Global Downstream Pathway Analysis Reveals a Dependence of Oncogenic NF-E2–Related Factor 2 Mutation on the mTOR Growth Signaling Pathway

During the multistep carcinogenesis, cancer cells are exposed to multiple environmental stresses such as carcinogens, hypoxia, host inflammation, or anticancer drugs (1). They also have to deal with intrinsic stresses induced by genetic instability, oncogene-induced abnormal proliferation, or aberrant protein folding (1, 2). Therefore, acquisition of resistance to these stresses by some cancer cells would be beneficial for their survival, leading to malignant progression and lethal characteristics such as invasion, metastasis, and chemotherapy resistance (3). Oxidative stress in cancer is one of the important metabolic stresses caused by both extrinsic (reperfusion of a hypoxic microenvironment) and intrinsic (uncoupled mitochondrial activity) factors (4, 5). From the viewpoint of carcinogenesis, oxidative stress is a double-edged sword, because it promotes genetic alterations through DNA damage to drive malignant progression while also causing cell damage and inducing apoptosis (6, 7). Under physiologic conditions, cells possess molecular mechanisms that enable them to adapt to these stresses. In multicellular organisms, adaptive responses to oxidative stresses are regulated by NF-E2–related factor 2 (NRF2), a master transcription factor of many antioxidant genes and phase II detoxifying enzymes (8).

NRF2 belongs to the Cap'nCollar subfamily of basic leucine zipper (bZIP)–type transcriptional factors. It recognizes the specific antioxidant recognition element (ARE) sequence by forming a heterodimer with the bZIP family and promotes the expression of a wide range of cytoprotective genes (9). The activity of NRF2 is finely controlled by ubiquitin-mediated protein degradation. In unstressed cells, NRF2 is recognized by an E3 ubiquitin ligase, KEAP1, and rapidly degraded in the proteasome (10, 11). When cells are exposed to oxidative stress, the cysteine residues of KEAP1 are rapidly modified by electrophiles, resulting in structural change and loss of association with NRF2 (12). The free NRF2 released from KEAP1 then moves into the nucleus and induces its specific targets.

Previous studies have reported that NRF2 activation is beneficial for prevention of cancer, because Nrf2-deficient mice are susceptible to cancer development in chemical-induced carcinogenesis models and NRF2 promotes detoxification of carcinogen and decreases DNA damage by inducing phase II enzymes and reducing oxidative stressors such as reactive oxygen species (ROS), both mechanisms protecting normal cells from carcinogenesis (13, 14). However, recent studies have challenged this concept and shown that aberrant activation of NRF2 through either loss-of-function mutations in the Keap1 gene or gain-of-function mutations in the Nrf2 gene occurs in a wide range of human cancers such as lung (∼40%), head and neck (∼20%), gallbladder (∼30%), liver, and breast cancers (15–20). Thus the KEAP1-NRF2 pathway seems to have a dual function in carcinogenesis, although the detailed biological consequences of NRF2 activation in cancer cells remain unknown.

Details of the experimental procedures are described in Supplementary Materials and Methods.

Cell lines used in this study have been described previously (15, 16) and were validated by short tandem repeat genotyping. Rapamycin (Nakarai) and NVP-BEZ235 (LC Laboratories) were chemically synthesized and purchased. Mutant NRF2 cDNA and small interfering RNAs (siRNA; purchased from Thermo Fisher Scientific) were transfected into cells by using Lipofectamine and RNAiMAX (Invitrogen), and stable clones were isolated after G418 selection. Cell proliferation and drug sensitivity were measured by bromodeoxyuridine incorporation (cell proliferation ELIZA, Roche Diagnostic). All experiments were done in triplicate.

For in vivo assessment of mammalian target of rapamycin (mTOR) inhibitor, we transplanted 5 × 10⁶ LK2 cells in the nude mice. NVP-BEZ235 (free base) was formulated in NMP/polyethylene glycol 300 (10:90, v/v). Solutions (5 mg/mL) were prepared fresh each day of dosing and were given daily (20 mg/kg) by oral gavage. The application volume was 10 mL/kg. Tumor volume was calculated by (tumor diameter)² × (tumor length/2).

Total RNA was extracted from cell lines, primary squamous cell carcinoma (SCC) of the lung, and the corresponding normal lung tissues by RNAeasy (Qiagen). Quantitative reverse transcription-PCR (RT-PCR) was performed in triplicate and evaluated using universal probes for each amplicon and the Light-Cycler system (Roche Diagnostic). Protein extraction and immunoblotting were performed as described previously (15). Antibodies and siRNA used in this study were listed in Supplementary Table S1.

Ten micrograms of total RNA were reverse-transcribed by Moloney murine leukemia virus reverse transcriptase, and a Cy3-labeled cRNA probe was synthesized using T7 RNA polymerase in accordance with the manufacturer's protocol. Probes were hybridized with a microarray containing 41,000 oligonucleotides covering the whole human genome (Whole Human Genome Oligo Microarray, G4112F, Agilent Technologies). All analyzed microarray data are in accordance with the MIAME guidelines.

Among the total of 41,000 probes, 12,091 stable probes corresponding to the MAQC common genes were used for statistical analysis. To characterize the molecular backgrounds of the gene list, enrichment analysis for Gene Ontology (GO) was performed using the GO Term Finder (http://go.princeton.edu/cgi-bin/GOTermFinder). Preranked gene set enrichment analysis was applied for the gene list with C2 pathway gene sets and 2,000 permutations (http://www.broadinstitute.org/gsea/index.jsp). Two independent sets of gene expression data from head and neck cancer and squamous lung cancer were classified into two groups based on the signatures of the gene list using K-means clustering (R cluster package). Then, stable clusters were obtained in 1,000 random trials to avoid artifacts of classification due to dependency of initial random seeds. Survival analysis was performed using the Kaplan-Meier method, and log-rank test was used to evaluate disease-free survival.

We established stable HEK293 (an immortalized human epithelial cell line) clones expressing two mutant NRF2 cDNAs (T80R and L30F mutants), both of which were detected in human cancer and behave as gain-of-function mutants (15). Clones expressing mutant and wild-type NRF2 cDNA showed accumulation of NRF2 in the nuclear fraction of these clones (Fig. 1A, top). The degree of nuclear NRF2 accumulation in these clones (samples 3 and 4) was less than the endogenous NRF2 accumulation in Nrf2- (sample 6) or Keap1- (sample 5) mutated cancer cell lines (refs. 15, 16; Fig. 1A, bottom).

We then tested the cell proliferation activity and found that clones expressing mutant NRF2 grew more rapidly than the mock or wild-type NRF2-transfected clones (especially T80R mutants; Fig. 1B). Downregulation of NRF2 by siRNA inhibited the proliferation of clones stably expressing the mutant and wild-type NRF2 gene, whereas growth of the mock control was little affected (Supplementary Fig. S1A and B). We also tested the anchorage-independent cell growth of these clones. As shown in Fig. 1C, both mutant NRF2-expressing clones produced colonies in soft agar, whereas the control clone formed no colonies. In addition, NRF2 mutant clones showed an increase of cell migration activity in vitro (Fig. 1D). We had previously reported two lung cancer cell lines (EBC1 and LK2) that harbored the Nrf2 mutation (EBC1 cell has the D77V mutation, and LK2 cell has the E79K mutation; ref. 15). Downregulation of NRF2 also reduced the proliferation of Nrf2 mutant cancer cells but exerted no effect on SQ5 cells that contain low wild-type NRF2 protein (Fig. 1A; Supplementary Fig. S1C and D), supporting the contention that Nrf2-mutated cancer cells depend on the Nrf2 activity for their proliferation as previously reported in Keap1-mutated cancer cells (17).

We further examined the in vivo tumorigenic activity of mutant NRF2-expressing cells by s.c. transplantation in immunodeficient mice. Both clones expressing mutant NRF2 produced tumors at markedly high frequency, whereas control clones did not form tumors (TR-1: 8/8, TR-2: 8/8, LF1: 5/8, LF2: 7/8, control: 0/10; Fig. 2A; Supplementary Table S2). Tumors expressing mutant NRF2 exhibited poorly differentiated carcinoma with increased mitotic activity (>70% cells being immunopositive for Ki67; Fig. 2B and C), accompanied in most cases by highly aggressive characteristics, such as invasion into the surrounding adipose and muscle tissues (Fig. 2D). Moreover, there was occasional metastasis to the liver, suggesting that mutant NRF2 confers invasive and metastatic activity (Fig. 2E and F), consistent with the increased cell migration activity observed in vitro (Fig. 1D). We also noted that these tumors contained many microvessels (as revealed by CD34 staining; Fig. 2B and G). As observed under in vitro culture conditions, the tumor cells overexpressed a NRF2 target, NQO1 in vivo (Fig. 2H).

Because NRF2 encodes a sequence-specific transcriptional factor, direct or indirect downstream targets of NRF2 should contribute to at least a proportion of the phenotypes described above. We attempted to overview the genome-wide changes in gene expression induced by mutant NRF2 and conducted a microarray analysis (Fig. 3A). Among individual downstream genes, we observed upregulation of several antioxidant genes (Nqo1, Hmox1, Abcc2, and enzymes relating to glutathione synthesis) that have been previously reported to be NRF2 target genes, and we validated the increased expression of these authentic targets by quantitative RT-PCR (Supplementary Fig. S2A and B). Among the other downstream targets, we also observed increased expression of osteopontin, which has been reported to be associated with metastatic activity and angiogenesis in cancer (21, 22), both in vitro and in vivo (Supplementary Fig. S1C; Fig. 2I). We also detected direct NRF2 binding in the promoter of the onsteopotin gene by chromatin immunoprecipitation (ChIP) sequencing.⁵

GO analyses of control and mutant NRF2-expressing clones revealed that persistent expression of mutant NRF2 affected multiple signal pathways (Supplementary Table S3). Consistent with its previously known roles in the adaptative response to oxidative stress, mutant NRF2 significantly induced genes belonging to categories such as “response to stress” and “response to chemical stimuli.” In addition, expression of genes related to inflammatory processes (“response to wounding,” “defense response,” “inflammatory response,” and “cytokine production”) and cell motility (“locomotor behavior” and “chemotaxis”) was also altered in NRF2 mutant clones.

Because mutant NRF2 promoted cell proliferation and anchorage-independent growth in vitro and metastasis activity and increased angiogenesis in vivo, which are all hallmarks of cancer malignancy, the molecular signature induced by mutant NRF2 may be associated with poor clinical outcome. To test this hypothesis, we extracted the mutant NRF2 signature (complete gene list is shown in Supplementary Table S4) and examined its association with clinical data using archives of previous microarray data. We first analyzed a large cohort (n = 60) of head and neck cancer samples (23) because NRF2 is frequently mutated in this type of cancer, and thus far little information has been available about its prognostic molecular signature. On the basis of the mutant NRF2 signature, we classified the samples into two clusters (Cluster1, n = 45; Cluster 2, n = 15; Fig. 3B) and compared the prognosis between them. As shown in Fig. 4B, cases containing the mutant NRF2 signature showed significantly poorer prognosis (P = 0.000118), supporting the contention that the gene profile extracted from HEK293 cells is comparable with the malignant signature of clinical SCC samples. To validate this result, we then analyzed another cohort of lung SCC cases (24). Although the difference did not reach statistical significance (P = 0.0803), probably because of the small size of the cohort (n = 16), we observed a trend for tumors harboring the mutant NRF2 signature to have a poorer prognosis than those without the signature (Supplementary Fig. S3).

To further elucidate the molecular pathways affected by mutant NRF2, we then conducted gene set enrichment analysis (GSEA). This revealed significant enrichment of several pathways in clones expressing mutant NRF2 (Table 1; all significant pathways are listed in Supplementary Table S5). Consistent with the above GO analysis, we detected significant enrichment of pathways associated with genes induced by oxidative stresses such as ROS (Houstis_ROS) or UV treatment (UVB_SCC_UP, UVB_NHEK3_C0, UVB_NHEK3_C3, and UVB_NHEK3_ALL). Another noteworthy finding was that the molecular signature induced by mutant NRF2 included pathways associated with mTOR signaling (Peng_Rapamycin_DN, Peng_Leucine_DN, Peng_Glutamine_DN; ref. 25; Fig. 3C). We also observed other molecular pathways associated with other stresses (hypoxia or viral infection), the progenitor cell signature, and lipid metabolism (Supplementary Fig. S4). Thus, it was revealed that mutant NRF2 induced a set of genes more diverse than had previously been thought. In particular, the mTOR pathway emerged as a new target of NRF2, in addition to various stress-regulated genes.

Because the mTOR pathway has been reported to be an attractive therapeutic target in many solid tumors, we then focused on the relationship between the NRF2 and mTOR pathways. We first examined activation of the mTOR pathway in HEK293 clones expressing mutant NRF2. Immunoblot analysis revealed increased phosphorylation of ribosomal S6 protein, which encodes one of the major downstream substrates of the mTOR pathway, and a mild increase of AKT phosphorylation in mutant NRF2-expressing cells relative to the control, whereas no strong activation of mitogen-activated protein kinase was observed in any of the clones (Fig. 4A). Downregulation of NRF2 expression in Nrf2-mutated cancer cells was associated with reduced S6 kinase activation (Fig. 4B).

To understand the molecular link between the two pathways, we returned to the gene expression profiling because NRF2 is a transcriptional factor and does not seem to interact directly with mTOR signaling. We searched for the downstream targets of mutant NRF2, which are responsible for modulating the mTOR pathway. Among the genes whose expression was found to be significantly altered in the microarray analysis, we found that a member of the small G protein family, RagD, which encodes a recently discovered activator of the mTOR pathway (26), was significantly upregulated in the HEK293 cells expressing mutant NRF2 (Fig. 4C). RagD knockdown reduced the activation of mTOR signaling (Supplementary Fig. S5) and NRF2 downregulation reduced RagD expression in Nrf2-mutated cancer cells (Fig. 4C). The expression of RagD was also quantified in the clinical samples of primary lung SCC and increased in >40% (10 of 24, 41.7%) of tumors relative to corresponding normal lung tissue (Fig. 4C). We found that cell proliferation of both mutant NRF2 expression HEK293 clones and Nrf2-mutated cancer cells was reduced by downregulation of RagD expression (Fig. 4D; Supplementary Fig. S6). Collectively, therefore, mutant NRF2 upregulates RagD expression and RagD plays an important role in the proliferation of Nrf2-mutated cancer cells.

These results prompted us to examine whether inhibition of the mTOR pathway has any effect on Nrf2 mutant cancer. We first examined whether the proliferation of clones expressing mutant NRF2 was sensitive to inhibition of the mTOR pathway. We treated mutant NRF2-expressing HEK293 clones with rapamycin, a small-molecule inhibitor of mTOR, and measured the resulting cell proliferation. As shown in Fig. 5A, rapamycin treatment more effectively reduced cell proliferation in Nrf2 mutant clones than in the mock and wild-type NRF2-expressing ones (IC₅₀ values: control >10 nmol/L, TR1 0.29 nmol/L, TR2 0.3 nmol/L, LF1 0.34 nmol/L, LF2 0.53 nmol/L, WT1 9.94 nmol/L, WT2 >10 nmol/L). We then tested low wild-type NRF2 expressing (SQ5 and QG56), high wild-type NRF2-expressing (A549), and high mutant NRF2-expressing (LK2 and EBC1) cancer cell lines (Fig. 1A). Nrf2 mutated cancer cell lines exhibited more sensitivity to rapamycin than did Nrf2 wild-type cancer cell lines (Fig. 5A). However, Nrf2 mutant cells showed resistance to rapamycin at a higher dose, probably because of the more transformed phenotypes such as feedback system involving phosphoinositide 3-kinase (PI3K) reported previously in other cancer cells (27, 28). Consistently high-dose (1 nmol/L) rapamycin treatment did not robustly reduce the phosphorylation of mTOR pathway targets in LK2 and EBC1 cells (Fig. 5B). To overcome this resistance in cancer cells, we then tested a dual kinase inhibitor (NVP-BEZ235), which targets both mTOR and PI3K and is currently being evaluated in clinical trials. NVP-BEZ235 robustly inhibited mTOR pathway activation and significantly reduced the cell proliferation of Nrf2-mutated lung cancer cells (Fig. 5A and B; IC₅₀ values: LK2 0.042 μmol/L, EBC1 0.099 μmol/L, SQ5 >5 μmol/L, QG56 >5 μmol/L, A549 >5 μmol/L). Similar drug sensitivity was also observed in an Nrf2-mutated head and neck cancer cell line (HO-1-U1; data not shown). NVP-BEZ235 also induced apoptosis in Nrf2-mutated cancer cells as revealed by an increase in the expression of cleaved poly(ADP-ribose) polymerase (PARP; Fig. 5B). Finally, we challenged an in vivo treatment model with NVP-BEZ235. NVP-BEZ235 was administered p.o. to mice bearing s.c. xenografts of LK2 or SQ5 cells, and we found that this drug significantly inhibited the growth of LK2 cells in comparison with placebo treatment (Fig. 5C). Immunohistologic analysis confirmed significant inhibition of mTOR activation (indicated by decreased phospho-S6) and induction of apoptosis (increase of cleaved PARP and caspase-3 activation) in NVP-BEZ235–treated LK2 tumors relative to the placebo-treated ones (Fig. 5C; Supplementary Fig. S7).

In this study, we found that mutant NRF2 confers improved cell proliferation, anchorage-independent growth, and in vivo tumorigenicity, which are the fundamental hallmarks of an oncogene, in an immortalized human epithelial cell line. Moreover, cells expressing mutant NRF2 showed malignant phenotypes such as enhanced local invasion and metastasis to distant organ. These phenotypes could be explained by acquisition of resistance to or reduction of the oxidative stress induced by the intrinsic cellular immortalization process and extrinsic environmental factors in vivo, but our gene expression analysis revealed that mutant NRF2 affects a broader range of biological processes in addition to the antioxidant pathway.

The aforementioned results imply that mutant NRF2 directly or indirectly regulates a set of genes related to the carcinogenesis process. To clarify this transcriptional network and identify therapeutic targets in Nrf2-mutated cancer, we performed gene expression profiling and conducted “pathway analysis” to connect the expression patterns of gene groups with specific phenotypes. This analysis revealed that, in addition to the previously known NRF2 targets, mutant NRF2 altered the expression of a battery of genes, many of which had not been previously assigned to the oxidative stress pathway, in human epithelial cells. This result partly overlaps (e.g., genes related to lipid metabolism) with the previous gene expression analyses of Keap1-deleted mouse hepatocytes (29, 30) and our findings of a genome-wide study to clarify the in vivo sites of NRF2 binding to chromatin in Keap1 mutated cancer cells.⁶

The finding that mutant NRF2 is implicated in the mTOR pathway had not been anticipated. This relationship is also strengthened by the fact that NRF2 mutant–expressing clones and cancer cells are sensitive to mTOR pathway inhibitors (rapamycin and NVP-BEZ235). The mTOR pathway sensitizes cells to changes in multiple extracellular signals/environments (depletion of growth factors and nutrients or cellular stresses such as hypoxia and DNA damage) and dynamically regulates translation, ribosome biogenesis, macroautophagy, and other processes (32, 33); thus, it would be possible that NRF2-mediated stress adaptation may converge with this environmental sensing system. To determine how NRF2 modulates the mTOR pathway, we focused on the gene expression data and have revealed one candidate, RagD, which is a positive regulator of the mTOR pathway (33, 34) and was induced by mutant NRF2. RagD is a member of the small G-protein family and enhances mTOR activity through direct binding and by recruiting the mTOR complex to the endosomal fraction (26), where mTOR is activated by Rheb, another small G protein, which has been shown to be an oncogene in a mouse model as well as in chick fibroblasts (35, 36). Although the expression of RagD was increased in primary lung cancers, the transcriptional regulation of the Rag gene family has been largely unexplored; it has only been reported that amino acid addition triggers its activation (26). RagD expression in cancer samples was significantly associated with Hmox1, another Nrf2 target gene (P = 0.025; Supplementary Fig. S8). However, the putative promoter region of the RagD gene contains no ARE sequence and our chromatin immunoprecipitates sequence (ChIP-seq) analysis has revealed that RagD is not a direct target of NRF2 (data not shown). These findings suggest that an additional regulatory mediator links the two molecules (Fig. 5D).

Continuous NRF2 activation probably stimulates the mTOR pathway on which cancer cells have become dependent, making it a potential therapeutic target. Identification of such indirect molecular addiction has been recently reported as NF-κB activation in Kras/Tp53 mutated cancer cells (37) and could expand a window of anticancer therapeutic strategy especially when targeting oncoproteins encoding transcriptional factors. Recently, mTOR inhibitors have caught much attention as promising anticancer drugs (38, 39). Previous studies elucidated the existence of feedback mechanism against mTOR inhibitors in cancer (27, 28) and a dual kinase inhibitor (NVP-BEZ235), which targets both the mTOR and PI3Ks, has been recently shown to be more effective in many solid tumors, including Kras mutated lung cancer or Pik3ca mutated cancers (40–42). The present study also supports that the dual kinase inhibition would be beneficial in treating Nrf2 mutated cancer. However, it remains unclear how mutant NRF2 activates AKT in the presence of the negative feedback and further studies to uncover the molecular interaction of NRF2 and PI3K/AKT/mTOR signaling would be required. Notably, we reported that Nrf2 mutation occurs more frequently in lung cancers without Egfr and Kras mutations (15). Therefore, the mTOR pathway may also be a promising therapeutic target in the Nrf2-mutated subtype of lung cancer.

Mutant NRF2 promoted cell motility, local invasion, angiogenesis, and liver metastasis. GO profiling also revealed significant enrichment of genes related to cell motility. Previous studies have shown that upregulation of osteopontin, an extracellular matrix-associated protein with multiple functional domains, is associated with metastatic activity of a range of cancers including head and neck and lung cancers (21, 43) and enhancement of angiogenesis (22). We found that osteopontin was highly expressed in cells expressing mutant NRF2 and that NRF2 seems to directly activate the osteopontin gene. In addition to osteopontin, mutant NRF2 induced a battery of genes such as matrix metalloproteinase 12 and heparin-binding epidermal growth factor-like growth factor, which are associated with tumor metastasis (44, 45). Thus NRF2-mediated adaptation to environmental stress seems to be tightly associated with tumor metastasis and progression. Because our experimental model showed that the presence of mutant NRF2 confers metastatic potential, the molecular signature resulting from NRF2 activation may be associated with poor outcome in clinical cases. We tested this hypothesis and confirmed that the mutant NRF2 signature is indeed a prognostic factor in SCC of the head and neck region and probably that of the lung. Interestingly, an independent study focusing on the prognostic molecular pathway in lung adenocarcinoma has identified the Peng_Rap_DN pathway, a top-ranking gene set enriched in mutant NRF2 clones, as a signature that is significantly associated with poor prognosis (46). This also supports our contention that the mutant NRF2 signature is a marker of poor prognosis.

In conclusion, our results show the existence of an intimate association between oxidative stress and the mTOR pathway in cancer and further highlight the potential of NRF2 activation as a novel biomarker for personalized molecular therapies or prognostic assessment. From a clinical viewpoint, it is important to note that NRF2 activation in cancer is significantly associated with response to anti-mTOR pathway treatments, which are being widely investigated in clinical trials for many cancers.

No potential conflicts of interest were disclosed.

Grant Support: A grant-in-aid for the Comprehensive 10-Year Strategy for Cancer Control, Ministry of Health, Labor and Welfare, Japan grant-in-aid for cancer research 19-1, Princess Takamatsu Cancer Research Fund research grant 08-24007, and Program for Promotion of Fundamental Studies in Health Sciences of the National Institute of Biomedical Innovation, Japan.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.