A Phase II Randomized Trial of Chemoradiation with or without Metformin in Locally Advanced Cervical Cancer

This article is featured in Highlights of This Issue, p. 5231

Cervical cancer ranks fourth for both cancer incidence and mortality worldwide (1). A significant proportion of patients present with locally advanced disease (FIGO stage IB–IVA) that is not amenable to (primary) surgical treatment and is treated with external beam radiation, concurrent cisplatin chemotherapy, and brachytherapy with curative intent. There are very few effective treatments for patients with persistent or recurrent cervical cancer after chemoradiation. Tumor hypoxia, which results from an imbalance between the cellular oxygen consumption rate and the oxygen supply, is associated with inferior survival in cervical and many other cancers, and with increased metastasis and resistance to chemotherapy and radiotherapy (2–8).

Although tumor hypoxia currently represents a significant therapeutic challenge, it also presents an exciting opportunity for personalized prognosis and directed intervention. Although a meta-analysis showed that hypoxia-modifying therapies improved locoregional control (HR 0.77) and overall survival (HR 0.87), they have not been adopted in clinical practice because of practical limitations and toxicities (9). In addition, some studies may have been negative due to the lack of selection for patients with hypoxic tumors. For example, the phase III chemoradiation ± tirapazamine trial was negative in patients with head and neck cancer not selected for the presence of hypoxia (10). However, a substudy from the phase II trial conducted by the same group found that patients with detectable tumor hypoxia on PET imaging had a high risk of locoregional failure (11). Of those, patients treated with tirapazamine had significantly lower locoregional failure than those without (11). This highlights the importance for trials to include the measurement of the extent of tumor hypoxia as an entry criterion.

PET imaging of tumor hypoxia is a noninvasive method that uses hypoxia-sensitive tracers, usually nitroimidazoles. The nitroimidazole tracer binds in hypoxic regions of the tumor and undergoes radioactive decay, which is detected by PET imaging and spatially localized to form a three-dimensional image of the distribution of hypoxia (i.e., evaluation of the entire tumor). Pretreatment PET imaging of tumor hypoxia in patients with head and neck, lung, brain, and cervical cancers has been shown to correlate with disease-/progression-free survival (12–14), cancer-specific survival (15, 16), and overall survival (17–20) following (chemo)radiation. Meta-analysis of hypoxia imaging studies demonstrates a uniform tendency for a poor response to radiotherapy for hypoxic tumors (OR 0.25; ref. 21). ¹⁸F-Fluoroazomycin arabinoside (FAZA) is a nitroimidazole tracer with a favorable tumor-to-background ratio (22) and the ability to monitor and/or predict response to hypoxia-targeted therapy (23, 24).

We investigated a novel, safe, and practical approach using metformin to decrease tumor hypoxia and improve response to chemoradiotherapy. Metformin, a well-established, inexpensive, first-line oral drug treatment for type 2 diabetes, was found to be associated with a significant reduction in the risk of cancer-related mortality (OR 0.66) in epidemiologic studies (25). Our population-based study of 385 diabetic patients with cervical cancer showed that the cumulative dose of metformin after cervical cancer diagnosis was independently associated with a decreased risk of cervical cancer–specific mortality (26). Metformin's primary mechanism of action in diabetic patients is direct inhibition of complex I activity in the mitochondrial respiratory chain. Metformin is thought to inhibit cancer growth by lowering circulating insulin level or directly activating AMP-activated protein kinase (AMPK; ref. 27). Our group discovered a novel mechanism of metformin—enhancing tumor radiation response by inhibiting tumor cell oxygen consumption (28). By inhibiting mitochondrial complex I and oxygen consumption in tumor cells close to vessels, metformin facilitated increased oxygen diffusion to previously hypoxic tumor cells, and this reoxygenation consequently increased radiation response in experimental tumor models. Metformin decreased tumor cell oxygen consumption in a dose-dependent manner in cell lines and xenografts (28). This metformin-induced decrease in tumor hypoxia [15% change in fractional hypoxic volume (FHV)] detected on FAZA-PET was sufficient to cause a meaningful improvement in tumor radiation response (28). These promising preclinical and population-based data provided a clear rationale for launching a phase II, window-of-opportunity trial to determine if metformin decreases tumor hypoxia in patients with hypoxic locally advanced cervical cancer.

This phase II randomized clinical trial comparing standard chemoradiation ± metformin was approved by the local Institutional Review Boards and accrued patients across 2 academic centers in Canada between May 2015 and October 2019. Female patients ages 18 to 75 with an ECOG performance status of 0 to 1, and histologically confirmed 2009 FIGO stage IB2–IVA cervical squamous cell carcinoma, adenocarcinoma, or adenosquamous carcinoma planned for definitive chemoradiation were eligible. Key exclusion criteria included evidence of distant metastases; prior anticancer treatment for cervical cancer; diabetes mellitus; current use of metformin, sulfonylureas, thiazolidinediones, or insulin; any condition associated with increased risk of metformin-associated lactic acidosis; HIV positivity; and history of another invasive malignancy, except for nonmelanoma skin cancer or tumors curatively treated with no evidence of disease for ≥5 years. All participants provided written informed consent prior to registration and did not receive compensation for the study. No commercial support was provided. The study was registered with ClinicalTrials.gov (NCT02394652). It was conducted in accordance with the Declaration of Helsinki and Belmont report.

All consenting patients underwent a screening FAZA-PET/CT scan. A punch biopsy of the primary cervical tumor was taken within 0–2 days of the FAZA-PET/CT scan. To avoid interfering with FAZA-PET/CT scan results, cervical biopsies were not performed immediately prior to the FAZA-PET/CT scan. Only patients with FAZA uptake (hypoxic tumor) on the PET/CT were officially enrolled in the study; patients without FAZA uptake were excluded. One patient later withdrew. Their age varied from 31 to 74 years (median 52).

Patients with FAZA uptake were randomized centrally by the Princess Margaret Drug Development Office in a 2:1 ratio (in favor of metformin) to receive either metformin in combination with standard chemoradiation or standard chemoradiation alone. The study statistician generated the random allocation sequence without blocks. Patients randomized to the metformin arm started metformin within 1 day of randomization 850 mg once daily × 3 days, followed by 850 mg twice daily throughout the entire duration of external radiotherapy. Metformin was held during brachytherapy. A second FAZA-PET/CT scan and tumor biopsy were performed ∼1 week from the first (i.e., after ∼7 days of metformin or no intervention in the control group), just before the start of chemoradiation. Both arms received identical definitive chemoradiation, consisting of external beam radiotherapy 45 Gy/25 fractions, concurrent weekly cisplatin chemotherapy 40 mg/m², intensity-modulated radiotherapy boost to enlarged nodes, and MRI-guided brachytherapy 28 Gy/4 fractions. Target and normal tissue constraints followed the EMBRACE guidelines (29). All patients were assessed weekly during treatment, every 6 months for the first 2 years to evaluate toxicities and disease status for study purpose, and then as per institutional practice. Acute (within 90 days from the date of randomization) and late (after 90 days) toxicities were graded using the Common Terminology for Common Terminology Criteria version 4.

FAZA-PET/CT scans were acquired and analyzed as described in our prior observational study (30). The diagnostic T2-weighted MRI was coregistered to the CT from FAZA-PET/CT for tumor delineation for each patient. A single, experienced observer blinded to the PET images contoured the primary cervical tumor, entire gluteus maximus (GM) muscle bilaterally on 6 consecutive CT slices (then contracted by 4 mm on each axial slice) at the level of tumor, and bladder for all patients. To avoid intraobserver contouring variation, the CT from the second FAZA-PET/CT was coregistered with the baseline CT/MR by soft-tissue match (i.e., match to cervix/tumor), and the contour delineated on the baseline CT was propagated onto the second CT without modification by the same observer blinded to the PET images. To minimize the spillover effect of radiotracer signal from the bladder, the Gaussian correction method was applied as previously described (30). Mutual voxels from the baseline and second FAZA-PET images remaining after Gaussian correction were analyzed to determine the FHV, defined as the ratio of the number of hypoxic voxels to the total number of tumor voxels (30). To determine whether the patient had a “hypoxic tumor” for study inclusion in real time, each tumor (T) voxel was classified as either oxic or hypoxic using a threshold T/M >1 adapted from the DAHANCA 24 trial by Mortensen and colleagues (12), where M = [SUVmean + 3 standard deviations (SD) of each patient's individual GM muscle] (i.e., patient-specific variation of the Mortensen method). After the conclusion of study accrual, FHV was also analyzed following the exact method by Mortensen and colleagues, where each trial patient's muscle histogram was normalized by its SUVmean such that all patients had a muscle SUVmean of 1, and then the muscle histograms were pooled to determine the SD of the population (i.e., the Mortsensen pooled method; ref. 12). Although the Mortensen pooled method could not be determined in real time (i.e., could only pool all patients’ muscle histograms after all patients were accrued), it usually generates a more conservative estimate of FHV. Therefore, FHV from both methods are presented: (i) the Mortensen pooled method (SUVmean + population 3SD = threshold of 1.34), and (ii) patient-specific variation of the Mortensen method. PET images and FHV were analyzed centrally by the UHN Quantitative Imaging for Personalized Cancer Medicine Program, an emeritus member of the National Institute of Health's Quantitative Imaging Network. The research associate performing the analysis was blinded to the randomization and clinical data.

Punch biopsies of the cervical tumor were snap-frozen in liquid nitrogen at the time of biopsy and then immediately stored in a −80°C freezer until processed for gene expression. RNA samples were sequenced using Illumina TruSeq Whole Transcriptome v1.5 library prep followed by sequencing on an Illumina NovaSeq. 10-μm sections were cut from frozen tissue at −20°C and stained for hematoxylin and eosin. A gynecologic pathologist identified regions of interest containing ≥ 40% tumor cells, which were then macrodissected. RNA was isolated from macrodissected tissue using the RNeasy Mini Kit (Qiagen, 74104) following the manufacturer's instructions. RNA samples were sequenced using Illumina TruSeq Whole Transcriptome v1.5 library prep followed by sequencing on an Illumina NovaSeq. Read quality was assessed using FastQC (RRID:SCR_014583) followed by alignment using the STAR (RRID:SCR_004463) aligner in two-pass basic mode with the GRCh38.p13 (hg38) reference genome and GENCODE v31 (RRID:SCR_014966) annotations. Gene expression was then quantified using RSEM v1.3.0 (RRID:SCR_013027).

Formalin-fixed tissue was paraffin-embedded, and 3-μm sections were stained on the Roche Ventana Benchmark using the MIB1 antibody (Agilent; cat. # M7240, RRID:AB_2142367) and counterstained with hematoxylin. Negative and positive controls were also performed for quality assurance. The Ki67 labeling index was calculated based on the number of positive cells expressing Ki67 divided by total tumor cells counted manually in the regions of high proliferative activity (hot spots). These regions were annotated by a gynecologic pathologist in each case and subsequently within each hot spot, areas of inflammatory cells, nontumorous stromal cells and cell debris were manually marked for exclusion. For more accurate and reproducible Ki67 labeling index results, 1,000 to 1,500 tumor cells were counted from multiple hot spots in each case.

The primary endpoint was change in FHV, defined as the average of the difference between the FHV of the cervical tumor from the first and second FAZA-PET scans, ∼1 week apart (after 7 days of metformin in the metformin arm). It was estimated that the average FHV would be 42% at baseline (30), and that a sample size of 32 patients in the metformin arm would have 83% (or 93%) power to detect a reduction of 12% (or 15%) in average FHV (biologically significant to improve tumor response to radiation in xenograft studies; ref. 28) after 1 week of metformin with an estimated SD of differences of 30%, using a one-sided paired t test with a significance level of 0.1 (0.1 given that this trial was planned as a phase II trial). The planned total sample size was 48 patients, with 32 in the metformin arm and 16 in the control arm. The control arm was used to confirm that the FHV did not vary significantly from baseline after 1 week such that the confidence interval of the difference in FHV encompasses 0 and not 12%. Although the original plan was to compare the FHV of the primary tumor on FAZA-PET pre- and post-metformin using a one-sided paired t test with a significance level of 0.1, the Wilcoxon signed-rank test was performed in the end given the small number of patients. Spatiotemporal variability of hypoxia in the cervical tumor and the influence of metformin in individual patients were explored post hoc by comparing individual voxel readings from baseline and 1 week using a one-sided test for proportion with a significance level of 0.0039 (0.05/13) to account for multiple comparisons. Secondary endpoints include disease-free survival (DFS), toxicity, change in Ki67, and predictors of metformin response for individualized therapy (metformin transporter expression (31)). DFS was defined as the duration of time from randomization to the time of relapse or death and compared using the log-rank test. The mean change in Ki-67 and predictors were compared using the Wilcoxon signed-rank test. Data for all clinical outcomes analyses were locked on August 3, 2021.

The data generated in this study are available upon request from the corresponding author.

Twenty patients underwent screening FAZA-PET scan (18 from center A and 2 from center B). Six patients were ineligible with no FAZA uptake (i.e., nonhypoxic tumor) and 1 withdrew after experiencing nausea from metformin, resulting in 13 evaluable patients: 10 randomized to the metformin arm and 3 to the control arm (Fig. 1). The study was closed prematurely due to poor accrual, decreasing availability of FAZA, and the COVID-19 pandemic.

Patient demographics and tumor characteristics are summarized in Table 1. The majority had squamous cell carcinoma (8/13, 62%) and stage IIB disease (8/13, 62%). All patients received treatment per protocol: one patient in the metformin arm refused further cisplatin chemotherapy after the first cycle, and another patient decided not to continue metformin after it was held for elevated creatinine due to new hydronephrosis 5 days after starting radiation. Metformin was held or the dose reduced to 850 mg daily (temporarily) in 3 patients as per protocol for diarrhea, nausea, and/or reduced creatinine clearance. Although it was not possible to distinguish whether these side effects were attributable to metformin or cisplatin, the protocol recommended modifying the metformin dose as the initial step to manage side effects before adjusting standard-of-care cisplatin chemotherapy. Acute grade 2+ toxicities are summarized in Supplementary Table S1. Late toxicities included grade 2 gastrointestinal (n = 1), grade 2 urinary (n = 1), grade 3 insufficiency fracture (n = 1), and grade 2 leg lymphedema (n = 1) in the metformin arm (latter 3 toxicities were observed in the same patient). No late toxicities were observed in patients from the control arm.

Figure 2A and B plot each patient's FHV. With the Mortensen pooled method (12), FHV of the 10 patients in the metformin arm decreased by an average of 10.2% ± SD 16.9% after 1 week of metformin, compared with an average increase of 4.7% ± 11.5% for the 3 patients in the control arm (P = 0.024). Using the patient-specific variation of the Mortensen method, the corresponding FHV numbers were an average decrease of 9.4% ± 18.8% and 4.9% ± 19.1%, respectively (P = 0.08). There was no strong correlation between tumor volume and change in FHV (Spearman correlation coefficient 0.14). Figure 3 and Supplementary Fig. S1 show examples of FAZA-PET images from patients in the metformin arm and control arm, respectively. Biological response to metformin in individual patients, defined for the purpose of this analysis as a statistically significant change in FHV (with a rigorously defined threshold of P < 0.0039 to account for multiple comparisons), is shown in Supplementary Table S2. There was a reduction in hypoxia in 5 of 10 patients treated with metformin using the Mortensen pooled method (or 6/10 using the patient-specific variation method), and in 1 patient in the control arm. The absolute hypoxic volume at baseline and 1 week is shown in Supplementary Fig. S2.

Three of the 10 patients in the metformin arm relapsed: 1 regional, 1 regional, and distant, and 1 distant relapse. All three patients in the control arm relapsed: 2 regional and 1 distant relapses. All regional relapses were nodal sites within the treated radiation volume. With a median follow-up of 3.3 years (range, 0.4–5.0 years), the 2-year DFS was 67% (95% CI, 27%–88%) for the metformin arm versus 33% (95% CI, 1%–77%) for control (P = 0.09; Fig. 4).

There was no significant difference in the Ki67 index following 1 week of metformin among the 7 patients with sufficient biopsy material for both timepoints (average Ki67 of 74% at baseline versus 75% after 1 week of metformin; P > 0.05); nor in the 2 control patients with sufficient sample (47% baseline vs. 44% after 1 week). Supplementary Fig. S3 plots each patient's Ki67 index. Representative photomicrographs of Ki67 staining from 2 patients are shown in Supplementary Figs. S4 and S5.

Metformin depends upon organic cation transporters and multidrug and toxin extrusion transporters (MATE) for its cellular uptake and excretion, respectively (31). Nine patients from the metformin arm had adequate tumor fraction at baseline for RNA-seq. Those with a reduction in FHV (using the pooled Mortensen method) after a week of metformin had significantly lower MATE2 expression (compared with those with no reduction in FHV, P = 0.024; Fig. 5) and a trend toward higher OCT3 expression (P = 0.17; Supplementary Fig. S6).

The use of hypoxia imaging to personalize RT is an emerging treatment paradigm. This trial selected upfront for patients with hypoxic locally advanced cervical cancer using FAZA-PET imaging and demonstrated that metformin decreased tumor hypoxia, with a trend toward improved DFS. The low survival rate observed in our control arm is consistent with the poor outcomes of patients with hypoxic cervical cancer reported in other studies (32, 33). Metformin in combination with pelvic (±paraaortic RT) was well tolerated in this trial, similar to that observed in a prostate trial (34).

Several window-of-opportunity studies have shown that metformin decreases Ki67 (35–40), whereas some have not observed a difference overall (41–44), akin to this study's finding. The lack of decrease in Ki67 is not as relevant to the main mechanism of metformin of interest (targeting hypoxia) in this trial. Our group previously showed metformin inhibited proliferation in most cancer cell lines and consistently inhibited oxygen consumption rate in a dose-dependent manner (31). Lower MATE2 (and higher OCT3) expression was associated with cancer cell response to metformin, consistent with the observations in this trial. MATE2 may be a biomarker of response to metformin.

Although retrospective data supporting the use of metformin in cancer therapy are abundant, prospective data are limited. The recent NRG-LU001 phase II RCT showed no difference in survival and the OCOG-ALMERA phase II RCT (stopped early due to slow accrual) reported worse overall survival with the addition of metformin to chemoradiation in locally advanced non–small cell lung cancer (45, 46). The dismal results observed in the metformin arm in the OCOG-ALMERA trial may be due to the larger tumor volume, only 23% of patients completing metformin per protocol, more radiation discontinuations, and fewer number of patients completing chemotherapy or receiving consolidation durvalumab in the metformin arm (47). In contrast, another phase II RCT demonstrated the addition of metformin to epidermal growth factor receptor tyrosine kinase inhibitors significantly improved progression-free survival in patients with advanced lung adenocarcinoma (48). The higher DFS observed in the metformin arm in this trial may be due to an improved response to chemoradiotherapy consequent to a reduction in tumor hypoxia. However, metformin has been reported to elicit anticancer effects through numerous mechanisms. We cannot rule out that these mechanisms, or simply chance, may have contributed to the superior outcomes in the metformin arm of this trial.

The main limitation of this study is the small sample size given premature closure. Accrual was challenging due to patient acceptance of the additional FAZA-PET/biopsies that were required at baseline and 1 week (especially given the discomfort associated with biopsies) and decreasing FAZA availability starting in 2019 due to increasing, competing production demand for other isotopes such as PSMA. The trial was also closed to accrual for long periods of time during the pandemic as per institutional policy. Other limitations include some patients with insufficient biopsy samples for translational analysis, and potential residual bladder spillover of FAZA signal despite Gaussian correction (30, 49). Histograms of SUV values before and after Gaussian correction are shown in Supplementary Fig. S7. The results should be interpreted with caution given that the trial was stopped early with about 30% of the original planned sample size accrued. Nonetheless, to our knowledge, this is one of the first trials with a hypoxia-targeted intervention that has selected upfront for patients with hypoxic tumors and demonstrated a response. Another trial evaluating dose escalation to the hypoxic region on PET in head and neck cancer also struggled with accrual (50), highlighting the importance for the scientific community to develop more efficient ways of measuring tumor hypoxia and/or boosting accrual.

In summary, metformin decreased cervical tumor hypoxia in this trial that selected for patients with hypoxic tumors. The additional FAZA-PET scans and tumor biopsies limited study accrual, and more pragmatic ways of measuring and selecting tumor hypoxia are needed for future studies.

K. Han reports grants from the Terry Fox Research Institute, Radiological Society of North America, and Canadian Institutes of Health Research during the conduct of the study as well as personal fees from AstraZeneca outside the submitted work; in addition, K. Han has a patent for circulating HPV DNA pending. T. Shek reports grants from CIHR-QIN and Ontario Institute for Cancer Research during the conduct of the study. T.-Y. Lee reports other support from GE Healthcare and Neusoft outside the submitted work. M. Pintilie reports grants from TFRI Program Project Grant during the conduct of the study. U. Metser reports personal fees from POINT biopharma Inc outside the submitted work. M. Milosevic reports grants from Terry Fox Research Institute during the conduct of the study. No disclosures were reported by the other authors.

The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

K. Han: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. A. Fyles: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, project administration, writing–review and editing. T. Shek: Data curation, software, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. J. Croke: Resources, data curation, investigation, writing–review and editing. N. Dhani: Resources, investigation, writing–review and editing. D. D'Souza: Resources, data curation, investigation, project administration, writing–review and editing. T.-Y. Lee: Resources, data curation, software, investigation, writing–review and editing. N. Chaudary: Resources, data curation, supervision, investigation, methodology, project administration, writing–review and editing. J. Bruce: Data curation, software, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. M. Pintilie: Resources, data curation, software, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. R. Cairns: Investigation, methodology, writing–review and editing. D. Vines: Investigation, visualization, methodology, writing–review and editing. S. Pakbaz: Data curation, investigation, visualization, methodology, writing–original draft, writing–review and editing. D. Jaffray: Conceptualization, resources, software, supervision, project administration, writing–review and editing. U. Metser: Investigation, writing–review and editing. M. Rouzbahman: Supervision, investigation, writing–review and editing. M. Milosevic: Conceptualization, resources, data curation, supervision, investigation, project administration, writing–review and editing. M. Koritzinsky: Conceptualization, resources, data curation, supervision, investigation, visualization, methodology, project administration, writing–review and editing.

This work was supported by Program Grants from the Terry Fox Research Institute, a Radiological Society of North America Research Scholar Grant (grant number RSCH1513) to K. Han, and in part through the Ontario Institute for Cancer Research and a Quantitative Imaging Network grant funded by the Canadian Institute of Health Research. The authors thank Pat Merante, the Radiation Medicine Clinical Research Program, and the Princess Margaret Phase II Consortium/Drug Development Program for coordinating this trial.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.