Qualification of a Noninvasive Magnetic Resonance Imaging Biomarker to Assess Tumor Oxygenation

Tumor hemodynamics has become a key target in preclinical and translational cancer research (1), involving both negative and positive modulations of tumor oxygenation/perfusion. On the one hand, attempts are made to target the established tumor vasculature or neovasculature by the use of antivascular or antiangiogenic agents (2–6), whose effects are traditionally assessed noninvasively not only using dynamic contrast MRI (DCE-MRI) after administration of an exogenous paramagnetic contrast agent (7–9), but also possibly using endogenous hemodynamic markers, including BOLD-MRI (blood oxygen level dependent MRI; ref. 10). On the other hand, positive modulations of tumor hemodynamics, tumor hypoxia, or environmental pH, for example, (11), are being considered in the field of tumor radiosensitization (12–14).

Most solid tumors contain regions of acute and chronic hypoxia that indicate a negative clinical outcome after radiotherapy (15). To bridge the gap between the occurrence of tumor hypoxia and clinical radiation practice, there is an essential need to predict the presence of hypoxic regions in tumors individually. On the basis of the individual tumor characteristics and/or the ability to alleviate the tumor hypoxia, it will become possible to adapt the individual treatment either by boosting optimal radiation doses in the resistant areas, by adapting the radiotherapy to each tumor throughout the course of treatment, or by administering an associated treatment to potentiate the effectiveness of the radiation treatment.

There is therefore a critical need to develop accurate, noninvasive and quantitative in vivo imaging methods of mapping tumor oxygenation in cancer management, both for the purpose of targeting tumor blood vessels and for the oxygen-induced resistance to radiation of tumors. Noninvasive, safe, and repeatable techniques to map tumor hypoxia are therefore required.

Direct quantitative methods, including Eppendorf microelectrodes (16), electron paramagnetic resonance (EPR) oximetry (17), ¹⁹F relaxometry (18), or Overhauser enhanced MRI (19), are either invasive or require the injection of a reporter probe, and are currently not clinically applicable. One technique for mapping hypoxia in the current clinical setting is PET using nitroimidazole-derived tracers accumulating within hypoxic areas (20, 21). However, cost, reduced availability, and radiation exposure preclude its use in routine longitudinal practice. Moreover, it does not allow the monitoring of acute changes in hypoxia because it requires an accumulation of the tracer in the tissue of interest for several hours, and does not allow the follow-up of dynamic changes induced by oxygenation modulators. Ideally, imaging hypoxia in humans should be noninvasive, sensitive, robust, quantitative, widely available, and able to probe tumor heterogeneity. To this end, two endogenous sources of contrast in MRI, namely functional MRI and oxygen-enhanced MRI, are currently being explored (22). T₂* (effective transversal relaxation rate) mapping, also referred to as functional MRI or “BOLD” imaging (23), is sensitive to changes in oxygenation in the vascular compartment and has demonstrated significant limitations in terms of quantitative relationships between response signal intensity and true changes in tumor tissue pO₂ (24–26). Moreover, it is also sensitive to changes in the basal hemoglobin content. More recently, changes in tissue oxygen concentrations have been shown to produce changes in the relaxation rate R₁ (= 1/T₁) of water (“oxygen enhanced MRI”; refs. 27–29). T₁ demonstrates sensitivity to dissolved oxygen, thereby acting as a potential T₁-shortening paramagnetic contrast agent (19). An oxygen-induced increase in the R₁ has the potential to provide noninvasive measurements of fluctuations in the oxygen level of tissue, as a complement to BOLD imaging (30, 31). Unfortunately, this technique is still hampered by insufficient sensitivity (only a few percentage changes) and the Δ R₁ measured may be biased owing to confounders unrelated to changes in the oxygen level in the tissue, such as an alteration in blood flow, and the H₂O content of the tissue (32).

We recently described a new noninvasive MRI method allowing rapid mapping of changes in tissue oxygenation and based on the higher solubility of oxygen in lipids than in water. By monitoring changes in the R₁ relaxation rate of the lipid peak rather than those in the water peak, sensitive estimates of variations in tissue oxygenation could be obtained (33, 34). This sequence, with the acronym MOBILE for “mapping of oxygen by imaging lipids relaxation enhancement,” has simultaneously been translated into the clinical setting because it enables noninvasive and sensitive measurements of changes in tissue oxygenation. Until now, MOBILE has been successfully applied in hypoxic mouse models mimicking physiopathological conditions such as tumor hypoxia, peripheral ischemia, cerebral ischemic stroke, and liver steatosis, as well as in the clinical setting in healthy volunteers (33) and in stroke patients (35).

The aim of the current work was to investigate the potentially quantitative aspect of MOBILE, as a biomarker for tumor oxygenation in two mammary tumor models (NT2 and MDA-MB-231), to benchmark MOBILE in comparison with the more traditional endogenous contrast using the global R₁, and to assess the ability of the technique to monitor positive and negative changes in tumor oxygenation. Hypoxic challenges were performed by the injection of the antivascular agent CA4, described to induce a vascular shutdown as early as 3 hours after the administration of the drug (36), whereas hypoxic challenges were induced by a carbogen breathing (95%O₂/5%CO₂) known to acutely increase tumor oxygenation in a wide range of tumor models/xenografts and in human tumors (36–40, 31). Oxygenation was assessed in matching tumors (the same tumors underwent both EPR and MRI experiments) using EPR oximetry as a method of reference for the quantitative assessment of the tumor pO₂.

To this end, the following questions will be addressed: “Is a change in the R₁ (ΔR₁) of lipids related to a change in the tumor pO₂ (ΔpO₂)?”; “Are intrinsic R₁ values of lipids related to absolute pO₂ values?”; and finally, “Are changes in the R₁ of lipids more sensitive than changes in the global R₁?”

A total of 7 × 10⁶ NT2 cells (provided by Elizabeth M. Jaffee, M.D., The Sidney Kimmel Cancer Center at Johns Hopkins, Baltimore, MD) or 10 × 10⁶ MDA-MB-231 (LGC Promochem), amplified in vitro, were collected by trypsinization, washed three times with Hanks Balanced Salt Solution (HBSS), and resuspended in 100 μL HBSS. These mammary tumor cells were injected subcutaneously into the right upper mammary fat pad of 6-week-old FVB/N or nude NMRI female mice (Janvier). The tumors were analyzed when they reached 5 mm in diameter. The animals were anesthetized by inhaling isoflurane (Forene) mixed with 21% oxygen (air) in a continuous flow (1.5 L/hour). This anesthetic has been shown not to interfere with tissue hemodynamics (41). The respiration rate and body temperature (37.0°C ±1.0°C) were monitored and maintained with a circulating water blanket. Studies were undertaken in accordance with the national and local regulations of the ethical committee (agreement number UCL/2010/MD/001).

To negatively modulate tumor oxygenation, CA4 was administered at a dose of 100 mg/kg and MR measurements were taken before injection and repeated 3 hours after the administration of CA4. EPR oximetry was performed on the same tumors, in a similar way, before injection and directly after the second MRI session. To positively modulate tumor oxygenation, carbogen breathing (95% O₂, 5% CO₂) was used. For that purpose, three MR measurements of each type (global R₁ and R₁ of lipids) were taken sequentially and repeated three times while the subject was breathing air. The gas was then switched to carbogen, and MR measurements were repeated at 10, 15, and 20 minutes after the switch, as it is known that oxygenation is significantly increased after 10 minutes' breathing (42). EPR experiments were subsequently performed on the same tumors. We considered five hyperoxic challenges and five hypoxic challenges in each tumor model. Both challenges could have been applied on the same tumor for three NT2 tumors and two MDA-MB-231 tumors. The remaining experiment was performed on different tumors.

T₁ measurements.

A segmented IR FISP (Inversion-Recovery Fast Imaging with Steady state Precession) sequence (SSFP FID mode) was used to acquire parametric images of T₁ relaxation time, as described previously (33). Briefly, the acquisition parameters were TR/TE/FA/BW/matrix = 4 ms/1.2 ms/5°/100 kHz/64 × 64, four segments, and a total acquisition time of 1 minute 20 seconds. For the global proton experiment (global R₁, essentially reflecting the water peak), a series of 100 images were taken, with a slice thickness of 1 mm. For the lipid experiments (MOBILE), the offset between water and lipid peaks was assessed experimentally with a single pulse sequence (the lipid peak of interest was ∼4.0 ppm) and then used as an imaging frequency offset in the same IR FISP protocol. We added a saturation pulse to spoil the water signal. A series of 40 images (spaced by scan repetition time, TR = 100 ms) with a slice thickness of 2 mm were acquired with a spatial resolution of 0.344 × 0.344 mm. The images were then fitted using a home-made program written in Matlab (The MathWorks, Inc.) to determine the T₁ relaxation time in the regions of interest (ROI), as described previously (33).

EPR oximetry.

The in vivo tumor pO₂ was monitored using EPR oximetry, with charcoal as the oxygen-sensitive probe (17). EPR spectra were recorded using a 1.1 GHz EPR spectrometer (Magnettech). Calibrations curves were made by measuring the EPR line width as a function of the pO₂. Mice were injected in the center of the tumor using the suspension of charcoal (100 mg/mL, 40 μL injected) 24 hours before the experiment. The tumor being studied was placed in the center of the extended loop resonator whose sensitive volume extends 1 cm into the tumor mass. The pO₂ measurements correspond to an average of pO₂ values in the volume. MOBILE and EPR measurements were taken the same day as we know from unpublished data that a carbogen challenge on one tumor is reproducible on a same day within a 3-hour interval.

ROIs were delineated by hand to select the tumor tissue without charcoal, although it has been previously shown that charcoal does not interfere with T₁ measurements (33). As we recorded three parametric maps of the global R₁, and three parametric maps of the R₁ of lipids, as well as three EPR measurements in each condition (at baseline, during carbogen breathing and 3 hours after CA4 administration), from each set of three measurements, we calculated the mean as well as the SD used for the correlations. Pearson correlations between the relaxation rates R₁ (or ΔR₁) and the actual pO₂ values (or ΔpO₂) were performed using the GraphPad software. Deming regression was performed to compare the sensitivities of the techniques.

Tumor oxygenation was modulated in two distinct tumor models, one syngenic (NT2 mammary tumors) and one xenograft (MDA-MB-231 xenografts), using a hypoxic challenge (i.e., administration of CA4 3 hours before measurement) or a hyperoxic challenge (i.e., carbogen breathing). Two MDA-MB-231 tumors underwent both challenges, whereas six other tumors underwent hyperoxic (n = 3) or hypoxic challenges (n = 3) only. Three NT2 tumors were then submitted to both challenges and four other tumors were submitted to carbogen breathing (n = 2) or were targeted by CA4 (n = 2). Typical maps of the R₁ of lipids at baseline, and both after a hypoxic and a hyperoxic challenge are shown in Fig. 1, together with similar challenges assessed using global R₁ mapping.

While pooling data from 20 different tumors (10 of each tumor model), we can observe a significant correlation between the ΔR₁ of lipids (MOBILE) and the ΔpO₂ in the same tumors (P = 0.0059, r = 0.5097; Fig. 2A), whereas similar analysis using the global R₁ does not give a significant correlation (Fig. 2B). Individual changes in the R₁ corresponding to an increase in absolute pO₂ values, as assessed by EPR oximetry, evidence that (i) the R₁ of lipids increased concomitantly (by 2%–17% of relative increase) when the pO₂ increased, except for one tumor (2.7% decrease; Fig. 2A) and (ii) the global R₁ increased from 0% to 6% with respect to the same pO₂ increase, except for two tumors which showed decreases of 0.5% and 7% (Fig. 2B). As regards changes in the R₁ with respect to a decline in the pO₂, the graphs in Fig. 2 show that (i) the R₁ of lipids falls systematically (by 0.1%–13.6%; Fig. 2A) and (ii) the global R₁ shows mixed results with both negative (down 4.4%) and positive (up 5.6%) changes in matching tumors (Fig. 2B).

To investigate the potentially quantitative properties of MOBILE, mean individual R₁ values of lipids were compared with individual pO₂ values within each tumor model. For this purpose, all pre- and postchallenge mean values of the R₁ of lipids (MOBILE) were plotted together with respect to all corresponding pre- and postchallenge pO₂ values (Fig. 3A and B).

Since some tumors underwent both challenges, only one basal value of each parameter (pO₂, global R₁ and R₁, of lipids) was assessed for these tumors, resulting in a number of points that differ from 20: we considered 17 points and 18 points for the NT2 and MDA-MB-231 models, respectively. We were able to observe a positive correlation between the R₁ values of lipids and absolute pO₂ values in both models [(P<0.0001, r = 0.8164 and P = 0.0378, r = 0.5069) for MDA-MB-231 and NT2 tumor models, respectively], and between global R₁ and pO₂ values in the MDA-MB-231 tumor model (P = 0.0025, r = 0.6673).

When all the data were pooled, a significant correlation was established between the R₁ values of lipids and the pO₂ values (P = 0.0275, r = 0.3726; Fig. 4). We could not pool the global R₁ data from both tumor models because there is no correlation in the NT2 tumor model between pO₂ values and global R₁ values. All these data argue in favor of qualifying MOBILE as a sensitive method of assessing tumor oxygenation.

Furthermore, according to Fig. 4, the relaxation rates of lipids calculated within the MDA-MB-231 tumor model seem to be superior to those obtained in the NT2 tumor model. This observation goes hand in hand with the difference in oxygenation levels that was evidenced by EPR oximetry: MDA-MB-231 tumors are well oxygenated at baseline (10.5 ± 6.5 mm Hg) compared with NT2 tumors (4.8± 3.1 mm Hg).

Sensitivities were assessed by comparing the slopes of the linear regression graphs for traditional oxygen-enhanced MRI and MOBILE data. This comparison can only be assessed within the MDA-MB-231 model because both techniques show a significant correlation in this tumor model, which is not the case for the NT2 model. When comparing the slopes providing by the Deming regressions (Y = 0.01486 X + 0.9143 for Lipids R₁ and Y = 0.01486 X + 0.9143 for global R₁), we observe that MOBILE is approximately 1.5 times more sensitive than global R₁ measurement.

Figure 5 presents the magnitude of response in terms of the global R₁, the R₁ of lipids, and the pO₂, in relation to basal means of these three parameters. Although the response to a carbogen breathing challenge does not depend on basal pO₂ (Fig. 5A), as already published (42), the fall in pO₂ induced by CA4 is greater when the basal oxygenation level is higher. This is also assessed using the R₁ of lipids (Fig. 5C). However, neither basal values of the R₁ of lipids nor the global R₁ can help predict the extent of the response to carbogen or CA4 (Fig. 5D–I).

This study demonstrates the ability of MOBILE to follow positive and negative changes in tumor oxygenation further to hypoxic or hyperoxic challenges, suggesting that the endogenous source of contrast relying on the R₁ of lipids in MRI can constitute a sensitive noninvasive marker of tumor hypoxia. The MOBILE technique enables the assessment of the relaxation parameter “R₁ of lipids” and is consecutive to a parent emerging technique assessing the “global R₁,” which is prominently influenced by the R₁ of water, but lacks good sensitivity to changes in oxygenation. As R₁ techniques are sensitive to tissue oxygenation, they appear to be complementary to the routinely used functional imaging or BOLD-MRI technique, assessing changes in the R₂* relaxation parameter (29), which is sensitive to changes in oxygenation in the vascular compartment, yet with significant limitations in terms of quantitative aspects and in sensitivity.

Further validation of the hypoxia marker requires a correlation of the new method with a quantitative method in the preclinical setting. This was assessed using EPR oximetry, an invasive but quantitative and highly sensitive method able to assess tissue oxygenation in vivo (17). Our study evidenced that (i) positive and negative changes in tumor oxygenation can be detected using MOBILE; (ii) a ΔR₁ of lipids is positively correlated with a ΔpO₂in vivo; (iii) individual R₁ values of lipids are positively correlated to absolute pO₂ values; and (iv) changes in the R₁ of lipids are more sensitive than changes in the global R₁. This makes MOBILE a sensitive method to assess changes in tumor oxygenation. This is not systematically the case for the global R₁, showing global R₁ changes that are not always correlated to the changes in the pO₂ and that are also smaller in magnitude (less sensitive), as observed on individual graphs.

The ability of the MOBILE technique to follow the effect of an antivascular agent (combretastatin A4) longitudinally as well as that of a hyperoxic challenge (carbogen breathing) noninvasively could find direct applications in the clinical setting for the individual monitoring of patients treated using anticancer agents. To this end, individual monitoring of the actual effect of a drug on individual tumor hemodynamics could help the clinician in therapeutic decisions. This has also prompted efforts to combine antiangiogenic or antivascular agents together or with other treatment modalities (43–46).

When investigating the potentially quantitative properties of MOBILE (i.e., comparison of individual R₁ values of lipids vs. the actual pO₂ values), the R₁ values of lipids were shown to be correlated in individual models and on pooled data from both tumor models, whereas the global R₁ was not able to show such correlation. The significant correlations between R₁ of lipids and pO₂ argue in favor of a potential quantitative aspect of the MOBILE technique. However, although EPR oximetry measurements have shown that MDA-MB-231 tumors exhibit higher oxygenation level than NT2 tumors, the difference between the R₁ values of lipids at baseline between the two tumor models could imply that this parameter is also tissue dependent. Therefore, we cannot exclude that the R₁ is also influenced by tissue type and composition, especially for tumor oxygenation lower than 5 mm Hg. It should be noted that tissue type dependency would also be applicable for the global R₁. Moreover, on the single MDA-MB-231 tumor model, the global R₁ also correlated positively with the absolute pO₂ values. Accordingly, MOBILE seems to be more adapted to assessing tumor oxygenation than the parent technique assessing the global R₁, in the tumor models included in this study, and is systematically more sensitive than the global R₁. Nevertheless, both techniques remain complementary in nature because the origin of the information is different for lipid and global R₁, and because it is not proven, yet that “MOBILE” could be applied on tumor models with a low content of lipids.

In the field of radiotherapy, individual monitoring and quantification of tumor oxygenation could find application in the identification of a therapeutic window during which oxygenation is effectively modified following treatment aimed at modifying oxygen supply and/or consumption. Overall modification of tumor hypoxia has been shown to significantly improve the efficiency of RT for locoregional control and for overall survival (47). Recent reports have also pointed out that the hypoxia-targeted approach ARCON (accelerated radiotherapy plus carbogen inhalation and nicotinamide) had an impact on the patients' outcome in hypoxic laryngeal tumors but not in well-oxygenated tumors (48), outlying that segmentation of the patients with respect to their basal oxygenation level is mandatory for the optimization of radiotherapy. The MOBILE technique could therefore help in the proper qualification of drugs targeting hypoxia. In addition, the technique could guide intensity modulated radiotherapy planning (IMRT; refs. 49–51) currently used in the clinical setting, which enables millimetric irradiation of each individual tumor to boost the radiation dose in the most hypoxic regions. Importantly, the time and spatial resolutions of the MOBILE technique are compatible with longitudinal monitoring of tumor oxygenation with sufficient spatial resolution to guide IMRT. IMRT is currently not guided in day-to-day clinical practice using hypoxia tracers (52), because none of them are as yet adapted for the purpose.

As far as we know, there is no direct, quantitative, dynamic, and accurate noninvasive/irradiating method that is part of routine clinical practice to assess tumor oxygenation. That is why, despite the dispersion of our data in correlation graphs, that prevents its use in routine clinical practice for exact pO₂ measurement at this stage, the significant correlation found between the R₁ of lipids and the pO₂ using MOBILE makes this a very interesting tool for assessing tumor oxygenation variations. This will need to be further evaluated in additional preclinical and clinical studies. Experiments showing positive correlations in a wider range of tumor models and on other types of tissues, as well as a cross-validation with an alternative clinically available method able to map tumor hypoxia (i.e., PET imaging with nitroimidazoles), would be required.

In conclusion, the current study qualifies MOBILE as a sensitive, noninvasive, and potentially quantitative endogenous marker of tumor oxygenation, as the technique is sensitive to both positive and negative modulations of tumor oxygenation and, more importantly, significantly correlated to actual pO₂ values in matching tumors for both tumor models being studied. As the MOBILE technique presents unique translational properties and can be successfully implemented in the clinical setting, it is promising for further clinical monitoring of individual tumor oxygenation to assess the response to antiangiogenic or antivascular treatments, for treatment combination and for planning radiotherapy. If further characterization is required for pure quantitative routine clinical applications, the technique can be considered to study modulations of tumor oxygenation in patients and clinical intratumoral heterogeneity.

B.F. Jordan, J. Magat, and B. Gallez have a pending patent on in vivo quantification of a variation in a tissue by using a MRI technique. No potential conflicts of interest were disclosed by the other authors.

Conception and design: B. Gallez, B.F. Jordan

Development of methodology: J. Magat, B. Gallez

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Colliez, M.-A. Neveu, T.T.C. Pham, B. Gallez

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Colliez, B. Gallez

Writing, review, and/or revision of the manuscript: F. Colliez, B. Gallez, B.F. Jordan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): F. Colliez, M.-A. Neveu

Study supervision: B. Gallez, B.F. Jordan

The authors thank Philippe Levêque for assistance with statistical analysis.

This study was supported by grants from the Belgian National Fund for Scientific Research (FNRS), the “Fournier-Majoie Foundation”, the Joseph Maisin Foundation, the “Actions de Recherches Concertées-Communauté Française de Belgique, Grant No. ARC 09/14-020,” and the “Belgian Cancer Plan.”

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.