ROS/KRAS/AMPK signaling determines gemcitabine-induced pancreatic cancer stem-like cell property in vitro

Poor prognosis of pancreatic cancer (PanCa) is partially due to chemoresistance to gemcitabine (GEM), the standard chemotherapy. Dysregulated glucose metabolism is revealed to contribute to therapeutic resistance and pluripotent state. Cancer stem cells (CSCs) rewire cellular metabolism to survive the chemotherapeutics and initiate cancer relapse. However, few studies focus on the effect of GEM on cancer cell metabolism, the stemness status, and the involved mechanisms which are crucial in PanCa treatment. In this study, we demonstrate that GEM treatment induces metabolic reprogramming from mitochondrial oxidation to glycolysis and promotes cancer stem-like status. However, inhibition of glycolysis using 2-deoxyD-glucose (2-DG) suppresses cancer stemness and strengthens the cytotoxicity of GEM. We further demonstrate that GEM-induced metabolic reprogramming is KRASdependent as knockdown of KRAS reverses the metabolic shift. In addition, GEMinduced metabolic reprogramming activates AMP-activated protein kinase (AMPK) which promotes glycolytic flux and cancer stemness. Our results further reveal the involvement of GEM-induced reactive oxygen species (ROS) in the activation of KRAS/ AMPK pathway in regulation of metabolic reprogramming and cancer stemness, while N-acetyl-L-cysteine (NAC), a ROS scavenger, inhibits the activated KRAS/AMPK pathway, which is further validated by introducing exogenous hydrogen peroxide (H2O2). Collectively, these findings reveal an undesired effect of GEM in PanCa treatment. Therefore, regulating cellular redox, targeting KRAS/AMPK signaling, or reversing metabolic reprogramming may be effective approaches to eliminate CSCs and enhance chemosensitivity of GEM to improve the prognosis of patients with PanCa. www.impactjournals.com/oncotarget/ Oncotarget, Advance Publications 2018


INTRODUCTION
Cytotoxic chemotherapy with gemcitabine (GEM) continues to be the first-line treatment for pancreatic cancer (PanCa), however, minimal progress has been made to improve the prognosis of PanCa patients [1]. The limited effectiveness has been attributed to the existence of relatively rare, but highly chemoresistant cancer stem cells (CSCs) [2,3]. CSCs are reported to have inherently high tumorinitiating potential, which is implicated in tumor relapse as well as establishment of metastases [4,5]. Accumulating evidence has shown that the therapeutic resistance of cancer may result from the dysregulation in glucose metabolism [6]. Tumor cells rewired their metabolic pathways to ensure energy supply and maintain survival to escape external lethal insult [7]. In addition, ionizing radiation could induce metabolic alteration which was closely associated with CSC phenotype [8], and promote cancer stem-like cell population expansion and resistance to anticancer drugs [9]. Therefore, it is interesting to investigate the effect of GEM on cancer www.impactjournals.com/oncotarget cell metabolism and stemness to illuminate the underlying mechanisms of chemoresistance in PanCa.
Oncogenic KRAS mutation functions as a key driver in the initiation and maintenance in approximately 90% of PanCa [10]. KRAS oncogene encodes a small GTPase (21 KDa) which is active in a GTP bound form and inactive when bound to GDP [10]. Accumulating evidence suggests that the altered metabolic switch is under the control of various oncogenic signals [11]. Aberrant activation of KRAS can cause dysfunction of oxidative phosphorylation (OXPHOS), leading to elevated glycolysis and tumor development [12]. The transition from OXPHOS into glycolysis is a prerequisite to fuel proficient reprogramming directing a differentiated cytotype back to the pluripotent state [13]. Such transition of cellular metabolism could cause energy crisis; to restore energy supply, AMP-activated protein kinase (AMPK), a key regulator of cellular energy metabolism exerts stimulatory effects on ATP generation via promoting glucose uptake and transport [14]. In addition to act as a stress-response molecule, AMPK is associated with drug resistance and CSCs enrichment [15].
Induction of reactive oxygen species (ROS) upon GEM treatment has been reported previously [16,17]. Although multiple studies have shown that high levels of ROS are cytotoxic while low levels of ROS may be tumor promoting [17,18], alteration of ROS levels in human cohorts do not produce the desired effects observed in preclinical models. Exploration of the underlying signaling linked with ROS might provide more clues on cancer management.
In this study, we demonstrate that GEM treatment induces metabolic reprogramming toward glycolysis, leading to the induction of PanCa cell stem-like property and chemoresistance. Mechanistically, GEM-induced metabolic reprogramming and cancer stemness are regulated by reactive oxygen species (ROS)-mediated activation of KRAS/AMPK pathway. These findings may shed light on how GEM induces therapeutic resistance and provide rationale and windows for better managing PanCa.

GEM treatment induces metabolic reprogramming toward glycolysis in PanCa cells
In our recent study, we demonstrated that GEMresistant PanCa cells underwent metabolic shift, becoming more glycolytic [19]. However, it remains unclear whether GEM treatment induces metabolic reprogramming. We first examined the effect of GEM on the expression of glycolytic genes. As revealed by qRT-PCR analysis, GEM treatment resulted in higher mRNA levels of the majority of glycolytic enzymes reported to contribute to the aggressiveness or drug resistance in cancer cells such as GLUT1, HK2, PFKP, ALDOA, TPI1, PGK1, PGAM1, PKM2, and LDHA in PANC-1, SW1990, and Patu8988 cells (Supplementary Figure 1A).
We then utilized (2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose) (2-NBDG), a fluorescent deoxyglucose analog, to monitor glucose uptake and found that GEM treatment increased 2-NBDG uptake in PanCa cells ( Figure 1A). To confirm the up-regulation of glycolysis at the functional level, we monitored the release of lactate in the extracellular medium. After GEM treatment, the lactate concentration in the medium was strongly augmented, indicating an increased pyruvate to lactate conversion ( Figure 1B). Importantly, when 2-deoxy-D-glucose (2-DG), a glucose analog which competes with glucose for uptake via GLUT1, as well as being phosphorylated by HKII, allosterically inhibits HKII and thus functions as a glycolytic inhibitor [20], was introduced, the elevated lactate production induced by GEM was significantly reduced ( Figure 1C). Likewise, GEM dose-dependently up-regulated the protein and mRNA expressions of glucose transporter 1 (GLUT1) which is responsible for the uptake of extracellular glucose, and lactate dehydrogenase A (LDHA), an enzyme responsible for the conversion of pyruvate into lactate ( Figure 1D). Collectively, the high glucose uptake and lactate production rate suggested that the glycolysis flux was up-regulated upon GEM treatment.
At the intersection of aerobic glycolysis and respiration, pyruvate dehydrogenase kinase (PDK), which inhibits pyruvate dehydrogenase (PDH) thereby prevents the entry of pyruvate into OXPHOS-based metabolism, plays a key role in cellular glucose metabolism. Interesting, we detected four genes encoding kinase (PDK1, PDK2, PDK3 and PDK4) that inhibit the PDH enzyme complex activity and found that PDK2 gene was significantly induced at transcript level in all three tested cell lines with GEM exposure (Supplementary Figure 1B). We therefore examined the protein levels of PDK2. As expected, GEM dose-dependently up-regulated the expression of PDK2 as indicated by Western blot analysis ( Figure 1E). Likewise, GEM clearly increased the phospho-PDHE1-α (Ser 293) (p-PDHE1-α) in a dose-dependent manner ( Figure 1E). PDK phosphorylates the PDHE1-α subunit and inactivates the PDH enzyme complex that converts pyruvate to acetylcoenzyme A, thereby inhibiting pyruvate metabolism via the tricarboxylic acid (TCA) cycle. In addition, mitochondrial membrane potential (ΔΨm) is generated by OXPHOS activity, thus, a decrease in ΔΨm is indicative of decreased electron transport and OXPHOS activity [21]. We also observed that GEM treatment decreased ΔΨm ( Figure 1F) without affecting the cell viability (Supplementary Figure 2A). All these findings suggested that GEM treatment induced metabolic reprogramming.

Inhibition of glycolysis abrogates GEM-induced enhancement of cancer stemness
Transition from OXPHOS, typical of somatic cells, into glycolysis is a prerequisite to fuel proficient reprogramming directing a differentiated cytotype back to the pluripotent state [13]. We therefore investigated the cancer stem-like property of GEM-treated PanCa cells. CSCs are defined by their ability to form tumor spheres in non-adherent conditions [3,22]. Three PanCa cells (PANC-1, SW1990, and Patu8988) were treated with GEM at a "harmless" dose (5 µM, 24 h) (Supplementary Figure 2A) exhibited enhanced ability to form tumor spheres in ultra-low adhesion plates (Supplementary Figure 2B). Since the pluripotent markers Nanog and Sox2 are functionally involved in the maintenance of stemness [22], we determined expression of Nanog and Sox2 and found increased mRNA and protein expressions in all three cell lines treated with GEM in a dose-dependent manner (Supplementary Figure 2C). We further found that GEM significantly increased gene expression of ALDH, OCT4, KLF4, CXCR4, and CD24 reported as CSC markers in all three cell lines [3,[23][24][25], and EPCAM in SW1990 (Supplementary Figure 2D). In addition, GEM up-regulated ABCG2 and MDR1 gene levels associated with drug resistance [26] (Supplementary Figure 2E). Additionally, ID-1, MUC-1 and MUC-4 were all up-regulated upon GEM treatment in pancreatic cancer cells, indicating the decreased All the three cell lines were treated with increasing concentrations of GEM for 24 h. Total protein or RNA was extracted for Western blot or qRT-PCR analyses. Ratio is expressed as fold change compared with control values normalized to 1 after normalized by β-Actin. (E) Western blot analysis showed increasing expressions of PDK2 and p-PDHE1-α. Cells were treated as indicated in (D). Ratio represents the intensity of band of PDK2 or p-PDHE1-α normalized β-Actin or total PDHE1-α, respectively, and then normalized by control. Densitometry was performed by Image Lab software. (F) ΔΨm in PANC-1 cells treated with GEM (5 µM, 24 h) by JC-1 method. GEM treatment induced a depolarized ΔΨm with more green than red JC-1 fluorescence. Representative images of fluorescence microscope (left), flow cytometry analysis (middle), and quantification (right) are shown. Scale bar, 100 µm. The data shown are from three independent experiments. Bars represent means ± SD. β-Actin was served an internal control. * P < 0.05; ** P < 0.01; *** P < 0.001. GEM: gemcitabine. differentiated extent [27,28] (Supplementary Figure 2F). Furthermore, GEM dose-dependently increased epithelialmesenchymal transition (EMT)-associated genes such as N-cadherin, Vimentin, Snail1, and Zeb1 in pancreatic cancer cells [29] (Supplementary Figure 2G).
We further investigated whether inhibition of glycolysis could abrogate GEM-induced cancer stemness in PanCa cells. We found that 2-DG, a glycolytic inhibitor, partially abrogated GEM-induced expression of pluripotent marker Nanog and Sox2 at protein and mRNA levels ( Figure  2A-2B). To further confirm that GEM-induced up-regulation of glycolysis promoted cancer stemness, we examined the expression of CD133, a CSC surface marker of PanCa [22,25,29,30]

KRAS oncogene is involved in GEM-induced metabolic reprogramming and cancer cell stemness
It has been reported that KRAS (G12V) transformation led to mitochondrial dysfunction and metabolic switch [12]. We therefore investigated the contributing role of KRAS in GEM-induced metabolic reprogramming. We first evaluated the effect of GEM on KRAS activation. As revealed in Figure 3A, GEM exposure effectively activated KRAS, showing a higher level of GTP-bound KRAS compared with the control group. We further assessed the effect of KRAS extinction on metabolic reprogramming. Knockdown of KRAS was first verified, which resulted in a decrease in the most of glycolytic genes (Supplementary Figure 3A-3B). Importantly, knockdown of KRAS reduced lactate production ( Figure 3B). Meanwhile, KRAS knockdown suppressed the expression of PDK2 and consequently decreased p-PDHE1-α ( Figure 3C), indicating the restoration of OXPHOS. Additionally, the ΔΨm was not decreased after KRAS knockdown as revealed by fluorescent microscope and flow cytometry analyses ( Figure 3D). The maintenance of ΔΨm indicates the functional OXPHOS. These results suggested that the activation of KRAS can inhibit the mitochondrial function and drive metabolic switch to glycolysis.
We further assessed the role of KRAS in cancer stemness maintenance and chemoresistance. KRAS knockdown decreased the sphere formation ability ( Figure 3E) and the expression of pluripotent markers in both protein and mRNA levels ( Figure 3F-3G). In addition, KRAS knockdown significantly enhanced the cytotoxicity of GEM, 2-DG, or both with decreased IC50 ( Figure 3H). Together, these results established that the metabolic switch from glycolysis to OXPHOS induced by KRAS knockdown inhibited cancer stemness and sensitized PanCa cells to chemotherapeutics.

GEM-induced metabolic shift activates AMPK
As observed above, GEM induced a metabolic shift from OXPHOS toward aerobic glycolysis which was a less efficient pathway for cellular energy supply. AMPK is an evolutionarily conserved energy sensor, modulating cellular energy flux in response to energy crisis [31]. We therefore examined the effect of GEM on AMPK activation. As expected, GEM dose-dependently increased the phosphorylation of AMPKα (Thr172) (pAMPKα) ( Figure 4A), which is thought to correlate with the enzyme activity [32]. To further examine whether the rewired cellular metabolism accounted for AMPK activation, we examined the effect of KRAS knockdown on AMPK activation. As expected, knockdown of KRAS resulted in a reduced pAMPKα expression ( Figure 4B), suggesting a regulatory role of KRAS-mediated cellular metabolism in AMPK activation.
It is well documented that AMPK is involved in the regulation of glycolytic flux [33]. We therefore investigated whether AMPK was involved in GEMinduced metabolic reprogramming. We pretreated cancer cells with compound C, an inhibitor of AMPK, followed by exposure to GEM. The results showed that GEMinduced pAMPKα were abrogated after introducing compound C which showed marginal effect on p-PDHE1-α ( Figure 4C). However, compound C significantly inhibited GEM-induced lactate production ( Figure 4D), suggesting the inhibition of glycolysis. We also observed that cancer cells treated with compound C exhibited reduced sphere formation ( Figure 4E), diminished expression of pluripotent markers ( Figure 4F). In addition, compound C strengthened the cytotoxicity of GEM toward PanCa cells, showing a decreased IC50 ( Figure 4G and Figure 2F).
To further illuminate the potential role of AMPK in glycolytic flux, we introduced an AMPK-dependent activator A-769662. Treatment with A-769662 activated AMPK, while had no obvious effect on p-PDHE1-α (Supplementary Figure 4A). We did not observe difference in lactate production after A-769662 treatment (data not shown). However, A-769662 treatment promoted the glycolytic gene expression such as GLUT1 and LDHA (Supplementary Figure 4B Figure 4E). Collectively, these findings demonstrated that AMPK www.impactjournals.com/oncotarget mediated GEM-induced up-regulation of glycolysis and cancer stemness induction.

GEM-induced ROS participates in the activation of KRAS/AMPK pathway, metabolic reprogramming, and cancer stemness
Induction of ROS upon GEM treatment has been reported previously [16,17]. To examine whether GEMinduced ROS activates the KRAS/AMPK pathway, we first examined ROS production upon GEM treatment. Consistent with previous findings, GEM treatment induced a dose-dependent increase in ROS levels which were abrogated after the introduction of N-Acetyl-cysteine (NAC) (a free radical scavenger) ( Figure 5A). Importantly, GEM-induced up-regulation of KRAS-GTP and pAMPKα were both abrogated by pretreatment with NAC ( Figure  5B). Next, we found that GEM-induced lactate production was reduced ( Figure 5C), and p-PDHE1-α expression was diminished by NAC ( Figure 5D), indicating that GEMinduced activation of KRAS/AMPK signaling was ROSdependent.
We further found that GEM-induced sphere formation ( Figure 5E), CD133 expression ( Figure 5F), and pluripotent markers ( Figure 5G-5H) were all inhibited by pretreatment with NAC. To avoid ROS-induced oxidative damage, cancer cells often up-regulated their antioxidant defense capacity. We observed a significant increase in the expression of antioxidant genes such as superoxide dismutase 1 (SOD1), superoxide dismutase 2 (SOD2), catalase (CAT), glutathione peroxidase 1 (GPX1), and nuclear factor (erythroid-derived 2)-like 2 (Nrf2) in PanCa cells (Supplementary Figure 5), suggesting that GEM induced oxidative stress accompanied with an increase of the antioxidant reactions.
To further establish a role of oxidative stress in the activation of KRAS/AMPK pathway, we introduced exogenous H 2 O 2 to induce an increase in ROS levels comparable to that induced by GEM in PanCa cells ( Figure 6A). Similar to GEM treatment, H 2 O 2 induced KRAS/AMPK activation, which was also inhibited by NAC ( Figure 6B). In addition, H 2 O 2 increased glycolytic gene expression (Supplementary Figure 6), and lactate production ( Figure 6C), indicating the elevation of glycolytic flux. Importantly, H 2 O 2 -induced lactate and p-PDHE1-α were both diminished with NAC treatment (Figure 6C-6D). Moreover, H 2 O 2 -induced sphere formation, CD133 expression, and pluripotent markers were all decreased by the introduction of NAC ( Figure 6E-6H). Collectively, these findings established that GEM induced metabolic reprogramming and cancer stemness via ROSdependent activation of KRAS/AMPK pathway (Figure 7).

DISCUSSION
GEM has been the standard of care chemotherapy for PanCa patients [1]. However, its therapeutic effect is limited and patients who do respond ultimately develop chemoresistance [34]. Accumulating evidence suggests that CSCs are resistant to conventional chemotherapy and radiotherapy, and are a major cause of cancer recurrence and metastasis [2,3,22,35,36]. Therefore, illumination of the underlying mechanisms involved in chemoresistance and cancer stem-like property would shed light on identifying potential therapeutic windows to improve PanCa treatment. In the present study, we for the first time demonstrate that GEM treatment induces metabolic reprogramming toward glycolysis via ROS-mediated activation of KRAS/AMPK pathway, attributing to the enhanced stemness in PanCa cells in vitro.
Both induced pluripotent stem cells and embryonic stem cells undergo metabolic transition, including upregulation of glycolysis and down-regulation of OXPHOS [13]. Similar to these phenomena, we found that GEM induces metabolic reprogramming with enhanced Cell viability was analyzed using MTT assay and quantified with IC50. All experiments were repeated at least 3 times. Bars represent means ± SD. β-Actin was served as an internal control. * P < 0.05; ** P < 0.01; *** P < 0.001. GEM: gemcitabine. glycolysis and decreased mitochondrial oxidation, accompanied by the enhanced cancer stemness. However, inhibition of glycolysis using 2-DG suppresses cancer stemness and strengthens the cytotoxicity of GEM, suggesting a rationale of combining glycolytic inhibitors with conventional chemotherapeutics for PanCa treatment.
Metabolic reprogramming has been recognized a core hallmark in cancer cells [37]. Oncogenic KRAS can lead to many of the phenotypic hallmarks of cancer including cellular metabolism [10]. KRAS transformation can promote glycolytic activity and decreased oxidative flux [38]. Ectopic KRAS expression could inhibit mitochondrial respiratory chain activity, leading to the suppression of mitochondrial respiration [12]. Viale et al. found that surviving cells cultured in semisolid medium with KRAS ablation exhibited decreased glycolysis and relied on mitochondrial oxidation for survival [39]. Our study shows that GEM induces Representative results (left) and quantification of IC50 (right). β-Actin was served as an internal control. All experiments were repeated at least 3 times. Bars represent means ± SD. Ratio represents the intensity of band of pAMPKα or p-PDHE1-α normalized total AMPKα or PDHE1-α, respectively, and then normalized by control. ** P < 0.01; *** P < 0.001. GEM: gemcitabine. www.impactjournals.com/oncotarget KRAS activation which in turn regulates metabolic reprogramming, as knockdown of KRAS can inhibit glycolysis flux and restore mitochondria respiration. The restored cellular metabolism after KRAS knockdown suppresses cancer cell stemness, and sensitizes PanCa cells to GEM, 2-DG, or both, suggesting a tumorigenic and chemoresistant role for KRAS. Similarly, activation of mitochondrial oxidation by PDK2 inhibition could activate mitochondrial apoptotic signaling, causing the death of chemoresistant cancer cells [40]. Mitochondrial OXPHOS was essential for efficient execution of cell death [41,42]. We did not observe the inhibitory effect of KRAS knockdown on cancer cell viability, which might due to limited efficiency of siRNA-mediated silencing. The oncogenic role of mutant KRAS prompts intensive efforts to explore the pharmacological approaches. However, the pursued anti-KRAS therapies have not been successful in clinic practice. The connection between oncogenic KRAS and metabolic reprogramming might provide promising directions [10]. It was reported that GEM-induced ROS was generated by activation of NADPH oxidase (NOX) through NF-κB activation as siRNA depletion of p22 -phox (a catalytic subunit of the NOX complex) alleviated GEM-induced ROS production [17]. We also found that GEM-induced ROS was NOX-dependent as inhibition of NOX using Apocynin (a NOX inhibitor) reduced ROS production [43]. ROS-induced activation of NF-κB can also promote HIF-1α and CXCR4 expression, which was involved in GEM resistance in PanCa cells [18]. Our findings show that GEM-induced ROS participates in the activation of KRAS/ AMPK pathway, which can be reversed by pre-treatment with NAC, a free radical scavenger. It was reported that CSCs have mechanisms to fine-tune ROS levels by tightly regulating metabolic pathways such as glycolysis rather than oxidative phosphorylation to reduce oxidative stress [44]. Nomura et al. found that CD133 + pancreatic CSCs showed increased glycolytic flux and decreased mitochondrial activity, and the metabolic switch resulted in decreased ROS production upon GEM treatment and favored their  in (B, H). The mRNA expressions of Nanog and Sox2 in PANC-1 cells treated as in B. β-Actin was served as an internal control. All experiments were repeated at least 3 times. Bars represent means ± SD. Ratio was consistent with previous description. * P < 0.05; ** P < 0.01; *** P < 0.001. GEM: gemcitabine.
survival [30]. Thus, the metabolic switch induced by GEM was likely a cellular response to oxidative stress and prerequisite for the induction of CSCs. In addition, GEMinduced expression of the antioxidant genes may also play a role in the maintenance of intracellular redox homeostasis. GEM treatment increased Nrf2 expression which regulated the expression of glutathione (GSH), a major cellular antioxidant, to maintain ROS levels within threshold in case of oxidative damage [17]. Additionally, ROS-dependent regulation of energy metabolism can drive the carbohydrate flux to pentose phosphate pathway, thereby increasing the NADPH level to counteract intracellular ROS [45]. Furthermore, fatty acid oxidation key enzyme carnitine palmitoyltransferase 1A (CPT-1A) maintains intracellular redox and energy homeostasis via generation of NADPH and GSH during energy stress [46]. Cancer cells can rewire metabolic pathways in response to changes in cellular energy and nutrient status, suggesting the complexity of targeting cellular metabolism.
AMPK is an conserved energy sensor and master regulator of cellular metabolism [31]. We found that GEM treatment activates AMPK, probably resulting from the energy crisis induced by GEM treatment as the restoration of mitochondria oxidation after KRAS knockdown inhibits AMPK activation. Interestingly, AMPK-dependent activator promotes glycolysis while AMPK inhibitor inhibits glycolysis flux, suggesting a direct regulatory role of AMPK in glycolysis, which was consistent with recent findings [33]. AMPK activation may also promote intracellular NADPH production against oxidative stress under energy stress conditions [33,47]. Besides, AMPK was critical in modulating the self-renewal and chemoresistance of CSCs [15]. We find that AMPK activation promoted glycolysis, cancer cell stemness and survival, suggesting AMPK as a potential target against the malignant behavior of PanCa.
In conclusion, we demonstrate an undesired effect of GEM in PanCa treatment. GEM induces KRAS activation, which contributes to metabolic reprogramming from mitochondrial oxidation to glycolysis and enhanced cancer cell stemness. The metabolic shift activates AMPK which further promotes glycolysis and enhances cancer cell pluripotency. Furthermore, GEMinduced ROS participates in the activation of KRAS/ AMPK signaling, metabolic reprogramming and cancer stemness maintenance. Considering that metabolic reprograming was implicated in pluripotent redirection and the contributing role of glycolysis in chemoresistance, interventions targeting cellular metabolism might provide new therapeutic opportunities to PanCa treatments. (Houston, TX) and was dissolved in DMSO according to the manufacturer's instruction.

Cell lines and culture conditions
The human PanCa cell lines PANC-1 and SW1990 were originally obtained from American Type Culture Collection (ATCC) (Manassas, USA). Patu8988 cell line was obtained from Keygen (Key Gen Bio Tech, China). All three cell lines have point mutation at codon 12 of KRAS gene (GGT→GAT in PANC-1 and SW1990 and GGT→GTT in Patu8988). The cells were maintained in culture media recommended. Media were supplemented with 1% penicillin/streptomycin and 10% fetal bovine serum (Gibco Invitrogen, Grand Island, NY, USA). The cells were maintained in water-saturated air with 5% CO 2 at 37° C.

Western blot analysis
Cells were lysed with a radioimmunoprecipitation assay lysis buffer (Beyotime Biotechnology, Shanghai, China) at 4° C. Equal amounts of protein were subjected to 10% SDS-PAGE and transferred to PVDF membranes (Millipore, Billerica, MA, USA) which were then blocked with 5% non-fat for 1 h and incubated with primary antibodies over night at 4° C. The membranes were washed and incubated with secondary horseradish peroxidase-coupled antibodies (Aspen, Wuhan, China), and visualized using ECL substrate (Thermo Fisher, Waltham, MA, USA). Primary antibodies against Nanog, Sox2, phospho-AMPKα, and AMPKα were acquired from Cell Signaling Technologies (MA, USA). The antibodies against GLUT1 and LDHA were bought form Cusa Bio (Wuhan, China). The antibodies against β-Actin and KRAS were purchased from ProteinTech (Wuhan, China). The antibodies against PDK2, phospho-PDHE1-α and PDHE1-α were obtained from Abcam (Cambridge, MA). Image Lab software (Version 3.0) (Bio-Rad Lab, Inc, Hercules, CA) was used to quantify the band density.

Quantitative real-time PCR assay
Total RNA was extracted using TRIzol (Invitrogen). Equal amount of RNA (0.5 μg) was reverse-transcribed to cDNA using PrimeScript RT Master Mix (Takara Bio). The qRT-PCR analysis was performed using a quantitative SYBR Green PCR Kit (Takara Bio) according to the manufacturer's instruction. The data were interpreted with the 2 -ΔΔCT method and normalized to β-Actin. Primer sequences are shown in Supplementary Table 1.

Flow cytometric analysis
Cells were dissociated into single cells and washed with PBS followed by stained with PE-conjugated CD133 antibody or isotype control IgG (Biolegend, San Diego, CA, USA) and were analyzed using a flow cytometer (BD, Biosciences, USA).

Determination of intracellular ROS levels and mitochondria membrane potential (ΔΨm)
Cells were treated as indicated and incubated with dichloro-dihydro-fluorescein diacetate (DCFH-DA) (Beyotime) at a final concentration of 10 μM in serumfree medium for 20 min at 37°C. ΔΨm of PanCa cells was measured using mitochondrial membrane potential assay kit with JC-1 (Beyotime) according to the manufacturer's instruction. The fluorescence was then measured with a flow cytometry or fluorescent microscope.

Small interfering RNA-mediated knockdown of KRAS
Small interfering RNA (siRNA) sequence targeting the coding region of KRAS (siKRAS: GGAAG CAAGTAATTGA) and negative control siRNA (ncKRAS) not matching any of the human genes were obtained (RiboBio Co. Guangzhou, China). Knockdown of KRAS was performed by transfecting cells with Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Efficiency of the transfection was confirmed by Western blot and qRT-PCR analyses.

Glucose uptake and lactate production assays
Glucose uptake was performed according to previously published methods [48]. For assessment of lactate production, cells (3 × 10 5 ) were cultured overnight. After different treatments, the supernant was collected and centrifuged. Lactate production was assessed using a lactic acid assay kit (Nanjing Jiancheng Bio. Nanjing, China) according to the manufacturer's protocol and was corrected for the total protein in each sample.

KRAS activation assay
The level of active KRAS was assessed utilizing a RAS activation assay kit (NewEast Biosciences, Malvern, PA). Briefly, cells were cultured in 10 cm plated to 80-90% confluence before different treatments. Total cell lysates were collected and incubated with anti-active RAS monoclonal antibody plus protein A/G agarose bead slurry at 4° C for 1 h with gentle agitation. The agarose beads were resuspended in 20 μL 2 × reducing SDS-PAGE sample buffer and boiled. The precipitated active RAS was detected by immunoblot analysis using anti-KRAS monoclonal antibody according to the manufacturer's instructions.

Statistical analysis
All statistical analyses were performed by GraphPad Prism 5.0 software (San Diego, CA). Unpaired t tests, one-way ANOVA with Bonferroni post-tests, or two-way ANOVA were used to calculate statistical significance. Results were presented as means ± standard deviation (SD). P values < 0.05 were considered statistically significant.