“Keaping” a lid on lung cancer: the Keap1-Nrf2 pathway

ABSTRACT

Lung cancer remains one of the world’s deadliest cancers, with effective targeted treatment options available for only a small subset of patients. The rapid expansion of cancer genomics in recent years has provided insight into the genetic landscape of all major lung cancer subtypes and led to new discoveries on the heterogeneous biology underlying lung tumorigenesis. Interestingly, these studies have revealed a high frequency of alterations in the Kelch-like ECG-associated protein 1 (KEAP1)–Nuclear factor erythoid–2–related factor 2 (NRF2) stress response pathway, for which no targeted treatments are currently available. In this review, we describe the molecular mechanisms underlying NRF2 pathway activation in lung cancer cells, with a focus on in vivo functional studies in genetically engineered mouse models. Importantly, potential avenues and implications for therapeutic targeting of KEAP1-NRF2 pathway vulnerabilities for lung cancer patients will be highlighted.

Keywords:

• KEAP1
• NFE2L2/NRF2
• lung cancer
• adenocarcinoma
• squamous cell carcinoma
• genetically engineered mouse models

Lung cancer

Lung cancers constitute the highest proportion of cancer-related deaths worldwide, accounting for over 1.5 million deaths annually, an estimated one in five of all cancer-related mortality [1]. Moreover, the predicted five-year survival rate for lung cancer is a mere 17.8%, a figure that is significantly lower than many other types of cancer, such as prostate (99.6%), breast (90.5%) and colon (65.4%) cancer. Lung cancer is classified into two main pathological subgroups: non-small cell lung cancer (NSCLC) and small cell lung cancer (SCLC), comprising 80% and 20% of lung cancers, respectively. NSCLC is a heterogeneous group of diseases encompassed of adenocarcinoma (ADC), squamous cell carcinoma (SqCC) and large cell carcinoma. Recently, comprehensive genomic datasets have been generated for all major lung cancer subtypes [2–4], with attention now focused on validating targetable driver mutations. Among these, mutations in the KEAP1-NRF2 stress response pathway have been identified in 23% of lung ADC [4] and 34% of SqCC [3], for which no targeted treatments are currently available. In this review, we describe recent advances in delineating KEAP1-NRF2 pathway biology in lung cancer. Specifically, we will discuss mechanistic insights learnt from in vivo genetically engineered mouse models (GEMMs) and address how these vulnerabilities can be harnessed for therapeutic intervention in this subset of lung cancer patients.

The KEAP1-NRF2 pathway

One of the most critical stress-response pathways in mammals is mediated by the transcription factor Nuclear factor erythoid-2-related factor 2 (NFE2L2/NRF2), which is negatively regulated by Kelch-like ECG-associated protein 1 (KEAP1). Under homeostatic conditions, NRF2 is degraded via the proteasome through binding to KEAP1, an adaptor protein of the CUL3 E3 ubiquitin ligase (Figure 1). Oxidative stress induces the oxidation of KEAP1 at key cysteine residues which causes a conformational change in KEAP1 releasing NRF2, resulting in translocation and nuclear accumulation of NRF2. In the nucleus, NRF2 forms a heterodimer with its partner sMAF (v-Maf avian musculoaponeurotic fibrosarcoma oncogene homolog) and binds to antioxidant responsive element (ARE) sequences to regulate the transcription of target genes (Figure 1). NRF2 binding activates expression of over 200 genes, many of which are involved in cellular antioxidant, detoxification and metabolic pathways [5]. A major NRF2 transcriptional target is NAD(P)H:quinone dehydrogenase 1 (NQO1), which serves as a robust marker of NRF2 activation in vitro and in vivo [6,7]. It is however important to note that, activation of NRF2 can also occur through KEAP1-independent mechanisms (Figure 1). These mechanisms include competitive binding of NRF2 by proteins such as p53-induced p21 [8] or nutrient sensor p62 [9,10]; inhibition of GSK3β-mediated NRF2 degradation via the activity of the PI3K pathway [11]; nuclear sequestration of NRF2 through phosphorylation at Serine 558 by AMP-activated protein kinase (AMPK) in response to cellular energy stress [12]; among others. For a more detailed description of the additional level of complexity of the regulation of the KEAP1-NRF2 pathway, we refer readers to relevent reviews [13–16].

Figure 1. The KEAP1-NRF2 pathway is altered in human lung cancers.

Schematic of KEAP1-NRF2 pathway under homeostatic conditions (left), oxidative conditions (center) or synergistic conditions, for example in a cancer cell (right). Transcriptional activity of NRF2 is highlighted in the green energy bar. Common mutation hotspots in adenocarcinoma (ADC; black circles), squamous cell carcinoma (SqCC; pink circles) or common to both (half black/pink circles) are highlighted in the protein constructs of KEAP1 and NRF2 (NFE2L2). KEAP1, Kelch-like ECH-associated protein 1; NRF2/NFE2L2, Nuclear factor erythoid-2-related factor 2; CUL3, Cullin 3; D, DLG domain; E, ETGE domain; ARE, antioxidant response element; MAF, v-Maf avian musculoaponeurotic fibrosarcoma oncogene homolog; AKT, RAC-alpha serine/threonine-protein kinase; AMPK, AMP-activated protein kinase; GSK3β, Glycogen synthase kinase 3β.

Deregulation of the KEAP1-NRF2 pathway in lung cancer

Comprehensive genomic sequencing studies have reported deregulation of the KEAP1-NRF2 pathway in human cancer, through both genetic and epigenetic mechanisms (reviewed in [17,18]). Since the initial identification of loss of function mutations in the KEAP1 gene in two lung ADC cell lines [19], multiple studies have revealed the presence of somatic alterations in NSCLC cell lines and lung tumor tissue from patients [3,4,20,21]. KEAP1 inactivating mutations have been detected throughout the protein, with an increased frequency within the Kelch domains (Figure 1), involved in mediating protein-protein interactions with NRF2 [22] and result in nuclear accumulation of NRF2 [19]. In human lung ADC, KEAP1 is the third most commonly mutated gene, behind the tumor suppressor TP53 and KRAS oncogene, where it is found mutated in 19% of patients [4,23]. Interestingly, emerging evidence also suggests that mutations in the KEAP1 gene are less prevalent in non-smokers compared to former or current smokers [24,25]. Recent genomic profiling studies of pulmonary large-cell neuroendocrine carcinomas (LCNECs) revealed the presence of KEAP1 somatic alterations (22%) [26]. Interestingly, KEAP1 alterations are mutually exclusive with inactivating mutations in RB1, and together with somatic mutations of STK11 define a distinct subgroup of LCNECs that exhibit an ASCL1^low/DLL3^low/NOTCH^high neuroendocrine profile, akin to SCLC [26,27].

In lung cancer, gain-of-function mutations in NRF2 appear to be largely restricted to lung SqCC and are mutually exclusive with mutations in KEAP1 and CUL3, which occur less frequently in this cancer subtype [3,28]. Interestingly, high mutation frequencies of NRF2 are also detected in squamous cell cancers of the esophagus, larynx and skin [29], suggesting that augmented NRF2 regulates transcriptional programs that are associated with squamous cell differentiation. Indeed, newborn Keap1-null mice die from severe hyperkeratosis of the esophagus and forestomach, rescued by loss of Nrf2 [30]. Notably, all mutations in the NRF2 gene are localized within the DLG (43%) and ETGE (57%) motifs, which are critical sites for the binding of NRF2 to KEAP1 [31] (Figure 1). Thus, mutations in NRF2 disrupt the high affinity binding to KEAP1, resulting in the stabilization and nuclear accumulation of NRF2 [31].

In lung ADC, recent studies have revealed a significant correlation between increased NQO1 expression and NRF2 pathway activation, suggesting that NQO1 is a clinically relevant biomarker for KEAP1– and/or NRF2-mutant tumors [6,7]. Furthermore, an NRF2 gene signature, composed of 108 high-confidence NRF2 target genes could stratify lung ADC patients that exhibited high-grade tumors and poor survival [7]. Together, these studies reveal an unappreciated role of the KEAP1-NRF2 pathway in lung cancer, suggesting that this pathway is a critical player in protecting lung epithelium from cellular transformation.

Unveiling the function of the KEAP1-NRF2 pathway in lung cancer using genetically engineered mouse models (GEMMs)

GEMMs of lung cancer have emerged as critical bridging strategies between understanding pathogenic mechanism and clinical translation. Importantly, they reveal insights on the events and processes underlying tumor initiation and progression, studies which are not possible utilizing transplantation or chemically-induced model systems. The “gold-standard” lung cancer models are based on Cre-LoxP recombination technology that enable the formation of autochthonous tumors from a limited number of somatic cells. Critically, tumors form in their natural location in the presence of their normal tissue microenvironment. This approach has not only allowed investigators to interrogate the functional consequences of genetic alterations found in human lung cancer but has also provided key insights into the cellular origin of the disease [32]. A loss-of-function (LOF) Keap1 conditional allele (Keap1^f/f) was generated by Blake et al, to study constitutive activation of the Keap1-Nrf2 pathway [33]. Using this GEMM, prolonged activation of the Nrf2 pathway was shown to result in no adverse events in lung function or pathology (aged up to 12 months) [6] (Table 1). However, when combined with mutations that enhance proliferative signals, Keap1 inactivation can drive the malignant transformation of lung epithelium, as discussed below.

Table 1. Mouse models engineered to harbor KEAP1-NRF2 pathway alterations.

Mouse mutant	Features of the GEMM	Histopathology	Reference
Keap1^f/f	Adenoviral Cre-dependent deletion of Keap1 (floxed exons 2 – 5) sporadically in lung epithelium.	Normal lung morphology, no tumor development (12–15 months post-infection)	Best et al., 2018
Keap1^f/f;Pten^f/f (K1P)	Adenoviral Cre-dependent deletion of Keap1 (floxed exons 2 – 3) and Pten (floxed exon 5) sporadically in lung epithelium	Multifocal bronchiolocentric ADC with a papillary-predominant pattern; latency 12 months. Tumors were sensitive to anti-PD-1/anti-CTLA4 inhibition.	Best et al., 2018
Kras^LSL-G12^D/+;Nrf2^−/-	Adenoviral Cre-dependent expression of Kras^G^12^D on a background of Nrf2 germline loss.	Nrf2 loss inhibits Kras^G^12^D-induced lung tumorigenesis.	DeNicola et al., 2012
Kras^LSL-G12D/+;p53^f/f (KP) with Keap1 LOF	Lentiviral sgKeap1-pSECC inhalation: Cre-mediated expression of Kras^G12^D, and deletion of p53 and CRISPR-induced mutation of Keap1.	Keap1 mutation promoted Kras^G12^D;p53^Δ/Δ tumorigenesis by Nrf2 stabilization. Tumors were sensitive to glutaminase inhibition.	Romero et al., 2017
Kras^LSL-G12^D/+;Roas26^LSL-tdTomato;H11^LSL-Cas9 (KT;Cas9)	Inhalation of a pool of barcoded Lenti-sgTS-Pool/Cre vectors: Cre-dependent expression of Kras^G12^D, dtTomato, Cas9 and an sgRNA targeting known/putative tumor suppressor genes^a.	sgRNAs targeting the Keap1 gene were not enriched in Kras^G12^D-induced lung tumors.	Rogers et al., 2017; Rogers et al., 2018
Kras^LSL-G12D/+;p53^f/f; Roas26^LSL-tdTomato;H11^LSL-Cas9 (KPT;Cas9)	Inhalation of a pool of barcoded Lenti-sgTS-Pool/Cre vectors: Cre-dependent expression of Kras^G12D, dtTomato, Cas9 and inactivation of p53 and CRISPR-Cas9-mediated mutation of known/putative tumor suppressor genes^a.	sgRNAs targeting the Keap1 gene were not enriched in Kras^G12D;p53^Δ/Δ lung tumors.	Rogers et al., 2017; Rogers et al., 2018
Kras^LSL-G12D/+;Lkb1^f/f; Roas26^LSL-tdTomato;H11^LSL-Cas (KLT;Cas9)	Inhalation of a pool of barcoded Lenti-sgTS-Pool/Cre vectors: Cre-dependent expression of Kras^G12D, dtTomato, Cas9 and inactivation of Lkb1 and CRISPR-Cas9-mediated mutation of known/putative tumor suppressor genes^a.	sgRNAs targeting the Keap1 gene were not enriched in Kras^G12D;Lkb1^Δ/Δ lung tumors.	Rogers et al., 2017; Rogers et al., 2018
Kras^LSL-G12D/+;Lkb1^f/f	Lentiviral Nrf2-Cre inhalation: expression of Nrf2, oncogenic Kras^G12D and concomitant deletion of Lkb1.	Nrf2 overexpression decreased the incidence of SqCC and increased ADC formation.	Li et al., 2015

^aLenti-sgTS-Pool/Cre contains 4 vectors with inert sgRNAs and 11 vectors targeting known and candidate tumor suppressor genes (Setd2, Lkb1, Rb1, Apc, Rbm10, Cdkn2a, Trp53, Keap1, Smad4, Arid1a and Atm)

Adenocarcinoma

KRAS, TP53 and KEAP1 are the three most common genetic mutations in lung ADC, and mutations in TP53 and KEAP1 occur in 26% and 18% of KRAS-mutant lung ADC, respectively [34]. The functional consequences of Keap1 inactivation on Kras^G12D-induced lung cancer formation has been explored using the Kras^G12D conditional knock-in mouse model, whereby a glycine to aspartic acid substitution at codon 12 was engineered in the endogenous Kras locus. Expression of oncogenic Kras^G12D is controlled by a LoxP-Stop-LoxP (LSL) cassette, which consists of transcriptional and translational stop elements flanked by LoxP sites [35] (Kras^LSL-G12D/+; hereafter Kras^G12D). Intra-nasal delivery of an adenovirus or lentivirus expressing Cre-recombinase under the control of the ubiquitous cytomegalovirus (CMV) promoter (Ad5-CMV-Cre) results in the removal of the stop elements and formation of lung ADC in Kras^G12D mice [36]. Critically, additional loss of the p53 tumor suppressor gene accelerates tumorigenesis. Tumors that develop in Kras^G12D;p53^f/f mice (hereafter KP) more closely mimic their human counterpart and exhibit an aggressive phenotype, which promotes the local dissemination of tumor cells [37,38]. While combined loss of Lkb1, detected in 17% of KRAS-mutant lung ADC [4,34], accelerates Kras^G12D tumorigenesis and promotes ADC to SqCC transformation [39,40].

A bespoke in vivo screening approach based on CRISPR-Cas9 gene editing combined with high-throughput barcode sequencing (Tuba-seq) [41] was employed to interrogate the functional consequences of validated and candidate tumor suppressor genes found commonly altered in human lung ADC [4,42]. Kras^G12D, KP and Kras^G12D;Lkb1^f/f mice engineered to carry a Cre-inducible tdTomato reporter allele, and a LoxP-Stop-LoxP-Cas9 allele were infected with pool libraries and lung tumor size was evaluated 15 weeks following Lenti-sgRNA/Cre infection. While this approach was successful in identifying new genetic interactions, Keap1 inactivation did not provide a significant growth advantage to Kras^G12D tumors, nor to Kras^G12D tumors that had concomitant loss of either p53 or Lkb1 [41,42] (Table 1). Interestingly, sgRNAs targeting p53 were also not enriched in KT;Cas9 mice under these experimental conditions, which contradicts findings using conventional Cre/loxP-based mouse models [37]. The absence of an effect of sgRNAs targeting Keap1 in the background of Lkb1-deficiency contradict recent bioinformatic analyses that revealed a strong co-occurrence of KEAP1 inactivation and loss of function mutations in STK11 (Lkb1) [28,34].

In a more targeted approach, Romero et al., intra-tracheally infected KP mice with pSECC lentiviral vector expressing a single sgRNA against Keap1 [7,43]. Notably, compared to KP mice infected with a control sgRNA (tdTomato), sgKeap1-infected mice exhibited increased tumor burden suggesting that Keap1 loss accelerates Kras^G12D-driven lung cancer (Table 1). Keap1 LOF tumors displayed augmented KEAP1-NRF2 pathway activation, characterized by increased NRF2 nuclear localization and elevated expression of Nqo1 [7]. Moreover, the levels of reactive oxygen species (ROS) were significantly lower in Keap1 LOF tumors. Tumorigenesis was also augmented in the presence of sgRNAs targeting Nrf2, which induced gain-of-function mutations. Together, these findings implicate a crucial role for the oxidative stress response in Kras^G12D-driven lung cancer. Curiously, LOF mutations in KEAP1 occur in 18% of KRAS-mutant lung ADC and are largely mutually exclusive with mutations in TP53 [34]. The findings of Romero et al., would therefore argue against the idea that TP53 and KEAP1 are synthetic lethal. Further research is therefore required to dissect whether the mechanisms of p53 and Keap1 tumor suppression overlap. An in-depth comparative analysis of tumorigenesis in Kras^G12D;Keap1^f/f and KP mice would assist in addressing these outstanding questions.

A functional interaction between the KEAP1-NRF2 and phosphatidylinositide 3-kinase (PI3K) pathways has been reported to promote the malignant evolution of tumors [25,44]. Given that genetic inactivation of both KEAP1 and PTEN have been identified in lung ADC and lung SqCC, a GEMM based on conditional loss of Keap1 and Pten was generated [6]. Consistent with previous reports, significant synergy in Nrf2 activity, measured by augmented expression of Nqo1, was identified in Keap1^f/f/Pten^f/f (hereafter K1P) tumor cells, compared to the modest increase in Nqo1 following the loss of Keap1 alone [6]. These findings are in line with the regulation of Nrf2 by GSK3β [11,45] (Figure 1). Critically, combined loss of Keap1 and Pten led to bronchiolocentric ADC formation [6] (Table 1). On a molecular level, the tumors that developed in this model resemble the salient features of human ADC stratified on the basis of high NQO1 mRNA expression [4], a robust biomarker of KEAP1-NRF2 pathway activity [6,7].

Squamous cell carcinoma

The transcriptome of lung basal cells most closely correlates with the transcriptional profile of human lung SqCC [46]. Indeed, landmark studies have revealed that different lung cancer subtypes arise from distinct lung epithelial cell lineages [47,48], suggesting that basal cells are the cells-of-origin of lung SqCC [49]. Given that recent studies demonstrated that Keap1 loss promoted the self-renewal of basal stem cells in vitro [50], we hypothesized that combined loss of Keap1 and Pten specifically in basal progenitor cells would drive lung SqCC formation in vivo. To directly test this, we infected K1P mice treated with polidocanol one day prior to infection with the Ad5-K14-Cre virus infection that directs Cre-recombinase expression to basal cells [51,52] (Figure 2A). We, and others have shown that pre-treatment with polidocanol causes damage of the luminal epithelium, enhancing the infection of the exposed basal cell layer [53,54]. Critically, when combined loss of Keap1 and Pten was restricted to basal progenitor cells, lung SqCC was observed, 12 months following Ad5-K14-Cre infection (Figure 2C). These tumors, like their human counterpart, stained positive for keratin 5 (K5), expressed high levels of Nqo1 and Sox2, but stained negative for TTF-1/Nkx2.1, a marker of lung ADC (Figure 2D). Interestingly, tumors exhibiting histological and morphological features of lung ADC were also observed in the lungs of K1P mice (Figure 2C, D), likely resulting from “switched” Club cells repopulated following polidocanol-induced injury (Figure 2E). Taken together, these findings suggest that combined loss of Keap1 and Pten has the ability to drive SqCC formation when inactivated in basal progenitor cells, highlighting the importance of the cell-of-origin in dictating the tumor phenotype.

Figure 2. Directed loss of Keap1 and Pten in lung basal progenitor cells promotes lung SqCC formation.

(A) Timeline of polidocanol (POL) administration and adenovirus infection in Keap1^f/f/Pten^f/f mice. Briefly, mice were intra-tracheally (i.t.) injected with 10 µl of 2% w/v polidocanol (POL) one day prior to intra-nasal infection of Ad5-K14-Cre virus. Mice were sacrificed, and lung tissue analyzed 12 months following adenoviral infection. (B) Timeline of Ad5-CMV-Cre virus infection in Keap1^f/f/Pten^f/f/LSL-Sox2 (K1PS) mice. (C) Quantification of histological phenotype of lung tumors in Ad5-CMV-Cre-infected Keap1^f/f/Pten^f/f (K1P) mice [6] (n = 6), Ad5-K14-Cre-infected K1P mice (n = 5) and Ad5-CMV-Cre-infected K1PS mice (n = 3). (D) Representative Hematoxylin & Eosin (H&E) and immunostained sections of lung ADC and SqCC tumors detected in Keap1^f/f/Pten^f/f mice 12 months following Ad5-K14-Cre infection. Scale bars; H&E-stained sections 200 µM; Nkx2.1, Nqo1, K5 and Sox2 50 µM. (E) Schematic representation of the consquences of combined loss of Keap1 and Pten in distinct cell types or when combined with enforced Sox2 expression. Combined loss of Keap1 and Pten in bronchiolar and/or alveolar type 2 (AT2) epithelium results in the formation of ADC tumors, while restricted loss of Keap1 and Pten to basal progenitor cells promotes SqCC formation. Interestingly, overexpression of Sox2 in K1P mice promoted the ADC to SqCC transdifferentiation, likely initiated from a “switched” Club or AT2 cell.

Genomic profiling studies have enabled the stratification of lung SqCC into 4 molecular subgroups: classical (36%), basal (25%), secretory (24%) and primitive (15%) [3,55]. The classical subtype is characterized by alterations in the KEAP1-NRF2 pathway, as well as inactivation and/or loss of the tumor suppressor gene PTEN and amplification of chromosome 3q, containing the SOX2, TP63 and PIK3CA genes [3]. SOX2 has been hypothesized to act as a lineage-specific oncogene in human lung SqCC, where is found to be amplified and/or overexpressed in 56% of cancers [56]. Indeed, more recently, we and others have shown that overexpression of SOX2 is a potent driver of lung SqCC formation when combined with either loss of Pten and Cdkn2ab [51] or loss of Lkb1 alone [57]. SOX2 overexpression, however, can induce SqCC formation from Club (CC10-expressing), alveolar type 2 (AT2; SPC-expressing), in addition to basal progenitor cells, suggesting that enforced SOX2 expression, rather than the initiating cell, dictates the resulting tumor phenotype [51]. We therefore hypothesized that overexpression of Sox2 in combination with loss of both Keap1 and Pten in basal cells, would promote lung SqCC with high penetrance. To address this, we crossed LSL-Sox2 mice [51] with Keap1^f/f;Pten^f/f mice [6] (LSL-Sox2;Keap1^f/f;Pten^f/f, hereafter K1PS mice) (Figure 2B). Consistent with observations made in K1P mice following Ad5-CMV-Cre infection, pronounced bronchiolar hyperplasia was observed throughout the large and small airways of K1PS mice (data not shown). In addition however, tumors exhibiting characteristics of both ADC and SqCC were observed in K1PS mice, suggesting that enhanced Sox2 expression promotes ADC-SqCC transdifferentiation [58] (Figure 2C), with the most progessed tumors detected in the lungs of K1PS mice exhibiting a pure SqCC tumor phenotype. These data suggest that alterations in the KEAP1-NRF2 pathway can promote SqCC development, but this is dependent upon on the cellular origins and/or the genotype of co-occurring mutations (Figure 2E).

NRF2 regulation of tumor metabolism

Tumor cells utilize a high rate of glycolysis to generate macromolecules essential for the synthesis of new DNA, RNA, cellular membranes and protein [59,60]. Indeed, lung cancers express high levels of enzymes involved in glycolysis, with a significant correlation between the glucose transporter GLUT1 and aggressive ADC [61]. In addition to regulating genes that confer the inducible resistance to xenobiotic and oxidative stress, NRF2 activity alters the expression of key metabolic enzymes [15,16]. Key studies of altered lung cancer metabolism have utilized the A549 cell line, a human lung ADC with KEAP1 LOF via point mutation (KEAP1^G333C) and loss of heterozygosity (19p13.2) [21]. Using A549 cells, in vitro studies identified a key role for NRF2-activated transcription in enhancing serine biosynthesis and pentose phosphate pathway (PPP) activity to generate key metabolic intermediates for tumor cell proliferation [44,62,63]. Recently, in vivo studies have detected the altered metabolism of Keap1-mutant lung tumors as a distinct metabolite signature peripherally in the plasma of tumor-bearing GEMMs [6]. Specifically, serum levels of lactate and sugar phosphates (Glc6P, Rib5P, and Ru5P), as well as a number of nucleotides and nucleosides, were significantly decreased in tumor-bearing mice, consistent with an increase in PPP enzyme transaldolase 1 (Taldo1) expression.

Taldo1 is a key enzyme in the non-oxidative arm of the PPP, generating ribose 5-phosphate (Rib5P) and NADPH, essential for cell proliferation. In rapidly proliferating cancer cells, over 85% of ribonucleotides are derived from the non-oxidative PPP [64]. Enhanced PPP activity is achieved through the constitutive activation of NRF2, and results in enhanced proliferation rates and ribose synthesis in A549 cells [44,63]. Tumor studies using a YFP reporter of Keap1 loss, and thus Nrf2 nuclear localization, have identified key regulatory patterns of enzyme regulation in NSCLC [6]. Though expression of PPP enzymes G6PD, Tkt, Pgd and Taldo1 were increased in the tumor as a whole compared to untransformed lung epithelial cells, only the expression of Taldo1 was specific to YFP positive tumor cells [6]. Indeed, the expression of G6PD, Tkt and Pgd are regulated generally by an oxidative stress axis involving HDAC4 with miR-1 and miR-206 [63]. Moreover, while G6PD, TKT and PGD promoters contain an ARE that can be bound by NRF2 [58], the location is distant from the transcription start site suggesting that additional elements are required for robust gene activation. TALDO1 alone has been confirmed in two independent reports to be directly regulated by NRF2 [44,63], highlighting the direct relationship between NRF2 and the non-oxidative arm of the PPP.

In addition to the PPP, NRF2 regulates the expression of enzymes involved in the serine/glycine biosynthetic pathway [62]. Serine and glycine are important metabolic intermediates in the synthesis of proteins, nucleic acids and lipids that support tumor processes [15,65]. Through the NRF2-dependent regulation of Activating Transcription Factor 4 (ATF4), the enzymes PHGDH, PSAT1 and SHMT2 are transcribed in KEAP1–mutant NSCLC cells that render these cancer cells self-sufficient in the biosynthesis of serine and glycine [62]. Importantly, the production of serine and glycine was capable of supporting glutathione (GSH) biosynthesis, purine biosynthesis, thymidine production and NADPH levels, supporting NSCLC cell growth [62]. Furthermore, KEAP1 LOF renders mouse and human cells dependent on the non-essential amino acid glutamine, crucial for the generation of the antioxidant GSH [66].

The NRF2 pathway is intimately linked to metabolic processes that are essential for cancer cell proliferation and survival. The enhanced utility of these pathways creates unique characteristics of KEAP1– and/or NRF2-mutant NSCLC, that can be harnessed as both a biomarker [6] and exploited for targeted therapy, discussed below.

Targeting the KEAP1-NRF2 pathway in the treatment of cancer

Cancers with high NRF2 levels are associated with poor prognosis [67], due to their chemotherapy and radiotherapy resistance and proliferative nature [50,68–70]. This may, in part, be due to the NRF2 mediated up-regulation of Multidrug Resistance-associated Protein 1 (MRP1/ABCC1) [71]. MRP1 is an energy-dependent efflux pump that exports bulky hydrophobic molecules from the cell, such as Paclitaxel, Methotrexate, Doxorubicin and Flutamide chemotherapeutic classes [72]. MRP1 is expressed frequently in lung cancer and is a predictor of poor response to chemotherapy [73,74]. NRF2-mediated upregulation of MRP1 may be a direct mediator of resistance to standard lung cancer treatment and thus targeted treatment to mitigate the activity of NRF2 are urgently required.

Direct inhibition of NRF2

The activity of NRF2 promotes resistance to platinum-based chemotherapy, the current first-line treatment of NSCLC, and is thus itself a molecular target for therapy prompting the identification of NRF2 inhibitors [75]. In cancer cell lines, Brusatol [76], Trigonelline [77], Ascorbic acid [78], Chrysin (5,7-digydroxyflavone) [79], Apigenin (4ʹ,5,7-trihydroxyflavone) [80] and Luteolin (3ʹ,4ʹ,5,7-tetrahydroxyflavone) [81] were found to be effective inhibitors of NRF2 pathway activity. Specifically, the activity of brusatol and luteolin were investigated in A549 lung ADC cells. Brusatol acts to reduce the abundance of NRF2 through enhanced ubiquitination and degradation, reducing NRF2-protective response in chemotherapy treated cell lines. Significantly, brusatol sensitized A549 xenograft models to the activity of cisplatin in a KEAP1-NRF2 dependent mechanism [76]. Similarly, the flavonoid luteolin accelerates NRF2 mRNA turnover and sensitizes A549 cells to oxaliplatin, bleomycin and doxorubicin in cell line viability assays [81]. Additionally, targeted chemical screening approaches have identified NRF2 inhibitors, such as ML385 [82] and AEM1 [83], that similarly sensitize KEAP1-mutant NSCLC cell lines to combination therapy. Though these compounds provide strong evidence of activity in cell lines, no clinical trials in lung cancer have yet been performed to validate the beneficial effect of NRF2 inhibition on sensitizing ADC to chemotherapy. These studies strongly suggest there is an advantage to decreasing NRF2 stabilization in promoting sensitivity of lung ADC cells to platinum-based chemotherapeutic agents.

Inhibition of NRF2 downstream metabolic pathways

Lung cancer with high NRF2 activity, determined via the expression level of NQO1 [6,7], have enhanced PPP and serine biosynthetic pathway enzyme expression, consistent with preclinical studies [6,7,62]. Abrogation of these pathways in culture have indicated that metabolic targets may be viable treatment options in lung cancer with highly active NRF2. Indeed, blocking the oxidative and non-oxidative arms of the PPP using combined knockdown of G6PD and TKT repressed growth of A549 in vitro and in xenograft models in vivo [44]. Similarly, knockdown of serine and glycine biosynthesis enzyme PHGDH reduced colony formation and xenograft growth in A549 KEAP1-mutant cell lines [62]. Utilizing a translational approach, a small molecule inhibitor of PHGDH has been developed (CBR-5884) and was found to be efficacious in cancer cells with high serine biosynthetic activity [84]. Compellingly, drug studies in cell lines and orthotopic transplant models of KP and Kras^G12D/p53/Keap1 spontaneous lung tumors found that Keap1-mutant lung tumors are sensitive to the glutaminase inhibitor CB-839 [7,66]. In addition to rational drug design, an unbiased chemical screen recently identified a GLUT8-dependent pathway that selectively sensitized KEAP1-mutant cells based on their consumption of glucose to support serine biosynthetic pathway activity [85]. These data, while still preliminary, provide evidence that targeting downstream GSH and PPP metabolic pathways are viable in NRF2-active lung cancers, and highlight the dependence of these cancers on NRF2-driven metabolic pathways.

Immunotherapy in KEAP1-mutant lung cancers

Immunotherapy provides additional promise in KEAP1-mutant lung tumors. Unleashing a patient’s autologous T cells to target tumors has received attention in lung cancer due to a high mutation load and level of immune infiltration [86,87]. In a recent clinical trial (KEYNOTE-189), non-squamous NSCLC (excluding EGFR or ALK mutant ADC) found that the addition of Pembrolizumab to chemotherapy enhanced both the progression-free and overall survival of lung cancer patients [88]. GEMMs are crucial for preclinical studies of immunotherapy, due to the spontaneous generation of a primary tumor in an intact immune microenvironment. The efficacy of checkpoint inhibition using monoclonal antibodies that target programmed cell death protein-1 (PD-1) has been tested in a number of lung cancer GEMMs, and while the efficacy was low in KP GEMMs [89,90], sensitivity to immunotherapy was achieved in the NRF2-active K1P GEMM [6]. This suggests that activation of the NRF2 pathway may provide additional immunomodulation that sensitizes these ADC to immunotherapy. Significantly, RNA sequencing data provide insight into the immune landscape of ADC [4]. Analysis of lung ADC with high NQO1 expression identified significantly reduced immune infiltration signatures [6]. Similarly, mutation in KEAP1 correlates with reduced leukocyte infiltration in lung ADC [91]. Together, these findings suggest that aberrant NRF2 pathway activity alters the immune microenvironment of lung ADC, which improves the efficacy of immunotherapy. Indeed, the K1P GEMM exhibited increased PD-L1 tumor cell expression [6], and enhanced serine biosynthesis results in phosphotidyl-serine (PS) exposure on epithelial cells [15], together suggesting that NRF2-active tumor cells can mediate immune evasion in the tumor microenvironment. Further studies are required to verify whether NRF2 pathway activation correlates with immunotherapy efficacy in clinical trials, however preclinical evidence suggests that there is significant merit in selecting patients with high NQO1 expression for treatment with immunotherapy.

On multiple levels of tumor progression and resistance to treatment, NRF2 transcriptional activity provides advantages to the cancer cell. Proliferative advantages arise from resistance to oxidative stress, enhanced NAPDH production and metabolic pathway activity. In addition to the advantages of these cellular systems under toxic chemotherapy insult, enhanced efflux pump expression exports from the cell the classes of chemotherapeutic drugs that are most commonly used to treat lung cancer. Thus, alternative treatments to target this aggressive subset of lung cancer are urgently required. While metabolic targets and immunotherapy provide limited tumor control, additional studies are required to identify synergistic combinations to control KEAP1-NRF2 altered tumors. This review highlights recent findings in the role of KEAP1-NRF2 pathway in lung cancer and provides hope for additional advances in the field for the treatment of this aggressive disease.

Acknowledgments

We are grateful to A. Kersbergen for technical assistance and C. Alvarado, L. Scott and H. Johnson for animal husbandry; E. Tsui in the WEHI Histology Facility for expert support; A. Berns (Netherlands Cancer Institute) for LSL-Sox2 conditional mice; and The University of Iowa Gene Transfer Core Facility for purified Ad5-Cre viruses. All experiments presented in this manuscript were conducted according to regulatory standards approved by the Walter and Eliza Hall Institute Animal Ethics Committee. We are thankful to A. Cox (Peter MacCallum Cancer Centre) and Giustina Ferone (Netherlands Cancer Institute) for critical reading of the manuscript. This work was supported by an Australian Health and Medical Research Council (NHMRC) grant awarded to K.D.S (APP1138275). S.A.B is supported by a Victorian Cancer Agency (VCA) Early Career Seed Grant (ECSG16001); K.D.S is supported by the Peter and Julie Alston Centenary Fellowship. This work is made possible through Victorian Government Operational Infrastructure Support and the Australian government.

Disclosure statement

No potential conflict of interest was reported by the authors.

Additional information

Funding

This work was supported by the National Health and Medical Research Council [APP1138275];Victorian Cancer Agency [ECSG16001].