LSL-Kras G12D , Pdx1-ires-Cre (KC) mice and LSL-Kras G12D , LSL-p53 R172H , Pdx1-ires-Cre (KPC) compound mice (Fig. 1c) using well-established Kras and p53 mutant alleles 18,19 . In accordance with the results in HA-KC mice, most pancreatic cells exhibited recombination in both Kras G12D and p53 R172H alleles in KPC mouse (Fig. 1d, Supplementary Fig. 9). However, the majority of pancreatic cells were histologically normal in both KC and KPC mice at 6 weeks of age, except for the spotted formation of early PanIN and the dysplastic area (Fig. 1e). Although PDAC development was observed in KPC mice at 8 weeks of age, the affected area was still limited, and most pancreatic cells remained histologically normal (Fig. 1e). Consistently, pERK immunostaining showed positive staining only in PanIN and PDAC lesions, while the majority of pancreatic cells were negative for pERK (Fig. 1e), demonstrating that compound mutations of Kras and p53 are insufficient for robust activation of the ERK pathway. Similarly, the majority of pancreatic cells in KPC mice were negative for Ki67, a cell proliferation marker (Fig. 1e), which indicates that compound mutations of Kras and p53 gene are insufficient for inducing the abnormal proliferation of pancreatic cells. These results suggest that aberrations beyond Kras and p53 mutations are required for PDAC development.
Reversible ductal metaplasia by in vivo reprogramming. In order to investigate the influence of in vivo reprogramming in the pancreas, we next generated mice in which reprogramming factors can be induced specifically in the pancreas upon doxycycline (Dox) treatment. We established mice harboring Pdx1-ires-Cre, Rosa LSL-rtTA3, and Col1a1::tetO-Oct3/4, Sox2, Klf4, c-Myc-ires-mCherry (hereafter C-OSKM) (Fig. 2a). After Dox treatment, we confirmed mCherry fluorescence in the pancreas of these mice  Fig. 1 Kras/p53 compound mutations are insufficient for PDAC development. a A schematic illustration of the genetic construct to activate HA-Kras G12D in the pancreas. b Representative histology and immunostaining for HA-Kras G12D . Scale bars, 50 μm. c Genetic strategy to study the influence of somatic activation of Kras and p53 mutations in the pancreas. d Southern blotting of LSL-Kras G12D and LSL-p53 R172H allele. ESCs containing Pdx1 ires-Cre, LSL-Kras G12D and LSL-p53 R172H alleles were used as control. Note that majority of two LoxP alleles are converted into one LoxP allele in the pancreas of KPC mouse (Cont control, Panc pancreas). e Immunostaining for pERK and Ki67 in the pancreas of 6-week-old wild-type mice, KC mice, and KPC mice. KPC mice at 8 weeks of age showed a focal PDAC area with pERK staining (bottom). Scale bars: HE and pERK (low magnification) staining, 200 μm; pERK (high magnification) staining, 50 μm; and Ki67 staining, 100 μm (Supplementary Fig. 2A). We also performed histological analysis of 4-week-old mice given Dox for 3 days (Fig. 2b).
It has been proposed that the transition of acinar cells into a duct-like state (acinar to ductal metaplasia (ADM)) is an early event of pancreatic cancer development [20][21][22] . Notably, Doxtreated mice often exhibited dilatation of the acinar glands and expression of CK19, a ductal marker (Fig. 2c). CK19 expression was sometimes observed in amylase-expressing cells (Fig. 2c), suggesting that ADM occurs by the short induction of reprogramming factors. Consistent with this notion, Oct3/4 expression was observed in newly formed ductal cells (Fig. 2d). Notably, after 1 week of Dox withdrawal, the histology of the pancreas was almost normal, and CK19-expressing ADM lesions were hardly detected (Fig. 2e, f) A recent study demonstrated that continuous Klf4 overexpression induces ADM 23 . To further investigate whether cellular reprogramming is indeed related to ADM formation, we next examined the effect of another reprogramming cocktail that lacks Klf4. For this purpose, we established ESCs with Doxinducible polycistronic cassette containing Sall4, Nanog, Esrrb, and Lin28 (hereafter SNEL), which was reported as an alternative reprogramming factor cocktail, using the KH2-ESC system (Fig. 2g) 24,25 . To investigate the effect of SNEL expressions on reprogramming, mouse embryonic fibroblasts (MEFs) containing Dox-inducible SNEL alleles were established from E13.5 embryos. After 2 weeks of Dox exposure to SNEL-MEFs, iPSC-like colonies were observed ( Supplementary Fig. 2B, C). Although these iPSClike colonies could not be expanded, they exhibited increased alkaline phosphatase activity ( Supplementary Fig. 2D), reflecting the partial reprogramming state. We then examined the in vivo effect of SNEL expression in the pancreas. After Dox treatment, SNEL-inducible chimeric mice exhibited mCherry fluorescence in various organs, including the pancreas ( Supplementary  Fig. 2E, F). Notably, Dox treatment longer than 2 weeks resulted in the formation of ADM lesions in the pancreas (Fig. 2h, i). Consistently, CK19-positive regions often co-expressed amylase (Fig. 2j), indicating that ADM was similarly induced by SNEL expression. Moreover, the ADM lesions were hardly detectable at 1 week after Dox withdrawal, suggesting that the SNEL-induced ADM phonotype is reversible ( Supplementary Fig. 2G). Collectively, these results indicate that the transient expression of reprogramming factor cocktails induces reversible ADM phenotype.
In vivo reprogramming represses acinar cell enhancers. To investigate the molecular basis of reprogramming-mediated ADM formation, we first examined transcriptional changes of the pancreas during in vivo reprogramming. The pancreas of C-OSKM mice given Dox for 3 days was analyzed for global gene expressions by microarray. In accordance with histological evidence of ADM formation, the expression of acinar cell-related TFs such as Ptf1a and Mist1 (Bhlha15) was reduced, while the expression of CK19 (Krt19), a ductal cell-specific gene, was markedly increased (Fig. 3a). In addition, ADM-related genes such as Muc5ac, Tff1, and Plac8 were similarly upregulated ( Fig. 3a) 26,27 . To examine the initial transcriptional response after OSKM induction, we next treated C-OSKM mice with Dox via i.p. injection and examined gene expressions after 24 h. Notably, marked repression of Ptf1a and Mist1 was already detectable at 24 h after Dox treatment (Fig. 3b). Conversely, CK19 and ADMrelated genes were upregulated in these samples (Fig. 3b), suggesting that the silencing of acinar cell-related genes is coupled with the induction of ADM-related genes.
Previous studies demonstrated that silencing of the somatic cell-specific enhancers is an initial epigenetic event during reprogramming 5,6 . We next examined the effect of reprogramming factor expressions on the enhancers of pancreatic cells. ChIP-seq analysis for H3K27ac was performed using the pancreas of C-OSKM mice given Dox for 3 days. A perfusion fixation protocol was employed for cross-linking chromatin in vivo. Consistent with reduced expressions, H3K27ac deposition was significantly reduced at the enhancers of several acinar cell-related genes including Ptf1a, Mist1 (Bhlha15), Nr5a2, and Cpa1 ( Fig. 3c and Supplementary Fig. 3A) after Dox treatment, but it was not altered at the control Actb gene ( Supplementary  Fig. 3B). In contrast, H3K27ac deposition was not increased in upregulated genes, such as CK19 or in ADM-related genes ( Supplementary Fig. 3C), which supports the notion that the repression of acinar cell enhancers is an initial epigenetic regulation step after reprogramming factor expression. Similarly, Sox9, a key TF in ductal cells 22,28 , did not exhibit an enrichment of H3K27ac deposition ( Supplementary Fig. 3C). Furthermore, genome-wide analysis revealed that enhancers including super enhancers (SEs) 29,30 are significantly repressed after Dox treatment for 3 days ( Fig. 3d and Supplementary  Fig. 3D, E). Consistently, the number of enhancers and SEs were substantially decreased by Dox treatment for 3 days ( Fig. 3d and Supplementary Fig. 3E). Moreover, the expressions of genes linked with Dox OFF-specific SEs were decreased by Dox treatment, while genes linked with common-SEs or Dox ONspecific SEs were not repressed (Fig. 3e). Ptf1a promoter and enhancer remained unmethylated in the pancreas after 3 days of Dox treatment ( Supplementary Fig. 3F), indicating that DNA methylation was not involved in the Ptf1a repression. Notably, the Dox-induced reduction of H3K27ac deposition was not observed at 2 weeks after Dox withdrawal, indicating the repression of SEs is reversible (Fig. 3d). Collectively, these results suggest that reprogramming factor expression causes transient epigenetic repression of acinar cell enhancers and loss of acinar cell identity, which lead to ADM formation. To further connect the loss of acinar cell enhancer with ADM formation, we induced acinar-cell-related TFs (Mist1 or Ptf1a) together with OSKM in the pancreas (Supplementary Fig. 4A-C). Notably, the co-expression of Ptf1a or Mist1 with OSKM attenuated the OSKM-induced expression of CK19 and ADM formation ( Fig. 3f and Supplementary Fig. 4D). Together, these results suggest that the silencing of acinar cell program causes the induction of ADM.
Finally, to connect OSKM-induced ADM formation with pathological conditions, we examined the effect of caerulein treatment, which is well-established pancreatitis stimuli and induces transient ADM formation, on pancreatic SEs. We H3K27ac  Fig. 3g and Supplementary Fig. 4G). Taken together, our results suggest that inflammatory stimuli induce ADM formation through repression of somatic enhancers.
In vivo reprogramming activates ERK in Kras mutant pancreas. To elucidate the influence of the loss of acinar cell identity on Kras-mutated pancreatic cells, we next established ESCs containing LSL-Kras G12D , Pdx1-ires-Cre, Rosa LSL-rtTA3, and Col1a1::tetO-OSKM-ires-mCherry (hereafter KC-OSKM) and generated chimeric mice to induce Kras mutation and reprogramming factors simultaneously in the pancreas (Fig. 4a). The experimental protocol of Dox treatment for OSKM induction is shown in Fig. 4b. Consistent with our findings that OSKM expression induces ADM, increased expression of CK19 was observed in both C-OSKM mice and KC-OSKM mice (Fig. 4c). Notably, pERK was strongly induced in CK19-positive area of KC-OSKM mice whereas such induction was modest in C-OSKM mice (Fig. 4c), suggesting that Kras mutation cooperatively activates the ERK pathway in conjunction with reprogramming factor expression (Fig. 4c). In sharp contrast to reversible ductal metaplasia in C-OSKM mice, CK19-expressing ducts were observed even at 1 week after Dox withdrawal in KC-OSKM mice ( Fig. 4d), suggesting that the ADM phenotype is sustained in the presence of Kras mutation. Consistently, increased expression of pERK was sustained in the ductal area at 1 week after Dox withdrawal in KC-OSKM mice (Fig. 4e, f). Furthermore, ductal lesions were histologically displayed PanIN/PDAC (Fig. 4d, e, Supplementary Fig. 5A). The affected lesions sometimes displayed positive staining for Alcian blue, a PanIN marker, and often exhibited widespread Sirius red staining, which showed that massive fibrosis simultaneously occurs in PanIN/PDAC lesions within 10 days of Dox exposure (Fig. 4e, f, Supplementary  Fig. 5B).
Previous studies have demonstrated that increased expression of endogenous EGFR activates the MAPK pathway in human and mouse PanIN/PDAC area 31,32 . To unveil the molecular mechanism for persistent activation of the ERK signaling pathway in KC-OSKM mice, we examined EGFR expression in Dox-treated C-OSKM and KC-OSKM mice. Although Dox treatment for 3 days increased EGFR expression in both C-OSKM and KC-OSKM mice, the increased EGFR levels were pronounced in KC-OSKM mice, while only modest expression was observed in C-OSKM mice ( Fig. 4g and Supplementary Fig. 5C). Furthermore, the increased EGFR expression was sustained in KC-OSKM mice even after Dox withdrawal, whereas C-OSKM mice showed almost no EGFR staining ( Fig. 4g and Supplementary  Fig. 5C, 5D), suggesting that Kras mutation is involved in robust and continuous EGFR expression through positive feedback regulation. Western blot confirmed the sustained activation of the ERK signaling pathway and increased EGFR expression in the pancreas of KC-OSKM mice even after Dox withdrawal (Fig. 4h, Supplementary Fig. 9). These results suggest that maintenance of the acinar cell trait prevents ERK signaling activation in the pancreas of Kras mutant mice.
p53 mutation accelerates Kras/OSKM-induced PDAC. An inactivating mutation at the p53 gene was reported to promote PDAC formation 33,34 . To investigate the effect of p53 mutation on in vivo reprogramming-mediated development of pancreatic lesions in Kras mutant background, we next generated chimeric mice containing LSL-Kras G12D , LSL-p53 R172H , Pdx1-ires-Cre, Rosa LSL-rtTA3, and Col1a1::tetO-OSKM-ires-mCherry (hereafter KPC-OSKM) (Fig. 5a). The time course of Dox treatment is shown in Fig. 5b. Transient induction of OSKM in the pancreas of KPC-OSKM mice significantly increased the CK19-and pERK-positive area, which persisted even after withdrawal of Dox ( Supplementary Fig. 5E, F) as similarly observed in KC-OSKM mice. Notably, KPC-OSKM mice developed predominantly PDAC within 10 days of OSKM induction, while KC-OSKM mice developed mixed lesions containing both PanIN and PDAC (Fig. 5c-e). Consistently, Dox-treated KPC-OSKM mice showed a significant reduction of PanIN-related Alcian blue-positive area when compared with Dox-treated KC-OSKM mice (Supplementary Fig. 5G). Dox-treated KPC-OSKM mice died within 3 months, and some mice developed cancerous ascites (Supplementary Fig. 5H). Moreover, cell lines established from pancreatic tumors in Dox-treated KPC-OSKM mice were able to develop secondary tumors in the subcutaneous tissue of nude mice ( Supplementary Fig. 5I), affirming that they have cancerinitiating properties. Together, these results demonstrate that compound mutations of Kras and p53 in cooperation with transient repression of acinar cell enhancers are sufficient for the rapid and widespread induction of PDAC.
We further examined mechanistic insights into the phenotypic difference between KC-OSKM and KPC-OSKM mice after Dox exposure. Given that escape from Kras-induced senescence is associated with progression of PanIN and that the p21-p53 pathway is involved in oncogene-induced senescence 35 , we particularly focused on the effect of p53 R172H mutation on cellular senescence. Consistent with the previous report that found OSKM expression in vivo induces cellular senescence 36 , widespread DNA damage response, which was measured by γH2AX staining, as well as increased p21 expression were detected after Dox treatment in C-OSKM mice ( Supplementary  Fig. 5J). Consistently, an increased deposition of H3K27ac was observed in Cdkn1a (p21) after Dox treatment for 3 days in C-OSKM mice (Supplementary Fig. 5K). Notably, the induction of γH2AX and p21 was indistinguishable among C-OSKM, KC-OSKM, and KPC-OSKM mice ( Supplementary Fig. 5J), suggesting that oncogenic Kras mutation and p53 mutation hardly affect the immediate senescence-associated response against reprogramming factor expressions. After withdrawal of   Positive area (%) (13) (14)  subsequent cancer growth in Kras-mutated acinar cells of the pancreas. Finally, we tried to maintain acinar cell enhancers by expressing acinar cell-specific TFs (Ptf1a or Mist1) in caeruleintreated KC mice, in which caerulein-induced inflammation initiates ADM/PanIN formation and activates ERK signaling ( Fig. 5f and Supplementary Fig. 4E). Notably, forced expression of either Ptf1a or Mist1, but not the control mCherry, strongly inhibited PanIN formation ( Fig. 5g and Supplementary Fig. 7A). Consistently, pERK staining was significantly reduced by the expression of acinar cell-related TFs (Fig. 5h). These results provide additional evidence that acinar cell enhancers safeguard Kras-induced cancer initiation.
Undifferentiated cancer development by in vivo reprogramming.
It has been shown that the acquisition of the pluripotency-related epigenetic landscape occurs at late-stage reprogramming 5,6 . We next extended the duration of reprogramming factor expression in KC-OSKM mice to examine the effect of late epigenetic regulation of reprogramming on cancer development. KC-OSKM mice were given Dox for 1-2 weeks, followed by Dox withdrawal for 1-2 weeks. In contrast to the PanIN/PDAC formation in KC-OSKM mice given Dox for 3 days, prolonged treatment of Dox caused the development of histologically undifferentiated cancer with higher proliferative activity in the pancreas of KC-OSKM mice (Fig. 6a). Notably, these cancers exhibited the expression of pluripotency-related proteins, Sall4 and Lin28a, independently of OSKM transgene expressions (Fig. 6a).
We found that KC-OSKM mice given Dox for 1-2 weeks often developed undifferentiated cancers in the stomach (Fig. 6b and Supplementary Table 1), which is consistent with the observation that Cre-LoxP recombination also occurs in the pyloric glands of the stomach with Pdx1-ires-Cre allele ( Supplementary Fig. 1B, C). These gastric cancers similarly exhibited the expression of Sall4 and Lin28a (Fig. 6c). In agreement with the acquisition of pluripotency-related epigenetic regulation, partial activation of ESC-Core and ESC-Myc modules was observed in gastric cancers in KC-OSKM mice (Supplementary Fig. 8A). C-OSKM mice treated with Dox for 1-2 weeks also developed undifferentiated cancer in the stomach ( Supplementary Fig. 8B, C), albeit at lower frequency than KC-OSKM mice (p < 0.01, Fisher's exact test, Supplementary Table 1), suggesting that the Kras mutation promotes but is not essential for the development of undifferentiated cancers.
In vivo reprogramming induces AFP-producing cancers. Human AFP-producing cancers predominantly arise in the stomach and display poorly differentiated histology and show dismal prognosis 37,38 . Notably, we observed focal expression of Afp in gastric cancers in KC-OSKM mice given Dox for 1-2 weeks (Fig. 6c, d), which raised the possibility that the acquisition of pluripotency-related epigenetic regulation is associated with the development of human AFP-producing cancers. Therefore, we next investigated the expression of pluripotency-related proteins in human AFP-producing gastric cancers. Twenty-five gastric cancers that were diagnosed as AFP-producing gastric cancers, were examined for the expression of SALL4, LIN28A, and LIN28B by immunostaining. Notably, as observed in gastric cancers in KC-OSKM mice, SALL4, LIN28A, and LIN28B were frequently co-expressed in human AFP-producing gastric cancers (Fig. 6e, f and Supplementary Table 2), suggesting that these cancers exhibit activation of the pluripotency-related regulatory network.
Finally, we re-examined the microarray data previously used to investigate 200 cases of human gastric cancers 39 . We found that AFP, LIN28B, and SALL4 were overexpressed in seven, eight, and nine cases of the 200 human gastric cancers, respectively. Consistent with the frequent co-expression of pluripotencyrelated proteins in human AFP-producing gastric cancers, partial activation of ESC-Core and ESC-Myc modules was observed in AFP-, LIN28B-, and SALL4-expressing human gastric cancers ( Fig. 6g and Supplementary Fig. 8D), while such activations were not obvious in ERBB2-highly expressed gastric cancers (n = 12) ( Supplementary Fig. 8D), which recapitulated our observations of reprogramming-induced gastric cancers in KC-OSKM mice.

Discussion
Cancer is believed to arise through an accumulation of genetic mutations 40,41 . Indeed, genome-wide sequencing analyses have revealed a number of driver mutations in multiple cancer types. The functional significance of such driver mutations has been demonstrated by reverse genetic approaches where rodents in which these mutations were introduced develop the same types of cancer observed in human patients. However, the genetically engineered cancer models always exhibit a substantial latency period for cancer development, indicating that additional events are required for the development of full-blown cancer. Therefore, which sets of genetic mutations are sufficient for cancer development remains unclear in most cancers. Indeed, in the present study we show that most pancreatic cells with compound mutations at Kras and p53 display no obvious activation of the ERK signaling pathway and exhibit normal histology at 6 weeks of age, which is consistent with previous findings that oncogenic KRAS mutations are detectable in human disease-free pancreas 42 . Notably, we demonstrate that short, transient expression of reprogramming factors in Kras mutant mice at 4 weeks of age results in persistent ERK activation and widespread PDAC development within 10 days, indicating that cellular reprogramming is sufficient for the initiation of cancer growth in pancreatic cells with Kras mutation.
Alteration of the epigenetic modifications is one of the most common aberrations detected in human cancers [43][44][45] . Decades of studies have demonstrated that epigenetic regulation plays a significant role in cancer development both in vitro and in vivo. However, recent sequencing analysis identified that most cancer types harbor mutations at epigenetic modifier genes at varying frequencies, which has raised the possibility that genetic abnormalities are the primary cause of the dysfunction or dysregulation of epigenetic modifiers that result in secondary alterations in the cancer epigenome. Therefore, direct evidence of bona fide epigenetic alterations that drive cancer development is still limited, especially in vivo. iPSC derivation is mediated through epigenetic reorganization without affecting genomic information. In the initial stage of d   (11) Ptf1a (5) Mist1 (6) reprogramming in vitro, reprogramming factors repress somatic enhancers, which causes the repression of cell-type specific gene expressions 6 . Consistently, the gradual repression of somatic cellspecific gene expressions is similarly observed during in vivo reprogramming 8 . Notably, the early reprogramming process in vivo is reversible up to 7 days, which reflects epigenetic memory of the cells at early-stage reprogramming 8,46 . In the present study, we show that in vivo reprogramming primarily represses acinar cell enhancers and induces remarkable downregulation of acinar cell-related genes such as Ptf1a and Mist1, suggesting that in vivo reprogramming induces the loss of acinar identity. The fact that this loss is reversible further supports the notion that the process is an epigenetic event. Together, the rapid and widespread development of PDAC by 3-day expression of reprogramming factors provide in vivo evidence that transient alteration of epigenetic regulation plays a critical role on cancer initiation. Our results also highlight the significance of dedifferentiation-related epigenetic regulation in cancer initiation, which may provide mechanistic insight into how dedifferentiation is involved in cancer development.
Previous studies using genetically engineered mouse models show that adult ductal cells are refractory to Kras mutationmediated oncogenic transformation. In contrast, adult acinar cells are competent to give rise to PDAC in the presence of Kras mutation in which ADM plays a role as an initiating event [20][21][22] . Although ADM is associated with the initiation of Kras-mediated abnormal cell growth, the cause of ADM formation remains to be fully understood. Here we demonstrate that the repression of acinar cell enhancers causes ADM formation, which is consistent with previous findings in which the genetic ablation of acinar cellrelated TFs such as Ptf1a, Mist1, and Nr5a2 similarly leads to ADM in the pancreas [47][48][49] . Our results showing that the forced expression of Ptf1a or Mist1 partially prevents reprogramminginduced ADM formation further support the conclusion that the repression of acinar cell enhancers induces ADM formation. Notably, the transient ADM formation becomes irreversible and proceeds to PDAC in Kras mutant background, which is accompanied by continuous activation of the ERK signaling pathway. The results suggest that transient repression of acinar cell identity can initiate continuous cancer growth in Krasmutated pancreatic cells. Consistently, forced expression of acinar cell-related TFs (Ptf1a or Mist1) successfully inhibits Krasmediated activation of the ERK signaling pathway and ADM formation in a caerulein-induced pancreatitis model. Given that ADM is observed in human pancreatitis, a representative risk factor for PDAC [50][51][52] , and pancreatitis promotes PanIN/PDAC formation in mice [53][54][55] , epigenetic fluctuations provoked by inflammatory stimuli may initiate the repression of acinar cell enhancers, which results in continuous cancer cell growth in human pancreas. Collectively, we propose that the stable maintenance of acinar cell enhancers is a roadblock for the cancer cell growth of Kras-mutated pancreatic cells.
Gradual activation of the ESC regulatory network occurs during late-stage reprogramming 5,6 . Here we show that prolonged expression of reprogramming factors longer than 1 week results in the development of different types of cancer in the pancreas and stomach that is characterized by the expression of pluripotency-related proteins. Notably, these cancers resemble human AFP-producing cancer, one of the most fatal adult cancers 37,38,56 , suggesting that epigenetic reorganization related to late-stage reprogramming may be involved in such cancers. Consistently, human AFP-producing gastric cancers often coexpress pluripotency-related proteins, such as SALL4, LIN28A, and LIN28B. Furthermore, these gastric cancers exhibit similar activation patterns of ESC modules as pluripotent stem cells. Our results suggest that the activation of pluripotency-related genes via in vivo reprogramming may be involved in development of certain types of adult cancers.
In summary, taking advantage of in vivo reprogramming systems, we show that reprogramming-related epigenetic regulation has a profound impact on Kras-induced cancer. We propose that epigenetic fluctuations provoked by environmental factors may cause a transient dedifferentiation state, which is sufficient for continuous ERK signaling activation and cancer growth of quiescent pancreatic cells with Kras mutation.
Southern blotting. Genomic DNA was collected from ESCs or pancreatic tissues using the Purelink Genomic DNA mini Kit (Invitrogen). The recombinated BAC plasmids were collected using NucleoBond Xtra BAC Kit (Marchery-Nagel). The extraction was performed according to the manufacturer's instructions. Three nanograms of each recombinated BAC as a positive control and 5 μg of sample genomic DNA were digested with specific restriction enzymes (Pdx1-ires-Cre: ScaI, LSL-HA Kras G12D : NheI, LSL-Kras G12D : MfeI, LSL-p53 R172H : NheI) at 37°C overnight. The restriction enzymes were obtained from New England Biolabs. The restricted products were ethanol precipitated and then dissolved into 15 μL of TE buffer. The products were mixed with 5 mL of 6× Loading Buffer Triple Dye (NIPPON Gene), and electrophoresis was performed in 0.8% agarose gel (Fast Gene), 50 V for 2 h. The products in the gel were depurinated in 250 mM HCl for 10 min, washed in distilled water for 5 min, and denaturated in 0.5 M NaOH,  59 . One hundred microliters of 2i-ESC medium was added into each well every day. One week later, the colonies were passaged and expanded in ESC medium on the MEFs. Male ESCs were selected by genotyping Sry allele using Go taq Master Mix (Promega). Genotyping primers are described in Supplementary  Table 5.
Alkaline phosphatase staining of Dox-treated KH2-SNEL MEFs. MEFs from E13.5 of KH2-SNEL chimeric mice were cultured with 2.0 μg/mL of Dox in ESC medium for 2 weeks. ALP staining was performed according to the manufacturer's instructions using the ALP stain Kit (Muto Pure Chemicals).
X-gal staining. The precise procedure was described previously 61 . Whole mount staining: Mice under anesthesia were perfused through the vascular system with 10 mL of phosphate-buffered saline (PBS) to clear the blood. Perfusion was performed with 10 mL of 4% PFA to obtain adequate fixation. The dissected pancreas of Pdx1ires-Cre, Rosa LSL-LacZ mice were collected and fixed with ice-cold 4% PFA for 1 h. The tissues were poured into PBS containing 10% of sucrose, and then into PBS containing 20% of sucrose (Nacalai Tesque) for 1 h, respectively, and then placed in PBS containing 30% of sucrose at 4°C for overnight. Next the pancreas was poured into permeabilization solution (5 mM EGTA, 2 mM MgCl 2 , 0.01% sodium deoxycholate, 0.02% Nonidet P-40 in PBS) and reacted with the X-gal solution (1 mg/mL X-gal, 5 mM potassium ferrocyanide, and 5 mM potassium ferricyanide in the permeabilization solution) at 37°C for 6 h. Frozen tissue staining: After fixation of the dissected tissues with PBS containing 30% sucrose, the pancreas was embedded in Tissue-Tek O.C.T compound (Sakura). Frozen tissues were cut serially into 10-μm-thick sections and stained with the procedure described above. Counterstaining was performed using Nuclear Fast Red (Merck Millipore).
Doxycycline treatment. Doxycycline (Sigma) was dissolved in drinking water at 2 mg/mL. For quantitative PCR (qPCR) and ChIP-seq of the pancreas of C-OSKM mice, Dox-containing PBS was intraperitoneally injected to induce uniform transgene expressions (1 mg/injection every consecutive 12 h until sacrifice). To maintain acinar cell enhancer in Ptf1a or Mist1-inducible mice, we started Dox administration to induce Ptf1a or Mist1 24 h before caerulein treatment.
Caerulein treatment. The precise procedure was described previously 28  Paraffin-embedded specimen preparation. The dissected tissue samples were fixed in 4% PFA (Wako) for overnight at room temperature. The next day, the samples were poured into 70% ethanol (Wako) for several hours, and then poured into 80% ethanol, 90% ethanol, 100% ethanol, and xylene (Wako), and embedded in paraffin using a spin tissue processor (STP120; Thermo Scientific). Sections were sliced into 3-4 μm thickness, and stained with hematoxylin and eosin (Muto Pure Chemicals). Serial sections were used for the immunohistochemical analyses.
Immunostaining. The samples were soaked in xylene for 5 min three times to remove paraffin and 100% ethanol for 5 min three times to hydrophilize. After washing with water for several min, the samples were soaked in epitope retrieval buffer (Nichirei Bioscience) and microwaved at 100 W for 10 min. The samples were then soaked in 1× PBS for several min, and incubated with 150 μL primary antibodies in PBS with 2% bovine serum albumin (BSA) (MP Biomedicals) for 1 h at room temperature. The lists of antibodies and dilutions are shown in Supplementary Table 7. After washing with PBS for 5 min two times, the samples were processed with secondary antibodies for 30 min at room temperature. The samples were washed in PBS for 5 min two times, and nuclear staining was performed with hematoxylin for 10-30 s. After washing with water for 10 min, the samples were poured into 100% ethanol three times and placed in xylene three times. The samples were evaluated under a microscope (Olympus BX41).
Immunofluorescent staining. The same procedure as immunostaining described above was performed until primary antibodies treatment. The samples were processed with 500× DAPI (Invitrogen) and 150× fluorescence-labeled secondary antibodies diluted with 0.5% BSA PBS, for 90 min at room temperature. The lists of antibodies and dilutions are shown in Supplementary Table 7. After washing in PBS for 5 min two times, the samples were mounted and evaluated with a confocal laser scanning microscope (Zeiss LSM700).
Alcian blue staining. Tissue samples were soaked into xylene for 5 min three times, 100% ethanol for 5 min three times, washed with water for several minutes, and finally stained with Alcian blue 8GX solution (Merck Millipore) and nuclear fast red (Merck Millipore), according to the manufacturer's protocol.
Sirius red staining. Tissue samples were soaked in xylene for 5 min three times and 100% ethanol for 5 min three times, washed with water for several minutes, and stained with Picro-Sirius Red Stain Kit (PSI).
Microarray analysis. Two hundred nanograms of total RNA prepared from the pancreatic tissues was subjected to cDNA synthesis with a WT Expression Kit (Ambion), and the resultant cDNA was fragmented and hybridized to a Mouse Gene 1.0 ST Array (Affymetrix). After hybridization, GeneChip arrays were washed and stained using GeneChip Fluidics Station 450 (Affymetrix) and detected with the Scanner 3000 TG system (Affymetrix) following the manufacturer's standard protocols. The data were analyzed using GeneSpring GX software (version 14.8; Agilent Technologies).
Chromatin immunoprecipitation from pancreatic tissue. Perfusion fixation with ice-cold 4% PFA (Wako) was employed for fixation of the pancreatic tissue of C-OSKM mice. The pancreas was dissected after 5 min of fixation and treated with glycine to quench the formaldehyde. ChIP was performed as described previously 62 . H3K27Ac antibody (Monoclonal Antibody Lab) was used for the ChIP analysis, and mouse IgG antibody (Abcam) was used as a control.
ChIP-seq. Purified ChIP DNA (5 ng per sample) was used for the generation of sequencing libraries with TruSeq ChIP Sample Prep Kit (Illumina). The resultant DNA libraries were assessed on Agilent Bioanalyzer and quantified with KAPA Library Quantification Kits (KAPA BIOSYSTEMS). The libraries were sequenced to generate single-end 86 bp reads using NextSeq 500 (illumina).
ChIP-seq data analysis. ChIP-Seq Peak calls were made as previously described 63 . Briefly, the low-quality bases at the 3′ read ends and the adaptors were trimmed using cutadapt-1.9.1 (ref. 63 ). Untrimmed and trimmed reads of 20 bp or longer were mapped to mouse genome (mm10) with BWA 0.7.12 (ref. 64 ). Duplicate reads and low mapping quality reads (MAPQ < 20) were removed using Picard (http:// broadinstitute.github.io/picard/) and SAMtools 65 for further analysis. Mapped reads were visualized using igv-2.3 (ref. 66 ). Peaks were called using MACS2 (ref. 67 ) plus IDR framework developed by the ENCODE project 68 . The reads mapped to the blacklisted regions (https://sites.google.com/site/anshulkundaje/projects/ blacklists) and sex chromosomes were excluded for the peak calling. The heatmaps for ChIP-Seq data were drawn using the ngsplot-2.47 program 69 . Super enhancers were identified by H3K27ac with ROSE pipelines 29,30 . Super enhancers identified in this study were assigned to the transcripts whose TSS were within the SEs.
Bisulfite sequencing from pancreatic tissue. Freshly collected pancreatic tissues were frozen in liquid nitrogen and grounded into powder using mortar. Genome DNA of the pancreas was extracted using Purelink Kit (Invitrogen) according to the manufacturer's protocol. Bisulfite treatment was performed using EZ DNA Methylation-Gold Kit (Zymo Research). The target sequence was cloned by Ex Taq HS (Takara Bio) and ligated into pCR4-Topo vector using Topo cloning technology (Invitrogen). These plasmids were transformed into DH5α at 37°C, overnight. Colony PCR was performed using Go taq Master Mix (Promega) and ascertained ligation of the cassettes. After cleanup of the PCR products by exonuclease I (New England Biolabs) and shrimp alkaline phosphatase (Takara Bio), the sequence reaction was performed using M13 Rv primer (5′-CAGGAAA-CAGCTATGAC-3′) and was sequenced with ABI 3500 xL (Applied Biosystems). Methylation rates were measured using Quantification tool for Methylation Analysis (QUMA) software (Riken).
Quantification analysis. To quantify the immunostained tissue samples, immunostained section was randomly photographed at ×200 magnification three times for each mouse. The pictures were processed with ImageJ software (NIH) to evaluate the positive area of each CK19, phospho-ERK, Alcian blue, Sirius red, and EGFR staining. The positive area was determined as the mean of positive area in three independent histological images. The results were evaluated with Graphpad Prism 7 software for quantification. To assess γH2AX-, p21-, Ki67-, Caspase 3-positive rates, we randomly photographed epithelial area of the pancreas at ×200 magnification. The positive nucleus of epithelial cells was counted. The number of each positive nucleus was divided by the number of total nucleus of epithelial cells. For quantification of immunofluorescent samples, section was randomly photographed at ×200 magnification. The ratio of CK19-positive cells in EGFP-expressing cells was determined in ten independent immunofluorescent images for each cohort.
Statistical analysis of tissue samples. Statistical parameters including statistical significance and n values are described in the figures. Statistical analyses were carried out using Prism 7 Software (GraphPad). For statistical comparison of two groups, we performed Mann-Whiteney U-test (two-tailed, 95% confidence intervals). A box-and-whisker plot is used to display information about the range, the median and the quartiles. Solid lines in each box indicate the median. Bottom and top of the box are lower and upper quartiles, respectively. Whiskers extend to ±1.5 interquartile range (IQR). Numbers in parentheses indicate the number of mice examined. *p < 0.05, **p < 0.01, ***p < 0.001. n.s., not significant.
Patients-derived paraffin-embedded tissue samples. Human patients-derived paraffin-embedded tissue samples were used in compliance with ethical regulations in Kyoto University Hospital.
Statistical analysis of survival. Kaplan-Meier survival curves were calculated using the survival time for each mouse. The Log-rank test was used to test for significant differences between the groups using Prism 7 Software (GraphPad).
Data availability. ChIP-seq and microarray data reported in this study have been deposited to the Gene Expression Omnibus (GEO) under accession GSE100842.