Recurrent Human Papillomavirus–Related Head and Neck Cancer Undergoes Metabolic Reprogramming and Is Driven by Oxidative Phosphorylation

Abstract

Purpose:

Human papillomavirus (HPV) infection drives the development of some head and neck squamous cell carcinomas (HNSCC). This disease is rapidly increasing in incidence world-wide. Although these tumors are sensitive to treatment, approximately 10% of patients fail therapy. However, the mechanisms that underlie treatment failure remain unclear.

Experimental Design:

We performed RNA sequencing (RNA-seq) on tissues from matched primary- (pHNSCC) and metachronous-recurrent cancers (rHNSCC) to identify transcriptional differences to gain mechanistic insight into the evolutionary adaptations of metachronous-recurrent tumors. We used HPV-related HNSCC cells lines to investigate the effect of (i) NRF2 overexpression on growth in vitro and in vivo, (ii) oxidative phosphorylation (OXPHOS) inhibition using IACS-010759 on NRF2-dependent cells, and (iii) combination of cisplatin and OXPHOS inhibition.

Results:

The OXPHOS pathway is enriched in recurrent HPV-associated HNSCC and may contribute to treatment failure. NRF2-enriched HNSCC samples from The Cancer Genome Atlas (TCGA) with enrichment in OXPHOS, fatty-acid metabolism, Myc, Mtor, reactive oxygen species (ROS), and glycolytic signaling networks exhibited worse survival. HPV-positive HNSCC cells demonstrated sensitivity to the OXPHOS inhibitor, in a NRF2-dependent manner. Further, using murine xenograft models, we identified NRF2 as a driver of tumor growth. Mechanistically, NRF2 drives ROS and mitochondrial respiration, and NRF2 is a critical regulator of redox homeostasis that can be crippled by disruption of OXPHOS. NRF2 also mediated cisplatin sensitivity in endogenously overexpressing primary HPV-related HNSCC cells.

Conclusions:

These results unveil a paradigm-shifting translational target harnessing NRF2-mediated metabolic reprogramming in HPV-related HNSCC.

Graphical Abstract

Introduction

While human papillomavirus (HPV)–related head and neck squamous cell carcinoma (HNSCC) responds favorably to combinations of surgery and/or platinum-based chemoradiation (1), treatment failure portends a grim outlook with limited treatment options including morbid surgical resection, chemotherapy (i.e., with platinum-based agents) and/or reirradiation, or clinical trials mainly with immuno-therapeutic agents (2). From a public health standpoint, the incidence of HPV-related HNSCC has surpassed that of cervical cancer and is projected to continue rising until at least 2060 (3, 4). Given the ongoing rise in HPV-related HNSCC and challenges of managing treatment nonresponders, it is critical to understand the biological mechanisms underlying recurrent HPV-related HNSCC so that we may mitigate recurrence at the time of initial therapy.

To better understand the signaling mechanisms that drive recurrent HPV-related HNSCC, our group sought to identify differences in the genomic landscapes between recurrent and primary HPV-related HNSCCs. Our prior work found that recurrent HPV-related HNSCC shares a genomic landscape with carcinogen-associated (HPV-unrelated) HNSCC (5). A subsequent study identified a subset of primary HPV-related HNSCC that exhibited poor treatment response and a similar transcriptional landscape to HPV-unrelated HNSCC (6). In our prior work, we also compared primary HPV-related HNSCCs that did or did not recur and identified mutations exclusive to the primary tumors that ultimately recurred including NFE2L2 (the gene encoding NRF2; ref. 5). HPV-unrelated HNSCCs in The Cancer Genome Atlas (TCGA) study were notable for genomic alterations in NFE2L2 (7). NFE2L2 plays a role in regulating oxidative stress and mitigating the efficacy of chemoradiation (8) and also interacts with the HPV E1 protein (9).

While prior work focused on uncovering the features of primary HPV-related HNSCC associated with a poor prognosis, in this study, we sought to elucidate targetable mechanisms underlying recurrent disease. Our hypothesis is that HPV-related HNSCC contains unique expression phenotypes in key signaling pathways which may be exploited as therapeutic targets for primary treatment or mitigation of recurrence. Intriguingly, we observed activation of the oxidative phosphorylation (OXPHOS) pathway in the background of metabolic gene dysregulation among matched patient samples from primary and subsequent recurrent HNSCCs. We further confirmed that NRF2 functions as a driver of growth in an OXPHOS-dependent manner in HPV-associated HNSCC. In vitro findings illustrated a functional dependence on OXPHOS and reactive oxygen species (ROS) among NRF2-overexpressing cells conferring a critical weakness to OXPHOS inhibition. Taken together, these findings implicate NRF2-dependent OXPHOS dysregulation as a driver of a subset of HPV-associated HNSCC raising the intriguing possibility of rationally-targeted therapy in this subset of patients.

Material and Methods

Data collection

Clinical data from the University of Pittsburgh tumor samples were abstracted by study investigators. Clinical data for the TCGA data were abstracted from FireBrowse (http://firebrowse.org/; gdac.broadinstitute.org_HNSCC.Merge_Clinical.Level_1.2016012800.0.0/HNSCC.clin.merged.txt). HPV status was derived from the variable: patient. hpv_test_results.hpv_test_result.hpv_status (levels: positive, negative, indeterminate).

Archival tissue specimens from the University of Pittsburgh tumor samples were processed via fixation in 10% neutral buffered formalin, dehydrated in ethanol, and embedded with paraffin wax [formalin-fixed, paraffin-embedded (FFPE)]. Hematoxylin and eosin (H&E) slides were prepared, and areas with high tumor density (>75% tumor cells) were marked for extraction. Two-mm punch biopsies were taken from the FFPE tumor-dense regions for downstream tumor RNA extraction. FFPE tissue was deparaffinized with xylenes, washed in consecutive ethanol rinses (100% and 70%), and heated to remove formalin cross-linking (10).

Tumor RNA was extracted using the Quick-RNA FFPE Miniprep extraction kit (Zymo Research). RNA was quantified using the Qubit RNA High Sensitivity Assay Kit (Thermo Fisher Scientific). Sample integrity was evaluated using the Agilent Bioanalyzer RNA Nano and Pico kits (Agilent Technologies).

RNA sequencing and alignment

Sequencing and alignment steps for TCGA data were described previously (7, 11). TCGA RSEM normalized-expression data were obtained through FireBrowse (http://firebrowse.org/; illuminahiseq_rnaseqv2-RSEM_genes_normalized). University of Pittsburgh sample libraries were manually prepared via standard protocols using the Illumina TruSeq RNA Exome Library Prep Kit (Illumina) and sequenced on an Illumina NextSeq sequencing system using NextSeq 500 Mid- and High-Output 150 cycle kits (75 bp, paired-end; Illumina). Quality control was performed on the raw reads using RNA-SeQC (v1.1.7). Of the original 12 sets of paired primary and recurrent tumors, gene-expression data from two pairs were not included in the downstream analyses due to low quality. Preprocessed short reads were aligned to the human genome reference sequence assembly (hg19) with the STAR2 aligner using a two-pass procedure to generate RNAseq BAM files.

RNA expression analyses

FeatureCounts from SubRead (v1.6.0) was used for counting reads. Further expression analysis was performed in Empirical Analysis of Digital Gene Expression Data in R program (edgeR; refs. 12, 13). Counts per million (CPM) were calculated to control for sequencing depth. Starting with a total of 26,364 genes, we excluded 3,098 genes containing no reads across samples. Further filtering was performed by retaining genes with a sum of ≥ 6 CPM from at least 2 samples. An additional 9,473 genes were excluded with 13,793 samples kept for downstream analyses. RNA library-size normalization was performed using trimmed mean of M values. Mean library size was 18,439,805 (per sample library size and normalization factors presented in Supplementary Table S10). A negative binomial-generalized log-linear model was applied to determine differential expression including patient as a variable in the model to control for intrapatient differences in expression (expression ~ group + patient; group = = primary or recurrence). A cut-off fold change of ≥ 1 or ≤ −1 and a FDR q value of less than 0.05 were applied to select the most differentially-expressed genes. A ranked list of differentially-expressed genes was generated based on the edgeR FDR q values and log₂ fold change and used as input into the PANTHER classification system. The top 500 most variable genes were identified and used as input into principal component analysis (PCA) using the ‘factoextra’ R program (14) and hierarchical clustering using the ‘ComplexHeatmap’ R program with Euclidean distance and complete hierarchical clustering (15).

Pathway and gene set enrichment analysis

Gene set enrichment analysis (GSEA; v20.0.5) was performed using TCGA RSEM expression data and Pittsburgh CPM data in GenePattern with minimum and maximum gene-set sizes set to 5 and 1,500, respectively (16, 17). Single-sample GSEA version 10.0.3 (18) was implemented in GenePattern using log₂-transformed, median-centered expression data with rank normalization. The PANTHER classification system (PANTHER version 14.0; released 2018–12–03) was used for pathway analysis of the differentially-expressed genes comparing the Pittsburgh recurrent with primary tumors as described above (19–21). The PANTHER GO-Slim biological-process overrepresentation test was applied using the Homo sapiens, all genes, reference gene list with the Fisher exact test option and no correction. FDR q values from the PANTHER overrepresentation test were calculated using R software to correct for multiple-hypothesis testing.

The Database for Annotation, Visualization, and Integrated Discovery (DAVID) program (22, 23) was used for annotating the top 500 most variably expressed genes to infer functional programs differentially enriched among the primary and recurrent tumors. Parameters selected were as follows: Identifier = = ‘OFFICIAL_GENE_SYMBOL’; Species = = ‘Homo sapiens’; List Type = = ‘Gene List’; Agilent Background = = ‘Human Genome’. Default parameters were used for the functional annotation analyses.

Whole-exome sequencing analysis

Whole-exome sequencing (WES) data from Harbison and colleagues (5) were utilized to evaluate the genomic profiles of NFE2L2, KEAP1, and CUL3. WES data from 4 paired primary and recurrent tumors corresponding to the tumors in this study were available as well as 2 additional pairs from the University of Washington. Sequencing and alignment as well as somatic nucleotide variant, indel, and copy-number variant analyses were previously performed for these 6 paired primary- and recurrent-tumor samples as described in detail (5).

TCGA survival analysis

TCGA genomic data were downloaded from cBioPortal (24, 25). Classification of samples by genomic aberration status (i.e., mutation, copy-number variant, expression) was extracted from cBioPortal. For classification by gene-set cluster, TCGA RSEM normalized-expression data were clustered using the Consensus Cluster Plus algorithm to facilitate gene-expression cluster assignment (26). Kaplan–Meier survival functions were stratified either by genomic-aberration status or expression-cluster status and estimated using the Survminer package in R (27). In order to manage confounding, Cox proportional hazards models were fit controlling for age and tumor stage (28) and forest plots used to present the HRs.

Transcription factor/motif-activity analysis

To analyze activities of transcription-factor binding motifs (TFBM) using RNA sequencing (RNA-seq) data, we used the Integrated System for Motif Activity Response Analysis (ISMARA; ref. 29).

Cell lines and tumor samples

HPV-16–expressing HNSCC cell lines used in the present study were: UMSCC47 (isolated from the primary tumor of the lateral tongue of a male patient, established by Dr. Thomas Carey, University of Michigan), UPCI:SCC90 (from tongue squamous cell carcinoma (isolated by Dr. Susanne Gollin, University of Pittsburgh), and SiHa (squamous cell carcinoma isolated from the cervix uteri). UMSCC47, 93VU147T, and SiHa cells were maintained in high-glucose DMEM while UPCI:SCC90 was cultured in Eagle Minimum Essential Medium (EMEM) supplemented with 10% FBS and 1% penicillin/streptomycin mixture. Cells were used for collecting data for 10 passages after thawing and then discarded. All cell lines were authenticated using human cancer–cell line short tandem repeat (STR) profiles once a year. All cell lines were tested for mycoplasma contamination once every 6 months and were maintained at 37°C in a humidified atmosphere of 95% air and 5% CO₂. All human tumor samples were obtained from the University of Pittsburgh Medical Center in accordance with established University of Pittsburgh Institutional Review Board (IRB) guidelines.

Reagents and chemicals

DMEM and EMEM were purchased from GE Healthcare Life Sciences). FBS was obtained from Gemini Bioproducts. Penicillin/streptomycin antibiotic mixture was purchased from Invitrogen-Life Technologies (now part of Thermo Fisher Scientific). DMSO, N-acetyl-L-cysteine (NAC), rotenone, zVADfmk, and necrostatin-1 (Nec-1) were from Sigma-Aldrich. Cisplatin was purchased from EMD Millipore (now part of Millipore Sigma). IACS-010759 was purchased from Selleckchem. Antibody to tubulin was bought from Invitrogen.

Stock solution of each compound was stored at −20°C and diluted in culture media before use. Antibody against NRF2 (ab62352), HMOX1 (ab68477), and NQO1 (ab80588) were from Abcam. Beta-tubulin (MA5–16308) antibody was purchased from Invitrogen-Life Technologies and actin (MAB1501) was from EMD Millipore. Primers were bought from IDT Technologies. NRF2 endoribonuclease-prepared siRNA (esiRNA) was purchased from Sigma-Aldrich. Kits for RNA isolation and QIAshredder were purchased from Qiagen. iScript RT supermix for qRT-PCR and iQ SYBR Green Supermix for qPCR was from Bio-Rad Laboratories.

NRF2-overexpressing cell lines

The NRF2-expressing retroviral plasmid (MSCV-NRF2) was a kind gift from Dr. David Tuveson’s laboratory. UMSCC47 and UPCI:SCC90 cells were stably infected with retrovirus-expressing empty vector (MSCV-Neo) or NRF2 (MSCV-NRF2). Cells were selected in culture with media supplemented with 200 μg G418 for more than 4 weeks. NRF2 overexpression was confirmed by qPCR and Western blot.

NRF2 esiRNA transfection

Cells were plated in 6-well plates and grown to 70% confluency. Ten μM esiRNA was transfected with Lipofectamine RNAiMAX reagent (Invitrogen) according to manufacturer’s instructions before replating and analyzing for qPCR or cell proliferation.

Colony-formation assay

About 2 × 10³ cells were plated in 12-well plates and left in the incubator for 10 to 12 days for the development of colonies with or without indicated treatments. Media was changed twice a week and cultures were monitored for 7 to 14 days (depending on growth-rate differences) to allow for Neo cells to reach more than 50 cells per colony. They were fixed in 4% buffered formalin and stained with crystal violet. The plates were scanned and analyzed in Image J using the colony area plugin.

Western blotting

Cells were collected and lysed using appropriate amount of Tris–HCl/EDTA buffer supplemented with protease and phosphatase inhibitor. Lysates were incubated on ice for 15 to 20 minutes, sonicated, and spun down at maximum speed for 20 minutes. After protein quantification (Bradford method), lysates were prepared in β-ME and boiled. Between 50 to 80 μg protein was loaded in each lane. Tubulin or actin normalization was performed for each experiment. Immunoreactive bands were visualized and quantified using the LiCor Odyssey system.

Quantitative real-time PCR (qRT-PCR)

Total RNA from cells were isolated using RNeasy kit as described before (30). Tumors were homogenized in lysis buffer provided in the kit. The tissue lysate was loaded onto the QIAshredder homogenizer after which RNA was isolated using the manufacturer’s protocol. First-strand cDNA was synthesized using iScript RT and qPCR was performed. Relative quantification was performed using the 2^−ΔΔCq method. Primer sequences for HMOX1 and NQO1 are described in (30). NRF2; Forward 5′-CGG TAT GCA ACA GGA CAT TG-3′ and Reverse 5′-ACT GGT TGG GGT CTT CTG TG-3′.

Cell proliferation assay

Approximately 2.5 to 5 × 10⁴ cells/well were plated in 96-well plates and allowed to attach overnight. WST-1 assay (Takara Bio Inc., Clontech Laboratories, Inc.) was performed as described previously (30) according to manufacturer’s instructions.

Amplex red assay

UMSCC47 and UPCI:SCC90 cells engineered to overexpress NRF2 or control were used to determine the extracellular H₂O₂ using Amplex Red reagent (Molecular Probes Inc.) as previously described (30). Briefly, confluent cells attached in 96–black well plates were treated with NAC (20 mmol/L) for 2 to 3 hours. Cells were then incubated for 30 minutes in reaction buffer containing horseradish peroxidase and Amplex Red reagent. Fluorescence was read on Synergy H1 Hybrid microplate reader at absorption/emission maxima: 530/590 nm. Fluorescence was normalized with corresponding crystal violet readings to correct for variations in cell densities.

MitoSOX assay

Detection of mitochondrial superoxide in live cells was done using MitoSOX Red (Molecular Probes Inc.) as described previously (30). Florescence was read at absorption/emission maxima: 510/595 nm.

ATP production

The ATP-monitoring ATPlite luminescence assay system from Perkin Elmer Inc. was used for quantitative evaluation of proliferation in cultured UMSCC47 and UPCI:SCC90 cells. Briefly, 25 × 10³ cells/well were plated in 96–black well plates and allowed to attach overnight. After appropriate treatments, luminescence was read on a Synergy H1 Hybrid microplate reader (BioTek). ATP production was measured by ATPlite assay using standard curve of ATP and further normalization with cell count. Following treatments were used: IACS was applied at 10 nmol/L (24 hours), rotenone at 10 μmol/L (2 hours followed by rescue in regular DMEM media for 24 hours), and 20 μmol/L carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (FCCP; 30′).

Tetramethylrhodamine ethyl ester assay

The Tetramethylrhodamine ethyl ester (TMRE)-Mitochondrial Membrane Potential Assay Kit was used for quantifying changes in mitochondrial-membrane potential (MMP) in live cells using a Synergy H1 Hybrid microplate reader. The assay implements FCCP, an ionophore uncoupler of OXPHOS, as positive control. Adherent and live cells are stained for 15 to 20 minutes with or without FCCP, after which TMRE staining is measured by microplate spectrophotometry (excitation/emission 549/575 nm).

In vivo xenograft studies

The animal experiment protocol was approved by the Institutional Animal Care and Use Committee of the University of Pittsburgh. A total of 2 × 10⁶ UMSCC47-Neo and UMSCC47-NRF2 overexpressing cells mixed with matrigel were injected subcutaneously into either flank of NOD SCID mice aged 5 to 8 weeks (N = 6–8 tumors/group). Tumor volume was measured once a week using calipers and was calculated using the formula: $\frac{V=W^2\times L}{2}$ (31). Two-way ANOVA with Sidak test was used to test for differences in mean tumor volume between Neo and NRF2 groups over time.

In vivo PDX study

A written informed consent was obtained from the patient after obtaining IRB approval, in accordance with ethical guidelines prior to inclusion in the study. The tumor sample after surgical resection was transported immediately in RPMI media supplemented with penicillin/streptomycin. The tumor was implanted into the flank of 6-week-old female NOD/SCID mice under isoflurane anesthesia. The tumor was passaged once for the purpose of expansion before being used for this experiment.

The mice were randomized when tumor volume reached an average of 200 to 250 mm³ to be grouped into 4 treatments. Treatment drugs were prepared as follows: IACS was dissolved in 0.5% CMC-Na. Cisplatin was dissolved in PBS. The IACS group received 10 mg/kg body weight, orally, each day except days 5 to 7. The cisplatin (CDDP) group received 3 mg/kg body weight, i.p., on days 1, 6, 9, and 13, and 1 mg/kg body weight on day 4. Tumor volume and body weights of mice were recorded each day.

Relative tumor volume was calculated as fold change of control. Tumor-growth inhibition percentage was calculated using the following formula: [1 − (mean volume of treated tumors)/(mean volume of control tumors)] × 100%. The differences in relative tumor volume for each day was calculated using one-way ANOVA with Kruskal–Wallis test.

Statistics

R programming software (version 3.6.1) was used to perform statistical analyses (32). To account for multiple-hypothesis testing, the FDR was controlled using the method of Benjamini and Hochberg. In GSEA analyses, hypothesis generating q-values < 0.25 were used to determine if findings were statistically significant. In survival analyses, log-rank test P values less than 0.05 were used to determine if findings were statistically significant. All other analyses used an alpha of 0.05 to test for statistical significance. All in vitro experiments were repeated 2 to 3 times. Representative data from CFA are shown with quantitation from all experiments. Difference between group means was tested with Student t test. *, P value < 0.05; **, P value < 0.001; ***, P value < 0.0001.

Study approval

This study protocol was reviewed and approved by the University of Pittsburgh IRB (IRB 99–069). Written consent was obtained for genomic characterization of tumor tissues for all participants prior to inclusion in this study. This study abided by the Declaration of Helsinki principles.

Data availability

The raw and processed expression data have been deposited in the Gene Expression Omnibus (GEO; GSE165883).

Results

Patient characteristics

We identified 10 patients with early-stage (I/II) HPV-related oropharyngeal cancer. We obtained tissues from matched primary (pHNSCC) and recurrent tumors (rHNSCC). Our first aim was to identify transcriptional differences between the primary and recurrent tumors to gain mechanistic insight into the evolutionary adaptations of recurrent tumors (Fig. 1A). The median age of the cohort was 60 years old (Supplementary Table S1). Based on the American Joint Committee on Cancer (AJCC) 8^th edition staging manual (33), 7 of 10 patients presented with stage I disease and 3 of 10 with stage II disease. Of the 10 patients in the cohort, 6 were never-smokers, 3 patients had a greater than 10 years smoking history, and 1 patient had a 15-year history of chewing-tobacco use. Median overall survival was 36.4 months among the cohort.

Figure 1.

NRF2- and OXPHOS-signaling pathways are enriched in recurrent HPV-related HNSCC. A, RNA expression assays performed on 10 paired pHNSCC and rHNSCC. B, PANTHER pathway analysis of differentially-expressed genes comparing rHNSCC with pHNSCC. Pathways with over-representation test FDR q value < 0.05 are displayed. C, GSEA of rHNSCC versus pHNSCC. Normalized enrichment scores (NES) plotted. FDR q values labeled within the bars. Top, GSEA using panel of metabolic gene sets. Bottom, GSEA using panel of oncogenic gene sets. D, Left, Hierarchical clustering of hallmark OXPHOS gene set among pHNSCC and rHNSCC. Rows represent genes. Columns represent tumors. Row and column dendrograms demonstrate clustering results. Normalized expression counts in cells. Color bars in the upper plot illustrate ssGSEA enrichment score. DDR, Reactome p53-dependent DNA damage repair; FA, hallmark fatty-acid metabolism; MYC, hallmark MYC v1; NRF2 Core, NRF2 core signature (12). Right, NRF2 ssGSEA enrichment scores between groups.

NRF2- and OXPHOS-signaling pathways are enriched in recurrent HPV-related HNSCCs

PCA of the top 500 most variably expressed genes revealed clustering of the expression data primarily by tumor status (primary vs. recurrence; Supplementary Fig. S1A). Recurrent tumors included 2 local, 2 distant, and 6 regional recurrences which did not cluster independent of the site of recurrence (Supplementary Fig. S1B). Unsupervised hierarchical clustering of the top 500 most variably expressed genes demonstrated differential clustering between the recurrences and the primary tumors (Supplementary Fig. S1C). Using these 500 genes as input into DAVID, the cluster containing mostly recurrences was driven by genes involved in mitochondrial and OXPHOS whereas the cluster represented mainly by the primary tumors was notable for upregulation of immune- and mitogenic-signaling genes (e.g., MTOR, TRAF3, IL2RG; Supplementary Tables S2 and S3). Comparing the primary and recurrent tumors, there were 13 differentially-expressed genes (e.g., IGFBP3, CPS1, TNIP3, LTF, and DEFB4A) out of 13,793 genes assayed (Supplementary Tables S4 and S5). PANTHER pathway analysis of the differentially-expressed genes revealed upregulation of biological processes involved with biosynthesis, metabolism, OXPHOS, and Wnt signaling (FDR q < 0.05; Fig. 1B).

To further understand the mechanisms driving recurrent HNSCC at the pathway level, we performed GSEA utilizing prespecified metabolic- and oncogenic-gene sets (Supplementary Tables S6 and S7). This demonstrated selective enrichment of OXPHOS among the rHNSCCs (FDR q < 0.25; Fig. 1C, top). In the oncogenic GSEA, rHNSCCs were enriched in MYC, p53-dependent DNA-damage repair, and NRF2 core-signaling (34) pathways (FDR q < 0.25; Fig. 1C, bottom). We next performed semisupervised hierarchical clustering of expression data limited to genes of the hallmark OXPHOS gene set revealing two main clusters (Fig. 1D, left panel). One cluster included 8 of the 10 rHNSCCs while the second cluster was populated by 6 of the 10 pHNSCCs. To assess signaling pathways associated with OXPHOS gene expression in these data, we performed single-sample GSEA (ssGSEA) on the paired primary- and recurrent-tumor samples using oncogenic gene sets that were significantly enriched among the rHNSCCs (Fig. 1D, bars above main heatmap). We also performed ssGSEA using the Hallmark fatty-acid metabolism gene set which was enriched in the metabolic GSEA and Hallmark ROS hypothesizing that these pathways would be associated with NRF2 target-gene expression. We compared the enrichment scores between the primary and recurrent tumors for each gene set and observed significantly greater enrichment scores among the rHNSCCs for the NRF2, fatty-acid metabolism, and MYC gene sets (Wilcoxon rank-sum P < 0.05; Fig. 1D, right panel; Supplementary Fig. S2).

To further assess context-specific gene-regulatory networks that may be enriched or diminutive in rHNSCCs, we inferred transcription factor activities based on RNA-seq data (29). We observed a significant increase in E2F3 and SMAD3 activity among the rHNSCCs (Supplementary Fig. S3). BACH1/NFE2/NFE2L2 activity was also marginally increased in rHNSCCs (Wilcoxon rank-sum P = 0.062). PKNOX1/TGIF2 activity was significantly decreased among the rHNSCCs (Wilcoxon rank-sum, P = 0.006).

Lastly, to compare genomic aberrations in NRF2 signaling with RNA expression, we analyzed WES data from our prior work (5). Four pairs of matched primary and recurrent tumors from our prior work corresponded with 4 pairs of primary and recurrent tumors from the current study (UPHN4A/B, UPHN5A/B, UPHN22A/B, UPHN23A/B). We also included data from an additional 2 pairs of samples from the University of Washington that were analyzed in our prior work (Supplementary Fig. S4). NFE2L2 was amplified in the primary but not the recurrent tumor in 2 patients. Another patient had two different NFE2L2 missense mutations between the primary and recurrent tumor as well as a deletion at recurrence. One patient had NFE2L2 deletions in both the primary and recurrent tumor. KEAP1 was deleted in both the primary and recurrence of 3 patients, amplified in the recurrent tumor of 1 patient (in addition to a CUL3 deletion), and deleted in the recurrent but not primary tumor in 1 patient. Similarly, CUL3 was deleted in the recurrent but not primary tumor in 3 patients and deleted in both the primary and recurrent tumor in 2 patients. These data suggest that loss of NFE2L2 regulation by KEAP1 or CUL3 may contribute to NRF2 upregulation and more aggressive tumor features.

NRF2 activation portends worse survival among HPV-related HNSCCs

Given that several signaling pathways, including MYC and NRF2, were dysregulated in the rHNSCCs, we sought to investigate the extent of these pathways on patient survival in the TCGA HNSCC data set (7). With a set of 99 HPV-related TCGA HNSCC samples, we performed a survival analysis stratified by NRF2 or MYC genomic-alteration status which was defined as expression more than 2 (relative to diploid), mutation, and/or copy-number gain or amplification. Two of the 15 patients with NRF2 genomic alterations had likely oncogenic mutations (E79K and D457G; Supplementary Table S8). Interestingly, these 2 patients did not have evidence of copy-number alteration or increased expression while 8 of 15 patients had expression more than 2, 9 of 15 had copy-number gain or amplification, and 9 of 15 had both increased expression and copy-number gain or amplification (Supplementary Table S9). We identified a significant difference in survival among the NRF2-altered (NRF2 Up) samples (log-rank test, P = 0.005) versus no difference in survival among the MYC-altered (MYC Up) samples (log-rank test, P = 0.16; Fig. 2A). These data suggest that NRF2, but not MYC, may functionally drive survival in HPV-related HNSCC. Next, we sought to interrogate signaling network-level associations with survival among the TCGA HPV-related HNSCC samples.

Figure 2.

NRF2 activation decreases survival in TCGA HPV-related HNSCC. A, Survival stratified by NRF2 (top) or MYC (bottom) genomic-alteration status (Up: gene expression > 2, mutation, copy-number gain or amplification; Basal: gene expression ≤ 2, no mutation, copy neutral). Log-rank global P value labeled. B, Left, Survival stratified by NRF2 status as determined by clustering TCGA HPV-related HNSCC samples on Singh NFE2L2 gene-set expression. Log-rank P value labeled. Right, Singh NFE2L2 ssGSEA enrichment scores stratified by gene-set cluster membership as assigned in clustering analysis. C, Cox regression HR by NRF2 gene-set expression cluster, age group, and tumor stage (AJCC 8th edition, clinical staging). Point estimates plotted with 95% confidence intervals represented by whiskers. P values for each HR by predictor are represented on the far right of plot. D, Left, GSEA of TCGA HPV-related HNSCC NRF2-upregulated tumors as defined in A (top). FDR q values labeled within bars. Right, Hallmark OXPHOS gene-set enrichment plot.

We performed clustering analyses on the TCGA HPV-related HNSCC expression data using genes within gene sets of interest. The Consensus Cluster Plus algorithm was used and the resulting clusters were used for survival analysis. ssGSEA was performed to test gene-set enrichment between clusters. Using the Singh NFE2L2 targets gene set (35), we found a significant difference in survival with the NRF2-enriched cluster (NRF2 Up) conferring worse survival (Fig. 2B). NRF2 ssGSEA scores were compared between clusters confirming a significant difference gene-set enrichment between the clusters (Fig. 2B, right plot). When controlling for NRF2 enrichment cluster, age, and stage, NRF2 enrichment conferred a 5.13-fold (HR 95% confidence interval: 2.02–13.0; P < 0.001) increased risk of mortality (Fig. 2C). NRF2-enriched TCGA HPV-related HNSCCs featured enrichment in MYC, MTOR, fatty-acid metabolism, OXPHOS, ROS, peroxisome, and glycolysis gene sets suggesting a relationship between NRF2 and critical metabolic-signaling pathways (Fig. 2D).

To extend these observations, we tested for an independent or synergistic effect of NRF2 transcriptional targets and other critical metabolic regulatory genes on survival. We focused on HMOX1, NQO1, and SQSTM1. SQSTM1 is an inactivator of the KEAP1 complex and creates a positive feedback loop by inducing antioxidant response-element gene transcription (36). HMOX1 and NQO1 were included as canonical transcription targets of NRF2 in the antioxidant response-element pathway that have unique functions converging on redox regulation (34). We performed a survival analysis among the TCGA HPV-related HNSCCs stratified by NRF2 genomic-alteration status in combination with HMOX1, NQO1, or SQSTM1 genomic alterations (defined by gene expression > 2, copy-number gain, or amplification). Interestingly, NRF2 and HMOX1 double-altered tumors had the worst survival among the NRF2 ± HMOX1 alteration group while NRF2-altered ± NQO1- or SQSTM1-altered tumors were note associated with poor survival (Supplementary Fig. S5). We also tested if HMOX1, NQO1, or SQSTM1 upregulation had an independent association with survival among the TCGA HPV-related HNSCCs. Only tumors with upregulation of HMOX1 had stastically significantly worse survival compared with those without HMOX1 upregulation (Supplementary Fig. S6). Taken together, these data implicate OXPHOS as a driver of HPV-related HNSCC and suggest that recurrent/metastatic tumors coopt this pathway during disease progression.

NRF2 promotes cell proliferation in HPV-positive cell lines

Next, we sought to investigate the effect of NRF2 overexpression on growth in HPV-related cell lines. Thus, we used both HPV-related head and neck and cervical cancer cell lines. First, we established stable NRF2 overexpression in HPV16-harboring UMSCC47 and UPCI: SCC90 cell lines. NRF2 overexpression induced cell proliferation in these cells (Fig. 3A–C). In order to investigate the generalizability of our results to HPV-associated carcinomas, we used the HPV-driven cervical cancer cell line SiHa, which is known to express detectable levels of NRF2 (37). We used siRNA targeting NRF2 in these cells and another HPV-driven HNSCC 93VU147T and evaluated the effect on cell proliferation. Knockdown of NRF2 reduced cell proliferation by approximately 50% (Fig. 3D and E).

Figure 3.

NRF2 promotes cell proliferation in OXPHOS-dependent manner in HPV-positive HNSCC cells. A and B, qPCR (A) and Western blot (B) demonstrate overexpression of NRF2 after retroviral stable infection. C, Representative image (top) of colony-formation assay (CFA) and quantification (bottom) in stably-expressing empty vector (Neo) or NRF2 cells. D and E, Representative picture (top) of CFA and quantification (bottom) in SiHa (D) and 93VU147T (E) after transient NRF2 knockdown. F, Cell-proliferation assay for IACS-treated UMSCC47-Neo and -NRF2 cells measured after 24 hours of treatment (representative data from 3 experiments). G, CFA after treatment with IACS in UPCI:SCC90-Neo and -NRF2 cells. H, Cell proliferation in UPCI:SCC90-Neo and -NRF2 cells after treatment with 0.1 or 10 μmol/L rotenone. Results represent mean ± SEM.

Our transcriptional analyses suggested a predilection for OXPHOS gene expression in tumors with poor survival characteristics. We therefore postulated that inhibition of OXPHOS would be preferentially toxic to NRF2-dependent cells. Therefore, we tested IACS-010759 (IACS), a potent and selective electron transport chain complex I inhibitor which is currently in phase I trials. As expected, IACS was significantly more cytotoxic to cells overexpressing NRF2 as compared with control cells (Fig. 3F and G; Supplementary Fig. S7A and S7B).

To further evaluate the role of OXPHOS in cellular proliferation, we tested the effect of rotenone (another complex I inhibitor and mitochondrial toxin) and observed similar effects. Rotenone treatment had a potent cytotoxic effect in NRF2-overexpressing, but not in control cells (approximately 30% reduction, P < 0.0001; Fig. 3H). To further evaluate the role of OXPHOS in cellular proliferation, we tested the effect of rotenone via CFA in UPCI:SCC90-Neo and -NRF2 cells (Supplementary Fig. S7C). Rescuing the cells from rotenone after 2 hours of treatment was irreversible in the NRF2-dependent cells and the cells failed to form colonies after 10 days in culture. Taken together, these data suggest that NRF2 promotes cell proliferation in an OXPHOS-dependent manner in HPV-related HNSCC.

NRF2-overexpressing xenografts exhibit enhanced growth

The in vivo effect of UMSCC47 stably expressing NRF2 and Neo was assessed in a NOD/SCID mouse model. Cells were injected into both flanks and allowed to establish for a week. The mean tumor weight and volume increased significantly over time in the mice implanted with NRF2cells relative to Neo.At day 21 post implantation, thetumor length and width had reached more than 20 mm in the NRF2 xenografts. The NRF2-expressing tumors had 3-fold greater tumor weight at time of harvest(P < 0.05, n = 3 or4;Fig. 4A and B). The tumorvolume of NRF2-overexpressing cells amplified by 3.5-fold at day 7 (P < 0.05), 6-fold at day 14, and 14-fold at day 21 (P < 0.0001; Fig. 4C). These data advocate that NRF2-expressing tumors have enhanced proliferative advantage over the low-NRF2–expressing tumors in vivo.

Figure 4.

NRF2-overexpressing xenografts exhibit enhanced growth. Bilateral flanks of NOD/SCID mice were injected subcutaneously with UMSCC47-Neo and -NRF2 cells and tumor volumes were measured weekly. A, Images of harvested UMSCC47 tumors are demonstrated for each group. B, Tumor weights collected on day 21 postimplantation. Difference in mean tumor weights between groups compared using Student t test. C, UMSCC47 tumor growth over time. Data on graph represent mean ± SEM.

NRF2 activation upregulates mitochondrial respiration in a ROS-dependent manner

In order to gain insight into how NRF2 modulates redox balance in HPV-positive cell lines, we measured ROS produced by mitochondria in the context of NRF2 expression. To quantify relative ROS production in NRF2-enriched cells, we measured hydrogen peroxide (H₂O₂) and superoxide (O2^•−) formation in intact and live cells. We used the free radical scavenger NAC to test the effect of abrogating ROS in these cells. We began by measuring H₂O₂ production SiHa and 93VU147T cells in the context of NRF2 knockdown using siNRF2. We found that siNRF2 treatment abrogated ROS by approximately 30% and approximately 20% in SiHa and 93VU147T cells, respectively. NAC treatment reduced H₂O₂ levels by a similar amount (Fig. 5A and B). Next, we measured markers of ROS in control and NRF2-over-expressing conditions. We hypothesized that NRF2 overexpression would abrogate oxidative stress. Indeed, NRF2-overexpressing cells demonstrated increased expression of the cytoprotective and detoxifying genes, heme oxygenase 1 (HMOX1) and NAD(P)H:quinone oxidoreductase-1 (NQO1) by 3- to 5-fold (Fig. 5C). Overexpression of HMOX1 and NQO1 were also confirmed in a NRF2-enriched HPV-positive primary HNSCC cell line (UPCI:UDSCC17–70; Supplementary Fig. S7D).

Figure 5.

NRF2 activation upregulates mitochondrial activity in a ROS-dependent manner. A and B, ROS (H₂O₂ released) measured in SiHa (A) and 93VU147T (B) cells treated with and without 10 mmol/L NAC for 3 hours after transient NRF2 knockdown. C, qPCR for HMOX1 and NQO1 in Neo and NRF2-expressing UMSCC47 cells. D, ROS (H₂O₂ released) measured in cells treated with and without 10 mmol/L NAC for 3 hours. E, Mitochondrial superoxide (MitoSOX) measured in control versus 10 mmol/L NAC (3 hours)-treated cells via MitoSOX Red. F, NRF2 promotes ATP production and is dependent on mitochondria: NRF2-driven ATP production is abrogated upon treatment with mitochondrial inhibitors. G, MMP measured using TMRE dye. H, NRF2 drives ATP production via ROS production. I, TMRE fluorescence measured in cells treated with and without 10 mmol/L NAC (3 hours).

Surprisingly, NRF2 expression increased the basal rate of ROS production by approximately 2-fold (Fig. 5D). Moreover, NAC mitigated ROS production to a greater extent in the context of NRF2 overexpression (by approximately 70%). However, NAC did not significantly decrease ROS in control cells (Fig. 5D).

Next, we measured superoxide levels (as measured by MitoSOX) to quantify ROS produced by mitochondria in the context of NRF2 expression. We found that NRF2 overexpression increased mitochondrial superoxide levels by approximately 50% (Fig. 5E). There was a significant reduction in MitoSOX signal in the NRF2-overexpressing cells treated with NAC (approximately 50%), but no significant changes were noted in control cells (Fig. 5E).

To further understand the impact of NRF2-mediated OXPHOS reprogramming, we assessed mitochondrial function by measuring ATP production and MMP. NRF2-overexpressing UMSCC47 cells exhibited approximately 1.7-fold higher ATP production which was mitigated after treatment with rotenone (P < 0.001), IACS (approximately 65% decrease), and FCCP (approximately 40% decrease; Fig. 5E). Since MMP is an essential component of energy storage during OXPHOS, we measured MMP in live cells using TMRE dye. NRF2 overexpression increased MMP by approximately 35%, but treatment with rotenone or IACS abrogated this effect (Fig. 5F). Similarly, FCCP significantly decreased MMP in the NRF2 cells (Fig. 5F). In contrast, ATP production and MMP in the control cells was not significantly affected by IACS, rotenone, or FCCP treatment.

To elucidate whether NRF2 affects mitochondrial respiration in a ROS-dependent manner, we tested the effect of a free radical scavenger, NAC, on ATP production and MMP. The increase in ATP noted upon NRF2 overexpression was abrogated by NAC treatment, returning ATP content to the level observed in control cells. In contrast, NAC had no effect on ATP levels in control (Fig. 5G and H). We confirmed these findings in the UPCI:SCC90 cell line engineered to stably overexpress NRF2 (Supplementary Fig. S8). Overall, these data suggest that increased ROS in the context of NRF2 overexpression is not merely a collateral phenomenon but may represent a positive feedback mechanism promoting OXPHOS.

OXPHOS inhibition synergizes with cisplatin to enhance cytotoxicity in NRF2-overexpressing cancer cells

Prior studies have demonstrated that NRF2 can mediate resistance to cisplatin (38–41). Since cisplatin is commonly used in the treatment of HPV-related HNSCC, we investigated the effect of NRF2 on cisplatin sensitivity. Genetic knockdown of NRF2 in SiHa cells led to increased sensitivity to cisplatin treatment (Fig. 6A). We confirmed this data in UMSCC47-Neo and NRF2 cells where NRF2 cells demonstrated 1.7-fold inflated resistance to cisplatin compared with Neo cells (Fig. 6B).

Figure 6.

NRF2 status mediates cisplatin sensitivity in cooperation with OXPHOS inhibition in HPV-positive HNSCC cells. A and B, IC₅₀ for CDDP measured in (A) SiHa after transient knockdown of NRF2 and (B) UMSCC47-Neo and –NRF2 cells. C, qPCR for NRF2 to demonstrate knockdown in primary cell line UPCI:UDSCC17–70. D, IC₅₀ for CDDP in UPCI:UDSCC17–70 cells in control and NRF2-knockdown conditions. E, Complex I inhibition is synthetically lethal with cisplatin in NRF2-overexpressing cells. UMSCC47-Neo and -NRF2 cells were treated with 10 μmol/L CDDP, 100 nmol/L IACS, or combination for 72 hours. F and G, Cell proliferation measured after 1 hour pretreatment with apoptotic inhibitor z-VADfmk (10 μmol/L; F) and 20 μmol/L Nec-1 (G). Results represent mean ± SEM.

Similar results were obtained using a primary cancer–cell line (UPCI:UDSCC17–70) harboring endogenously-elevated NRF2 expression (Fig. 6C). In these primary cells, NRF2 knockdown led to a 2.2-fold decrease in the IC₅₀ of cisplatin (Fig. 6D). As cisplatin induces oxidative stress, we hypothesized that cotreatment with IACS would be synergistically cytotoxic, especially in the context of NRF2 overexpression. To test this hypothesis, we treated UMSCC47-control and NRF2-overexpressing cells with cisplatin alone (CDDP) or in combination with IACS. IACS and CDDP yielded synergistic cell death in the context of NRF2 expression (Fig. 6E). To elucidate the mechanism of cell death induced by the dual combination of cisplatin and IACS, we treated the cells with standard pan-caspase inhibitor z-VADfmk and necrotic cell-death inhibitor Nec-1. Both these drugs contributed to partial recovery from cell death induced by cisplatin+IACS (Fig. 6F and G), indicating that there may be other mechanisms contributing to synergistic cell killing. Taken together, these data indicate that NRF2 overexpression mediates a delicate balance between ROS and antioxidant production through OXPHOS reprogramming suggesting a targetable weakness in NRF2-enriched head and neck cancer.

In vivo growth inhibition in HPV-related HNSCC PDX mice treated with a CDDP and IACS combination

The overexpression of NRF2 in UMSCC47 cells increases the proliferative advantage in xenograft studies. Furthermore, we also show synergistic interaction between cisplatin and IACS in context of NRF2. To address the translational impact of these data in vivo, we used an HPV-related HNSCC PDX model expressing high NRF2 (Supplementary Fig. S9B and S9C) derived from a patient with HPV-related HNSCC. The patient had experienced multiple recurrences and the cancer was refractory to a multitude of cytotoxic and immune-modulating therapies. Patient characteristics are provided in Supplementary Fig. S9A. Mice bearing this PDX were treated with cisplatin, IACS, and cisplatin + IACS for 15 days. As single agents, cisplatin and IACS significantly reduced PDX tumor growth over time, compared with control (Fig. 7A and B). IACS was significantly more effective as a single agent (mean TGI approximately 56%) compared with cisplatin (mean TGI approximately 11%). The IACS-alone–treated mice showed stasis after an initial dip in the tumor volume (Fig. 7A). The combination group was significantly different (P = 0.01; one-way ANOVA using Kruskal–Wallis test) from the IACS-treated group from day 13 through the end of the treatment. The cisplatin-alone– and the combination-treated mice showed toxicity throughout the treatment period as evident from their body weights (Fig. 7C). The experiment was culled after 15 treatment days because the mice in the cisplatin and the combination groups were apparently sick, probably caused by collateral toxicity from cisplatin. The IACS-treated mice did not show visible sign of distress. A summary of the treatment schedule and relative tumor volume on day 15 is given in Fig. 7D.

Figure 7.

In vivo tumor inhibition in mice bearing HPV-positive HNSCC PDX treated with CDDP and IACS. A, Relative tumor volume for each treatment day. Data shown is mean ± SEM for control (n = 8), CDDP (n = 11), IACS-010759 (n = 10), and CDDP + IACS-10759 (n = 14). B, Representative tumor images at the end of treatment. C, Body weights for each treatment group. D, Table presenting antitumor activity on day 15 of treatment. wt, weight.

Discussion

Efforts to deescalate treatment for patients with HPV-associated HNSCC have been hampered by a lack of knowledge of the molecular determinants of poor outcome. This study sought to address this gap in knowledge. We find that gene-expression analyses of paired recurrent versus primary HPV-related HNSCCs demonstrated enrichment of the NRF2- and OXPHOS-signaling pathways amidst a milieu of metabolic gene dysregulation specific to the recurrences. These data allowed us to identify the complex I inhibitor IACS as a potential therapeutic agent for these patients. Interestingly, we found translational potential in NRF2-mediated synergy between cisplatin and OXPHOS inhibition. Thus, targeting OXPHOS in NRF2-mediated recurrent HPV-related HNSCC in conjunction with cisplatin therapy may provide a synergistic precision target for improving oncologic outcomes in this devastating disease.

In this study, we identified gene-expression enrichment of NRF2, OXPHOS, MYC, fatty-acid metabolism, and DNA damage-response signaling pathways among recurrent HPV-related HNSCCs. MYC activation can reprogram cancer-cell metabolism by activating genes involved with glycolysis, glutaminolysis, and mitochondrial biogenesis (42). Moreover, a substantial body of evidence illustrates the effect of NRF2 in driving MMP, mitochondrial biogenesis, fatty-acid oxidation, and OXPHOS (43–46). This led us to query existing datasets for a relationship between these pathways and overall survival. The NRF2-related gene-expression signature was associated with poor survival among the TCGA HNSCC data and in multiple non–small cell lung cancer cohorts (47, 48). Our analysis of the TCGA HPV-related HNSCC data recapitulated and expanded upon these findings, in that we observed NRF2 dysregulation at the genomic and pathway levels correlated with overall survival. In contrast, we did not identify a survival difference among the TCGA HPV-related HNSCCs when stratified by MYC gene dysregulation, though at the MYC-pathway level, there was an association with pathway enrichment and worse survival. Lastly, GSEA of the NRF2-upregulated TCGA HPV-related HNSCCs demonstrated enrichment in OXPHOS signaling, glycolysis, fatty-acid metabolism, and MYC-, MTOR-, and ROS-signaling pathways suggesting an interaction among these pathways in NRF2-altered tumors that ultimately converge on increased tumor-cell fitness and worse overall patient survival. Taken together, our data implicate NRF2 as a critical ‘lynch-pin’–signaling molecule in HPV-associated HNSCC. We therefore focused our subsequent efforts to further elucidate the role of NRF2 in the context of tumor recurrence.

NRF2 is a biological double-edged sword acting both as a tumor suppressor and oncogene in part by regulating redox homeostasis which is a delicate balance. NRF2 functions as a transcriptional regulator of antioxidant response-element genes and modulates mitochondrial function and OXPHOS efficiency abrogating the effect of ROS (46, 49). NRF2 impacts OXPHOS efficiency (46) by affecting basal MMP, basal ATP levels, and oxygen-consumption rates (44) while knockdown of NRF2 decreases oxygen consumption and ATP production in cancer cells (50). NRF2 provides substrate for OXPHOS and regulates the expression of complex IV cytochrome c oxidase subunits as well as nuclear respiratory factor 1 which regulates the expression of respiratory complexes (44, 51–56).

Activation of the NRF2 pathway induces expression of proteins involved in xenobiotic metabolism and clearance, inhibition of inflammation, repair and removal of damaged proteins, as well as transcription and activation of growth factors (57), thus permitting cells to acquire features for therapeutic resistance. Upregulation of the NRF2-mediated survival pathway protects tumor cells from chemotherapeutic agents including etoposide, doxorubicin, and cisplatin (38, 40, 58, 59). Constitutive activation of NRF2 accelerates recurrence and induces metabolic reprogramming to reestablish redox homeostasis and upregulate de novo nucleotide synthesis in breast cancer cells (34). Ultimately this acquired NRF2 activation confers lowered cancer-therapeutic efficacy and materialization of therapeutic resistance. Despite prior literature, little is known about the role of NRF2 in the context of HPV-associated HNSCC.

Our in vitro and in vivo work describe for the first time that NRF2 drives the proliferation of HPV-related cancers. Prior studies in other model systems have found similar results. For example, Keap^−/− cells proliferate faster than parental cells while NRF2^−/− cells proliferate more slowly (60, 61) modulated by variation in growth factors (57). Oncogenic proteins that regulate proliferation such as KRAS, BRAF, and MYC increase the expression of NRF2 (62, 63), which corroborates our findings.

Mitochondrial respiration depends on NRF2 activity. We found that complex I inhibitors IACS and rotenone decrease ATP production in NRF2-overexpressing cells indicating a role for NRF2 in mitochondrial respiration. These OXPHOS inhibitors also decreased cellular proliferation in a NRF2-mediated manner indicating a NRF2-dependent relationship between OXPHOS and the cell cycle. This suggests that in context of NRF2 overexpression, glucose consumption, and energy demand, needs are met in part through upregulating OXPHOS and likely other components of central–carbon signaling. Abundance of NRF2 also led to increased ROS production which may independently trigger cellular proliferation (64–66). Inhibition of mitochondrial respiration with NAC suggests that ROS is not only a byproduct of OXPHOS but also acts in a positive feedback mechanism to regulate OXPHOS.

This is the first report identifying a targetable role for OXPHOS inhibition in NRF2-driven head and neck cancer. Several studies have illustrated a role for either OXPHOS or NRF2 in tumorigenesis. NRF2 dysregulation in an esophageal cancer model demonstrated an association with increased cell proliferation and altered metabolism (67). Prior work analyzing the HPV-host protein network identified an HPV E1-KEAP1 interaction that phenocopies inactivating mutations in the KEAP1-NRF2 pathway leading to the expression of cytoprotective genes (9). However, OXPHOS can drive treatment resistance independent of NRF2; OXPHOS promotes platinum-based chemoresistance in ovarian cancer models (68). Genomic analysis of mantle cell lymphoma found that metabolic reprogramming towards OXPHOS and glutaminolysis is associated with resistance to ibrutinib, and that resistance can be overcome with OXPHOS inhibition using IACS (69). Complex I inhibition via IACS also delays regrowth of neoadjuvant chemotherapy-resistant triple-negative breast cancer (70). This data suggests that IACS may potentiate the effects of cytotoxic chemotherapy. Taken together, there is much to be understood regarding the interaction of NRF2 and OXPHOS in the overall metabolic signaling milieu and their role in treatment resistance. Our study illustrates NRF2-dependent sensitization to OXPHOS inhibition.

Limitations of the current study include the impact of tumor heterogeneity on our expression results. Head and neck cancers harbor genomic heterogeneity accounting in part for their ability to resist our current treatment regimens (71). It is possible that other genomic, epigenomic, metabolic, immune dysregulation, or virally-mediated (9) mechanisms are driving recurrence aside from those identified in our differential expression and GSEA and that we are capturing a subset of highly-expressed genes in certain tumor clones. For example, the phosphoinositide 3-kinase and AKT-signaling pathway is frequently enriched in HPV-related HNSCCs and regulates redox metabolism in cancer (72, 73). Evidence from the TCGA HNSCC dataset demonstrate an association between NRF2 dysregulation and survival lending support to the role of NRF2 in recurrence. Moreover, our bulk RNA-seq approach limits the conclusions that can be drawn regarding the contribution of tumor-intrinsic versus stromal contributions to the expression findings observed in this study. To mitigate this limitation, we curated tumor samples by histologically identifying tumor-rich portions of patient samples with at least 75% tumor and using this to guide excision of tumor-enriched regions. It is notable that the majority of primary tumors cluster separate from recurrent tumors as opposed to primary and recurrent tumor pairs clustering together. This further suggests that the observed signatures may reflect tumor-intrinsic expression patterns rather than stromal-expression patterns, although single-cell RNA-seq would yield further insight. Lastly, there is potential for misclassification of HPV status given that we used p16 status as a surrogate marker for HPV-driven disease. Further work will aim to increase our understanding of mechanisms underlying NRF2-dependent OXPHOS-inhibitor sensitivity and evaluate the effects of metabolic adaptations on the tumor microenvironment in driving treatment resistance.

In summary, we observed that recurrent HPV-related HNSCCs are enriched in NRF2 and OXPHOS-signaling dysregulation in an overall metabolic dysregulated milieu and that NRF2-enriched HPV-related HNSCC displays increased sensitivity to OXPHOS inhibition. Treatment options for recurrent HPV-related HNSCC are limited. IACS is currently under investigation in a phase I clinical trial for metastatic or unresectable malignancies (74). Our data may be used to support the rational use of OXPHOS inhibitors in patients with NRF2-enriched, recurrent HPV-related HNSCC cancer in a clinical trial providing expanded options for patients with this devastating disease.

Translational Relevance.

The human papillomaviruses (HPV) are thought to be the etiologic agents for the majority of oropharyngeal cancers. Recent advances in treatment have significantly improved the outcomes for these patients. Despite this, up to 10% of patients experience tumor recurrence. Improved understanding of the genomic characteristics of recurrent HPV-associated cancers may help us to define better strategies to treat these patients. Here, we show that genes in the oxidative phosphorylation (OXPHOS) pathways are highly expressed in recurrent cancers. Further, we find that NRF2 plays a critical role in regulating OXPHOS in vitro and promotes tumor growth in a murine xenograft model. Cisplatin treatment had an additive effect on cell death when combined with a translationally relevant OXPHOS inhibitor. Taken together, these data provide a scientific rationale to evaluate combination therapy with cisplatin and OXPHOS inhibition in HPV-associated oropharyngeal cancer.