Enhanced O-GlcNAc modification induced by the RAS/MAPK/CDK1 pathway is required for SOX2 protein expression and generation of cancer stem cells

Cancer stem cells (CSCs) have tumour initiation, self-renewal, and long-term tumour repopulation properties, and it is postulated that differentiated somatic cells can be reprogrammed to CSCs by oncogenic signals. We previously showed that oncogenic HRASV12 conferred tumour initiation capacity in tumour suppressor p53-deficient (p53−/−) primary mouse embryonic fibroblasts (MEFs) through transcription factor NF-κB-mediated enhancement of glucose uptake; however, the underlying mechanisms of RAS oncogene-induced CSC reprogramming have not been elucidated. Here, we found that the expression of the reprogramming factor SOX2 was induced by HRASV12 in p53−/− MEFs. Moreover, gene knockout studies revealed that SOX2 is an essential factor for the generation of CSCs by HRASV12 in mouse and human fibroblasts. We demonstrated that HRASV12-induced cyclin-dependent kinase 1 (CDK1) activity and subsequent enhancement of protein O-GlcNAcylation were required for SOX2 induction and CSC generation in these fibroblasts and cancer cell lines containing RAS mutations. Moreover, the CDK inhibitor dinaciclib and O-GlcNAcylation inhibitor OSMI1 reduced the number of CSCs derived from these cells. Taken together, our results reveal a signalling pathway and mechanism for CSC generation by oncogenic RAS and suggest the possibility that this signalling pathway is a therapeutic target for CSCs.


Results
Oncogenic RAS induces SOX2-initiated CSC generation. Previously, we reported that oncogenic HRAS (HRAS G12V mutant: HRAS V12 ) induced tumorigenic properties in p53 −/− MEFs 26 , suggesting that HRAS V12 promoted the generation of CSCs. To confirm this hypothesis, we generated HRAS V12 -expressing p53 −/− MEFs (Fig. 1A) and demonstrated that these cells formed tumours in nude mice (Fig. 1B). Previous studies have shown that CSCs display tumorigenic activity, and cancer cells with CSC-like properties will form spheres in low attachment culture conditions in medium containing growth factors 27,28 . As shown in Fig. 1C, approximately 0.4% of HRAS V12 -expressing p53 −/− MEFs developed spheres in low attachment plates. It has been reported that the reprogramming factors OCT4, KLF4, and SOX2 are important for generation and maintenance of CSCs and are used as CSC markers in various cancers [29][30][31] . Therefore, we analysed the mRNA expression levels of these factors and found that the sphere-forming cells exhibited enhanced expression of OCT4, KLF4, and SOX2 compared with that of parental adherent cells (Fig. 1D). Moreover, although there may have been differences in reactivities between human and mouse antibodies, SOX2 protein expression in sphere-forming cells from HRAS V12 -expressing MEFs was not significantly different from that observed in the human colon cancer line HCT116 ( Supplementary Fig. S1A online). Furthermore, the protein expression level of OCT4 was 8.7-fold higher in sphere-forming cells from HRAS V12 -expressing MEFs than that in HCT116. Therefore, these results suggested that sphere-forming cells from HRAS V12 -expressing p53 −/− MEFs were CSCs. We also compared expression levels of these genes between adherent HRAS V12 -expressing p53 −/− MEFs cells and p53 −/− MEFs control cells. Interestingly, although OCT4 expression was unchanged, the expression of SOX2 was markedly enhanced in HRAS V12 -expressing p53 −/− MEFs cells ( Fig. 1E and F). These results suggest that the expression of SOX2 as well as KLF4 was induced by HRAS V12 expression in p53 −/− MEFs, and OCT4 was induced in the process of CSC reprogramming.
It has been shown that tumorigenic properties are efficiently induced by co-expression of oncogenes in primary rodent cells but not in human cells, suggesting a mechanistic difference between human and rodent oncogenesis 32 . As previously shown, HRAS V12 and simian virus 40 large T antigen (SV40 LT), which functionally inactivates p53 and the retinoblastoma tumour suppressor protein, were unable to induce tumorigenic properties in human cells 32 ; SOX2 expression and sphere-forming activity were not induced by p53 knockout (KO) or HRAS V12 expression in the human lung fibroblast cell line TIG-3 (Supplementary Fig. S1B and C online). However, tumorigenic properties were induced in TIG-3 cells by the combination of SV40 T-ag, c-MYC, and HRAS V12 (TIG-3-SMR cells) 33,34 , and sphere-forming cells were present in the TIG-3-SMR cell population ( Supplementary  Fig. S1D online). Notably, SOX2 expression was also elevated in TIG-3-SMR cells when compared with control TIG-3 cells ( Supplementary Fig. S1E online). Moreover, in TIG-3-SM cells expressing only SV40 T-ag and c-MYC, SOX2 expression was not induced, and the SOX2 protein level was identical to that in control TIG-3 cells ( Supplementary Fig. S1F online). These data indicated that RAS signalling was also required for the induction of SOX2 in TIG-3 cells. Therefore, these results suggest that the induction of SOX2 by oncogenic signalling is regulated differently between mice and humans. However, the difference in tissue origin cannot be ruled out.  35,36 . Therefore, from these studies and the above results, it was possible that SOX2 functioned as an initiation factor for CSC reprogramming. To test this hypothesis, we deleted the SOX2 gene in HRAS V12expressing p53 −/− MEFs with three independent gRNAs using the CRISPR-Cas9 gene knockout system. Single cell-derived SOX2 KO clones from each gRNA were isolated, and immunoblotting and quantitative PCR (qPCR) showed a lack of SOX2 expression in each clone (Supplementary Fig. S2A and B online). Inhibition of SOX2 expression was reported to attenuate cancer cell proliferation 37 . SOX2 KO cells showed suppressed cell growth properties; however, proliferation was not completely inhibited ( Fig. 2A). Next, we analysed the effect of SOX2 KO on the CSC properties found in HRAS V12 -expressing p53 −/− MEFs. The appearance of sphere-forming cells and an increase in anchorage-independent colony-forming cells correlate with tumourigenicity 38 ; however, these properties were not observed after SOX2 KO ( Fig. 2B and Supplementary Fig. S2C online). Furthermore, we analysed the effect of SOX2 KO on tumour-initiating activity in vivo. HRAS V12 -expressing p53 −/− MEFs with or without SOX2 KO were subcutaneously injected into nude mice and tumour growth was monitored. Consistent with data presented in Fig. 1B, we confirmed tumour progression in mice injected with the HRAS V12 -expressing p53 −/− MEFs (Fig. 2C). However, SOX2 KO in these cells completely inhibited tumour development ( Fig. 2C and D). Because the cell growth rate was suppressed in vitro by SOX2 KO (Fig. 2A), we measured tumorigenic activity for an extended period; however, we did not detect tumour formation up to 13 weeks after injection (Fig. 2C). Moreover, forced expression of SOX2 in p53 −/− MEFs and p53 −/− TIG-3 cells promoted sphere formation and tumour development ( Supplementary Fig. S2D-H online). These results suggest that SOX2 expression is required for CSC generation in p53-deficient cells.  24,39 . Therefore, we next examined whether these effectors were involved in promoting SOX2 expression. Constitutively active forms of RAF (BRAF V600E ) and PI3K (PI3K CAAX ) were stably expressed in p53 −/− MEFs. Although the increase in SOX2 mRNA expression was relatively weakly in PI3K CAAX -expressing cells, the levels of SOX2 and KLF4 mRNA in BRAF V600E -expressing cells were similar to that in HRAS V12 -expressing cells (Fig. 3A). These results suggest that HRAS V12 induces the expression of SOX2 mRNA through RAF and its downstream effectors MEK/ERK. Furthermore, the expression of HRAS V12 enhanced the activating phosphorylation of ERK and both the mRNA and protein expression of SOX2. These enhancements were significantly reduced by the MEK inhibitor U0126 but not by the PI3K inhibitor LY294002 (Fig. 3B-D). Similar results were obtained after treatment with the MEK inhibitor PD184352 (Fig. 3E). Immunofluorescent staining confirmed that the fluorescence intensity of SOX2 was significantly reduced by the MEK inhibitor but not by the PI3K inhibitor ( Fig. 3F and G). Additionally, the number of sphere-forming cells was reduced after treatment of HRAS V12 -expressing p53 −/− MEFs with the MEK inhibitors (Fig. 3H). These results suggest that the RAF/MEK/ERK pathway is required for SOX2 induction and reprogramming of normal fibroblasts to CSCs.

CDK1 activity is required for SOX2 induction and generation of CSCs in RAS-activated cells.
Activation of ERK enhances expression of cyclin D1, which binds to and activates cyclin-dependent kinase 4 and 6 (CDK4/6) during the G1 phase of the cell cycle 40 . Aberrant regulation of the CDK4/6-cyclin D1 axis is associated with the development of metastatic melanoma and breast cancer 41,42 . From these findings, it is possible that HRAS V12 induced SOX2 expression through ERK-mediated CDK4/6 activation. To test this possibility, we analysed the effect of the CDK4/6 inhibitor palbociclib 43 on SOX2 expression and confirmed that the expression of SOX2 mRNA and protein levels were slightly reduced in HRAS V12 -expressing p53 −/− MEFs after treatment with a high-dose of palbociclib (Supplementary Fig. S3A and B online). Similar results were obtained with TIG-3-SMR cells ( Supplementary Fig. S3C online). These results suggest that CDK4/6 activity alone is insufficient for SOX2 induction. In addition to CDK4/6, the BRAF/MEK/MAPK pathway is also essential for   [44][45][46] . Recent studies have reported a role for CDK1 and CDK2 in the maintenance of stem cell pluripotency and CSC-like properties in breast cancer [47][48][49] . These reports suggested that CDK1 and/or CDK2 are required for SOX2 induction. Therefore, we examined the effects of the CDK1/2 inhibitors dinaciclib 50 and roscovitine 51 on SOX2 expression. We found that SOX2 mRNA and protein expression was strongly suppressed by dinaciclib ( Fig. 4A and B) and roscovitine (Supplementary Fig. S3D and E online) in HRAS V12 -expressing p53 −/− MEFs. Similar results for dinaciclib were observed in TIG-3-SMR cells (Fig. 4C). Moreover, the number of sphere-forming cells in HRAS V12 -expressing p53 −/− MEFs was largely suppressed by dinaciclib but only partially suppressed by palbociclib (Fig. 4D), which correlated with suppression of SOX2 expression levels ( Fig. 4A and Supplementary Fig. S3A online). To further analyse whether CDKs were essential for SOX2 expression, we analysed SOX2 levels in HRAS V12 -expressing p53 −/− MEFs after knockdown of CDK1 or CDK2 using siRNA. Immunoblot analysis showed that CDK1, but not CDK2, knockdown reduced the expression level of SOX2 protein (Fig. 4E). Treatment with dinaciclib, U0126, and PD184352, which suppressed the expression of SOX2 in this study, reduced the phosphorylation of CDK1 T161 , an indicator of CDK1 activation 52 (Supplementary Fig. S4A-C online). These results indicate that CDK1 activity is required for SOX2 expression in HRAS V12 -expressing p53 −/− MEFs. Additionally, we analysed the effect of dinaciclib on human cancer cell lines, including KRAS-mutated colon cancer (HCT116, SW480, and DLD1) and lung cancer (H460 and A549) cells. Dinaciclib suppressed SOX2 mRNA expression in these cancer cells (Fig. 4F) and the number of sphere-forming cells from the HCT116, SW480, and H460 lines (Fig. 4G). These results suggest that CDK1-mediated SOX2induction promotes the generation of CSCs in human cancer cells with RAS mutations.    53 . Previously, we reported that enhanced O-GlcNAcylation is important for SOX2 expression and maintenance of CSC properties, including sphere-and tumour-forming activities, in colon and lung cancer cells 21 . These findings suggested the possibility that O-GlcNAc modifications are involved in acquisition of CSC properties. Therefore, we determined the O-GlcNAc levels in HRAS V12 -expressing p53 −/− MEFs and TIG-3-SMR cells and found elevated levels of protein O-GlcNAcylation compared with those in the respective control cells (Fig. 5A and B). Next, we analysed the cells after treatment with OSMI1, a cell-permeable, small molecule OGT inhibitor 37 . OSMI1 treatment suppressed total O-GlcNAcylation levels of proteins and SOX2 expression in these cells ( Fig. 5A and B). In contrast, treatment of HRAS V12 -expressing p53 −/− MEFs with thiamet G, a specific OGA inhibitor that increases O-Glc-NAcylation 38 , enhanced total O-GlcNAcylation levels and SOX2 expression (Fig. 5C). Interestingly, the OSMI1mediated reduction in SOX2 levels was attenuated by treatment with the proteasome inhibitor MG-132 ( Fig. 5A  and B). The mRNA levels of SOX2 in HRAS V12 -expressing p53 −/− MEFs were not significantly changed by OSMI1 and thiamet G (Fig. 5D), indicating that O-GlcNAcylation regulated SOX2 expression at the post-transcriptional level. Consistent with these results, the numbers of sphere-forming cells decreased and increased after treatment with OSMI1 and thiamet G, respectively (Fig. 5E). These results suggest that increased O-GlcNAcylation is required for SOX2 protein expression and sphere-forming activity in these cells.  Fig. S4A and B online). The MEK inhibitors U0126 and PD184352 reduced the phosphorylation of CDK1 T161 and O-GlcNAcylation levels in HRAS V12 -expressing p53 −/− MEFs ( Supplementary Fig. S4C online). Furthermore, knockdown of CDK1, but not CDK2, with siRNA inhibited O-GlcNAcylation levels in these cells (Fig. 5F). In contrast, treatment with palbociclib had no effect on O-GlcNAcylation levels (Supplementary Fig. S3B and C online). In addition, SOX2 expression (as shown in Fig. 4F and G) and O-GlcNAcylation levels in KRAS-activated cancer cells were suppressed by dinaciclib (Fig. 5G). These results suggest that RAS/RAF/MAPK pathway-induced CDK1 activation is important for induction of O-GlcNAcylation, and this activation pathway is required for SOX2 expression and subsequent CSC generation.

Discussion
Accumulating evidence has revealed that serial oncogenic mutations in stem cells and even in differentiated somatic cells may induce the generation of CSCs [54][55][56][57][58] . The dedifferentiation of somatic cells into CSCs is thought to be induced by a reprogramming mechanism similar to that observed in the production of iPSCs 59,60 . Viewed through the "hallmarks of cancer" 7,8 , CSCs may be generated by iPSC reprogramming factors that are induced by gain-of-function oncogenic and loss-of-function tumour suppressor gene mutations. Therefore, the HRAS V12expressing p53 −/− MEFs provided a simple and easy-to-analyse model for evaluating the signalling system used by oncogenes to regulate reprogramming factors. In our previous studies, we found that the HRAS V12 mutation conferred tumour-initiating activity, a hallmark of CSCs, in p53 −/− MEFs 26 , and this phenomenon was completely dependent on NF-κB-induced aerobic glycolysis 13 . In this study, we found that HRAS V12 induced MAPK-CDK1 signal-dependent induction of SOX2 mRNA transcription and O-GlcNAcylation-mediated SOX2 protein accumulation. Although the transcriptional induction mechanism of SOX2 by CDK1 was not elucidated in this experimental system, CDK1-mediated induction of SOX2 has been analysed in other studies. For example, it was reported that CDK1-induced phosphorylation directly activated transcription factor CP2-like protein 1 (TFCP2L1), which is an activator of pluripotency-associated genes, including SOX2 61 . CDK1 phosphorylated and inhibited the histone lysine demethylase KDM5B, which is a transcriptional suppressor of the pluripotency genes SOX2 and NANOG 62 .
It has been shown that the hexosamine biosynthetic pathway shunts glycolysis toward the production of a key substrate for O-GlcNAcylation and is activated by enhanced glycolysis 63 . The RAS/MAPK pathway induces metabolic reprogramming, including enhanced glycolysis 64 . Therefore, in addition to NF-κB-induced enhancement of glycolysis in p53 −/− MEFs 13 , our present results suggest that the RAS/MAPK/CDK1 pathway further promoted glycolysis and resulted in enhanced protein O-GlcNAcylation. Although SOX2 expression was induced at the mRNA level in transformed cells (Fig. 1E), we also found that the O-GlcNAc modification regulated SOX2 at the post-transcriptional level (Fig. 5A-D). For this issue, we demonstrated that depletion of the SOX2 gene in HRAS V12 -expressing p53 −/− MEFs reduced O-GlcNAc levels ( Supplementary Fig. S4D online), and exogenous SOX2 expression induced protein O-GlcNAcylation in p53 −/− MEFs ( Supplementary Fig. S4E online). These data suggest that transcriptionally elevated SOX2 expression promotes O-GlcNAc levels, and enhanced O-GlcNAcylation could promote expression of SOX2 at post-transcriptional levels. At present, the mechanism by which SOX2 protein expression is induced by O-GlcNAcylation has not been elucidated. In this study, we found that O-GlcNAcylation inhibited proteasomal degradation of SOX2. Moreover, the O-GlcNAc modification of SOX2 was confirmed in HRAS V12 -expressing p53 −/− MEFs ( Supplementary Fig. S4F online). However, because the band intensities were relatively low, it was difficult to conclude that only direct O-GlcNAcylation mediated SOX2 protein induction. These results suggested that direct O-GlcNAcylation of SOX2 and/or other factor(s) is involved in the ubiquitin-mediated degradation system that regulates SOX2 protein induction.
The tumour suppressor protein p53 is referred to as "the guardian of the genome" because it continuously surveys damaged genomic DNA and facilitates DNA repair 65 . Moreover, in response to oncogenic signalling, the p53 tumour surveillance system eliminates cells through p53-mediated apoptosis or induction of senescence 66 . In addition to these functions, accumulating evidence has indicated that regulation of cellular metabolism by p53 is involved in tumour suppression 67 . It was demonstrated that mice expressing acetylation-defective p53 mutants lost the ability to induce apoptosis and senescence but retained tumour suppressive function and the www.nature.com/scientificreports/ ability to modulate the expression of metabolic genes 68 . In p53 −/− MEFs, we previously found that oncogenic transformation by HRAS V12 was dependent on enhanced aerobic glycolysis through NF-κB-mediated induction of the glucose transporter GLUT3 13 . In this previous study, we found that GLUT3 (also known as SLC2A3) functioned as an oncogene because p53 −/− MEFs underwent oncogenic transformation after overexpression of GLUT3 without HRAS V12 . These results suggested that enforced glucose flux induced CSC generation in cells that lacked p53 function. Furthermore, we found that O-GlcNAcylation increased with accelerated glucose consumption in p53 −/− MEFs and was further enhanced by HRAS V12 expression 14 . In the present study, CSC generation in HRAS V12 -expressing p53 −/− MEFs was completely dependent on the induction of SOX2, and enhancement of O-GlcNAcylation by the HRAS V12 -MAPK-CDK1 pathway was important for SOX2 induction. Therefore, it is possible that glycolysis may be hampered by p53, which prevents excessive O-GlcNAcylation-mediated induction of SOX2. Indeed, it has been shown that oncogenes, such as HRAS V12 , activate p53 in untransformed cells 66 , and activation of p53 results in inhibition of glucose consumption and glycolysis 13,69 . Moreover, it has been reported that the OKSM reprogramming factors induce p53-mediated senescence 70 , suggesting the possibility that these reprogramming factors limit their own function using p53-mediated negative feedback. Therefore, at least in MEFs, p53 may exert a tumour suppressor function through the regulation of O-GlcNAcylation by limiting glucose metabolism.
In MEFs, induction of SOX2 was essential for CSC reprogramming by HRAS V12 , but this pathway may not be the only signal for CSC generation. For example, in the context of inflammation-induced cancer, we found that the activated form of MYD88, a regulator of inflammatory signalling, promoted CSC generation in p53 −/− MEFs through activation of the NF-κB-HIF1-OCT4, but not the SOX2, induction pathway 71 . Furthermore, in the present study, expression of HRAS V12 alone was insufficient to induce SOX2 in human TIG-3 cells. Although it is possible that this difference was because of cell type specificity, differences between human and rodent cells cannot be ruled out. In contrast to human cells, primary rodent cells are efficiently converted to tumorigenic cells by the co-expression of oncogenes, suggesting a fundamental difference between humans and rodents 32 . Therefore, these results suggest the possibility that an unidentified tumour suppressive mechanism other than p53 exists in humans and prevents the induction of SOX2 by RAS. Further analyses are required to clarify this mechanism.
siRNA and transfection. The predesigned short interfering RNA (siRNA) for mouse Cdk1 and Cdk2 were obtained from Sigma-Aldrich: siCDK1#1, SASI_Mm01_00179321; siCDK1#2, SASI_Mm01_00179322; siCDK2#1, SASI_Mm02_00323492; siCDK2#2, SASI_Mm01_00151932. We performed reverse transfection of siRNA using Lipofectamine™ RNAiMAX Transfection Reagent (Thermo Fisher Scientific, Waltham, MA, USA). www.nature.com/scientificreports/ Immunoprecipitation and immunoblot analysis. Cells were lysed with TNE buffer (10 mM Tris-HCl, pH 7.4; 1% NP-40; 150 mM NaCl; 1 mM EDTA; 1 mM DTT, and protease inhibitor cocktail) (Nacalai Tesque, Japan). Protein concentrations were measured using the Bradford assay. For immunoprecipitation, lysates were pre-cleared with Protein A/G PLUS agarose (Santa Cruz Biotechnology), and proteins were immunoprecipitated with the anti-SOX2 antibody. For immunoblot analysis, cellular proteins (20µ) or immunoprecipitates were separated using sodium dodecyl sulfate-polyacrylamide gel electrophoresis and then transferred to polyvinylidene difluoride membranes (Merck). The membranes were probed with primary antibodies, followed by incubation with horseradish peroxidase-conjugated mouse or rabbit immunoglobulin G (GE Healthcare, England) and visualisation using Chemi-Lumi-One Super or Ultra assay kits (Nacalai Tesque). The protein bands were digitalised using the LAS-3000 mini image analyser (Fujifilm, Japan), and the intensity of each band was quantified using ImageJ software. For quantification, the intensity of the β-actin band was used to normalise each protein signal.
Quantitative real-time PCR. Total RNA was extracted using the NucleoSpin RNA kit (Macherey-Nagel, Germany) following the manufacturer's instructions. Double-stranded cDNA was prepared from total RNA using oligonucleotide (dT), random primers, and Superscript III reverse transcriptase (Invitrogen, Carlsbad, CA, USA). Quantitative real-time PCR (qPCR) analysis was performed as previously described 23  Alexa Flour 488-conjugated anti-rabbit antibody (Thermo Fisher Scientific) was used as the secondary antibody for 1 h at room temperature. VECTASHIELD Mounting Medium with DAPI (Vector Laboratories, Burlingame, CA, USA) was used to stain nuclei and for mounting the cells on slides. Images were acquired using a fluorescence microscope (BioZero BZ-8100; Keyence, Japan), and the fluorescence intensities were quantified with ImageJ software.
Cell growth analysis. HRAS V12 -expressing p53 −/− MEFs (1 × 10 5 ) expressing each SOX2 sgRNA separately were seeded in 6-well plates. Cell numbers were determined using a Vi-CELL cell analyser (Beckman, Brea, CA, USA) on the indicated days after plating, and cell growth curves were created.
Sphere formation assay. Cells (5 × 10 3 or 1 × 10 4 ) were plated in 6-well, ultra-low attachment plates and grown in serum-free DMEM/F12 medium containing epidermal growth factor (20 ng/mL) and basic fibroblast growth factor (10 ng/mL) for 7 days. The numbers of spheres were counted in each treatment group. Sphere images were captured using the BZ-8100 microscope.
Colony formation assay. A layer of 1.5% (weight/volume) agarose prepared in DMEM containing 10% FBS was added to the wells of 6-well plates. Agarose (0.6%) containing 3 × 10 4 HRAS V12 -expressing p53 −/− MEFs containing each SOX2 sgRNA was added to the top of the first layer. After 30 days, each well was stained with 0.005% crystal violet (Sigma-Aldrich), and the colonies were counted.
Animal experiments and cell line xenografts. The animal experiment protocol was approved by the Ethics Committee on Animal Experiments of Nippon Medical School (ethics approval number 26-020, . Animal experiments were carried out in accordance with the guidelines for Animal Experiments of Nippon Medical School and the guidelines of The Law and Notification of the Government of Japan as well as the ARRIVE guidelines. Mice were maintained at 20-24 °C in a facility with a 12 h light/ dark cycle and 40%-70% humidity. The mice were allowed free access to water and standard MF laboratory mouse chow (Oriental Yeast Co., ltd. Tokyo, Japan) and housed at a maximum number of five per cage. All mice were checked for stress each day. For the xenograft experiments, 5-week-old male BALB/cAJcl-nu/nu mice were purchased from CLEA Japan, Inc. (Tokyo, Japan) and assigned at random to the experiments. These mice were subcutaneously injected with 1 × 10 5 cells of p53 −/− MEFs, HRAS V12 -expressing p53 −/− MEFs or HRAS V12 -expressing p53 −/− MEFs containing SOX2 gRNA. The number of mice used are indicated for each experiment. Tumour growth was monitored every 3 days for 3 weeks and each week thereafter by caliper measurements, and tumour size was determined using the following formula: (Length × Width 2 )/2. At the end of the experiments (3 or 13 weeks post-injection), mice were euthanized by cervical dislocation, then each tumour were removed and weighed, and also collected for further experiments.

Statistical analysis.
All experiments were repeated at least three times independently. Data are presented as means ± standard deviation (SD). Statistical analysis was performed with the Student's t-test using Microsoft Excel (Microsoft, DC, USA). P < 0.05 was considered statistically significant.