NFE2L2 mutations enhance radioresistance in head and neck cancer by modulating intratumoral myeloid cells

Abstract

Radiotherapy is one of the primary treatments of head and neck squamous cell carcinoma (HNSCC), which has a high risk of locoregional failure (LRF). Presently, there is no reliable predictive biomarker of radioresistance in HNSCC. Here, we found that mutations in NFE2L2, which encodes Nrf2, are associated with a significantly higher rate of LRF in patients with oral cavity cancer treated with surgery and adjuvant (chemo)radiotherapy but not in those treated with surgery alone. Somatic mutation of NFE2L2 led to Nrf2 activation and radioresistance in HNSCC cells. Tumors harboring mutant Nrf2^E79Q were substantially more radioresistant than tumors with wild-type Nrf2 in immunocompetent mice, while the difference was diminished in immunocompromised mice. Nrf2^E79Q enhanced radioresistance through increased recruitment of intratumoral polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs) and reduction of M1-polarized macrophages. Treatment with the glutaminase inhibitor CB-839 overcame the radioresistance induced by Nrf2^E79Q or Nrf2^E79K. Radiotherapy increased expression of PMN-MDSC-attracting chemokines, including CXCL1, CXLC3 and CSF3, in Nrf2^E79Q-expressing tumors via the TLR4, which could be reversed by CB-839. This study provides insights into the impact of NFE2L2 mutations on radioresistance and suggests that CB-839 can increase radiosensitivity by switching intratumoral myeloid cells to an anti-tumor phenotype, supporting clinical testing of CB-839 with radiation in HNSCC with NFE2L2 mutations.

Introduction

Head and neck squamous cell carcinoma (HNSCC), which accounts for 90% of all head and neck malignancies, are mucosal tumors arising in the upper aerodigestive tract, which includes the oral cavity (OC), larynx (LX), oropharynx (OP), and hypopharynx (HP)(1, 2). Radiation therapy (RT), which induces DNA damage and causes accumulation of endogenous reactive oxygen species (ROS), is one of the primary local therapies for HNSCC. It is estimated that 75% of patients with HNSCC will benefit from RT as part of their definitive treatment or as adjuvant treatment after surgery(3). Despite aggressive surgery and RT, ~20–30% of these patients continue to develop locoregional failure (LRF) and 40–50% will die from the cancer(4). Thus, identify a biomarker of RT resistant is important in order to distinguish a patient group who would need either more intensified adjuvant treatment (chemoradiation) or novel treatment to mitigate recurrence.

Recent advances in cancer genomics have led to an understanding of mutational and gene expression profiles in HNSCC that might shed light on RT resistance in these cancers. The Cancer Genome Atlas (TCGA) project identified mutations in the KEAP1/NFE2L2 pathway in ~6–15% of HNSCC, depending on the primary tumor location, with ~6% in the OC(5). An increasing body of evidence indicates that the somatic gene mutations of KEAP1 and NFE2L2 are frequently mutated in many types cancers(6, 7). These mutations phenotypically converge at the constitutive activation of Nrf2(8–11). Since Nrf2 is a principal regulator of cytoprotective responses to endogenous and exogenous stresses caused by ROS and electrophilic stresses, the Keap1/Nrf2 pathway is a promising potential therapeutic target to combat RT resistance. As such, clarifying the relationship between constitutive NFE2L2 mutations and radiation resistance has important prognostic and therapeutic implication.

In this study, we sought to identify biomarker(s) of RT resistance in oral cavity (OC) cancer patients, a large subset of HNSCC, and underly the mechanisms driving RT resistance. We found that NFE2L2 mutations were significantly associated with local regional failure (LRF) only in the patients treated with surgery followed by adjuvant (chemo)radiotherapy (SRT) but not in patients treated with surgery (S) alone, suggesting that it is a potential predictive marker of radioresistance. Using a combination in vitro and in vivo studies in immune compromised and immune competent hosts, we found that Nrf2^E79Q and Nrf2^E79K mutations led to radiation resistance through enhanced recruitment of polymorphonuclear myeloid-derive suppressor cells (PMN-MDSCs) to the tumor environment in immune competent mice via upregulation of MDSC homing chemokines (CXCL1, CXLC3, and CSF3). In addition, we showed that treatment with a glutaminase inhibitor, CB-839, overcame the Nrf2^Mut-induced RT resistance through a reduction of PMN-MDSCs recruitment associated with chemokine reduction. Collectively, our results suggest that NFE2L2 mutation is a potential predictive marker of RT resistance in OC tumors, and that it mediates RT resistance in part by modulating myeloid cells in the tumor environment that can be overcome by CB-839. Our data also provide mechanistic insights into how CB-839 can overcome NFE2L2 mutation-mediated radiation resistance.

Material and Methods

Patient cohorts.

The oral cavity cancer (OC) cohort comprises of 135 patients: 75 patients received Surgery + (chemo)RT (SRT) and 60 patients received Surgery (S) alone at Stanford Medical Center. Patient characteristics and treatment details for both cohorts are provided in Supplemental Table 1.

DNA extraction from FFPE blocks.

Tumor or matched normal tissue from the FFPE block were marked by a head and neck pathologist based on the H&E staining of the matched slides. Tissues were then punched from patient FFPE tissue blocks using Integra^™ Miltex^™ Standard Biopsy Punches (Catalog#3331AA, Fisher Scientific). Genomic DNA from FFPE tissues were obtained using the AllPrep DNA/RNA FFPE kit (Catalog# 80234, Qiagen). In brief, tumor rich tissues were dissolved in the Xylene to remove the paraffin, residual xylene was removed by washing with ethanol. Next, proteinase K was applied to digest the samples at 56°C for15 minutes to loosen the tissue pellet. The samples were cooled down on ice and then centrifuged to obtain RNA-containing supernatant and DNA-containing tissue pellet. The DNA-containing tissue pellet was next lysed with proteinase K in Buffer ATL at 56°C for 12 hours followed by incubating at 90°C for 2 hours without agitation. Subsequently, the samples were cool down at room temperature and mixed with lysis buffer (Buffer AL) and then immediately applied to a QIAamp MinElute spin column followed by washing steps with the buffer supplied. Pure DNA was eluted in 50μl of DNase free distilled water.

DNA sequencing and analysis.

For the SRT group, 68 tumor samples were successfully sequenced by targeted next generation sequencing (NGS), whereas 7 samples were subjected to targeted NFE2L2 Sanger sequencing to assess for NFE2L2 mutation status at the E79 position after failing NGS. For the Surgery alone group, all samples were subjected to targeted NFE2L2 Sanger sequencing to assess for the mutation status at the E79 position.

For targeted NGS, libraries were generated using the Kapa LTP Library Preparation Kit for Illumina Platforms (Catalog#K8232, Roche). Targeted NGS DNA sequencing was performed using a hybrid capture-based approach and a previously prepared targeted library as published(12). Matched normal tissues from uninvolved lymph node, benign mucosa, or muscle in the resection specimens were sequenced at the same time. Sequencing result was analyzed as described in previously(12).

Cell culture.

Human HNSCC SAS and FaDu cells were obtained from the American Type Culture Collection (ATCC) and cultured in DMEM containing 10% fetal calf serum (FCS). MOC1 cells were generous provided by Dr. Ravindra Uppaluri(13) and cultured in medium with 5% FCS, 40ng/ml hydrocortisone (Catalog# H0135, Sigma-Aldrich), 5 ng/ml EGF (Catalog#01–107, Sigma-Aldrich), 5ug/ml insulin (Catalog#I0516, Sigma-Aldrich), and 1% penicillin-streptomycin. Cells were maintained at 37°C in a saturated humidity atmosphere containing 95% air and 5% CO2.

Stable NFE2L2-knockout SAS, FaDu and MOC1 cells were generated with CRISPR/Cas9 technology using pSpCas9–2A-Puro (PX459) from Addgene. The sgRNA of human/murine NFE2L2 were designed using web-based biology tool (chopchop: https://chopchop.cbu.uib.no/). Reconstruction of Nrf2^WT or Nrf2^E79Q or Nrf2^E79K cells was conducted as the following. NFE2L2 knockout SAS, FaDu or MOC1 cells were infected with same amount of Nrf2^WT or Nrf2^E79Q or Nrf2^E79K lentiviral particles and then selected with 500 – 600 ug/ml G418 to make stable cell lines. All cells were tested negative for the presence mycoplasma in November 2022. Gene mutagenesis was conducted using Phusion site-directed mutagenesis kit (Catalog#F541, ThermoFisher). The mutagenesis primers were listed in Supplemental Table 2.

RNA sequencing and analysis.

Nrf2^WT or Nrf2^E79Q SAS cells were pretreated with 0.1μM CB-839 or vehicle for 24 hours, then irradiated (4 Gy). RT was performed as previously described(14) using a 225kVp cabinet x-ray irradiator filtered with 0.5 mm Cu (IC-250, Kimtron Inc.). Thus 4 groups were created, including Nrf2^WT+RT, Nrf2^E79Q+RT, Nrf2^WT+RT+CB-839, Nrf2^E79Q+RT+CB-839. All the experiments were performed in triplicate, generating a total of 12 samples. Next, RNA was isolated using Trizol (Catalog# 15596026, Invitrogen) 24 hours after RT. Library preparation and RNA sequencing were performed by Novogene via Illumina platforms based on mechanism of SBS (Sequencing by synthesis). Raw FASTQ files were analyzed by Novogene using a combination of programs. We performed STAR(15) (v2.5) mapping reads to the reference human genome GRCh38. The method of Maximal Mappable Prefix (MMP) was used to generate a precise mapping result for junction reads. HTSeq(16) v0.6.1 was used to count the read numbers mapped of each gene. Next, the differential expressed genes were determined through edgeR(17) (v3.36) in R v4.1 platform at significant cutoff FDR (False Discovery Rate) <0.05 and log2 FC (Fold Change) > 1.5. KEGG pathway enrichments were implemented by Enrichr(18) and cutoff FDR < 0.05 was used to identify significant pathways. The R package ComplexHeatmap(19) was used to draw all heatmap figures in this study.

Western blot.

Western blot was performed using standard methods as described(20).The following antibodies were used: Nrf2 (Catalog #ab62352, Abcam), NQO1 (Catalog #A180, Cell Signaling Technology), pS6 Ribosomal Protein (S240/244) (Catalog #2215S, Cell Signaling Technology ), S6 Ribosomal Protein (54D2) (Catalog #2317, CST), TLR4 (Catalog #ab13556, Abcam), β-Actin (Catalog #47778, Santa Cruz).

qRT-PCR analysis.

Total RNA was isolated with Trizol reagent (Catalog #15596026, Invitrogen) according to the manufacturer’s instructions. Reverse transcription was performed using High-Capacity cDNA Reverse Transcription Kit (Catalog #4368813, Applied Biosystems). Real-time PCR reactions were performed using Applied Biosystems^™ PowerUp^™ SYBR^™ Green Master Mix (Catalog #A25742, Applied Biosystems). Relative quantification was achieved by normalization to the amount of β-Actin. Primers used for real-time PCR are listed in Supplemental Table 3.

Cell transfection and luciferase activity assay.

SAS or FaDu cells stably expressing Nrf2^WT or Nrf2^Mut were seeded in 24-well plates. Cells were transfected with the ARE constructs purchased from BPS bioscience (Catalog #60514, BPS bioscience). The transfection was performed using Lipofectamine 3000 transfection reagent (Catalog #L3000015, ThermoFisher). After 48 hours of transfection, cells were washed with PBS and lysed in 200 μl harvest buffer. The luciferase reporter gene assay was performed and analyzed using ARE reporter kit (Catalog #60514, BPS bioscience) according to the manufacturer’s instructions. The experiment was carried out in triplicate and expressed as the mean ± the standard deviation (SD). For transfection of TLR siRNA in MOC1 cells: Nrf2^WT or Nrf2^Mut MOC1 cells were seeded in 6-well plates. When the cells reach 80% confluency, cells were transfected with 75 pmol of TLR4 siRNA or Scramble using Lipofectamine 3000 transfection reagent. After 48 hours, cells were collected for analysis. Silencer Pre-designed mouse TLR4 siRNAs were purchased from Thermo (siRNA ID: 188775). The sequences were listed below. Sense (5’-3’): GCUUGAAUCCCUGCAUAGAtt. Anti-sense (5’-3’): UCUAUGCAGGGAUUCAAGCtt.

Determination of reactive oxygen species (ROS).

SAS or FaDu cells stably expressing Nrf2^WT or Nrf2^E79Q were seeded in 6-well plates for 24 hours then irradiated (4 Gy). Six hours after radiation, cells were incubated with 10μM 2’, 7’- dichlorofluorescein diacetate (DCFH-DA) (Catalog #C6827, Invitrogen) at 37°C for 30 min. DCFH-DA was washed away after incubation, and cells were immediately measured by flow cytometry using BD LSRFortessa^™ X-20 cell analyzer.

Clonogenic survival assay.

SAS, FaDu cells were exposed to a range of radiation doses (0, 1, 2, 4 Gy) and plated in triplicates at different densities ranging from 300 to 3,000 cells per 6 cm dish. Cells plating efficiency were determined by harvesting untreated cells. After 7 to 10 days, the cells were fixed and stained with a solution containing PBS with 0.05% crystal violet, 1% methanol, and 1% of 37% formaldehyde. Surviving fraction were determined by counting the number of colonies with more than 50 cells. Surviving fraction were normalized by the plating efficiency.

Tumor xenograft experiment.

Six- to eight-week-old female C57BL/6 mice or NRG (NOD-Rag1^null IL2rg^null, NOD rag gamma) mice were purchased from the Jackson Laboratory. 5×10⁶ MOC1 cells that stably expressed Nrf2^WT or Nrf2^E79Q or Nrf2^E79K were subcutaneously injected into the dorsal flanks of the C57BL/6 mice. 10⁶ MOC1 cells were injected into the dorsal flanks of the NRG mice. When the tumor volume reached ~100–150 mm³, the mice were randomized to either irradiation or observation. Thus, 4 murine groups were created: Nrf2^WT, Nrf2^E79Q, Nrf2^WT+RT, Nrf2^E79Q+RT. Similar murine groups were used for Nrf2^E79K vs. Nrf2^WT experiments. RT was performed using a 225kVp cabinet x-ray irradiator filtered with 0.5 mm Cu (IC-250, Kimtron Inc.). Radiation was administered in 5 fractions of 3 Gy each, to a total dose of 15 Gy. Bidimensional tumor measurements (the product of the longest diameter and its longest perpendicular diameter) with a caliper were performed every 3–5 days for each tumor. Mice were sacrificed when the tumor reached ~1500 mm³. Next, Nrf2^WT or Nrf2^E79Q or Nrf2^E79K xenografts were established in C57BL/6 mice as above. When the tumor volume reaches 100–150 mm³, the mice are randomized into four groups: Nrf2^WT + RT + vehicle, Nrf2^E79Q + RT + vehicle, Nrf2^WT + RT + CB-839, Nrf2^E79Q + RT + CB-839. Similar murine groups were used for Nrf2^E79K vs. Nrf2^WT experiments. RT (3 Gy × 5 fractions over 5 days) was delivered as described above. For CB-839 or vehicle treatment, the mice received either CB-839 (200mg/kg) or vehicle deliver through oral gavage twice daily starting 2 days before RT, during the 5 days of RT, and one more day after RT. CB-839 was provided by Calithera Biosciences. Bidimensional tumor measurements were taken as described above. The mice were sacrificed when the tumor reached ~1500 mm³.

FACS analysis.

At the time of sacrifice, the tumors were removed and dissociated to single cell using murine tumor dissociation kit (Catalog# 130–096-730, Miltenyi Biotec) according to the manufacturer’s instructions. Nonspecific binding was blocked using an anti–mouse CD16/CD32 antibody (Catalog# 101319, BioLegend). The cells were then processed for staining using the following antibodies purchased from BioLegend: CD45 (30-F11, Catalog #103147), CD4 (RM4–5, Catalog #100512), CD8 (53–6.7, Catalog #100712), CD11b (M1/70, Catalog #101206), CD11c (N418, Catalog #117333), F4/80 (BM8, Catalog #123131), Ly6G (1A8, Catalog #127607), Ly6C (HK1.4, Catalog #128041), CD206 (MMR, Catalog #141720). Antibody for MHC Class II (M5/114.15.2, Catalog #25–5321-82) was purchased from eBioscience. The cells were washed and fixed in 4% paraformaldehyde after staining. Cell acquisition was performed using LSR II flow cytometer (BD Biosciences). FlowJo and Cytobank software was used to analyze the data.

Statistics.

The time to event outcome of locoregional failure (LRF) were calculated as the time from the surgery date until either local or regional failure for LRF. Patients who did not experience an event were censored at the date of last follow up. Death was considered as a competing event and the patients who died before experiencing an event were designated as having succumbed to the competing event. These time to event outcomes were analyzed in a competing risk model and the patients with NFE2L2 mutations were compared to those without NFE2L2 mutations using Gray’ test. The time to event outcomes of overall survival (OS) and progression free survival (PFS) were calculated as the time from the surgery date until death for OS, and until death or progression for PFS. Patients who did not experience an event were censored at the date of last follow up. The data was summarized using Kaplan-Meier curves and the patients with NFE2L2 mutations were compared to patients without NFE2L2 mutations using a log-rank test.

Comparisons of measurement data between two groups were analyzed using an unpaired, two-tailed Student’s t test. Tumor growth curves were analyzed by in a mixed effects model to account for the within subject correlation. All tests are two-sided with an alpha level of 0.05. All analyses of patient’s data were performed in SAS v9.4 (SAS Institute Inc, Cary, NC) and all graphs were generated in Prism (GraphPad Software LLC, San Diego, CA).

Study approval.

Studies related to human OC cancer patients were approved by the Stanford Institutional Review Board (IRB-10564). Written informed consent was received from all OC patients prior to their inclusion in this study. All the animal studies have been approved by the Institutional Animal Care and Use Committee (IACUC) of Stanford University, under APLAC protocol number 15106.

Data availability statement

The TCGA dataset we used in this study was Head and Neck Squamous Cell Carcinoma (TCGA, Firehose Legacy) which was obtained from the cBioportal at www.cbioportal.org. The raw RNA sequencing data were deposited in the GEO database with accession number GSE222658. The raw DNA sequencing data generated in this study are not publicly available due to privacy restrictions related to dissemination of germline sequencing information included in the informed consent forms used to enroll study subjects but are available upon reasonable request from the corresponding author. The other data generated in this study are available within the article and its supplementary data files.

Results

KEAP1/NFE2L2 gene mutation in oral cavity cancer.

To study the genetic alterations involved in radiation resistance, we first performed targeted next generation sequencing in a unique cohort of 75 oral cavity cancer patients treated with SRT for 325 genes (Figure 1A); 68 tumor samples were successfully sequenced. 55 (81%) of 68 sequenced tumors had at least one mutation in our selected gene set. Among the 325 genes, 23 were found to be recurrently mutated (Figure 1B and Supplemental Table 4), while 41 genes were singleton (Supplemental Figure 1A and Supplemental Table 4). Recurrently mutated genes were sorted and tabulated according to their frequency of occurrence (Figure 1B). The most commonly mutated genes with mutations identified above 5% variant allele fraction (VAF) were TP53 (53%), PIK3CA (12%), ZNF536 (10%), NFE2L2 (7%), PABPC5 (6%) and CDKN2A (6%) (Figure 1C). As shown in Figure 1C, the mutation frequency of our patients was similar to that found in the CTGA oral cavity cohort and the CTGA overall HNSCC except for somewhat lower TP53 and CDKN2A mutation frequency noted in our patient group.

Figure 1. KEAP1/NFE2L2 gene mutation in patients (n=75) with oral cavity cancer treated with surgery and postoperative (chemo)radiation at Stanford University.

(A) Schematic model of the experiment design. (B) Landscape of recurrent gene mutations identified by targeted next generation sequencing and clinical characteristics of these patients. (C) Comparison of somatic mutation frequency of the most commonly mutated genes, including TP53, ZNF536, PIK3CA, NFE2L2, PABPC5 and CDKN2A, in the TCGA head and neck squamous cell carcinoma (HNSCC) cohort and the Stanford OC cancer patient. (D) Observed NFE2L2 mutations in patients with OC cancer cohort treated with surgery and (chemo)radiation at Stanford University.

Since mutations in the NFE2L2/KEAP1/CUL3 pathway have been linked to radiation resistance, we focused on these mutations. We found that 7% of the sequenced tumor had mutation in the NFE2L2 gene, which is identical to the rate noted in the TCGA OC cohort. Interestingly, there was no KEAP1 mutation and only 3 CUL3 mutations in this group. For the NFE2L2 mutations, all were found at the 79^th residue where the acidic glutamic acid was mutated to either a neutral glutamine (E79Q, C to G mutation in the DNA) or a basic lysine (E79K, in C to T mutation in the DNA) (Figure 1D). This mutational pattern is similar to that found in the TCGA for oral cavity cancer. Because NFE2L2 mutations occurred at a unique residue, we were able to validate all mutated cases with sanger sequencing using DNA from the tumors and their respective normal tissues (Supplemental Figure 1B).

The relationship between NFE2L2 mutation and radiation resistance in OC cancer.

To identify the genes involved in the RT resistance in OC, we firstly investigated the association between individual mutated genes and locoregional failure (LRF) in the above sequenced 75 OC patients treated with SRT. Among the top-ranking mutated genes, NFE2L2 mutation was the only factor associated with increased LRF with a hazard ratio (HR) of 4.24 (95% confidence interval (CI): 2.35–7.63, adjusted p<0.001). There was a trend for higher LRF with PABPC5 mutation, but the difference did not reach statistical significance (HR 3.07; 95% CI: 1.11–8.48, adjusted p=0.059). There was no association with LRF for mutations in other commonly mutated genes including PIK3CA (HR 0.36; 95% CI: 0.05–2.79, adjusted p=0.413), TP53 (HR 0.40; 95% CI: 0.17–0.94, adjusted p=0.059), ZNF536 (HR 1.08; 95% CI: 0.23–5.00, adjusted p=0.922) (Figure 2A). CDKN2A mutation was not included in this Cox proportional hazards regression analysis because there was no LRF events in the patients with this mutation. NFE2L2 mutation was the only significant factor when analyzed in Cox models that included relevant prognostic clinical factors such as treatment (Surgery + RT vs. Surgery + chemoRT), T-stage, N-Stage, age, smoking status, and gender (Figure 2B). Figure 2C shows the cumulative incidence of time-to-LRF by NFE2L2 mutation status in these patients. The 5-year cumulative incidence of LRF was 80% for those with NFE2L2 mutated tumor vs. 32% for those without (p = 0.007). There was a trend for lower progression-free survival (PFS) with NFE2L2 mutation (p = 0.075) but no difference for overall survival (p = 0.333) (Supplemental Figure 2A, B). These results indicate that NFE2L2 mutation is a significant risk factor for RT resistance, leading to locoregional relapse. We also verified that all relapses (local and regional) were within the radiation fields.

Figure 2. The relationship between NFE2L2 mutation and radiation resistance in OC cancer.

(A) Forest plot showing the impact of individual gene mutation, displayed as hazard ratio (HR), on locoregional failure (LRF) for NFE2L2, PABPC5, PIK3CA, TP53 and ZNF536. HR was calculated by Cox regression model. CI: confidence interval. (B) Forest plot showing hazard ratio of NFE2L2 mutation on locoregional failure (LRF) across different clinical subgroups, including treatment, age, T-stage, N-stage, smoking status, and gender. Treatment stands for Surgery + RT vs. Surgery + chemoRT. (C-D) Cumulative incidence of LRF in patients with or without NFE2L2 mutation from OC patients treated with surgery and postoperative (chemo)radiation (C) or surgery alone (D). p-value was calculated by the log rank test. (E-F) The recurrent rate of patients with (right panel) or without (left panel) NFE2L2 mutation in OC patients treated with surgery and postoperative (chemo)radiation (E) or surgery alone (F). Lower panel showing the patient number in the indicated groups.

To determine whether NFE2L2 is simply a prognostic marker or whether it is a predictive marker of RT resistance, we analyzed an independent cohort of 60 OC patients treated primarily with surgery alone (S). There was no association between NFE2L2 mutation and time-to-LRF in this S group (p = 0.611, Figure 2D). Collectively, in the SRT cohort, all patients (100%) with NFE2L2 mutation developed LRF, whereas this was seen in only 36% of the patients without NFE2L2 mutation (Figure 2E). In S cohort, 20% of the patients with NFE2L2 mutation developed LRF, whereas this was seen in 35% of patients without mutation (Figure 2F). Taken together, these data suggested that NFE2L2 mutation is a predictive marker of RT resistance rather than a prognostic marker of poor outcome.

Somatic mutation of NFE2L2 leads to Nrf2 activation and promotes intrinsic RT resistance of human HNSCC cells.

Given that NFE2L2 mutation is a potential predictive marker of RT resistance in OC cancer patients, we investigated the biological function of NFE2L2 mutation in RT response in HNSCC cells. To remove the confounding effects of basal Nrf2, we generated NFE2L2 knockout (KO) SAS and FaDu cells using CRISPR-Cas9 technology; the gRNA is listed in Supplemental Figure 3A. Consistent with previous reports(21), higher ROS level (Supplemental Figure 3B), and lower mRNA level of Nrf2 downstream targets, including HMOX1 and NQO1 (Supplemental Figure 3C), were detected in Nrf2^KO cells. In addition, clonogenic survival assay revealed that Nrf2^KO cells were more sensitive to radiation (Supplemental Figure 3D). Next, we reconstituted the NFE2L2 transcript that codes for Nrf2^WT, Nrf2^E79Q or Nrf2^E79K mutation into these NFE2L2 knockout cells (Figure 3A). Immunoblots show stable expression of the mutated Nrf2 protein and its downstream target protein (NQO1) in 2 different cell lines (Figure 3B, Supplemental Figure 3E). ARE-dependent firefly luciferase assays in SAS cells stably expressing the Nrf2^E79Q or the Nrf2^E79K showed these cells had a significantly increase in luciferase activity compared to those expressing Nrf2^WT (Figure 3C). Consistent with the luciferase reporter data, SAS or FaDu cells expressing Nrf2^E79Q or Nrf2^E79K showed a significant increase in mRNA expression of Nrf2 downstream targets genes, including NQO1, HMOX1, ABCC2, GCLC and AKR1C1 compared to their wild-type counterpart (SAS cells - Figure 3D, FaDu cells - Supplemental Figure 3F). Nrf2 mutant protein translocated into the nucleus as reflected in immunofluorescent staining (Figure 3E, Supplemental Figure 3G).

Figure 3. Somatic mutation of NFE2L2 leads to Nrf2 activation and promotes intrinsic RT resistance in human HNSCC cells.

(A) Schematic Model of the experiment design for generating cell lines stably expressing Nrf2^WT or Nrf2^Mut. (B) Immunoblot showing the expression of Nrf2 and HMOX1 proteins in SAS cells stably expressing Nrf2^WT, Nrf2^E79Q or Nrf2^E79K. (C) Nrf2 mediated ARE-promoter luciferase activity in Nrf2^WT, Nrf2^E79Q and Nrf2^E79K SAS cells. (D) qRT-PCR analyses showing the expression level of Nrf2-downstream target genes including NQO1, HMOX1, ABCC2, GCLC, and AKR1C1 in Nrf2^WT, Nrf2^E79Q and Nrf2^E79K SAS cells. (E) Immunofluorescent staining of Nrf2 in indicated SAS cells. Cell nuclei were counterstained with DAPI (blue). (F) Nrf2^WT or Nrf2^E79Q SAS cells were exposed to the increasing radiation doses of 0, 1, 2, and 4 Gy. ROS levels were measured by flow cytometry after 6 hours. (G) Nrf2^WT or Nrf2^E79Q SAS cells were subjected to increasing radiation doses of 0, 2, 4, and 6 Gy. Cells were plated, incubated, and the number of colonies was counted after 7 days. The survival fractions were calculated after normalizing to the plating efficiency of the un-irradiated cells. The results are represented as the mean ± SD of three independent experiments. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

Since both mutations behave similarly and since Nrf2^E79Q is a mutational hotspot that has been shown to impair Keap1’s binding of Nrf2(22), we used Nrf2^E79Q as a representative to further test the function of NFE2L2 mutations in RT resistance. Given that Nrf2 is major mediator of the antioxidant response cancer cells(23), we assess the level of reactive oxygen species (ROS) with radiation in NFE2L2 mutant and wildtype cells. There was a significant reduction in the intracellular ROS levels with RT in Nrf2^E79Q cells compared to Nrf2^WT cells using SAS (Figure 3F) and FaDu model (Supplemental Figure 3H). Importantly, clonogenic survival assay revealed that Nrf2^E79Q cells were more radioresistant than Nrf2^WT cells treated with the increasing doses of ionizing radiation (IR) in SAS (Figure 3G) and FaDu cell lines (Supplemental Figure 3I). Collectively, these results demonstrate that the detected NFE2L2 mutation in OC leads Nrf2 activation, which mediates antioxidant response, resulting in intrinsic RT resistance in vitro.

NFE2L2 mutation leads to extrinsic radiation resistance through recruitment of PMN-MDSCs in tumors.

To identify additional mechanism by which Nrf2^E79Q mediates RT resistance, we performed RNA sequencing of SAS cells stably expressing either Nrf2^WT or Nrf2^E79Q treated with RT and analyze the differentially expressed genes in Nrf2^E79Q vs. Nrf2^WT cells. As expected, the expression of antioxidant genes was increased in Nrf2^E79Q cells (Supplemental Figure 4A). To uncover other potential signaling pathways, we performed Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis. Interestingly, we found that the IL17 pathway was highly activated in Nrf2^E79Q compared to Nrf2^WT cells post RT (Figure 4A), with numerous upregulated cytokines including CXCL1, CXCL3, CXCL8, CSF2, CSF3 etc. as showed in the heat map (Figure 4B). Consistent with the RNA sequencing data, MOC1 cells expressing Nrf2^E79Q showed a significant increase in mRNA expression of CXCL1, CXCL3, CSF2 and CSF3 (Figure 4C). Since many of these cytokines, including CXCL1, CXCL3, CSF3 and CSF2, can influence the tumor microenvironment (TME) and affect the tumor development(24)(25)(26), we hypothesize that Nrf2^E79Q might affect the radiation resistance through an extrinsic approach by regulating the TME.

Figure 4. NFE2L2 mutation leads to extrinsic radiation resistance through recruitment of PMN-MDSCs in tumors.

(A) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of Nrf2^WT and Nrf2^E79Q SAS cells treated with RT. Gradient color bar showed –log10 FDR (False Discovery Rate); circle size indicates the ratio of the number of differentially expressed genes in a pathway over the total number of genes in this pathway. Gradient color bar showed Z-score normalized CPM (counts per million) across each group. (B) Heat map of differential gene expression from the IL-17 signaling pathway in A. Genes that had significantly different expression (FDR < 0.05) and at least 1.5-fold change in expression were plotted in heat map. (C) qRT-PCR analyses of the gene expression level of CXCL1, CXCL3, CSF2 and CSF3 at 24 hours after 4 Gy irradiation of Nrf2^WT or Nrf2^E79Q MOC1 cells. (D-E) Subcutaneous xenograft tumors established in NRG (NOD-Rag1^null IL2rg^null, NOD rag gamma) mice (D) and C57BL/6J mice (E), respectively treated with RT (3 Gy × 5 fractions). Tumor volumes were monitored by direct caliper measurement. The upper panel shows the experiment design. The lower panel shows the tumor volume (Left) and Kaplan Meier survival curves (Right) for each group using 1500mm³ tumor size as a proxy for survival. N=5 mice/group. (F-H) Comparison of PMN-MDSCs (F), M1 polarized macrophages (G) and M2 polarized macrophages (H) levels (expressed as % of live cells) in Nrf2^WT and Nrf2^E79Q MOC1 tumors treated with RT. Cells isolated from Nrf2^WT and Nrf2^E79Q tumors treated with RT were stained for the indicated markers and analyzed by flow cytometry. Scatter plot (left panel) shows the difference in Nrf2^WT and Nrf2^E79Q tumors with each dot representing 1 mouse. Flow cytometric data were merged to create single t-distributed stochastic neighbor embedding (tSNE) map to show PMN-MDSC and M1 macrophages in Nrf2^WT and Nrf2^E79Q tumors (right panel). Error bars indicate the mean ± SD. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

To explore the effect Nrf2^E79Q in the TME with radiation, we knocked out NFE2L2 in a mouse oral squamous cell carcinoma cell line MOC1 using CRISPR-Cas9 (Supplemental Figure 4B), then generated MOC1 cells stably expressing either Nrf2^WT, Nrf2^E79Q or Nrf2^E79K. Consistent with the results in SAS and FaDu cells, higher ROS level (Supplemental Figure 4C) and lower mRNA level of Nrf2 downstream targets (Supplemental Figure 4D) were seen in Nrf2^KO compare to Nrf2^WT MOC1 cells. Similarly, Nrf2^KO MOC1 cells were more sensitive to radiation than Nrf2^WT MOC1 cells by clonogenic survival assay (Supplemental Figure 4E). As expected, the downstream targets HMOX1 and NQO1 were upregulated in Nrf2^E79Q MOC1 cells (Supplemental Figure 4F). Next, the Nrf2^WT or Nrf2^E79Q MOC1 cells were implanted in the flank of either immune competent C57BL/6J mice or immune deficient NRG (NOD-Rag1^null IL2rg^null, NOD rag gamma) mice to establish xenografts. Nrf2^E79Q and Nrf2^WT MOC1 tumors had a similar growth rate in NRG mice without RT treatment (Supplemental Figure 4G). In contrast, the growth of Nrf2^E79Q MOC1 tumors was slightly faster than Nrf2^WT counterparts in C57BL/6J mice (Supplemental Figure 4H), suggesting that Nrf2^E79Q mutation in tumor cells has an effect on the TME that supports faster tumor growth in immune competent mice.

Intriguingly the effect of Nrf2^E79Q on radiation resistant was substantially larger in immune competent than in immune-compromised mice. Although Nrf2^E79Q tumors grew slightly faster than Nrf2^WT tumors in NRG mice after radiation (Figure 4D, Supplemental Figure 4I), they grew much faster than Nrf2^WT tumors in C57BL/6J mice (Figure 4E, Supplemental Figure 4J). These data suggested that Nrf2^E79Q exerts an effect on the immune function that leads to radiation resistance. To dissect the potential contribution of immune cells in the TME in promoting radiation resistance of Nrf2^E79Q tumors, we analyzed the composition of immune cells in Nrf2^WT and Nrf2^E79Q tumors from C57BL/6J mice after RT. The gating strategy for CD45⁺ cells, tumor-associated macrophages (TAMs, CD11b⁺Ly6G⁻F4/80⁺), polymorphonuclear myeloid-derived suppressor cells (PMN-MDSCs, CD11b⁺F4/80⁻Ly6G^highLy6C^mid), M1 macrophages (CD11b⁺ F4/80⁺CD206⁻MHC Class II⁺), M2 macrophages (CD11b⁺F4/80⁺CD206⁺), CD4⁺ T cells, CD8⁺ T cells are showed in Supplemental Figure 5A–C. There was no difference in the percentage of CD45⁺, CD4⁺, CD8⁺ and TAMs cells between Nrf2^WT and Nrf2^E79Q tumors (Supplemental Figure 5D–G). However, Nrf2^E79Q tumors displayed a significantly higher level of PMN-MDSCs (Figure 4F) and lower M1 polarized macrophages after RT compared to Nrf2^WT tumors (Figure 4G). In addition, Nrf2^E79Q tumors showed higher M2 macrophages after RT compared with Nrf2^WT tumors (Figure 4H). These data suggest that Nrf2^E79Q mutation in the tumor cells can upregulate PMN-MDSCs and down-regulate M1 polarized macrophages, contributing to extrinsic radiation resistance.

CB-839 overcomes RT resistance in Nrf2^E79Q MOC1 cells by modulating myeloid response.

Considering the gain-of-function of NFE2L2 mutation and its role in radiation resistance in HNSCC, therapeutic approaches targeting the Nrf2 activity of cancer cells, especially in HNSCC with NFE2L2 mutations, deserved further investigation. Recent studies have shown that Nrf2-activated lung cancer cells via Keap1 loss have altered metabolic requirements that render them markedly sensitive to the glutaminase inhibitor CB-839, which converts glutamine into glutamate(10, 27). Therefore, we hypothesized that Nrf2^E79Q cells are vulnerable to CB-839 and that the addition of this drug to RT would increase RT sensitivity in Nrf2^E79Q compared to Nrf2^WT tumors in vivo. To address this, we treated Nrf2^E79Q and Nrf2^WT MOC1tumors grown in immune competent C57BL/6J mice with RT + vehicle vs. RT + CB839 (200mg/kg via oral gavage, twice daily) as indicated in Figure 5A. Treatment with CB839 alone without RT had a delay effect on tumor growth that was similar between Nrf2^E79Q and Nrf2^WT tumors (Supplemental Figure 6A). However, it abolished the radiation resistant effect of Nrf2^E79Q tumors. Irradiated Nrf2^E79Q tumors grew substantially faster than irradiated Nrf2^WT tumors; however, both tumor types grew at similar rates when treated with RT and CB-839 (Figure 5B, Supplemental Figure 6B). As shown in Supplemental Table 5, the treatment with CB839 alone had a delay effect on tumor growth that was similar between Nrf2^E79Q and Nrf2^WT tumors. However, the combination of RT + CB839 led to a much larger tumor growth delay for Nrf2^E79Q tumors than for Nrf2^WT tumors. The effect of tumor growth delay is reflected in survival, where mice bearing Nrf2^E79Q tumors treated with RT alone had the shortest survival and those treated with RT + CB-839 had the best (Figure 5C). We also performed the same experiments on MOC1 tumors bearing the Nrf2^E79K mutation, which is another common NFE2L2 mutation seen in the OC cohort and the CTGA dataset and observed similar results as shown for the Nrf2^E79Q mutation (Supplemental Figure 6C–F). Taken together, these data suggested that CB-839 treatment could overcome the radiation resistance of Nrf2^Mut.

Figure 5. CB-839 overcomes RT resistance in Nrf2^E79Q MOC1 cells through regulating TME.

(A) Schematic representation of experimental design for MOC1 mouse tumor model. (B-C) Nrf2^WT or Nrf2^E79Q MOC1 xenografts were treated with either RT alone or RT + CB-839. Average tumor volumes (B) and Kaplan Meier survival (C) using the threshold of 1500mm³ tumor volume as a proxy for survival. (D) tSNE plots of flow cytometry data sets showing different clusters of immune cell populations in Nrf2^WT and Nrf2^E79Q MOC1 tumors treated with RT or RT + CB-839. Cells are color-coded and represent the mean of cell density in each cluster. Red circles outline the PMN-MDSC cluster. (E-G) Percentage of PMN-MDSC (E), M1 polarized macrophages (F) and M2 polarized macrophages (G) among live cells from indicated group. Data are represented as mean ± SD. *P<0.05, ****P<0.0001.

Recent studies showed that targeting glutamine metabolism enhanced tumor-specific immunity via modulating suppressive myeloid cells(25). Thus, we hypothesize that glutaminase inhibition with CB-839 leads to more favorable changes in the TME of Nrf2^Mut tumors especially in the context of radiation therapy. To address this hypothesis, we used multi-parametric flow cytometry to phenotype immune cells in the indicated tumors (Figure 5D). These studies showed a significantly higher numbers of PMN-MDSCs cells in irradiated Nrf2^E79Q (Figure 5D–E) or Nrf2^E79K (Supplemental Figure 6G) compared to irradiated Nrf2^WT tumors. Treatment with CB-839 did not affect the number of MDSC in irradiated Nrf2^WT tumors; however, it substantially reduced the number of PMN-MDSCs in irradiated Nrf2^E79Q (Figure 5D–E) or Nrf2^E79K tumors (Supplemental Figure 6G). Of note, CB-839 treatment did not significantly affect the proportion of CD45⁺ cells (Supplemental Figure 6H–I), CD4⁺ T cells (Supplemental Figure 6J–K) or CD8⁺ T cells (Supplemental Figure 6L–M). Treatment with CB-839 increased the level of tumor associated M1 macrophages in irradiated Nrf2^WT tumors compared to Nrf2^E79Q (Figure 5F) or Nrf2^E79K (Supplemental Figure 6N). This was associated with a corresponding decrease in M2 macrophages in irradiated Nrf2^E79Q (Figure 5G) or Nrf2^E79K tumors (Supplemental Figure 6O). Taken together, CB-839 treatment appears to overcome the Nrf2^mut-induced radiation resistance in part through enhanced PMN-MDSCs recruitment and a shift in balance between M1 and M2 macrophages.

CB-839 reverses the Nrf2^E79Q-induced cytokine changes after radiation through regulating TLR4 expression.

To get insight into the mechanism by which CB-839 regulates PMN-MDSCs recruitment to overcome Nrf2^E79Q-induced RT resistance, we performed RNA sequencing of Nrf2^WT and Nrf2^E79Q SAS cells treated with either RT alone or RT + CB-839. In total, 371 genes were upregulated in Nrf2^E79Q cells compared with Nrf2^WT cells after RT treatment, indicating that some of these genes may contribute to the Nrf2^E79Q-induced radiation resistance. 402 genes were downregulated after CB-839 treatment in irradiated Nrf2^E79Q cells compared to irradiated Nrf2^WT cells. 69 genes overlapped between these two groups (Figure 6A, Supplemental Figure 7A), suggesting that these genes may participate in CB-839’s overcoming Nrf2^E79Q-induced RT resistance. These differentially expressed 69 genes were significantly enriched in the IL17 pathway (Figure 6B) that was previously seen to be induced by NFE2L2 mutation with RT (Figure 4A). Some of the upregulated genes include CXCL1, CXCL3, CSF3, MMP1, MMP13 (Figure 6C). To confirm that the changes in these genes are also seen in MOC1 cells, we treated the Nrf2^WT and Nrf2^E79Q MOC1 cells with RT or RT+CB-839 and perform qPCR for the above genes. The results showed that CXCL1, CXCL3, CSF3, MMP1 and MMP13 were all upregulated in Nrf2^E79Q MOC1 cells with RT and restored with the combined CB-839 + RT treatment (Figure 6D). Unlike SAS cells, there was no difference in Ptgs2 expression between Nrf2^E79Q and Nrf2^WT MOC1 cells (Supplemental Figure 7B). We also confirmed that the noted changed in CXCL1, CXCL3, and CSF3 also occurred in vivo (Supplemental Figure 7C). Multiple studies have shown that CXCL1, CXCL3, and CSF3 are involved in regulating the recruitment of PMN-MDSCs to the TME(24–26), suggesting that RT induced the expression of these chemokine in Nrf2^E79Q tumors, leading to MDSC recruitment to these tumors and subsequent RT resistance, and this was reversed by CB-839.

Figure 6. CB-839 reverses the Nrf2^E79Q-induced cytokine changes after radiation through regulating TLR4 expression.

(A) Vent diagram showing the differential gene expression regulated by Nrf2^E79Q and CB-839 in combination with RT. 371 genes were upregulated in Nrf2^E79Q cells compared with Nrf2^WT cells after RT. 402 genes were down regulated by CB-839 in Nrf2^E79Q cells treated with RT (4Gy). 69 overlapping genes between the two group were identified. (B) KEGG pathway enrichment analysis of 69 genes from (A) showing IL17 signaling pathway as the top upregulated pathway. (C) Heat map of differential gene expression from the IL-17 signaling pathway in (B). Genes with at least 1.5-fold change in expression were plotted in the heat map. (D-F) Nrf2^WT or Nrf2^E79Q MOC1 cells were pretreated with 0.5 μM CB-839 or vehicle for 24 hours, followed by RT (4 Gy) for another 24 hours. qRT-PCR analysis showing the expression level of CXCL1, CXCL3 and CSF3 in indicated MOC1 cells (D). qRT-PCR analysis showing the expression level of TLR4 in indicated MOC1 cells (E). Immunoblot showing protein expression of TLR4, phospho-S6, and total-S6 in MOC1 cells stably expressing Nrf2^WTor Nrf2^E79Q treated with either RT or RT + CB-839 (F). (G) qRT-PCR analyses showing the expressional level of CXCL1, CXCL3, and CSF3 in cells with indicated treatment. Error bars indicate the mean ± SD; n=3 independent experiments. **P<0.01, ***P<0.001, ns means no significant difference.

To understand the molecular link between NFE2L2 mutation and cytokines upregulation after radiation, we returned to the RNA-seq data (Supplemental Figure 7A) and searched for potential candidates that may be involved in chemokine modulation. We noted that that TLR4 was significantly upregulated in Nrf2^E79Q cells after RT, and was restored after CB-839 treatment (Supplemental Figure 7A). Since TLR4 has been shown to positively regulate the expression of CXCL1, CXCL3 and CSF3(28–30), we speculate that it may be involved in the chemokine changes in Nrf2^E79Q MOC1 cells after RT. qPCR (Figure 6E) and Western Blot (Figure 6F) confirmed an increase in both mRNA and protein expression of TLR4 in Nrf2^E79Q MOC1 cells after RT. These increases in expression were reversed with the addition of CB-839 (Figure 6E, F). Silencing of TLR4 expression (Supplemental Figure 7D) prevented to the induction of CXCL1, CXCL3 and CSF3 (Figure 6G) by RT in Nrf2^E79Q MOC1 cells. Similarly, the addition of CB-839 to RT in Nrf2^E79Q and Nrf2^E79Q cells that had down-regulated TLR4 expression failed to further reduce cytokines expression in the absence of TLR4 (Figure 6G), suggesting that TLR4 is involved in CB-839 mediated cytokine reduction in Nrf2^E79Q cells after RT. Taken together, these data suggest TLR4 and TLR4-mediated cytokines are induced in Nrf2^E79Q cells after RT and CB-839 reverses the Nrf2^E79Q-induced cytokine changes after RT through regulating TLR4 expression.

Discussion

HNSCC is a significant public health burden worldwide and RT is a major treatment for this disease. Presently, definitive biomarkers that can predict RT resistance are lacking. The only existing biomarkers are those of prognosis with the most established one being the human papillomavirus (HPV) in oropharyngeal carcinoma(31). Thus, having predictive biomarker(s) of RT resistance and specific strategies to target some of these radioresistant phenotypes is crucial to move the needle in HNSCC. Pan-cancer analyses have revealed frequent mutations of the KEAP1/NFE2L2 pathway in different solid cancers, including HNSCC (32, 33). The relationship between KEAP1/NFE2L2 mutations and RT resistance in HNSCC is of crucial interest. Since HNSCC is a heterogenous group of squamous cell carcinoma arising along the aerodigestive track from the oral cavity to the hypopharynx, we focused our investigation on oral cavity cancer (OC) for several reasons. First, its initial treatment is surgery, thus providing large amount of tumor and match normal tissues for sequencing. Next, OC are free of HPV, thus removing the potential confounding effect of HPV. Most importantly, because some OC patients are treated with surgery alone and some are with surgery and postoperative radiation, it allows us to determine whether the marker of interest is a prognostic marker of poor outcome or a potential predictive marker of RT resistance. If a marker were a prognostic marker of aggressive tumor behavior, then its presence should be associated with worse outcomes for both treatment groups. However, if it were a marker of radiation resistance, then its presence should be associated with worse outcome only in the patients who received RT as part of their therapy. This was indeed the case for NFE2L2 activating mutation, whose presence was associated with a high risk of LRF. In fact, all patients with NFE2L2 mutation developed LRF whereas only 35% of those without did in the SRT group. In contrast, the rate of LRF was somewhat lower for patient with NFE2L2 mutation compared to those without in the surgery only group. The association of NFE2L2 mutation and LRF in the SRT group was independent of known prognostic factors such as tumor stage, nodal stage, cigarette use or treatment, indicating that it is a predictive marker of RT resistance in oral cavity cancer. Of note, we focused on LRF because it is the best reflection of radiation failure since the radiation fields covered only the primary tumor site and the draining lymph nodes. PFS, which includes LRF, distant metastasis and death as events, is not a direct reflection of RT failure.

Our rate of NFE2L2 mutations is similar to that observed in the TCGA data set, which also shows the most common mutation to be at the 79th residue. However, unlike TCGA, we did not observe any other NFE2L2 mutations. Similarly, we did not observe any KEAP1 mutations in our patient cohort. This may be due to our limited sample size. Another genetic alteration that can lead to Nrf2 activation is NFE2L2 gene amplification; however, this is only seen in 1% of oral cavity cancers in the TCGA database based on our analysis. Thus, we did not evaluate the NFE2L2 copy number changes in our study.

It is well known that activation of the Nrf2-Keap1 pathway would lead to intrinsic RT resistance in cancer via its modulation of RT-generated ROS and electrophilic stresses (34). However, less is knowns about the crosstalk between NFE2L2 activation and the tumor microenvironment in radiation resistance. Our data revealed a distinct extrinsic mechanism that NFE2L2 mutation can lead to radiation resistance through the recruitment of PMN-MDSCs to the TME. Using both immune competent and compromised hosts, we showed that RT can induce TLR4, leading to the production of chemokines (CXLC1, CXCL3, CSF3) that recruit PMN-MDSCs to the TME. It is well known that PMN-MDSCs are capable of supporting tumor growth through inhibiting T cell activation and function(35, 36). We showed that treatment with CB-839 reversed PMN-MDSCs recruitment and enhanced RT sensitivity in Nrf2 tumors. CB-839, a glutaminase inhibitor, has been reported to preferentially sensitizes KEAP1-mutant non-small cell lung cancer cells to RT in vitro because of the dependence of cells with KEAP1 loss on glutamine metabolism(27) that is needed for the production of GSH, a critical antioxidant that has been linked to RT resistance. Thus, glutaminase inhibition with CB-839 can lead to reduced GSH production and preferentially radiosensitize KEAP1/NFE2L2-mutant cells in vitro. However, limited work has been done to characterize the immune microenvironment of NFE2L2 mutant tumors with CB-839 treatment. To our knowledge, we are the first to show that CB-839 treatment, in addition to its intrinsic effect on NFE2L2 mutant HNSCC, reduces recruitment of PMN-MDSCs to the TME, leading to enhanced RT sensitivity in these tumors. Thus, preferential inhibition of glutamine metabolism in tumor cells with NFE2L2 mutations may represent a promising treatment strategy that enhances antitumor immune responses with RT in these mutated HNSCC.

In addition to changes in PMN-MDSC levels, we also noted changes in the M1/M2 macrophage ratio in the TME when CB-389 was added to RT. Compared to Nrf2^WT MOC1 tumors, RT led to more M2 and fewer M1 macrophages in the TME of Nrf2^E79Q MOC1 tumor, which was abolished with the addition of CB-839. This phenomenon may be due to the change in MDSC levels in the TME. In addition to inhibiting T cell activation, MDSC also impact antitumor immunity by polarizing macrophages toward a tumor-promoting phenotype (M2 macrophages) (37–39). Thus, our observation of M1-M2 polarization regulated by Nrf2^E79Q and CB-839 could be the consequence of PMN-MDSCs changes. Moreover, it has been shown that CB-839, by inhibiting glutamine metabolism, can prevent M2 polarization since glutamine metabolism is fundamental for this process(40); however, in vitro study did not show that it can repolarize M2 to M1 macrophages(40). Additional studies are needed to confirm these observations in other tumors and determine the mechanism behind this action.

Weaknesses of this study include the retrospective nature of the patient cohorts, its small sample size, the low percentage of NFE2L2 mutation. Because of the small size of this study, further investigation with independent datasets or larger databases is warranted. Even though NFE2L2 activating mutation are found in only 5–10% of patients with oral cavity cancer, these mutations are associated with enhanced radiation resistant, leading to a significant decrease in locoregional control in these patients when treated with traditional therapy. Knowing the mutation status upfront will help to determine whether these patients should receive more intensive adjuvant therapy or be enrolled in clinical trials testing novel therapy targeting this pathway. Moreover, inclusion of these patients in radiation-based clinical trials can potentially lead to spurious findings as their worse outcome may impact the results if they are not evenly distributed across the treatment arms. Moreover, with the advent of precision medicine, identifying a reliable predictive biomarker of radiation resistance that can be overcome with a specific therapy is crucial to move the field forward in HNSCC. To date, most biomarkers in HNSCC are prognostic markers. There is a published pan-tumor predictive gene expression signature of radiation resistance; however, the dataset to support this signature in HNSCC is small and it does not have a companion specific treatment strategy to surmount this resistance. In addition, CB839 is being tested with radiation in high grade gliomas (NCT03528642) or platinum-based chemotherapy in first line recurrent/metastatic lung cancer (NCT04265534). Thus, with such extensive human safety data, it is highly feasible to conduct a trial to test the combination of CB839 and postoperative radiotherapy in patients with NFE2L2 mutated oral cavity cancer. Finally, our observation that RT magnifies the recruitment of PMN-MDSCs to NFE2L2 mutated HNSCC provides mechanistic basis to test novel MDSC targeting strategies in these tumors.

In summary, we provide new insights into the impact of gain-of-function NFE2L2 mutation on radiation resistance, and propose that CB-839 can overcome the NFE2L2 mutation induced RT resistance through modulating the TME. Although this study focusses on oral cavity cancer, it can potentially be applicable to other types of tobacco-induced squamous cell carcinoma, such as larynx and lung cancers, which have a higher percentage of activating mutation in the NFE2L2-KEAP1 pathway. We are presently working to sequence a cohort of early stage glottic cancer patients treated with RT in order to generalize our findings to other HNSCC. If validated, these findings have clinical implications and support the future clinical testing of combining RT +/− chemotherapy and CB-839 in HNSCC with NFE2L2 mutations.

Significance:

NFE2L2 mutations are predictive biomarkers of radioresistance in head and neck cancer and confer sensitivity to glutaminase inhibitors to overcome radioresistance.