Role of KEAP1/NRF2 and TP53 Mutations in Lung Squamous Cell Carcinoma Development and Radiation Resistance

Lung cancer is the second most common cancer in both men and women and the most common cause of cancer death in the United States (1). Lung squamous cell carcinoma (LSCC) comprises a large fraction of non–small cell lung cancers (NSCLC) and annually accounts for 50,000 deaths in the United States. Unlike for lung adenocarcinoma, for which multiple targeted therapies are available, no first-line targeted therapies are currently clinically available for LSCCs. Furthermore, their pathogenesis and cell of origin remain poorly understood, and biomarkers that predict therapeutic responses are lacking. Progress in the field has been slow, in part due to the lack of reliable animal models of LSCCs. Recently, several LSCC mouse models have been reported, involving inactivation of IKKα, Lkb1, and Pten or activation of Sox2 (2–4). However, a large fraction of mutations that are recurrently found in LSCCs are not included in these studies, and a need remains for the development of additional mouse models of the disease (5).

The Cancer Genome Atlas (TCGA) project has provided a comprehensive landscape of genomic alterations of LSCCs (5). This revealed, as expected, that TP53 is the most frequently mutated gene, occurring in more than 80% of all LSCCs. In addition to mutations affecting several other pathways, KEAP1/NRF2 pathway mutations were found in over one third of patients with LSCC. The KEAP1–NRF2 pathway is involved in protection of cells from oxidative and toxic stresses. NRF2 (also known as NFE2L2) is a transcription factor and master regulator of phase II detoxifying and antioxidant genes (6). At homeostasis, NRF2 is bound by the adapter protein KEAP1, which recruits the CUL3 ubiquitin ligase, leading to proteasomal degradation of NRF2 (7, 8). In response to oxidative stress, NRF2 is released from KEAP1, translocates into the nucleus, and promotes the transcription of genes involved in defenses against reactive oxygen species (ROS), such as GCLM, GCLC, G6PD2, and NQO1. Recent studies have reported that the KEAP1–NRF2 pathway and levels of ROS contribute to the development and progression of lung cancer (9–13). However, the effects of somatic alterations in KEAP1/NRF2 and TP53 on airway stem cells and LSCC tumorigenesis and metastasis have not been deeply explored. Importantly, although a number of groups have examined the effects of KEAP1 and NRF2 on treatment resistance in lung cancer cell lines (14), no studies have done so in genetically engineered mouse models and no studies have demonstrated a clinical association between mutations in KEAP1/NRF2 and response to radiotherapy in patients with lung cancer.

Here, we explore the role of the KEAP1–NRF2 pathway and Trp53 in the self-renewal of airway basal stem cells (ABSC), LSCC pathogenesis, and prediction of radiation resistance. We found that deletion of Trp53 or Keap1 in tracheal epithelial cells promotes ABSC self-renewal. Furthermore, deletion of Trp53 with or without Keap1 in tracheal cells leads to the formation of lung cancer with features of squamous cell carcinoma (SCC), whereas the same deletions in peripheral lung cells lead to adenocarcinoma formation. We further demonstrate that ABSCs are the cell of origin for the LSCC in these models. Also, constitutive NRF2 activation and ROS suppression by Keap1 deletion promoted tumor aggressiveness, metastasis, and resistance to oxidative stress and radiotherapy (RT). Treatment with sulfasalazine, an inhibitor of the antiporter system x_c⁻, overcame Keap1 deletion–mediated radioresistance. Consistently, KEAP1/NRF2 mutation status in patients with NSCLC was predictive of local recurrence after RT in human patients, and these mutations could be noninvasively identified in circulating tumor DNA (ctDNA). Our findings suggest that system x_c⁻ is a potential target for personalized radiosensitization of patients with NSCLC harboring KEAP1/NRF2 mutations and that KEAP1/NRF2 mutation status is a potential predictive marker for clinical decision making in the treatment of patients with NSCLC.

In order to explore the role of mutations in Keap1 and Trp53 in LSCC pathogenesis, we began by examining the effects of loss of these genes on the self-renewal of ABSCs, the hypothesized cell of origin for LSCC (15). Bulk tracheal epithelial cells from wild-type (WT) R26^tdTomato, Keap1^f/f;R26^tdTomato or Trp53^f/f;R26^tdTomato mice were transduced with adeno-Cre (Ad-Cre) virus to mimic inactivating mutations of Keap1 or Trp53 (Supplementary Fig. S1A and S1B). In vitro tracheosphere assays revealed that Keap1 deletion or silencing increased primary tracheosphere formation by 2- to 3-fold compared with WT cells, consistent with a previous study (ref. 16; Fig. 1A and Supplementary Fig. S1C). To further analyze self-renewal in vitro, we dissociated primary tracheospheres into single cells, FACS-sorted tdTomato⁺ or GFP⁺ cells, and replated the same number of cells for secondary tracheosphere formation. Secondary tracheosphere formation of Keap1-deleted or silenced cells was ∼2-fold greater than WT cells (Fig. 1B and Supplementary Fig. S1D). Consistently, upon the single-cell sorting, Keap1^−/− basal cells from tamoxifen-injected Krt5^CreERT2;Keap1^f/f; R26^tdTomato mice formed ∼70% more tracheospheres than WT basal cells from tamoxifen-injected Krt5^CreERT2;R26^tdTomato mice (Supplementary Fig. S1E). Similarly, Trp53 deletion enhanced primary and secondary tracheosphere formation by 60% to 80% (Fig. 1C and D). These data indicate that inactivation of Keap1 or Trp53 promotes ABSC self-renewal in vitro.

Because ∼80% of LSCCs with mutations in the KEAP1–NRF2 pathway also carry inactivating mutations in TP53, we next examined the effects of simultaneous deletion of Keap1 and Trp53. Homozygous deletion of both genes led to an additive increase in tracheosphere number (Fig. 1E). Furthermore, tracheospheres derived from Keap1^−/−;Trp53−/−; cells displayed aberrant morphologic features manifesting as solid spheres devoid of a central acellular lumen, whereas tracheospheres derived from WT, Keap1^−/−, and Keap1^f/f;Trp53^f/f cells displayed the expected cystic shape. Tracheospheres derived from Trp53−/− cells displayed a mixed phenotype, with a majority of cystic spheres and a minority of compact spheres (Fig. 1F and G and Supplementary Fig. S1F). These data indicate that combined deletion of Keap1 and Trp53 leads to greater enhancement of ABSC self-renewal than deletion of either gene alone, and promotes the deregulated expansion of ABSCs.

To confirm these findings in vivo, we developed a population-based lineage-tracing approach for tracheal epithelial cells. Although intranasal instillation of Ad-Cre induces recombination of peripheral lung cells (17), tracheal epithelial cells are not efficiently transduced, likely due to action of the mucociliary apparatus (Fig. 2A and Supplementary Fig. S2A). We hypothesized that denuding of luminal tracheal cells by SO₂ would allow transduction of tracheal basal cells. Indeed, Ad-Cre intranasal instillation at 20 hours after SO₂ injury led to recombination in up to 40% of tracheal cells. At 3 days post-injury (dpi), tdTomato⁺ cells uniformly expressed KRT5 but not acetylated tubulin (ACT), indicating denudation of luminal cells and transduction of basal cells (Fig. 2B). At 9 dpi, however, tdTomato⁺ cells expressing either KRT5 or ACT were observed, indicating that transduced basal cells had generated luminal cells (Fig. 2C). We used this approach to test if the ratio of basal and luminal tracheal epithelial cells was altered by deletion of Keap1 or Trp53. Specifically, we administered Ad-Cre viruses intranasally to WT R26^tdTomato, Keap1^f/f;R26^tdTomato, and Trp53^f/f;R26^tdTomato immunocompetent mice after SO₂ injury and analyzed the ratio of basal and luminal cells by flow cytometry at 7 dpi. Deletion of either Keap1 or Trp53 increased the basal-to-luminal ratio about ∼2.5-fold compared with WT or nonrecombined Keap1^f/f or Trp53^f/f cells (Fig. 2D–F). In addition, FACS-sorted Keap1^−/− tdTomato⁺ cells formed ∼3 fold more tracheospheres than WT tdTomato⁺ cells (Supplementary Fig. S2B). Thus, deletion of Keap1 or Trp53 in bulk tracheal epithelial cells leads to expansion of ABSCs in vivo.

These findings prompted us to examine whether deletion of Trp53 and/or Keap1 in tracheal epithelial cells leads to LSCC formation. We first attempted to establish tumors in vivo using SO₂ injury followed by Ad-Cre intranasal instillation into Keap1^f/f;Trp53^f/f;R26^tdTomato immunocompetent mice (Fig. 3A). However, as previously observed with Trp53^f/f mice treated with endotracheal Ad-Cre without SO₂ injury (18), Trp53^f/f;R26^tdTomato or Keap1^f/f;Trp53^f/f;R26^tdTomato mice developed SFTPC⁺NKX2-1⁺KRT5⁻ adenocarcinomas in the lung periphery that necessitated euthanasia, regardless of SO₂ injury (Fig. 3B and Supplementary Fig. S3A). No LSCCs were observed, suggesting that the development of adenocarcinomas may not have allowed sufficient time for tumors to emerge in central airways. This observation is consistent with latency times of 8 to 10 months observed in other genetically or chemically induced LSCC models (2, 19, 20). To further test whether Trp53-deleted cells in the peripheral lung form lung adenocarcinomas (P-LUAD), we administered Ad-Cre to Trp53^f/f;R26^tdTomato mice in the absence of SO₂ injury and sorted tdTomato⁺ epithelial cells from the lung periphery after 1 to 2 weeks. When these cells were transplanted subcutaneously into NOD/SCID IL2Rgamma^null (NSG) mice, we observed formation of SFTPC⁺NKX2-1⁺KRT5⁻ adenocarcinomas after ∼3 months (Fig. 3C and D and Supplementary Fig. S3B). Finally, when we sorted type II pneumocytes from Ad-Cre transduced Trp53^f/f;R26^tdTomato mice based on expression of EpCAM and SCA1 (21) and transplanted these into NSG mice, we observed adenocarcinoma formation (data not shown). Thus, deletion of Trp53 with or without Keap1 in peripheral lung cells leads to lung adenocarcinoma and not squamous cell carcinoma.

We reasoned that the development of lung adenocarcinoma in these mice may not allow enough time for development of LSCCs. As an alternate strategy, we next treated Krt5^CreERT2;Keap1^f/f;Trp53^f/f;R26^tdTomato mice with tamoxifen and monitored animals for tumor formation. Unfortunately, these mice developed aggressive skin cancers requiring euthanasia as early as 1 month after treatment due to expression of KRT5 in epidermal basal cells, confounding analysis of LSCC development. Therefore, we next sought to directly test if tracheal cells can give rise to LSCCs. Specifically, we harvested bulk tracheal epithelial cells from WT R26^tdTomato, Keap1^f/f;R26^tdTomato, Trp53^f/f;R26^tdTomato, and Keap1^f/f;Trp53^f/f;R26^tdTomato tracheae, transduced between 10,000 and 30,000 cells with Ad-Cre overnight, and transplanted these subcutaneously into NSG mice (Fig. 3E). Within 2 to 4 months, tumors appeared in >80% of animals transplanted with either Trp53^−/− or Keap1^−/−;Trp53^−/− cells, but never from WT or Keap1^−/− cells (Fig. 3F). Both Trp53 and Keap1 were confirmed to be homozygously deleted in these tumors (Supplementary Fig. S3C). Additionally, these tumors were orthotopically transplantable into lungs (Supplementary Fig. S3D).

Both Trp53-deleted (P-LSCC) and Keap1/Trp53–double-deleted (K/P-LSCC) tumors displayed histologic features of poorly differentiated squamous cell carcinomas, including squamous morphology, nuclear pleomorphism, and dense eosinophilic keratin deposits (Fig. 3G). Consistent with their poorly differentiated histologic appearance, both P-LSCC and K/P-LSCC showed high tumor-initiating cell frequency of ∼1/10 in in vivo–limiting dilution analyses (Supplementary Table S1). These histologic features were very similar to those observed in patient-derived LSCC xenografts. By immunostaining, K/P- and P-LSCCs, as well as their orthografts, were found to express KRT5, ITGA6, and SOX2, but not TTF1 and SFTPC, again similar to findings from LSCC patient–derived xenografts (Fig. 3G and Supplementary Fig. S3E). Furthermore, K/P-LSCC and P-LSCC tumors expressed high levels of Krt5, p63, and Sox2, and minimal levels of Sftpc, Nkx2-1, and Cgrp mRNA (Fig. 3H and Supplementary Fig. S3F). Additionally, compared with WT lung tissue, Krt5, Krt14, and p63 mRNAs were significantly more highly expressed in K/P-LSCCs whereas Sftpc expression was negligible (Supplementary Fig. S3G). Additionally, these tumors metastasized to the lung, and the metastatic deposits also expressed KRT5 but not SFTPC (Fig. 3I and Supplementary Fig. S3H and S3I). Thus, whereas deletion of Trp53 with or without deletion of Keap1 in peripheral lung cells leads to the formation of tumors closely resembling human LUAD, the same genetic alterations in bulk tracheal epithelial cells lead to formation of tumors closely resembling human LSCC (Fig. 3J).

One of the goals of our study was to identify the cell of origin for LSCC, which has not been experimentally identified, although has been proposed to be central airway basal cells/ABSCs (15). Because LSCCs were initiated from bulk tracheal epithelial cells in our models, we next tested the ability of purified basal or luminal tracheal epithelial cells to give rise to LSCCs, using a similar strategy as previously used in cell-of-origin studies for human and mouse prostate and esophageal cancers (Fig. 4A; refs. 22, 23). First, we sorted tracheal basal and luminal cells from Keap1^f/f;Trp53^f/f;R26^tdTomato mice, transduced these with Ad-Cre in vitro, and transplanted them into NSG mice. Second, we treated Krt5^CreERT2;Keap1^f/f;Trp53^f/f;R26^tdTomato mice with tamoxifen, allowed ∼4 months for luminal cell generation from recombined basal cells, and sorted 500 to 2,000 tdTomato⁺ tracheal basal and luminal cells for transplantation (Supplementary Fig. S4A). Despite similar cell viability at the time of transplantation (Supplementary Fig. S4B), only basal cells were able to generate tumors, and these displayed similar histologic features as LSCCs arising from bulk tracheal epithelial cells (Fig. 4B and C). Taken together, these data indicate that ABSCs are the likely cell of origin for LSCCs with TP53 mutation.

Because both Trp53- and Trp53;Keap1-deleted tracheal cells generated LSCCs, we next sought to examine the role of Keap1 loss in LSCC pathogenesis by comparing the two types of tumors. Compared with P-LSCC cells, K/P-LSCC cells proliferated more rapidly in vitro (Fig. 5A). Similarly, in vivo tumor growth studies revealed that K/P-LSCCs grew significantly more rapidly than P-LSCCs when equal numbers of tumor cells were implanted subcutaneously (Fig. 5B). In addition, K/P-LSCC cells displayed significantly higher migratory potential in vitro (Fig. 5C) and significantly higher numbers of lung metastases in vivo when injected into the tail vein (Fig. 5D).

We next investigated the genome-wide gene expression changes induced by Keap1 loss by FACS-sorting tumor cells and performing RNA sequencing (RNA-seq). Unsupervised hierarchical clustering of global gene expression profiles displayed concordance between paired replicates and revealed significant differences between the two types of tumors (Supplementary Fig. S5A). Using a 2-fold cutoff and corrected P-value threshold < 0.01, 298 genes were upregulated and 463 genes were downregulated in K/P-LSCCs compared with P-LSCCs. Using gene set enrichment analyses (GSEA; ref. 24), we found that K/P-LSCCs overexpressed genes upregulated in stem cells, consistent with our findings in Figs. 1 and 2, showing that Keap1 deletion enhances ABSC self-renewal (Fig. 5E). In addition, K/P-LSCCs overexpressed gene sets related to NRF2 target genes, phase II conjugation, and biological oxidation. In agreement with these gene expression changes, intracellular ROS levels were significantly lower in K/P-LSCC cells than in P-LSCC cells (Fig. 5F and G).

Previous studies have shown that both NRF2 and NF-κB signaling can serve as downstream mediators of Keap1 loss (3, 25, 26). Although NRF2 target genes were upregulated by KEAP1 loss, NF-κB target genes were not differentially enriched in either K/P-LSCC or P-LSCC cells (Fig. 5E). Consistently, the expression of selected NF-κB target genes measured by qRT-PCR was not higher in K/P-LSCC compared with P-LSCC cells (Fig. 5H), whereas the expression of NRF2 target genes was significantly increased (Fig. 5I). Furthermore, Nrf2 silencing in K/P-LSCC cells decreased cell proliferation in vitro and tumor growth in vivo to levels similar to those seen in P-LSCC cells (Fig. 5J and K and Supplementary Fig. S5B), suggesting that the increased proliferation seen in K/P-LSCCs is largely mediated via NRF2. Also, NOTCH1 target genes, which were previously shown as a downstream mediator NRF2 (16), were not differentially expressed in either of two tumor types (Fig. 5L). These data indicate that loss of Keap1 in LSCCs leads to increased cell proliferation, invasion, and expression of NRF2 target genes and suggest that NRF2 activation and subsequent ROS decrease is the main mechanistic mediator of phenotypes observed upon Keap1 loss.

To establish the functional relevance of differences in ROS levels observed between P-LSCC and K/P-LSCC cells, we examined whether inhibition of ROS could mimic the effect of Keap1 loss on LSCC proliferation. Consistent with baseline differences in ROS, treatment with the ROS inhibitor N-Acetylcysteine (NAC) increased tumorsphere formation in P-LSCC but not in K/P-LSCC cells (Fig. 6A and Supplementary Fig. S6A). Next, we asked whether treatment with NAC could mimic the prometastatic effects of Keap1 deletion. NAC treatment significantly increased the number of metastatic lung nodules formed by P-LSCC cells after tail-vein injection (Fig. 6B and Supplementary Fig. S6B and S6C). Thus, effects of Keap1 deletion on cell proliferation, clonogenicity, and metastasis can be mimicked by reduction of cellular ROS.

Lower ROS levels induced by Keap1 deletion should protect K/P-LSCC from oxidative stress. Congruent with this hypothesis, we found that K/P-LSCC cells were significantly more resistant to H₂O₂ treatment than P-LSCC cells (Fig. 6C and Supplementary Fig. S6D). Because the main mechanism of ionizing radiation (IR)–induced cell killing involves DNA damage caused by ROS induction (27), we next explored the effect of Keap1 deletion on LSCC radiosensitivity. In agreement with our findings of enhanced ability to survive ROS stress, K/P-LSCC cells were significantly more radioresistant than P-LSCC cells in tumorsphere clonogenic assays (Fig. 6D and E). This effect was not unique to our LSCC models, because Keap1/Trp53-deleted LUAD (K/P-LUAD) cells were also significantly more radioresistant than LUAD cells in which only Trp53 was deleted (P-LUAD; Supplementary Fig. S7A and S7B). Furthermore, IR-induced phosphorylated histone 2AX (γ-H2AX) nuclear foci were significantly fewer in K/P-LSCC cells than in P-LSCC cells, indicating that K/P-LSCC cells develop less DNA double-strand breaks after IR (Supplementary Fig. S7C and S7D). Additionally, NAC pretreatment significantly protected P-LSCC cells from IR while having no effect on K/P-LSCC cells, consistent with the hypothesis that radioresistance induced by Keap1 loss is largely mediated through reduction of intracellular ROS (Fig. 6F and Supplementary Fig. S7E). Similarly, in vivo growth of K/P-LSCC tumors was significantly less inhibited by IR than that of P-LSCC tumors (Fig. 6G). Also, animals bearing K/P-LSCC tumors showed significantly worse overall survival after in vivo irradiation than littermates bearing P-LSCC tumors (Fig. 6H). These data indicate that activation of the KEAP1–NRF2 pathway induces radioresistance in NSCLCs.

Finally, we sought to identify a potential strategy for overcoming Keap1-mediated radioresistance. One of the genes most highly overexpressed by K/P-LSCCs compared with P-LSCCs was Slc7a11 (>15-fold; Supplementary Fig. S8A), which encodes the light chain of the system x_c⁻ antiporter that plays a critical role in glutathione synthesis via cystine import (28). Therefore, we reasoned that inhibition of system x_c⁻ may overcome the radioresistant phenotype of K/P-LSCC tumors. Treatment with sulfasalazine, a system x_c⁻ inhibitor (29–31), in combination with IR, sensitized K/P-LSCCs and resulted in similar cell killing as in P-LSCCs (Fig. 6I and Supplementary Fig. S8B). These data suggest that targeting system x_c⁻ represents a potential strategy for overcoming Keap1-mediated radioresistance.

Based on the findings that both K/P-LSCCs and K/P-LUADs are more radioresistant to IR than P-LSCCs and P-LUADs, respectively, we hypothesized that KEAP1/NRF2 mutations lead to increased rates of local recurrence in patients with localized NSCLCs treated with RT. We therefore genotyped a cohort of 42 tumors from patients with stage I–III NSCLC treated with RT, with or without concurrent chemotherapy, and identified 9 patients with mutations in NRF2 or KEAP1 (Fig. 7A and Supplementary Table S2). All seven KEAP1 mutations are predicted to be deleterious to protein structure/function and six are located in the Kelch repeat domain that mediates interaction with NRF2 (Supplementary Table S3). Strikingly, the cumulative incidence of local failure at 30 months was 70% in patients whose tumors carried mutations in KEAP1/NRF2 and 18% in patients with WT tumors (P < 0.008; Fig. 7B). Additionally, out of 12 total patients who developed local recurrence, 6 had tumors with KEAP1/NRF2 mutations, implicating the pathway in half of RT failures in our cohort. Restricting analysis to patients with stage II–III NSCLC also revealed significantly higher rates of local failure in patients with KEAP1 or NRF2 mutations (P < 0.04; Fig. 7C). Of note, mutations in TP53, which were found in 19 patients, were not associated with risk of local recurrence (Supplementary Fig. S9).

Because patients with NSCLC receiving radiotherapy often undergo only limited tumor tissue sampling via needle biopsies and can therefore have insufficient tissue for mutation testing (32), it would be advantageous to be able to evaluate KEAP1/NRF2 mutation status noninvasively. To facilitate such analyses, our group developed an ultrasensitive approach for detection of mutant ctDNA called CAPP-Seq (33), which we have recently optimized for noninvasive mutation detection (34). We were able to obtain pretreatment plasma for 7 of the patients in our initial cohort and found that CAPP-Seq analysis correctly identified KEAP1 mutation status in all of them (Fig. 7D). Furthermore, in 1 patient for whom we analyzed serial plasma samples, the KEAP1 mutation as well as other co-occurring mutations tracked with treatment response (Fig. 7E).

Finally, in order to test the association of KEAP1 mutation status with local recurrence after RT in an independent cohort, we analyzed 20 patients for whom we did not have access to tumor tissue. We applied CAPP-Seq to noninvasively genotype each patient's tumor using the pre-RT plasma sample and identified 5 patients with KEAP1 mutations (Supplementary Table S4). As in our initial cohort, the cumulative incidence of local recurrence was significantly higher in patients with KEAP1 mutations (P = 0.02; Fig. 7F). Taken together, these results suggest that KEAP1/NRF2 mutation status is a predictor of local recurrence after RT in patients with NSCLC.

In this study, we demonstrate that deletion of Keap1 or Trp53 in murine ABSCs promotes their self-renewal, leading to expansion of mutant stem cell clones. Furthermore, we show that deletion of Trp53 with or without Keap1 leads to LSCC formation and that ABSCs are the cell of origin in these LSCC models. The effects of Keap1 loss appear to be predominantly mediated by NRF2 activation and subsequent lowering of intracellular ROS, leading to increased tumor proliferation, metastasis, and resistance to oxidative stresses and irradiation. Finally, we show for the first time that KEAP1/NRF2 mutations strongly predict clinical resistance to RT in patients with NSCLC. Although several prior studies have suggested a role for KEAP1 and NRF2 in radioresistance, none have examined the association of mutations in these genes with radiation resistance. Instead, the prior studies either did not contain clinical data and focused on preclinical experiments (35–37), analyzed other clinical outcomes and not local control (38–40), or analyzed gene expression signatures (41). Because mutation status can be readily established in clinical laboratories using widely available genotyping methods, our findings have potential clinical implications.

In order to study the role of Trp53 and Keap1 in ABSC self-renewal in vivo, we developed a novel lineage-tracing method specific for ABSCs. Although intranasal Ad-Cre is used routinely to knock out genes in the lung, we found that it is unable to transduce upper airway epithelial cells in homeostatic condition, likely due to the mucociliary barrier. Importantly, these data indicate that previous Ad-Cre–induced lung cancer models have likely not explored central airway basal cells as a potential cell of origin. By combining SO₂ injury (42) and Ad-Cre intranasal instillation, we could efficiently transduce tracheal basal cells of immunocompetent mice and lineage-trace them during regeneration. Our lineage-tracing method is complementary to conventional Krt5-driven lineage tracing in airway stem cells. One advantage of our approach is that unlike the former methods, it does not lead to recombination in other Krt5-expressing tissues such as the skin or esophagus.

We established LSCC models based on loss of Trp53 and Keap1, which mimic mutations occurring in >80% and >30% of human LSCCs, respectively, and which have not been previously explored in murine LSCC models. We found that the type of lung cancer formed by deletion of Trp53 is dependent on the normal cell type that is targeted. Specifically, inactivation of Trp53 by Ad-Cre inhalation, which leads to recombination in peripheral lung cells but not in the central airway, led to SFTPC⁺TTF1⁺KRT5⁻ LUAD formation. This finding is consistent with that of previous studies (18, 43, 44), which found that inactivation of Trp53 induced by Ad-Cre inhalation led only to LUAD formation. However, when we specifically induced deletion of Trp53 with or without Keap1 in tracheal epithelial cells, LSCCs were generated.

Our experimental approach allowed us to explore the cell of origin of LSCC using a similar strategy as has been successfully applied in prostate and esophageal cancers (22, 23). However, one drawback of this approach is that we could not induce LSCCs orthotopically in immunocompetent mice due to formation of LUADs when using Ad-Cre or skin tumors when using mice with Krt5^CreERT2, both with a latency of ∼2 months. This suggests that LSCCs may have a relatively long latency in mice. In support of this notion, time to tumor formation in previously published LSCC models has been reported to be ∼8 to 10 months (2, 19, 20). Additionally, our use of immunodeficient mice makes our approach suboptimal for studies on tumor immunology and immunotherapy. However, we believe that our model, using either subcutaneous or orthotopic transplantation, will be useful for studies focused on other aspects of cancer biology, including tumor initiation, metastasis, and treatment resistance.

We found that only tracheal ABSCs and not more differentiated tracheal luminal cells, which contain secretory and ciliated cells, could form LSCCs, indicating that ABSCs are the cell of origin in our model. Interestingly, a recent study showed that Scgb1a1⁺ secretory cells can dedifferentiate into ABSCs (45), raising the question of why luminal cells could not give rise to tumors in our model. One potential explanation for this is that the process of dedifferentiation may not function properly in the presence of the gene expression changes induced by loss of Trp53 with or without loss of Keap1. Another possible explanation is that, although basal cells arising from secretory cells are able to self-renew and repair tissue upon injury, they may not be entirely identical to basal cells at homeostasis and may differ in ability to undergo tumorigenesis. Lastly, considering the importance of stem cell and stromal interaction, the transplanted microenvironment may not favor dedifferentiation.

Nonetheless, our results are consistent with findings from other LSCC models. For example, in a recent study (3), Xiao and colleagues provided data indicating that IKKα inactivation in KRT5-expressing lung cells, but not other types of lung cells, leads to LSCC formation, consistent with our findings of ABSCs as the cell of origin in our LSCC models. Separately, in an Lkb1;Pten deletion–based LSCC model, SPC-Cre and CCSP-Cre failed to produce tumors, and endotracheal Ad-Cre administration led to only LSCC formation (2). This indicates that alveolar type II pneumocytes and club cells are not the cell of origin in this model and suggests that another cell type, possibly rare KRT5-expressing peribronchiolar cells (46), is the source of these tumors.

However, other data suggest that murine LSCCs do not originate only from ABSCs. Rather, a complex interplay between the cell of origin and the driver mutation appears to determine the type of lung cancer that develops. For example, recent studies revealed that Kras activated/Lkb1-deficient LUADs, which originated from type II pneumocytes, can transdifferentiate into LSCC (47–49). Also, CCSP-Cre induced SOX2 overexpression, and Kras activation can lead to squamous hyperplasia (50). Additionally, some data suggest that KRT5-expressing cells may be able to give rise to LUADs. Specifically, Krt5^CrePR-mediated Kras activation or Pten deletion resulted in mixed LSCC and LUAD tumors, although as the authors point out, this may have been due to leakiness of the promoter that was used (51). Thus, it appears that the cell of origin may dictate the lung cancer subtype that forms for certain mutations such as Trp53 but not for others such as Kras and Sox2. Interestingly, the cell of origin appears to determine lung cancer subtype when the driver gene is commonly mutated in both LSCC and LUAD (e.g., Trp53). Conversely, driver mutations that are predominantly observed in human LSCC, such as Sox2 and Lkb1, appear to be able to induce LSCC even in bronchiolar club cells or alveolar type II cells, whereas mutations that are predominantly observed in human LUADs, such as Kras, appear to be able to drive LUAD formation in murine ABSCs. Thus, at least in the mouse, lung cancer subtypes appear to be determined by the interplay of cell type and genotype.

Our results suggest a critical role for mutations in TP53 and KEAP1 during LSCC oncogenesis, because deletion of either gene leads to increased self-renewal of ABSCs. Significantly, the impact of Trp53 loss on ABSC self-renewal and the fact that loss of both Trp53 and Keap1 leads to increased self-renewal compared with loss of either gene alone have not been previously demonstrated and are of clinical relevance, because TP53 is mutated in over 80% of human LSCCs harboring KEAP1 mutations (5). Congruent with our findings, Paul and colleagues recently also reported a role for ROS regulation in ABSC homeostasis that was mediated through increased Notch pathway signaling and reflected by gene expression changes of Notch pathway components upon NRF2 modulation (16). We did not observe any differences in expression of the same genes by Keap1 deletion, suggesting that this Notch-based mechanism does not play a major role in mediating the effects of NRF2 activation in our LSCC models.

Our findings suggest that once an ABSC acquires mutations in either TP53 or KEAP1, it will be able to outcompete WT stem cells, thus creating an expanding pool of mutated cells that are at risk for acquiring additional genetic hits and ultimately leading to LSCC formation. Indeed, evidence consistent with such clonal expansion of mutant ABSCs has been demonstrated in humans. Specifically, it has been shown that preinvasive lesions consisting of keratin 14–expressing basal cells that contain TP53 mutations can expand through large portions of the bronchial tree, including spreading from one side to the other (52, 53). Although the mutation status of KEAP1 has not been examined in these studies, we speculate that a premalignant clone containing a TP53 mutation would gain significant additional fitness if it acquired mutations in KEAP1, mediated by the increased proliferation, migration, and resistance to oxidative stress that we observed in our model.

From a clinical standpoint, our finding of most immediate potential impact is the observation that KEAP1/NRF2 mutation status predicts the rate of local failure after radiotherapy in patients with NSCLC. Because radiotherapy kills cells by induction of ROS that lead to DNA double-strand breaks, KEAP1/NRF2-mutant tumors are likely resistant to RT due to enhanced expression of ROS scavengers and detoxification pathways. However, we cannot rule out that additional somatic or epigenetic alterations that we did not interrogate may play a role in the higher incidence of local recurrence in KEAP1/NRF2-mutant tumors. Thus, future studies should include human models with more complex genetics and examine the effects of KEAP1 loss in TP53 WT tumors in order to further explore the role of KEAP1/NRF2 in radiation resistance.

KEAP1/NRF2 mutation status was not associated with overall survival in our cohorts or in the LUAD (54) and LSCC (5) cohorts from TCGA (data not shown). However, Kim and colleagues previously found that NRF2 mutations were statistically significantly associated with worse overall survival in LSCC (55). We hypothesize that these conflicting results likely relate to statistical power, differences in clinical covariates among the different cohorts, and the potential impact of additional somatic or epigenetic modifiers of tumor aggressiveness whose prevalence may differ between studies.

All of the KEAP1 mutations we identified in human patients were somatic and appeared to be heterozygous. This observation is consistent with prior studies that have shown that mutation of KEAP1 on both alleles is not required to achieve functional effects. For example, mutant KEAP1 proteins can form heterodimers with WT proteins and function as dominant-negatives, inhibiting the association with NRF2 (56, 57). Separately, epigenetic modifications of KEAP1 promoter associated with loss of expression have been reported in a variety of malignancies, including lung cancer (58–60).

Our observation of KEAP1/NRF2 mutations in the plasma of patients with NSCLC being treated with RT suggests that noninvasive mutation assessment using ctDNA could be a valuable approach for facilitating personalized management of patients with NSCLC who have limited tissue samples available for molecular analysis. Given the high rate of local failure we observed in patients with KEAP1/NRF2 mutations, mutation testing for these genes in tumor or plasma could potentially be used to personalize treatment strategies for patients with NSCLC. Although personalized medicine approaches based on mutation testing are routinely used to identify the most appropriate systemic therapy in advanced NSCLC, similar approaches are not being used to select treatments for patients with localized NSCLC. We speculate that KEAP1/NRF2 mutation status could potentially be used to select the best choice of local therapy for such patients or to identify patients who might benefit from radiation dose escalation. Furthermore, inhibition of critical NRF2 targets, such as system x_c⁻, may allow targeted radiosensitization in patients with KEAP1/NRF2 mutations. Prospective studies to validate the association of KEAP1/NRF2 mutations with radioresistance and to test personalized treatment approaches will, of course, be critical. Lastly, because KEAP1/NRF2 mutations are found in many other cancer types (61–65), we hypothesize that they may contribute to radioresistance in a substantial fraction of patients. Therefore, strategies for countering NRF2 activation may improve personalized therapy and outcomes for a large number of patients with cancer.

Mouse tracheae were harvested and incubated in dispase (Invitrogen; #17105-041) for 40 minutes at 37°C and longitudinally opened. The epithelium was peeled off and incubated in 0.25% trypsin for 5 minutes at 37°C. Trypsinized epithelium was centrifuged, treated with ammonium–chloride–potassium (ACK) lysis buffer to lyse the red blood cells, and filtered through a 40-μm cell strainer (BD Biosciences). After centrifugation, cells were resuspended in blocking buffer (Hank's Balanced Salt Solution with 2% bovine calf serum) and stained with anti-mouse CD31, CD45, CD140a, CD49f, and CD326 (Biolegend; antibody lists are in Supplementary Table S4). Cells were sorted using the BD FACS Aria.

Primary tracheal epithelial cells were resuspended in MTEC/Plus (66) mixed at a 1:1 ratio with growth factor—reduced Matrigel (67). Cell/media/matrigel mixture (100 μL) was plated on top of a 24-well cell culture insert for air–liquid interface culture. Media (0.4 mL) were provided to the lower chamber and changed every other day. Sphere formation and growth were followed for at least 10 to 14 days. Spheres (>50 μm in diameter) were counted manually. For the secondary sphere formation assay, spheres were dissociated with dispase for 40 minutes, digested with trypsin/0.05% EDTA for 5 minutes, and passaged through a 27-G needle five times to dissociate into single cells (68). After centrifugation, cells were resuspended and filtered through a 40-μm cell strainer for further analysis or culture. The same number of tdTomoto⁺ cells was FACS-sorted and cultured for the secondary tracheosphere culture.

Keap1^+/f mice (C57BL/6J background) were a kind gift from T. Kensler (University of Pittsburgh, PA; refs. 69, 70). Trp53^+/f; R26^tdTomato mice (B6/129 background) were a kind gift from M. Winslow (Stanford University, CA; ref. 71). Mouse cohorts of WT R26^tdTomato, Keap1^f/f;R26^tdTomato, Trp53^f/f;R26^tdTomato, and Keap1^f/f; Trp53^f/f;R26^tdTomato were generated by mating Keap1^+/f mice and Trp53^+/f;R26^tdTomato mice. NSG mice, Krt5^CreERT2 mice, and nude mice were obtained from The Jackson Laboratory and Charles River Laboratories. All mice were genotyped by previously reported methods (69, 71, 72), and mice between 4 weeks and 9 months of age were used for experiments. The LSCC was established by implanting in vitro or in vivo recombined tracheal cells or tumor cells into NSG mice. Mice were housed in a designated pathogen-free area in a facility at the Stanford University School of Medicine accredited by the Association for the Assessment and Accreditation of Laboratory Animal Care. All care and treatment of experimental animals were in accordance with Stanford University School of Medicine institutional animal care and use committee guidelines. For SO₂ injury, adult mice were exposed to 500 ppm SO₂ in air for 5 hours (73). After 18 to 24 hours, Ad-Cre viruses were intranasally inhaled into mouse lung. For in vivo recombination for the cell-of-origin experiments, 250 mg/g tamoxifen was intraperitoneally injected into Krt5^CreERT2;Keap1^f/f;Trp53^f/f;R26^tdTomato mice. After ∼4 months, recombined basal and luminal cells were sorted and subcutaneously transplanted into NSG mice for tumor generation.

For de novo murine tumor generation, 10k to 30k tracheal epithelial cells from WT R26^tdTomato, Keap1^f/f;R26^tdTomato, Trp53^f/f;R26^tdTomato, and Keap1^f/f; Trp53^f/f;R26^tdTomato mice were harvested and transduced by Ad-Cre viruses in vitro overnight at 37°C. The next day, cells were resuspended in 200 μL of a 1:1 mixture of the culture medium and Matrigel and subcutaneously transplanted into NSG mice. For tumor growth assays, primary lung tumor cells were suspended in a 1:1 mixture of the culture medium and Matrigel and were subcutaneously injected into nude mice. Tumor sizes were measured every 2 to 3 days after tumor inoculation and continued throughout the experiment. Tumor volumes were calculated using the formula (length × width²)/2. To analyze the effect of irradiation on tumor growth, when tumor volumes reached ∼100 mm³, mice were randomly divided into two groups, and irradiation was focally delivered. Tumor volumes were measured every other day. Mice displaying severe radiation toxicity (i.e., dramatic weight loss within ∼3 days after irradiation attributed to inadequate shielding) were excluded from the analysis.

Tumors generated from Trp53- or Keap1;Trp53-null tracheal cells were minced with a razor blade and suspended in 10 mL of L-15 Leibovitz medium (Thermo Fisher Scientific Inc.) supplemented with 0.5 mL of collagenase/hyaluronidase (Stem Cell Technologies). Tumors were digested for 1.5 to 2 hours at 37°C and 5% CO₂ with manual dissociation by pipetting every 30 minutes. Once digested, 40 mL of blocking buffer was added and tumor cells were collected by centrifugation. Tumor cells were resuspended in 5 mL of trypsin/0.05% EDTA for 5 minutes and centrifuged with the addition of blocking buffer. The cell pellet was incubated with 100 Kunitz units of DNase I (Sigma) and Dispase (Stem Cell Technologies) for 5 minutes at 37°C and centrifuged again with the addition of blocking buffer. Once digested, tumor cells were treated with ACK lysis buffer and filtered through a 40-μm cell strainer. After centrifugation, tumor cells were resuspended in blocking buffer, blocked with rat IgG for 10 minutes, and stained with rat anti-mouse CD31, CD45, and SCA1, and rat anti-human/mouse CD49f. CD31- and CD45-negative viable cells with strong red signal were sorted for further analysis.

Tumor cells (2,000–5,000) were injected into NSG mice via the tail vein after preincubation with PBS or N-acetylcysteine (NAC) for 2 hours. The detailed information is described in Supplementary Methods.

Tumor cell invasion assays were performed according to the manufacturer's instruction (Trevigen; see Supplementary Methods).

Tissues or spheres grown in Matrigel were embedded in OCT compound (Sakura) and immediately frozen on dry ice for frozen section or fixed in 4% paraformaldehyde overnight, transferred to 70% ethanol, and embedded in paraffin. The detailed information is described in Supplementary Methods.

Thirty minutes after irradiation, irradiated and nonirradiated cells were fixed in ice-cold 50% ethanol and 50% methanol for 20 minutes for γ-H2AX foci evaluation. The detailed information is described in Supplementary Methods.

We analyzed tumor specimens from patients undergoing definitive treatment for newly diagnosed NSCLC who were enrolled in a study approved by the Stanford University Institutional Review Board and who had provided informed consent in accordance with the Declaration of Helsinki. All patients analyzed had biopsy-confirmed primary NSCLC, stage IA–IIIB according to the seventh edition American Joint Committee on Cancer staging manual. Patients who underwent surgery were excluded. Patients with stage I NSCLC were treated with stereotactic ablative radiotherapy (SABR), whereas patients with stage II–III NSCLC were treated with conventionally fractionated radiotherapy or concurrent chemoradiotherapy. Local failure was defined as previously described (74). For the initial cohort, tissue blocks were obtained and clinical outcomes were collected from patient records. A subset of these patients also donated blood samples for biomarker analysis. For the independent validation cohort, plasma samples were collected prior to treatment.

Tumor genotyping was performed using a hybrid capture-based approach on formalin-fixed, paraffin-embedded tumor samples to determine KEAP1, NRF2, and TP53 mutation status, as previously described (33). Matched germline was also sequenced when available. Briefly, single-nucleotide variants (SNV) were called using an in-house pipeline for processing targeted sequencing data that used BWA sampe 0.5.9-r16 and SAMtools mpileup 0.1.18-r579. The GRCh37/hg19 reference genome was used. Only properly paired reads and bases with a phred quality score of at least 30 were considered, and base alignment quality adjustment (BAQ) was enabled for calling SNVs. When matched germline was available, variants at ≥2.5% were called, and when only tumor DNA was available, variants at ≥5% were called.

For ctDNA analyses, CAPP-Seq was performed on cell-free DNA isolated from plasma samples as recently described (34). Briefly, cell-free DNA was isolated, and sequencing libraries were prepared using adapters containing molecular barcodes. To further reduce sequencing errors, data were background-polished. Noninvasive genotyping was performed to call nonsynonymous SNVs and indels in KEAP1, NRF2, and TP53 that were present in pretreatment plasma but not matched germline DNA obtained from plasma-depleted whole blood. For one patient, plasma samples were also available immediately posttreatment and at the time of local recurrence. The global amount of ctDNA in each sample was calculated using SNVs and indel reporters detected in the tumor biopsy.

The detailed information from RNA isolation of FACS-sorted cells to RNA-seq analysis is described in Supplementary Methods.

shNrf2 expression plasmids were purchased from OriGene (TL515053). Lentiviral generation and transduction of primary lung cancer cells were described in Supplementary Methods.

RNA from primary sorted cells was isolated using the RNeasy Micro Kit (Qiagen), and cDNA was prepared using the High-Capacity cDNA Reverse Transcription Kit (ABI). The detailed information is described in Supplementary Methods.

Statistical analyses were performed using unpaired two-sided Student t test between groups at each time point after checking that variances were similar between groups, and significance was defined based on P < 0.05. Data are presented as mean ± SEM of 3 or more independent biological replicates. Competing risk analyses were performed in the R statistical program, and Gray's tests were used to compare strata. Sample numbers were determined empirically using estimation of sample size considering the variation and mean of the samples. For animal experiments, this ranged from 4 to 10 animals per group. No statistical method was used to predetermine sample size. Animals were randomly assigned to groups for in vivo studies. Assessment was performed without blinding.

A. Lovejoy is a senior scientist at Roche. A.M. Newman has ownership interest (including patents) in Capp Medical, Inc., and is a consultant/advisory board member for Roche. B.W. Loo is a board member at TibaRay, Inc., reports receiving commercial research grants from Varian Medical Systems and RaySearch Laboratories, and has ownership interest (including patents) in TibaRay, Inc. M. Diehn is a consultant/advisory board member for Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Jeong, A.A. Alizadeh, M. Diehn

Development of methodology: Y. Jeong, W. Kong, R.B. West, P. Beachy, A.A. Alizadeh, M. Diehn

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Jeong, N.T. Hoang, A. Lovejoy, D. Truong, S. Martin, D. Heiser, L. Zhou, C. Say, J.N. Carter, S.M. Hiniker, B.W. Loo Jr, R.B. West, A.A. Alizadeh, M. Diehn

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Jeong, N.T. Hoang, A. Lovejoy, H. Stehr, A.M. Newman, A.J. Gentles, A. Chaudhuri, S.M. Hiniker, A.A. Alizadeh, M. Diehn

Writing, review, and/or revision of the manuscript: Y. Jeong, N.T. Hoang, A. Lovejoy, H. Stehr, A.J. Gentles, S.M. Hiniker, B.W. Loo Jr, R.B. West, P. Beachy, A.A. Alizadeh, M. Diehn

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Jeong, H. Stehr, A. Chaudhuri, L. Zhou, J.N. Carter, S.M. Hiniker, M. Diehn

Study supervision: Y. Jeong, A.A. Alizadeh, M. Diehn

Other (SO2 injury): W. Kong

Other (provided injury system): P. Beachy

We thank J. Sage, A. Sweet-Cordero, M. Winslow (Stanford University), T. Kensler (University of Pittsburgh), and members of their labs for supplying reagents and suggestions. We also thank R. von Eyben for statistical advice.

This work was supported by grants from the California Institute for Regenerative Medicine (TG2-01159, Y. Jeong; TB1-01194, N.T. Hoang), the NIH (P30CA147933 and P01CA139490, M. Diehn; R01CA188298, A.A. Alizadeh and M. Diehn), the V Foundation (V Scholar Award, M. Diehn), the CRK Faculty Scholar Fund (M. Diehn), and the Virginia and D.K. Ludwig Foundation (M. Diehn). M. Diehn was also supported by a Doris Duke Clinical Scientist Development Award and an NIH New Innovator Award (1-DP2-CA186569).

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.