FMISO-PET-derived brain oxygen tension maps: application to glioblastoma and less aggressive gliomas

Quantitative imaging modalities for the analysis of hypoxia in brain tumors are lacking. The objective of this study was to generate absolute maps of tissue ptO2 from [18F]-FMISO images in glioblastoma and less aggressive glioma patients in order to quantitatively assess tumor hypoxia. An ancillary objective was to compare estimated ptO2 values to other biomarkers: perfusion weighted imaging (PWI) and tumor metabolism obtained from 1H-MR mono-voxel spectroscopy (MRS). Ten patients with glioblastoma (GBM) and three patients with less aggressive glioma (nGBM) were enrolled. All patients had [18F]-FMISO and multiparametric MRI (anatomic, PWI, MRS) scans. A non-linear regression was performed to generate ptO2 maps based on normal appearing gray (NAGM) and white matter (NAWM) for each patient. As expected, a marked [18F]-FMISO uptake was observed in GBM patients. The ptO2 based on patient specific calculations was notably low in this group (4.8 ± 1.9 mmHg, p < 0.001) compared to all other groups (nGBM, NAGM and NAWM). The rCBV was increased in GBM (1.4 ± 0.2 when compared to nGBM tumors 0.8 ± 0.4). Lactate (and lipid) concentration increased in GBM (27.8 ± 13.8%) relative to nGBM (p < 0.01). Linear, nonlinear and ROC curve analyses between ptO2 maps, PWI-derived rCBV maps and MRS-derived lipid and lactate concentration strengthens the robustness of our approaches.

Hypoxia is a critical component of the glioblastoma (GBM) microenvironment and has been associated with both poor prognosis and resistance to various therapies 1 . There is a marked correlation between hypoxia, vascular dysfunction and tumor aggressiveness 2 . It is also known, since the 1950's, that the low oxygenation observed in the tumor plays a major role in the resistance to X-ray radiation therapy 3 . The conventional treatment of GBM is surgical resection when possible, followed by radiotherapy along with concomitant chemotherapy. Despite the rigorous treatment protocol, the median survival fails to exceed 15 months 4 . Multi-parametric MRI or CT imaging are routinely used for tumor diagnosis, for the planning of radiation treatment but also for tumor follow-up after treatment. However, molecular imaging with positron emission tomography (PET) tracers sensitive to various biological parameters have now enlarged the range of functional information available for more targeted treatments and treatment efficacy determination 5 . For instance, biological target volume (BTV) used in radiotherapy could provide additional relevant data for dose painting in biologically active sub-regions or to boost radiotherapy 6 . Several approaches have been published over the last decade to estimate tumor hypoxia, or tumor oxygenation, including Eppendorf probes, electron paramagnetic resonance (EPR), 19 7 . Until now, PET imaging has been demonstrated to be the most rigorous (and "gold standard") approach to map tumor hypoxia. Yet, PET is a quantitative imaging technique and remains the most sensitive and specific approach. One of the most common hypoxia-specific tracers in PET imaging is 3-[18 F]-fluoro-1-(2-nitro-1-imidazolyl)-2-propanol ([ 18 F]-FMISO) 8 . However, it is well known that the relationship between oxygen tension and [ 18 F]-FMISO uptake is non-linear 9 and estimation of p t O 2 (oxygen pressure in the tissue) remains mandatory for advanced use of FMISO maps since ptO 2 maps are of greater biological pertinence. For instance, the accurate calculation of dose modulation for radiotherapy based on the Oxygen Enhancement Ratio effect requires ptO 2 mapping 10 . Two main models have been proposed in the literature to convert [ 18 F]-FMISO images into p t O 2 maps 11,12 . The former is a complex approach that uses many parameters to relate FMISO uptake to oxygen partial pressure but that was validated at the preclinical level. The latter is a simpler approach but needs to be validated in various situations.
To the best of our knowledge, adapted and validated mathematical models have never been employed in patients suffering from glioma. Here, we have initially employed in GBM patients the mathematical model previously used for head and neck cancers 12 . Nonetheless, this model turned out misfit. Therefore, the aim of the present study was to propose a new method of mathematical modeling to build-up p t O 2 maps from [ 18 F]-FMISO images for individual patient. Ten patients with GBM (GBM group) and three others patients with less aggressive glioma (nGBM group) were included and in which the proposed model was validated through the use of indirect estimates of tumor oxygenation from multimodal imaging; namely MRS and rCBV maps which were used to reflect anaerobic tumor metabolism and vascular status, respectively.

Results
Description of patients and images. GBM patients were characterized by the extravascular leakage of Gd-DOTA, hyperintensities on FLAIR images, a well-defined [ 18 F]-FMISO uptake, an elevated rCBV and a decrease in the NAA signal concomitant to an increase in the lactate and lipid signal (Fig. 1A, two uppers panels and Fig. 1B). In contrast, nGBM were characterized by no extravasation of Gd-DOTA, hyperintensities on FLAIR, basal [ 18 F]-FMISO uptake corresponding to the washout of the radiopharmaceutical in oxic conditions, nearly normal rCBV and a MRS profile compatible with that of less aggressive tumors (

Estimation of p t O 2 .
We firstly used the unknown parameters initially described 12 to calculate the p t O 2 value from the [ 18 F]-FMISO uptake. With this empirical approach, the mean p t O 2 in NAGM and NAWM were 19.7 ± 0.5 and 20.9 ± 0.4 mmHg.
In the contrast-enhanced tumor of GBM group, the mean p t O 2 was 15.2 ± 1.1 mmHg; in the nGBM patient, the volume of edema was delineated and the mean p t O 2 was found to be 19.4 ± 0.4 mmHg. Necrotic foci of the tumors in GBM patients also appeared abnormally elevated (arrows Fig. 1A(e)). Given that the retention of [ 18 F]-FMISO in a hypoxic cell occurs only when the partial pressure of oxygen is less than 10 mmHg 13 and that no [ 18 F]-FMISO metabolism should have occurred in the necrotic region of the tumor as well as in the health tissue, we hypothesized that the initial parameters reported for the operational equation would require further adaptation for normal and pathological brain tissues. Adaptation of the model for intracerebral tissues. We consequently determined the equation by fixing the p t O 2 in two regions (NAGM and NAWM) which allowed a hyperbolic function to convert the [ 18 F]-FMISO uptake to absolute values of p t O 2 . For each patient, a, b and c were estimated. The mean values were 1.5, 0.7 and 9.9 respectively.
After fitting the data and generating p t O 2 maps ( Fig. 1A(f)), the mean p t O 2 was 29.0 ± 5.7 and 41.5 ± 6.5 mmHg in NAGM and NAWM, respectively (p < 0.001). The mean estimated p t O 2 in the tumors was 4.8 ± 1.9 mmHg in the GBM group and 33.3 ± 11.3 mmHg for the nGBM group (p < 0.001) ( Fig. 2A). No significant differences were found between the oxygen status in NAGM and nGBM tissues.

ROC analyses for tumor and tissue type detection.
The ROC curves were performed to assess the ability of the various estimates to discriminate between GBM and nGBM patients (Fig. 3). The AUC value was 0.94 for CBV (threshold = 1.24); 0.96 for [ 18 F]-FMISO (threshold = 1.16). Interestingly; the AUC was 1.00 for p t O 2 (threshold = 8.07 mmHg).

Discussion
In GBM, hypoxia has been identified as an independent factor of poor prognosis 1 . We present here an approach to estimate p t O 2 maps calibrated for individual patients and based on [ 18 F]-FMISO PET that markedly discriminates severely hypoxic GBM from nGBM; p t O 2 maps correlate well with other indices of oxygen status and may suggest avenues to explore personalized radiotherapy.
In the present study, we demonstrate that the model initially proposed by Toma-Dasu and his group 12 for head and neck cancer to calculate intracerebral p t O 2 requires further adaptation in the specific instance of brain tumors. After an ad hoc calculation of the parameters for individual patients based on the values of [ 18 F]-FMISO retention in their corresponding NAGM and NAWM, we calculated p t O 2 in NAGM as 29.0 ± 5.7 mmHg and that of NAWM as 41.5 ± 6.5 mmHg. Measurements in the tumor of GBM subjects revealed a mean p t O 2 of 4.8 ± 1.9 mmHg that is in good agreement with the measurements obtained with oxygen sensitive probes 14 . However, one should mention that the assessment of tissue oxygen with probes remains problematic due to the high spatial heterogeneity observed in GBM patients. Additionally, in the nGBM group, the calculated p t O 2 was 33.3 ± 11.3 mmHg in the tumoral tissue. This value is closed to that of NAGM and NAWM values and also concords with previous probe-based results 15 . Our findings clearly demonstrate the existence of severe hypoxia in GBM tumors, which may, in turn, explain the limited response to radiation therapy. In terms of rCBV status, a marked and significant difference was observed between nGBM and GBM, which is in agreement with the literature 16 . The reproducibility of rCBV maps is reliable in non-pathological brain regions (NAGM and NAWM) in the two groups of patients (rCBV GM = 1.1 ± 0.2 and rCBV WM = 0.7 ± 0.1 respectively). Additionally, we failed to detect outright perfusion abnormalities in the nGBM group. In contrast, rCBV was increased in the GBM group (p < 0.05). The correlation between rCBV and p t O 2 revealed an inverse linear relationship. These results confirm those of Cher and colleagues 17 and those of Gerstner and colleagues 18 based on histological and MRI markers of the vasculature and hypoxia assessed by [ 18 F]-FMISO PET and reinforce the idea that the vessels observed in GBM are poorly functional; the result is a drastic fall in p t O 2 values. We also applied the ROC curve analysis on GBM and nGBM patients discrimination, and observed that the sensitivity and specificity were improved when p t O 2 was used instead of [ 18 F]-FMISO and rCBV maps.
A correlation was also found between p t O 2 values and the lactate and lipid concentrations, which may reflect an intense anaerobic metabolism resulting with lactate production in highly hypoxic tissues.
In the present study, we have developed a mathematical algorithm to enable the calculation of absolute p t O 2 in tumoral tissues at the individual level. The values so obtained are in accordance with other, albeit indirect, estimations of oxygenation status, namely blood volume and indices of anaerobic metabolism. As radiosensitivity parameters that take into account the OER, and the non-linearity between p t O 2 and [ 18 F]-FMISO, we believe that the present p t O 2 maps should be used rather than [ 18 F]-FMISO to evaluate the impact of dose escalation on tumor control probability (TCP) models.
The principal features of our investigation are undernoted. i) The determination of the three factors necessary to calculate p t O 2 from [ 18 F]-FMISO maps and, these, in individual ROI and in individual subjects. ii) Our innovative approaches allow in terms of tissue oxygenation, a marked discrimination between the values in GBM (severely hypoxic) and less aggressive glioma as well as apparently normal brain tissue, be it white or gray matter. The overall findings would encourage prospective study.
However, our proof of concept has some limitations. First, the calculation of p t O 2 is based on a mathematical model. Although other models have been proposed in the literature 9,11 , the relationship between [ 18 F]-FMISO uptake and p t O 2 remains hyperbolic. The advantage of our approach is that it requires two regions of interest for calibration of the model.

Another limitation of this study is that [ 18 F]-FMISO is trapped into hypoxic cells when p t O 2 is less than 10 mmHg 19 rendering the calculation of p t O 2 less adapted for low grade gliomas.
Finally, in our small cohort, [ 18 F]-FMISO uptake was only observed in GBM but not in less malignant gliomas (grade II and III). These findings concord with numerous published studies in which, GBM display invariably an increased uptake of [ 18 F]-FMISO 20,21 . Nonetheless, it is less evident that less aggressive glioma develop hypoxia. Cher and colleagues 22 reported that all the GBM showed a high uptake of [ 18 F]-FMISO, however they reported that only one patient of grade III shows a positive uptake of [ 18 F]-FMISO. In another study 23 , the authors reported a positive uptake in grade III but the uptake observed in grade IV was significantly higher. Our results concord with these observations since a very severe hypoxia was obtained for GBM and not for less aggressive grades, however, our proposed model needs to be tested in a larger cohort of GBM and less aggressive gliomas. Additionally, this hypoxic marker and its conversion could be a powerful tool for stratification in radiation treatment.
In the present study, the main goal was to analyze the feasibility of adapting a model, formerly developed for head and neck to brain tumors, as a proof of concept. Future investigations are needed to test this hypothesis on the tumor control improvement with radiotherapy ought to benefit from the wealth of information based on functional imaging. Molecular imaging with positron emission tomography (PET) tracers is sensitive to various biological parameters have now enlarged the range of functional information available for more targeted treatments planning by defining a so called " biological target volume " (BTV) treatment efficacy determination. The BTV could provide additional relevant data for dose painting in biologically active sub-regions or to boost radiotherapy.

Methods
Patients. This study was part of a prospective monocentric clinical trial called HYPONCO, funded by INCa (Institut National du Cancer). The study was approved by both the ethics committee (CPP Nord-Ouest III) and AFSSAPS (Agence française de sécurité sanitaire des produits de santé) and registered under EUDRACT number 2009-015543-16. More information's are available bellow: (NTC01200134 -https://www.clinicaltrials.gov/ct2/ show/NCT01200134). All procedures performed in the study were in accordance with ethical standards of both research committees. All patients are included between September 10, 2010 and November 16, 2012 after having been fully informed of the study and provided signed, informed consent.
Ten patients with confirmed GBM (based on the classification of the World Health Organization (WHO)) and three patients with less aggressive glioma were investigated. The characteristics of the patients are summarized (Table 1). After the discovery of a glioma (based on anatomical MRI), PET imaging was employed to detect regions of hypoxia. The day following PET imaging, an MRI study (anatomic, vascular and metabolic) was performed. PET images and anatomical 3D T1w-Gd images were used to guide the placement of the volumes of interest (VOI) for MRS.

Positron emission tomography/CT Scan. [ 18 F]-FMISO was synthesized by the LDM-TEP group
(UMR6301-ISTCT, Cyceron) based on methods previously described 24 . Acquisitions performed on a General Electric Discovery VCT 64 PET/CT camera, lasted 20 min and were carried out 120 minutes after the intravenous injection of ≈5 MBq/kg of [ 18 F]-FMISO. The attenuation-corrected images were reconstructed with an OSEM 2D algorithm (9 subsets and 2 iterations) and filtered in 3D with a Butterworth filter. Two blood samples were taken and the plasma averaged at the time of imaging for the calculation of T/B maps according to the following formula:  . The selection of these bandwidths is in accord with previously described in vitro and in vivo NMR studies of brain metabolites 25 . Areas of each frequency band of interest were then expressed as a percentage of the total area of the real value of spectra from 0 to 5.9 ppm. These proportions of area were taken as an index for the concentration of the metabolites. Lactate and lipid are pooled as it is difficult to distinguish the lactate doublets from lipids at 1.3 ppm. Accordingly we chose not to dissociate these compounds and to use a scale expressed as a percentage of the total area in order to obtain more precise indices of concentration, albeit with lower specificities. Pre-processing of the spectra was carried out on JMRUI 5.0 software.  26 . Active tumor areas were segmented manually on 3D T1w-Gd images with no necrotic foci (which is formerly defined as an anoxic region) and used to assess vascular abnormalities. For the nGBM subjects, defined by the absence of both an increased [ 18 F]-FMISO uptake and, vascular abnormalities, the region of interest (ROI) corresponding to the FLAIR hyperintensities was taken as the tumor. The edema is defined as a region with hyperintensities on T2w images, which encompasses the tumor core but also invaded tissue that cannot be considered anymore as a healthy tissue.

MRI, MRS and PET imaging co-registration.
Conversion of FMISO uptake into absolute values of the partial pressure of oxygen (p t O 2 ) in brain tissue. The mathematical model 12 initially proposed to convert [ 18 F]-FMISO images into absolute p t O 2 maps is: where u is the uptake as assessed by the T/B ratio and a, b and c are tissue-dependent, reaction-specific parameters. Based on the model described by Toma-Dasu and colleagues in neck and head tumors, which are also hypoxic 12 , a, b and c were 10.9, 10.7 and 2.5 respectively.
Scientific RepoRts | 7: 10210 | DOI:10.1038/s41598-017-08646-y The novel approach to convert [ 18 F]-FMISO to ptO2 maps. After co-registration of all images, a processing step was performed to segment normal appearing gray matter (NAGM) and normal appearing white matter (NAWM) so as to generate masks (SPM8 software). The edematous zones were avoided. Three patients are exemplified in Fig. S1. The median T/B value of [ 18 F]-FMISO uptake was measured in NAGM and NAWM region and p t O 2 was fixed at 30 mmHg 27 and 60 mmHg 28 respectively. A least-square error minimization (Levenberg-Marquardt non-linear regression) was then performed under Matlab 2012b ® to adjust a, b and c to the fixed p t O 2 in NAGM and NAWM respectively to calibrate the equation. Thereafter, we calculated p t O 2 in the whole brain by the model based on the estimates in normal appearing regions.
Statistical analyses. Box plots, linear and logarithmic regressions, Tuckey's HSD test and ROC curves analysis were performed using JMP Pro10 software (SAS). Further statistical analyses are detailed in the legends to figures. Data are presented either as mean ± sd or median and extremes of range.
Data availability. The datasets generated and/or analyzed during the study are available from the corresponding author on reasonable request.