Abstract
Tumor hypoxia fuels an aggressive tumor phenotype and confers resistance to anticancer treatments. We conducted a clinical trial to determine whether the antimalarial drug atovaquone, a known mitochondrial inhibitor, reduces hypoxia in non–small cell lung cancer (NSCLC).
Patients with NSCLC scheduled for surgery were recruited sequentially into two cohorts: cohort 1 received oral atovaquone at the standard clinical dose of 750 mg twice daily, while cohort 2 did not. Primary imaging endpoint was change in tumor hypoxic volume (HV) measured by hypoxia PET-CT. Intercohort comparison of hypoxia gene expression signatures using RNA sequencing from resected tumors was performed.
Thirty patients were evaluable for hypoxia PET-CT analysis, 15 per cohort. Median treatment duration was 12 days. Eleven (73.3%) atovaquone-treated patients had meaningful HV reduction, with median change −28% [95% confidence interval (CI), −58.2 to −4.4]. In contrast, median change in untreated patients was +15.5% (95% CI, −6.5 to 35.5). Linear regression estimated the expected mean HV was 55% (95% CI, 24%–74%) lower in cohort 1 compared with cohort 2 (P = 0.004), adjusting for cohort, tumor volume, and baseline HV. A key pharmacodynamics endpoint was reduction in hypoxia-regulated genes, which were significantly downregulated in atovaquone-treated tumors. Data from multiple additional measures of tumor hypoxia and perfusion are presented. No atovaquone-related adverse events were reported.
This is the first clinical evidence that targeting tumor mitochondrial metabolism can reduce hypoxia and produce relevant antitumor effects at the mRNA level. Repurposing atovaquone for this purpose may improve treatment outcomes for NSCLC.
Translational Relevance
The antimalarial drug atovaquone, a known inhibitor of mitochondrial oxygen consumption by targeting oxidative phosphorylation, rapidly reduces tumor hypoxia in patients with non–small cell lung cancer (NSCLC). This is the first clinical report that targeting mitochondrial metabolism is an effective strategy to alter the tumor microenvironment in this way. As tumor hypoxia is a well-recognized barrier to radiation, chemotherapy, and immunotherapy, combining this agent with such treatments may improve outcomes for many patients. In NSCLC, this is particularly important as novel radiosensitizers are needed to improve radiation outcomes. Atovaquone reduces oxygen consumption in many tumor cell lines, therefore hypoxia modification is expected in patients with various tumor types. As atovaquone is FDA approved and well tolerated, repurposing this agent as an anticancer treatment has the potential for rapid adoption into clinical practice. Utilizing established, as well as putative tumor hypoxia measures, findings presented here contribute to the development of clinically useful biomarkers to enable patient selection for hypoxia-reducing therapies.
Introduction
In contrast to normal tissues, hypoxia is a common finding in solid tumors resulting from imbalance between high oxygen demand and poor delivery (1). Principally, through hypoxia-inducible factor signaling, hypoxia contributes significantly to tumorigenesis by directly contributing to the majority of cancer hallmarks, including sustained proliferation, immune system evasion, invasion, and metastasis (2–4). Consequently, hypoxia results in a highly aggressive tumor phenotype.
Of further clinical relevance is that tumor hypoxia is a significant barrier to numerous anticancer treatments. In particular, hypoxia confers profound resistance to radiation treatment with up to three times the dose of radiation required to treat hypoxic tumors successfully (5), and is associated with inferior clinical outcomes following radiotherapy (6, 7). Hypoxia-related resistance to cytotoxic chemotherapy and immunotherapy is also well recognized (8, 9).
An emerging novel strategy to tackle hypoxia is to reduce tumor oxygen consumption rate (OCR) through metabolic reprogramming (10). Recently, the commonly prescribed and well-tolerated antimalarial drug, atovaquone, has been discovered as a highly promising agent for this purpose. We reported previously that atovaquone reduces OCR in numerous cancer cell lines and alleviates hypoxia in xenograft models leading to tumor radiosensitization (11). Reduction in OCR occurs by inhibiting oxidative phosphorylation (OXPHOS) at complex III of the mitochondrial electron transport chain (ETC). Interestingly, besides causing hypoxia through oxygen consumption, tumor mitochondria are increasingly recognized as important mediators of drug resistance, invasion, and metastasis (12).
However, demonstration that mitochondrial function can be modified in tumors using mitochondrial inhibitors has never been shown previously in the clinical setting. We therefore conducted a translational clinical trial to determine whether by using the pharmacodynamics endpoints of tumor hypoxia and gene expression analysis, we could prove the concept and effectiveness of atovaquone as a clinically useful mitochondrial inhibitor. Patients with non–small cell lung cancer (NSCLC) were chosen for this study given the urgent clinical need to improve treatment outcomes for this patient group, especially with regards to radiotherapy.
Noninvasive PET-CT imaging of tumor hypoxia is reliable and the most accepted technique for measuring hypoxic change in patients (13–18), and was the primary method used. The most common radiolabeled tracers used for this purpose are nitroimidazole-based compounds, such as [18F]-fluoromisonidazole (FMISO) and [18F]-fluoroazomycin arabinoside (FAZA), which under low oxygen tensions undergo irreversible intracellular enzymatic reduction. By utilizing such scans in this study, assessment of hypoxia in viable and, therefore, clinically relevant tumor regions was ensured.
Gene expression signatures measure the complex tumor transcriptome response to hypoxia (19) and were considered a key endpoint to demonstrate that reduction of hypoxia had a desired biological effect. Because development of validated and deliverable clinical hypoxia biomarkers is needed, putative circulating and tumor sample–based measures of hypoxia were also assessed. Assessment of changes in tumor perfusion and vasculature was also performed, thus ensuring a comprehensive assessment of tumor oxygenation.
We report the first clinical evidence that hypoxia can be reduced through targeting tumor metabolism and highlight atovaquone as a highly promising agent for combination with anticancer treatments in NSCLC.
Patients and Methods
Study design, patients, and treatment
This was an open-label, nonrandomized, two-cohort, window-of-opportunity study (NCT02628080) conducted at the Oxford Cancer and Haematology Centre (Oxford, England, United Kingdom). Ethical approval was obtained from National Research Ethics Service Committee South Central Oxford B (16/SC/0012). Trial conduct adhered to all regulatory requirements and was in full accordance with the provisions of the Declaration of Helsinki and Good Clinical Practice guidelines.
Eligible patients were aged ≥18 years, had a pathologic or radiological diagnosis of NSCLC, were scheduled for surgical resection, had disease >2 cm in diameter, and had Eastern Cooperative Oncology Group (ECOG) performance status 0–2. Patients were excluded if taking known ETC inhibitors. Complete eligibility criteria are provided in the trial protocol (Supplementary Materials and Methods). All patients provided written informed consent.
The trial was designed to recruit 30 patients sequentially into two equal-sized cohorts: cohort 1 received atovaquone and cohort 2 did not. All patients attended for two visits for evaluation of tumor hypoxia and clinical assessment. For cohort 1, oral atovaquone (Wellvone, 750 mg/5 mL micronized suspension, GlaxoSmithKline) was self-administered at a standard clinical dose of 750 mg twice daily for 7–16 days between visits, as permitted by the surgical date. Patients were asked to take atovaquone together with fat-containing food to aid absorption. Shortly following the second visit, all patients underwent surgery with 500 mg/m2 oral pimonidazole hydrochloride (Hydroxyprobe-1, Natural Pharmacia International) administered 16–24 hours before as an exogenous marker of hypoxia.
This study utilized multiple imaging and sample-based methods to assess tumor oxygenation with numerous prespecified primary, secondary, and exploratory endpoints. All endpoints are listed in the trial protocol and the trial schema is shown in Supplementary Fig. S1 (Supplementary Materials and Methods).
Tumor imaging outcomes
The primary imaging-related trial endpoint was change in tumor hypoxic volume (HV) measured using hypoxia PET-CT. Patients were imaged using GE Discovery 690 or 710 PET-CT Scanners (GE Healthcare). An activity of 370 MBq of FMISO (University of Cambridge, Cambridge, England, United Kingdom) or FAZA (University of Manchester, Manchester, England, United Kingdom) was injected and immediately a 45-minute dynamic PET acquisition and then 10-minute acquisitions at 2 and 4 hours postinjection were obtained. CT was performed for localization and attenuation correction. The same tracer and scanner were mandated for both study visits per individual patient.
Four-hour hypoxia PET-CT images were analyzed using Hermes Hybrid Viewer Software (Hermes Medical Solutions AB). Tumors were outlined by an experienced PET radiologist (R.E. Macpherson). Blood in the descending aorta was outlined to provide the background mean standardized uptake value (SUVmean). To determine the HV, voxel-by-voxel SUVs were divided by background SUVmean to provide tumor-to-blood ratio (TBR) values for each tumor voxel and TBR ≥ 1.4 voxels were classified as hypoxic, as described previously (15). Volumes of hypoxic voxels pre- and post-atovaquone were compared and ≥10% reduction was defined as a meaningful reduction, as published previously (ref. 14; further explained in Supplementary Materials and Methods).
Secondary imaging endpoints were change in tumor perfusion using dynamic hypoxia PET, perfusion CT (pCT), and dynamic contrast-enhanced (DCE) MRI. Non-prespecified T2* map and T1 shMOLLI MRI analysis were also performed to investigate changes in tumor venous oxygenation and biochemistry, respectively. Methodology for all these imaging techniques is provided in Supplementary Materials and Methods.
Tumor sample and circulating marker outcomes
Resected tumors were formalin-fixed and paraffin-embedded (FFPE) using a standard clinical protocol. Intercohort comparison of IHC for the exogenous hypoxia marker pimonidazole was a coprimary trial endpoint and performed using, whenever possible, three entire tumor cross-sections. IHC staining for endothelial markers, CD31 and CD146, as markers of vascular density, was also performed (Supplementary Materials and Methods).
Transcriptomic analysis for Buffa (20), Hallmark (21), Harris (22), and Winter (23) hypoxia signatures was conducted from the same tumor regions with OXPHOS, glycolysis, reactive oxygen species (ROS), angiogenesis, VEGF, epithelial-to-mesenchymal transition (EMT), integrated stress response, and STAT3 signaling pathway signatures also studied. Genes in each signature are listed in Supplementary Tables S1–S4. RNA isolation was performed using RecoverAll Total Nucleic Acid Isolation Kit for FFPE (Invitrogen), as per the manufacturer's instructions (Supplementary Materials and Methods). Paired end library preparations and RNA 3′ end sequencing (HiSeq4000 75 bp) were performed by Oxford Genomics Centre, Wellcome Centre for Human Genetics (Oxford, England, United Kingdom). FastQC was used to evaluate quality of sequencing reads at both base and read levels (24). Reads with low quality bases (phred score < 20) were removed. Filtered reads were aligned to the human reference genome (hg19) using STAR (25). Gene expression was quantified using RSEM (26) and normalized counts were logged base 2.
Change in plasma levels of VEGF, carbonic anhydrase IX (CAIX), and osteopontin were additional coprimary endpoints and change in plasma miR-210 was a secondary endpoint (Supplementary Materials and Methods). Circulating levels of these hypoxic cellular response proteins have been shown to correlate with other measures of tumor hypoxia, and thus represent putative hypoxia measures, while also functioning as negative prognostic biomarkers for patients treated with radiotherapy (27–31).
Atovaquone plasma levels were also measured on the morning of the second trial visit, with no time interval mandated between plasma sampling and preceding dose as treatment duration was sufficient to achieve steady-state levels (Supplementary Materials and Methods).
Clinical assessment
Adverse event (AE) monitoring occurred until surgery. AEs were graded according to NCI Common Terminology Criteria v4.0. Given the established excellent safety profile of atovaquone, only grade ≥3 AEs or grade ≥2 atovaquone-related AEs were reportable.
Evaluability and statistical analysis
Primary analyses were on a modified intention-to-treat basis. Hypoxia PET-CT and plasma marker analyses included patients with interpretable data from both visits. Primary imaging analysis using hypoxia PET-CT excluded patients with insufficient baseline HV to reliably measure change (predefined as <1.5 mL). Patients were included for pimonidazole analysis if pimonidazole was administered 16–24 hours prior to surgery.
Statistical analyses used Stata v15.0 (StataCorp) and R v3.6.3 (Comprehensive R Archive Network, https://CRAN.R-project.org).
Percentage change in HV, plasma markers, and other measures was calculated between trial visits for each patient. The binomial method was used to derive 95% confidence intervals (CI) for median changes within cohorts. For primary imaging analyses, a linear regression model was used to assess mean difference between cohorts at the second imaging visit adjusted for cohort, baseline measure, and tumor volume, with values log-transformed to improve model validity. Spearman coefficients were reported for correlation analysis. Statistical tests were performed at the 5% level.
For RNA sequencing (RNA-seq), normalized read count data for all samples were processed by Limma-Voom method to identify differentially expressed genes (32). Gene set enrichment analysis (GSEA) was carried out using gene sets from the Molecular Signatures Database v7.1 (21, 33). GSEA3.0.0 was used with default settings (33). Gene set variation analysis (GSVA) and single sample gene set enrichment analysis (ssGSEA) were carried out using the GSVA Bioconductor package (34). sigQC protocol was used to evaluate signature applicability and compute signature summary scores (35). Heatmaps were created using the Pheatmap R package (https://CRAN.R-project.org/package=pheatmap). Intercohort differences in expression were compared using independent samples t test for hypoxia signature analysis and Wilcoxon rank-sum test for all other expression analysis (equal variances).
Results
From May 2016 to August 2018, 46 patients were recruited, 23 per cohort. Cohorts were well balanced in key clinical characteristics, except for male predominance in untreated patients (Table 1). CONSORT diagrams are shown in Supplementary Figs. S1 and S2. Median length of atovaquone treatment between imaging timepoints was 12 [interquartile range (IQR), 7–13] days and the median length prior to tumor resection was 14 (IQR, 12–15) days. Median atovaquone plasma concentration was 62.3 μmol/L (IQR, 50.2–94.2), with all patients achieving a level shown preclinically to reduce tumor hypoxia (ref. 11; Supplementary Fig. S3). Median length between visits for untreated patients was 13 (IQR, 6–13) days.
Clinical characteristics of patients at baseline.
. | Atovaquone cohort . | No atovaquone cohort . |
---|---|---|
Characteristic . | n = 23 . | n = 23 . |
Age (year) | 69.7 (58.5–74.1) | 68.6 (62.4–71.4) |
Sex | ||
Male | 11 (47.8) | 17 (73.9) |
Female | 12 (52.2) | 6 (26.1) |
Ethnicity | ||
White | 23 (100) | 23 (100) |
Smoking status | ||
Never smoker | 0 (0.0) | 2 (8.7) |
Ex-smoker | 22 (95.7) | 21 (91.3) |
Current smoker | 1 (4.4) | 0 (0.0) |
ECOG performance status | ||
0 | 16 (69.6) | 16 (69.6) |
1 | 7 (30.4) | 7 (30.4) |
Histologya | ||
Adenocarcinoma | 15 (65.2) | 11 (47.8) |
Squamous cell carcinoma | 7 (30.4) | 8 (34.8) |
Adenosquamous carcinoma | 1 (4.4) | 0 (0.0) |
Large cell | 0 (0.0) | 1 (4.4) |
Giant cell carcinoma | 0 (0.0) | 1 (4.4) |
Pleomorphic carcinoma | 0 (0.0) | 1 (4.4) |
Not otherwise specified | 0 (0.0) | 1 (4.4) |
TNM stagea | ||
T1b | 0 (0.0) | 1 (4.4) |
T1c | 2 (8.7) | 1 (4.4) |
T2vis | 1 (4.4) | 2 (8.7) |
T2a | 3 (13.0) | 4 (17.4) |
T2b | 5 (21.7) | 3 (13.0) |
T3 | 7 (30.4) | 5 (21.7) |
T4 | 5 (21.7) | 7 (30.5) |
N0 | 13 (56.5) | 12 (52.2) |
N1 | 7 (30.4) | 8 (34.8) |
N2 | 3 (13.0) | 3 (13.0) |
M0 | 23 (100.0) | 22 (95.7) |
M1b | 0 (0.0) | 1 (4.4) |
Tumor maximum diameter (mm) | 45 (33–60) | 44 (29–60) |
Hemoglobin (g/dL) | 13.2 (12–14.3) | 13.9 (12.9–14.7) |
. | Atovaquone cohort . | No atovaquone cohort . |
---|---|---|
Characteristic . | n = 23 . | n = 23 . |
Age (year) | 69.7 (58.5–74.1) | 68.6 (62.4–71.4) |
Sex | ||
Male | 11 (47.8) | 17 (73.9) |
Female | 12 (52.2) | 6 (26.1) |
Ethnicity | ||
White | 23 (100) | 23 (100) |
Smoking status | ||
Never smoker | 0 (0.0) | 2 (8.7) |
Ex-smoker | 22 (95.7) | 21 (91.3) |
Current smoker | 1 (4.4) | 0 (0.0) |
ECOG performance status | ||
0 | 16 (69.6) | 16 (69.6) |
1 | 7 (30.4) | 7 (30.4) |
Histologya | ||
Adenocarcinoma | 15 (65.2) | 11 (47.8) |
Squamous cell carcinoma | 7 (30.4) | 8 (34.8) |
Adenosquamous carcinoma | 1 (4.4) | 0 (0.0) |
Large cell | 0 (0.0) | 1 (4.4) |
Giant cell carcinoma | 0 (0.0) | 1 (4.4) |
Pleomorphic carcinoma | 0 (0.0) | 1 (4.4) |
Not otherwise specified | 0 (0.0) | 1 (4.4) |
TNM stagea | ||
T1b | 0 (0.0) | 1 (4.4) |
T1c | 2 (8.7) | 1 (4.4) |
T2vis | 1 (4.4) | 2 (8.7) |
T2a | 3 (13.0) | 4 (17.4) |
T2b | 5 (21.7) | 3 (13.0) |
T3 | 7 (30.4) | 5 (21.7) |
T4 | 5 (21.7) | 7 (30.5) |
N0 | 13 (56.5) | 12 (52.2) |
N1 | 7 (30.4) | 8 (34.8) |
N2 | 3 (13.0) | 3 (13.0) |
M0 | 23 (100.0) | 22 (95.7) |
M1b | 0 (0.0) | 1 (4.4) |
Tumor maximum diameter (mm) | 45 (33–60) | 44 (29–60) |
Hemoglobin (g/dL) | 13.2 (12–14.3) | 13.9 (12.9–14.7) |
Note: Data shown as patient number with percentage or mean with IQR.
Abbreviation: TNM, tumor–node–metastasis.
aHistology and staging are following resection, unless this was not performed, in which case diagnostic histology and imaging were used, respectively.
Tumor hypoxia imaging analysis
Thirty patients were evaluable for assessing change in HV using hypoxia PET-CT, 15 per cohort. Thirteen patients were deemed unevaluable due to insufficient baseline HV (predefined as <1.5 mL) and 1 patient had uninterpretable scan images. Eleven (73.3%) atovaquone-treated patients had >10% reduction in HV from baseline and the median change was −28% (95% CI, −58.2 to −4.4). In contrast, median change in untreated patients was +15.5% (95% CI, −6.5 to 35.5), with >10% reduction observed in only 2 (13.3%) patients. Estimated geometric mean HV was 55% (95% CI, 24%–74%) lower at the second visit for atovaquone-treated patients compared with untreated patients (P = 0.004), after adjusting for baseline HV and tumor volume. HV changes per individual patient are shown in Fig. 1A and B and summarized in Table 2.
Change in tumor HV measured by hypoxia PET-CT. Waterfall plots of percentage change in HV from baseline for atovaquone-treated (A) and -untreated patients (B) using TBR ≥1.4 to define HV. More than 10% reduction in HV represents a meaningful reduction and is indicated by the dashed line. Changes in HV using TBR ≥1.6 (C) and ≥1.8 (D) are shown for atovaquone-treated patients with sufficient HV (≥1.5 mL) to reliably measure change. E, Representative FMISO PET-CT scans for 3 patients (A, B, and C) at baseline and post-atovaquone treatment (SUV scale, 0.5–3). Arrows on the pretreatment scans indicate tumor location. Areas with high tracer uptake (red) represent hypoxic regions. Treatment length between scans was 14 days for patients A and C, and 13 days for patient B. The reduction in HV from baseline for patients A, B, and C was 33%, 61%, and 66%, respectively.
Change in tumor HV measured by hypoxia PET-CT. Waterfall plots of percentage change in HV from baseline for atovaquone-treated (A) and -untreated patients (B) using TBR ≥1.4 to define HV. More than 10% reduction in HV represents a meaningful reduction and is indicated by the dashed line. Changes in HV using TBR ≥1.6 (C) and ≥1.8 (D) are shown for atovaquone-treated patients with sufficient HV (≥1.5 mL) to reliably measure change. E, Representative FMISO PET-CT scans for 3 patients (A, B, and C) at baseline and post-atovaquone treatment (SUV scale, 0.5–3). Arrows on the pretreatment scans indicate tumor location. Areas with high tracer uptake (red) represent hypoxic regions. Treatment length between scans was 14 days for patients A and C, and 13 days for patient B. The reduction in HV from baseline for patients A, B, and C was 33%, 61%, and 66%, respectively.
Summary of 4-hour hypoxia PET-CT results.
. | Atovaquone cohort . | No atovaquone cohort . | ||
---|---|---|---|---|
. | n = 15 . | n = 15 . | ||
. | Tumor volume . | HV . | Tumor volume . | HV . |
. | (mL) . | (mL) . | (mL) . | (mL) . |
Baseline scan | 59.3 | 28.9 | 43.7 | 11.4 |
(30.7–179.5) | (5.4–50.8) | (20.6–80.7) | (2.6–34.0) | |
Presurgery scan | 53.7 | 13.7 | 43.8 | 13.1 |
(28.5–186.6) | (1.7–37.4) | (23.1–79.3) | (3.2–32.7) | |
Change from baseline | +4.0% | −28.0% | +2.8% | +15.5% |
(−2.8 to 6.2) | (−58.2 to −4.4) | (−1.7 to 14.1) | (−6.5 to 35.5) | |
Treatment estimate for HV | 0.45 (0.26–0.76) |
. | Atovaquone cohort . | No atovaquone cohort . | ||
---|---|---|---|---|
. | n = 15 . | n = 15 . | ||
. | Tumor volume . | HV . | Tumor volume . | HV . |
. | (mL) . | (mL) . | (mL) . | (mL) . |
Baseline scan | 59.3 | 28.9 | 43.7 | 11.4 |
(30.7–179.5) | (5.4–50.8) | (20.6–80.7) | (2.6–34.0) | |
Presurgery scan | 53.7 | 13.7 | 43.8 | 13.1 |
(28.5–186.6) | (1.7–37.4) | (23.1–79.3) | (3.2–32.7) | |
Change from baseline | +4.0% | −28.0% | +2.8% | +15.5% |
(−2.8 to 6.2) | (−58.2 to −4.4) | (−1.7 to 14.1) | (−6.5 to 35.5) | |
Treatment estimate for HV | 0.45 (0.26–0.76) |
Note: Median values with IQR for scan visit data and median percentage change from baseline with 95% CI. Treatment estimate is geometric mean of HV in atovaquone-treated group in relation to untreated group and is equivalent to 55% (95% CI, 24%–74%) reduction in the atovaquone-treated group. Treatment estimate was adjusted for baseline tumor volume and HV with values log-transformed. All patients had FMISO PET-CT, except for 1 patient in the atovaquone treatment cohort who had FAZA PET-CT.
Volumes of baseline tumor hypoxia did vary in both cohorts, however, this did not appear to affect the degree or direction of response. On review of individual patient hypoxia PET-CT data (Supplementary Table S1), the 4 atovaquone-treated patients who did not have a meaningful reduction in HV had a range of small (24.6 mL), medium (59.3 and 45.9 mL), and large (366.9 mL) HVs at baseline. Similarly, in the 2 untreated patients in whom HV decreased, one had significant baseline HV (44.8 mL), while the other had very little (3.5 mL).
Given the reduction in HV in atovaquone-treated patients, unplanned analysis using higher hypoxia PET tracer uptake thresholds to define hypoxia, thus measuring more intensely hypoxic regions, was performed. Using TBR ≥ 1.6 and ≥1.8 to define tumor hypoxic voxels, even more marked reduction following treatment was observed in patients whose tumors contained sufficient (>1.5 mL) HV at this level to reliably measure change (Fig. 1C and D).
Representative example hypoxia PET-CT images before and after atovaquone treatment are shown in Fig. 1E.
Dynamic hypoxia PET results indicated reduction in tumor hypoxia (k5) following treatment, −11.8% (95% CI, −23.8 to 6.8) versus 11.4% (95% CI, −25.1 to 22.6) for atovaquone-treated and -untreated patients, respectively (Table 3). Complete 4-hour and dynamic hypoxia PET-CT data obtained for each patient are provided in Supplementary Table S1.
Summary of dynamic hypoxia PET-CT results.
. | Atovaquone cohort . | No atovaquone cohort . | ||||
---|---|---|---|---|---|---|
. | n = 13 . | n = 13 . | ||||
. | vB . | K1 . | k5 . | vB . | K1 . | k5 . |
. | (%) . | (mL/minute/g) . | (1/minute) . | (%) . | (mL/minute/g) . | (1/minute) . |
Baseline scan | 3.7 | 0.42 | 0.0038 | 2.1 | 0.29 | 0.0044 |
(2–5.7) | (0.36–0.54) | (0.0027–0.0050) | (0.1–4.7) | (0.27–0.47) | (0.0038–0.0062) | |
Presurgery scan | 4.9 | 0.43 | 0.0037 | 2.8 | 0.37 | 0.0046 |
(2.7–5.1) | (0.36–0.50) | (0.0029–0.0069) | (0.1–4.6) | (0.24–0.42) | (0.0033–0.0058) | |
Change from baseline | +6.8% | −5.7 | −11.8% | +7.4% | −10.4% | +11.4% |
(−16.9 to 142.2) | (−23.3 to 15.3) | (−23.8 to 6.8) | (−7.6 to 43.5) | (−29.3 to 6.9) | (−25.1 to 22.6) |
. | Atovaquone cohort . | No atovaquone cohort . | ||||
---|---|---|---|---|---|---|
. | n = 13 . | n = 13 . | ||||
. | vB . | K1 . | k5 . | vB . | K1 . | k5 . |
. | (%) . | (mL/minute/g) . | (1/minute) . | (%) . | (mL/minute/g) . | (1/minute) . |
Baseline scan | 3.7 | 0.42 | 0.0038 | 2.1 | 0.29 | 0.0044 |
(2–5.7) | (0.36–0.54) | (0.0027–0.0050) | (0.1–4.7) | (0.27–0.47) | (0.0038–0.0062) | |
Presurgery scan | 4.9 | 0.43 | 0.0037 | 2.8 | 0.37 | 0.0046 |
(2.7–5.1) | (0.36–0.50) | (0.0029–0.0069) | (0.1–4.6) | (0.24–0.42) | (0.0033–0.0058) | |
Change from baseline | +6.8% | −5.7 | −11.8% | +7.4% | −10.4% | +11.4% |
(−16.9 to 142.2) | (−23.3 to 15.3) | (−23.8 to 6.8) | (−7.6 to 43.5) | (−29.3 to 6.9) | (−25.1 to 22.6) |
Note: Median values with IQR for scan visit data and median percentage change from baseline with 95% CI. All patients had FMISO PET-CT, except 1 patient in the atovaquone treatment cohort who had FAZA PET-CT.
For patients with complete MRI datasets, increase in T2* was observed in 5 (42%) of 12 atovaquone-treated patients in keeping with increased venous oxygenation, while no such change was observed in any untreated patients (Supplementary Fig. S4A and S4B).
Tumor tissue hypoxia transcriptomic and IHC analysis
Hypoxia transcriptomic analysis revealed significant differences between atovaquone-treated and -untreated tumors. Expression of the Buffa hypoxia signature was lower in treated tumors regardless of whether GSVA (P = 0.01), ssGSEA (P = 0.04), or median (P = 0.02) summarization scores were used (Fig. 2A). This was confirmed by intercohort comparison of GSVA using Hallmark (P < 0.001), Harris (P < 0.001), and Winter (P < 0.001) hypoxia signatures (Fig. 2B; Supplementary Fig. S5). GSEA confirmed all four hypoxia gene signatures were significantly enriched in nontreated patients with normalized enrichment score 1.58–2.07, P < 0.05 in all cases. Expression of OXPHOS and glycolysis genes was altered (Fig. 2C) and a striking lower expression of genes activated in response to ROS in the group of patients treated with atovaquone was observed (Fig. 2D). Although it may seem paradoxical, it is well known that ROS production is enhanced in hypoxia (36), therefore, a lower level of ROS in atovaquone-treated patients may be a direct consequence of increased oxygen levels. Hypoxia signature validity was confirmed by assessing expression metrics (Supplementary Fig. S6) and correlation with hypoxia PET-CT (Fig. 2E). The Buffa signature correlated with hypoxic tumor percentage (Spearman rho = 0.40; P = 0.04), with significance maintained in a mixed-effects linear model accounting for replicate RNA-seq samples (effect size = 0.62; P = 0.03).
Altered expression of tumor hypoxia- and metabolism-related genes in atovaquone-treated patients as measured by RNA-seq. A, Box and whisker plots of the Buffa hypoxia gene signature expression score in atovaquone-treated and -untreated tumor samples using GSVA, ssGSEA, and median summarization methods. Each data point represents an expression score obtained from tissue from an entire cross-section of tumor. B, GSVA scores for Hallmark, Harris, and Winter hypoxia gene signatures. Median values, IQR, and absolute range (minus outliers) are shown with independent sample t test significance provided. C, Heatmaps illustrating differentially expressed KEGG glycolysis and OXPHOS genes between treatment groups. Genes present in any of the four hypoxia signatures are indicated by the pink bar. D, Heatmap showing lower expression of ROS-associated genes in atovaquone-treated patients using Hallmark and gene ontology (GO1 and GO2) ROS signature GSVA scores. E, Positive correlation between Buffa GSVA hypoxia score and hypoxic tumor percentage as measured by hypoxia PET-CT.
Altered expression of tumor hypoxia- and metabolism-related genes in atovaquone-treated patients as measured by RNA-seq. A, Box and whisker plots of the Buffa hypoxia gene signature expression score in atovaquone-treated and -untreated tumor samples using GSVA, ssGSEA, and median summarization methods. Each data point represents an expression score obtained from tissue from an entire cross-section of tumor. B, GSVA scores for Hallmark, Harris, and Winter hypoxia gene signatures. Median values, IQR, and absolute range (minus outliers) are shown with independent sample t test significance provided. C, Heatmaps illustrating differentially expressed KEGG glycolysis and OXPHOS genes between treatment groups. Genes present in any of the four hypoxia signatures are indicated by the pink bar. D, Heatmap showing lower expression of ROS-associated genes in atovaquone-treated patients using Hallmark and gene ontology (GO1 and GO2) ROS signature GSVA scores. E, Positive correlation between Buffa GSVA hypoxia score and hypoxic tumor percentage as measured by hypoxia PET-CT.
Fifteen patients in cohort 1 and 11 patients in cohort 2 were evaluable for tumor pimonidazole analysis. Pimonidazole scores were comparable between cohorts, in keeping with equivalent HV tumor percentage for both cohorts at the presurgery visit (Supplementary Figs. S7 and S8; Supplementary Table S2). Significant positive correlations were observed between pimonidazole score and HV percentage defined using the higher PET tracer uptake thresholds (Supplementary Fig. S9).
Circulating hypoxia markers
Changes in plasma levels of all four putative markers of tumor hypoxia could be assessed in only 10 patients as VEGF was undetectable in 9 patients, CAIX was undetectable in 1 patient, and miR-210 was unmeasurable in 15 patients due to sample hemolysis (Supplementary Fig. S10; Supplementary Tables S3–S6). No agreement was observed between individual markers or between any marker and hypoxia PET-CT or pimonidazole IHC.
Tumor perfusion and vascular analysis
Chronic hypoxia is known to stimulate sustained angiogenesis. As a result, tumor vascularization becomes abnormal leading to impaired tumor perfusion and consequently restricted delivery of oxygen, thus reinforcing hypoxia (37). We investigated whether reversal of hypoxia by atovaquone may improve tumor perfusion through vascular normalization. Using pCT, no meaningful changes in tumor blood volume (Fig. 3A), blood flow (Fig. 3B), or blood mean transit time (Fig. 3C) were observed following treatment, therefore suggesting that hypoxic reduction was not because of increased tumor perfusion. pCT data obtained for each patient are provided in Supplementary Table S7. A number of patients were unevaluable for pCT imaging analyses, with the most common reason being uninterpretable scan images due to variations in the rate of contrast injection as measured by the analysis input function and significant interference of signal from adjacent structures, such as heart and great vessels (n = 4). DCE-MRI Ktrans (Fig. 3D), as well as dynamic PET-derived vB and K1 (Table 3), were also not altered for either cohort, signifying that tumor perfusion was unchanged. We also analyzed the endothelial markers, CD31 and CD146, by IHC to assess a potential decrease in vessel density as indicative of vascular normalization (37). In agreement with the lack of increased perfusion, the endothelial marker analysis did not show a lower vessel density in atovaquone-treated patients, with in fact a trend for the reverse effect observed (Supplementary Fig. S11). In addition, MRI T1 shMOLLI analysis indicated that no changes in tumor biochemistry occurred to suggest significant alteration in cellularity either (Supplementary Fig. S4C and S4D), in keeping with preclinical data, showing that as monotherapy, atovaquone does not have direct cytotoxic effects (11).
Effect of atovaquone on tumor perfusion and gene expression signatures. Change in pCT-derived blood volume (A), blood flow (B), and mean transit time (C), as well as DCE MRI–derived Ktrans (D) for atovaquone-treated and -untreated patients. The few cases where a meaningful change in imaging parameter was observed are indicated by red lines. Only data for patients with complete scan datasets and evaluable for primary imaging analysis are shown. E–H, Box and whisker plots showing GSVA scores in atovaquone-treated and -untreated tumor samples. E, Angiogenesis-related gene signatures: Hallmark and gene ontology (GO) angiogenesis signature scores, and KEGG and BIOCARTA VEGF signaling pathway expression scores. F, EMT signature. G, Integrated stress response (STRESS) genes. H, STAT3-induced gene expression. Each data point represents an expression score obtained from tissue from an entire cross-section of tumor. The median value, IQR, and absolute range (minus outliers) are shown with Wilcoxon rank-sum test significance provided.
Effect of atovaquone on tumor perfusion and gene expression signatures. Change in pCT-derived blood volume (A), blood flow (B), and mean transit time (C), as well as DCE MRI–derived Ktrans (D) for atovaquone-treated and -untreated patients. The few cases where a meaningful change in imaging parameter was observed are indicated by red lines. Only data for patients with complete scan datasets and evaluable for primary imaging analysis are shown. E–H, Box and whisker plots showing GSVA scores in atovaquone-treated and -untreated tumor samples. E, Angiogenesis-related gene signatures: Hallmark and gene ontology (GO) angiogenesis signature scores, and KEGG and BIOCARTA VEGF signaling pathway expression scores. F, EMT signature. G, Integrated stress response (STRESS) genes. H, STAT3-induced gene expression. Each data point represents an expression score obtained from tissue from an entire cross-section of tumor. The median value, IQR, and absolute range (minus outliers) are shown with Wilcoxon rank-sum test significance provided.
Despite the reported absence of an improvement in vascular function, RNA-seq analysis revealed a significantly lower expression of angiogenesis signature genes in atovaquone-treated patients (Fig. 3E): Hallmark angiogenesis (P = 0.002) and gene ontology angiogenesis (P = 0.012). Although there was no significant overlap between the hypoxia and angiogenesis signatures (Supplementary Fig. S12), many of the genes in the angiogenesis signature have been linked to the hypoxia response (38–40). Preclinical and clinical evidence has demonstrated that inhibition of the VEGF angiogenic pathway effectively increases tumor oxygenation linked to features of vascular normalization (37). Of note, we did not find differences in the expression of VEGF signaling signatures between untreated and atovaquone-treated patients (Fig. 3E): Kyoto Encyclopedia of Genes and Genomes (KEGG) VEGF (P = 0.88) and BIOCARTA VEGF (P = 0.25).
Altogether, these results suggest that atovaquone treatment reduces angiogenic signaling in patients with NSCLC, but this has no impact on vessel functionality, at least within the timeframe of the study.
EMT, integrated stress response, and STAT3
Upon OXPHOS inhibition, cancer cells are able to shift to a more glycolytic metabolism for energy production. Because increased glycolysis has been linked to enhanced tumor aggressiveness (41), we investigated whether atovaquone was able to promote EMT by analyzing a previously published gene expression signature (42). As shown in Fig. 3F, the expression of the EMT signature was similar in untreated and atovaquone-treated patients (P = 0.36), suggesting that atovaquone does not induce a more aggressive tumor phenotype.
Aside from inhibiting OXPHOS, atovaquone has recently been shown to modulate two signaling pathways in cancer cells. In preclinical models of acute myeloid leukemia (AML), atovaquone consistently upregulated four genes involved in the integrated stress response (REDD1, ATF3, CHOP, and CHAC1; ref. 43), suggesting that atovaquone is able to induce a state of cellular stress, which may have antitumoral effects. In contrast to these preclinical findings, we found that these four stress response genes were not induced following atovaquone treatment (P = 0.13; Fig. 3G). A gene signature of STAT3 activation was used previously to demonstrate that atovaquone inhibited STAT3 function in multiple myeloma models (44). Using the same signature, we also found that atovaquone treatment strongly diminished the expression of STAT3-induced genes (P < 0.001; Fig. 3H).
Safety and tolerability
In keeping with the established favorable safety profile of atovaquone, only one AE was reported (grade 3 hypertension) in the atovaquone treatment cohort and was deemed unrelated to treatment.
Discussion
This study demonstrates that the mitochondrial inhibitor atovaquone reduces tumor hypoxia in patients with NSCLC, therefore providing the first clinical confirmation for the strategy of targeting tumor metabolism to alter the tumor microenvironment in this way.
Hypoxia PET-CT was the primary method used to measure tumor HV, with a rapid reduction observed in the majority of atovaquone-treated patients. The increase in median HV observed in the nontreated cohort, indicating a worsening of tumor hypoxia over time, made the reduction in treated patients all the more striking. Because this imaging modality is the most accepted method for measuring hypoxic change in the clinical setting (13, 16, 17), our findings provide the best available evidence that atovaquone is a clinically useful hypoxia modifier. Furthermore, as the nitroimidazole-based hypoxia PET tracers used in this study accumulate only in viable tissue at oxygen levels below the threshold at which hypoxia-related radioresistance becomes apparent (45), reduction in uptake should translate into improved tumor radiation response. In support of this, clinical studies have repeatedly demonstrated high uptake of such tracers functions as a negative prognostic biomarker for patients treated with radiation (7, 46). Our observation that the most pronounced reduction was in the most intensely hypoxic, and thus radiobiologically resistant tumor regions, provides particular impetus for developing atovaquone as a novel tumor radiosensitizer.
Confirmation of the pharmacodynamics effect of hypoxia reduction by the reduced expression of key hypoxia-related tumor gene signatures not only further establishes atovaquone as a hypoxia modifier, but also demonstrates the alteration in oxygenation was of biological relevance. Furthermore, transcriptomic analysis revealed altered expression of OXPHOS and glycolysis pathways, as well as decreased ROS signature expression, in keeping with effects on tumor metabolism more widely. The downregulation of ROS-induced genes in the tumors of patients treated with atovaquone is in agreement with higher oxygen levels, as it is well established that ROS production is enhanced under hypoxia (36).
Preclinical evidence suggests that the mechanism of action by which atovaquone reduces tumor hypoxia is through decreased tumor oxygen consumption (11). It is possible, however, that reversal of tumor hypoxia could cause a normalization of the vasculature, through a decrease in angiogenic signaling, and consequently improve tumor perfusion, which in turn could contribute to tumor oxygenation (37). Our observation that atovaquone treatment was not associated with changes in tumor perfusion, as assessed by multimodality perfusion imaging, strongly suggests that vascular normalization does not contribute to the hypoxia alleviation reported in patients with NSCLC treated with atovaquone. This was further supported by the lack of decreased vessel density, considered a feature of vascular normalization (37), in atovaquone-treated patients. Therefore, the decrease in hypoxia is likely the consequence of increased oxygenation because of diminished oxygen consumption.
RNA-seq analysis revealed reduced angiogenic signaling in atovaquone-treated patients. The absence of improved perfusion with atovaquone treatment may indicate that the reported decrease in angiogenic signaling does not suffice or requires a longer time to cause a significant vascular normalization. In addition, the fact that VEGF signaling, regarded as one of the main drivers of tumor angiogenesis (37), remained unaltered with atovaquone treatment may also partly explain the lack of substantial vascular normalization and increased perfusion, despite an overall decrease in angiogenic signaling.
Atovaquone did not lead to an increased expression in genes involved in EMT, suggesting that treatment did not make the tumors more aggressive. This is important because of the possibility that the inhibition of OXPHOS and subsequent increased glycolysis could lead to increased tumor aggressiveness and metastasis (41). Our study precluded long-term follow-up, therefore we were unable to assess whether incidence of metastasis or other clinical outcomes differed between cohorts, and these are important endpoints for future clinical testing of this agent as an anticancer treatment.
Aside from modulating metabolism, atovaquone has recently been shown to inhibit STAT3 signaling and upregulate integrated stress response genes, as demonstrated in multiple myeloma and AML models, respectively (43, 44). While we did not see a change in the expression of stress response genes, we did find a decrease in genes affected by STAT3 signaling. Interestingly, unlike the above-mentioned studies in hematologic malignancies, our preclinical studies did not show an antitumor effect of atovaquone as single agent (11). In this study, we did not observe a reduction in tumor volume with atovaquone treatment, but it is feasible this may occur with longer treatment duration.
Although this study demonstrates hypoxia alleviation in NSCLC, preclinical evidence demonstrates that atovaquone inhibits OCR in a wide range of tumor cell lines (11), and therefore it is expected this effect would occur in patients with different tumors types. Because hypoxia confers resistance to radiation, chemotherapy, and immunotherapy (5, 7–9), coadministration with numerous such well-established treatments widens clinical applicability further.
Several FDA-approved drugs inhibit OXPHOS through off-target effects and have been proposed as clinical hypoxia modifiers (10), of which metformin is most studied. This mitochondrial complex I inhibitor (47) reduces OCR in cancer cell lines and ameliorates hypoxia in xenografts, leading to improved tumor radiation response (11, 48, 49). Although metformin has been shown to alter tumor metabolism in patients with breast cancer (50), its effect on tumor oxygenation in patients has not been reported. Importantly, concentrations required to reduce tumor OCR in preclinical models are higher than clinically achievable. In keeping with this, a recent phase II study combining metformin with chemoradiotherapy in NSCLC failed to improve outcomes (51). In contrast, all patients in our study had atovaquone plasma levels at or above that known to reduce tumor OCR in preclinical experiments, with levels observed to be in-keeping with that stated in the summary of product characteristics for the oral micronized preparation used in this study (52).
Favorable toxicity is paramount when developing radiosensitizers because standard (chemo)radiotherapy regimes are already significantly toxic. The established excellent safety profile of atovaquone, as was observed in this study, further highlights this drug as highly promising for this purpose. As atovaquone is safe, FDA approved, and inexpensive, repurposing this agent as an anticancer treatment has the potential for rapid adoption into routine clinical practice, while helping to contribute to more sustainable healthcare systems.
Our study had limitations. First, this was a nonrandomized study and, although sizable for a translational study, patient numbers were relatively small. For this reason, it was decided not to investigate the mutational status of these tumors as the patient numbers would be too small to meaningfully assess whether certain mutations could be associated with response to atovaquone. Second, as baseline tumor tissue was not collected, assessment of change in hypoxia signatures and IHC per patient was not possible. For pimonidazole IHC, analysis was impeded due to surgery occurring outside the narrow time-window required following pimonidazole administration for a number of patients. For evaluable patients, comparable scores between cohorts were in keeping with hypoxia PET-CT–derived HV equivalence prior to surgery, with correlation between these measures observed. As atovaquone-treated patients appeared to have more baseline HV, the lack of difference in pimonidazole staining may in fact represent additional evidence for hypoxia modification.
Validated and deliverable clinical hypoxia biomarkers are needed to enable selection of patients with hypoxic tumors for treatment, and the challenge of identifying such markers was apparent in this trial. Putative circulating hypoxia markers previously shown to correlate with radiotherapy outcomes in NSCLC (28) were often undetectable or provided conflicting results. Despite difficulties correlating hypoxia measures, observed agreement between hypoxia PET-CT and gene expression signatures suggests a transcriptomic approach may be a particularly promising future strategy. Agreement between the different hypoxia signature scores and correlation with hypoxia PET-CT was also observed, supporting signature validity as hypoxia biomarkers in NSCLC.
Here, we provide the first clinical report that tumor hypoxia can be reduced by targeting tumor metabolism. Repurposing atovaquone for this purpose has the potential to improve the outcomes of numerous established treatments in NSCLC, as well as in other tumor types. In light of these highly promising findings, we have initiated phase I clinical testing combining this agent with chemoradiation in NSCLC.
Authors’ Disclosures
D.R. McGowan reports grants from NIHR during the conduct of the study. P.S. Macklin reports grants from The Pathological Society of Great Britain and Ireland/Jean Shanks Foundation during the conduct of the study. H. Flight reports grants from Cancer Research UK during the conduct of the study. J.P.B. O'Connor reports grants from Cancer Research UK during the conduct of the study. G.S. Higgins reports grants from Cancer Research UK and The Howat Foundation during the conduct of the study. No disclosures were reported by the other authors.
Authors' Contributions
M. Skwarski: Conceptualization, resources, data curation, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. D.R. McGowan: Conceptualization, resources, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. E. Belcher: Resources, investigation, writing–review and editing. F. Di Chiara: Resources, investigation, writing–review and editing. D. Stavroulias: Resources, investigation, writing–review and editing. M. McCole: Data curation, formal analysis, investigation, writing–review and editing. J.L. Derham: Resources, investigation, writing–review and editing. K.-Y. Chu: Investigation, writing–review and editing. E. Teoh: Investigation, writing–review and editing. J. Chauhan: Data curation, formal analysis, investigation, visualization, writing–review and editing. D. O'Reilly: Investigation, writing–review and editing. B.H.L. Harris: Investigation, writing–review and editing. P.S. Macklin: Data curation, formal analysis, investigation, writing–review and editing. J.A. Bull: Data curation, formal analysis, investigation, writing–review and editing. M. Green: Investigation, writing–review and editing. G. Rodriguez-Berriguete: Data curation, formal analysis, investigation, visualization, writing–review and editing. R. Prevo: Data curation, investigation, writing–review and editing. L.K. Folkes: Formal analysis, investigation, writing–review and editing. L. Campo: Formal analysis, investigation, writing–review and editing. P. Ferencz: Formal analysis, investigation, writing–review and editing. P.L. Croal: Formal analysis, investigation, writing–review and editing. H. Flight: Project administration, writing–review and editing. C. Qi: Data curation, formal analysis, investigation, writing–review and editing. J. Holmes: Data curation, formal analysis, investigation, writing–review and editing. J.P.B. O'Connor: Investigation, writing–review and editing. F.V. Gleeson: Investigation, writing–review and editing. W.G. McKenna: Investigation, writing–review and editing. A.L. Harris: Investigation, writing–review and editing. D. Bulte: Formal analysis, investigation, writing–review and editing. F.M. Buffa: Data curation, formal analysis, investigation, writing–review and editing. R.E. Macpherson: Formal analysis, investigation, writing–review and editing. G.S. Higgins: Conceptualization, resources, supervision, funding acquisition, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.
Acknowledgments
We thank all patients and their families for taking part and the staff at Oxford University Hospitals National Health Service Foundation Trust. This trial was sponsored by the University of Oxford. Trial management was provided by the Oncology Clinical Trials Office at the University of Oxford as part of the UKCRC Oxford Clinical Trials Research Unit. We thank Alix Hampson for assistance with measurement of plasma hypoxia marker and atovaquone levels. This work was funded by the Howat Foundation (C5255/A25069). Additional support was obtained from CRUK & EPSRC Cancer Imaging Centre in Oxford (C5255/A16466), CRUK/MRC Oxford Institute for Radiation Oncology (C5255/A23755), and the NIHR Oxford Experimental Cancer Medicine Centre. D.R. McGowan was funded by a NIHR/HEE Clinical Lectureship (ICA-CL-2016-02-009). P.S. Macklin was supported by the Jean Shanks Foundation/Pathological Society of Great Britain & Ireland Clinical Research Training Fellowship. J.A. Bull was supported by the EPSRC/MRC Centre for Doctoral Training in Systems Approaches to Biomedical Science (EP/G037280/1) and the EPSRC Impact Acceleration Account (EP/R511742/1). G.S. Higgins was supported by a CRUK Clinician Scientist Fellowship (C34326/A19590).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
References
Supplementary data
Trial Protocol
Supplementary Methods