Oxygen-enhanced MRI Is Feasible, Repeatable, and Detects Radiotherapy-induced Change in Hypoxia in Xenograft Models and in Patients with Non–small Cell Lung Cancer

Most solid tumors contain subregions of hypoxia. The presence and degree of hypoxia has long been recognized as an important negative prognostic factor in cancer (1, 2). Furthermore, hypoxia predicts poor outcome following surgery (3), radiotherapy (4), and chemotherapy (5), and is associated with relapse and progression (6). Therefore, patients with highly hypoxic tumors tend to have decreased overall cancer survival following conventional therapies (7). Recently, there is evidence that targeted therapies such as immunotherapy may be less effective in hypoxic tumors (8, 9). For these reasons, there is an unmet need to target hypoxia to improve cancer outcomes (10).

Positron emission tomography (PET) methods have shown that imaging can provide serial noninvasive in vivo measurement of tumor hypoxia, can track hypoxia modification, and has potential to stratify patients and personalize their therapy (11–13). However, hypoxia PET approaches are limited currently to specialist centers, hindering widespread use in clinical trials and adoption into clinical practice (14). Proton MRI is used widely in the clinic, making it an attractive alternative to PET for measuring oxygen delivery and hypoxia in tumors. Oxygen-enhanced MRI (OE-MRI) measures the change in longitudinal relaxation rate of tissue protons (R₁; ref. 15). Preclinical studies have shown that the OE-MRI biomarker “perfused Oxy-R” distinguishes hypoxic tumor regions from well-oxygenated regions, where analysis is performed in perfused tumor subregions [identified by dynamic contrast-enhanced MRI (DCE-MRI); ref. 16].

We hypothesized that perfused Oxy-R could spatially map and quantify changes in tumor hypoxia induced by radiotherapy and chemoradiotherapy. To test this, we evaluated changes in perfused Oxy-R induced following radiotherapy or chemoradiotherapy in two xenograft models. We validated our findings with ex vivo IHC and oxygen consumption rate (OCR) assays. We then performed a first-in-human study in patients with non–small cell lung cancer (NSCLC) to demonstrate feasibility and repeatability and to see whether equivalent findings were observed. Finally, we investigated whether perfused Oxy-R could detect intersubject heterogeneity in hypoxia modification between patients to support a potential role for OE-MRI in personalized medicine.

The studies complied with UK guidelines on animal welfare in cancer research (17) and received approval from the Animal Welfare and Ethical Review Body. Preclinical experiments were performed in Calu-6 NSCLC xenografts and in U87 glioma xenografts.

Tumors were propagated by injecting either 0.1 mL of Calu-6 NSCLC cells (2 × 10⁷ cells/mL) or 0.1 mL of U87 cells (5 × 10⁶ cells/mL) intradermally in the lower back of CD1 nude mice. When tumors reached >200 mm³ by caliper measurement, mice were entered into the study (assigned as day 0). They were imaged while anesthetized using 2% isoflurane carried initially in medical air (21% oxygen), before switching the 100% oxygen as part of the OE-MRI protocol. Core temperature was controlled at 36°C.

After initial imaging (day 0), mice were randomized to pathology validation only, control/sham, or given treatment with tumor-localized radiotherapy. This was administered using a metal-ceramic MXR-320/36 X-ray machine (320 kV, Comet AG) under ambient conditions to restrained, nonanesthetized mice held in a lead-shielded support perpendicular to the source. Irradiation was delivered at a dose rate of 0.75 Gy/minute. Mice were turned around halfway through the procedure to ensure uniform tumor dose.

For Calu-6 xenograft MRI experiments (Supplementary Fig. S1A), groups were:

• (a) MRI pathology validation (N = 7) with single MRI scan, followed by tumor harvest;

• (b) Treatment effect: control group (N = 9); treatment with single 10 Gy fraction of radiotherapy (N = 9); treatment with fractionated 5 × daily 2 Gy radiotherapy with concurrent cisplatin on day 0; chemoradiotherapy (N = 6). Initial MRI was at day 0 after which therapy was administered. Subsequent MRI was performed at days 3, 6, and 10 in all groups and then at day 14 (control), day 18 and 24 (radiotherapy), and day 18 (chemoradiotherapy).

For U87 xenograft MRI experiments (Supplementary Fig. S1B), groups were:

• (a) MRI pathology validation (N = 10) with single MRI scan, followed by tumor harvest;

• (b) Treatment effect: control group (N = 10); treatment with single 10 Gy fraction of radiotherapy (N = 13). Initial MRI was at day 0 after which therapy was administered. Subsequent MRI was performed at day 3 only (control) or at days 3, 6, 10, and 24 (radiotherapy).

Imaging was performed using a volume transceiver coil on a 7 T Magnex instrument interfaced to a Bruker Avance III console and gradient system. After initial localizer and T₂-weighted anatomy scans, mice underwent coronal imaging. OE-MRI consisted of a variable flip angle (VFA) spoiled gradient echo (SPGR) acquisition to calculate native tissue R₁ (1/T₁; unit/s; TR = 30 ms; TE = 1.44 ms; α = 5°/10°/20°). This was followed by 42 dynamic T₁-weighted spoiled gradient recalled acquisitions in the steady state (SPGR; TR = 30 ms; TE = 1.44 ms; α = 20°; temporal resolution: 28.8 seconds). After 18 acquisitions, gas delivered through a nose cone was switched to 100% oxygen. Details of preclinical DCE-MRI acquisition are described in Supplementary Methods.

Regions of interest (ROI) were defined by a research radiographer (Y. Watson, 18 years’ experience) using Jim 7 (Xinapse Systems). For OE-MRI, voxel-wise ΔR₁ were calculated for each voxel, where ΔR₁ = R₁ while breathing oxygen (last 18 of 24 time points on 100% oxygen) minus R₁ on breathing air (18 time points on medical air). Tumor average ΔR₁ was calculated as the median value. Voxels were classified as oxygen-enhancing (Oxy-E) if the ΔR₁ was positive and significant (one-sided paired sample t test between preoxygen and oxygen-breathing time points, P < 0.05). All other voxels were defined as refractory to oxygen challenge (Oxy-R).

DCE-MRI was then used to classify tumor voxels as nonperfused [where initial area under the Gd contrast agent concentration–time curve at 60 seconds (IAUC₆₀; units mmol/L) was ≤0)] or perfused (where IAUC₆₀ >0). Together, these coregistered voxel-wise OE-MRI and DCE-MRI data were combined to distinguish perfused Oxy-E (normoxia), perfused Oxy-R (hypoxia), and nonperfused (necrosis) voxels, based on previously published methods (ref. 18; Supplementary Fig. S2A).

For Calu-6 xenografts, three datasets were used. The first pathology experiment compared the hypoxic fractions measured on an IHC image from the center of each tumor with a single MRI slice acquired in the same plane from the corresponding tumor region, to validate OE-MRI measurements of hypoxia (Supplementary Fig. S1A). The next two pathology experiments evaluated change in hypoxia following therapy. We evaluated tumor size–matched data obtained at tumor harvest from the mice that underwent the MRI study (details above; Supplementary Fig. S1A) and separate time-matched data acquired in nonimaged mice at day 10, guided by the MRI experiment. This additional experiment had control (N = 9), and tumors treated with single 10 Gy fraction of radiotherapy (N = 6; Supplementary Fig. S1C). For U87 xenografts, size-matched data were obtained at tumor harvest from the mice that underwent MRI (details above; Supplementary Fig. S1B). Pimonidazole data acquisition and analysis details are described in Supplementary Methods.

A fourth Calu-6 xenograft cohort was used separate to those used for MRI and pathology analysis. Tissue biopsy preparation protocols were adapted from a previously published method (19) and are described in Supplementary Methods. Separate mouse cohorts were used: Calu-6 xenograft groups were time-matched control (N = 7), size-matched control (N = 5), and treated with single 10 Gy fraction of radiotherapy (N = 6) (Supplementary Fig. S1D). Five consecutive basal OCR measurements were performed for all samples.

Because previous work (19) has shown that percentage viability and necrotic fraction of tumors is directly proportional to observed oxygen consumption, we used hematoxylin and eosin (H&E) staining to quantify necrotic fraction for each sample using a scoring method based on the level of nuclear staining. Following OCR measurements, the samples were imaged using Oxford Optronix GelCount (Oxford Optronix Ltd.). Necrotic areas were calculated using ImageJ software. Whole-field H&E images, representative biopsy sections, and the scoring method devised to quantify necrosis are shown in Supplementary Fig. S3.

Patients with NSCLC eligible for curative-intent radiotherapy alone or combined chemoradiotherapy (either concurrent or sequential) were recruited prospectively from The Christie NHS Foundation Trust following local research and development and institutional review board approval (reference: 15/NW/0264, CPMS ID: 18870). This study was conducted in accordance with the Declaration of Helsinki.

All patients were ≥18 years old, had histologically or cytologically confirmed NSCLC, and provided written informed consent. Eligible participants also had Eastern Cooperative Oncology Group (ECOG) performance score of ≤2, serum creatinine <120 μmol/L, or calculated creatinine clearance (Cockcroft−Gault) ≥30 mL/minute. Patients with distant metastasis were included only if eligible for curative-intent therapy. Patients with contraindications to MRI were excluded. Coexistent chronic obstructive pulmonary disease was allowed, but patients were required to have adequate lung function as part of standard radiotherapy workup (forced expiratory volume in 1 second greater than 1 liter, or >40% predicted value).

We examined safety (defined as no adverse events reported or detected on clinical examination) and tolerability (defined as completion of the imaging protocol) in a development cohort (N = 6 patients). On the basis of these data, the study recruited an expansion cohort to evaluate MRI biomarker repeatability and sensitivity to treatment effect (N = 17 patients). The study design is summarized in Supplementary Fig. S4.

Radiotherapy planning was performed using three-dimensional or four-dimensional CT. Treatment was delivered with intensity-modulated radiotherapy on a linear accelerator (55 Gy in 20 daily fractions or 60–66 Gy in 30–33 daily fractions). Patients receiving concurrent chemoradiotherapy had cisplatin and etoposide administered. Patients receiving sequential chemoradiotherapy had cisplatin or carboplatin administered with either gemcitabine (if squamous cell carcinoma) or pemetrexed (if adenocarcinoma). In sequential chemoradiotherapy-treated patients, repeat imaging was performed after completion of chemotherapy. Toxicity was assessed by a clinical oncologist (A. Salem; 8 years’ experience) prior to and after each scan. No predefined standardized follow-up diagnostic scans were mandated after completion of the research imaging protocol.

All MRI data were acquired free breathing on a 1.5 T whole body scanner (Philips Achieva, Philips Medical Systems) using the Q Body (OE-MRI) and Sense XL Torso (DCE-MRI) coils. After initial localizer and anatomic scans, patients underwent OE-MRI and DCE-MRI. Scan duration was approximately 45 minutes (summarized in Supplementary Fig. S5A). All sequences were colocalized and acquired in the coronal plane with FOV 450 mm × 450 mm × 205 mm; and 5-mm thick slices. Anatomic scans’ in-plane resolution was 1.76 mm × 1.76 mm, whereas OE-MRI and DCE-MRI images had in-plane resolution of 4.69 mm × 4.69 mm.

For OE-MRI, R₁ was calculated using a series of coronal 3D single-shot inversion recovery–prepared SPGR acquisitions (TR = 2.1 ms; TE = 0.50 ms; α = 6°, number of excitations = 5, TI: 10, 50, 300, 1,100, 2,000, 5,000 ms). Dynamic images were acquired at TI of 1,100 ms. For the first 6 patients in the protocol development cohort, 12 measurements were acquired while breathing medical air (21% oxygen) to determine native R₁ prior to oxygen challenge. This was increased to 18 for all subsequent patients following an interim study review. Measurements on air were followed, in all patients, by 48 measurements after gas switch to 100% oxygen (flow rate: 15 L/minute) and a final 30 following switch back to medical air (temporal resolution: 10 seconds). Gases were delivered using a tight-sealed, nonrebreathing Intersurgical EcoLite Hudson facemask (Intersurgical Ltd) via a gas blender, allowing switching between gases. A two-step quality assurance procedure was applied to ensure successful delivery of the gas challenge. First, 100% oxygen challenge delivery was confirmed in all scans at the time of image acquisition via a gas-sensing probe placed inside the facemask and connected to a gas analyzer (ADInstruments Pty Ltd; Supplementary Fig. S5B). Second, R₁ was measured in the descending thoracic aorta, outlined by a clinical oncologist (A. Salem), to provide a positive control by detecting indirect oxygen input function to the tumor. Details of clinical DCE-MRI acquisition are described in Supplementary Methods. Representative preprocessed OE-MRI and DCE-MRI acquisitions are shown in Supplementary Video S1A.

Tumor ROIs were defined by a clinical oncologist (A. Salem) and reviewed by a board-certified clinical radiologist (J.P.B. O′Connor; 14 years’ experience), using Jim 7 on coronal pre- and post-Gd T₁-weighted images, guided by diagnostic [¹⁸F]fluorodeoxyglucose (FDG) PET CT images.

For OE-MRI, whole tumor and voxel-wise ΔR₁ were calculated, where ΔR₁ = R₁ while breathing oxygen (mean of last 18 time points on 100% oxygen) minus R₁ on breathing air (mean of acquisitions on medical air). Voxels were classified as Oxy-E if the ΔR₁ was positive and significant (one-sided paired sample t test between preoxygen and oxygen-breathing time points, P < 0.05). All other voxels were defined as Oxy-R.

For DCE-MRI, tumor median IAUC₆₀ was also calculated. Voxels were classified as perfused when the area under the Gd contrast agent concentration curve was >0 (one-sided paired sample t test, P < 0.005; ref. 20). Voxel-wise OE-MRI and DCE-MRI data were combined to distinguish perfused Oxy-E (normoxia), perfused Oxy-R (hypoxia), and nonperfused voxels (necrosis), based on previously published methods and with an identical approach to that used in the preclinical experiments (ref. 18; Supplementary Fig. S2B), using data corrected for breathing and bulk patient motion (details in Supplementary Methods; Supplementary Video S1B).

All statistical analyses were performed in IBM SPSS Statistics version 22.0 (IBM Corp.). For all preclinical data, tumor growth rate comparisons were assessed using a log-rank test for time to double tumor volume. Pearson correlation analysis was used to assess relationships between pretreatment MRI parameters and IHC. Two-sample unpaired t tests were performed to assess differences in mean subregion fractions and volumes (perfused Oxy-R, perfused Oxy-E, and nonperfused), IHC or OCR assay between control and treated groups at specific timepoints of interest. Corresponding P values are reported and were considered statistically significant when P < 0.05 (two-sided). No corrections for multiple comparisons were made.

For both preclinical and clinical MRI, changes in perfused Oxy-R volume of >50% were considered significant. This threshold was based on similar studies of other functional imaging biomarkers, where parameter changes of 40%–50% are considered significant (21). For clinical data, descriptive statistics were presented using median and 95% confidence intervals (CI) or SD.

For the clinical study, prestudy sample size calculations were not performed as this was a first-in-human proof-of-concept study. Single measures two-way mixed effects model absolute agreement interclass correlation coefficient (ICC) and coefficient of variation (CoV) were calculated for summary OE-MRI parameters and the hypoxia biomarker perfused Oxy-R between repeat acquisitions (together with their respective 95% CI). Recommendations (22) that interpret ICC values between 0.75 and 0.9 and greater than 0.9 as indicative of good and excellent repeatability, respectively, were used. In addition, Bland–Altman analysis was performed to calculate bias and 95% limits of agreement (LoA = 1.96 × SD).

Calu-6 xenografts (23, 24) and U87 xenografts (25) were chosen for the study because these models contain moderate to high levels of hypoxia. We measured the MRI perfused Oxy-R fraction (Supplementary Fig. S2A) on one central slice of the tumor to provide in vivo quantification of hypoxia. Perfused Oxy-R fraction correlated significantly with the hypoxic fraction measured by pimonidazole IHC, for both Calu-6 xenografts (R² = 0.700; P = 0.019; Fig. 1A) and U87 xenografts (R² = 0.447; P = 0.035; Fig. 1B). No relationship was detected between pimonidazole IHC measurement of hypoxia and IAUC₆₀ (Supplementary Fig. S6A and S6B). These data provide pathologic validation that the OE-MRI biomarker perfused Oxy-R fraction, but not DCE-MRI–derived IAUC₆₀, quantifies tumor hypoxia in Calu-6 and U87 xenografts.

Data were obtained across a size range that was representative of all subsequent experiments (tumor sizes: 191–974 mm³ measured by MRI volumetrics). No significant relationship was seen between pimonidazole IHC measurement of hypoxia and tumor size (representative images in Fig. 1C and D; Supplementary Fig. S6C and S6D). This highlights that tumor size and hypoxia are independent of one another in both xenograft models used in this study.

Calu-6 xenografts treated with either radiotherapy (single fraction of 10 Gy) or chemoradiotherapy (5 × 2 Gy fractions plus cisplatin) exhibited significant growth delay relative to control xenografts (P < 0.001), assessed by measuring time to double in tumor volume.

To understand the dependence of three-dimensional growth inhibition on tumor hypoxia, we measured change in perfused Oxy-R volume within each tumor. Because growth inhibition was clearly apparent by day 10 (Fig. 2A), we assessed OE-MRI at this time point. By day 10, perfused Oxy-R volume was reduced in xenografts treated with radiotherapy (P = 0.029) or fractionated chemoradiotherapy (P = 0.047), relative to control (Fig. 2B). Hypoxia modification persisted until tumor harvest (chemoradiotherapy group at day 18; radiotherapy group at day 24; both P < 0.001). OE-MRI maps showed spatially coherent changes in hypoxia (Fig. 2C). In distinction, median change in ΔR₁, a commonly reported biomarker in OE-MRI (26), showed borderline significance only at day 6 and 10 (Supplementary Fig. S7).

We analyzed pimonidazole IHC performed at day 10 in a separate cohort of Calu-6 xenografts. Lower hypoxic fraction was seen in tumors treated with radiotherapy (P = 0.026), relative to time-matched controls (Fig. 2D). Next, we performed pimonidazole IHC analysis of the xenografts undergoing MRI. Lower hypoxic fractions in radiotherapy (P = 0.042) and chemoradiotherapy (P = 0.041) were found in treated tumors, relative to size-matched control (Fig. 2E and F; Supplementary Fig. S8).

Experiments were repeated in U87 tumors. Xenografts treated with single fraction of 10 Gy radiotherapy exhibited significant growth delay, relative to control xenografts (P < 0.001; Supplementary Fig. S9A). Perfused Oxy-R volume was decreased in radiotherapy treated (P = 0.017) xenografts, relative to control by day 3 and this reduction persisted until day 10 (Supplementary Fig. S9B). Sample OE-MRI maps are shown (Supplementary Fig. S9C). Pimonidazole IHC confirmed that lower hypoxic fractions were observed in radiotherapy-treated tumors (P = 0.002), relative to size-matched control (Supplementary Fig. S9D). Collectively, these data provide pathologic validation that the OE-MRI biomarker perfused Oxy-R volume detects hypoxia modification induced by radiotherapy in two xenograft models.

To investigate the mechanistic basis for the reduction in hypoxia following treatment, we examined dynamic change in tumor vascular status measured on DCE-MRI from day 0 to tumor harvest and also single measurement of OCR at day 10. DCE-MRI data showed that IAUC₆₀ was increased significantly at days 6–10 in Calu-6 xenografts in both the radiotherapy- and chemoradiotherapy-treated groups, relative to control (P < 0.001). These changes also persisted to day 24 in the group treated with single 10 Gy radiotherapy (Fig. 3A and B). Equivalent data were observed in U87 xenografts (Supplementary Fig. S10).

Fresh biopsy samples were taken from Calu-6 xenografts in three groups: control time-matched tumors (average volume 637 ± 52 mm³) at day 10 after sham treatment; control size-matched tumors (average volume 242 ± 27 mm³); and radiotherapy-treated tumors at day 10 post single 10 Gy fraction (average volume 271 ± 15 mm³). The mean of the five consecutive basal OCR measurements performed over 30 minutes under ambient conditions was significantly reduced by radiotherapy compared with time- (P = 0.008) and size-matched (P < 0.001) controls (Fig. 3C). Necrotic scoring revealed no significant differences between the irradiated and time- or size-matched controls (Fig. 3D), indicating equivalent levels of tissue viability between the three groups.

Large intra- and intertumor heterogeneity of OCR measurements were observed within both time- and size-matched control Calu-6 xenografts. Both types of heterogeneity were markedly reduced in tumors treated with radiotherapy (Fig. 3E). In all, 31.3% of control biopsy samples at day 10 (time-matched) had OCR of >25 pmol/minute per normalized unit and this increased to 59.4% in size-matched control biopsy samples. In distinction, only 13.0% of radiotherapy-treated biopsy samples had residual OCR of >25 pmol/minute per normalized unit. These data provide evidence that the hypoxia modification detected by OE-MRI and pimonidazole IHC in Calu-6 is likely due to reduced overall OCR, in particular removing those subregions with very high localized OCR.

Twenty-three patients with stage I–IV NSCLC were recruited (Supplementary Fig. S4; Supplementary Table S1). The protocol was safe and well-tolerated (Supplementary Table S2). Significant changes in R₁ were observed following oxygen inhalation in the aorta of all patients demonstrating technique feasibility and providing quality control (Supplementary Fig. S11).

Whole tumor median R₁ increased with oxygen inhalation in 11 of 15 patients (individual all P < 0.05 for 11 patients, oxygen inhalation versus air). However, all tumors demonstrated some degree of intratumor spatial heterogeneity in oxygen-induced ΔR₁, with three patterns of tissue classification revealed. In 3 of 15 tumors, the whole tumor median ΔR₁ was significant and had similar temporal evolution and magnitude as the ΔR₁ seen in the aorta. These tumors had no measurable regional hypoxia (absence of perfused Oxy-R; representative example in Fig. 4A). The remaining 12 of 15 tumors had spatially coherent regions of hypoxia; this included 8 of 15 tumors with an overall significant ΔR₁ that was partially attenuated compared with aorta ΔR₁ (representative example in Fig. 4B), and 4 of 15 tumors where the proportion of hypoxic tissue was high enough to substantially attenuate the median whole tumor ΔR₁ so that it was not significant (representative example in Fig. 4C).

Ten patients underwent repeat MRI (within 7 ± 5 days; Supplementary Fig. S4) before radiotherapy was administered to measure biomarker precision. The perfused Oxy-R volume, measuring tumor hypoxic volume, demonstrated excellent repeatability with ICC of 0.961 (95% CI, 0.858–0.990) and CoV of 25.880% (Fig. 4D). Additional repeatability results can be found in Supplementary Table S3.

Consistent classification of tumors as either entirely normoxic (n = 3; perfused Oxy-R volume = 0) or containing some hypoxia (n = 7; perfused Oxy-R volume >0) was concordant between the two preradiotherapy scans. Finally, visual inspection revealed that MRI hypoxia mapping was spatially repeatable in tumor, nodal, and distant metastatic lesions across a range of tumor and hypoxic volumes (Fig. 4E). Collectively, these data demonstrate that OE-MRI can identify and map hypoxia in clinical NSCLC tumors, when subregional tissue classification is performed using perfusion data.

Twelve patients were imaged at day 14 ± 4 of radiotherapy, in addition to pretreatment imaging (Supplementary Fig. S4). No significant change was detected in tumor volume at this time point (P = 0.159; Table 1), but we hypothesized that radiation would induce measurable changes in hypoxia within this window, based on previous clinical PET studies in head and neck cancer and lung cancer (11–13).

The perfused Oxy-R volume, indicating hypoxic tumor volume, decreased in the patient cohort from 4.16 cm³ (95% CI, 0–10.6 cm³) at baseline to 3.23 cm³ (95% CI, 0–9.41 cm³) at mid-treatment (P = 0.015). In distinction, the increase in median ΔR₁ at day 14 was not significant (P = 0.097). MRI parameter changes are summarized in Table 1. These data show that the MRI biomarker perfused Oxy-R volume detected reduction in tumor hypoxia in patients with NSCLC consistent with data in the two xenograft models of cancer.

Previous PET studies have reported variable hypoxia modification in patients with NSCLC and other cancer types (11, 27, 28). We examined the variation in hypoxia modification in Calu-6 xenografts at day 10. Those xenografts with >50% change in hypoxic volume were designated as exhibiting significant hypoxia modification. While radiotherapy-treated tumors showed overall reduction in hypoxia at the cohort level, relative to control, a >50% decrease in hypoxic volume was seen in only 9 of 15 tumors (Fig. 5A). These “hypoxia-modified tumors” had higher perfusion and permeability (denoted by median DCE-MRI IAUC₆₀) at baseline than tumors that did not demonstrate hypoxia modification (P = 0.035; Fig. 5B).

Similarly, variation was seen in response to (chemo)radiotherapy in the clinical tumors, with 8 of 12 tumors having a decrease in hypoxic volume >50%. In distinction, two tumors had an increase in hypoxic volume >50% and two tumors showed no change above the 50% threshold (Fig. 5C). Tumors with significant reduction in hypoxia had higher median IAUC₆₀ (P = 0.011) at baseline than other tumors (Fig. 5D). There was no difference in baseline tumor size or hypoxic volume between the hypoxia “modified” and nonmodified tumors. These data show that OE-MRI can distinguish tumors that demonstrate hypoxia modification following radiotherapy from those that have persistent hypoxia.

There is an unmet need to develop noninvasive biomarkers of tumor hypoxia to monitor response for anticancer treatments. This is particularly important for radiotherapy and chemoradiotherapy, as meta-analysis has identified the negative impact of hypoxia detected via imaging, on radiotherapy outcome (29).

Imaging enables repeated whole tumor sampling that overcomes the limitations of tissue-based hypoxia quantification (subsampling and single measurement) and biofluid assays (inability to distinguish heterogeneity between different tumors in an individual). Therefore, translational imaging tests could enable patient selection and stratification in trials of combined radiotherapy and hypoxia-targeted therapies (10), to adapt radiotherapy dose intensification, and to monitor emergence of radio-resistant hypoxic cancer cells (14).

Here, we report the first-in-human evidence that perfused Oxy-R can identify, map, and quantify change in hypoxia induced by a therapeutic intervention. We chose to evaluate patients with NSCLC as a proof-of-principle study because lung cancer is the leading cause of cancer mortality worldwide (30); around 90% of cases are of the NSCLC subtype (31, 32); and tumor hypoxia is associated with poor survival in patients with NSCLC (33). Furthermore, radiotherapy plays an important role in the treatment of all stages of NSCLC (34) and imaging biomarkers are attractive in this setting due to limited access to tumor tissue material in radiotherapy-treated patients.

Our bench-to-bedside approach began by testing whether OE-MRI biomarkers could detect radiotherapy-induced changes in tumor biology in two xenograft models of NSCLC (Calu-6) and high-grade glioma (U87). We showed that OE-MRI, combined with assessment of perfusion, identified and mapped regional differences in hypoxic and normoxic tumor in both models, using IHC validation (18, 35, 36). Next we showed that radiotherapy resulted in significant reduction in perfused Oxy-R volume in Calu-6 xenografts after 10 days, relative to control tumors, with both high-dose single fraction radiotherapy and a fractionated chemoradiotherapy regimen that more closely mimics clinical therapy. These differences persisted for 18–24 days, indicating that while regrowth occurred eventually in the radiotherapy and chemoradiotherapy-treated tumors, they had less hypoxic tissue than untreated tumors of a similar size. Our findings were validated by time- and size-matched IHC quantification of hypoxia and quantification of oxygen consumption. Confirmatory data were obtained in U87 tumors. In all studies, an OE-MRI biomarker that was sensitive to spatial heterogeneity (37) between hypoxic and nonhypoxic tumor subregions, perfused Oxy-R, was most closely related to hypoxia on pathology validation and was most sensitive to therapeutic modification of hypoxia.

These data are the first to show that perfused Oxy-R can track group differences in hypoxia modification following therapy in vivo. Mechanistic insight was provided through analysis of the DCE-MRI data acquired during MRI treatment experiments in Calu-6 and U87 xenografts showed that increases in tumor perfusion (measured by IAUC₆₀) were detected with radiation-based therapies, relative to control. Furthermore, ex vivo OCR analysis in Calu-6 xenografts provided evidence that oxygen consumption was reduced at day 10 following radiotherapy. These data suggest that both increased oxygen delivery and reduced oxygen consumption contribute to radiation-induced hypoxia modification.

We then translated the OE-MRI technique to humans, demonstrating safety, tolerability, and measurement feasibility in patients with NSCLC. Measurement precision was evaluated by assessing biomarker repeatability. Perfused Oxy-R (hypoxic) volume had excellent test–retest precision with high ICC and compared favorably with previously published studies of MRI biomarker precision in lung and other tumors (38–41). The intratumor spatial distribution of hypoxia was comparable between repeat scans on visual inspection. Lack of absolute visual repeatability could be due to methodologic factors (including imperfect image registration or inconsistent target volume definition) or biological factors (including cyclical hypoxia; ref. 14). Furthermore, OE-MRI provided consistent classification of tumors as hypoxic or not; in this respect, it outperformed PET imaging with the tracer [¹⁸F]FMISO in patients with NSCLC (42). These data are the first to demonstrate repeatability of OE-MRI biomarkers that quantify tumor hypoxia in patients with cancer.

Our clinical data provide the first evidence that OE-MRI can track tumor hypoxia modification induced by a therapy in patients. This finding concurs with previous PET imaging studies in patients with NSCLC and other cancers which have shown that the volume of hypoxic tumor can increase, decrease, or remain unchanged within the initial few weeks following chemoradiation (11).

Finally, because PET studies in patients with head and neck cancer suggest that the persistence of hypoxia during chemoradiotherapy, rather than pretreatment levels of hypoxia, predict clinical outcome (12, 13), we investigated the potential for OE-MRI to have application in personalized medicine. We demonstrated that perfused Oxy-R distinguished those xenografts and NSCLC patient tumors whose hypoxic tumor volume decreases following radiotherapy-based treatments from those that remained hypoxic (and in some cases worsened). Of note, hypoxia modification was observed in tumors with relatively high pretreatment vascular perfusion and permeability, measured by IAUC₆₀. This suggests that a multimodal imaging, genomic, and tissue pathology biomarker panel, performed before and during treatment, may be required to fully understand, predict, and monitor hypoxia modification in the clinic (43). However, further prospective powered studies are required to confirm any role for hypoxia imaging in stratified medicine.

While this study focused on the role of perfused Oxy-R in measuring hypoxia changes induced by radiotherapy, these findings have broad implications. Perfused Oxy-R has potential to monitor direct pharmacologic targeting of hypoxia (44) and to inform how resistance to immunotherapy relates to the hypoxic tumor stem cell niche (9). Furthermore, perfused Oxy-R may guide radiotherapy dose intensification (dose painting; refs. 45, 46) and has unique potential for clinical applications such as real-time adaptive radiotherapy on MR Linac systems (47). For each of these applications, large studies are required to qualify the prognostic value of perfused Oxy-R as a biomarker of hypoxia and its ability to predict therapy response.

In summary, this study provides substantial new information to advance clinical translation of the OE-MRI and DCE-MRI biomarker perfused Oxy-R volume (48) and supports further effort to qualify this biomarker for use in clinical trials.

G.J.M. Parker is an employee of and holds ownership interest (including patents) in Bioxydyn Limited. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. Salem, A. Jackson, G.J.M. Parker, C. Faivre-Finn, K.J. Williams, J.P.B. O′Connor

Development of methodology: A. Salem, R.A. Little, A. Latif, Y. Watson, G.J.M. Parker, C. Faivre-Finn, J.P.B. O′Connor

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. Salem, A. Latif, M. Babur, S. Cheung, Y. Watson, V. Tessyman, G. Ashton, K.J. Williams, J.P.B. O′Connor

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Salem, R.A. Little, A. Latif, A.K. Featherstone, I. Peset, Y. Watson, H. Mistry, J.C. Matthews, R.G. Bristow, C. Faivre-Finn, K.J. Williams, J.P.B. O′Connor

Writing, review, and/or revision of the manuscript: A. Salem, R.A. Little, A. Latif, M. Babur, S. Cheung, V. Tessyman, H. Mistry, J.C. Matthews, M.-C. Asselin, R.G. Bristow, A. Jackson, G.J.M. Parker, C. Faivre-Finn, K.J. Williams, J.P.B. O′Connor

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Salem, A.K. Featherstone, C. Behan, A. Jackson, J.P.B. O′Connor

Study supervision: A. Salem, G.J.M. Parker, C. Faivre-Finn, K.J. Williams, J.P.B. O'Connor

Other (supervision of A.K. Featherstone): J.C. Matthews

This study was funded by the Engineering and Physical Sciences Research Council and Cancer Research UK (CRUK; grant C8742/A18097; 2013–2018). This study was supported by researchers at the NIHR Manchester Biomedical Research Centre. A. Salem was supported by an Early Careers Fellowship from Manchester Cancer Research Centre and The Christie Charity and by a University of Manchester Presidential Fellowship for Clinicians. J.P.B. O′Connor is supported by Cancer Research UK (C19221/A15267, C19221/A22746). Clinical OE-MRI was supported by Philips through a research agreement with The University of Manchester.

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.