Oxygen-Enhanced MRI Accurately Identifies, Quantifies, and Maps Tumor Hypoxia in Preclinical Cancer Models

Hypoxia is a common feature of most solid malignancies, resulting from an imbalance between oxygen delivery and consumption (1). Tumor hypoxia is associated with the activation of angiogenesis and with metastatic potential (2). Consequently, tumor hypoxia is an important negative prognostic factor (3–5). Tumor hypoxia also mediates resistance to radiotherapy and to some chemotherapy agents and is an independent predictor of treatment failure. Strategies to counteract tumor hypoxia using either radiosensitizers or hypoxia-activated cytotoxic agents are currently being evaluated (6).

A noninvasive imaging biomarker that identifies the presence of hypoxia and measures its extent and spatial distribution within a tumor would help facilitate personalized medicine. However, no such tool is currently available. Proton (¹H) MRI is used routinely in clinical medicine, making it an attractive noninvasive modality for measuring oxygen delivery and hypoxia in tumors. Current ¹H MRI methods of imaging hypoxia have focused on either dynamic contrast–enhanced MRI (DCE-MRI) or R₂*-based intrinsic susceptibility imaging, also referred to as blood oxygenation level dependent (BOLD) imaging (7).

In DCE-MRI, administration of a gadolinium-based contrast agent allows estimation of blood vessel flow and permeability, providing an indirect measurement of oxygen delivery and necrosis (8). In BOLD imaging, paramagnetic deoxyhemoglobin molecules in erythrocytes create magnetic susceptibility perturbations around blood vessels, which increase the local transverse MRI relaxation rate (R₂*; units ms⁻¹). The value of tumor R₂* decreases when blood oxygen saturation increases following inhalation of hyperoxic gas (9). Unfortunately, both DCE-MRI and BOLD imaging have significant limitations that have hindered implementation as clinical biomarkers of hypoxia (10). Neither measure hypoxia directly. Further, BOLD measurements are affected by the presence of hemorrhage, by change in vessel geometry and by artifact in air/soft tissue interfaces, such as in the lungs and bowel (7).

Oxygen-enhanced MRI (OE-MRI) is a distinct ¹H MRI method for quantifying tumor oxygen delivery. Here, the MRI longitudinal relaxation rate (R₁; units s⁻¹) is sensitive to changes in the level of molecular oxygen (O₂) dissolved in blood plasma or interstitial tissue fluid (11, 12). When hyperoxic gas is inhaled, excess oxygen is carried in the blood plasma in tissues with adequate perfusion. Because well-oxygenated tissue has near complete saturation of hemoglobin molecules (13), the excess delivered oxygen remains dissolved in blood plasma and interstitial tissue fluid, where it increases the R₁ value (Fig. 1A). The change in R₁ (ΔR₁) observed is theoretically proportional to magnitude of change in dissolved O₂ concentration for a given voxel (Fig. 1B; ref. 11). Several previous studies have reported positive average values of ΔR₁ following oxygen inhalation in preclinical models of cancer (14–22) and in human tumors (20, 23–25). Importantly, although oxygen relaxivity, baseline R₁, and signal-to-noise (SNR) ratios show some variation with field strength, the technique is feasible on both preclinical and clinical MRI platforms (22, 26).

Measuring positive ΔR₁ in OE-MRI quantifies and maps oxygen delivery by identifying tissue with fully saturated hemoglobin, but does not directly identify tissue hypoxia per se. Tumor subregions refractory to oxygen challenge are considered to have low hemoglobin oxygen saturation and so excess delivered O₂ molecules bind preferentially to hemoglobin molecules but do not significantly alter plasma pO₂ (Fig. 1C; ref. 10). Several preclinical OE-MRI studies have reported that some tumor subregions are refractory to hyperoxic gas challenge because they had no positive ΔR₁ change (14, 19–21). If these regions are perfused yet lacking in oxygen enhancement, then OE-MRI should identify regions of tumor hypoxia (Fig. 1D). In this study, we tested the hypothesis that measuring the fraction of each tumor refractory to oxygen challenge—determined by having absent positive ΔR₁ (hereafter termed “Oxy-R fraction”)—would enable the identification, quantification, and mapping of tumor hypoxia with MRI in vivo.

In vitro validation experiments were performed on the same 7T system as used for the in vivo studies. The phantom consisted of four 5 mm test tubes (Sigma-Aldrich) with solutions of gadopentetate dimeglumine (Magnevist; Schering) serially diluted in water to yield final concentrations of 0.01, 0.05, 0.07, and 0.09 mmol/L gadolinium and an expected range of T₁ between 1,400 and 2,500 ms based on the relaxivity of Magnevist at 7T at 37°C (27). Test tubes were placed inside a plastic container filled with dental paste to reduce susceptibility effects and to ensure efficient heat transfer for constant temperature. Three repeated measurements of R₁ were measured for each tube on three separate days.

All experiments were performed in compliance with licenses issued under the UK Animals (Scientific Procedures) Act 1986 and following local ethical review. Studies were compliant with the United Kingdom National Cancer Research Institute Guidelines for Animal Welfare in Cancer Research (28) and with the Animals in Research: Reporting In Vivo Experiments guidelines (29).

Three cell lines exhibiting differing vascular and hypoxic phenotypes in vivo were used. Parental 786-0 renal carcinoma cells (ATCC, LCG Standards; purchased 2011; 786-0-par), and cells established from a sunitinib refractory subcutaneous 786-0 xenograft (786-0-R; ref. 30), were cultured in RPMI supplemented with 10% (v/v) FCS (Gibco, Life Technologies). Tumors were propagated by injecting 3 × 10⁶ cells in 100 μL of sterile PBS into the flanks of 8-week-old female C.B17-scid mice under isoflurane anesthesia. SW-620 colorectal carcinoma cells (ATCC; lot #8924081, purchased 2005) were cultured in DMEM supplemented with 10% FCS. Tumors were propagated by injecting 5 × 10⁶ cells in 100 μL of sterile PBS into the flanks of 8-week-old female athymic NCr-Foxn1^nu mice. Immediately prior to in vivo implantation, all cells tested negative for mycoplasma infection and the number of short tandem repeats (STR) present at 7 to 10 loci were assessed by PCR to provide STR profiles, from which cell line authenticity was confirmed.

Tumor size was monitored using callipers and the formula for ellipsoid volume, (π/6) × L × W × D, where L, W, and D are the largest orthogonal dimensions of the ellipsoid. Tumors were typically used for experimentation at a volume of approximately 400 mm³.

The fiber optic oxygen sensing device OxyLite (Oxford Optronix; ref. 31) was used to measure tumor partial pressure of oxygen (pO₂; units mm Hg). A heat mat maintained core body temperature, and gas delivery was at 2l/min via a nosepiece. Two dual pO₂ and temperature probes were positioned within the tumor tissue. Animals initially breathed medical air, followed by 100% oxygen to mimic the gas changes induced during the MRI experiments. Data acquisition and averaging were performed using Chart version 5 (AD Instruments). Tissue pO₂ data were collected the day after corresponding MRI data.

Anesthesia was induced with a 10 mL/kg intraperitoneal injection of fentanyl citrate (0.315 mg/mL) plus fluanisone (10 mg/mL; Hypnorm; Janssen Pharmaceutical Ltd.), midazolam (5 mg/mL; Hypnovel; Roche), and sterile water (at 1:1:2 ratio; ref. 16). Mice were positioned in a 3 cm birdcage coil on a custom-built platform to isolate the tumor, which was surrounded by dental paste (3M) to minimize motion and susceptibility artifacts. Gas delivery (medical air or 100% oxygen) was continuous at 2l/min through a nosepiece. Warm air maintained animal core temperature at 37°C. Lateral tail vein cannulation was performed with a heparinized 27G butterfly catheter (Venisystems; Hospira) to enable the remote intravenous administration of gadolinium contrast agent in DCE-MRI studies or when vasomodulator was administered.

All MRI data were acquired on a 7T horizontal bore microimaging system (Bruker Instruments). Localization was performed using a multislice turboRARE T₂-weighted sequence and was followed by shimming over the tumor (32). All sequences were acquired using 30 mm × 30 mm FOV (128 × 128 matrix; in plane resolution 0.234 mm) for a single 1-mm-thick slice. Experimental protocols are detailed in Supplementary Fig. S1. Sequences used were as follows:

OE-MRI: Inversion recovery (IR) True-FISP images were used to calculate R₁ (TR, 2.4 ms; scan TR, 10 seconds, 48 inversion times spaced 38.8 ms apart with initial inversion time of 106.2 ms; TE, 1.2 ms; α 60°). High SNR images were used to obtain highly accurate R₁; these required 8 signal averages and took 10 minutes and 40 seconds to perform. Dynamic images were used to quantify the temporal onset of R₁ changes induced by switching between air and 100% oxygen. The dynamic images required 2 signal averages and took 2 minutes and 40 seconds to perform.

BOLD: Multiple gradient echo images were used to calculate R₂* (TR, 200 ms; 8 echo times; TE, 6.2 to 28.2 ms with 3.1 ms echo spacing; 8 signal averages); duration 3 minutes and 25 seconds.

DCE-MRI: Data were collected using a modified True-FISP sequence (TR, 2.4 ms; scan TR, 10 seconds, 8 inversion times spaced 155 ms apart, initial inversion time 108 ms, TE, 1.2 ms; α 60°; one signal average, temporal resolution 20 seconds). After five baseline measurements, 0.1 mmol/kg bolus of the gadolinium-based contrast agent Magnevist (2 mL/kg 50 mmol/L solution) was injected intravenously at 2 mL/min using a power injector with duration of 10 minutes.

Regions of interest (ROI) were drawn around the tumor on the T₂-weighted images by an experienced operator (Y. Jamin). ROIs were transferred to all OE-MRI, BOLD, and DCE-MRI data. Voxel-wise and median values of R₁ and R₂* were calculated for each map using a Bayesian maximum a posteriori approach, with in-house software.

For OE-MRI, the voxel-wise ΔR₁ was calculated, where ΔR₁ = R₁ (O₂) – R₁ (air). The initial R₁ (air) was derived from the average of the first two high SNR R₁ maps. The R₁ (O₂) was derived from the high SNR R₁ map acquired during oxygen breathing. Oxygen enhancement was measured as 2 × tumor baseline R₁ × cohort CoV. Voxels were then classified as enhancing (termed “Oxy-E”) or refractory (termed “Oxy-R”) to oxygen challenge. Voxel-wise BOLD ΔR₂* was calculated, where ΔR₂* = R₂* (O₂) – R₂* (air).

Where DCE-MRI was performed, the initial area under the contrast agent concentration curve from 0 to 60 seconds (IAUC₆₀) was calculated (16), and voxels were classified as enhancing when IAUC₆₀ >0. Where both OE-MRI and DCE-MRI were acquired, data were combined to distinguish three tumor subregions from one another: (i) perfused Oxy-E voxels; (ii) perfused Oxy-R voxels; and (iii) nonperfused voxels.

Hydralazine acts directly on vascular smooth muscle in vessels of normal tissues, causing vasodilation and reduced mean arterial blood pressure. Tumor blood vessels lacking smooth muscle do not dilate in response to hydralazine. Hence, blood is redistributed away from the tumor, reducing blood flow and increasing hypoxia within 30 minutes (33–36). Hydralazine challenge has, therefore, been used as a tool to manipulate acute hypoxia.

An initial air-to-oxygen gas challenge was performed. Then, gas delivery was switched back to air breathing, and intravenous injection of either 5 mg/kg hydralazine hydrochloride (Sigma-Aldrich Co.) or saline was performed. Finally, a second air-to-oxygen gas challenge was performed. Spatial differences between the ΔR₁ defined Oxy-E and Oxy-R voxels on the two air-to-oxygen challenges were assessed by mismatch mapping. Formal randomization was not employed, rather during each day's scanning, mice were prospectively assigned with intent to balance treatment groups according to tumor size.

Intraperitoneal injection of 60 mg/kg pimonidazole (Hypoxyprobe) was performed 55 minutes before 100% O₂ inhalation began to allow for maximal bioreduction of the agent in hypoxic tumor regions. Hoechst 33342 (15 mg/kg; Sigma-Aldrich Co.) was administered intravenously 1 minute prior to rapid tumor excision. Tumors were excised whole and bisected along the imaging plane so that the cut surface approximated to the MRI region of interest. Half the tumor was snap frozen, and half was formalin fixed and paraffin embedded.

Frozen tissue sections (5 μm) were obtained from snap-frozen tumor material and scanned using fluorescent microscopy on a Panoramic 250 Flash system (3DHistech) to determine the number of Hoechst-stained (perfused) vessels (excitation 350 nm/emission 480 nm). Pimonidazole binding was determined in the same sections using Hypoxyprobe-1 (Hypoxyprobe), a mouse-monoclonal, followed by rabbit anti–mouse-FITC conjugated secondary antibody (excitation 488 nm/emission 525 nm). Paraffin-embedded sections (5 μm) were obtained from tumor tissue dehydrated after fixation in 10% phosphate-buffered formalin. Sections were stained with hematoxylin and eosin (H&E). Data were analyzed using ImageJ software (NIH, Bethesda). The perfused vessel area, hypoxic fraction, and percentage necrosis were calculated, as described elsewhere (37).

IBM SPSS Statistics v.22 was used for all statistical analyses. In all cases, P values of <0.05 were considered significant. Comparison of median values of R₁ during medical air only breathing and during oxygen challenge experiments were evaluated using a one-way ANOVA with a post hoc Bonferroni correction for multiple comparisons.

Data comparing median values of R₁, R₂*, and IAUC₆₀ and fractions of Oxy-E or Oxy-R between different tumor cohorts (786-0-R, 786-0-par, or SW620) were evaluated with independent t tests corrected for multiple comparisons. Data for different cohorts were assumed to be distributed normally and have unequal variance. Correlations between MRI and pathology were assessed by the nonparametric Spearman rho. Formal sample size calculations were not performed.

We sought to demonstrate OE-MRI signal precision, stability, and sensitivity to altered oxygen tension, because these factors have not been well documented. Because OE-MRI signal changes are based on the longitudinal relaxation rate, R₁, we compared the expected R₁ against the measured R₁ in an in vitro phantom. These data showed that our R₁ measurement technique is accurate (Supplementary Fig. S2).

To test measurement precision and stability in vivo, we then performed an experiment in four mice implanted with subcutaneous 786-0-R tumors (786-0-R is a fast-growing subline of the 786-0 renal cancer cell line). Mice initially breathed medical air for 40 minutes. This was followed by 100% oxygen challenge for 20 minutes. Finally, the mice breathed medical air again for 25 minutes. Multiple R₁ maps were acquired (Supplementary Fig. S1A). The within-scan coefficient of variation (CoV) of baseline voxel-wise R₁ was 0.41% in vivo, indicating high measurement precision. No appreciable difference was observed in the spatial mapping of tumor R₁ during air breathing (see time points 1–7 in Fig. 2A). No significant change was seen in tumor R₁ in the four mice (Fig. 2B), indicating signal stability while breathing air.

Challenge with 100% oxygen was then performed in the same mice, to increase tumor pO₂. All four tumors showed rapid, heterogeneous and significant increase in median R₁ following oxygen challenge. These changes were clearly visible in the first map after 2 minutes and 40 seconds of oxygen breathing and persisted for the duration of the oxygen challenge (P < 0.008 for all time points 8–12 compared with air breathing) showing consistent spatial arrangement (Fig. 2A and B). Across all four tumors, 89.3% of voxels were oxygen enhancing (Oxy-E), and 10.7% of voxels were oxygen refractory (Oxy-R). Sample traces of ΔR₁ change are shown for Oxy-E and Oxy-R voxels (Supplementary Fig. S3). OxyLite measurement in the same tumors showed pO₂ increase (beyond the limit of detection of the OxyLite equipment, namely at 100 mmHg) in 7 of 8 measured regions over the same time frame (sample trace in Fig. 2C). The OxyLite data provide independent evidence that OE-MRI detects real-time increases in the oxygen concentration in tumor interstitial tissue fluid. BOLD MRI, performed in the same mice, revealed a significant reduction in R₂* during oxygen challenge (mean reduction, 39.8 ms⁻¹; SE, 11.7 ms⁻¹; P = 0.043), consistent with previous studies (38).

Tumors showed rapid reversal of R₁ changes when oxygen challenge ended and air breathing was resumed (see time points 13–17 in Fig. 2A and B). There was no significant difference between baseline and end-of-study mean R₁ values for the cohort. OxyLite measurement showed pO₂ return to prechallenge levels over a period of 2 to 6 minutes mirroring ΔR₁ change. Collectively, these data confirm that R₁ measurements used for this study are accurate, precise, stable in the absence of perturbation, and sensitive to change in tumor pO₂.

We investigated whether Oxy-R fraction (the fraction of each tumor refractory to oxygen challenge) could detect different levels of hypoxia in cancer models. We measured the Oxy-R fraction in two isogenic cell lines: the slow growing parental 786-0 cell line (786-0-par), which took 206.3 days ± 66.2 SD to reach a tumor volume of approximately 400 mm³ (n = 8 mice), and the much faster growing 786-0 subline (786-0-R), which only took only 32.1 days ± 6.9 SD to reach the same tumor volume (n = 9 mice). From these mice, we acquired OE-MRI, ΔR₂*, and DCE-MRI data along with histopathologic analysis of pimonidazole adduct formation (for hypoxia) and Hoechst 33342 uptake (for perfusion; Supplementary Fig. S1B).

Staining for pimonidazole adduct formation showed a significantly higher hypoxic fraction in 786-0-R tumors compared with the 786-0-par tumors (P = 0.008; Fig. 3A). This finding was mirrored by the Oxy-R fraction being significantly higher in the 786-0-R xenografts compared with the 786-0-par xenografts (P = 0.047; Fig. 3B). In distinction, there was no significant difference in the median values of ΔR₁, IAUC₆₀, or ΔR₂* between the 786-0-R and 786-0-par xenografts (Fig. 3C–E). No significant difference was seen in the perfused vessel area measured by Hoechst 33342 staining (data not shown). These data confirm that Oxy-R fraction is sensitive to differential levels of hypoxia, but that median values of ΔR₁, IAUC₆₀, and ΔR₂* were insensitive to this difference.

We sought to test if Oxy-R fraction could estimate tumor-hypoxic fraction in two tumor models originating from different cell lines. In particular, we wanted to test if the presence of avascular or necrotic tumor tissue—which does not receive any delivery of oxygen gas in the blood plasma—would confound the relationship between Oxy-R and hypoxia. Therefore, we analyzed data from the perfused portion of each tumor.

To do this, we first used the 786-0-R data already collected. We examined the within-cohort correlation between Oxy-R fraction and the pimonidazole adduct formation–based measurement of hypoxic fraction. This model is known to be highly vascularized. Further, there was a wide dynamic range of hypoxia in this model. We then performed an equivalent experiment using SW620 tumors. This model was chosen because these tumors are relatively poorly vascularized (39).

Anticipated differences in the perfusion status of the two models were confirmed; 786-0-R xenografts had significantly greater perfused tumor area compared with SW620 xenografts (n = 8 mice), measured by DCE-MRI (P = 0.011; Table 1) and by Hoechst 33342 (P = 0.004). Pathology analysis showed that perfused vessel area measured by Hoechst 33342 (Supplementary Fig. S4A and S4B) correlated with perfused Oxy-E fraction (those voxels showing positive enhancement with both oxygen and gadolinium) and that nonperfused fraction correlated with necrosis on H&E in both models (Supplementary Fig. S4C and S4D).

Oxy-R fraction correlated strongly with pathologic hypoxic fraction in 786-0-R tumors (rho, 0.810; P = 0.028; Fig. 4A), irrespective of whether the entire tumor was analyzed or if analysis was restricted to gadolinium-enhancing tumor only (perfused Oxy-R fraction). In SW620 tumors, the Oxy-R fraction did not correlate with hypoxic fraction (measured by pimonidazole adduct formation). However, as anticipated, the perfused Oxy-R fraction and hypoxic fraction (measured by pimonidazole adduct formation) correlated strongly (rho, 0.929; P = 0.002; Fig. 4B). In both models, visual inspection revealed that perfused Oxy-R voxels were located at an interface between the peripheral perfused Oxy-E voxels and the centrally located nonperfused voxels. Collectively, these data provide evidence that Oxy-R fraction can quantify hypoxic fraction accurately, but requires perfusion data in poorly vascularized tumors.

In order to provide evidence that Oxy-R fraction can detect reduced oxygen delivery to tumors, we administered the vasomodulator hydralazine to a cohort of mice bearing 786-0-R xenografts (n = 4 mice) and compared with control mice receiving saline (n = 6 mice). We performed an air-to-oxygen challenge before and after the administration of hydralazine or control (Supplementary Fig. S1C). In the initial air-to-oxygen challenge, overall ΔR₁ changes were positive, as expected in all 10 mice (Fig. 5A).

Importantly, the second air-to-oxygen challenge was initiated 15 minutes after administration of hydralazine or saline. The magnitude of ΔR₁ was reduced significantly in mice receiving hydralazine, relative to control (Fig. 5A; P < 0.01). This difference in hydralazine-treated tumors was driven by the Oxy-R fraction increasing significantly from 17.3% to 30.5% (P = 0.045; Fig. 5B).

To further understand these changes, we used “mismatch mapping” to identify tumor subregions that changed oxygen enhancement status. In control tumors, the majority of voxels remained consistently either Oxy-E or Oxy-R. The number of voxels changing from Oxy-E to Oxy-R or vice versa was equal for both directions of change and was small (Fig. 5C). In hydralazine-treated tumors, substantially more voxels turned from Oxy-E to Oxy-R than vice versa. Some of these newly Oxy-R tumor subregions were immediately adjacent to established Oxy-R regions, but other new areas of Oxy-R were detected at spatially distant subregions (Fig. 5D). These data confirm that OE-MRI can detect dynamic change in tumor hypoxia. Further, tumor regions with appearing or resolving hypoxia can be mapped using this technique.

Hypoxia is an important prognostic factor for solid tumors and mediates resistance to radiotherapy and to some chemotherapy agents. Noninvasive clinical imaging has potential to improve patient care by producing biomarkers that alter clinical decision making (40, 41). In the case of hypoxia, a clinically translational imaging biomarker has potential to be prognostic and/or predictive (identifying presence of hypoxia and quantifying hypoxic fraction/volume prior to therapy), to assist treatment planning (mapping targets for radiation boost or adaptive therapy; ref. 42), or to monitor therapy-induced changes in tumor hypoxia, for example following radiosensitization or hypoxia-activated cytotoxic therapy (10, 43).

To date, clinical strategies to image tumor hypoxia have been dominated by PET methods (7). Several investigational PET tracers have been evaluated, each with slightly different underlying relationship to tissue oxygen tension (44, 45). However, at present, no imaging method identifies and maps the extent of tumor hypoxia that is validated, cost-effective, widely available and used for patient benefit (7).

OE-MRI is an emerging technique that detects the presence of excess dissolved O₂ molecules in plasma and tissue fluid. Most OE-MRI studies in cancer have been small cohort–based studies, and without corresponding pathology or intervention, tending to report changes in median signal intensity or median ΔR₁. In well-oxygenated tumor tissue, oxygen-induced ΔR₁ signals are positive, similar to signal changes observed in healthy tissues (46, 47). Advances in OE-MRI have focused largely around providing dual ΔR₁ and ΔR₂* acquisitions (25) or in enhancing the magnitude of oxygen-induced ΔR₁, for example by developing methods sensitive to tumor lipid signals (17).

In this study, we adopted a different approach. We focused on the fact that tumors are biologically heterogeneous, composed of subregions with distinct pathophysiology (48, 49). Previous OE-MRI studies have suggested that some tumor subregions are refractory to the positive ΔR₁ induced by oxygen challenge (14, 20). We built on this observation by partitioning the OE-MRI signal into regions with enhancing (Oxy-E) or refractory (Oxy-R) signatures. We conducted a series of experiments to test the hypothesis that Oxy-R fraction tumor can identify, quantify, and map tumor hypoxia.

We first used the renal carcinoma model 786-0-R (rapid growing, vascular tumors with anticipated large oxygen-induced ΔR₁) to develop a protocol that was not only stable while breathing medical air, but also sensitive to oxygen-induced signals. This is the first study to provide detailed evidence of measurement stability and sensitivity for serial R₁ mapping in OE-MRI. The temporal evolution of oxygen-enhanced–positive ΔR₁ occurred within 3 minutes and was spatially stable. The OE-MRI data mirrored OxyLite measurement of pO₂ changes, concurring with data from a previous study where rapid increases in ΔR₁ were observed following gas challenge (15) and with data comparing OE-MRI signal changes with ¹⁹F oximetry (19).

Next, we demonstrated that Oxy-R fraction could distinguish differences in hypoxic fraction within tumors derived from two isogenic cell lines: parental 786-0 cells and 786-0-R cells. Here, the relatively rapid growing 786-0-R tumors had significantly more hypoxia than the slow growing 786-0-par tumors. Crucially, Oxy-R fraction distinguished between cell lines, mirroring differences observed in values of hypoxic fraction quantified through pimonidazole staining.

We then tested if Oxy-R fraction could estimate hypoxic fraction. We chose the 786-0-R and SW620 models because of their differing perfusion characteristics. Significant correlation was seen between Oxy-R fraction and hypoxic fraction (defined on pimonidazole adduct formation) in the relatively well perfused 786-0-R xenografts. This association was not replicated in the poorly perfused SW620 xenografts because many Oxy-R voxels had negligible perfusion, so received insufficient oxygen gas to alter voxel R₁. Having accounted for these voxels using DCE-MRI, we measured Oxy-R fraction in perfused tissue and observed a significant association with hypoxia in the SW620 tumors.

We then increased the fraction of tumor hypoxia in a further cohort of 786-0-R tumors using the vasomodulator hydralazine. Oxy-R fraction increased 2-fold, providing direct evidence that OE-MRI can track acute changes in tumor oxygenation. Some signal fluctuation was observed in both directions in hydralazine and control tumors. This may reflect changes in acute transient hypoxia (50), but further work is required to understand these signal changes.

In these experiments, median ΔR₁ (the most common biomarker derived from OE-MRI data) was unable to distinguish the differing levels of hypoxia between the 786-0 sublines and did not correlate with the degree of hypoxic fraction within the SW620 tumors. We also compared the Oxy-R analysis approach with BOLD MRI. Oxygen-induced ΔR₂* did mirror changes in tissue pO₂, similar to previous reports (38), but ΔR₂* did not differ between the 786-0 sublines with differing levels of hypoxia, nor did ΔR₂* relate to degree of hypoxic fraction in 786-0-R or SW620 tumors, suggesting that R₂* measurements were confounded by other factors, such as vessel flow. These findings are consistent with previous reports of inconsistent relationships between ΔR₂* and pimonidazole-based assay of hypoxic fraction (38).

Our findings are important for at least four reasons. First, these data demonstrate that positive ΔR₁ changes on OE-MRI imaging are sufficiently precise, stable, and strong enough to be measured reliably. This enables those portions of tumor refractory to positive signal change (i.e., Oxy-R subregions) to be identified with confidence. Second, these data suggest that in OE-MRI, Oxy-R fraction relates more closely to tumor hypoxia than median ΔR₁. This supports further evaluation of Oxy-R fraction or Oxy-R tumor volume as a prognostic and/or predictive biomarker. Third, Oxy-R voxels were located in spatially distinct subregions, suggesting that, if replicated in humans, this technique may facilitate the planning of localized boost and adaptive radiotherapy delivery strategies. Fourth, clinical tumors vary in their extent of perfusion, and this may further change with therapy. Our data suggest that subsequent studies of OE-MRI in patients with cancer should evaluate Oxy-R fraction in perfused tumor regions (using a combined analysis with DCE-MRI or an equivalent technique), rather than using OE-MRI in isolation.

In conclusion, these data provide substantial evidence to validate the measurement of Oxy-R fraction to identify, quantify, and map tumor hypoxia. Unlike other nuclear magnetic resonance techniques such as electron paramagnetic resonance oximetry, ¹⁹F oximetry, and Overhauser enhanced MRI—which are either invasive or require use of an investigational diagnostic agent, hindering clinical translation (7)—Oxy-R fraction can be readily quantified on clinical MRI scanners (24), making the technique suitable for rapid clinical translation.

Geoff J.M. Parker is founder and CEO at and has ownership interest (including patents) in Bioxydyn Limited. He is also a consultant/advisory board member for GlaxoSmithKline. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.P.B. O'Connor, J.K.R. Boult, Y. Jamin, K.J. Williams, A. Jackson, G.J.M. Parker, J.C Waterton, S.P Robinson

Development of methodology: J.P.B. O'Connor, J.K.R. Boult, Y. Jamin, R.A. Little, G.J.M. Parker, A.R. Reynolds, J.C Waterton, S.P Robinson

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.P.B. O'Connor, J.K.R. Boult, Y. Jamin, A.R. Reynolds, S.P Robinson

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.P.B. O'Connor, Y. Jamin, M. Babur, K.G. Finegan, K.J. Williams, R.A. Little, G.J.M. Parker, S.P Robinson

Writing, review, and/or revision of the manuscript: J.P.B. O'Connor, J.K.R. Boult, Y. Jamin, M. Babur, K.J. Williams, R.A. Little, A. Jackson, G.J.M. Parker, A.R. Reynolds, J.C Waterton, S.P Robinson

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.P.B. O'Connor, K.G. Finegan

Study supervision: J.P.B. O'Connor, G.J.M. Parker, S.P Robinson

The authors thank Allan Thornhill and his staff (ICR) for animal maintenance and Dr. Damian McHugh (University of Manchester) for generation of maps used in the figures.

This study was supported by Cancer Research UK (CRUK) Clinician Scientist award (grant C19221/A15267 to J.P.B. O'Connor), CRUK and EPSRC Cancer Imaging Centre in Cambridge and Manchester funding to The University of Manchester (grant C8742/A18097 to J.P.B. O'Connor, K.J. Williams, A. Jackson, and G.J.M. Parker), CRUK Cancer Imaging Centre funding to The Institute of Cancer Research (grants C1060/A10334 and C1060/A16464 to S.P. Robinson), The Wellcome Trust funding (grant 091763Z/10/Z to S.P. Robinson), Paul O'Gorman Postdoctoral Fellowship funded by Children with Cancer UK (grant 2014/176; Y. Jamin), and Breakthrough Breast Cancer Senior Fellowship (A.R. Reynolds). Breakthrough Breast Cancer has recently merged with Breast Cancer Campaign to form Breast Cancer Now.