Metabolomics

, Volume 11, Issue 6, pp 1834–1847 | Cite as

The three glioma rat models C6, F98 and RG2 exhibit different metabolic profiles: in vivo 1H MRS and ex vivo 1H HRMAS combined with multivariate statistics

  • Nicolas Coquery
  • Vasile Stupar
  • Régine Farion
  • Severine Maunoir-Regimbal
  • Emmanuel L. Barbier
  • Chantal Rémy
  • Florence Fauvelle
Original Article

Abstract

Glioblastomas are the most malignant subtypes of glioma and many efforts are currently made to improve their characterization though molecular, microvascular, immunogenic and metabolomic approaches. The variability within pre-clinical tumor models may mimic glioma heterogeneity and force the development of innovative analytical methodologies. In this study, we investigate the metabolic variability within three rat models of glioma: C6, RG2 and F98, using in vivo magnetic resonance spectroscopy (1H MRS) and ex vivo high resolution magic angle spinning (1H HRMAS MRS). We used a multivariate statistic approach with orthogonal projection to latent structure-discriminant analysis (OPLS-DA) that was compared with univariate statistic. OPLS-DA reveals a clear separation between C6, RG2 and F98 tumors and, with the help of shared and unique structure plot (SUS-Plot), promotes a comprehensive view of their metabolic differences. Both in vivo and ex vivo analyses are similar but ex vivo 1H HRMAS MRS provides more robust results. In conclusion, MRS-based OPLS-DA appears sensitive enough to correctly predict the classification of tumors and to investigate the relationship between the host brain metabolism and the grafted tumor.

Keywords

Metabolomic Magnetic resonance spectroscopy High resolution magic angle spinning Glioma models OPLS-DA Radiomic 

Abbreviations

Ace

Acetate

Ala

Alanine

Asp

Aspartate

Bet

Betaine

Cho

Choline

GABA

Gamma-amino-butyric acid

Gln

Glutamine

Glu

Glutamate

Gly

Glycine

GPC

Glycerophosphocholine

Gsh

Glutathione

Hyp

Hypotaurine

Lac

Lactate

M-ins

Myo-inositol

NAA

N-acetylaspartate

PC

Phosphocholine

tCr

Total creatine (phosphocreatine and creatine)

PE

Phosphoethanolamine

S-Ins

Scyllo-inositol

Tau

Taurine

MM

Macromolecules

MRS

Magnetic resonance spectroscopy

HRMAS

High resolution magic angle spinning

jMRUI

Java based version of the Magnetic Resonance User Interface

OPLS-DA

Orthogonal projection to latent structure-discriminant analysis

PCA

Principal component analysis

SUS-plot

Shared and unique structure plot

GBM

Glioblastomas

PRESS

Point RESolved spectroscopy

CPMG

Carr-Purcell-Meiboom-Gill CPMG

CRLB

Cramer Rao lower bounds

FID

Free induced decay

NMR

Nuclear magnetic resonance

1 Introduction

Among gliomas, glioblastomas (GBM) are the most malignant subtypes and, since the protocol from Stupp et al. (2005) few improvements have been made regarding therapy outcome in patients. Besides the development of new therapeutically strategies, efforts are currently made to characterize molecular, microvascular, immunogenic and metabolic subclasses of GBM that, depending on their response to treatment, might ultimately lead to personalized medicine (Weller et al. 2012).

In rat, several cell lines, e.g. C6, F98 and RG2, induce reliable models for in vivo studies in orthotopic scenario (Barth and Kaur 2009). C6 tumor was isolated from Wistar rats and RG2, F98 from Fisher rats. The couple tumor/host in orthotopic and syngeneic model is of high importance in order to avoid severe immune effect that could lead to bad tumor growth or even tumor rejection, and to respect the proper complexity of tumor with respect to human pathology (Huszthy et al. 2012). These models also present distinct phenotypes at different levels that could be used to mimic human glioma heterogeneity. Barth and Kaur (2009) described C6 as invasive astrocytomas, F98 as anaplastic or undifferentiated glioma and RG2 as a good model of high-grade glioma regarding the induced vascular alterations. Despite the few number of studies comparing these tumor models, several differences have been found in vivo in their metabolism (Doblas et al. 2012), vascular (Doblas et al. 2010) and microvascular features (Coquery et al. 2012, 2014; Lemasson et al. 2013; Valable et al. 2008). Cuperlovic-Culf et al. (2012) showed recently that the metabolic profiles of nine human cell lines could be associated to the 4 relevant subtypes of GBM, that have been identified in the Cancer Genome Atlas network (Verhaak et al. 2010). However, the impact of brain metabolism and physiology is not taken into account when using cell lines. Doblas et al. (2012) characterized by in vivo magnetic resonance spectroscopy (MRS) several rat and mouse glioma models, however it can be difficult to extract the most important features from multiple group comparisons with univariate statistics. In this study, we used multivariate statistics to find the metabolites that most clearly discriminate the C6, F98 and RG2 tumor models.

1H MRS is a powerful analytical method to investigate the complex metabolic consequences of cancer disease (Griffin and Shockcor 2004; Kauppinen and Peet 2011). In vivo 1H MRS can provide information regarding glioma growth, tumor grading and response to treatment (Bulik et al. 2013; Glunde et al. 2011). However, its clinical use remains so far limited given the duration of the examination and the restricted number of provided metabolites. A wider range of metabolites can be obtained ex vivo in biopsies using 1H high resolution magic angle spinning (1H HRMAS) MRS. Besides validation of in vivo 1H MRS method (Glunde et al. 2011; Opstad et al. 2010; Wilson et al. 2009), ex vivo 1H HRMAS MRS may also be used on tumor biopsy in a clinical context (Erb et al. 2008). Multivariate statistical methods including principal component analysis (PCA) and orthogonal projection to latent structure-discriminant analysis (OPLS-DA), allow to reduce the number of variables to 2 or 3 latent variables, that describe data variance. Compared with metabolite level/ratio analysis, OPLS-DA might refine diagnosis (Holmes et al. 2006) or potentially the prediction of treatment response.

Here, we aimed to assess whether the 3 tumor models C6, F98 and RG2 could be distinguished on the basis of metabolic profiles, by using two 1H MRS methods, in vivo MRS and ex vivo HRMAS MRS on the same animals. Chemiometric methods like PCA and OPLS-DA were used to find the most discriminant metabolites. Since two rat strains were needed for syngeneic models, we also analyzed contralateral tissue in order to dissociate the impact of tumor from the impact of rat strain on tumor tissue metabolism.

2 Materials and methods

2.1 Animal tumor models

C6 cells (1 × 105 cells, ATCC: CCL-107), RG2 cells (1 × 105 cells, ATCC: CRL-2433), and F98 cells (1 × 103 cells, ATCC: CRL-2397) were implanted in the right caudate nucleus (coordinates from Bregma: AP = 0, ML = 3.5, DV = 5.5 mm) of male Wistar rats (for C6 model, n = 12, 175-200 g, Janvier, France) and male Fischer 344 rats (for RG2 model, n = 8; for F98 model, n = 12; 175–200 g, Charles River, France) with a stereotaxic frame, as previously described (Coquery et al. 2012). Briefly, five microliters of tumor cell suspension in serum-free alpha-MEM medium were inoculated under anesthesia: 5 % isoflurane for induction and 2.5 % for maintenance in 100 % air. Bupivacaine (8 mg/kg, Centravet) was subcutaneously injected before incision to prevent postoperative pain.

2.2 In vivo 1H MRS acquisition

Twenty-four animals (10 C6, 8 F98 and 6 RG2) were submitted to in vivo MRS. 1H MRS experiments were performed on a 7 T MRI system (Biospec Advance III; Bruker, Ettlingen Germany). A 72 mm inner diameter linear volume coil was used for RF transmission and an actively decoupled quadrature surface coil was used for signal detection. Animals were analysed between 21 and 24 days after tumor cell implantation. Animals were anesthetized as described above. The body temperature was maintained at 37 °C using a warm water circuit. The breath rate was monitored during the entire acquisition protocol and kept between 40 and 60 breaths per minute. Localizer images, covering the entire tumor core, were acquired in axial plane with a multi-slice fast spin echo T2-weighted sequence (TR/TE = 4000/33 ms, in-plane resolution = 117 × 117 µm, slice thickness = 1 mm, 19 slices) and were used to determine tumor volume by delineating the tumor core. For each animal, two MRS data sets were acquired from tumor and contralateral striatum. The volume of interest was set to 27 mm3 (3 × 3 × 3 mm3) and was placed either in the center of the tumor in order to cover the most of tumor core or in the center of the contralateral striatum in order to cover the most of homogenous tissues as seen with the localizer image. Prior to data acquisition, the static magnetic field homogeneity was adjusted across the volume of interest using the MAPSHIM technique (Kanayama et al. 1996). The Point RESolved Spectroscopy (PRESS) localization sequence (Bottomley 1987) was used with TE/TR = 20/2500 ms, VAPOR water suppression and outer volume suppression modules (Tkáč et al. 1999). The acquisition bandwidth was set to 4 kHz around the water frequency to avoid aliasing. For each data set, 400 averages were performed for a total acquisition time of 16 min 40 s. Resonance assignment was performed as previously described (Govindaraju et al. 2000, de Graaf 2007), and with the help of ex vivo assignments (Supplementry Table 1).

2.3 Ex vivo 1H HRMAS MRS acquisition

2.3.1 Preparation of tumor and brain samples

One day after 1H MRS acquisitions, rats were deeply anesthetized with 5 % isoflurane and brains were rapidly isolated after decapitation. One brain slice of 2 mm was cut at the tumor site with a dedicated brain sectioning frame, the center of the entire visible tumor or striatum was dissected out, immediately frozen in liquid nitrogen, and stored at −80 °C to prevent biochemical and structural degradation. This procedure was performed at 4 °C and never exceeded 3 min. For 1H HRMAS MRS analysis, 30 μL of a cold 1 mM D2O solution of 3-(trimetylsilyl) propionic-2,2,3,3-d4 acid (TSP) was added to approximately 15 mg of frozen biopsy in a 50 μL zirconium rotor. The rotor was closed with a cap and inserted in the pre-cooled HRMAS probe at 4 °C. All HRMAS experiments were performed at 4 °C and were started immediately after the temperature has reached the equilibrium inside the probe (10 min).

As for in vivo 1H MRS, a total of 24 tumor spectra and their corresponding 24 contralateral spectra were obtained. Ex vivo 1H HRMAS MRS was also performed on tumor and contralateral samples from 8 additional rats (2 C6, 4 F98, 2 RG2).

2.3.2 1H HRMAS MRS acquisition

Spectra were recorded at a proton frequency of 400.13 MHz on a Bruker Avance III 400 spectrometer (BrukerBiospin, Wissembourg, France) equipped with a 1H–13C–31P 4 mm gradient HRMAS probe. A 4 kHz spinning rate was used to keep spinning sidebands out of the spectral region of interest. The standard one-dimensional Carr-Purcell-Meiboom-Gill (CPMG) Bruker pulse sequence was used, synchronized with the spinning rate (interpulse delay 250 µs, total spin echo time 30 ms) (Wieruszeski et al. 2001; Piotto et al. 2009). This sequence allows to reduce lipid and macromolecule signals. Residual water peak was pre-saturated with low power irradiation during the 2 s relaxation delay. A 14 ppm bandwidth was used, with 32 K points. Each acquisition of 256 scans lasted 16 min.

Resonance assignment was performed as previously described (Fauvelle et al. 2012; Govindaraju et al. 2000; Rabeson et al. 2008; see Supplementry Table 1). Additional 2D experiments were also performed when needed (i.e. hypotaurine and betaïne in tumor tissue) (Wright et al. 2010; Ziegler et al. 2001). Addition of metabolite solutions to biopsies was also performed to confirm betaïne assignments.

2.4 Metabolite quantification

Quantification was performed in the time domain as previously described (Fauvelle et al. 2012; Rabeson et al. 2008) with the jMRUI software package (http://www.mrui.uab.es/mrui/) using the “subtract-QUEST” procedure (Ratiney et al. 2005). The procedure involves a simulated database composed of all assigned metabolites, and truncation of the first FID points for background estimation (baseline, lipids and macromolecules).

2.4.1 Ex vivo 1H HRMAS MRS quantification

The database included acetate (Ace), alanine (Ala), aspartate (Asp), creatine and phosphocreatine (tCr), choline (Cho), gamma-amino-butyric acid (GABA), glutamate (Glu), glutamine (Gln), glutathione (GSH), glycerophosphocholine (GPC), glycine (Gly), lactate (Lac), myo-inositol (M-ins), N-acetylaspartate (NAA), phosphoethanolamine (PE), phosphocholine (PC), scyllo-inositol (S-ins), and taurine (Tau). Hypotaurine (Hyp) and betaïne (Bet) were detected only in tumor tissue (Wright et al. 2010) and were then added to the database of tumor. The first sixteen points of FID were used to estimate the non-parametric part of the signal, i.e. lipids and macromolecules, in order to generate pure-metabolite spectra. The amplitude of metabolites calculated by QUEST was normalized to the total spectrum signal, so only relative concentrations were produced.

The Cramer Rao lower bounds (CRLB) determined by the jMRUI algorithm are estimates of the standard deviation of the fit for each metabolite. We obtained CRBL ≤5 % for most metabolites and for all quantifications, and CRBL ≤25 % for Ace and Gsh. However, Asp and S-ins could not be reliably quantified in most tumor spectra, so these two metabolites were excluded from statistical analysis.

2.4.2 In vivo 1H MRS quantification

Eleven metabolites were quantified: tCr, total choline (tCho) composed of GPC and PC, GABA, Glu, Gln, Hyp, Lac, M-ins, NAA, Asp and Tau. Hyp was added to the database of tumor only. Three broad peaks were also included in the database at 0.9, 1.3 and 1.7 ppm as lipid and macromolecule compounds (see Fig. 1, MM compounds), and the first ten points of free induced decay (FID) were also used for a better fit of the baseline. Normalization to the sum of metabolites was used as seen in ex vivo quantification, and the three lipid and macromolecule peaks were excluded from the sum of metabolites at this normalization step.
Fig. 1

a Representative anatomical T2-weighted images of C6, F98 and RG2 tumors. b In vivo 1H MRS and c ex vivo 1H HRMAS MRS spectra of contralateral striatum and tumor of the F98 tumor model. An anatomical T2-weighted image is depicted with the two regions of interest used for in vivo acquisition: tumor and contralateral striatum

CRLB were higher than in ex vivo quantification (<50 %) due to the poor resolution of in vivo spectra, we consequently did not exclude metabolites based on CRLB. Only Hyp, which could only be quantified in two rats, was excluded from statistical analysis.

2.5 Statistical analyses

To find metabolic profiles that are specific to each tumor C6, F98 and RG2, multivariate statistical models were built separately either with quantified data from in vivo 1H MRS spectra or with quantified data from ex vivo 1H HRMAS MRS spectra. Classical univariate statistical analyses were also performed to validate our multivariate statistical analyses

2.5.1 Multivariate statistical analysis

In vivo or ex vivo quantified data were analysed with the SIMCA-P software version 13 (Umetrics, Umea, Sweden) and scaled to unit variance before analysis. This constitutes the “X” variable matrix.

First, global unsupervised Principal component analysis (PCA) was carried out on all data from both tumor and contralateral striata grouped to ensure good homogeneity of data and possibly to exclude outliers. For this aim, data were visualized by score plots, where each point represents a MRS spectrum and thus a sample.

Then, a supervised approach was used. OPLS-DA models were built with tumor data only, or with contralateral data only (ex vivo only), using the three tumor: C6, F98 and RG2 as membership classes (the “Y” variable matrix), or the strains (Fisher-Wistar) (Wold et al. 1984). Compared to PLS-DA, this method allows separating systemic intra-class variations in X that are unrelated to Y (orthogonal) from linear variations in X that are related to Y (inter-class variations). At this step, exclusion of outliers could be possibly performed after examination of the residual error; i.e. the distance to the model.

All OPLS-DA models were cross-validated in order to find the appropriate number of components for the final model and to allow evaluation of its statistical significance. Cross-validation is a procedure during which the model is iteratively rebuilt using a part of the data as training set. The model is then used to predict the class of the remaining data that serves as a test sample. The iterative classical SIMCA procedure was used for repartition of samples between test and training groups (seven groups). Cross-validation leads to the calculation of the R2Y and Q2 factors, Q2 estimates goodness of prediction while R2Y estimate goodness of fit. A model is considered as robust when Q2 > 0.5 and a R2Y > 0.5. The reliability of our OPLS-DA models was assessed by a CV-ANOVA test (analysis of variance, test of cross-validated predictive residuals).

All OPLS-DA models were visualized by plotting the score of individuals relative to the two first components, either first predictive and first orthogonal components or two predictive components, depending on the models.

The variable loadings were also plotted, however they are often uneasy to interpret in a multiple class OPLS-DA, i.e. multiple group comparison. Three Pair-wise OPLS-DA: C6-F98, C6-RG2 and RG2-F98 were then performed for identification of the most discriminant metabolites in each comparison. All data were synthesized in a shared and unique structure plot (SUS-plot) (Wiklund et al. 2008), corresponding to the superimposition of two loadings plots, with loadings expressed as correlation instead of covariance. Metabolites with absolute correlations of the same order of magnitude in the two OPLS-DA models will then be found on diagonals, they correspond to metabolites that vary in the same direction or in opposite direction in the two statistical models, depending on the sign of correlations. On the contrary, metabolites found outside the diagonals have a unique evolution in the corresponding statistical model (depending on their coordinates). In this study we chose to plot C6-F98 versus C6-RG2, since C6 was raised in a rat strain different from that of F98/RG2. The SUS-plot will then highlight both shared variations of F98/RG2 relative to C6 and their unique variations relative to C6. We can ultimately deduce from this SUS-plot the particularities of each tumor type.

Variable influence on projection (VIP) was also calculated for each metabolite and allows ranking the metabolites according to their contribution to the separation between groups. Metabolites with VIP > 1 are considered as the most important for separation.

Finally, we compared the output of the ex vivo 3 class model with the actual results obtained with 8 animals that were submitted to the same analytical protocol that the other (C6, n = 2; RG2, n = 2; and F98, n = 4) but were not used for building the OPLS-DA model. Quantified data were imported in the model with no a priori class assignment, and the score scatter plots drawn for graphical inspection. These observations were automatically assigned to the nearest class and the classification rate, in percentage, was produced to evaluate the predictability of the model.

2.5.2 Univariate statistical analysis

A non-parametric paired Wilcoxon test was performed to compare quantified data in tumor samples in comparison with their contralateral counterpart. The Kruskal–Wallis ANOVA by ranks was performed to compare the different tumor models or the contralateral groups. These tests were followed by a post hoc multiple comparison test (Statistica v9 software, Statsoft, Inc.). A p value inferior to 0.05 was considered as significant.

3 Results and discussion

3.1 Animal number

Statistical multivariate analyses were performed on the data from the 24 animals that were submitted to both in vivo and ex vivo MRS. Following their distance to the model (see Sect. 2), two C6 rats were identified as outliers in ex vivo multivariate statistics and were then excluded from all further statistical analysis, both multivariate and univariate. One of these excluded animals had a hemorrhagic tumor. The 22 remaining rats were thus: 8 C6, 8 F98 and 6 RG2 rats.

3.2 Tumor anatomical description

Tumor size was similar between groups (C6: 56.8 ± 16.9 mm3; F98: 46.5 ± 15.9 mm3; RG2: 48.9 ± 8.4 mm3). RG2 tumors were homogenous with good intra tumor homogeneity as already shown by Coquery et al. (2012). F98 and C6 tumors were heterogeneous (Fig. 1a). C6 tumors were mostly bipolar with a cortical part having hemorrhagic/necrotic characteristics and a striatal part with diffuse characteristics of cell infiltration as already described (Coquery et al. 2014). In nearly half of F98 tumors, a necrotic signature was detected in the center of the tumor, i.e. white areas within tumor bulk on T2 weighted images. RG2 and F98 tumors were mostly found in striatum and were surrounded by a ring with high intensity on anatomical T2 weighted images, which might be regarded as an inflammation and/or edema around the tumor core.

3.3 Metabolic changes in tumor versus contralateral striatum

Ex vivo and in vivo spectra shared many features although the resolution of ex vivo 1H HRMAS MRS spectra was higher (Fig. 1b, c).

Contralateral striata exhibited a typical brain metabolic profile with the most intense resonances attributed to NAA, tCr and tCho whereas tumors exhibited a global decrease of NAA and tCr resonances and increase of tCho/tCr ratio.

Common features of the three tumors were highlighted with both in vivo and ex vivo analysis (Table 1 and for more details Supplementry Table 2) such as decreased NAA and Glu, and less consistently decreased tCr (except C6 in vivo and ex vivo) and increased M-ins (except F98 ex vivo) and Lac (except C6 in vivo). Ex vivo 1H HRMAS MRS yielded additional information compared to in vivo 1H MRS. GABA was significantly decreased and Gly was increased in all tumor models whereas Cho and Gsh were not statistically different from their contralateral level. Gln was decreased only in C6 and RG2 tumors. Hyp, Ala, and choline compounds (GPC and PC) were increased only in RG2 and F98 tumors. Specific features of RG2 tumors were detected such as a decrease of Ace and an increase of PE. The main characteristics observed in vivo for the 3 glioma models (decrease of NAA, Glu, tCr and increase of Lactate,) are consistent with those generally described in the literature in preclinical and clinical studies (Bulik et al. 2013; Cuperlovic-Culf et al. 2012; Doblas et al. 2012). Previously, only Doblas et al. (2012) compared the metabolism of different glioma models in vivo. Their data are mostly similar to ours but some differences can be found: we showed that NAA was decreased in F98 and Gln was not modulated in all tumor types. These differences can be explained by the difference in experimental set-up: site of tumor cell implantation (cortical versus striatal), difference on tumor stage at the time of acquisition, difference in quantification method. Bet, an end-product of choline catabolism in glioma (Bansal et al. 2008) was detected here in all tumor models while not detected in contralateral striata. Interestingly, a high level of Bet has previously been documented in human GBM in comparison with other solid brain tumors (Wright et al. 2010). Overall, this is consistent with good face validity of these tumor models in regard to human gliomas.
Table 1

Metabolic variation in each tumor versus contralateral striatum

 

C6 (n = 8)

F98 (n = 8)

RG2 (n = 6)

In vivo

Ex vivo

In vivo

Ex vivo

In vivo

Ex vivo

Ace

n. a

=

n. a

=

n. a

Ala

n. a

=

n. a

n. a

Bet

n. a

n. a

n. a

tCho

=

=

=

Cho

n. a

=

n. a

=

n. a

=

GPC

n. a

=

n. a

n. a

PC

n. a

=

n. a

n. a

GABA

=

=

=

Gln

=

=

=

=

Glu

Gly

n. a

n. a

n. a

Gsh

n. a

=

n. a

=

n. a

=

Hyp

=

=

=

=

Lac

=

M-ins

=

NAA

PE

n. a

=

n. a

=

n. a

Tau

=

=

=

=

=

tCr

=

=

Paired Wilcoxon test was used, and show respectively statistically significant increase and decrease for a p value < 0.05

n.a. non-admitted quantification

3.4 Can we separate the three tumor types?

To address this issue, and to identify which metabolites are the most important for the characterization of a given tumor model, multivariate statistical analyses were performed as described above with ex vivo data and then with in vivo data.

3.4.1 Ex vivo MR-based separation

On the score scatter plot of the PCA analysis conducted with the 24 tumor and 24 contralateral data (Fig. 2a), the first component PC1 clearly separates contralateral striatum from tumor, while the second component PC2 separates the three tumor models.
Fig. 2

a Score scatter plot of the PCA analysis of ex vivo 1H HRMAS MRS spectra, relative to the two first components PC1 and PC2 of the model; n = 24 (C6 n = 10, F98 n = 8, RG2 n = 6). b, c Score plots of the OPLS-DA model built with tumor data only n = 22 (C6 n = 8, F98 n = 8, RG2 n = 6) relative to the two first predictive components Pred1 and Pred2 of the model (b), and to the first predictive and the orthogonal components Tpred1 and Torth (c). d Corresponding VIP, black bars depicted metabolites with VIP ≥ 1. The  % of x variance explained by each component is reported in brackets in score plots

A first OPLS-DA was then performed on tumor data only, and 2 outliers were excluded resulting in 22 tumor observations. The final model generated with 2 significant OPLS predictive components and 1 orthogonal component has a cumulative R2Y of 0.68 and a cumulative Q2 of 0.50. X variance was explained by the predictive components for 39 % and by the orthogonal component for 36 %. The CV-ANOVA test is significant (p = 0.011).

The score scatter plot relative to the two first predictive components (Fig. 2b) shows a good separation between the 3 tumor types. Interestingly, the first predictive component separates F98 and RG2 observations (left) from C6 tumors (right). The score scatter plot relative to the first predictive component and the orthogonal component (Fig. 2c) clearly shows along the orthogonal axis the intra-class variance of F98 and C6 rats, while RG2 observations are well grouped. Nine metabolites had VIP values greater than 1 (Fig. 2d) but the exclusion of metabolites with VIP < 1 did not improve the models so all variables were retained. Gln, M-Ins, Gsh, tCr, GABA, Hyp, Tau, Ace, and Ala were the most important variables for separation between the 3 tumor models.

The loading plot including the 3 tumors was not easy to interpret so pair-wise OPLS-DA were performed. The main parameters of the models obtained are given in Supplementry Table 3. All models have high R2Y and Q2 values. For the F98-RG2 model, the main part of X variance (60 %) was explained by the two orthogonal components while only 16 % was explained by the predictive component. SUS-plots were then plotted to synthesize the loadings of the three OPLS-DA in a single figure (Fig. 3a, see Sect. 2 paragraph for the complete description of the SUS-plot). tCr was higher in C6, and on the other side Hyp, Ace, Ala, Tau and Gsh were lower in C6 than in F98 and RG2. M-Ins (Fig. 3a, green quadrants left) was lower and Gln higher in F98 than in C6. Since these 2 metabolites did not differ between C6 and RG2, one can deduce that these characteristics, low M-Ins and high Gln, are specific to F98 tumor type. On the other side, the level of Choline compounds, mainly GPC, were higher in RG2 than in C6, while the level of GABA was lower in RG2 than in C6, making those two characteristics specific to RG2 tumor (Fig. 3a, red quadrants). However these two metabolites had lower p(corr) values than M-Ins and Gln.
Fig. 3

a SUS-plot built from ex vivo data (left) and b in vivo data (right), corresponding to superimposition of pair-wise OPLS-DA models C6-F98 versus C6-RG2. Scores and loadings for C6 were always negative and positive for RG2/F98. In red: metabolites that have a unique evolution in RG2 relative to C6, in green: metabolites that have a unique evolution in F98 relative to C6; within diagonals: from bottom left to top right: metabolites that share a common evolution in F98/RG2 compared to C6, implying a specific evolution in C6. Metabolites with highest pcorr values (close to +1 or −1) are the most reliable, while metabolites located in the center have less weight in tumor model separation and are less reliable. c Mean amplitude of metabolites quantified in 1H HRMAS MRS spectra in tumor. Mean ± SD, *p < 0.05, Kruskal–Wallis ANOVA by ranks comparison between tumor models. d Score scatter plot of the ex vivo OPLS-DA model with prediction of new observations, in triangle, n = 8 (C6 n = 2, F98 n = 4, RG2 n = 2)

Group comparison of metabolite levels between tumor models using Kruskal–Wallis ANOVA by ranks partially confirmed these results (Fig. 3c). C6 tumor showed the greatest differences compared to the other tumor models: tCr was higher, and, in the group of metabolites that appeared lower in the SUS-plot (Hyp, Ace, Ala, Tau and Gsh) the same tendency was observed with univariate statistic but with only Gsh and Tau being statistically different from both F98 and RG2 (p = 0.056 for Tau between C6 and F98). F98 also showed a specific pattern as described in the SUS-plot with a lower M-ins level and a higher Gln level than in RG2 and C6 tumors (with p = 0.066 for this latter comparison). Regarding RG2 tumor, GABA level was only lower than that of the F98 tumor.

Interestingly, some metabolites commonly used for in vivo MRS-based diagnostic since they strongly vary compared with normal tissue, e.g. lactate and NAA, had few contribution to the ex vivo MRS-based tumor type discrimination. Indeed the level of these metabolites varied in the same manner between the 3 tumors, making them not useful for tumor type separation.

3.4.2 In vivo MR-based separation

Unsupervised PCA analysis was performed with all 48 spectra (24 rats: 24 tumors and 24 contralateral spectra) and the 10 variables as X matrix. A relative separation can be observed on the scatter plot of the PCA analysis between tumor and contralateral tissue (Fig. 4a). The first component separated tumor from contralateral samples, with 30 % of X variance explained. The second component explained 25 % of X variance but was not able to add better separation between tumor and contralateral striatum or to clearly separate tumor from each other.
Fig. 4

a Score scatter plot of the PCA analysis conducted with both contralateral and tumor in vivo 1H MRS spectra, related to the two first components PC1 and PC2 of the model; n = 24 (C6 n = 10, F98 n = 8, RG2 n = 6). b Score plot of the OPLS-DA model built with tumor data only, relative to the two first predictive components Tpred1 and Tpred2 of the model; n = 22 (C6 n = 8, F98 n = 8, RG2 n = 6). c Corresponding VIP, black bars depicted metabolites with VIP ≥ 1. The  % of x variance explained by each component is reported in brackets in score plots

An OPLS-DA analysis of tumor data only was then performed with the 22 tumor observations (same animals than in ex vivo analysis). The model obtained, built with 2 predictive components, had a moderately good cumulative quality (R2Y = 0.54) and a poor predictability (Q2 = 0.38). However, the CV-ANOVA was significant (p = 0.0042). In the score scatter plot represented in Fig. 4b, each tumor type shows a tendency to separate from the others with only few overlaps. Four metabolites had VIP values greater than 1: M-ins, tCr, Glu, and tCho, are the most important variables for separation between the 3 tumor models (Fig. 4c in black).

As for ex vivo analysis, three pair-wise OPLS-DA were performed for a better identification of the most discriminant metabolites and SUS-plot were plotted (Fig. 3b). The resulting model parameters are summarized in Supplementry Table 3. C6-RG2 and C6-F98 models have high R2Y and Q2 values while the F98-RG2 model is not reliable. In accordance with ex vivo SUS-plot, but with less reliability, a higher level of Choline compounds in RG2 compared to C6 was detected, and, to a less extent, a lower level of M-ins specific to F98, and a higher level of tCr associated with the lower level of Tau in C6 relative to F98 and RG2.

3.4.3 Specific metabolic profile of each tumor type

In line with the study of Cuperlovic-Culf et al. (2012) which was focused on several human tumor cell lines, we observed differences in metabolic signature between the three rodent tumor models. C6 tumor was characterized by lower levels of Gsh, Ace, Ala, Hyp and Tau and higher level of tCr compared to RG2 and F98. tCr is known to decrease with necrosis (Bulik et al. 2013). Opstadt et al. (2009) suggested that Tau could be an apoptotic biomarker independent of tumor necrotic status. Note that our results for tCr and Tau are in line with the in vivo study of Doblas et al. (2012). Ala, similarly to Lac, can also increase along with hypoxia (Griffin and Shockcor 2004). According to that, in our conditions, C6 tumor should then be less necrotic, less hypoxic and less apoptotic that the 2 other tumors. Paradoxically C6 tumors are usually considered more necrotic than the 2 others, but few sign of necrosis were depicted in T2-weighted anatomical images in our study. Given that necrosis increased during the progression of tumor, we suppose that our study was performed with C6 tumor having not reached their full necrotic potential.

F98 tumor was characterized by a lower level of M-ins and a higher level of Gln relative to C6 and RG2. In this study, the MRS methods used did not provide any dynamic parameter but only global static metabolic profiles. Accordingly, higher Gln consumption as well as lower Gln production might explain the decrease of this metabolite observed in the three tumors compared to contralateral tissue. Indeed, it has been recently reviewed that tumor cell metabolism is characterized by a high degree of glutamine consumption (Dang 2010). On the other hand, M-Ins, a major organic cell osmolyte (Yancey 2005), was lower in F98 compared to C6 and RG2. We could then suggest that, in accordance with anatomical observation based on T2-weighted images, F98 tumor were more necrotic than C6 and RG2, and presented an apparent lower glutamine consumption, and lower M-ins level due to a lower cell number.

RG2 tumor had a specific signature of increased level of GPC. GPC increase in RG2 might be related to the increased cell membrane turnover (Glunde et al. 2011), in good accordance with the high aggressive invasion characteristics of this model (Barth and Kaur 2009). For instance, tumor volume doubling time in orthotopic models is reported as 55 h for F98 (data from our laboratory), 51 h for RG2 and 87 h for C6 (Valable et al. 2008), which is also well related to the overall animal survival (Doblas et al. 2012).

3.5 Basal brain metabolism in contralateral striatum and its interaction with tumor

Differences in brain metabolism and functions between rat strains have previously been reported both in non-pathological (Clemens et al. 2014; Hong et al. 2011) and in pathological conditions (Golden et al. 1995; Herz et al. 1996). Given that RG2 and F98 were raised in Fischer rats and C6 in Wistar rats, the host tissue might participate to the separation between RG2/F98 and C6 tumor. In order to investigate this hypothesis, OPLS-DA were conducted with contralateral data only (ex vivo data), and two groups as Y classes: one group for Fischer rats with RG2 and F98 observations, and one group for Wistar rats with C6 observations. The resulting OPLS-DA (Fig. 5a), built with one predictive and three orthogonal components, was very robust (R2Y = 0.96, Q2 = 0.87), and significant (CV-ANOVA, p = 0.0003) and depict a clear distinction between Wistar and Fischer contralateral striatum. Among the 4 metabolites having VIP value >1 (Glu, NAA, Gly and Ace), only Ace has VIP value >1 in tumor separation (Fig. 2), suggesting that the tumor separation cannot be explained by the host metabolism, i.e. the genotype of the host rat strain. The first two metabolites explaining the clustering (highest VIP, not shown) were NAA higher in Wistar rat, and Glu higher in Fisher rats. Kruskal–Wallis ANOVA confirmed this observation with these two metabolites being significantly different when comparing C6 and RG2/F98 contralateral striata (Fig. 5d).
Fig. 5

a, b and c multivariate statistical models built with contralateral 1H HRMAS MRS data. a Score plot of the rat strain-based OPLS-DA model built with ex vivo contralateral only data (n = 22, Wistar rat strain: C6 n = 8, and Fischer rat strain: F98 n = 8, RG2 n = 6), relative to the first predictive components Tpred1 and the first orthogonal components Torth1. b Score plot of tumor type-based OPLS-DA model built with the same ex vivo contralateral data as in (a) but with three groups as Y matrix (C6, F98 and RG2), relative to the two first predictive components Tpred1 and Tpred2. c Corresponding VIP of (b). In black metabolites with VIP ≥ 1 in the OPLS-DA model built with tumor data. d Mean amplitude of metabolites quantified from 1H HRMAS MRS spectra of contralateral striatum. Mean ± SD, *p < 0.05, Kruskal–Wallis ANOVA by ranks comparison between tumor models. The  % of x variance explained by each component is reported in brackets in score plots

However, we cannot exclude that contralateral metabolic profiles are impacted by the presence of tumor. In order to analyse this effect, OPLS-DA were conducted on the same data matrix but with the three tumors as Y variables: contra C6, contra F98 and contra RG2. The statistical model was generated with 2 significant predictive components and 4 orthogonal components (Supplementry Table 3). The model has a good cumulative quality (R2Y = 0.88), a good predictability (Q2 = 0.65), and is significant (CV-ANOVA, p = 0.023). The score scatter plot shows that the first predictive component separates contralateral C6 striatum from the two other contralateral striata (Fig. 5b) that is similar to the separation of C6 tumor from the F98-RG2 (Fig. 2b). Although this suggests a strain effect that might be superimposed to the tumor effect, only 19 % of X variance is explained by the predictive components, i.e. tumor type, and 69 % is explained by the orthogonal components (variance that is unrelated to the tumor type).

Within metabolites with VIP ≥ 1 (Fig. 5c), 3 metabolites: Ace, Gsh and tCr, have also VIP ≥ 1 in the tumor OPLS-DA model (in black Fig. 5c). This suggests that the presence of tumor might have an impact on contralateral metabolism. Some causes can be suggested: (i) discrete migration of tumor cells in contralateral tissue, (ii) metabolism modifications of healthy cells in contralateral tissue due to tumor presence.

Further studies should be performed to compare striatum of healthy Fisher and Wistar rats. Furthermore, immune depressed animals could be a good alternative in order to avoid the use of different rat strains. However this would have a strong impact on tumor/host interaction since the immune cells are of high importance for the validity of brain tumor model (Huszthy et al. 2012). This immune effect also explains why injecting C6 cell line in other rat strains than Wistar (Grobben et al. 2002) would result in tumor -if properly growth- that are less reliable with regard to human glioblastoma (Grobben et al. 2002).

3.6 Advantages of MRS-based multivariate statistic

Both in vivo and ex vivo data exhibited variability which can be explained by the sensitivity of the technic, and/or the heterogeneity of the tumors, within each tumor core, between animals for each tumor type and between tumors types. On the one hand, in vivo MRS method suffers from low frequency resolution and low Signal to Noise Ratio (SNR), which has a negative impact on inter-individual variability. On the other hand, intra-tumor variability is limited in vivo due to the voxel size that includes nearly the whole tumor. However, these limitations might be overcome by increasing animal number in preclinical studies. Concerning ex vivo 1H HRMAS MRS, which produced highly resolved spectra with high SNR, the sampling step necessarily introduced intra-tumor variability, even if the center of the tumor was always selected.

Among this data variance, we aimed to extract systemic variance that was only related to the tumor type, i.e. C6, F98 and RG2. For this purpose, the resulting quantified data were submitted to the same multivariate statistical protocol, i.e. supervised OPLS-DA and SUS-plot. With these statistical tools, all metabolites are taken into account for the statistical modeling of tumor properties, making the tumor description more complete. This highlights metabolites that are the most important for separation between groups. SUS-plot gives a direct visualization of shared and unique metabolic variations between tumor types. This is particularly well suited for a 3-groups comparison, while it is more difficult to extract a simple and global interpretation from classical univariate ANOVA.

Variance that is not related to tumor type was extracted from our data with orthogonal filtering. Interestingly, we observed that only RG2 observations were well grouped in the tumor OPLS-DA model while the other were scattered along the orthogonal component (see Fig. 2c). In line with previous microvascular analysis (Valable et al. 2008), this indicates that RG2 tumors were less heterogeneous than C6 and F98 tumors.

Despite this high data variability, the resulting OPLS-DA tumor model was robust enough to allow the proper classification of 8 new observations (C6 n = 2, F98 n = 4, RG2 n = 2). Based on clinical information in association with training data, one can then expect that OPLS-DA could ameliorate interpretation of MRS-based information in pathologies. Such improvement was previously reported by Erb et al. (2008) who showed that a similar PLS-DA approach applied to 1H HRMAS MRS spectra of biopsies was able to improve the grading of malignancy in oligodendrogliomas.

OPLS-DA was also sensitive enough to separate contralateral metabolic profiles whereas few differences were detected with ANOVA. OPLS-DA seems to be suitable for the detection of these subtle variations in normal brain that could interfere with the description of pathologies.

All our results show that OPLS-DA, a powerful method for classification and prediction purposes, could also be used to sort the data in order to focus on the most relevant information contained in data variance. In the future, this strategy for data analysis could therefore be very useful for the discovery of new biomarkers.

4 Concluding remarks

In this study, we investigated the metabolic profiles of the tumor and the corresponding contralateral striata in three glioma models C6, F98 and RG2 in rats. We used 1H MRS, 1H HRMAS MRS and chemometric tools to extract the most discriminant information in this three-group comparison.

With the help of OPLS-DA, the amount of information provided by in vivo 1H MRS and more robustly by ex vivo 1H HRMAS MRS was able to highlight metabolic differences that were not detected with classical statistical analysis. F98 was characterized by higher level of glutamine and lower level of myo-inositol, while RG2 tumor had higher level of phosphocholine and glycerophosphocholine. C6 had higher level of total creatine and lower level of taurine, hypotaurine, alanine, acetate and glutathione.

OPLS-DA can be very useful both for metabolic-based classification and prediction and to sort the data in order to extract from variance all the relevant information. Additionally, other MR-based features, such as microvascularization parameters (Barbier et al. 2001; Christen et al. 2014; He and Yablonskiy 2007; Tofts et al. 1999; Troprès et al. 2001), could be added to the MR-examination and subsequently yield additional information that would further improve the OPLS-DA-based metabolic classification.

Notes

Acknowledgments

We thank the animal care facility of GIN.

Funding

This study was funded by the French Service de Santé des Armées, the Fondation ARC (“Association pour la Recherche sur le Cancer”) and ANR (“Agence Nationale pour la Recherche”) Imoxy grant. IRMaGe was partly funded by the French program “Investissement d’Avenir” run by the “Agence Nationale pour la Recherche”; grant “Infrastructure d’avenir en Biologie Santé”—ANR-11-INBS-0006.

Compliance with ethical standards

Conflict of interest

All authors declare that they have no conflict of interest.

Ethical approval

The study design was approved by the local ethical committee for animal care and use (C2EA-04: “Comité d’éthique en expérimentation animale GIN”). Experiments were performed under permits (No. 38 12 63 and B 38 516 10 008 for experimental and animal care facilities) from the French Ministry of Agriculture.

Supplementary material

11306_2015_835_MOESM1_ESM.tif (1.4 mb)
Supplementary material 1 (TIFF 1409 kb)Supplementary Table 1. Metabolite assignment
11306_2015_835_MOESM2_ESM.tif (2.4 mb)
Supplementary material 2 (TIFF 2445 kb)Supplementary Table 2. Details of statistical values presented in Table 1
11306_2015_835_MOESM3_ESM.tif (996 kb)
Supplementary material 3 (TIFF 996 kb)Supplementary Table 3. OPLS-DA models built with quantified 1H MRS data (in vivo) or 1H HRMAS MRS data (ex vivo) as x variables and tumor types as classes (C6, F98 and RG2). Cumulative values of R2X, R2Y and Q2 are given

References

  1. Bansal, A., Shuyan, W., Hara, T., Harris, R. A., & DeGrado, T. R. (2008). Biodisposition and metabolism of [18F] fluorocholine in 9L glioma cells and 9L glioma-bearing fisher rats. European Journal of Nuclear Medicine and Molecular Imaging, 35(6), 1192–1203.CrossRefPubMedPubMedCentralGoogle Scholar
  2. Barbier, E. L., Lamalle, L., & Décorps, M. (2001). Methodology of brain perfusion imaging. Journal of Magnetic Resonance Imaging, 13(4), 496–520.CrossRefPubMedGoogle Scholar
  3. Barth, R. F., & Kaur, B. (2009). Rat brain tumor models in experimental neuro-oncology: the C6, 9L, T9, RG2, F98, BT4C, RT-2 and CNS-1 gliomas. Journal of Neuro-oncology, 94(3), 299–312.CrossRefPubMedPubMedCentralGoogle Scholar
  4. Bottomley, P. A. (1987). Spatial localization in NMR spectroscopy in vivo. Annals of the New York Academy of Sciences, 508(1), 333–348.CrossRefPubMedGoogle Scholar
  5. Bulik, M., Jancalek, R., Vanicek, J., Skoch, A., & Mechl, M. (2013). Potential of MR spectroscopy for assessment of glioma grading. Clinical Neurology and Neurosurgery, 115(2), 146–153.CrossRefPubMedGoogle Scholar
  6. Christen, T., Bouzat, P., Pannetier, N., Coquery, N., Moisan, A., Lemasson, B., et al. (2014). Tissue oxygen saturation mapping with magnetic resonance imaging. Journal of Cerebral Blood Flow and Metabolism, 34(9), 1550–1557.CrossRefPubMedPubMedCentralGoogle Scholar
  7. Clemens, L. E., Jansson, E. K. H., Portal, E., Riess, O., & Nguyen, H. P. (2014). A behavioral comparison of the common laboratory rat strains Lister Hooded, Lewis, Fischer 344 and Wistar in an automated homecage system. Genes, Brain and Behavior, 13(3), 305–321.CrossRefGoogle Scholar
  8. Coquery, N., Francois, O., Lemasson, B., Debacker, C., Farion, R., Rémy, C., & Barbier, E. L. (2014). Microvascular MRI and unsupervised clustering yields histology-resembling images in two rat models of glioma. Journal of Cerebral Blood Flow and Metabolism, 34(8), 1354–1362.CrossRefPubMedPubMedCentralGoogle Scholar
  9. Coquery, N., Pannetier, N., Farion, R., Herbette, A., Azurmendi, L., Clarencon, D., et al. (2012). Distribution and radiosensitizing effect of cholesterol-coupled dbait molecule in rat model of glioblastoma. PLoS One, 7(7), e40567.CrossRefPubMedPubMedCentralGoogle Scholar
  10. Cuperlovic-Culf, M., Ferguson, D., Culf, A., Morin, P., & Touaibia, M. (2012). 1H NMR metabolomics analysis of glioblastoma subtypes correlation between metabolomics and gene expression characteristics. Journal of Biological Chemistry, 287(24), 20164–20175.CrossRefPubMedPubMedCentralGoogle Scholar
  11. Dang, C. V. (2010). Rethinking the warburg effect with Myc micromanaging glutamine metabolism. Cancer Research, 70(3), 859–862.CrossRefPubMedPubMedCentralGoogle Scholar
  12. de Graaf, R. A. (2007). Front matter. In vivo NMR spectroscopy (pp. i–xxi). John Wiley & Sons, Ltd.Google Scholar
  13. Doblas, S., He, T., Saunders, D., Hoyle, J., Smith, N., Pye, Q., et al. (2012). In vivo characterization of several rodent glioma models by 1H MRS. NMR in Biomedicine, 25(4), 685–694.CrossRefPubMedGoogle Scholar
  14. Doblas, S., He, T., Saunders, D., Pearson, J., Hoyle, J., Smith, N., et al. (2010). Glioma morphology and tumor-induced vascular alterations revealed in seven rodent glioma models by in vivo magnetic resonance imaging and angiography. Journal of Magnetic Resonance Imaging, 32(2), 267–275.CrossRefPubMedPubMedCentralGoogle Scholar
  15. Erb, G., Elbayed, K., Piotto, M., Raya, J., Neuville, A., Mohr, M., et al. (2008). Toward improved grading of malignancy in oligodendrogliomas using metabolomics. Magnetic Resonance in Medicine, 59(5), 959–965.CrossRefPubMedGoogle Scholar
  16. Fauvelle, F., Carpentier, P., Dorandeu, F., Foquin, A., & Testylier, G. (2012). Prediction of neuroprotective treatment efficiency using a HRMAS NMR-Based statistical model of refractory status epilepticus on mouse: A metabolomic approach supported by histology. Journal of Proteome Research, 11(7), 3782–3795.CrossRefPubMedGoogle Scholar
  17. Glunde, K., Bhujwalla, Z. M., & Ronen, S. M. (2011). Choline metabolism in malignant transformation. Nature Reviews Cancer, 11(12), 835–848.PubMedPubMedCentralGoogle Scholar
  18. Golden, G. T., Smith, G. G., Ferraro, T. N., & Reyes, P. F. (1995). Rat strain and age differences in kainic acid induced seizures. Epilepsy Research, 20(2), 151–159.CrossRefPubMedGoogle Scholar
  19. Govindaraju, V., Young, K., & Maudsley, A. A. (2000). Proton NMR chemical shifts and coupling constants for brain metabolites. NMR in Biomedicine, 13(3), 129–153.CrossRefPubMedGoogle Scholar
  20. Griffin, J. L., & Shockcor, J. P. (2004). Metabolic profiles of cancer cells. Nature Reviews Cancer, 4(7), 551–561.CrossRefPubMedGoogle Scholar
  21. Grobben, B., Deyn, P. D., & Slegers, H. (2002). Rat C6 glioma as experimental model system for the study of glioblastoma growth and invasion. Cell and Tissue Research, 310(3), 257–270.CrossRefPubMedGoogle Scholar
  22. He, X., & Yablonskiy, D. A. (2007). Quantitative BOLD: Mapping of human cerebral deoxygenated blood volume and oxygen extraction fraction: Default state. Magnetic Resonance in Medicine, 57(1), 115–126.CrossRefPubMedPubMedCentralGoogle Scholar
  23. Herz, R. C. G., Gaillard, P. J., de Wildt, D. J., & Versteeg, D. H. G. (1996). Differences in striatal extracellular amino acid concentrations between wistar and fischer 344 rats after middle cerebral artery occlusion. Brain Research, 715(1–2), 163–171.CrossRefPubMedGoogle Scholar
  24. Holmes, E., Tsang, T. M., & Tabrizi, S. J. (2006). The application of NMR-based metabonomics in neurological disorders. NeuroRx, 3(3), 358–372.CrossRefPubMedPubMedCentralGoogle Scholar
  25. Hong, S.-T., Balla, D. Z., Choi, C., & Pohmann, R. (2011). Rat strain-dependent variations in brain metabolites detected by in vivo 1H NMR spectroscopy at 16.4T. NMR in Biomedicine, 24(10), 1401–1407.CrossRefPubMedGoogle Scholar
  26. Huszthy, P. C., Daphu, I., Niclou, S. P., Stieber, D., Nigro, J. M., Sakariassen, P. O., et al. (2012). In vivo models of primary brain tumors: pitfalls and perspectives. Neuro-Oncology, 14(8), 979–993.CrossRefPubMedPubMedCentralGoogle Scholar
  27. Kanayama, S., Kuhara, S., & Satoh, K. (1996). In vivo rapid magnetic field measurement and shimming using single scan differential phase mapping. Magnetic Resonance in Medicine: Official Journal of the Society of Magnetic Resonance in Medicine/Society of Magnetic Resonance in Medicine, 36(4), 637–642.CrossRefGoogle Scholar
  28. Kauppinen, R. A., & Peet, A. C. (2011). Using magnetic resonance imaging and spectroscopy in cancer diagnostics and monitoring. Cancer Biology & Therapy, 12(8), 665–679.CrossRefGoogle Scholar
  29. Lemasson, B., Valable, S., Farion, R., Krainik, A., Rémy, C., & Barbier, E. L. (2013). In vivo imaging of vessel diameter, size, and density: A comparative study between MRI and histology. Magnetic Resonance in Medicine, 69(1), 18–26.CrossRefPubMedGoogle Scholar
  30. Opstad, K. S., Bell, B. A., Griffiths, J. R., & Howe, F. A. (2009). Taurine: A potential marker of apoptosis in gliomas. British Journal of Cancer, 100(5), 789–794.CrossRefPubMedPubMedCentralGoogle Scholar
  31. Opstad, K. S., Wright, A. J., Bell, B. A., Griffiths, J. R., & Howe, F. A. (2010). Correlations between in vivo 1H MRS and ex vivo 1H HRMAS metabolite measurements in adult human gliomas. Journal of Magnetic Resonance Imaging, 31(2), 289–297.CrossRefPubMedGoogle Scholar
  32. Piotto, M., Moussallieh, F.-M., Dillmann, B., Imperiale, A., Neuville, A., Brigand, C., et al. (2009). Metabolic characterization of primary human colorectal cancers using high resolution magic angle spinning 1H magnetic resonance spectroscopy. Metabolomics, 5(3), 292–301.CrossRefGoogle Scholar
  33. Rabeson, H., Fauvelle, F., Testylier, G., Foquin, A., Carpentier, P., Dorandeu, F., et al. (2008). Quantitation with QUEST of brain HRMAS-NMR signals: Application to metabolic disorders in experimental epileptic seizures. Magnetic Resonance in Medicine, 59(6), 1266–1273.CrossRefPubMedGoogle Scholar
  34. Ratiney, H., Sdika, M., Coenradie, Y., Cavassila, S., van Ormondt, D., & Graveron-Demilly, D. (2005). Time-domain semi-parametric estimation based on a metabolite basis set. NMR in Biomedicine, 18(1), 1–13.CrossRefPubMedGoogle Scholar
  35. Stupp, R., Mason, W. P., van den Bent, M. J., Weller, M., Fisher, B., Taphoorn, M. J. B., et al. (2005). Radiotherapy plus concomitant and adjuvant temozolomide for glioblastoma. The New England Journal of Medicine, 352(10), 987–996.CrossRefPubMedGoogle Scholar
  36. Tkáč, I., Starčuk, Z., Choi, I.-Y., & Gruetter, R. (1999). In vivo 1H NMR spectroscopy of rat brain at 1 ms echo time. Magnetic Resonance in Medicine, 41(4), 649–656.CrossRefPubMedGoogle Scholar
  37. Tofts, P. S., Brix, G., Buckley, D. L., Evelhoch, J. L., Henderson, E., Knopp, M. V., et al. (1999). Estimating kinetic parameters from dynamic contrast-enhanced t1-weighted MRI of a diffusible tracer: Standardized quantities and symbols. Journal of Magnetic Resonance Imaging, 10(3), 223–232.CrossRefPubMedGoogle Scholar
  38. Troprès, I., Grimault, S., Vaeth, A., Grillon, E., Julien, C., Payen, J.-F., et al. (2001). Vessel size imaging. Magnetic Resonance in Medicine, 45(3), 397–408.CrossRefPubMedGoogle Scholar
  39. Valable, S., Lemasson, B., Farion, R., Beaumont, M., Segebarth, C., Remy, C., & Barbier, E. L. (2008). Assessment of blood volume, vessel size, and the expression of angiogenic factors in two rat glioma models: a longitudinal in vivo and ex vivo study. NMR in Biomedicine, 21(10), 1043–1056.CrossRefPubMedGoogle Scholar
  40. Verhaak, R. G. W., Hoadley, K. A., Purdom, E., Wang, V., Qi, Y., Wilkerson, M. D., et al. (2010). An integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR and NF1. Cancer Cell, 17(1), 98.CrossRefPubMedPubMedCentralGoogle Scholar
  41. Weller, M., Stupp, R., Hegi, M., & Wick, W. (2012). Individualized targeted therapy for glioblastoma. The Cancer Journal, 18(1), 40–44.CrossRefPubMedGoogle Scholar
  42. Wieruszeski, J.-M., Montagne, G., Chessari, G., Rousselot-Pailley, P., & Lippens, G. (2001). Rotor synchronization of radiofrequency and gradient pulses in high-resolution magic angle spinning NMR. Journal of Magnetic Resonance, 152(1), 95–102.CrossRefPubMedGoogle Scholar
  43. Wiklund, S., Johansson, E., Sjöström, L., Mellerowicz, E. J., Edlund, U., Shockcor, J. P., et al. (2008). Visualization of GC/TOF-MS-Based metabolomics data for identification of biochemically interesting compounds using OPLS class models. Analytical Chemistry, 80(1), 115–122.CrossRefPubMedGoogle Scholar
  44. Wilson, M., Davies, N. P., Grundy, R. G., & Peet, A. C. (2009). A quantitative comparison of metabolite signals as detected by in vivo MRS with ex vivo1H HR-MAS for childhood brain tumours. NMR in Biomedicine, 22(2), 213–219.CrossRefPubMedGoogle Scholar
  45. Wold, S., Ruhe, A., Wold, H., & Dunn, W. J, I. I. I. (1984). The collinearity problem in linear regression. The partial least squares (PLS) approach to generalized inverses. SIAM Journal on Scientific and Statistical Computing, 5(3), 735–743.CrossRefGoogle Scholar
  46. Wright, A. J., Fellows, G. A., Griffiths, J. R., Wilson, M., Bell, B., & Howe, F. A. (2010). Ex-vivo HRMAS of adult brain tumours: metabolite quantification and assignment of tumour biomarkers. Molecular Cancer, 9(1), 66.CrossRefPubMedPubMedCentralGoogle Scholar
  47. Yancey, P. H. (2005). Organic osmolytes as compatible, metabolic and counteracting cytoprotectants in high osmolarity and other stresses. Journal of Experimental Biology, 208(15), 2819–2830.CrossRefPubMedGoogle Scholar
  48. Ziegler, A., von Kienlin, M., Décorps, M., & Rémy, C. (2001). High glycolytic activity in rat glioma demonstrated in vivo by correlation peak 1H magnetic resonance imaging. Cancer Research, 61(14), 5595–5600.PubMedGoogle Scholar

Copyright information

© Springer Science+Business Media New York 2015

Authors and Affiliations

  • Nicolas Coquery
    • 1
    • 2
  • Vasile Stupar
    • 3
    • 4
    • 5
  • Régine Farion
    • 3
    • 4
    • 5
  • Severine Maunoir-Regimbal
    • 6
  • Emmanuel L. Barbier
    • 1
    • 2
  • Chantal Rémy
    • 1
    • 2
  • Florence Fauvelle
    • 3
    • 4
    • 5
    • 6
    • 7
  1. 1.Inserm, U836GrenobleFrance
  2. 2.Univ Grenoble Alpes, Grenoble Institut des NeurosciencesGrenobleFrance
  3. 3.Univ Grenoble Alpes, IRMaGeGrenobleFrance
  4. 4.CNRS, UMS 3552GrenobleFrance
  5. 5.Inserm, US17GrenobleFrance
  6. 6.IRBABrétigny Sur OrgeFrance
  7. 7.U836 – Grenoble Institut des Neurosciences, Chemin Fortuné Ferrini, BP 217 – CHU GrenobleGrenoble CedexFrance

Personalised recommendations