Evaluation of the Lactate-to-N-Acetyl-aspartate Ratio Defined With Magnetic Resonance Spectroscopic Imaging Before Radiation Therapy as a New Predictive Marker of the Site of Relapse in Patients With Glioblastoma Multiforme
Alexandra Deviers, DVM, PhD,*,y,z Sole´akhe´na Ken, PhD,*,y Thomas Filleron, PhD,x Benjamin Rowland, PhD,* Andrea Laruelo, MSc,* Isabelle Catalaa, MD, PhD,y,jj Vincent Lubrano, MD, PhD,y,jj Pierre Celsis, MD, PhD,y Isabelle Berry, MD, PhD,y,jj,# Giovanni Mogicato, DVM, PhD,y,z Elizabeth Cohen-Jonathan Moyal, MD, PhD,*,{,# and Anne Laprie, MD, PhD*,y
Abstract
Purpose: Because lactate accumulation is considered a surrogate for hypoxia and tumor radiation resistance, we studied the spatial distribution of the lactate-to- N-acetyl-aspartate ratio (LNR) before radiation therapy (RT) with 3D proton magnetic resonance spectroscopic imaging (3D-1H-MRSI) and assessed its impact on local tumor control in glioblastoma (GBM).
Methods and Materials: Fourteen patients with newly diagnosed GBM included in a phase 2 chemoradiation therapy trial constituted our database. Magnetic resonance im- aging (MRI) and MRSI data before RT were evaluated and correlated to MRI data at relapse. The optimal threshold for tumor-associated LNR was determined with receiver-operating-characteristic (ROC) curve analysis of the pre-RT LNR values and MRI characteristics of the tumor. This threshold was used to segment pre-RT normal- ized LNR maps. Two spatial analyses were performed: (1) a pre-RT volumetric com- parison of abnormal LNR areas with regions of MRI-defined lesions and a choline (Cho)-to- N-acetyl-aspartate (NAA) ratio 2 (CNR2); and (2) a voxel-by-voxel spatial analysis of 4,186,185 voxels with the intention of evaluating whether pre-RT abnormal LNR areas were predictive of the site of local recurrence.
Results: A LNR of 0.4 (LNR-0.4) discriminated between tumor-associated and normal LNR values with 88.8% sensitivity and 97.6% specificity. LNR-0.4 voxels were spatially different from those of MRI-defined lesions, representing 44% of contrast enhancement, 64% of central necrosis, and 26% of fluid-attenuated inversion recovery (FLAIR) abnormality volumes before RT. They extended beyond the overlap with CNR2 for most patients (median: 20 cm3; range: 6-49 cm3). LNR-0.4 voxels were significantly predictive of local recurrence, regarded as contrast enhancement at relapse: 71% of voxels with a LNR-0.4 before RT were contrast enhanced at relapse versus 10% of voxels with a normal LNR (P<.01).
Conclusions: Pre-RT LNR-0.4 in GBM indicates tumor areas that are likely to relapse. Further investigations are needed to confirm lactate imaging as a tool to define addi- tional biological target volumes for dose painting. © 2014 Elsevier Inc.
Introduction
Glioblastoma (GBM) is associated with a poor prognosis as local relapse occurs several months after chemoradiation therapy (RT) (1-3). The failure to achieve sustainable local control in this tumor emphasizes the need to develop innovative treatment strategies. An attractive approach is to define new RT target volumes that include active disease, which can be highlighted by functional imaging (4, 5).
In this context, in vivo 1H magnetic resonance spectro- scopic imaging (MRSI) shows significant promise (6-9). This technique measures the concentration and spatial distribution of tissue metabolites like choline (Cho) and N-acetyl-aspartate (NAA), respectively, membrane and neuronal markers. An elevated Cho-to-NAA ratio (CNR) indicates increased cellular proliferation and reduced neuron density and is assumed to highlight a metabolically active part of the tumor in high-grade gliomas (HGG) (7, 9). This metabolic tool is a useful predictor of survival (10, 11) and relapse location in GBM patients (6, 8).
Lactate (Lac), the end-product of nonoxidative glycol- ysis, is another metabolite of interest that can be detected by MRSI. Lac production by tumor cells can be significantly increased in hypoxic environments (12) but also occurs under normoxic conditions due to molecular alterations, a phenomenon known as the Warburg effect (13, 14). Ste- reotactic microdialysis and MRSI performed in GBM pa- tients has shown high intratumoral levels of this metabolite (10, 11, 15-18). As Lac may be related to hypoxia, a factor in radiation resistance, and tumor malignancy, its accumula- tion could be associated with a poorer response to RT. High initial levels of Lac have actually been associated with a poor survival (10, 11, 18), and studies of the Lac-to-NAA ratio (LNR) in HGG have brought to the foreground the potential prognostic value of this metabolic tool. As Lac signal is usually not detectable in healthy tissue, voxels with clearly resolvable elevated Lac peaks, whether or not asso- ciated with a concomitant decrease of NAA signal, can be characterized by an elevated LNR (19). In HGG, initial high values of LNR in the tumor bed or a rise in LNR during treatment have, respectively, been correlated to poor prog- nosis and failure to respond to chemotherapy (16, 19). However, the assessment of a potential association between areas with an increased LNR and the site of relapse has not been previously performed.
The working hypothesis of this study was that GBM areas with an elevated LNR before treatment have a higher risk of local relapse. So far, no specific LNR value indicative of an underlying neoplastic process has been documented. The present study aimed to determine a cut-off value that could discriminate between normal and tumor-associated LNRs, to characterize the spatial distribution of LNR by 3D-MRSI in GBM before treatment, and to assess the impact of abnormally elevated LNR on local control and survival.
Methods and Materials
Patients
Between December 2005 and January 2009, 27 patients with newly diagnosed GBM were enrolled in a bicentric phase 2 clinical trial that associated 60-Gy conformal fractionated RT with continuous administration of the farnesyl-transferase inhibitor, tipifarnib (Zarnestra; Johnson & Johnson, New Brunswick, NJ), as described previously (20, 21). We analyzed the 14 patients who were treated at our center and underwent a combined MRSI and MRI examination after surgery or biopsy but before RT and then every 2 months after the end of RT until a relapse.
Acquisition of imaging data
Our planning computed tomography (CT) scan and MRI acquisition protocols, including 1H-MRSI, have been described previously (5). The volume of interest (VOI) in the chemical shift imaging (CSI) box led to the acquisition of 248 MRSI voxels per patient (on average), which repre- sented a total of 3472 MRSI voxels acquired before RT for the whole cohort of patients. MR images and 3D-MRSI were acquired during the same examination and in the same plane.
MRSI data processing
Spectral processing (water subtraction, low-pass filtering, frequency-shift correction, baseline and phase correction, and curve fitting in the frequency domain) and the computation of CNR and LNR maps were performed with the Syngo MR B17 spectroscopy application (Siemens, Erlangen, Germany). To avoid wrongly elevated metabolite ratio values, each MRSI voxel in the VOI was thoroughly reviewed, and voxels exhibiting a poor signal, insufficient water-signal suppression, bad spectral resolution, presence of acquisition artifacts, or contamination of the lactate peak by lipid signal (indicated by the disappearance of the negative doublet at 1.33 ppm, typically observed for a TE of 135 ms) were excluded from the analyses (22).
The pre-RT LNR and CNR maps were processed ac- cording to our previously published method (5). The snapshots of the metabolite ratio maps (LNR and CNR) coregistered to MR anatomical images and saved in 8-bit RGB were (1) separated from the MR anatomical images; (2) converted from RGB to hue-saturation-value color- space; (3) normalized across the entire 3D-MRSI volume of acquisition according to the maximum values of CNR and LNR given by the Siemens spectroscopy application; and (4) segmented according to specific threshold values. Thresholded areas were then re-mapped onto respective anatomical MR images with smooth linear interpolation to the final resolution of anatomical MR images, leading to anatomical-metabolic images that could be fused to plan- ning CT scans. A threshold value of 2 was used for seg- mentation of pre-RT CNR maps because CNR 2 (CNR2) is assumed to correspond to tumors according to biopsy correlation studies and to predict the site of relapse in GBM (6, 23). A threshold for LNR which allows discrimination between normal tissue and tumors has not been described in the literature. We determined such a value with receiver operating characteristic (ROC) curve analysis.
Determination of the threshold value for LNR
LNR values were collected from 2 types of tissue: (1) healthy tissue, defined as white matter distant from the neoplastic process, without contrast enhancement, without fluid-attenuated inversion recovery (FLAIR) ab- normality, and without CNR abnormalities (total of 210 MRSI voxels, ie, 15 per patient); and (2) tumor tissue, defined as a contrast-enhanced area that exhibited CNR2 (total of 134 MRSI voxels). ROC curve analysis was then carried out with these LNR values to determine the optimal threshold value that allowed discrimination be- tween tumor and nontumor tissue. This threshold was defined to maximize Youden’s index. Internal validation was performed using the bootstrap resampling method. The upper and lower confidence limits of the area under the ROC curve and the sensitivity and specificity at the estimated threshold were determined based on the boot- strap results.
Delineation of regions of interest for spatial analyses
For each patient, a total of seven regions of interest (ROI; ie, 5 anatomic ROIs and 2 metabolic ROIs) were manually delineated once the anatomic MRI modalities and segmented pre-RT LNR and CNR maps were rigidly cor- egistered and automatically fused to the planning CT scan, using iPlan RT image, version 3.0.1, contouring software (Brainlab, Feldkirchen, Germany). The anatomic ROIs were defined as CE, indicated by contrast-enhancement on T1-gadolinium (T1-Gd)-weighted images; Nec, hypo- intense necrotic regions within the contrast-enhanced areas; FLAIR*, hyperintensity on FLAIR images excluding CE and Nec; and site of relapse (SR), CE plus Nec at relapse excluding CE plus Nec before RT or resection cavity for patients without initial CE. When the CSI box could not cover the entire anatomic ROI, only the portion of the lesion included in the box was considered for analysis. Normal appearing white matter (NAWM) was defined as brain parenchyma in the CSI box that was not contrast- enhanced on T1-Gd MRI or hyperintense on FLAIR images.
The 2 metabolic ROIs were defined as CNR2 and abnormal LNR, respectively, brain regions with CNR 2 and a LNR beyond the threshold value determined by ROC curve analysis. MRSI abnormalities located in NAWM at >2 cm from the initial CE lesion were not included in the final analyses (2, 3). Contours of the anatomic and metabolic ROIs were linked to the planning CT scan and exported in RTStructSet files (ie, radiation therapy structure set, object of the DICOM RT standard) for spatial analyses.
Spatial analyses
For all patients, the contours were extracted from the exported RTStructSet files and rasterized to voxels using the resolution of the individual RTStructSet contour data and custom script written in Python Considering only rasterized voxels (r-voxels) inside the validated spectroscopy box led to a pool of 4,186,185 r-voxels across the 11 patients with local relapse inside the initial CSI box. For the 2 metabolic abnormality thresholds, each of these r-voxels was then classified as either “posi- tive” or “negative” depending on whether they were inside or outside of the metabolic ROI. Positive r-voxels were further classified as “true positive” if they were also spatially located inside the tumor volume at relapse (SR or CE at relapse). The positive predictive value of the pre-RT metabolic abnormalities was then defined as the number of true positive r-voxels over the total number of positive r-voxels.
Statistical analyses
Comparison of the group characteristics of the 4,186,185 r-voxels was performed using the c2 or Fischer exact test. Correlations between the 2 continuous variables were investigated using Spearman coefficients. All survival times were calculated from the date of registration, using the following first-event definitions: progression for time to progression (TTP) and death from any cause for overall survival (OS), and were estimated by the Kaplan-Meier method. Univariate analyses were performed using the log-rank test. Discrimination between high and low burden of Lac groups was defined according to the pre-RT maximum value of LNR (LNRmax). A minimum P value approach was used to dichotomize LNRmax referring to the TTP (100 bootstrap internal validation). All P-values were 2-sided. For all statistical tests, differences were considered significant at the 5% level. Statistical analyses were per- formed using STATA 12.0 software.
Results
Patients’ characteristics
The patients’ characteristics are provided in Table 1. By the end of the clinical trial, 10 of the 13 patients who had relapses had died. Relapse was found to occur within the 60-Gy isodose region for 12 patients, whereas 1 patient presented with a contralateral new lesion.
Definition of LNR abnormalities
ROC analysis resulted in a threshold value for the Lac-to- NAA ratio of 0.4 (area under the curve, 94.8%; 95% confidence interval, 91.08%-97.8%; sensitivity, 88.8%; specificity, 97.6%; Youden’s index, 0.864 [Supplemental Fig. 1]). Lac abnormalities were thus indicated as a Lac- to-NAA ratio of ≥0.4 (LNR-0.4).
Volumetric analysis before RT
Volumes of LNR-0.4 and anatomic ROIs are displayed in Table 2. LNR-0.4 was present in 11 patients, with volumes ranging from 2 to 50 cm3. The volume of LNR-0.4, on average, represented 44% of CE volume, 64% of Nec volume, 26% of FLAIR* volume, and 5% of the NAWM volume within the CSI box.
Distribution of LNR-0.4 volumes within each anatomic ROI
The distribution of LNR-0.4 volumes according to the different anatomic ROIs was compared to that of CNR2 and found to be similar (Fig. 1 A and B). Although most Lac abnormalities were found in the volume of the gross tumor (ie, CE and Nec), an important portion was distributed in microscopic disease or edema (FLAIR*) and peritumoral NAWM. Figure 2 A-C shows represen- tative images of the spatial location of LNR-0.4 before treatment in 2 patients with morphologically different diseases.
Comparison between LNR-0.4 and CNR2 volumes
The volume of LNR-0.4 was greater than the CNR2 volume for the 8 patients who had both Lac and CNR2 abnormal- ities before RT; results of the spatial comparison between these 2 ratios of metabolites are displayed in Figure 1C. A large volume of LNR-0.4 extended beyond the overlap with CNR2 for most of the patients (median: 20 cm3; range: 6-49).
Evaluation of LNR-0.4 as a predictive marker of local relapse
The positive predictive value of pre-RT LNR-0.4 (ie, ability to predict CE at relapse) was assessed and compared to that of CNR2 in 11 patients (patient 14 did not have a relapse, and patients 4 and 11 presented with a distant relapse). As shown in Figure 3A, 71% of the r-voxels with pre-RT LNR-0.4 spatially corresponded to CE at relapse, as opposed to only 10% of the r-voxels with an initial LNR <0.4. The positive predictive value was equivalent to that of CNR2 (Fig. 3B). Difference in the distribution of r-voxels ac- cording to their MRI and MRSI characteristics before RT and at relapse were statistically significant (P<.01).
The probability of matching SR was significantly higher in pre-RT FLAIR* with LNR-0.4 than pre-RT FLAIR* with no LNR-0.4 (Fig. 3C). Such outcomes were not found for pre-RT CNR2 (Fig. 3D). The positive predictive value of LNR-0.4 and CNR2 in NAWM was also significantly higher than the probability of having a relapse in meta- bolically normal NAWM (Fig. 3 E and F). Representative images of the predictive ability of pre-RT LNR-0.4 are presented in Figure 2 D and E.
Survival analyses
We assumed that the burden of Lac in the tissue could be characterized by both the extension and intensity of Lac accumulation (ie, the volumes of LNR-0.4 and LNRmax, respectively). LNRmax was significantly correlated with the volume of LNR-0.4 (Spearman rho Z 0.9204, P<.001), and the patients could be divided into 2 categories: those with a high burden of Lac (high LNRmax; LNR-0.4 volumes ranging from 20 to 50 cm3) and those with a low burden of Lac (low LNRmax; LNR-0.4 volumes ranging from 0 to 19 cm3). Distinction between a high and low burden of Lac was defined according to the value of LNRmax. Kaplan- Meier plots showed shorter TTP in patients with a high burden of Lac (3.4 months) than in patients with a low burden of Lac (5.5 months) (Fig. 4), but this difference was only at the limit of statistical significance (PZ.059). Considering OS, no significant difference was found be- tween the 2 groups of patients.
Discussion
In this study, we focused on the analysis of LNR in 14 postsurgery, pre-RT GBM patients included in a prospec- tive clinical trial that assessed the association of a targeted drug (tipifarnib) with RT. Because Lac is a marker of both hypoxia and tumor malignancy, we wanted to assess whether the pre-RT spatial distribution of LNR in GBM could have a significant impact on tumor local control and response to RT. Pre-RT areas with LNR of 0.4 were found to be significantly predictive of CE at relapse, thus high- lighting the radioresistant parts of the tumor. Although previous studies have analyzed Lac in HGG and have considered the role of this metabolite in defining malignant behavior (10, 11, 18, 25), this is the first time, to our knowledge, that the spatial distribution of LNR before RT has been compared to the site of relapse.
The first step of our work was to determine a threshold value for LNR that indicated abnormal areas. Previous studies using 3D-MRSI described elevated Lac signals in GBM but did not document a specific threshold value for LNR. An increase in LNR was assumed to be largely associated with increased Lac levels as this metabolite is not detectable in normal tissue, and only voxels with a good spectral quality were included in our study. The cut-off value of 0.4 was determined by ROC curve analysis, which discriminated tumor from nontumor LNR with high sensi- tivity and specificity. Because this analysis was achieved using MRI to distinguish tumors from healthy tissue, a large number of voxels collected from our patient cohort could be evaluated. A biopsy correlation study, although more accurate in discriminating tumor from nontumor tis- sue, would have led to the analysis of a considerably smaller number of voxels. Anatomic imaging, however, has the disadvantage of not being an exact reflection of the underlying pathologic processes. To overcome this limita- tion, we defined tumors as CE with a high cellular density (Cho-to-NAA ratio of 2) and healthy tissue as a subpart of NAWM distant from the gross lesion and without anatomic or metabolic abnormalities. The subsequent stages of this study were considered evidence of the reliability of our threshold in this patient cohort; a consistent pattern of spatial distribution and predictive value was shown for LNR-0.4 compared to CNR2.
The second step of our work aimed at spatially comparing pre-RT volumes of LNR-0.4 with pre-RT vol- umes of anatomic abnormalities and CNR2. This analysis showed a decreasing percentage of LNR-0.4 volumes from the center of the tumor (necrosis) to the periphery, which is in accordance with Lac production by hypoxic cells and Lac accumulation in the necrotic center (26, 27). In agreement with this process, we found maximum values of LNR in areas of Nec or blood-brain barrier breakdown (CE) in most of the patients. Only a small proportion (5%) of the NAWM investigated was found to have LNR-0.4; these metabolically abnormal areas were located close to anatomic lesions, either adjacent to CE or to FLAIR hyperintensity. Although these regions may represent false positives, their peritumoral locations suggest neoplastic infiltration that is not visible with MRI. This latter expla- nation is consistent with biopsy correlation studies, which demonstrate isolated neoplastic cells in peritumoral NAWM and highlight the interest of MRSI in obtaining additional information about tumor infiltration (8, 9, 23).
The spatial comparison between LNR-0.4 and CNR2 showed a common volume of metabolic abnormalities. This finding might be explained by a decreased NAA level in these areas (collapse of nervous tissue due to the neoplastic process) but might also reflect the metabolic profile of cancer cells in which Lac is assumed to provide the necessary energy for their proliferation (28). LNR-0.4 volumes were shown to extend largely beyond CNR2 volumes, suggesting that Lac may provide additional spatial information compared to CNR2.
The third step of our work aimed to evaluate the potential impact of tissue Lac accumulation on the local control of GBM. In this context, we found that LNR-0.4 areas before RT were significant predictors of tumor presence at relapse, as defined according to radiographic criteria. The association of LNR-0.4 areas with post-RT persistent and/or new CE sug- gests that the spatial distribution of this metabolic tool could map radiation-resistant areas.
In a final step, the impact of LNR abnormalities on sur- vival was evaluated because pre-RT LNR-0.4 volumes seemed to be associated with a lack of local response to treatment. Although a high burden of Lac was associated with a shorter TTP in this study, this was not statistically significant, probably because of the small size of our cohort.
We plan to extend our investigation using a larger cohort of GBM patients currently enrolled in a multicentric phase 3 clinical trial promoted by our institution (trial registration ID: NCT01507506). This trial is evaluating whether tar- geting CNR2 volumes with a simultaneous integrated boost of intensity modulated RT can improve survival of GBM patients.
Conclusions
In conclusion, we determined a tumor-associated threshold value for LNR of 0.4, using 3D-MRSI in pre-RT GBM patients. We found that the spatial distribution of LNR-0.4 could significantly predict the site of relapse, suggesting that this metabolic tool may characterize radiation-resistant areas. Although these findings must be confirmed in a larger cohort, LNR may represent an additional target for further RT dose-painting trials.
References
1. Stupp R, Mason WP, van den Bent MJ, et al. Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. N Engl J Med 2005;352:987-996.
2. Sherriff J, Tamangani J, Senthil L, et al. Patterns of relapse in glio- blastoma multiforme following concomitant chemoradiotherapy with temozolomide. Br J Radiol 2013;86:2012-2014.
3. Wallner KE, Galicich JH, Krol G, et al. Patterns of failure following treatment for glioblastoma multiforme and anaplastic astrocytoma. Int J Radiat Oncol Biol Phys 1989;16:1405-1409.
4. Bentzen SM, Gregoire V. Molecular imaging-based dose painting: A novel paradigm for radiation therapy prescription. Semin Radiat Oncol 2011;21:101-110.
5. Ken S, Vieillevigne L, Franceries X, et al. Integration method of 3D MR spectroscopy into treatment planning system for glioblastoma IMRT dose painting with integrated simultaneous boost. Radiat Oncol 2013;8:1.
6. Laprie A, Catalaa I, Cassol E, et al. Proton magnetic resonance spectroscopic imaging in newly diagnosed glioblastoma: Predictive value for the site of postradiotherapy relapse in a prospective longi- tudinal study. Int J Radiat Oncol Biol Phys 2008;70:773-781.
7. McKnight TR, von dem Bussche MH, Vigneron DB, et al. Histo- pathological validation of a three-dimensional magnetic resonance spectroscopy index as a predictor of tumor presence. J Neurosurg 2002;97:794-802.
8. Park I, Tamai G, Lee MC, et al. Patterns of recurrence analysis in newly diagnosed glioblastoma multiforme after three-dimensional conformal radiation therapy with respect to pre-radiation therapy magnetic resonance spectroscopic findings. Int J Radiat Oncol Biol Phys 2007;69:381-389.
9. Pirzkall A, Li X, Oh J, et al. 3D MRSI for resected high-grade gliomas before RT: Tumor extent according to metabolic activity in relation to MRI. Int J Radiat Oncol Biol Phys 2004;59:126-137.
10. Crawford FW, Khayal IS, McGue C, et al. Relationship of pre-surgery metabolic and physiological MR imaging parameters to survival for patients with untreated GBM. J Neurooncol 2009;91:337-351.
11. Saraswathy S, Crawford FW, Lamborn KR, et al. Evaluation of MR markers that predict survival in patients with newly diagnosed GBM prior to adjuvant therapy. J Neurooncol 2009;91:69-81.
12. Kim J, Tchernyshyov I, Semenza GL, et al. HIF-1-mediated expres- sion of pyruvate dehydrogenase kinase: A metabolic switch required for cellular adaptation to hypoxia. Cell Metab 2006;3:177-185.
13. Gatenby RA, Gillies RJ. Why do cancers have high aerobic glycol- ysis? Nat Rev Cancer 2004;4:891-899.
14. Warburg O. On the origin of cancer cells. Science 1956;123:309-314.
15. Tabatabaei P, Bergstro¨m P, Henriksson R, et al. Glucose metabolites, glutamate and glycerol in malignant glioma tumours during radio- therapy. J Neurooncol 2008;90:35-39.
16. Tarnawski R, Sokol M, Pieniazek P, et al. 1H-MRS in vivo predicts the early treatment outcome of postoperative radiotherapy for malignant gliomas. Int J Radiat Oncol Biol Phys 2002;52:1271-1276.
17. Li X, Vigneron DB, Cha S, et al. Relationship of MR-derived lactate, mobile lipids, and relative blood volume for gliomas in vivo. AJNR Am J Neuroradiol 2005;26:760-769.
18. Li Y, Lupo JM, Parvataneni R, et al. Survival analysis in patients with newly diagnosed glioblastoma using pre- and postradiotherapy MR spectroscopic imaging. Neuro Oncol 2013;15:607-617.
19. Sankar T, Caramanos Z, Assina R, et al. Prospective serial proton MR spectroscopic assessment of response to tamoxifen for recurrent ma- lignant glioma. J Neurooncol 2008;90:63-76.
20. Ducassou A, Uro-Coste E, Verrelle P, et al. avb3 Integrin and Fibroblast growth factor receptor 1 (FGFR1): Prognostic factors in a phase I-II clinical trial associating continuous administration of Tipifarnib with radiotherapy for patients with newly diagnosed glio- blastoma. Eur J Cancer 2013;49:2161-2169.
21. Cohen-Jonathan Moyal E, Laprie A, Delannes M, et al. Phase I trial of tipifarnib (R115777) concurrent with radiotherapy in patients with glioblastoma multiforme. Int J Radiat Oncol Biol Phys 2007;68:1396- 1401.
22. Kreis R. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR Biomed 2004;17: 361-381.
23. Guo J, Yao C, Chen H, et al. The relationship between Cho/NAA and glioma metabolism: Implementation for margin delineation of cerebral gliomas. Acta Neurochir (Wien) 2012;154:1361-1370. discussion 1370.
24. Rossum, G. V. & Drake, F. L. J. The Python Language Reference Manual. (Network Theory Ltd., 2011). Available at http://www. network-theory.co.uk/python/language/.
25. Laprie A, Pirzkall A, Haas-Kogan DA, et al. Longitudinal multivoxel MR spectroscopy study of pediatric diffuse brainstem gliomas treated with radiotherapy. Int J Radiat Oncol Biol Phys 2005;62:20-31.
26. Garcia-Martin ML, Herigault G, Remy C, et al. Mapping extracellular pH in rat brain gliomas in vivo by 1H magnetic resonance spectro- scopic imaging: Comparison with maps of metabolites. Cancer Res 2001;61:6524-6531.
27. Manzoor AA, Schroeder T, Dewhirst MW. One-stop-shop tumor imaging: Buy hypoxia, get lactate free. J Clin Invest 2008;118: 1616-1619.
28. Sonveaux P, Ve´gran F, Schroeder T, et al. Targeting lactate-fueled respiration selectively kills hypoxic tumor cells in mice. J Clin Invest 2008;118:3930-3942.