Dynamic ROS Control by TIGAR Regulates the Initiation and Progression of Pancreatic Cancer

1Many studies and clinical trials have shown ROS to both enhance and retard tumor progression. This study seeks to clarify the effects of ROS over the full course of development of a single tumor type, pancreatic ductal adenocarcinoma. In this model we show that ROS limitation by the antioxidant protein TIGAR is important for the development of premalignancies, while ROS promotion can drive increased metastasis through activation of ERK. We also show that the effects of ROS on tumor invasiveness are reversible, allowing for the dynamic switching from a proliferative to an invasive phenotype, and back. This work rationalizes the complexity of ROS regulation during cancer progression and may help to guide the use of antioxidants in cancer therapy.

2Reactive oxygen species (ROS) play important and diverse roles in regulating many aspects of cell behavior, from signaling proliferation and survival to promotion of oxidative damage and cell death. Not surprisingly, therefore, control of ROS production and defects in antioxidant defense have been associated with many aspects of human health and disease. The contribution of ROS to cancer development has been somewhat controversial and is clearly highly complex. Numerous lines of evidence support the concept that ROS contribute to cancer initiation and development. ROS can induce both genotoxic damage and chronic inflammation, while membrane-associated ROS generated through NADPH oxidases (such as NOX4) are important contributors to the activation of signaling pathways that drive proliferation and metastasis. Furthermore, mitochondrial ROS were also shown to be necessary for KRAS-induced cancer development. It was therefore surprising that no beneficial effect of antioxidants was detected in cancer prevention studies, some of which even revealed an increase in cancer development in the antioxidant-treated group. Subsequent work has shown that many of the steps involved in malignant progression are associated with increased oxidative stress and an increased sensitivity of cells to succumb to ROS-induced death. Activation of antioxidant defense pathways is critical for successful tumor development, leading to a revised model in which ROS limitation may function to enhance tumorigenesis. Many studies have now supported the concept that tumor cells inherently carry a high burden of oxidative stress, reflecting abnormal oncogenic signaling and loss of normal environment. Survival of these cells depends on a concomitant increase in ROS scavenging pathways that is not needed in normal cells, suggesting that interference with these antioxidant pathways or simply additional ROS burden may selectively kill cancer cells. Interestingly, many commonly used chemotherapeutic agents effectively induce ROS.

3Many mechanisms through which tumor cells limit ROS exposure have been described. One of the most ubiquitous is the activation of NRF2, a transcription factor that regulates a program of genes involved in antioxidant defense. Activation of oncogenes such as KRAS or environmental signals such as lack of oxygen (hypoxia) can induce NRF2, with evidence that this response is necessary for tumor development. Direct activation of NRF2 through overexpression of the protein or loss of the negative regulator KEAP1 is frequent in cancer development. Another protein that contributes to ROS limitation is TIGAR, a protein with bisphosphatase activity that can support the activation of the oxidative pentose phosphate pathway (PPP) in response to oxidative stress and so enhance the production of NADPH for antioxidant defense. Functions of TIGAR that limit ROS have been shown to contribute to damage resolution in intestinal epithelium and protect from pathologies such as cerebral ischemia. However, elevated expression of TIGAR has also been detected in many cancer types, consistent with a role for antioxidants in tumor progression. In mice, loss of TIGAR leads to increased survival and slower tumor development in intestinal and lymphoma models.

4While ROS limitation has a clear role in supporting cancer development, the contribution of ROS to different stages of cancer development remains unclear. Early studies suggested that increased ROS may be linked to enhanced metastasis. However, several recent studies in melanoma and other tumor types have elegantly demonstrated an importance of ROS limitation in allowing metastatic spread. These observations are consistent with studies showing that detachment of cells from the extracellular matrix, as may occur when cells leave the primary tumor site and enter the circulation, results in increased ROS, which is limited by various metabolic adaptations. These recent data have led to the suggestion that ROS are required for early stages but suppress later stages of cancer development, a model supported by studies in melanoma.

5In this study, we examine the role of TIGAR in the development of pancreatic cancer. Using mutant KRAS-driven pancreatic ductal adenocarcinoma (PDAC) mouse models, we show that loss of TIGAR delays the emergence of premalignant pancreatic intraepithelial neoplasia (PanIN) lesions, but enhances the metastatic capacity of the tumor cells, leading to decreased survival. Tigar null tumors and cells show higher ROS levels and increased mesenchymal characteristics, accompanied by enhanced capacity for migration and invasion. The responses to loss of TIGAR are plastic and reverted by treatment of cells with antioxidants. Consistently, the pattern of TIGAR expression in both human and mouse PDACs also suggests a role for ROS limitation in the establishment of the primary malignancy and distant metastasis, with a role for enhanced ROS during the process of metastatic spread.

6To examine the role of TIGAR in the development of PDAC, we utilized well-established mouse models that use Pdx1-Cre to drive pancreas expression of mutant KRAS (LSL-KrasG12D/+) alone (KC) or mutant KRAS with mutant p53 (LSL-p53R172H/+; KPC) or with loss of p53 (p53fl/+; KFC). In the KFC model, CRE-mediated deletion of one p53 allele is accompanied by loss of the remaining wild-type allele during tumor development. Each of these models was crossed into a Tigarfl/fl strain to generate pancreatic tumors that retained Tigar expression (CTR) or deleted (KO) for Tigar. Initial analysis of preneoplastic PanIN in the KC model showed that loss of TIGAR delayed the appearance of each stage of PanIN progression (PanIN1, 2, and 3), accompanied by lower proliferation in the Tigar null lesions, measured by Ki67 staining (Figures 1A–1D). Using the KFC model, PanIN lesions were detected more rapidly, and again, the loss of TIGAR retarded the appearance of PanIN and lowered proliferation of these preneoplastic lesions (Figures 1E–1H). These results are consistent with our work showing that loss of TIGAR delayed the appearance of intestinal adenomas in response to APC loss and previous work showing decreased PanIN development following loss of the antioxidant factor NRF2 in a PDAC model. Using anti-malondialdehyde (MDA) staining of peroxidized lipids as a marker of oxidative stress, we confirmed an increase of ROS in the Tigar KO PanINs (in the KC and KFC models) as well as Tigar KO PDAC (in the KFC model) (Figures 1I–1L). Cell lines were derived from tumors from three Tigar wild-type (C1, C2, C3) and three Tigar-deleted (K1, K2, K3) KFC mice. Consistent with an antioxidant role for TIGAR, mitochondrial ROS levels were increased in the Tigar KO cell lines and could be lowered by treatment with the antioxidant N-acetyl-L-cysteine (NAC) (Figure S1A). The Tigar KO cells also showed increased death following exposure to the ROS-inducing chemotherapeutic Adriamycin (Doxorubicin), which was limited by treatment with NAC (Figure 1M). Importantly, introduction of recombinant TIGAR to the Tigar null cells (Figure S1B), which decreased ROS levels in Tigar KO cells (Figure S1C), also rescued the sensitivity to Adriamycin (Figure 1M). TIGAR has been shown to support flux through the oxidative PPP, which generates NADPH for antioxidant defense. Both oxidative and non-oxidative PPPs produce ribose 5-phosphate (R5P), and previous studies have shown that these mutant KRAS-expressing PDACs increase R5P generation through the non-oxidative pathway. Interestingly, no consistent differences in R5P levels were detected between Tigar wild-type or null cells (Figure S1D), suggesting that any defect in oxidative PPP in Tigar null cells is compensated for by an increase in non-oxidative PPP flux. Taken together, these results show that TIGAR limits oxidative stress, a function that correlates with the ability of TIGAR to support the initial stages of PDAC development.

7Whereas tumors in the KC mice progress rather slowly, PDAC development is more rapid in KFC and KPC mice. Surprisingly, however, despite the delay in PanIN development, loss of TIGAR reduced survival of both KPC and KFC mice (Figures 2A and 2B). This reduced survival was not reflected in any obvious difference in the differentiation status of CTR and KO primary tumors (Figure S2A), but was accompanied by widespread tumor dissemination to multiple organs. Staining for CK-19 confirmed that the lung lesions represented metastatic spread of PDAC (Figure S2B). In the KPC model, where most CTR mice showed evidence of metastases, the additional effect of TIGAR loss did not reach significance (Figure 2C). Nevertheless, the total number of organs showing evidence of metastasis was significantly increased by Tigar deletion (Figures 2F and S2C). In the less metastatic KFC model, loss of TIGAR resulted in significantly more mice carrying metastases (Figures 2D and 2G), with a striking increase in the number of lung lesions (Figure 2E). In support of these observations, loss of the antioxidant defense protein NRF2 in KPC tumors (which has been shown to reduce PanIN development due to increase of ROS; ) also failed to extend overall survival (Figure 2H) and PDAC-free survival (Figure S2D) and instead promoted increased lung metastases (Figure 2I). (In this NRF2 KPC model, the mutant p53 allele was LSL-p53R270H.) Taken together, our data indicate that increased oxidative stress due to loss of TIGAR in KFC and KPC pancreas cancer models delays initial tumor development but enhances metastatic progression at later stages.

8Previous studies have shown that metastatic capacity in PDAC tumors is associated with the acquisition of a more mesenchymal phenotype. While the exact contribution of various drivers of epithelial-mesenchymal transition (EMT) has been a topic of discussion, more recent data support the importance of a switch from an epithelial phenotype—characterized by E-CADHERIN staining—to a more mesenchymal, VIMENTIN-positive appearance of the cancer cells in driving PDAC metastasis. Histological analysis of KFC tumors null for Tigar (KO) showed a reduction of E-CADHERIN expression (Figures 3A and 3B) and an increase in VIMENTIN expression (Figures 3C and 3D) compared with KFC tumors that retained TIGAR expression (CTR). This shift in phenotype corresponded with increased expression of the mesenchymal marker SLUG in the KO tumors (Figures 3E and 3F). In support of these observations, loss of the antioxidant protein NRF2 also led to a reduction in E-CADHERIN expression (Figures S3A and S3B) and an increase in SLUG expression (Figures S3C and S3D) in PDAC tumors. In culture, cell lines derived from Tigar null KFC tumors also showed a clear shift to a more mesenchymal phenotype (Figure 3G), losing E-CADHERIN expression while gaining expression of SLUG and SNAIL (another marker associated with a mesenchymal phenotype) (Figure 3H). Immunofluorescence analysis confirmed the switch from E-CADHERIN to VIMENTIN expression in the Tigar null PDAC cells (Figure 3I). Consistent with the acquisition of a mesenchymal phenotype and increase in metastasis, Tigar null cells showed an increased rate of wound healing in a scratch assay (Figure 3J) and increased migration and invasion in transwell assays (Figure 3K). Reintroduction of exogenous TIGAR in these cells reversed the enhanced invasion (Figures S3G and S3H). These results indicate that TIGAR deficiency allows PDAC cells to increase invasiveness and switch to a more mesenchymal phenotype.

9Previous studies have described numerous mechanisms responsible for the switch to a mesenchymal phenotype in metastatic and therapy-resistant tumor cells, several of which have been shown to be induced by increased ROS. Analysis of pNF-κB, pSRC, pSTAT3, pAKT, and HO1 expression failed to demonstrate a convincing and consistent upregulation of the NF-κB, SRC, STAT3, AKT, or hypoxia pathway in the Tigar KO cells compared with CTR cells (Figure S4A). Furthermore, there were no consistent changes in the expression of proteins encoded by HIF1 target genes such as GLUT1 and BNIP3 (Figure S4B). Previous studies have suggested that depletion of TIGAR results in a decrease in MET expression in lung cancer cell lines, but we were unable to detect any difference in MET expression in our PDAC cells (Figure S4B). Recent studies have also shown that increased ROS lead to the degradation of BACH1 to limit metastasis, but this was not evident in the TIGAR null PDAC cells (Figure S4B). These data suggest that direct regulation of MET or BACH1 expression does not play a role in the response to loss of TIGAR in our model, although they do not preclude a more general involvement of these signaling pathways. By contrast, a clear increase in phosphorylated ERK was seen in Tigar KO cells, accompanied by a decreased expression of DUSP6/MKP-3, the phosphatase responsible for dephosphorylating and inactivating ERK (Figure 4A). This shift in expression correlated with the switch from E-CADHERIN to VIMENTIN expression in these cells (Figure 3I). A similar increase in pERK and decrease in DUSP63 expression was also seen in the PDAC lesions in vivo (Figures 4B–4E). Consistently, inhibition of the ERK signaling pathway using the MAPKK inhibitor PD98059 led to a reversion of Tigar null PDAC cells to a more epithelial phenotype (Figure S4C). Furthermore, a similar increase in phospho-ERK (Figures S3E and S3F) was also observed in NRF2 null PDAC tumors.

10To determine the effect of ERK activation on the invasive capacity of Tigar KO PDAC cells, we treated the cells with the ERK pathway inhibitor PD98059. The enhanced wound closure, migration, and invasion seen in cells derived from Tigar null PDACs were significantly decreased in PD98059 treated cells (Figures 4F–4K). Tigar null tumor cells showed an increased collagen-degrading activity indicative of increased invadopodia function (Figures 4L and 4M) that was also dependent on ERK pathway activity. To confirm the contribution of ERK in driving invasion, we showed decreased migration of Tigar null PDAC cells depleted of ERK by small interfering RNA (Figures S4D–S4F). Re-expressing DUSP6 in Tigar null cells decreased ERK phosphorylation (Figure S4G), as expected, and also decreased the migration of these TIGAR-deficient cells (Figures 4N and 4O). These data are consistent with a model in which Tigar null PDAC cells enhance MAPK signaling through a decrease in DUSP6 expression, which drives enhanced migration.

11Having shown increased ROS levels in tumors and cell lines deficient in TIGAR (Figures 1I–1L and S1A) and enhanced sensitivity to oxidative stress for Tigar null PDAC cell lines (Figure 1M), we sought to determine the contribution of ROS to the acquisition of the invasive phenotype of these cells. Treatment of Tigar KO cells with the antioxidant NAC reduced migration in both wound-scratch and transwell migration assays (Figures 5A–5D), while also limiting invasion (Figures 5E and 5F) and collagen degradation (Figures 5G and 5H). Importantly, the background levels of migration and invasion shown by Tigar expressing cells were not significantly decreased by NAC treatment (Figures 5A–5H). This activity in Tigar KO cells was also clearly reduced in response to limitation of mitochondrial ROS following treatment with mito-TEMPO (Figures S5A and S5B), suggesting that mitochondrially derived ROS are responsible for the increase in invasiveness in TIGAR-deficient cells. A dose of piercidin that inhibits complex I and so lowers mitochondrially derived ROS also decreased migration in TIGAR-deficient cells (Figures S5C and S5D). Additional treatment with antimycin, which re-establishes mitochondrial ROS production by inhibiting complex III, abrogated this reduction in migration (Figures S5C and S5D), underscoring the importance of mitochondrially derived ROS in the migratory phenotype in TIGAR-deficient cells. Long-term treatment with NAC (Figures 5I and 5J) also reverted the phenotype of the cells, which regained E-CADHERIN and DUSP6 expression and lost SNAIL expression and phosphorylated ERK. TIGAR KO cells also regained E-CADHERIN and lost SNAIL expression after long-term treatment with mito-TEMPO (Figures S5E and S5F). Interestingly, the ROS-dependent control of PDAC morphology and function was highly plastic. Tigar KO cells that reverted to epithelial morphology and reduced migration/invasion in response to long-term antioxidant treatment were able to regain these features following removal of the antioxidant (Figures 5I–5L, S5E, and S5F). Taken together, these results show that increased ROS in response to loss of TIGAR can promote a mesenchymal shift accompanied by increased invasive capacity, but that modulating ROS levels can allow the tumor cells to toggle between the two phenotypes.

12To test directly whether loss of TIGAR and ROS regulation could have an impact on metastatic capacity, we turned to an experimental model of metastasis in which lung colonization of tumor cells following tail vein injection was assessed. Tigar null PDAC cells showed a clearly increased lung colonization capacity compared with Tigar wild-type cells (Figures 6A and 6B), which was decreased following treatment of cells and mice with NAC (Figures 6A and 6B). As expected, Tigar KO metastases showed increased ROS (measured by MDA staining) (Figure S6A) that was limited following NAC treatment. Consistent with results in cell lines (Figures 4 and 5), decreased DUSP6 expression and increased phosphorylated ERK were detected in the Tigar null lung lesions, a response that was reversed by NAC (Figures 6C–6F). Intriguingly, however, proliferation rates in the Tigar KO lung deposits were significantly lower than those seen in the Tigar wild-type lesions, a response that was also reversed, to some extent, by NAC treatment (Figures 6G and 6H). While the data are consistent with a ROS-driven increase in the ability of Tigar null PDAC cells to colonize the lung following tail vein injection, this decrease in proliferation may indicate that maintenance of high ROS levels becomes detrimental to the proliferation of the lung lesions once the cells have become established at this site.

13Immunohistochemical analysis of TIGAR at various stages of PDAC tumorigenesis showed an increase in TIGAR expression during the early stages of tumor development in both the KFC mouse model (Figures 7A and 7B) and human PDAC samples (Figures 7C and 7D), consistent with a role for TIGAR in limiting ROS and promoting the survival of these preinvasive cells. However, in both mouse and human cancers, progression to invasive primary tumors was accompanied by a clear decrease in TIGAR expression (Figures 7A–7D), consistent with the selection for cells with higher ROS and higher invasive capacity during these stages of tumorigenesis. Interestingly, TIGAR levels were found to be slightly increased in the metastatic deposits from these tumors, consistent with a role for TIGAR re-expression in limiting ROS and supporting proliferation in the established metastases. Analysis of ROS levels in these different tumor stages showed the expected correlation between high TIGAR and low ROS in PDAC lesions, with higher ROS accumulation in later stage, invasive tumors expressing lower TIGAR levels (Figures 7E and 7F).

14Many studies in various models have indicated a role for both ROS limitation and ROS promotion in driving tumor initiation and metastatic spread. We showed previously that the antioxidant protein TIGAR can support cell survival and proliferation during tissue regeneration and adenoma development, and several studies have shown increased TIGAR expression in various tumor types. Using a pancreatic tumor model, we show that deletion of Tigar drives increased ROS and a decrease in the development of premalignant PanIN lesions, consistent with an enhanced proliferative capacity of cancer cells with higher TIGAR expression. However, despite the limitation of tumor initiation, Tigar null lesions showed an increased ability to metastasize, especially to the lung. Enhancing ROS through loss of NRF2 in a similar tumor model also limited PanIN development while promoting enhanced metastasis to the lung. Increased metastatic capacity of the Tigar null cells was mirrored by the acquisition of a more mesenchymal phenotype, activation of ERK signaling through the decreased expression of the phosphatase Dusp6/MKP-3, and increased migration and invasion in vitro. Activation of ERK, while associated with tumor proliferation, migration, and survival, can also limit growth through mechanisms such as senescence or increased sensitivity to other stresses, especially when hyperactivated in the context of an increase of ROS level. It seems likely that DUSP6 expression can be fine-tuned to modulate ERK at different stages of tumor progression and that this ability is lost in TIGAR null cells. Interestingly, expression of the phosphatase Dusp6/MKP-3 has been shown previously to be lost in some human PDAC samples, and its expression can be regulated by ROS. ROS limitation reversed the more aggressive phenotype of these cells and reduced the experimental metastatic capacity in vivo.

15Our data indicate a complex role for ROS in regulating cancer initiation, growth, and metastatic progression that may help to explain some of the apparently contradictory results seen in previous studies. It is clear that loss of antioxidant defense can limit some stages of tumor development while enhancing others. The exact response of any tumor system is likely to depend on numerous factors, including the cell of origin, the genetic landscape of the tumor cell, and the tumor environment. For example, a failure to limit ROS is clearly detrimental to the metastatic ability of melanoma and lung tumor cells, while ROS limitation is associated with enhanced EMT in breast cancer models. The contribution of different environments in determining the response to ROS is also evident in the increased metastasis seen in both Tigar and Nrf2 deletion models. In cell culture models, the ability of tumor cells to survive matrix detachment—an event that is thought to mirror the success of cells in surviving in the circulation—has also been related to an ability to modulate metabolism to limit ROS. However, in the PDAC model described here, enhanced ROS clearly promotes the acquisition of certain mesenchymal phenotypes that can be important in metastasis.

16The differential ability of ROS regulation to modulate different steps in the progression of a single tumor type highlights the challenges that would accompany the development of anti- or pro-oxidant approaches for cancer therapy. Although the response to ROS may reflect the overall level of oxidative stress, previous studies have shown that different ROS species, or different locations of ROS production in cells, can have differential effects on proliferation and survival. Whether the responses to ROS during tumor progression reflect overall levels of ROS, differences in the cell's ability to respond to ROS or a subtler contribution of different types of ROS that may be controlled by different pathways remain to be determined. We note that in this and previous studies, TIGAR was shown to more effectively limit mitochondrial than cytosolic ROS. Whereas in intestinal regeneration studies, this activity clearly functioned to support cell survival and proliferation (and thereby tissue repair), in the context of PDAC progression, the limitation of mitochondrial ROS by TIGAR dampens the activation of signaling pathways that promote migration and invasion.

17Our study underscores the multifaceted role of ROS in controlling disease progression. Therapies that either increase or decrease ROS could lead to very different outcomes in different tumor types or at different stages of cancer progression. In intestinal adenomas, which are not metastatic, deletion of TIGAR limits the development of the tumor and improves survival. However, despite the delay in premalignant tumor development in the PDAC model, the capacity of these cells to metastasize is enhanced by TIGAR deletion and decreases overall survival. We speculate that while therapies to increase ROS could be beneficial in some contexts (for example, in lung cancer or melanoma), an initial effect of such treatment on locally confined tumors (such as pancreas) could lead to an inadvertent increase in invasiveness, which is the more likely cause of death. Additional studies to determine the roles of different antioxidants in other genetic mouse models of cancer will be important to inform therapy decisions relating to ROS level manipulation and the timing/sequence of combined treatments involving ROS.

18In the PDAC model, ROS can be a powerful regulator of cell phenotype and behavior, allowing cells to toggle between an epithelial/less invasive state to a more mesenchymal/invasive state. The plasticity of this switch is of interest, suggesting that the response is driven by events such as modulation of chromatin modifications that can be reversed when the oxidative signal is removed, and it is important to note that the switch in phenotype requires several weeks in culture in the presence or absence of an antioxidant. A wealth of previous data shows that a switch to a more mesenchymal phenotype can promote successful metastasis. However, the ability to reverse this transition—to undergo mesenchymal-to-epithelial transition—has been linked with the success of cancer cells to expand and proliferate once they have reached and become established at a distant metastatic site. The ability of ROS regulation to toggle cells between these two states suggests this could be an important mechanism to regulate many steps of malignant progression. Our analyses of TIGAR expression correlate with a role for ROS limitation in the outgrowth of both the premalignant and the metastatic lesions (selecting for cells with high TIGAR expression), but a contribution of increased ROS in cells during the process of invading or moving to a distant site (selecting for cells with lower TIGAR expression). Our data also hint at a role for a second round of TIGAR upregulation to support proliferation of cells once they are established at the metastatic site. Further studies will be required to understand the mechanisms of TIGAR regulation that can result in this complex pattern of expression.

19The Lead Contact is Karen H Vousden (karen.vousden@crick.ac.uk). All unique/stable reagents generated in this study are available from the Lead Contact with a completed MTA. Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact Karen H Vousden (karen.vousden@crick.ac.uk).

20All animal experiments were performed under the Animals (Scientific Procedures) Act 1986 in accordance with UK Home Office licenses (Project License 70/8645, P319AE968) and the EU Directive 2010 and sanctioned by local ethical review process (University of Glasgow and The Francis Crick Institute). Mice were housed in an area free of pathogens as defined by FELASA recommendations in IVC ages at 5 per cage at constant temperature (19-23°C) and humidity (55% ± 10%), with a 12-hour light/dark cycle (lights on at 7:00 am) and were allowed access to food and water ad libitum. Mice were allowed to acclimatize for at least 2 days (for mice bred on site) or 7 days (for imported mice) prior to the experiment and were randomly assigned to experimental groups. For the PDAC GEMMs, both male and female mice were used, roughly matched between CTR and KO groups. For the tail vein lung colonization models only female mice were used. Mice had not been involved in any previous procedures.

21Trp53+/LSL-R172H, Kras+/LSL-G12D, Trp53+/fl, Pdx1-Cre strains were interbred to obtain KC (Pdx1-Cre; Kras+/LSL-G12D), KFC (Pdx1-Cre; Kras+/LSL-G12D; Trp53+/fl), and KPC (Pdx1-Cre; Kras+/LSL-G12D; Trp53+/LSL-R172H) mice. To introduce TIGAR deficiency in these models, TIGARfl/fl strain was used to breed into the above strains to obtain Tigarfl/+ or Tigar+/+ for control (CTR) and Tigarfl/fl for Tigar knockout (KO) in KC, KFC and KPC in a mixed background. Pdx1-Cre; Kras+/LSL-G12D; Trp53+/LSL-R270H; Nrf2-/- mice were also used. Mice were monitored two times weekly and tissues were collected when exhibiting symptoms of PDAC.

222X105 PDAC KFC cells per mouse (n≥4 in each group) in 100μl PBS were injected (tail vein) into athymic nu/nu mice (Jackson Laboratory). After 14 days, lung tissues were collected for histological analysis. For antioxidant treatment, a week before the injection NAC (N-Acetyl-L-cysteine, Sigma A7250) was administered to the mouse (1g/L drinking water, pH 7) and throughout the duration of experiment. PDAC cells were pre-treated overnight with NAC before trypsinized and used at the day of injection.

23PDAC tumor cell lines were derived from the KFC tumors from three TIGAR WT and three TIGAR KO animals. Tumor tissues were collected in PBS with 1% penicillin-streptomycin and then finely minced. Minced tissues were then incubated with collagenase type 1 (200U/ml, Gibco) and dispase (2.4U/ml, Gibco) in HBSS for 1 hour in 37°C for cell dissociation. After washing 2X in HBSS, cell pellets were resuspended and grown in growth media (Dulbecco's modified Eagle medium containing 10% fetal bovine serum, 2 mM l-glutamine, 1% penicillin-streptomycin).

24Transwell Migration assays (Corning® BioCoat™ Control Insert-No ECM, 8 micron pore size) and Invasion assays (Corning® BioCoat™ Matrigel Invasion chamber) were performed according to the manufacturer’s instructions. Briefly, cells were pre-treated overnight in 1% serum with or without the indicated drugs/treatments before seeding onto the upper chamber of the transwell the following day in the presence of the drug treatment. Media with 10% serum (with and without the drugs) were used in the lower chamber. After 16 hours, cells that remained on the top of the membrane were removed by a cotton-tipped applicator. Cell that were migrated/invaded to the bottom of the insert were then fixed in 70% ethanol and stained with 0.5% Crystal Violet. Migrated/Invaded cells were photographed under an inverted microscope, quantified using ImageJ, and represented as percentage of total area.

25Confluent monolayer of cells was scratched using a p20 pipette tip to create a scratch. Debris were removed by washing the cells gently with 2X complete media. Image were taken at the start of the assay and after 16 hours under a phase contrast microscope. The width of the gap was measured by ImageJ and the reduction of the width is represented as percentage (%) wound closure.

26Tissues were fixed in 10% neutral buffered formalin and were embedded in paraffin and processed by standard histological techniques. PanIN was defined by previously published guildelines in genetically modified mouse models of PDAC. Briefly, PanIN is a lesion that arises in native pancreatic ducts measuring <1 mm and not on a background of acinar-ductal metaplasia. PanIN lesions are graded as PanIN-1, 2 or 3 according to the cytological and architectural characteristics as described previously. Quantitation of PanIN number was done on 5 of 20X fields of view from at least 4 mice. Heat induced epitope retrieval with sodium citrate buffer (Antigen Unmasking Solution, Citric Acid Based, Vector Laboratories, cat # H-3300) followed by blocking endogenous peroxidase and Avidin/Biotin (BLOXALL Endogenous Peroxidase and Alkaline Phosphatase Blocking Solution, Vector Laboratories, cat # SP-6000) were used for immunohistochemistry prior to primary antibody incubation (diluted in 10% normal horse serum in 1XTBST, 4°C overnight). For immunohistochemistry, primary antibodies used were anti-Ki67 (1:1000 Thermo Scientific SP6), anti-MDA (1:300, Abcam Ab6463), anti-TIGAR (1:500 Millipore AB10545), anti-phospho-ERK (Cell Signalling), anti-DUSP6 (1:300 Abcam Ab76310), anti-Snail (1:300 Cell Signalling #3879), anti-Slug (1:300 Cell Signalling #9585), anti-E-Cadherin (1:300 Cell Signalling), anti-Vimentin (1:300, Cell Signaling #5741), anti-Cytokeratin 19 (CK-19) (1:500, Abcam Ab52625). Expression levels were scored based on staining intensity and area of tumor cells using a weighted histoscore calculated from the sum of (1 × % weak staining) + (2 × % moderate staining) + (3 × % strong staining).

27TMA of human pancreatic cancer was obtained from US Biomax (PA2081b) and was stained with anti-human TIGAR antibody.

28Cell death was quantified using LIVE/DEAD Viability Kit (Molecular Probes) 18 hours after adriamycin (1μg/ml, Sigma) alone or with either NAC (1mM) or recombinant TIGAR (rTIGAR, 5μg/ml, Peprotech). Mito tempo (50μM, Sigma SML0737), Piericidin (1μM, Caymen Chemical 15379), Antimycin A (1μM, Caymen Chemical 19433) and PD98059 (50μM, Tocris cat# 1213) were used at the indicated times and duration. Mitochondrial ROS was measured by MitoSOX™ Red Mitochondrial Superoxide Indicator (Invitrogen cat # M36008). Protein lysates were isolated in RIPA-buffer (Millipore) with complete protease inhibitors (Roche) and phosphatase inhibitor cocktail (Thermo Fisher Scientific), volume adjusted according to protein concentration measurements (Quick Start™ Bradford 1X Dye Reagent, Bio-Rad, Cat #500-0205), separated using precast NuPAGE 4-12% Bis-Tris Protein gels (Invitrogen, Life Technologies), and transferred to nitrocellulose membranes. ECL chemiluminescence detection kits (Pierce) with appropriate species-specific horseradish peroxidase−conjugated secondary antibodies were used to detect the proteins. Each blot shows one representative out of at least three. Antibodies used are: anti-TIGAR (1:1000, Santa Cruz sc-67273), anti-p-ERK1/2 Thr202/204 (1:1000, Cell Signaling), anti-DUSP6 (1: 500 Abcam Ab76310), anti-Snail (1:1000, Cell Signaling #3879), anti-Slug (1:1000, Cell Signaling #9585), anti-E-Cadherin (1:5000, Cell Signaling #3195), anti-p-FAK Tyr576/577 (1:1000, Cell Signaling #3281), anti-p-Akt Ser 473 (1:1000, Cell Signaling #13038), anti-GCLC (1:1000 Abcam Ab190685), anti-HO1 (1:500, Santa Cruz sc10789), anti p-NF-kB p65 (1:1000, Cell Signaling #3033), anti-p-Src family (1:500, Cell Signaling #6943), anti-p-Stat3 Tyr 705 (1:1000, Cell Signaling #9145), anti-Actin (1:10000, Abcam Ab20272), anti-ERK (1:1000, Cell Signaling #9102), anti-BACH1 (1:1000, Santa Cruz sc-271211), anti-BNIP3 (1:1000, Cell Signaling #3769), anti-MET (1:1000, Cell Signaling #8198), anti-Vinculin (1:1000, Santa Cruz sc-25336), anti-Histone H3 (1:1000, Cell Signaling #9715).

29Re-expression of DUSP6 was achieved by transfecting a mouse tagged ORF clone of DUSP6 (Origene, cat#MR222688) with empty vector as control. Knockdown of ERK1/2 was achieved by transfecting siRNA targeting ERK1/2 (Origene, cat#SR412074, 412814) with the corresponding scrambled siRNA as negative control. Transfection was performed using Lipofectamine 2000 for DNA (Invitrogen) or Lipofectamine RNAiMAX for siRNA (Invitrogen) according to the manufacturer’s instructions.

30Cells were fixed in 4% paraformaldehyde in 1X PBS, followed by permeabilization with 0.4% Triton X 100 in 1XPBS with 10% normal donkey serum. Primary antibody was prepared in 1XPBS with 0.4% Triton X-100 and 2% normal donkey serum. Antibody used were E-Cadherin (1:500, BD Biosciences, #610182) and Vimentin (1:500, Cell Signaling, #5741). Fixed cells were incubated overnight in 4°C with the primary antibody, washed 3X in 1XPBS, followed by incubation of secondary antibody for 1 hour at room temperature (1:500 in 1XPBS with 0.4% Triton X-100 and 2% normal donkey serum, Alexa Fluor 488 donkey anti-mouse Alexa Fluor 488 or Alexa Fluor 594 donkey anti-rabbit, Thermo Fischer Scientific).

31Invadopodia assay (QCM™ Gelatin Invadopodia Assay, Millipore) was performed according to the manufacturer’s instructions. Briefly, glass chamber slides were coated with Poly L Lysine, activated by a diluted glutaldehyde solution, and then fluorescently coated with fluorescent gelatin as the substrate for invadopodia. After disinfecting with 70% alcohol and quenching of free aldehydes with growth medium, cells were seeded onto the gelatin surface for 16 hours in complete media. Cells were then fixed in 4% formaldehyde in DPBS and visualized by nuclear (DAPI) and cytoskeleton staining (TRITC-Phalloidin) by fluorescent microscopy. Degraded area/puncta of fluorescent gelatin (devoid of green fluorescence) indicated the presence of invadopodia and was quantified per number of cells (at least 100 cells were analysed per experiment).

32Cells were seeded in 6 well plates at a density of 1x105 in DMEM. The media was refreshed after 24 hours and replaced after 48 hours with media containing 1,2-13C2-glucose. Cells were cultured in this media for 5 hours before harvesting the cells for metabolomics and isotope tracing analysis. Cells were harvested by removing the media and rinsing with cold PBS before cells were lysed by adding cold (-20°C) extraction buffer containing methanol, acetonitrile and water (50:30:20) directly on the cells followed by scraping and collecting everything in a clean 1.5 ml centrifuge tube. The tubes were vortexed for 1 min before centrifugation at full speed for 15 minutes at 4°C. The supernatant was collected into mass spectrometry tubes and analysed by liquid chromatography high resolution mass spectrometry as described before. Briefly, liquid chromatography was performed on a Dionex Ultimate 3000 LC system coupled to a Q Exactive mass spectrometer (Thermo Scientific). Metabolites were separated on a Sequant ZIC-pHILIC column (2.1 x 150 mm, 5mM) (Merck) with mobile phase A consisting of 20 mM (NH4)2CO3, 0.1% NH4OH in H2O and mobile phase B consisting of 100% acetonitrile. A linear gradient from 80% to 20% A was applied over 17 minutes at a flow rate of 200 ul/min. Eluents were ionized in a HESI probe connected to the Q Exactive which scanned a mass range between 75 and 1000 m/z with polarity switching. Data were analysed with using Thermo TraceFinder software.

33Data were analysed using GraphPad Prism 7 software (GraphPad Software). The survival data were analysed by log-rank Mantel-Cox test. Fisher’s exact test was used to compare frequency of metastasis. Other data represent mean values ± SEM from at least three independent experiments (n ≥ 3). Student’s t test (comparisons between two groups), one-way ANOVA with Tukey post hoc (comparisons of three or more groups with one independent variable) were used as indicated in the legends. p <0.05 was considered statistically significant.

34This study did not generate any unique datasets or code.