ATM-deficiency increases genomic instability and metastatic potential in a mouse model of pancreatic cancer

Germline mutations in ATM (encoding the DNA-damage signaling kinase, ataxia-telangiectasia-mutated) increase Familial Pancreatic Cancer (FPC) susceptibility, and ATM somatic mutations have been identified in resected human pancreatic tumors. Here we investigated how Atm contributes to pancreatic cancer by deleting this gene in a murine model of the disease expressing oncogenic Kras (KrasG12D). We show that partial or total ATM deficiency cooperates with KrasG12D to promote highly metastatic pancreatic cancer. We also reveal that ATM is activated in pancreatic precancerous lesions in the context of DNA damage and cell proliferation, and demonstrate that ATM deficiency leads to persistent DNA damage in both precancerous lesions and primary tumors. Using low passage cultures from primary tumors and liver metastases we show that ATM loss accelerates Kras-induced carcinogenesis without conferring a specific phenotype to pancreatic tumors or changing the status of the tumor suppressors p53, p16Ink4a and p19Arf. However, ATM deficiency markedly increases the proportion of chromosomal alterations in pancreatic primary tumors and liver metastases. More importantly, ATM deficiency also renders murine pancreatic tumors highly sensitive to radiation. These and other findings in our study conclusively establish that ATM activity poses a major barrier to oncogenic transformation in the pancreas via maintaining genomic stability.

PDAC develops in only a fraction of KC mice through a process that is reminiscent to the human disease 13 . Accordingly, roughly 40% of KC mice in a C57/NMRI genetic background showed symptoms indicative of tumor formation (Supplementary Table 1) after 1 year of age and succumbed shortly thereafter. Post-mortem analyses revealed that in the KC group the incidence of pancreatic tumors and liver metastases was 42% and 28%, respectively (Fig. 1A). This tumor frequency parallels our findings in a previous study using KC mice of a similar C57/ NMRI mixed background 14 . On the other hand, more than half of the KCATMΔ+ and KCATMΔΔ mice showed the former symptoms before 1 year of age and displayed an average life expectancy that was almost 5 months shorter than that of KC mice (the median survival was: 9 months [KCATMΔ+], 9 months [KCATMΔΔ], and 14 months [KC]; Fig. 1A). Results of post-mortem analysis identified pancreatic tumors in 62% of KCATMΔ+ mice and in nearly 100% of KCATMΔΔ mice (Fig. 1A). Additionally, this analysis also uncovered liver metastases in 62% of KCATMΔ+ mice and in 78% of KCATMΔΔ mice (Fig. 1A). These data demonstrate that ATM deficiency synergizes with Kras G12D to promote the formation of highly metastatic pancreatic tumors.
Histopathology analysis and cytokeratin-19 immunostaining of 3 KC metastatic tumors and of representative KCATMΔ+ (n = 6) and KCATMΔΔ (n = 5) tumors identified invasive PDAC with focal or extensive desmoplasia and variable morphology in all specimens (Fig. 1B,C). As has been described, KC tumors exhibited mostly sarcomatoid histology whereas KCATMΔ+ and KCATMΔΔ tumors had lesions with glandular histology and moderate or poor differentiation (Fig. 1B,C). These results indicate that ATM deficiency (partial or total) probably influences the spectrum of PDAC histology in Kras G12D -induced pancreatic tumors. Immunostaining results also showed scattered cells expressing the ATM target phospho-Kap1 6,7 in both the KC and KCATMΔ+ pancreatic tumor epithelium (Fig. 1E), and absence of this protein in KCATMΔΔ pancreatic tumor epithelial cells (Fig. 1E). This finding indicates that the pancreatic tumors of our KCATMΔ+ mice retained a functional copy of Atm.
Activation of ATM by DSBs results in phosphorylation of the histone H2A variant H2AX (to generate γH2AX) and localization of this phospho-protein to sites of DNA damage. Since ATM activity has been implicated in oncogene-induced DNA damage responses 4,15 we examined the expression of γH2AX in ATM-proficient and ATM-deficient pancreatic tumors of our study group to compare the extent of DNA damage in those tissues. This analysis revealed a few cells expressing γH2AX in tumor specimens from KC mice (Fig. 1F), and considerably more cells expressing γH2AX in tumor specimens from KCATMΔΔ mice (Fig. 1F). Quantitative results corroborated that the pancreatic tumors of KCATMΔΔ mice had significantly more γH2AX + cells than those tumors of KC mice (Fig. 1F). Similar analyses performed in human pancreatic specimens showed that γH2AX expression was virtually undetected in both, 1 control (healthy) specimen (Fig. 1G) and 1 specimen harboring a benign adenoma (Fig. 1G). In contrast, cells expressing γH2AX were noticed in focal areas of 1 tumor specimen from a patient with a heterozygous ATM germline inactivating mutation (Fig. 1G) and in large areas of 1 human pancreatic tumor that carried homozygous ATM somatic inactivating mutations (Fig. 1G). Thus, it is conceivable that ATM signaling is activated in pancreatic tumors of mice and humans to prevent the propagation of unrepaired DNA damage and to preserve genome integrity.
ATM-proficient and ATM-deficient primary tumors display a 'pancreatic progenitor/classical' phenotype and have similar alterations in Trp53 and p16 INK4A /p19 Arf expression. To further understand how ATM loss accelerates the formation of metastatic pancreatic tumors, we established low-passage cell cultures from primary PDACs and liver metastases of all 3 genotypes using published methods [16][17][18] . PCR analysis of RNA extracted from 3 KC cell lines, 3 KCATMΔ+ cell lines and 4 KCATMΔΔ cell lines showed that all KC lines expressed the Atm wildtype transcript, all KCATMΔΔ lines expressed the deleted (floxed) Atm transcript, and all KCATMΔ+ cell lines expressed both the wildtype and deleted Atm transcript (Supplementary Figure 2).
Likewise, FISH analysis revealed that no large deletions involving the Atm locus had occurred in KCATMΔ+ cell lines generated from PDAC and liver metastases (Supplementary Figure 2). These combined results further indicate that Atm loss-of-heterozygosity (LOH) is not a prerequisite for the development of murine metastatic pancreatic tumors that already carry an inactive copy of Atm.
Two recent studies used genome wide expression methods to classify human pancreatic tumors based on their specific expression profile. In their study, Bailey et al. 19 identified 4 subtypes of PDA that were named squamous, pancreatic progenitor, immunogenic and aberrantly differentiated endocrine exocrine or ADEX. Similarly, Collisson et al. 20 identified 3 tumor subtypes that were named classical, quasimesenchymal and exocrine-like. With the exception of the immunogenic subtype, there is broad overlap amongst the transcriptome classifiers of the pancreatic progenitor/classical, squamous/quasimesenchymal and ADEX/exocrine-like subtypes 19,20 . We subjected our primary tumor cell lines to qPCR analysis to investigate the expression of transcripts that the former studies used as tumor subtype classifiers. These results showed higher expression of the 'pancreatic progenitor' markers Pdx1, Hnf1β and Lgals4 19 across all our KC, KCATMΔ+ and KCATMΔΔ tumor cell lines (both from primary tumors and liver metastasis), in comparison to markers of the other subtypes ( Fig. 2A and Supplementary Table 2). Moreover, the 'classical' tumor marker Gata6 20 was uniformly expressed at high levels in all KC, KCATMΔ+ and KCATMΔΔ tumor cell lines ( Fig. 2A and Supplementary Table 2). Concurrent with the former results, immunostaining analysis of the 'pancreatic progenitor' marker Hnf1β showed extensive expression of this transcription factor in both primary tumors and liver metastases of all 3 genotypes (Fig. 2B). Interestingly, our generated tumor cell lines also expressed moderate to low levels of the 'progenitor' marker Hnf4 and the 'squamous' markers Egfr and Prxx1 (Fig. 2A). These results agree with the former histopathology results (Fig. 1B,C) showing that all the resected tumors presented variable glandular/sarcomatoid histology. Therefore, we conclude that the Kras G12D -driven tumors of our mouse models mainly acquired a pancreatic progenitor/ classical phenotype irrespective of the status of ATM.
We also investigated the expression of transcripts encoding the tumor suppressors p53, p16 Ink4a and p19 Arf 21 in our generated pancreatic cancer cell lines using qPCR. This analysis showed similar expression of Trp53 across all our generated KC, KCATMΔ+ and KCATMΔΔ cell lines (Fig. 2C). In addition, data from exon sequencing using genomic DNA or cDNA showed no mutations in Trp53 coding regions 22 in any of the 3 KCATMΔΔ PDAC cell lines that we tested (data not shown). On the other hand, qPCR analysis revealed that  Lack of ATM increases chromosomal instability and promotes radiosensitization in pancreatic tumor cells. As we determined that ATM-deficient tumors have more widespread DNA damage than ATM-proficient tumors, we examined the extent of chromosomal alterations (an indicator of genome instability) in our established KC and KCATMΔΔ tumor cell lines using SKY analysis (Fig. 3A,B). SKY results showed similar proportion of clonal alterations (numerical/structural) in 3 individual KC primary tumor cell lines and 3 individual KCATMΔΔ primary tumor cell lines (6-15% and 9-16% respectively; Fig. 3C). In contrast, the frequency of non-clonal (random) structural alterations was higher in the KCATMΔΔ primary tumor cells in comparison to KC primary tumor cells (68-83% and 16-47%, respectively; Fig. 3C). In addition, the proportion of breakage type alterations (chromatid gaps/breaks) was also higher in the KCATMΔΔ primary tumor cells than in KCATMΔΔ primary tumor cells (24-53% and 0-11%, respectively; Fig. 3C). Furthermore, the comparison of clonal alterations between a primary tumor cell line and a liver metastasis cell line from 3 KC mice and 3 KCATMΔΔ mice underscored more numerical and newly acquired structural abnormalities in the ATM-deficient metastases in comparison to the ATM-proficient metastases (Fig. 3D). These collective SKY results conclusively demonstrate that ATM-deficiency increases chromosomal instability in Kras G12D -induced pancreatic tumors.
Tumors that have defects in DDR components are highly sensitive to agents that cause DNA damage 4 , and published studies showed that ATM-nullizygous cells display radiosensitivity 7 . Thus, we investigated if partial or total ATM-deficiency affects the response of pancreatic cells to DNA damage-inducing agents by exposing our KC and KCATMΔΔ primary tumor cells to X-ray or γ-irradiation (IR). As reported by others 23 , we detected activation of signals downstream of ATM (i.e., induction of the ATM-specific target phospho-Kap1, p53 [Ser-15] phosphorylation, and p21 increase) in mouse embryonic fibroblasts (MEFs) exposed to 4Gy (1-4 hours post-IR) (Supplementary Figure 3). Similarly, those markers were also induced in KC 215.1 cells (either from primary PDAC or liver metastasis [LM]) 1-4 hours post-IR (4Gy; Fig. 4A-C). Interestingly, KCATMΔ+ 255.9 cells (PDAC and LM-derived) exposed to 4Gy IR also showed phospho-Kap-1 induction and upregulation of phospho-p53 and p21 proteins 1-4 hours post-treatment ( Fig. 4A-C), and a similar result was also noticed in 2 additional KCATMΔ+ cell lines after irradiation (Supplementary Figure 3). In contrast, KCATMΔΔ 149.4 cells (both PDAC and LM) exposed to similar IR doses did not express phospho-Kap1 and only increased phospho-p53 slightly after 24 hours (although p21 induction was similar to that of 215.1 and 255.9 cells; Fig. 4A-C). These results were reproduced using the KC cell lines 121.3 and 92.8, and the KCATMΔΔ cell lines 72.2 and 161.3 (data not shown) and were validated using quantitative methods (Fig. 4D). Together, these data demonstrate that DNA-damage responses downstream of ATM are functional in KC and KCATMΔ+ pancreatic tumors, and absent in KCATMΔΔ pancreatic tumors.
In agreement with the previous Western blot results, immunostaining analysis revealed expression of phospho-Kap1 in KC 215.1 cells exposed to 4Gy IR and lack of this protein in KCATMΔΔ 149.4 cells subjected to irradiation (Fig. 5A). In addition, immunostaining with anti-γH2AX antibodies uncovered numerous cells containing γH2AX+ foci in KCATMΔΔ 149.4 cultures (Fig. 5B) and less cells showing γH2AX+ foci in KC 215.1 cultures (Fig. 5B), 24h post-IR. Quantification of these results corroborated that significantly more γH2AX+  Figure 3). In conclusion, these novel results demonstrate that ATM deficiency confers a vulnerability (radiosensitivity) to pancreatic cancer cells that can be exploited therapeutically.
ATM deficiency increases DNA damage in Kras G12D -induced pancreatic precancerous lesions. Kras G12D -driven murine pancreatic tumors arise from precancerous lesions that include small ductal lesions and mPanINs (Fig. 6A,B), and these epithelial entities are increasingly abundant in the pancreas of KC mice after 1 month of age 14,[24][25][26] . Interestingly, a published study 27 reported activation of the ATM-CHK2 checkpoint and mounting DNA damage (in the form of γH2AX expression) in early human PanINs. Therefore, we investigated the expression of ATM-downstream effectors in precancerous lesions of KC mice using immunostaining methods. We detected expression of γH2AX (Fig. 6C), the specific ATM target phospho-Kap1 6, 7 (Fig. 6D), and the p53 target p21 (Fig. 6E), in small ductal lesions -and to a lesser extent in mPanINs -in pancreatic tissues of 2-months old KC mice. Additionally, we uncovered extensive expression of the proliferation marker Ki67 in small ductal lesions (Fig. 6F), and more limited expression of this marker in mPanINs (Fig. 6F), in 2-months old KC pancreata. These collective results demonstrate that ATM signaling is activated in murine pancreatic precancerous lesions in the context of cell proliferation and DNA damage. The previous results prompted investigating whether, similar to our findings in tumors (Fig. 1F,G), ATM deficiency increases DNA damage in pancreatic precancerous lesions. Immunofluorescence results showed expression of phospho-Kap1 in the pancreatic epithelium (Ecadherin + ) of 2-4 months old KC and KCATMΔ+ mice, and lack of this protein in the pancreas of KCATMΔΔ littermates (Fig. 6G). This data validated that Atm was effectively deleted in KCATMΔΔ pancreatic tissues. More importantly, double-immunofluorescence results revealed that ductal epithelial lesions exhibiting γH2AX + foci were more abundant in KCATMΔ+ and KCATMΔΔ pancreata than in KC pancreata (Fig. 6H). Quantification of these results corroborated that γH2AX + cells were more numerous in pancreatic epithelial cells of 2-months old KCATMΔ+ and KCATMΔΔ mice in comparison to those cells of KC littermates (Fig. 6I). These results conclusively establish that ATM activity is necessary to prevent the accumulation of excessive DNA damage in pancreatic precancerous lesions induced by Kras G12D . On the other hand, our finding that p21 was expressed in the pancreas of KCATMΔ+ and KCATMΔΔ mice (Fig. 6J) indicated that ATM activity is dispensable or redundant for the induction of p53-dependent checkpoints in pancreatic precancerous lesions.  ATM deficiency does not prevent Kras G12D -induced senescence but enhances mPanIN formation. Oncogene-induced senescence is a major tumor barrier in human precancerous lesions and it has been suggested that this process follows activation of a robust DDR 28,29 . On the other hand, the role of ATM in oncogene-induced senescence is controversial. For instance, a published study demonstrated that ATM inhibition prevents Kras-induced senescence in human fibroblasts 30 . In contrast, a separate report showed that lack of ATM does not affect senescence induction in Kras V12 -driven mouse lung adenomas 31 . These discrepancies prompted us to investigate the effects of ATM deficiency in senescence induction in murine Kras G12D -induced pancreatic precancerous lesions.
We stained the pancreas of 2-4 months old KC mice for SA-βgal detection to identify senescent cells. To complement this analysis, we also stained those tissues with antibodies recognizing p16 Ink4a because a previous study reported concomitant expression of p16 Ink4a and SA-βgal in mPanINs induced by Kras V12 32 . We observed broad expression of SA-βgal in pancreatic precancerous epithelial lesions (Fig. 7A) of KC mice. In addition, we detected p16 Ink4a and p19 Arf (the 2 gene products of Cdkn2a) protein expression in both small ductal lesions and mPanINs (Fig. 7B,C) in the pancreas of KC mice. Interestingly, our SA-βgal staining results showed no differences in the expression of this marker amongst precancerous lesions of KC, KCATMΔ+ and KCATMΔΔ mice (Fig. 7A). Also, both p16 Ink4a (Fig. 7B) and p19 Arf (Fig. 7C) had comparable expression in pancreatic precancerous lesions of the 3 genotypes. These results demonstrate that ATM activity is dispensable for oncogene-induced senescence in Kras G12D -driven pancreatic precancerous lesions. In addition, our finding that precancerous lesions but not pancreatic tumors express p16 Ink4a and p19 Arf in our mouse models (Fig. 2C) further argues that the checkpoint controlled by these proteins is lost late in pancreas carcinogenesis".
We also stained the pancreata of our mouse models with H&E and uncovered focal areas harboring ductal epithelial structures surrounded by stroma in those tissues of 2 months old KC mice (Fig. 7D), and more extensive areas containing stroma and ductal epithelial lesions in the pancreas of KCATMΔ+ and KCATMΔΔ littermates (Fig. 7D). To further this observation, the pancreata of 2-4 months old KC, KCATMΔ+ and KCATMΔΔ mice was stained with Alcian Blue 33 and the mPanIN lesions were counted. Results of this quantitative analysis corroborated that mPanINs were more numerous in the pancreas of both KCATMΔ+ and KCATMΔΔ mice than in this organ of KC littermates (Fig. 7E,F). Likewise, histologic and immunostaining results confirmed that the fibrotic  (Fig. 7G) and the macrophage infiltrates (Fig. 7H) were more prominent in the pancreas of KCATMΔ+ and KCATMΔΔ mice than in this organ of KC littermates. These results underscore that partial or total ATM deficiency directly or indirectly enhances Kras G12D -driven mPanIN formation and the associated inflammation.

Discussion
The activation of DNA damage responses downstream of oncogene-induced replication poses a major barrier to cancer progression and contributes to maintain genomic stability 5,34 . ATM plays an important role in cancer protection because its activity engages the molecular machinery that repairs DSBs 6,7 . In addition, ATM induces p53-dependent responses that promote cell cycle arrest, senescence or apoptosis to prevent the propagation of excessive DNA damage 5,15,35 . In this study we demonstrate that ATM activity is necessary for genome surveillance in the pancreas. First, we show that partial or total ATM deficiency cooperates with oncogenic Kras to promote highly metastatic murine pancreatic cancer. Second, we establish that lack of ATM causes persistent DNA damage in pancreatic tumors of mice and possibly also in pancreatic tumors of humans. Third, we show that ATM activity is necessary in established pancreatic tumors to prevent the accumulation of deleterious DNA lesions and to preserve genome integrity. These and other results suggest a model whereby oncogene-induced replication stress promotes DSBs in pancreatic precancerous cells and elicits activation of an ATM-mediated signaling cascade to repair the damaged DNA. In the absence of ATM activity, DNA damage (in the form of unrepaired DSBs) escalates, new oncogenic mutations are acquired, genomic instability accumulates, and pancreatic cancerous cells emerge. Thus, our study conclusively establishes that ATM's main tumor suppressor role in the pancreas is to maintain genomic stability. Moreover, the marked differences in the number of clonal alterations uncovered between our KC and KCATMΔΔ metastatic cell lines indicate that ATM activity restrains PDAC in a locally advanced state, and that once its function is lost the accumulation of genomic instability enhances distant metastasis. This notion concurs with published data showing that human sporadic tumors that exhibit unstable genomes, chemoresistance, and poor prognosis (including those arising in the pancreas), frequently carry ATM somatic mutations 9,32 .
Similar to our study, Russell et al. 36 showed that loss of ATM accelerates Kras G12D -induced pancreatic cancer in mice. However, those authors did not report chromosomal alterations or excessive DNA damage in pancreatic tumors that lack ATM. Instead, they concluded that ATM deficiency enhances pancreatic tumor potential by increasing EMT and BMP4 signaling in precancerous lesions 36 . While these parameters were not evaluated in our analysis, a direct role of ATM in EMT or BMP signaling is yet to be proven. Moreover, our SKY results strongly indicate that the primary driver of metastasis in ATM-deficient pancreatic tumors is chromosomal instability. This conclusion concurs with the recognized role of ATM in sensing DNA damage 6,7 , and with published results showing that aggressive, unstable forms of PDAC often carry somatic ATM mutations 32 . Therefore, our hypothesis that ATM activity is key to preserve genomic stability in pancreatic cells under oncogenic stress is well supported. On the other hand, we find intriguing that our generated ATM-deficient and ATM-proficient tumor cell lines shared a similar 'pancreatic progenitor/classical' subtype expression profile (including high Gata6 expression), because those features were found to correlate with better outcome in PDA patients 30,31 . Thus, it is possible that the increased genomic instability of ATM deficient tumors contributes to disable some of the transcriptional networks and gene programs operating in 'pancreatic progenitor/classical' tumors. Addressing this and other outstanding questions using genome-wide approaches is necessary to understand how ATM deficiency increases pancreatic metastatic disease.
Our new finding that murine pancreatic precancerous lesions express phospho-Kap1 coincides with similar observations in humans 27 and argues that ATM is activated very early during Kras G12D -driven oncogenesis, probably in response to DSBs induced by replication stress 4,35 . However, ATM could accomplish functions that do not involve signaling DNA damage in pancreatic precancerous lesions. For instance, ATM activity has been implicated in cellular redox homeostasis independent from its role in DNA damage responses 8 . Also, the activation of DNA damage (by ROS) and ATM signaling was shown to suppress the production of cytokines 37 . A potential regulatory role of ATM in both cellular redox and cytokine production in pancreatic tissues would agree with our finding that both KCATMΔ+ and KCATMΔΔ mice present increased DNA damage, extensive fibrosis and abundant macrophage infiltrates in the pancreas. Moreover, overabundance of cytokines could enhance a stromal reaction that favors the formation of mPanINs 38,39 . Investigating these issues should help disclosing how ATM activity protects precancerous pancreatic cells from oncogenic transformation.
The discovery that our generated ATM-deficient pancreatic tumor cells expressed p53 but not p16 Ink4A /p19 Arf concurs with the observation that Kras G12D -induced murine pancreatic tumors have frequent loss of p16 Ink4A and p19 Arf or loss of p53, but not simultaneous loss of the 3 tumor suppressors 21 . Moreover, our findings also agree with a whole-genome sequencing study 22 showing more frequent mutations or losses of CDKN2A (6/9 tumors) than TP53 (3/9 tumors) in human pancreatic tumors carrying ATM mutations. The preferential loss of p16 Ink4A /p19 Arf expression over that of p53 suggests that the need to inactivate the p53/p21 barrier is bypassed in ATM-deficient tumors because p53 is a target of this kinase. In contrast, inactivating p16 Ink4A /p19 Arf is probably necessary in Kras G12D -induced tumors to overcome the checkpoints induced by these tumor suppressors and the senescence barrier imposed by the p16 Ink4A -Rb pathway 21 . Whether genomic instability facilitates the inactivation of p16 Ink4A /p19 Arf in ATM-deficient pancreatic tumor cells or other mechanisms are involved in p16 Ink4A /p19 Arf downregulation warrants further investigation (e.g., ATM loss could enable the DNMT1-mediated methylation of the p16 Ink4A promoter) 40 .
The results from FISH, gamma-irradiation and qPCR analyses showed that our generated KCATMΔ+ pancreatic cancer cells retained a functional wildtype Atm allele. These findings are intriguing because in a study reporting germline heterozygous ATM mutations in FPC kindreds the analyzed tumors showed ATM LOH inactivation 3 . These results argue that transformed human and murine pancreatic cells respond differently to the dosage of ATM, although it is also possible that in our KCATMΔ+ tumor cells additional mutations or oncogenic events bypassed the need to inactivate the wildtype Atm allele. While these hypotheses remain to be tested, it is noteworthy that similar to our study other investigators reported acceleration of PDGF-induced murine gliomas in the context of Atm heterozygosis without LOH 41 .
In addition to providing new information on DNA damage-induced ATM activation in the pancreas, and the effects of ATM deficiency in the context of Kras G12D expression, our study introduces a novel mouse model that should be clinically relevant. First, the pancreatic tumors of our mice have the genetic features of some aggressive forms of human pancreatic cancer, as it was shown that patients carrying tumoral loss of ATM and normal p53 expression tend to have very poor prognosis after tumor resection 42 . Second, the fact that our established ATM-deficient murine pancreatic tumor cells display radiosensitivity is quite relevant because other studies showed that human pancreatic tumor cells become radiosensitized upon treatment with an ATM inhibitor 43 or ATM siRNA 44 . Therefore, our mice are suitable models to test drugs or combined treatments that confer synthetic lethality (e.g., PARP1-or ATR inhibitors 45,46 ) to pancreatic tumors carrying ATM-deficiencies.
In summary, our study determined that the main tumor suppressor role of ATM in the pancreas is to maintain genomic stability. This finding agrees with the canonical genome surveillance role of ATM and has clinical relevance, since human pancreatic tumors that carry ATM somatic mutations have 'unstable genomes' and are very difficult to overcome. In addition, our discovery that ATM signaling is activated in pancreatic precancerous lesions to prevent the propagation of unrepaired DNA suggests that pathologic conditions that promote DNA damage (e.g., chronic inflammation) 47 could increase the risk of pancreatic cancer in individuals carrying ATM germline mutations.

Methods
Mice. Generation of Atm +/LoxP , Ptf1a +/cre and LSL-Kras G12D mice was described previously 13,48,49 . Mice were maintained in a mixed C57BL/6/NMRI genetic background. Animals were monitored at least twice weekly for signs of illness and euthanized when they had early signs of distended abdomen or lethargy. Weight loss was not observed at this stage. At necropsy hemorrhagic ascites and several metastatic nodules in the liver, diaphragm and pleural surfaces (mainly in KCATMΔ+ and KCATMΔΔ animals) always accompanied the distended abdomen. Immunohistochemistry. Immunostaining on paraffin and frozen was performed as previously described 14 .
All primary and secondary antibodies used in this study are listed in Supplementary Table 3. Biotinylated anti-rabbit IgG antibodies were detected by using the VECTASTAIN Elite ABC kit (Vector Laboratories). Prolong Gold with DAPI (40,6-diamidino-2-phenylindole; Invitrogen) was used for nuclear staining. Images were obtained with a Zeiss Axioskop 2 microscope, or with a confocal/Multiphoton laser-scanning Zeiss LSM 510 META microscope.
SA-β-gal/Ki67 staining. Detection of SA-β-gal was performed using published methods 50 . Following this step, sections were post-fixed with 4% PFA (10 min, RT) and processed for immunostaining using anti-Ki67 antibodies. Sections were counterstained with nuclear Fast Red.
Morphometric analysis. Low and high power confocal images were analyzed using Image J and γH2AX-Ecadherin double positive cells were counted using the "cell counter" plugin. High power confocal images were used to count γH2AX + foci in at least 100 cells from each primary cell line. Low magnification images covering the entire pancreas were captured using the Aperio scanner and Alcian Blue positive lesions were counted using Image J (3 or more sections spanning the whole pancreas were used per genotype).

Primary cell lines. Primary pancreatic ductal cell cultures were established from KC, KCATMΔ+ and
KCATMΔΔ primary and metastatic tumors, according to established protocols [16][17][18] . Cells were passaged serially 4-5 times to insure homogeneity before analysis. All in vitro analyses were performed on low passage number lines (<P8).
Quantitative PCR and RNA isolation. Total RNA was extracted using TRizol reagent (Thermo Fisher Scientific) and PureLink mini kit (Invitrogen). The purity and integrity of the isolated RNA were determined using a spectrophotometer NanoDrop One/One (Thermo Fisher Scientific). cDNA synthesis was performed using Protoscript first strand cDNA synthesis kit (New England Biolabs) and random hexamers according to the manufacturer's instructions. iTaq ™ Universal SYBR ® Green Supermix (Bio-rad) was used for qPCR and the detection was performed in a Mastercycler ® RealPlex2 (Eppendorf). Gene expression was normalized against the expression of β-Actin using the Relative Standard Curve Method. Results were expressed as the Mean ± SE (standard error) from three different replicates. The primers used in this study are listed in Supplementary Table 4. γ-irradiation of cultured cells. Cells after passage 4 were plated in 24-well plates at a concentration of 50,000 cells per well. The cells were irradiated using a 137 Cs-source (Gammacell 40-exactor, MDS Nordion, Ottawa, Ontario, Canada) at a dose of 4Gy and collected at the designated time points.
Clonogenic assay. Clonogenic assays on tumor cells exposed to different x-ray doses (RS 2000 X-ray Irradiator, Rad Source) were performed following published methods 51 . Briefly, confluent 100 mm plates of primary tumor cells from KC, KCATMΔ/+ and KCATMΔΔ mice were trypsinized, counted with a hemocytometer and diluted with complete media to obtain 160, 400 or 800 cells per well of each genotype respectively. Colonies were stained with crystal violet after 7 days, and those containing at least 50 cells were counted as surviving colonies. The plating efficiency and the survival fraction for each cell line after each treatment were calculated as described in Franken et al. 51 . Survival was calculated in comparison with non-treated samples, using an average of three determinations of the same dose rate of cells (±SE).
Western blot. Cells (~800,000) from primary tumor and liver metastasis were seeded in 60 mm plate dishes.
After 22 hours the culture medium was changed and 2 hours later the cells were irradiated (4Gy) using an RS 2000 X-ray Irradiator (Rad Source). Cell extracts were collected at 0, 1, 4 and 24 hours post-irradiation. Cells were harvested in cold PBS, and then lysed in 250 µL of RIPA Buffer (Sigma Aldrich). Protein quantification was performed using the Bradford reagent (Bio Rad) and samples were prepared by adding NuPAGE LDS Sample Buffer (Thermo Fisher Scientific) and β -mercaptoethanol. 30 µg of whole cell extracts were loaded into 10% NuPAGE Bis-Tris gel (Thermo Fisher Scientific) and electrophoresed at 110 volts in NuPAGE MES SDS Running Buffer (Thermo Fisher Scientific). Proteins were transferred to PDVF membrane using 10% Methanol NuPAGE Transfer Buffer (Thermo Fisher Scientific) for 16 hours at 30 volts and 4 °C. Primary antibodies were incubated 2 hours at room temperature and secondary donkey Horseradish Peroxidase-conjugated antibodies (Jackson) were incubated 1 hour at room temperature. Membranes were developed using SuperSignal ™ West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific). The primary antibodies used here are listed in Supplementary Table 3. Three independent hybridizations were performed for each primary antibody. The corresponding bands were scanned and quantified by densitometry using the CLIQS software from TotalLab (www.http://totallab.com/). The median ratio of p-p53/total p53, p-Kap1/total Kap1 or p21/β-actin from the three experiments ± SE (standard error) was calculated.

Histochemistry. Alcian Blue staining for visualization of PanINs' mucinous content and Picrosirius
Red staining for collagen fibers were performed according to manufacturer's instructions (Abcam).
Immunohistochemically and histochemically stained slides were further scanned with an Aperio ® slide scanner (Leica).
Immunofluorescence of cultured cells. Cells were cultured, fixed and stained as previously described 14 . Fluorescence in situ hybridization (FISH). Cells were grown to 60-70% confluence, treated with colcemid (0.02 µg/mL) for a 4 h incubation at 37 °C, and harvested using routine cytogenetic methods. The cells were then trypsinized for 5 minutes at 37 °C and centrifuged at 900 RPM for 5 minutes. The cell pellet was resuspended in 0.075 M KCL for 8 minutes at RT and then centrifuged at 600 RPM for 5 minutes, resuspended in 3:1 Carnoy's fixative (3 parts methanol: 1 part acetic acid), and incubated for 15 minutes at RT. The cells were centrifuged at 600 RPM for 5 minutes, resuspended in Carnoy's fixative, and incubated for 10 minutes at RT. The cells were spread onto a glass slide using a Pasteur pipette. The slides were air dried at RT. Purified DNA from Atm (RP23-456L9/9A5.3) was labeled with a red-dUTP (AF594, Molecular Probes) and purified DNA from a chromosome 9 control (RP23-284E19/9A5.2) was labeled with a green-dUTP (AF488, Molecular Probes) by nick translation. The probes were combined and hybridized to interphase and metaphase cells derived from the two samples using routine cytogenetic methods. Cells were stained with DAPI and total of one hundred interphase nuclei were scored for the number of red and green signals per cell.
Spectral Karyotype (SKY) Analysis. Slides containing cells in metaphase were prepared as described before. Spectral karyotyping of chromosomes was performed using a mouse SkyPaint probe (Applied Spectral Imaging, Carlsbad, CA) per the manufacturer's recommendation. Probes were detected using Applied Spectral Imaging's concentrated antibodies detection kit (CAD), as described by the manufacturer. Cells were stained with 4, 6-diamidino-2-phenylindole (DAPI) and images were acquired with a Nikon Eclipse E600 fluorescence microscope equipped with an interferometer (Spectra Cube: Applied Spectral Imaging) and a custom designed filter cube (Chroma Technology Corporation, Rockingham, VT). HiSKY software version 7.2 (Applied Spectral Imaging) was used for SKY analysis.
Mutation analyses of Trp53 and Cdkn2a. For Trp53 exon sequence analysis, genomic DNA was isolated from 3 KC and 3 KCATMΔΔ primary cell lines (pancreatic tumor and liver metastases, <P8) using the Wizard DNA purification kit (Promega). For Trp53, p16 Ink4a and p19 Arf cDNA sequencing, total RNA was extracted and cDNA was synthesized as described 13 . Accuprime high fidelity polymerase (Life Sciences) was used for PCR. PCR products were purified using the QIAquick gel extraction kit (Qiagen science, Maryland, USA), and submitted for sequencing using the same forward and reverse primers used for amplification. Supplementary Table 5 lists the primers used for this analysis.
MTT assay. Human cell lines and murine PDAC and liver metastasis primary cell lines maintained in the logarithmic phase of growth were trypsinized, counted, and plated at 100 cells/well in 96 well plates in a total volume of 100 ml of complete medium. Twenty-four hours later, cells were irradiated and viability was assessed starting on the same day and for 6-7 more days using the Cell Titer Aqueous Viability assay (Promega) following the manufacturer's instructions. Cell titer aqueous is based on an improved version of the standard MTT assay where the MTS tetrazolium compound is bioreduced by cells into a colored formazan product that is soluble in tissue culture medium in contrast to medium-insoluble MTT product. Therefore, the colorimetric assessment of proliferation can be performed in one step.
Statistical Analyses. In all graphs, data are presented as mean ± SEM. Significance was accepted at a P value < 0.05. PRISM software was used to perform statistical analysis for all datasets. For survival analysis the Log-rank (Mantel-Cox) test was performed. For comparison of Alcian Blue or γH2AX staining results in the 3 genotypes, ordinary ANOVA one-way was performed. For Clonogenic and MTS data from different time points, ANOVA 2-way analysis was performed with Bonferroni's test for multiple comparisons.