Abstract

   SNAIL is a key transcriptional regulator in embryonic development and
   cancer. Its effects in physiology and disease are believed to be linked
   to its role as a master regulator of epithelial-to-mesenchymal
   transition (EMT). Here, we report EMT-independent oncogenic SNAIL
   functions in cancer. Using genetic models, we systematically
   interrogated SNAIL effects in various oncogenic backgrounds and tissue
   types. SNAIL-related phenotypes displayed remarkable tissue- and
   genetic context-dependencies, ranging from protective effects as
   observed in KRAS- or WNT-driven intestinal cancers, to dramatic
   acceleration of tumorigenesis, as shown in KRAS-induced pancreatic
   cancer. Unexpectedly, SNAIL-driven oncogenesis was not associated with
   E-cadherin downregulation or induction of an overt EMT program.
   Instead, we show that SNAIL induces bypass of senescence and cell cycle
   progression through p16^INK4A-independent inactivation of the
   Retinoblastoma (RB)-restriction checkpoint. Collectively, our work
   identifies non-canonical EMT-independent functions of SNAIL and unravel
   its complex context-dependent role in cancer.

   Subject terms: Cancer genetics, Oncogenes, Cancer models, Senescence,
   Checkpoints
     __________________________________________________________________

   SNAIL promotes tumour metastasis through inducing epithelial to
   mesenchymal transition (EMT). Here the authors report that SNAIL
   bypasses senescence and regulates cell cycle progression to promote
   pancreatic carcinogenesis and this is independent of EMT induction.

Introduction

   SNAIL is overexpressed in > 70% of human pancreatic ductal
   adenocarcinomas (PDAC)^[76]1 and a wide number of other tumour
   entities, such as intestinal, breast, lung and liver
   cancer^[77]2–[78]6. SNAIL expression is believed to be a key driver of
   tumour aggressiveness and metastasis formation via the induction of an
   epithelial-to-mesenchymal transition (EMT) program and the subsequent
   acquisition of stem cell-like features^[79]5,[80]7–[81]12. Accordingly,
   it is often correlated with poor prognosis and shortened survival of
   cancer patients. However, the specific in vivo functions of SNAIL and
   the role of EMT during tumour progression in different tumour types
   remain largely unexplored^[82]10,[83]13–[84]16. In an in vivo selection
   model of highly metastatic PDAC cells, we have demonstrated previously
   that SNAIL drives EMT and subsequently metastasis formation^[85]17.
   This runs counter to recent findings in an autochthonous mouse model of
   PDAC showing that Snail deletion does not influence the metastatic
   phenotype, but sensitizes tumours to chemotherapy^[86]13. This prompted
   us to re-investigate and mechanistically probe the function of SNAIL in
   vivo by systematic and comprehensive genetic gain- and loss-of-function
   approaches using a variety of disease-relevant genetically engineered
   autochthonous mouse models as well as human cancers.

   Here, we show context-dependent oncogenic functions of the
   transcriptional regulator SNAIL in cancer, which are independent of its
   role as a regulator of the EMT process. SNAIL-induced phenotypes depend
   on both, the genetic context of the tumour and its tissue of origin,
   ranging from protective effects that delay tumour onset in intestinal
   cancer models to a dramatic acceleration of pancreatic cancer
   development and aggressiveness. Mechanistically, we demonstrate that
   SNAIL acts as transcriptional activator that bypasses oncogenic
   KRAS-induced senescence and drives the cell cycle by
   p16^INK4A-independent inactivation of the Retinoblastoma
   (RB)-restriction checkpoint of senescence, thereby inducing
   context-dependent cancer progression. This knowledge provides
   opportunities to target SNAIL-driven cancers, which are a major
   clinical problem due to their high aggressiveness and lethality.

Results

SNAIL-driven cancer progression is highly context-specific

   To investigate the function of SNAIL in different cancer types in vivo,
   we created a latent Snail allele silenced by a lox-stop-lox (LSL)
   cassette as a knock-in (KI) at the mouse Rosa26 locus (LSL-R26^Snail/+
   mouse line, termed Snail^KI/+; Supplementary Fig. [87]S1a–c). SNAIL
   expression was then activated in several genetically engineered murine
   autochthonous cancer models: i) a PDAC model that depends on
   Cre-induced expression of oncogenic KRAS^G12D in the Ptf1a lineage of
   the pancreas (Ptf1a^Cre/+;LSL-Kras^G12D/+, termed PKras^G12D/+); ii) a
   classical WNT-driven intestinal cancer model, induced by loss of the
   tumour suppressor adenomatosis polyposis coli (Apc) due to Cre-mediated
   deletion of a floxed Apc allele in intestinal epithelial cells
   (Villin-Cre;Apc^lox/+, termed VApc^ΔInt); and iii) two different models
   of serrated intestinal cancer driven by either oncogenic KRAS^G12D or
   BRAF^V637E, based on Villin-Cre-induced activation of latent oncogenic
   Kras^G12D (Villin-Cre;LSL-Kras^G12D/+ termed VKras^G12D/+) or
   Braf^V637E (Villin-Cre;LSL-Braf^V637E/+, termed VBraf^V637E/+). In this
   way we mimicked the acquisition of SNAIL expression in different tumour
   types and subtypes driven by distinct oncogenes and signalling
   pathways.

   Concomitant transgenic expression of SNAIL and activation of oncogenic
   KRAS^G12D in the pancreas of
   Ptf1a^Cre/+;LSL-Kras^G12D/+;LSL-R26^Snail/+ mice (termed
   PKras^G12D/+;Snail^KI/+) accelerated the formation of acinar to ductal
   metaplasia (ADM) and PDAC precursor lesions (pancreatic intraepithelial
   neoplasia (PanIN)), and substantially increased cancer development
   (Fig. [88]1a–h and Supplementary Fig. [89]S1d–f). Consistent with this,
   PDAC gene set enrichment was already apparent in one-month old mice
   with aberrant SNAIL expression (Fig. [90]1f). All animals in the tumour
   watch cohort developed PDAC with a median survival of 190 days,
   compared to 465 days in PKras^G12D/+ animals. Biallelic Snail
   expression in Ptf1a^Cre/+;LSL-Kras^G12D/+,LSL-R26^Snail/Snail mice
   (termed PKras^G12D/+;Snail^KI/KI) drastically reduced survival further
   to a median of only 64 days (Fig. [91]1g, h).

Fig. 1. Aberrant SNAIL expression dramatically accelerates KRAS^G12D-driven
PDAC formation.

   [92]Fig. 1
   [93]Open in a new tab

   a Genetic strategy to activate SNAIL and KRAS^G12D expression in the
   pancreas. b Immunoblot analysis of SNAIL protein expression in pancreas
   of 1-month-old Ptf1a^Cre/+;LSL-Kras^G12D/+ (PKras^G12D/+) and
   Ptf1a^Cre/+;LSL-Kras^G12D/+;LSL-R26^Snail/+ (PKras^G12D/+;Snail^KI/+)
   compound mutant mice. c Representative hematoxylin and eosin (H&E),
   Alcian blue (AB), Muc5a, CK19 and BrdU stains of acinar to ductal
   metaplasia (ADM) and pancreatic intraepithelial neoplasia (PanIN) in
   1-month-old PKras^G12D/+ and PKras^G12D/+;Snail^KI/+ mice. Note the
   almost complete remodelling of pancreatic tissue in
   PKras^G12D/+;Snail^KI/+ animals. Scale bars, 50 μm. d Quantification of
   ADM and PanIN progression in % of total lesions at age of one-month
   (error bars, mean ± SEM; n = 4 per genotype; 3 representative slides
   per mouse; *p = 0.037, Mann-Whitney two-tailed test). e Quantification
   of ductal and acinar structures after in vitro culture of acinar
   explants for 5 days (n = 3 control/PKras^G12D/+ mice, n = 5
   PKras^G12D/+;Snail^KI/+ mice; mean ± SEM; *p = 0.036, Mann-Whitney
   two-tailed test). f Gene-set enrichment analysis (GSEA) using mRNA
   expression profiles of PKras^G12D/+;Snail^KI/+ (red) and PKras^G12D/+
   (blue) pancreata of 1-month-old mice (n = 2 per genotype) computed and
   corrected for multiple testing using the Benjamini–Hochberg procedure
   (for statistical details, please see methods section). Genes were
   ranked using Signal-to-Noise ratio statistics according to their
   correlation. Vertical black lines mark the position of each gene in the
   data set. Normalized Enrichment Score: 3.15; Nominal p-value < 0.0001;
   False Discovery Rate (FDR) q-value < 0.0001. g Kaplan-Meier survival
   curves of PKras^G12D/+ (n = 125; 465 days median survival),
   PKras^G12D/+;Snail^KI/+ (n = 42; 190 days median survival) and
   PKras^G12D/+;Snail^KI/KI (n = 28; 64 days median survival) mice
   (***p < 0.0001, log-rank test, Bonferroni correction). h Macroscopic
   view and representative H&E and BrdU staining of pancreata with PDAC of
   three endpoint mice per genotype (tu tumour; st stomach; sp spleen).
   Scale bars, 50 µm. Note: The PKras^G12D/+ cohort in (g) is the same
   shown in Figs. [94]4b, [95]6e, [96]f and the PKras^G12D/+;Snail^KI/+
   cohort is the same shown in Figs. [97]6e, [98]l. Source data of Fig. 1
   are provided in the Source Data file.

   To assess the impact of SNAIL on tumour development, progression and
   survival of intestinal cancer, where SNAIL is aberrantly expressed in
   78% of cases^[99]18, we used three different genetically engineered
   models that mirror major histopathological and molecular disease
   subtypes^[100]19–[101]21 (Fig. [102]2 and Supplementary Fig. [103]S2).
   Surprisingly, in the classical Apc loss-of-function model (VApc^ΔInt)
   that is driven by activation of canonical WNT signalling^[104]20, there
   was a trend towards prolonged survival (median survival of 419 vs. 355
   days; p = 0.12) and reduced number of adenomas and carcinomas per
   animal upon aberrant SNAIL expression (Fig. [105]2a–e; p = 0.08 and
   0.14, respectively). SNAIL had no major effect on tumour morphology and
   histopathology (Fig. [106]2f). This stands in sharp contrast to the
   dramatic pro-tumorigenic effects observed in the KRAS-driven PDAC model
   and suggests that SNAIL has context-specific functions in different
   tumour entities and/or oncogenic backgrounds. To test the hypothesis
   that SNAIL cooperates specifically with KRAS, but not WNT pathway
   activation, to induce cancer progression across tumour entities, we
   activated SNAIL in vivo in a KRAS^G12D-driven model of serrated
   intestinal cancer (Fig. [107]2g). This specific CRC subtype is
   characterized by a serrated histopathological morphology and progresses
   through a hyperplasia - serrated adenoma - serrated carcinoma sequence
   distinct from the classical WNT-driven CRC progression model described
   by Vogelstein and colleagues^[108]22, which is characterized by an
   adenoma-carcinoma sequence without hyperplasia and serrated
   morphology^[109]23,[110]24. Unexpectedly, aberrant SNAIL expression
   failed to accelerate oncogenesis. Instead, we observed a weak trend
   towards prolonged survival in the KRAS-driven intestinal cancer model
   (median survival of 502 days (VKras^G12D/+;Snail^KI/+) vs 354 days
   (VKras^G12D/+); p = 0.29; Fig. [111]2h, i). Accordingly, we detected no
   obvious change in the number of adenomas and carcinomas per animal, as
   well as in the grading of the tumours (Fig. [112]2j–n). These data
   indicate that the tissue of origin and not the oncogenic driver
   dictates the functional role of SNAIL in tumour progression in vivo.

Fig. 2. SNAIL does not promote classical APC loss-of-function and serrated
KRAS^G12D-driven intestinal cancer progression.

   [113]Fig. 2
   [114]Open in a new tab

   a Strategy to activate Snail expression in intestinal epithelium in the
   Villin-Cre;Apc^lox/+ (termed VApc^ΔInt) model of intestinal cancer. b
   Kaplan-Meier survival curves of Villin-Cre;Apc^lox/+;LSL-R26^Snail/+
   (termed VApc^ΔInt;Snail^KI/+; n = 10, median survival 419 days) and
   VApc^ΔInt mice (n = 8, median survival 355 days); p = 0.12, log-rank
   test. c qRT-PCR of Snail mRNA expression normalized to Cyclophilin A in
   intestinal tumours of VApc^ΔInt (n = 3) and VApc^ΔInt;Snail^KI/+
   (n = 4) endpoint mice (Mean ± SEM; **p = 0.0035, unpaired two-tailed
   t-test with We’ch’s correction). d Number of adenomas in VApc^ΔInt
   (n = 8) and VApc^ΔInt;Snail^KI/+ (n = 9) endpoint mice. Mean ± SEM,
   p = 0.08, Mann-Whitney two-tailed test. e Percentage of
   carcinoma-bearing mice (left) and carcinoma number (right) in VApc^ΔInt
   (n = 8) and VApc^ΔInt;Snail^KI/+ (n = 9) endpoint mice. Left,
   two-tailed Fisher’s exact test, p > 0.9999; right, Mann-Whitney
   two-tailed test, mean ± SEM, p = 0.14. f Representative H&E and Ki67
   staining of invasive intestinal carcinoma of three VApc^ΔInt and
   VApc^ΔInt;Snail^KI/+ mice. g Strategy to express Snail in intestinal
   epithelium in the Villin-Cre;LSL-Kras^G12D/+ model (termed
   VKras^G12D/+) of intestinal cancer. h Kaplan-Meier survival curves of
   VKras^G12D/+ (n = 5, median survival 354 days) and
   Villin-Cre;Kras^G12D/+;LSL-R26^Snail/+ (termed VKras^G12D/+;Snail^KI/+)
   mice (n = 5, median survival 502 days); p = 0.29, log-rank test. i
   qRT-PCR of Snail mRNA expression in the intestine of VKras^G12D/+
   (n = 5) and VKras^G12D/+;Snail^KI/+ endpoint mice (n = 3). Mean ± SEM,
   **p = 0.005, unpaired two-tailed t-test with Welch’s correction. j
   Number of adenomas in VKras^G12D/+ (n = 5) and VKras^G12D/+;Snail^KI/+
   (n = 5) endpoint mice. Mean ± SEM, Mann-Whitney two-tailed test;
   p = 0.795 (ns, not significant). k Representative H&E and Ki67 staining
   of intestinal adenoma in three VKras^G12D/+ and Vkras^G12D/+;Snail^KI/+
   mice. l Percentage of carcinoma-bearing mice (left) and carcinoma
   number (right) in VKras^G12D/+ (n = 5) and VKras^G12D/+;Snail^KI/+
   (n = 5) endpoint mice. Left, two-tailed Fisher’s exact test
   (p > 0.9999); right, two-tailed Student’s t-test (p = 0.72),
   mean ± SEM. m Representative H&E staining of intestinal carcinoma in
   two VKras^G12D/+ and three VKras^G12D/+;Snail^KI/+ mice. n Pathological
   grading of intestinal carcinomas from VApc^ΔInt (n = 44),
   VApc^ΔInt;Snail^KI/+ (n = 21), VKras^G12D/+ (n = 2),
   VKras^G12D/+;Snail^KI/+ (n = 4), VBraf^V637E/+ (n = 8) and
   VBraf^V637E/+;Snail^KI/+ (n = 6) mice; two-tailed Fisher’s exact test.
   FC, fold change; ns, not significant; scale bars, 50μm for all images.
   Source data are provided in the Source Data file.

   In the Braf^V637E-driven model of serrated intestinal cancer, we
   observed more invasive carcinomas in animals with aberrant SNAIL
   expression (Villin-Cre;LSL-Braf^V637E/+;Snail^KI/+; termed
   VBraf^V637E/+;Snail^KI/+) than in VBraf^V637E/+ animals, and also more
   adenomas per animal (Supplementary Fig. [115]S2a–f). Median survival
   was reduced to 392 days in the VBraf^V637E/+;Snail^KI/+ cohort vs. 481
   days in VBraf^V637E/+; mice (Supplementary Fig. [116]S2b; p = 0.0321).
   However, there was no overt consistent change in the grade of
   carcinomas observed towards more undifferentiated tumours in any of the
   three intestinal cancer models (Fig. [117]2f, m, n and Supplementary
   Fig. [118]S2f).

   These observations support the notion that SNAIL acts as a classical
   cancer-promoting oncogene particularly in KRAS-driven pancreatic and to
   a far lesser extent in BRAF-driven intestinal cancer, demonstrating its
   context-specific functions in cancer.

Snail activation fails to repress E-cadherin and does not induce an EMT
program in PDAC

   SNAIL is a master regulator of EMT that is associated with cancer
   aggressiveness, metastasis and decreased patient survival in various
   cancer types, such as PDAC^[119]5,[120]7,[121]14,[122]25,[123]26. We
   therefore analysed the effect of Snail activation on morphological and
   transcriptional EMT readouts in our autochthonous cancer models in
   vivo.

   H&E, E-cadherin (Cdh1) and CK19 stainings revealed differentiated and
   undifferentiated cancers regardless of genotype in our three different
   PDAC models (Fig. [124]3a, b and Supplementary Fig. [125]3a–c).
   However, histopathological grading indicated a trend towards less
   undifferentiated/sarcomatoid cancers (Grade 4) in the
   PKras^G12D/+;Snail^KI model in a Snail gene dose-dependent fashion
   (Fig. [126]3a and Supplementary Fig. [127]S3a–c). Morphologically,
   PKras^G12D/+;Snail^KI/+ and PKras^G12D/+;Snail^KI/KI mice displayed a
   phenotype characterized by budding of epithelial tumour cells with the
   formation of small solid tumour cell nests (Supplementary
   Fig. [128]S3b, c). These areas retained epithelial differentiation,
   visible as CK19 and E-cadherin positivity, adjacent to tubular ductal
   structures (Supplementary Fig. [129]S3b). Aberrant SNAIL expression
   therefore did not drive PDAC development to a more undifferentiated or
   sarcomatoid phenotype.

Fig. 3. SNAIL does not induce epithelial to mesenchymal transition (EMT) in
PDAC.

   [130]Fig. 3
   [131]Open in a new tab

   a Grading of PKras^G12D/+ (n = 32), PKras^G12D/+;Snail^KI/+ (n = 19)
   and PKras^G12D/+;Snail^KI/KI PDAC mice (n = 17). Grade
   4=undifferentiated/sarcomatoid^[132]86. b Representative staining of
   SNAIL and E-cadherin in PDAC sections of endpoint mice (n = 3 per
   genotype). c E-cadherin western blot of pancreas of 1-month-old mice
   (n = 2 per genotype). d qRT-PCR of Cdh1 mRNA expression normalized to
   Cyclophilin A (CypA) in pancreas of 1-month-old mice (Control, n = 3;
   PKras^G12D/+ n = 5; PKras^G12D/+;Snail^KI/+ n = 6;
   PKras^G12D/+;Snail^KI/KI n = 4). Mean ± SEM, unpaired two-tailed t-test
   with Welch’s and Bonferroni correction. e qRT-PCR of Snail (left) and
   Cdh1 (right) mRNA expression of PDAC cells with or without transgenic
   Snail expression (epi, epithelial (n = 27); mes, mesenchymal (n = 11)).
   Each dot represents one PDAC cell line. Mean ± SEM. ***p = 0.0005,
   unpaired two-tailed t-test with Welch’s correction. f Percentage of
   PKras^G12D/+ (n = 20), PKras^G12D/+;Snail^KI/+ (n = 12) and
   PKras^G12D/+;Snail^KI/KI (n = 8) PDAC cell lines with indicated
   morphology. g E-cadherin immunocytochemistry (green) of PDAC cells with
   or without transgenic Snail expression (n = 3 independent experiments).
   DAPI counterstain (blue). h SNAIL and E-cadherin western blot of
   PKras^G12D/+;Snail^KI/+ (n = 2), PKras^G12D/+;Snail^KI/KI (n = 1)
   (left) and Snail-transduced (RCAS-TVA system) PDAC cells (right)
   (n = 1). i PDAC cells from PKras^G12D/+, PKras^G12D/+;Snail^KI/+ and
   floxed PKras^G12D/+;Snail^KO/KO knock-out (KO) mice (n = 2 per
   genotype) treated with TGFβ for 72 h. j Total liver (left) and lung
   (right) metastasis rate of endpoint PKras^G12D/+ (n = 16),
   PKras^G12D/+;Snail^KI/+ (n = 17) and PKras^G12D/+;Snail^KI/KI (n = 17)
   PDAC mice. **p = 0.01 (liver) and p = 0.005 (lung), two-tailed Fisher’s
   exact test with Bonferroni correction. Nd. Not detected. k
   Representative H&E, CK19 and Ki67 staining of liver and lung metastases
   of three PKras^G12D/+;Snail^KI/+ mice. l Representative E-cadherin
   staining of intestinal tumours of VApc^ΔInt and VApc^ΔInt;Snail^KI/+
   endpoint mice (n = 3 per genotype). m qRT-PCR of Cdh1 mRNA expression
   in intestinal tumours of VApc^ΔInt (n = 3) and VApc^ΔInt;Snail^KI/+
   (n = 4) endpoint mice. Mean ± SEM, unpaired two-tailed t-test with
   Welch’s correction. n–o qRT-PCR of Cdh1 mRNA expression in colon
   samples of (n) VKras^G12D/+ (n = 5) and VKras^G12D/+;Snail^KI/+
   (n = 3), and (o) VBraf^V637E/+ (n = 5) and VBraf^V637E/+;Snail^KI/+
   (n = 7) mice. Mean ±SEM, unpaired two-tailed t-test with Welch’s
   correction. FC Fold change, ns not significant; scale bars 50 μm.
   Source data are provided in the Source Data file.

   SNAIL is known to repress the transmembrane glycoprotein CDH1
   (E-cadherin) in several contexts, thereby inducing EMT, migration,
   invasion and metastasis^[133]7,[134]27. Surprisingly, expression of
   SNAIL had no effect on CDH1 expression in vivo, nor in cultured primary
   PDAC cell isolates in vitro (Fig. [135]3b–e). E-cadherin protein and
   mRNA expression levels were comparable in cells with and without
   transgenic Snail expression (Fig. [136]3b–e). There was a marked
   decrease in the proportion of mesenchymal, and an increase in the
   epithelial phenotype in tumour cell lines from the
   PKras^G12D/+;Snail^KI/+ model compared to PKras^G12D/+ controls
   (Fig. [137]3f and Supplementary Fig. [138]S3d). Thus, there was no
   shift to a mesenchymal phenotype in the Snail-transgenic cell lines,
   not even with biallelic Snail expression, although we observed
   a > 5-fold higher expression of Snail in mesenchymal vs. epithelial
   PDAC cells without transgene (Fig. [139]3e, f and Supplementary
   Fig. [140]S3d). Furthermore, global mRNA expression profiles revealed
   no EMT-related signatures and epithelial cancer cell isolates from
   PKras^G12D/+;Snail^KI transgenic mice clustered exclusively with
   epithelial PDAC cells from PKras^G12D/+-driven models, irrespectively
   of the Trp53 mutational status (Supplementary Fig. [141]S3e). To
   activate SNAIL expression in established epithelial PDAC cells, we
   transduced them with a retroviral Snail expression cassette. E-cadherin
   localization, expression levels, and cell morphology remained unchanged
   (Fig. [142]3g, h, Supplementary Fig. [143]S3f), indicating that SNAIL
   expression alone is not sufficient to induce an overt EMT program in
   PDAC cells. To investigate whether SNAIL expression influences the
   ability of PDAC cells to undergo EMT, epithelial PDAC cells isolated
   from PKras^G12D/+, PKras^G12D/+;Snail^KI/+ and from conditional
   pancreas-specific Snail knock-out (KO) mice (PKras^G12D/+;Snail^KO/KO;
   see also Fig. [144]4a–d) were treated with the strong EMT-inducer TGFβ.
   All TGFβ-treated cells, regardless of genotype, underwent rapid EMT and
   displayed mesenchymal morphology (Fig. [145]3i), providing genetic
   evidence that SNAIL is dispensable for EMT-induction.

Fig. 4. Downregulation of Cdh1 expression drives undifferentiated PDAC
formation.

   [146]Fig. 4
   [147]Open in a new tab

   a Upper panel: Genetic strategy to conditionally delete a floxed Snail
   allele (Snail^lox) in the pancreas of Kras^G12D expressing mice. Lower
   panel: Genotyping PCR to test Snail-deletion (Snail-KO) using DNA from
   PDAC cells (cells) and tumour tissue with non-recombined stroma (tu) of
   PKras^G12D/+ and PKras^G12D/+;Snail^KO/KO mice (n = 2 per genotype).
   Lower left panel: Floxed Snail allele (fl): 480 bp, Snail WT allele
   (WT): 395 bp, deleted Snail allele: no band. Lower right panel: deleted
   Snail allele (del): 492 bp; floxed Snail and WT allele: no band. b
   Kaplan-Meier survival curves of indicated genotypes of
   PKras^G12D/+;Snail^KO/KO mice (n = 8; median survival 380 days),
   compared to PKras^G12D/+ (n = 125; median survival 465 days). ns, not
   significant, log-rank test. c Representative H&E-stained PDAC tissue
   sections of PKras^G12D/+;Snail^KO/KO mice with undifferentiated (upper
   panel) and differentiated (lower panel) morphology (n = 6). d
   Pathological grading of PdACs of PKras^G12D/+ (n = 32) and
   PKras^G12D/+;Snail^KO/KO mice (n = 6). e Genetic strategy to
   conditionally delete Cdh1 and express Snail in the pancreas of
   PKras^G12D/+ mice. f Kaplan-Meier survival curves of indicated
   genotypes of Pdx1-Cre;Kras^G12D/+;Snail^KI/+;Cdh1^KO/+ (n = 8; median
   survival 78 days), compared to Pdx1-Cre;Kras^G12D/+;Snail^KI/+ mice
   (n = 8; median survival 166 days). ***p = 0.0008, log rank test. g qPCR
   analysis of Cdh1 mRNA expression in PDACs of
   Pdx1-Cre;Kras^G12D/+;Snail^KI/+;Cdh1^KO/+ (n = 5) and
   Pdx1-Cre;Kras^G12D/+;Snail^KI/+ (n = 3) endpoint mice. Cdh1 mRNA levels
   were normalized to Cyclophilin A. Mean ± SEM, *p = 0.036, Mann-Whitney
   two-tailed test. FC, fold change. h Representative H&E-stained PDAC
   tissue sections of indicated Pdx1-Cre;Kras^G12D/+;Snail^KI/+ and
   Pdx1-Cre;Kras^G12D/+;Snail^KI/+;Cdh1^KO/+ mice with differentiated and
   undifferentiated morphology (n = 8 per genotype). i Pathological
   grading of PDACs in Pdx1-Cre;Kras^G12D/+;Snail^KI/+ (n = 8) and
   Pdx1-Cre;Kras^G12D/+;Snail^KI/+;Cdh1^KO/+ mice (n = 8). Note: The
   PKras^G12D/+ cohort in (b) is the same shown in Figs. [148]1g, [149]6e,
   [150]f. Source data of Fig. 4 are provided in the Source Data file.

   Beside the unchanged tumour differentiation status, what is also
   surprising is that aberrant SNAIL expression did not increase
   metastasis into liver and lung. Indeed, heterozygous transgenic
   PKras;Snail^KI/+ mice had a trend towards reduced metastases, and
   biallellic PKras;Snail^KI/KI mice significantly fewer (Fig. [151]3j,
   k). The capacity of autochthonous tumours to metastasize to liver and
   lung is thus independent of SNAIL expression, as demonstrated
   earlier^[152]13.

   Results from the three different models of intestinal cancer were
   notably similar. Adenomas and carcinomas were mainly well
   differentiated and showed no histopathological signs or features of EMT
   induction regardless of SNAIL expression. Consistent with this,
   E-cadherin expression was retained in SNAIL-expressing adenomas and
   carcinomas (Fig. [153]3l–o), indicating that aberrant SNAIL expression
   is insufficient to drive a full EMT program in the intestinal
   epithelium.

   To validate our findings in genetic loss-of-function models in vivo, we
   deleted Snail in KRAS-driven PDAC using the Cre/loxP system
   (Fig. [154]4a). As previously demonstrated^[155]13, this did not
   significantly alter PDAC development. Snail knock-out animals displayed
   similar tumour burden and overall survival to control mice
   (Fig. [156]4b–d). Importantly, and in line with our previous findings,
   loss of Snail did not drive PDAC to a well-differentiated epithelial
   phenotype nor block development of undifferentiated cancers that had
   already undergone an EMT program (Fig. [157]4c, d). Deletion of Snail
   does not therefore block EMT in vitro or in vivo.

   Using a genetic approach to assess E-cadherin/Cdh1 function in the
   aberrant Snail expression model, we knocked out one floxed allele of
   the Cdh1 tumour suppressor in Pdx1-Cre;Kras^G12D/+;Snail^KI/+ mice
   (Fig. [158]4e). E-cadherin mRNA levels were reduced in
   Pdx1-Cre;Kras^G12D/+;Snail^KI/+;Cdh1^lox/+ animals, which showed
   dramatically shortened median survival (78 days) compared to Pdx1-Cre;
   Kras^G12D/+;Snail^KI/+ (166 days) mice (Fig. [159]4f, g) and a clear
   shift of the tumours towards an undifferentiated mesenchymal phenotype
   (Fig. [160]4h, i). These data reveal that E-cadherin expression and
   function is independent of SNAIL in PDAC, and suppresses tumour
   progression and mesenchymal transition in vivo.

SNAIL bypasses senescence during pancreatic carcinogenesis

   To discern how aberrant SNAIL expression promotes rapid tumour
   progression in the PKras^G12D/+;Snail^KI/+ PDAC model, independent of
   overt EMT induction, we investigated early tumour barriers and events
   in tumour formation. Oncogene-induced senescence (OIS), a feature of
   KRAS-driven premalignant PanIN lesions of the pancreas^[161]28,[162]29,
   is a cellular stress response, which blocks proliferation so protecting
   cells from neoplastic transformation^[163]30. As reported previously,
   nearly all PanINs of the PKras^G12D/+ model displayed positive
   senescence-associated β-galactosidase (SA-β-gal) staining, the most
   reliable marker of OIS in the pancreas^[164]31. In contrast, we
   observed a dramatically reduced rate of OIS in PKras^G12D/+;Snail^KI/+
   mice (Fig. [165]5a, b), which displayed an almost complete loss of OIS
   in premalignant PanIN lesions (Fig. [166]5b). In addition, we observed
   loss of senescence-associated gene sets controlled by the
   retinoblastoma (RB) tumour suppressor gene in expression profiles of
   PanIN bearing PKras^G12D/+;Snail^KI/+ pancreata (Fig. [167]5c). In
   contrast, deletion of Snail in the PKras^G12D/+ model
   (PKras^G12D/+;Snail^KO/KO) induced a strong senescence phenotype as
   evidenced by SA-β-gal staining of PanIN bearing pancreatic tissue
   sections (Fig. [168]5d), indicating that SNAIL is indeed capable of
   impacting OIS.

Fig. 5. SNAIL bypasses senescence to drive pancreatic carcinogenesis.

   [169]Fig. 5
   [170]Open in a new tab

   a Representative images of senescence-associated β-galactosidase
   (SA-β-gal) staining of pancreata with PanIN lesions of PKras^G12D/+
   (n = 13) and PKras^G12D/+;Snail^KI/+ (n = 12) mice. Scale bars, 50 μm.
   LG, low grade; HG, high grade. b Quantification of SA-β-gal-stained
   PanIN lesions from PKras^G12D/+ (n = 13) and PKras^G12D/+;Snail^KI/+
   (n = 12) mice. Mean ± SEM, *p < 0.0001, Mann-Whitney two-tailed test.
   LG, low grade; HG, high grade. c Gene set enrichment analysis (GSEA) of
   mRNA expression profiling of 1-month-old mice (n = 2 per genotype)
   computed and corrected for multiple testing using the
   Benjamini–Hochberg procedure (for statistical details, please see
   methods section) shows significant enrichment of Rb1 targets senescent
   genes (CHICAS_RB1_TARGETS_SENESCENT) in PKras^G12D/+;Snail^KI/+ (red)
   vs. PKras^G12D/+ (blue) pancreata. Normalized Enrichment Score: 2.68;
   Nominal p-value < 0.001; False Discovery Rate (FDR) q-value < 0.001. d
   Representative images of SA-β-gal staining of pancreata with PanIN
   lesions of PKras^G12D/+;Snail^KO/KO mice (n = 2). Scale bars, 50 μm. LG
   Low grade, HG High grade. e Viability of Human Pancreatic Duct
   Epithelial (HPDE) cells after activation of KRAS^G12D alone or in
   combination with SNAIL. HPDE cells transduced with lentiviral
   constructs for doxycycline-inducible expression of EGFP, KRAS^G12D
   ( + mock vector) or KRAS^G12D + SNAIL were treated with 100 ng ml^-1
   doxycycline. Viability was assessed by CellTiter-Glo assay after 72 h
   and is displayed as % of the respective untreated controls. Mean ± SEM.
   n = 3 independent experiments; *p = 0.033, unpaired two-tailed t-test
   with Welch’s correction. f Representative images of SA-β-Gal staining
   of HPDE cells treated for 3 days with doxycycline (100 ng ml^-1) to
   induce activation of EGFP, KRAS^G12D + mock or KRAS^G12D + SNAIL. n = 3
   independent experiments. The percentage of SA-β-gal^+ cells is
   indicated in the upper right corner. Scale bar, 10 µm. Source data of
   Fig. 5 are provided in the Source Data file.

   To probe SNAIL function in a human model and validate the relevance of
   our findings for human PDAC development, we employed immortalized human
   pancreatic duct epithelial (HPDE) lineage cells and engineered them
   with doxycycline-inducible human KRAS^G12D and SNAIL expression vectors
   (Fig. [171]5e, f and Supplementary Fig. [172]S4a). Activation of
   oncogenic KRAS^G12D in HPDE cells induced strong morphological changes
   as well as OIS, as demonstrated by SA-β-gal staining, accompanied by a
   decrease in cell viability compared to EGFP transduced controls. In
   contrast, co-expression of KRAS^G12D and SNAIL reverted this phenotype
   almost completely, resulting in a loss of OIS and increased HPDE
   proliferation (Fig. [173]5e, f, Supplementary Fig. [174]S4a). PDAC
   cells isolated from full blown tumours displayed no SA-β-gal staining
   and senescence phenotype, indicating that OIS is bypassed during early
   steps of pancreatic carcinogenesis as described previously^[175]28
   (Supplementary Fig. [176]S4b). Thus, we demonstrate in genetic mouse
   and engineered human pancreatic epithelial lineage cells that SNAIL
   expression bypasses OIS, thereby driving pancreatic cancer initiation.

SNAIL overcomes senescence without inactivating the TRP53/p21^CIP1 axis

   Oncogene-induced senescence is associated with induction of the tumour
   suppressor TRP53 and/or p16^INK4A depending on cellular context and
   their loss facilitates tumor formation (Supplementary
   Fig. [177]S4c)^[178]30,[179]32. As shown by immunohistochemistry, TRP53
   and its functional downstream target p21^CIP1 are expressed in
   premalignant PanIN lesions and full-blown PDAC cells of
   PKras^G12D/+;Snail^KI/+ mice in vivo (Fig. [180]6a, b). Because mutant
   gain- or loss-of-function TRP53 is unable to transcriptionally activate
   p21^CIP1 ^[181]28, these data support the notion that i) expression and
   function of the TRP53/p21^CIP1 axis is intact in our model, and ii) the
   observed bypass of senescence in PKras^G12D/+;Snail^KI/+ pancreata is
   independent of the TRP53 pathway. To further substantiate these
   findings, we show upregulation of TRP53 and p21^CIP1 protein abundance
   in full-blown PKras^G12D/+;Snail^KI/+ PDAC cells upon treatment with
   the topoisomerase inhibitor etoposide in vitro (Fig. [182]6c) and
   enrichment of TRP53 downstream targets in pancreatic gene expression
   profiles of PanIN-bearing PKras^G12D/+;Snail^KI/+ mice (Fig. [183]6d).

Fig. 6. SNAIL overcomes senescence and the p16^INK4A cell cycle restriction
checkpoint without altering the Trp53/p21^CIP1 axis.

   [184]Fig. 6
   [185]Open in a new tab

   a, b Representative Trp53 and p21^CIP1 stainings of PanINs (a) and PDAC
   (b) of PKras^G12D/+;Snail^KI/+ animals (n = 3 each). c Western blot of
   TRP53 and p21^CIP1 expression in PKras^G12D/+;Snail^Ki/+ (n = 3) and
   PKras^G12D/+ (n = 1) PDAC cell lines after 6 h 20 μM etoposide (Eto) or
   vehicle (DMSO) treatment. d GSEA of mRNA expression of 1-month-old mice
   (n = 2 per genotype) computed and corrected for multiple testing using
   the Benjamini–Hochberg procedure (for statistical details, see methods
   section) shows significant enrichment of KEGG p53 signalling pathway
   genes in PKras^G12D/+;Snail^KI/+ (red) vs. PKras^G12D/+ (blue).
   Normalized Enrichment Score: 1.98; Nominal p-value < 0.001; False
   Discovery Rate (FDR) q-value < 0.001. e, f Kaplan-Meier survival curves
   of PKras^G12D/+ (n = 125; 465 days), PKras^G12D/+;Snail^KI/+ (n = 42;
   190 days), PKras^G12D/+;Trp53^R172H/+ (n = 28; 117 days) and
   PKras^G12D/+;Snail^KI/+;Trp53^R172H/+ (n = 22; 90 days) animals.
   ***p < 0.0001, log-rank test with Bonferroni correction. g
   Representative p16^INK4A staining of PanINs (left) and PDAC (right) of
   PKras^G12D/+;Snail^KI/+ mice (n = 3 each). h qRT-PCR analysis of
   p16^Ink4a (left) and p19^Arf (right) mRNA expression in PDAC of
   endpoint mice (PKras^G12D/+ n = 5; PKras^G12D/+;Snail^KI/+ n = 12;
   PKras^G12D/+;Snail^KI/KI n = 12). Mean ± SEM, **p = 0.0094,
   *p = 0.0365, Mann-Whitney two-tailed test. i Scheme of Cdkn2a gene
   locus and p16^Ink4a genotyping strategy. The non-related proteins
   p16^INK4A and p19^ARF are encoded both by the Cdkn2a locus. Red arrows,
   primer positions. UTR, untranslated region. Scheme according
   to^[186]35. j PCR of p16^Ink4a genomic sequence integrity in PDAC cell
   lines of PKras^G12D/+ (n = 30), PKras^G12D/+;Snail^KI/+ (n = 13);
   PKras^G12D/+;Snail^KI/KI (n = 9) endpoint mice. Gabra, internal
   positive control. k Quantification of PCR analysis of p16^Ink4a genomic
   sequence integrity of data in panel (j). **p = 0.0016, two-tailed
   Fisher’s exact test. l Kaplan-Meier survival curves of
   PKras^G12D/+;Snail^KI/+ (n = 42; 190 days),
   PKras^G12D/+;Snail^KI/+;p16^Ink4a*/+ (n = 22; 156 days) and
   PKras^G12D/+;Snail^KI/+;Cdkn2a^lox/+ with loss of p16^INK4A and p19^ARF
   (n = 17; 108 days). ***p < 0.0001, log-rank test with Bonferroni
   correction. Note: PKras^G12D/+ cohort in panel (e, f) is the same shown
   in Figs. [187]1g, [188]4b, and PKras^G12D/+;Snail^KI/+ cohort of panel
   e and l is the same shown in Fig. [189]1g. Source data of Fig. 6 are
   provided in the Source Data file. Scale bars, 50 μm. ns, not
   significant.

   To test the functional relevance of the TRP53 pathway in an in vivo
   model, we used a genetically engineered p53 mutant allele, which lacks
   canonical TRP53 function and p21^CIP1 induction^[190]33. These animals
   harbour the murine orthologue of the Li-Fraumeni hotspot mutation R175H
   at the endogenous murine Trp53 locus (Trp53^R172H). In this p53 mutant
   background, PKras^G12D/+;Snail^KI/+;Trp53^R172H/+ mice displayed a
   dramatically accelerated PDAC formation and showed in vivo evidence for
   synergy between aberrant SNAIL expression and functional p53
   inactivation. All animals in the tumour watch cohort developed invasive
   PDAC. Median survival of PKras^G12D/+;Snail^KI/+;Trp53^R172H/+ mice (90
   days) was significantly less than PKras^G12D/+;Snail^KI/+ (190 days)
   and PKras^G12D/+;Trp53^R172H/+ (117 days) mice (Fig. [191]6e, f). These
   data demonstrate that TRP53 and SNAIL function, at least in part, via
   non-overlapping and non-redundant pathways and tumour barriers.

   The cell cycle regulator p16^INK4A, which blocks G1 to S phase
   progression of the cell cycle, is implicated in OIS and tumour
   suppression and is frequently lost during KRAS-driven pancreatic
   carcinogenesis (Fig. [192]6g–k and Supplementary
   Fig. [193]S4c)^[194]32. It is encoded by the Cdkn2a locus together with
   p19^ARF, which is involved in TRP53 activation by inhibiting
   Mdm2^[195]34. Immunohistochemistry revealed strong expression of
   p16^INK4A in PanIN lesions and full-blown PDAC of
   PKras^G12D/+;Snail^KI/+ mice (Fig. [196]6g), indicating intact
   p16^INK4A function. Furthermore, p16^Ink4a/p19^Arf mRNA expression in
   PDAC tissue was greater in PKras^G12D/+;Snail^KI/+ than PKras^G12D/+
   animals, and increased further in PKras^G12D/+;Snail^KI/KI animals
   (Fig. [197]6h). We therefore tested the integrity of the genomic
   sequence of the p16^Ink4a locus in cell lines isolated from the three
   different models (Fig. [198]6i, j). We observed loss of p16^Ink4a in
   60% of PKras^G12D/+ PDAC cell lines tested. In contrast, only 38,5% of
   PKras^G12D/+;Snail^KI/+ and none of the PKras^G12D/+;Snail^KI/KI cell
   lines were deficient for p16^Ink4a (Fig. [199]6j, k). These findings
   suggest that Snail might inactivate the p16-RB controlled cell
   cycle/OIS restriction checkpoint downstream of p16^Ink4a, to block OIS
   and drive PDAC progression.

   To test this hypothesis, we again used genetic in vivo models of
   pancreatic tumour evolution. We crossed PKras^G12D/+ and
   PKras^G12D/+;Snail^KI/+ animals with either p16^Ink4a* mutant loss of
   function mice^[200]35, or Cdkn2a^lox mice with conditional knock-out of
   both p16^Ink4a and p19^Arf gene products^[201]36. All animals of the
   tumour watch cohort developed invasive PDAC. However and as
   hypothesized, PKras^G12D/+;Snail^KI/+;p16^Ink4a*/+ mice showed no
   statistically significant difference in median survival compared to
   PKras^G12D/+;Snail^KI/+ mice (Fig. [202]6l). These data demonstrate at
   the level of genetics that p16^Ink4a and SNAIL function via overlapping
   pathways/shared tumour barriers. In line,
   PKras^G12D/+;Snail^KI/+;Cdkn2a^lox/+ mice with loss of both, the
   p16^Ink4a and the p19^ARF tumour suppressor barriers, had dramatically
   reduced median survival (Fig. [203]6l), confirming our data obtained
   with the Trp53^R172H mutant in vivo model (Fig. [204]6e, f).

   Taken together, our mechanistic dissection of three important tumor
   suppressor genes of KRAS-driven carcinogenesis provides strong in vivo
   genetic evidence that SNAIL bypasses senescence and drives tumour
   development independent of TRP53 inactivation and downstream of the
   p16^Ink4a cell cycle restriction check-point, e.g., via blocking the
   RB-controlled senescence-pathway.

   Consistent with this hypothesis, we observed increased proliferation of
   premalignant PanIN lesions, as evidenced by BrdU labelling and Ki67
   staining (Fig. [205]7a, b and Supplementary Fig. [206]S4d). In
   addition, we demonstrate the enrichment of genes that promote cell
   cycle progression by gene expression profiling of PanIN bearing
   PKras^G12D/+;Snail^KI/+ vs. PKras^G12D/+ pancreata of one-month-old
   mice (Fig. [207]7c), as well as increased DNA damage and apoptosis
   induction by aberrant SNAIL expression (Supplementary Fig. [208]S4e–i).
   It has been previously shown that E2F activation in response to RB
   inactivation leads to p53-dependent apoptosis^[209]37–[210]41. Thus,
   combining p53 inactivation with aberrant SNAIL expression accelerates
   tumorigenesis most likely due to the prevention of p53-dependent
   apoptosis.

Fig. 7. SNAIL drives tumour progression downstream of p16^Ink4A by direct
activation of cell cycle regulators.

   [211]Fig. 7
   [212]Open in a new tab

   a Representative BrdU stainings of ADMs and PanINs of PKras^G12D/+
   (n = 11) and PKras^G12D/+;Snail^KI/+ (n = 5) mice. b Percentage of BrdU
   positive cells in ADMs/PanINs of PKras^G12D/+;Snail^KI/+ (n = 5) and
   PKras^G12D/+ (n = 11) mice. Mean ± SEM, **p = 0.006, unpaired
   two-tailed t-test with Welch’s correction. c GSEA 1-month-old mice
   (n = 2 per genotype) computed and corrected for multiple testing using
   Benjamini–Hochberg procedure (statistical details see methods) shows
   significant enrichment of KEGG cell-cycle genes in
   PKras^G12D/+;Snail^Ki/+ (red) vs. PKras^G12D/+ (blue) pancreata.
   Normalized Enrichment Score (NES): 2.38; Nominal p-value < 0.001; False
   Discovery Rate (FDR) q-value < 0.001. d, e Representative stainings (d)
   and quantification of pRb-S807/811 positive PanINs (e) of
   PKras^G12D/+;Snail^KI/+ and PKras^G12D/+ animals (n = 4 per genotype).
   Mean ± SEM, *p = 0.029, Mann-Whitney two-tailed test. f GSEA corrected
   for multiple testing using Benjamini–Hochberg procedure shows
   significant enrichment of hallmark E2F target genes in
   PKras^G12D/+;Snail^KI/+ (red) vs. PKras^G12D/+ (blue) in 1-month-old
   mice (n = 2 per genotype). NES: 2.44; Nominal p-value < 0.001;
   FDRq-value < 0.001. g Chromatin-immunoprecipitation (ChIP) of SNAIL
   binding to E-boxes of indicated promoters in PKras^G12D/+ (n = 3) and
   PKras^G12D/+;Snail^KI/+ (n = 3) PDAC cell lines ± Trp53 mutation as
   indicated. %input calculation; IgG, negative control. Mean ± SEM.
   *p = 0.05, Mann-Whitney one-tailed test. h Ccnb1 and E2f3 promoter
   activity in PKras^G12D/+ (n = 3) and PKras^G12D/+;Snail^KI/KI (n = 3)
   PDAC cells (three independent experiments). Mean ± SEM, *p = 0.026,
   unpaired one-tailed Student’s t-test. i Volcano-plot representing
   enriched proteins in PKras^G12D/+;Snail^KI/+ PDAC cells upon Snail or
   IgG ChIP, respectively, followed by mass-spectrometry based
   quantification of co-precipitated proteins (two independent experiments
   in triplicate for each condition). x-axis, log2-fold change; y-axis,
   adjusted p-value of the two-sample t-test (two-tailed, FDR < 0.05,
   s0 = 1). 141 of 1039 proteins were significant vs. IgG control.
   Pathway-enrichment analysis of significant proteins with MSigDB
   Hallmarks (upper right panel) and Reactome (lower panel) databases. j
   Scheme of genome-scale CRISPR/Cas9 negative-selection screen
   (PKras^G12D/+;Snail^KO/KO, PKras^G12D/+;Snail^KI/+,
   PKras^G12D/+;Snail^KI/KI cells; n = 4). k Differences in β-scores
   (Snail^KI overexpression (OE) - Snail^KO knock-out (KO) cells) were
   used for Reactome database enrichment analysis (FDR ≤ 0.05; difference
   in β score < −1). Scale bars, 50 μm. ns not significant. LG low grade;
   HG high grade. Source data of Fig. 7 provided in Source Data file.

   Phosphorylation and thereby inactivation of the tumour suppressor RB by
   cyclin dependent kinase (CDK)/cyclin complexes, which is negatively
   regulated by p16^Ink4a, is an essential step to bypass the firm
   G1-phase arrest of senescent cells and initiate cell cycle activation
   and proliferation^[213]42. RB inactivation leads to dissociation of the
   E2F complex thereby activating the expression of E2F target genes,
   which can then drive cell cycle progression as well as apoptosis, as
   observed in our model^[214]40,[215]41,[216]43. In line, we detected
   phosphorylation of RB and thus inactivation of the RB-controlled cell
   cycle/senescence checkpoint (Fig. [217]7d, e) in PanIN lesions together
   with concomitant enrichment of E2F target genes (Fig. [218]7f) in the
   PKras^G12D/+;Snail^KI/+ model. Although gene expression profiling of
   bulk tissues is confounded by the increased number of PanIN lesions in
   PKras^G12D/+;Snail^KI/+ mice at an age of one month, and gene sets that
   promote cell cycle progression overlap substantially with E2F target
   genes, our various different in vivo models and datasets provide strong
   evidence for the hypothesis that SNAIL might bypass senescence
   downstream of p16^Ink4a via interference with the RB-controlled cell
   cycle/senescence restriction check-point.

SNAIL is a transcriptional regulator of the cell cycle

   Cyclins, such as cyclin A1 (CCNA1) or cyclin B1 (CCNB1) interact with
   cyclin-dependent kinases (CDK) to phosphorylate and thereby inactivate
   RB, which releases E2F transcription factors to enter the nucleus and
   activate transcription of target genes essential for the transition
   from G1 to S phase and progression of the cell cycle. Because our data
   provide genetic evidence that SNAIL bypasses senescence and drives the
   cell cycle in vivo, we validated the upregulation of important cell
   cycle regulators and downstream effectors. Transcriptomic profiling and
   qRT-PCR analysis revealed a marked increase in mRNA expression of
   several cell cycle-related genes, including cyclins and
   cyclin-dependent kinases, in pancreata with aberrant SNAIL expression
   compared with PKras^G12D/+ controls (Supplementary Fig. [219]S5a, b),
   in line with the proliferation and E2F signature shown in Fig. [220]7c,
   f.

   SNAIL, as a transcription factor, functions primarily via binding to
   promoter and enhancer regions of the target genes. To test whether
   SNAIL binds to promoter regions of genes that positively regulate the
   cell cycle, we analysed publicly available chromatin
   immunoprecipitation (ChIP)-seq cancer cell line datasets^[221]11 and
   compared them to significantly enriched genes in pancreas of
   one-month-old PKras^G12D/+;Snail^KI/+ mice. Of 69 enriched genes
   implicated in proliferation and cell cycle progression, 62 were bound
   by SNAIL in their promoter region (Supplementary Fig. [222]S5c).
   Calculation of the odds ratio for this enrichment (8.53; Supplementary
   Fig. [223]S5c) strongly suggested that the presence of the vast
   majority (89.9%) of genes in the SNAIL-bound fraction was not due to
   chance. Thus, SNAIL has a clear preference for binding to genes that
   promote cell cycle progression. To validate these findings functionally
   in the PKras^G12D/+;Snail^KI/+ PDAC model, we performed ChIP
   experiments using cell lines isolated from PKras^G12D/+ and
   PKras^G12D/+;Snail^KI/+ mice with and without Trp53 mutation, and
   selected SNAIL targets from the ChIP-seq study. This revealed binding
   of SNAIL to E-boxes in promoter regions of multiple genes, such as
   Ccnb1, Ccnb2, Ccnd1, E2f2 and E2f3 (Fig. [224]7g and Supplementary
   Fig. [225]S5d), all known to drive the cell cycle and bound by SNAIL in
   the published ChIP-seq data (Supplementary Fig. [226]S5c). In line with
   the ChIP-seq data set, we did not observe binding of SNAIL to the Ccna1
   promoter region, even though this cyclin mRNA is overexpressed in the
   PKras^G12D/+;Snail^KI/+ model (Supplementary Fig. [227]S5a). To test
   whether SNAIL is indeed capable of activating the expression of the
   identified cell cycle regulators, we performed promoter reporter assays
   using again our primary PDAC cell cultures. While aberrant SNAIL
   expression did not increase reporter gene activity of cyclin D1 (Ccnd1)
   and E2f2 promoter reporter constructs, cyclin B1 (Ccnb1; p = 0.026),
   cyclin B2 (Ccnb2; p = 0.0423) and E2f3 (p = 0.065) demonstrated
   evidence of gene activation (Fig. [228]7h and Supplementary
   Fig. [229]S5e).

   These findings support a context-specific function of SNAIL in vivo,
   which bypasses senescence by direct binding and activating important
   positive regulators of the cell cycle. To gain more insights into
   SNAIL-mediated gene activation, we studied potential co-regulators from
   chromatin cross-linked to SNAIL by proteomics and performed chromatin
   immunoprecipitation coupled to mass spectrometry (ChIP-MS). This
   allowed us to identify 141 significantly enriched putative
   chromatin-bound partners of SNAIL (Fig. [230]7i and Supplementary
   Data [231]1). Activating transcription factors, such as NFKB2 and
   SMAD2, nuclear receptors and coactivators, chromatin remodelers and
   histone modifiers, such as KDM1A were enriched together with SNAIL
   (Fig. [232]7i and Supplementary Data [233]1). Subsequent pathway
   analysis of significantly enriched genes revealed regulators of the
   cell cycle, such as MYC and E2F targets, and genes involved in
   progression through the G2M checkpoint (Fig. [234]7i). Further, several
   genes, e.g., RNA-binding proteins, involved in RNA pol II transcription
   and transcription termination, RNA metabolism, processing, splicing and
   transport were significantly enriched together with SNAIL, indicating a
   role of SNAIL in alternative splicing and RNA biology, which might
   contribute to its function in regulating cell cycle progression. Of
   note, NFKB/E2F interactions have previously been shown to control the
   timing of cell proliferation^[235]44 and SMAD2 silencing decreased PDAC
   cell division^[236]45. In addition, KDM1A has been recently linked to
   gene activation^[237]46 and PDAC cell cycle progression^[238]47. These
   data suggest that multiple interactions of SNAIL might contribute to
   its context-dependent role as a transcriptional activator and regulator
   of the cell cycle.

   To identify functional relevant targets of SNAIL driving PDAC
   progression and maintenance, we performed pooled genome-wide
   CRISPR/Cas9 loss-of-function (viability) screens with cell lines
   isolated from PKras^G12D/+;Snail^KI mice with aberrant SNAIL expression
   compared to PDAC cells from SNAIL-deficient PKras^G12D/+;Snail^KO/KO
   animals (Fig. [239]7j, k, Supplementary Fig. [240]S6 and Supplementary
   Data [241]2). We determined differential sensitivity scores^[242]48 by
   calculating the difference in β-score between SNAIL overexpressing and
   deficient cells and further analysed genes displaying a negative
   differential sensitivity score, pointing to enhanced depletion in
   PKras^G12D/+;Snail^KI cells. This allowed us to identify 238
   statistically significant genes, whose depletion led to the specific
   drop-out of cells with aberrant SNAIL expression.

   Pathway analysis of these hits enabled us to uncover the specific
   genetic dependencies and vulnerabilities of cells with aberrant SNAIL
   expression, such as cell cycle and checkpoint regulation, E2F targets,
   NF-kB signaling, RNA pol II transcription and chromatin modifications
   (Fig. [243]7k and Supplementary Fig. [244]S6b, c). Importantly, we
   observed that these pathways and processes correlated to a high degree
   with the ChIP-MS analysis of Fig. [245]7i, thereby cross-validating our
   findings by functional genetic screens. To probe the contribution of
   the differentially expressed cell cycle regulators for cell viability
   of SNAIL-driven PDAC, we correlated the β-scores of the
   PKras^G12D/+;Snail^KI cells with their gene expression levels. As shown
   in Supplementary Fig. [246]S6d, 32 out of a total of the 49
   differentially expressed cell cycle-related genes of Supplementary
   Fig. [247]S5 displayed a significant differential β-score indicating
   selective depletion in cells with aberrant SNAIL expression.

Aberrant SNAIL expression is prognostic in human PDAC

   To assess a potential EMT-independent link of aberrant SNAIL expression
   with cell cycle progression in human PDAC, we analysed SNAIL and CDH1
   abundance in differentiated and undifferentiated human PDAC specimens
   and cell lines, and analysed data of resected primary tumours^[248]49.
   We observed high levels of SNAIL expression in both differentiated and
   undifferentiated human PDAC specimens and cell lines in accordance with
   our findings in genetic mouse models corroborating EMT-independent
   functions of SNAIL also in human PDAC (Fig. [249]8a, b). In addition,
   we discerned undifferentiated specimens, which lack both, SNAIL and
   CDH1 expression (Fig. [250]8a). Gene expression profiling of primary
   resected differentiated human PDAC specimen with high CDH1 expression
   revealed no correlation with SNAIL abundance (p = 0.77; Fig. [251]8c).
   However, we observed a strong positive correlation between SNAIL and
   the expression of important cell cycle regulators, such as CDK4, CCNA1,
   CCND1, CCND2, CCNE1 and E2F3 (Fig. [252]8d). Strikingly, many of these
   genes have been identified as direct targets of SNAIL in murine PDAC
   (Fig. [253]7 and Supplementary Fig. [254]S5) and functionally validated
   by genome wide CRISPR/Cas9-based negative selection screens
   (Fig. [255]8d right panel). In addition, tumours with high SNAIL
   expression were strongly associated with a poorer disease-free survival
   (DFS) and overall survival (OS) after surgical resection (Fig. [256]8e,
   f). Further, we observed a trend towards resistance against
   chemotherapy with gemcitabine in a small cohort with available clinical
   data of the resected PDAC patients (6 gemcitabine sensitive and 24
   resistant PDAC cases; p = 0.065) (Fig. [257]8g). While intriguing and
   consistent with published experimental studies in mice, reporting that
   Snail knockout sensitizes PDAC tumours to gemcitabine
   treatment^[258]13, these human studies will require larger sample sets
   and prospective analyses in future.

Fig. 8. SNAIL expression in human PDAC is independent of EMT and associated
with poorer survival and chemoresistance.

   [259]Fig. 8
   [260]Open in a new tab

   a Left: Representative staining of SNAIL in human PDAC sections of
   differentiated (G1/2) and undifferentiated tumours (G3/4) of the Human
   Protein Atlas version 20.1 ([261]http://www.proteinatlas.org)^[262]85.
   Images and clinical data are available from
   [263]https://www.proteinatlas.org/ENSG00000124216-SNAI1/pathology/pancr
   eatic+cancer#ihc. Right: Representative SNAIL and CDH1 staining in
   serial sections of an independent PDAC patient cohort with G1/2 and
   G3/4 tumours (n = 11). Scale bars, 50 µm. b qRT-PCR of SNAIL (SNAI1;
   left) and E-cadherin (CDH1; right) mRNA expression of human PDAC cell
   lines (epithelial, n = 9; mesenchymal, n = 16). Mean ± SEM. Left: ns,
   not significant (p = 0.835) unpaired two-tailed t-test; Right:
   p < 0.0001, Mann-Whitney two-tailed test. c SNAI1 (left) and CDH1
   (middle) expression across SNAI1 quartile group of resected primary
   human PDAC samples (Q1 to Q4, n = 88). Mean ± SEM. p = 1.7e-26 (SNAI1)
   and 0.37 (CDH1), one-way ANOVA-test. Right: Pearson correlation of
   SNAIL and CDH1 expression across all PDAC samples (n = 88). Two-tailed
   Pearson correlation coefficient r = 0.031, r^2 = 0.0009949, p = 0.7705
   (not significant), 95% confidence interval −0.1791 to 0.2395. d Left:
   Heatmap of top 20 significant cell cycle related genes with the highest
   variance across SNAI1 quartile groups. Colour code, row-wise scaling of
   RNA expression. Row clustered using hierarchical clustering on
   Euclidean distance. Note: 11 out of the 20 human genes overlap with the
   cell cycle related genes of Supplementary Fig. [264]S5b identified in
   the murine model (depicted in red). Right: Cross-species validation of
   cell cycle regulators. β-scores from genome-wide CRISPR/Cas9
   negative-selection screen of PKras^G12D/+;Snail^KI cell lines are
   indicated. Genes with FDR-q value > 0.05 are marked with an X on the
   bar. e, f Kaplan-Meier analysis of PDAC patients (n = 111). e
   Disease-Free Survival (DFS) p = 0.0178 log-rank test and (f) Overall
   Survival (p = 0.0094 log-rank test), in samples with aberrant high
   SNAI1 expression (Q4) compared to the rest (Q1-3). g Correlation of
   SNAIL expression with gemcitabine treatment resistance of human PDAC
   patients. Density distribution of SNAI1 mRNA expression across
   gemcitabine resistant (n = 6) or sensitive (n = 24) samples, p = 0.065,
   two-tailed Wilcoxon rank test. Source data of Fig. 8 are provided in
   the Source Data file.

Discussion

   Understanding the specific in vivo functions of SNAIL, which is
   aberrantly expressed in a wide variety of epithelial cancers and often
   correlated with poor patient outcome, is crucial for improving patient
   stratification and clinical interventions^[265]5,[266]14. SNAIL has
   been extensively and convincingly characterized as a master regulator
   of the embryonic EMT program, which triggers cancer cell plasticity,
   migration and metastatic spread in various tumour
   types^[267]5,[268]8,[269]11. In contrast, little is known about
   EMT-independent oncogenic functions of SNAIL in cancer initiation and
   progression. Specifically, non-redundant functions of this EMT
   transcription factor (TF) in autochthonous tumours remain
   elusive^[270]26. We employed complex genetic in vivo modelling to
   address this important question in a comprehensive and systematic
   manner across different cancer types, oncogenic drivers and pathways.
   This enabled us to discover a complex non-redundant context-specific
   EMT-independent framework of SNAIL function in epithelial PDAC that
   bypasses oncogenic KRAS-induced senescence and drives the cell cycle by
   p16^INK4A-independent inactivation of the RB-restriction checkpoint of
   senescence and the cell cycle. Importantly, our data demonstrate that
   SNAIL acts in this context as a transcriptional activator, rather than
   via its canonical function as a transcriptional repressor^[271]50; it
   binds directly to the canonical E-boxes of a variety of important cell
   cycle regulators, such as cyclins, CDKs and E2F TFs to drive the cell
   cycle as evidenced by ChIP experiments and reporter gene assays. This
   allows sustained proliferation of epithelial PDAC cells and thus tumour
   progression independent of overt EMT induction and contrasts with WNT-
   and KRAS-driven intestinal cancer subtypes, which are refractory
   towards aberrantly expressed SNAIL. In line with our findings, SNAIL
   has recently been shown to be dispensable for the EMT process in PDAC,
   which is controlled by the EMT transcription factor Zeb1^[272]51.

   Collectively, our studies constitute a comprehensive analysis of SNAIL
   function in cancer. SNAIL has been identified and validated as an
   intrinsic cancer driver, and there are strong indications that both,
   the cell and tissue of origin as well as the genetic context dictates
   the function of SNAIL as a cancer driver. This improves our
   understanding of the diverse in vivo functions of SNAIL and will enable
   SNAIL downstream targets to be defined within the cell cycle machinery
   in epithelial PDAC. Our discovery has potentially important clinical
   implications, since it provides a framework for patient stratification
   and opens avenues for therapeutic interventions. Therapeutics targeting
   the cell cycle have been developed in recent years, which provide
   efficient opportunities to block cell cycle progression, e.g., via
   blockade of CDK4/6 activity^[273]52. Furthermore, considering the
   association of SNAIL with Gemcitabine resistance, it would seem
   worthwhile evaluating whether targeting SNAIL downstream effectors can
   improve the efficacy of current therapies for PDAC. Such treatment
   options are urgently needed. PDAC is a highly lethal and refractive
   disease with overall 5-year survival rates below 9%^[274]53.

Methods

Mouse strains and tumour models

   LSL-Kras^G12D/+ ^[275]54,[276]55, Pdx1-Cre^[277]54, Ptf1a^Cre/+
   ^[278]56,[279]57, LSL-Trp53^R172H/+ ^[280]33,[281]58,
   LSL-R26^TvailacZ/+ ^[282]59, Cdkn2a^lox/+ ^[283]36, p16^Ink4a*/+
   ^[284]35, Villin-Cre^[285]60, LSL-Braf^V637E/+ ^[286]21, Apc^lox/+
   ^[287]20, Cdh1^lox/+ ^[288]61, Snail^lox/+ ^[289]62 mice have been
   previously reported. All strains were on a mixed C57Bl/6 J;129S6/SvEv
   genetic background and interbred to obtain compound mutant mice of both
   sexes that develop autochthonous tumours in the pancreas and intestine.
   The sex, substrain, age and number of all animals analysed in this
   study in every experiment is provided in the “Source Data” Excel file.
   All animal studies were conducted in compliance with European
   guidelines for the care and use of laboratory animals and were approved
   by the Institutional Animal Care and Use Committees (IACUC) of the
   local authorities of Technische Universität München and the Regierung
   von Oberbayern (animal protocol number: ROB-55.2-2532.Vet_02-17-79 and
   55.2-1-54-2532-31-11). The maximal tumour size/burden permitted by the
   IACUC and the local authorities (Regierung von Oberbayern) is 1.5 cm in
   diameter, which was not exceeded in our study. Euthanasia was performed
   by cervical dislocation. Animals were housed under specific pathogen-
   free conditions (SPF) in a dedicated facility, with a light-dark cycle
   of 12:12 hours, a relative air humidity between 45 and 65% and a
   temperature between 20 and 24 °C.

Construction of the targeting vector and generation of the LSL-Rosa26^Snail
mouse line

   Rosa26 targeting by a knock-in strategy was performed based on the
   pROSA26–1 plasmid^[290]59. Murine Snail cDNA (Snai1 cDNA; Library: IRAV
   MGC Mouse verified full length amplified cDNA; Clone: IRAVp968A0443D6,
   German Science Centre for Genome Research) was cloned into the
   targeting vector 3´ of a loxP-flanked transcriptional and translational
   stop element (loxP-stop-loxP, LSL) with a neomycin resistance cassette
   (Supplementary Fig. [291]S1a). The targeting vector was linearized,
   electroporated into W4/129S6 embryonic stem cells, selection with
   250 µg/ml geneticin imposed, and correctly targeted cell clones
   identified by PCR^[292]59. Gene targeting was verified by Southern blot
   with an external ^32P-labeled 5´ probe and EcoRV digested genomic DNA
   (Supplementary Fig. [293]S1b). The Southern blot images were processed
   with an Amersham automatic Hyperprocessor (Amersham Biosciences). Two
   verified cell clones were injected into C57BL/6 J blastocysts
   (Polygene). Germ-line transmission was achieved in 2/2 clones
   harbouring the targeted allele. The mice were genotyped using a
   3-primer PCR strategy (ref. ^[294]59, Table [295]1 and Supplementary
   Fig. [296]S1c).

Table 1.

   Recombination PCR primers for LSL-Rosa26^Snail allele
   Name of PCR Name of primer Sequence (5’ – 3’)
   LSL-Rosa26^Snail recombination R26-GT forward AAAGTCGCTCTGAGTTGTTAT
   R26-Stop cassette reverse TGAATAGTTAATTGGAGCGGCCGCAATA
   Snail-cds reverse GCGCTCCTTCCTGGT
   [297]Open in a new tab

Transduction of tumour cells using the RCAS-TVA system

   To overexpress SNAIL in cell lines via the RCAS-TVA
   system^[298]59,[299]63, the murine Snail cDNA was amplified and cloned
   into the pCR-Blunt II-TOPO vector (Invitrogen). After AatII/NdeI
   digestion, Snail cDNA was ligated to a modified pENTR/D-TOPO
   (Invitrogen) vector carrying dsRed under the control of the EF1α
   promoter 3’ to the Snail insertion side. Further cloning into
   RCASBP(A)-Att-CCDB-Att (modified from RCASBP(A), kindly provided by
   Stephen H. Hughes) was performed using the GatewayR LR Clonase
   (Invitrogen) mix to generate the final retroviral construct.

   To generate RCAS vectors, the chicken fibroblast cell line DF-1
   (American Type Culture Collection # CRL-12203 (RRID:CVCL_0570) was
   transfected using Superfect (Qiagen) with 2.5 µg purified RCAS plasmid.
   Fresh virus-containing supernatant was filtered through 0.45 μm pores
   and added to the medium of murine tumour cells carrying the TVA
   receptor^[300]59. Transduction with fresh supernatant was repeated
   daily until 80% cells showed expression of the dsRed reporter gene
   (Supplementary Fig. [301]S3f).

Histology and immunohistochemistry

   Murine tissue specimens were fixed overnight in 4% buffered formalin,
   dehydrated, embedded in paraffin and sectioned (2.5 µm thick). ADM,
   PanIN lesions and intestinal adenomas and carcinomas were quantified
   using haematoxylin and eosin (H&E)-stained sections^[302]64.
   Quantification was carried out blinded to the genotype.

   TUNEL staining was conducted using the In Situ Cell Death Detection
   Kit, POD (Roche). Alcian blue staining and immunohistochemistry were
   performed using standard procedures^[303]64. If not stated otherwise,
   antigen retrieval was performed in citrate buffer, pH 6.0 in a
   microwave oven. The following primary antibodies were used: Muc5a
   (antigen retrieval Tris/EDTA pH 9.0, 1:200, 45M1 #MS-145-P1,
   Neomarkers), Cytokeratin 19 (1:300, TROMA 3 Developmental Studies
   Hybridoma Bank), E-cadherin (1:100, #610181, BD Biosciences), Rabbit
   anti-Ki67 (1:50, #MA5-14520, SP6, ThermoFischer), p-γ-H2AX (1:500,
   #05-636, Millipore), Cleaved Caspase 3 (1:250, #9664, Cell Signaling
   Technology), BrdU (1:500, #MCA2060, AbD Serotec), pRB (1:100, #8516,
   Cell Signaling Technology), p16^INK4A (1:50, #sc-1661, Santa Cruz
   Biotechnology), TRP53 (1:400, #NCL-p53-CM5p, Novocastra/Leica
   Microsystems), p21^CIP1 (1:50, #sc-397, Santa Cruz Biotechnology),
   SNAIL (1:50, #3879, Cell Signaling Technology).

   For BrdU assay, 5 mg/kg 5-bromo-2’-deoxyuridine (BrdU), dissolved in
   sterile PBS, was injected intraperitoneally into animals 2 h before
   sacrifice.

   Images were acquired with AxioVision Rel 4.8 and Aperio ImageScope
   v12.3.3. For counting of BrdU-, pRB-, and p-γ-H2AX-positive cells in
   ADMs and PanINs, one- to three-months old PKras^G12D/+;Snail^KI/+ mice
   and one-month to two-year old PKras^G12D/+ animals were used.
   Quantification was carried out blinded to the genotype.

Metastasis quantification

   At sacrifice, abdominal organs and lungs were investigated
   macroscopically for metastases^[304]17,[305]65. Macroscopic pictures
   were taken using a Stemi SV 11 stereomicroscope (Zeiss) and processed
   with AxioVision Rel 4.8 software. For microscopic quantification, at
   least ten series of sections (100 µm between each series) of
   paraffin-embedded lungs and livers were prepared, H&E stained and
   investigated for the presence of metastases. Quantification was carried
   out blinded to the genotype.

Senescence-associated β-galactosidase (SA-β-gal) analysis

   To obtain cryosections, tissue was fixed in 4% buffered formalin for
   2 h, dehydrated in a sucrose series (15% sucrose for 4 h and 30%
   sucrose overnight), embedded in Tissue-Tek (Sakura Finetek),
   snap-frozen and sectioned (6 µm). Sections were dried overnight, and
   staining performed using the Senescence β-Galactosidase Staining Kit
   (Cell Signaling Technology)^[306]66. ADMs and PanINs from three
   different slides per pancreas were assessed for SA-β-gal
   quantification. The number of cells displaying positive SA-β-gal
   staining was counted and divided by the total number of cells per PanIN
   lesion, and expressed as % positive cells per lesion in the respective
   graphs. Quantification was done blinded to the genotype. SA-β-gal
   staining of cells in culture was performed as recommended by the
   manufacturer of the the Senescence β-Galactosidase Staining Kit and
   quantified blinded to the genotype as % positive cells.

Immunocytochemistry

   Cells were washed 3 times in cold PBS and fixed 10 min in cold
   methanol. Washing was repeated following permeabilisation in 0.3%
   Triton X-100 in PBS for 10 min. Blocking was done for 30 min at 37 °C
   with 5% donkey serum before incubation with the primary E-cadherin
   antibody (1:80, #AF748, R&D Systems) for 2 h at 37 °C. After washing 3
   times with PBS, cells were incubated with secondary antibody (Alexa
   Fluor® 488 donkey anti-goat, 1:100, # A-11055, Invitrogen) for 30 min
   at 37 °C. Washing was repeated and cells were covered with a cover
   glass using Vectashield mounting medium with DAPI. Images of the slides
   were acquired with AxioVision Rel 4.8 and Aperio ImageScope v12.3.3.

Cell lines and cell culture

   Primary PDAC cell cultures were isolated from autochthonous mouse PDAC
   tumours and cultured in DMEM medium with 10% Fetal Bovine
   Serum^[307]17,[308]65. The following human PDAC cell lines from from
   the American Type Culture Collection (ATCC), German Collection of
   Microorganisms and Cell Cultures (DSMZ) or Cell bank were used: AsPC-1
   (CVCL_0152) ATCC# CRL-1682; Capan-2 (CVCL_0026) ATCC# HTB-80; CFPAC-1
   (CVCL_1119) ATCC# CRL-1918; DAN-G (CVCL_0243) DSMZ# ACC 249; HPAC
   (CVCL_3517) ATCC# CRL-2119; HPAF-II (CVCL_0313) ATCC# CRL-1997; Hs
   766 T (CVCL_0334) ATCC# HTB-134; HuP-T4 (CVCL_1300) DSMZ# ACC 223;
   IMIM-PC1 (CVCL_4061) [309]https://www.cellosaurus.org/CVCL_4061; KP-4
   (CVCL_1338) Cell bank# [310]RCB1005; MIA PaCa-2 (CVCL_0428) ATCC#
   CRL-1420; Panc 02.03 (CVCL_1633) ATCC# CRL-2553; Panc 03.27 (CVCL_1635)
   ATCC# CRL-2549; Panc 04.03 (CVCL_1636) ATCC# CRL-2555; Panc 05.04
   (CVCL_1637) ATCC# CRL-2557; Panc 08.13 (CVCL_1638) ATCC# CRL-2551;
   PANC-1 (CVCL_0480) ATCC# CRL-1469; Panc 10.05 (CVCL_1639) ATCC#
   CRL-2547; PaTu 8902 (CVCL_1845) DSMZ# ACC 179; PaTu 8988 s (CVCL_1846)
   DSMZ# ACC 204; PL45 (CVCL_3567) ATCC# CRL-2558; PSN1 (CVCL_1644) ATCC#
   CRL-3211; SU.86.86 (CVCL_3881) ATCC# CRL-1837; SW1990 (CVCL_1723) ATCC#
   CRL-2172; YAPC (CVCL_1794) DSMZ# ACC 382^[311]67. The Human Pancreatic
   Duct Epithelial (HPDE) cells (H6c7; RRID: CVCL_0P38) were obtained from
   Kerafast (#ECA001-FP), and the avian TVA receptor positive chicken
   embryonic fibroblast cell line DF-1 (RRID:CVCL_0570) from ATCC#
   CRL-12203.

   All human cell lines were authenticated through STR or SNP profiling
   (last correct authentication in 2022). All murine cell lines were
   re-genotyped and tested for correct recombination of the respective
   alleles (last re-genotyping in 2022). The chicken fibroblast cell line
   DF-1 was authenticated by genotyping PCR for presence of the avian TVA
   receptor. All cells used were cultivated for less than 30 passages and
   tested negative for mycoplasma contamination by PCR. The
   electrophoresis DNA gel pictures were acquired with the Gel Doc™
   XR + system (Biorad).

PDAC cell doubling time calculation

   For PDAC cell doubling time calculation, 1000–2000 cells per well were
   seeded out in triplicates in 96 Well plates. Cell viability was
   determined on the following day (Day 0) and again 72 hours after the
   initial measurement (Day 3) by CellTiter-Glo assay (Promega). Doubling
   times were calculated blinded to the genotype by the formula given in
   equation number 1:

   Equation Number 1:
   [MATH:
   <mi>D</mi><mi>o</mi><mi>u</mi><mi>b</mi><mi>l</mi><mi>i</mi><mi>n</mi><
   mi>g</mi><mi>T</mi><mi>i</mi><mi>m</mi><mi>e</mi><mo>=</mo><mn>72</mn><
   mspace
   width="0.25em"></mspace><mi>h</mi><mi>o</mi><mi>u</mi><mi>r</mi><mi>s</
   mi><mo>*</mo><mfrac><mrow><mi>log</mi><mfenced close=")"
   open="("><mrow><mn>2</mn></mrow></mfenced></mrow><mrow><mi>log</mi><mfe
   nced close=")"
   open="("><mrow><mfrac><mrow><mi>m</mi><mi>e</mi><mi>a</mi><mi>n</mi><ms
   pace
   width="0.25em"></mspace><mi>C</mi><mi>e</mi><mi>l</mi><mi>l</mi><mi>T</
   mi><mi>i</mi><mi>t</mi><mi>e</mi><mi>r</mi><mspace
   width="0.25em"></mspace><mi>G</mi><mi>l</mi><mi>o</mi><mspace
   width="0.25em"></mspace><mi>v</mi><mi>a</mi><mi>l</mi><mi>u</mi><mi>e</
   mi><mspace width="0.25em"></mspace><mi>o</mi><mi>n</mi><mspace
   width="0.25em"></mspace><mi>D</mi><mi>a</mi><mi>y</mi><mspace
   width="0.25em"></mspace><mn>3</mn></mrow><mrow><mi>m</mi><mi>e</mi><mi>
   a</mi><mi>n</mi><mspace
   width="0.25em"></mspace><mi>C</mi><mi>e</mi><mi>l</mi><mi>l</mi><mi>T</
   mi><mi>i</mi><mi>t</mi><mi>e</mi><mi>r</mi><mi>G</mi><mi>l</mi><mi>o</m
   i><mspace
   width="0.25em"></mspace><mi>v</mi><mi>a</mi><mi>l</mi><mi>u</mi><mi>e</
   mi><mspace width="0.25em"></mspace><mi>o</mi><mi>n</mi><mspace
   width="0.25em"></mspace><mi>D</mi><mi>a</mi><mi>y</mi><mspace
   width="0.25em"></mspace><mn>0</mn></mrow></mfrac></mrow></mfenced></mro
   w></mfrac> :MATH]

Stimulation of PDAC cells with TGFβ

   Cells at 50% confluence were cultured for 24 h in FCS-free DMEM before
   treatment with 10 ng/ml TGFβ or vehicle (10 nM citric acid, 2 mg/ml
   BSA). Cell morphology was documented after 72 h.

Acinar explants and acinar-ductal metaplasia (ADM) assay

   Directly after sacrifice, pancreata of one-month old mice were injected
   with 2 ml Collagenase P solution (1.33 mg/ml Collagenase P (Roche) in
   HBSS (Gibco)), cut out, minced with a scalpel and gently shaken for
   30 min at 37 °C in 5 ml Collagenase P solution. All subsequent steps
   were performed at 4 °C in a laminar flow cabinet and all centrifugation
   steps were carried out for 3 min at 180 x g. Cells were resuspended in
   10 ml 5% FBS in HBSS and incubated 10 min for sedimentation of the
   cellular fraction. Supernatant was aspirated carefully, and the pellet
   was washed 3 times with 5% FBS in HBSS. Cells in 10 ml 5% FBS in HBSS
   were transferred into a new tube through a 100 µm cell strainer, slowly
   laid over 20 ml 30% FBS in HBSS and centrifuged. Cells were resuspended
   in 2 ml recovery medium (acinar cell medium, see below, with 30% FCS),
   incubated at 37 °C for 1 h, centrifuged and resuspended in a 1:1
   mixture of acinar cell medium (containing 0.1% bovine serum albumin,
   0.2 mg/ml soybean trypsin inhibitor (Sigma), 1% ITS premix (Corning),
   50 µg/ml bovine pituitary extract (ThermoFisher), 0.1% FBS, 0.5%
   penicillin/streptomycin, 0.25 µg/ml Fungizone antimycotic
   (ThermoFisher) in Waymouth’s medium (Gibco) and rat tail collagen type
   I (Corning). Per pancreas, cells were seeded into 16 wells of a 48-well
   plate on a previously prepared collagen layer (final collagen
   concentration 2.5 mg/ml) and covered with another collagen layer before
   adding acinar cell medium. Medium was changed every 24 h.

   Five days after seeding, images were acquired with AxioVision Rel 4.8
   software and the percentage of ductal structures of the total amount of
   acinar explants was determined by counting 5 microscopic fields of view
   at 100x magnification for each pancreas. Quantification was blinded to
   the genotype.

Whole cell lysates and western blot

   Whole cell lysates and proteins from tissue were harvested and
   subjected to western blotting using the following primary antibodies:
   SNAIL (1:500, #3895, Cell Signaling Technology), E-cadherin (1:2000,
   #610181, BD Biosciences), HSP90 (1:250, #sc-13119, Santa Cruz
   Biotechnology), TRP53 (1:1000, #NCL-p53-CM5p, Novocastra/Leica
   Microsystems), p21^CIP1 (1:200, #sc-397, Santa Cruz Biotechnology),
   β-Actin (1:4000, #A5316, Sigma-Aldrich) and α-Tubulin (1:5000, #T9026,
   Sigma-Aldrich). The western blot images were collected using the
   Odyssey infrared imaging system with the Odyssey Software V1.2 (Li-Cor
   Biosciences).

Quantitative real-time PCR (qPCR)

   Total RNA was isolated from tissues and cell lines with the RNeasy Kit
   (Qiagen) following reverse transcription (Applied Biosciences). 1 µg
   RNA was used for generation of 50 µl cDNA. qPCR was performed with the
   StepOnePlus real time PCR system (Applied Biosystems) by using the
   StepOne Software v2.3. Power SYBR Green PCR Master Mix was used in a
   25 µl mixture containing 100 nM of each primer. Only primers with an
   amplification efficiency between 1.8 and 2.2 were applied. qPCR primers
   are given in Table [312]2. mRNA expression was analysed on 5 µl of 1:5
   diluted cDNA in either duplicate or triplicate. All expression values
   were normalized to the housekeeping gene Cyclophilin A (CypA) or GAPDH.
   A melt curve was performed after each run to check for unwanted primer
   dimerization. Data analysis was carried out using Excel version 16.65
   (Microsoft Corporation) according to 2^-ΔΔCt method.

Table 2.

   qPCR primers for testing mRNA expression level
           Gene         Name of primer Sequence (5’ – 3’)
   CDH1 human

   hCDH1-forward

   hCDH1-reverse

   CCGAGAGCTACACGTTC

   TCTTCAAAATTCACTCTGCC
   Cdh1 mouse

   mCdh1-forward

   mCdh1-reverse

   GAGCGTGCCCCAGTATCG

   CGTAATCGAACACCAACAGAGAGT
   p16^Ink4a

   mp16-forward

   mp16-reverse

   CCCAACGCCCCGAACT

   GTGAACGTTGCCCATCATCA
   p19^Arf

   mp19-forward

   mp19-reverse

   TCGCAGGTTCTTGGTCACTGT

   GAACTTCACCAAGAAAACCCTCTCT
   Ccna1

   mCcnA1-forward

   mCcnA1-reverse

   GCTGTCTCTTTACCCGGAGCA

   ACGTTCACTGGCTTGTCTTCTA
   Ccna2

   mCcnA2-forward

   mCcnA2-reverse

   CACTGACACCTCTTGACTATCC

   CGTTCACTGGCTTGTCTTCT
   Ccnb1

   mCcnB1-forward

   mCcnB1-reverse

   TTGTGTGCCCAAGAAGATGCT

   GTACATCTCCTCATATTTGCTTGCA
   Ccnb2

   mCcnB2-forward

   mCcnB2-reverse

   TGAAGTCCTGGAAGTCATGC

   GAGGCCAGGTCTTTGATGAT
   SNAIL human

   hSNAIL-forward

   hSNAIL-reverse

   CTCTAATCCAGAGTTTACCTTC

   GACAGAGTCCCAGATGAG
   SNAIL mouse

   mSNAIL-forward

   mSNAIL-reverse

   GCCGGAAGCCCAACTATAGC

   GGTCGTAGGGCTGCTGGAA
   KRAS human

   hKRAS-forward

   hKRAS-reverse

   GACTGAATATAAACTTGTGGTAGTTGGA

   CATATTCGTCCACAAAATGATTCTG
   Cyclophilin A (CypA)

   CypA-forward

   CypA-reverse

   ATGGTCAACCCCACCGTGT

   TTCTTGCTGTCTTTGGAACTTTGTC
   GAPDH human

   hGAPDH-forward

   hGAPDH-reverse

   AATCCCATCACCATCTTCCA

   TGGACTCCACGACGTACTCA
   [313]Open in a new tab

Analysis of p16^Ink4a genomic sequence integrity

   Genomic DNA was isolated from PDAC cell lines using the DNeasy Blood &
   Tissue Kit (Qiagen). The integrity of the p16^Ink4a locus was tested by
   PDAC amplification and gel electrophoresis using 10 ng DNA and the
   primers given in Table [314]3. GABRA was used as positive control.

Table 3.

   Primers for testing p16^Ink4a genomic sequence integrity
       Name of PCR     Name of primer Sequence (5’ – 3’)
   p16^Ink4a integrity

   p16^Ink4a-forward

   p16^Ink4a-reverse

   AGTTCGGGGCGTTGGG

   GCACAGGCTCTGGAATGCA
   Gabra

   Gabra-forward

   Gabra reverse

   AACACACACTGGAGGACTGGCTAGG

   CAATGGTAGGCTCACTCTGGGAGATGATA
   [315]Open in a new tab

Quantitative chromatin immunoprecipitation (ChIP)

   ChIP was performed using SimpleChIP Enzymatic Chromatin IP Kit (#9003,
   Cell Signaling Technology) according to the manufacturer’s protocol
   using SNAIL antibody (1:50, #3879, Cell Signaling Technology) and
   rabbit IgG (1:50, #2729, C15D3, Cell Signaling Technology) as negative
   control and H3 (1:50, #2650, Cell Signaling Technology) as positive
   control. Binding of SNAIL to the DNA regions of interest was determined
   by qPCR using the primers listed in Table [316]4 and analysed by the
   percent input method^[317]68.

Table 4.

   qPCR primers for testing E-box binding by ChIP
   Gene  Name of primer Sequence (5’ – 3’)
   Ccna1

   CcnA1-Ebox-forward

   CcnA1-Ebox forward

   TTAAAGCCCATTCAGCCATTGTT

   TGTCCCAACTTCCCGACAAAC
   Ccnb1

   CcnB1-Ebox-forward

   CcnB1-Ebox-reverse

   CATTGCTGCCACCTGCCTTA

   ATGCGTACTCCCCACAGTCA
   Ccnb2

   CcnB2-Ebox-forward

   CcnB2-Ebox-reverse

   CATCGTCTCCAGGTCGTTCA

   ATGACTCTGCTGGGGATCTGT
   Ccnd1

   CcnD1-Ebox-forward

   CcnD1-Ebox-reverse

   AGCGTCCCTGTCTTCTTTCAA

   GTCTGGCATCTTCGGGTGTT
   E2f2

   E2F2-Ebox-forward

   E2F2-Ebox-reverse

   TGCCTCAGTTTCGCCTACTG

   ACAGCGATTACGACAGGAGC
   E2f3

   E2F3-Ebox-forward

   E2F3-Ebox-reverse

   GCGCAAGTTTCGGTTTTGG

   CTACACTGCTTGGTTACAGGA
   [318]Open in a new tab

Chromatin immunoprecipitation coupled to mass spectrometry (ChIP-MS)

   ChIP was performed using freshly prepared cell lysates of murine
   primary PDAC cells (P144) isolated from the PKras^G12D/+;Snail^KI/+
   model. For each condition, three biological replicates were used.
   Briefly, 10^7 cells were fixed in 1% v/v formaldehyde (FA) in Phosphate
   buffered saline (PBS) at room temperature (RT) for 10 min. After
   incubating with 1.25 M glycine and washing twice with PBS, the samples
   were resuspended in IP buffer (50 mM Tris-HCl pH 8.0, 100 mM NaCl, 5 mM
   EDTA pH 8.0, 1.7% v/v Triton X-100, 0.3% v/v SDS, protease and
   phosphatase inhibitors). Subsequently, chromatin was sonicated (4 × 10
   cycles at 4 °C; 30 s ON, 30 seconds OFF each cycle) using a Bioruptor
   Plus (Diagenode, Denville, NJ, United States) to an average size of
   500 bp. The samples were then centrifuged at 3500 x g for 20 min at
   4 °C, and the supernatant was used for estimation of protein
   concentration with the bicinchoninic acid (BCA) assay based on the
   manufacturer’s instructions (Pierce™ BCA Protein Assay Kit, Thermo
   Fisher Scientific). 1 mg of extract from each sample was used for
   immunoprecipitation with anti-Snail Rabbit mAb (1:50, #3879, C15D3,
   Cell Signaling) or anti-IgG Rabbit Ab (#2729, Cell Signaling, 5 µg) by
   incubating overnight at 4 °C on a rotating wheel. The next day, the
   antibody-bound complexes were precipitated with protein A + G-coupled
   magnetic beads (Millipore, Sigma) washed three times with low salt
   buffer (50 mM HEPES pH 7.5, 140 mM NaCl, 1% v/v Triton X-100), once
   with high salt buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 1% v/v Triton
   X-100) and once with TBS. Immunoprecipitates were eluted from the beads
   and digested after incubating with the freshly prepared elution buffers
   I (2 M Urea, 50 mM Tris-HCl pH 7.5, 2 mM Dithiothreitol, 20 µg/mL
   Trypsin) and II (2 M Urea, 50 mM Tris-HCl pH 7.5, 10 mM
   Chloroacetamide) for 30 min and 5 min at 37 °C, respectively. Both
   eluates were combined and further incubated overnight at 25 °C.
   Subsequently, the tryptic peptides were acidified with 1% v/v
   Trifluoroacetic acid (TFA) solution, and transferred on top of
   styrene-divinylbenzene reverse-phase sulfonate (SDB-RPS; three layers)
   in-house made StageTips for desalting. Finally, they were concentrated
   (45 °C, 20 min) using a centrifugal evaporator (Eppendorf) until
   dryness, and analszed by liquid chromatography-coupled to mass
   spectrometry (LC-MS/MS) after reconstitution in MS compatible buffer
   [2% acetonitrile (ACN) v/v, 0.1 % v/v TFA]^[319]69.

LC-MS/MS analysis and data processing

   All peptide samples were measured in a single-shot manner in a
   Q-Exactive HF-X hybrid quadrupole-orbitrap mass spectrometer (Thermo
   Fisher Scientific) after peptide separation by high-performance liquid
   chromatography (nanoLC 1200, Thermo Fisher Scientific) using a 50 cm
   reversed-phase column (made in house, packed with 1.9 µm C18 ReproSil
   particles). Peptides were eluted over a 90-minute-gradient from 0% to
   95% buffer B (0.1% formic acid and 80% ACN) with a flow rate of
   300 nL/minute.

   Full scans were obtained from 300 to 1650 m/z with a target value of
   3 × 10^6 ions at a resolution of 60,000 at 200 m/z. The fifteen most
   intense ions (Top15) of each full scan were fragmented with
   higher-energy collisional dissociation (HCD) (target value
   1 × 10^5 ions, maximum injection time 120 ms, isolation window 1.4 m/z,
   underfill ratio 1%), and fragments were detected in the Orbitrap mass
   analyzer at a resolution of 15,000 at 200 m/z.

Mass spectrometry data analysis

   Raw MS data files were processed using MaxQuant (version 1.6.1.0) to
   calculate peptide intensities with the integrated Andromeda search
   engine with FDR < 0.01 both at the protein and peptide levels. Oxidized
   methionine (M) and acetylation (protein N-terminus) were set as
   variable modifications, and carbamidomethyl (C) as fixed modification.
   Only peptides with a minimal length of seven amino acids were
   considered and the “match between runs” option was enabled for the
   biological replicates within each condition with a matching time window
   of 0.7 min. For protein and peptide identification, the UniProt
   database from mouse (September 2014) including 51,210 entries were
   used. Each raw file and biological replicate was treated as one
   independent experiment.

   For bioinformatics analyses, the Perseus platform^[320]70 (version
   1.6.7.0) was used. The R environment (version 3.6.2) was used for data
   visualization. Pre-processing of the label-free proteomics data
   included: (a) exclusion of reverse, potential contaminants and peptide
   identified only by site, (b) log2 transformation of peptide
   intensities, and (c) peptides without intensity values in less than 67%
   of the values in at least one bait group were filtered out. Missing
   values were replaced from a normal distribution window (width 0.3,
   downshift 1.8 standard deviations).

   For statistical analysis, the two-sample t-test was implemented
   (FDR < 0.05, s0 = 1) and identified 141 proteins as significant (out of
   1039 quantified proteins) between the two conditions (bait vs. negative
   control).

Lentivirus production and transduction

   For lentivirus production, HEK293FT cells were transfected using
   TransIT-LT1 (Mirus Bio LLC) transfection reagent according to the
   manufacturer’s instructions with lentiviral packaging plasmids psPAX2
   and pMD2.G and the respective lentiviral transfer plasmids.
   Virus-containing supernatant was collected 48 and 72 h after
   transfection, filtered through a 0.45 µm filter and stored at −80 °C.
   Cell lines were transduced in the presence of 8 μg ml^−1 polybrene and
   selected with the respective selection antibiotic (Puromycin or
   Blasticidin).

Inducible activation of KRAS^G12D and Snail in HPDE Cells

   Inducible expression vectors for GFP and mutant Kras^G12D based on the
   pInducer20 vector system have been used in HPDE cells^[321]32. To
   generate an inducible expression system for Snail, cDNA of human SNAIL
   was cloned into the pInducer20-Blast (RRID:Addgene 109334)^[322]71 and
   verified by sequencing.

   HPDE cells were cultivated in Keratinocyte-SFM medium (ThermoFisher).
   To induce expression of the respective target genes, cells were treated
   for the indicated time points with 100 ng ml^−1 doxycycline.

Promoter reporter assays

   The promoter reporter constructs for E2F2 (#MPRM38445-LvPG04-GC), E2F3
   (MPRM40957-LvPG04-GC), CCNB1 (MPRM49947-LvPG04-GC) and CCNB2
   (MPRM39222-LvPG04-GC) were purchased from BioCat GmbH (Heidelberg,
   Germany) and transduced into PKras^G12D/+ and PKras^G12D/+;Snail^KI/KI
   PDAC cell lines.

   The Secrete-Pair Dual Luminescence Assay Kit (#LF032-GC, BioCat GmbH,
   Heidelberg, Germany) was used according to the manufacturer’s
   instructions to analyse reporter activity blinded to the genotype. In
   brief, cell lines transduced with the reporter constructs were seeded
   in 6-well plates and medium was collected after 24 h. For measurement
   of Gaussia Luciferase (GLuc), 10 µL of the collected culture medium
   were pipetted in duplicates into a white opaque 96-well plate. GLuc
   Assay Working Solution was prepared with Buffer GL-S according to the
   manufacturer’s instructions and 100 µL was added per well. After
   incubation for 1 min at room temperature, luminescence was measured
   using a CLARIOstar microplate reader (BMG Labtech GmbH). For
   transduction normalization, Secreted Alkaline Phosphatase (SEAP) was
   measured. Therefore, medium was heated at 65 °C for 15 min and SEAP
   Assay Working Solution prepared according to the manufacturer’s
   protocol. 100 µL SEAP Assay Working Solution were added to 10 µL medium
   per sample in duplicates in a white opaque 96-well plate. Luminescence
   was measured after 5 min incubation at room temperature in a CLARIOstar
   microplate reader (BMG Labtech GmbH). Ratios of the mean Gaussia
   Luciferase (GLuc) to the mean Secreted Alkaline Phosphatase (SEAP) were
   calculated to determine reporter activity.

   pGL3 Basic luciferase reporter plasmids containing Cyclin D1 (Ccnd1)
   promoter fragments (RRID:Addgene_32727 and RRID:Addgene_32726)^[323]72
   were used to determine Cyclin D1 promoter activity blinded to the
   genotype. PKras^G12D/+ and PKras^G12D/+;Snail^KI/KI PDAC cell cultures
   were transfected with the reporter constructs using Effectene
   Transfection Reagent (Qiagen, Hilden, Germany) according to the
   manufacturer’s recommendations. In each sample, 40 ng phRL-TK (Promega)
   Renilla Luciferase control reporter vector was co-transfected as an
   internal control for transfection efficiency. The medium was changed on
   the next day and the Dual-luciferase reporter assay system (Promega)
   was used according to the manufacturer’s instructions to determine
   luciferase activity 48 h post transfection.

Genome-wide CRISPR/Cas9 negative selection screens

   PKras^G12D/+;Snail^KI and PKras^G12D/+;Snail^KO/KO Cas9-expressing cell
   lines were used to perform the genome-wide CRISPR/Cas9 loss-of-function
   screens at 500x coverage, as in^[324]48. Briefly, cells were transduced
   with the Brie library (Addgene #73633) and the screens were performed
   in side-by-side duplicates. At the end of the experiment, cells were
   harvested, genomic DNA was isolated using the Blood & Cell Culture DNA
   Maxi Kit (Qiagen), and sgRNA libraries were generated. The pooled sgRNA
   libraries were sequenced using an Illumina NextSeq 500 (custom read and
   indexing primers spiked in) with a depth of 35 Mio reads^[325]48.
   MAGeCK v0.5.9.4^[326]73 was used for downstream analysis and β-scores
   calculated with the maximum likelihood estimation (mle) method by
   employing data of non-targeting control sgRNAs. The obtained β-score
   depicts enrichment (positive score) or depletion (negative score) of
   the sgRNAs compared to their initial abundance. To calculate
   selectively depleted genes in the PKras^G12D/+;Snail^KI model, the
   difference in the β-score between PKras^G12D/+;Snail^KI and
   PKras^G12D/+;Snail^KO/KO PDAC cells was determined. Enrichments were
   performed on the genes annotated as non-essential, presenting an
   FDR ≤ 0.05 in both PKras^G12D/+;Snail^KI and PKras^G12D/+;Snail^KO/KO
   cells, and showing a difference in beta score ≤ −1 by using the MSigDB
   v7.1 gene sets provided by Broad Institute, Massachusetts Institute of
   Technology and Harvard University as in^[327]48.

Gene expression profiling and gene set enrichment analysis (GSEA)

   mRNA was extracted using the RNeasy Kit (Qiagen). Quality was checked
   using the Experion RNA StdSens analysis Kit (Bio-Rad). For mRNA
   analysis of PDAC cell lines, 250 ng of each sample were processed with
   the GeneChip 3’ IVT express Kit (Affymetrix). Fragmentation and
   hybridization to GeneChip mouse genome 430 2.0 array chips (Affymetrix)
   was performed by the Institute for Medical Microbiology, Immunology and
   Hygiene, Technische Universität München. For mRNA analysis of
   pancreatic and PDAC tissues, 500 ng isolated mRNA was processed with
   the Ambion WT expression Kit (Applied Biosystems). Fragmentation and
   labelling were performed using the GeneChip WT terminal labelling Kit
   (Affymetrix). Hybridization to GeneChip mouse gene 1.0 ST array chips
   (Affymetrix) was carried out by the Institute for Medical Microbiology,
   Immunology and Hygiene, Technische Universität München. Data were
   collected with an Affymetrix Scanner 3000 7 G and the Affymetrix
   GeneChip Command Console Software (AGCC). The expression intensity of
   each gene was determined by using the Affymetrix Microarray Analysis
   Suite (MAS) 5.0 software.

   All analyses were carried out using R version 3.1.2^[328]74 and
   Bioconductor version 3.0^[329]75. Microarray data were processed with
   the RMA method^[330]76, following quantile normalization^[331]77. For
   initial correlation analysis, pairwise Pearson correlation was computed
   on the normalized intensity values. Differential gene expression
   between mesenchymal and epithelial cell lines was analysed with Limma
   version 3.22.0^[332]78. A probe set was considered to be differentially
   expressed with a Benjamini-Hochberg adjusted^[333]79 p-value of 0.05
   and an absolute fold change >2. Annotations were downloaded from
   ENSEMBL (GRCm38.p3)^[334]80. The top 50 upregulated or downregulated
   genes were hierarchically clustered with Ward’s minimum variance
   method^[335]81. The dissimilarities between samples were squared before
   cluster updating as implemented in R.

   We performed gene set enrichment analysis (GSEA)^[336]82 using GSEA
   v3.0 jar package and MSigDB v6.2 gene sets provided by Broad Institute
   of Massachusetts Institute of Technology and Harvard University. GSEA
   was conducted with RMA normalized microarray data. Parameters were set
   as follows: phenotype was defined as “PKras^G12D/+;Snail^KI/+“ versus
   “PKras^G12D/+“; gene sets were permuted for 1000 times; enrichment
   statistic for scoring was set as “weighted” and genes were ranked based
   on “t-Test” metric; other parameters were set as default. The cut-off
   for a significant FDR q-value as well as NOM p-value was set at 0.05.

Human primary PDAC cohort

   RNAseq data from resected primary PDAC tumours are accessible via
   International Cancer Genome Consortium (ICGC), as reported in Connor et
   al.^[337]49 were analysed using R version 4.2.12. Adapters and bad
   quality reads were trimmed with Trimmomatic version 2.38^[338]83.
   Filtered reads were aligned to human genome (hg19) and quantified using
   STAR version 2.6.0^[339]84. Raw read count per gene was normalized to
   the library size using Counts Per Million (CPM). The resected cohort
   (n = 177) was divided into epithelial-like (n = 88) and
   mesenchymal-like samples (n = 89) based on CDH1 expression. Samples
   with high CDH1 expression, i.e., above CDH1 median expression, were
   classified as epithelial subtype, whereas low CDH1 expression samples
   were considered as mesenchymal subtype. Epithelial samples were further
   divided according to the expression of SNAI1 mRNA using quartile
   distribution were Q1 and Q4 contain the samples with the lowest and
   highest SNAI1 expression respectively. Differential expression of cell
   cycle-related genes across Q1 to Q4 subgroups was determined using an
   ANOVA test. Top 20 significantly altered cell-cycle genes across SNAI1
   quartiles are depicted on the heatmap (see Fig. [340]8d).

   Among the epithelial-like PDAC cohort, 32 patients received adjuvant
   chemotherapy with Gemcitabine. 24 were sensitive and 6 resistant. The
   response of 2 patients is unknown. The differential expression of SNAI1
   mRNA between Gemcitabine resistant (n = 6) and sensitive (n = 24)
   samples was assessed using Wilcoxon rank test.

   Survival analysis was performed on the complete resected dataset
   (111/177 samples with follow-up annotation). Samples with aberrant
   SNAI1 expression (Quantile 4, Q4) were compared to the other samples
   (Q1 to Q3). Difference of survival was determined with a Cox
   proportional hazards regression model. P-value below 0.05 was
   considered as significant.

   To assess the expression of SNAIL in human PDAC sections of
   differentiated (G1/2) and undifferentiated tumours (G3/4), we used
   publicly available immunohistochemical stainings of the Human Protein
   Atlas version 20.1^[341]85, which are available from
   [342]https://www.proteinatlas.org/ENSG00000124216-SNAI1/pathology/pancr
   eatic+cancer#ihc, as well as a cohort of PDAC tissue samples purchased
   from Biomax.us (PA961a Pancreatic cancer tissue array with normal
   pancreatic tissue, [343]https://www.biomax.us/PA961a) and stained for
   SNAIL and CDH1.

Additional statistical methods and data analysis

   No statistical method was used to determine sample size a priory. In
   Supplementary Fig. [344]1e, one outlier in the PKras^G12D/+;Snail^KI/+
   cohort that differed significantly from the other observations, has
   been removed from the analysis (please see Source Data file,
   Supplementary Fig. [345]1e, for detailed information on outlier
   definition). No other data were excluded from other datasets.
   Randomization was not appropriate for experiments described in this
   study. The investigators were blinded to allocation during experiments
   and outcome assessment. Graphical depiction and statistical analysis
   were performed with GraphPad Prism v5 and v8. Unless otherwise
   indicated, all data were determined from at least 3 independent
   experiments and expressed as mean values ± SEM. For comparisons between
   data sets, log-rank test, Fisher’s exact test, one- or two-tailed
   t-test with or without Welch’s correction or Mann-Whitney test were
   employed and resulting p-values are indicated in the respective
   figures. The significance level was set to 0.05. If more than one
   statistical test was performed simultaneously on a single data set, a
   Bonferroni-adjusted significance level was calculated to account for
   the increased possibility of false-positive results. Percentage of mice
   with intestinal carcinoma, cell morphology and p16^Ink4a genomic
   sequence integrity were compared by Fisher’s exact test. Metastasis
   rates were compared by Fisher’s exact test followed by multiple testing
   correction with Benjamini Hochberg procedure. Survival analysis of the
   mouse models was carried out by the log-rank test.

Reporting summary

   Further information on research design is available in the [346]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [347]Supplementary Information^ (10.4MB, pdf)
   [348]41467_2023_36505_MOESM2_ESM.pdf^ (91KB, pdf)

   Description of Additional Supplementary Files
   [349]Supplementary Data 1^ (69.4KB, xlsx)
   [350]Supplementary Data 2^ (2.7MB, xlsx)
   [351]Reporting Summary^ (95.3KB, pdf)

Acknowledgements