PI3K/mTOR inhibition of IDH1 mutant glioma leads to reduced 2HG production that is associated with increased survival

70–90% of low-grade gliomas and secondary glioblastomas are characterized by mutations in isocitrate dehydrogenase 1 (IDHmut). IDHmut produces the oncometabolite 2-hydroxyglutarate (2HG), which drives tumorigenesis in these tumors. The phosphoinositide-3-kinase (PI3K)/mammalian target of rapamycin (mTOR) pathway represents an attractive therapeutic target for IDHmut gliomas, but noninvasive indicators of drug target modulation are lacking. The goal of this study was therefore to identify magnetic resonance spectroscopy (MRS)-detectable metabolic biomarkers associated with IDHmut glioma response to the dual PI3K/(mTOR) inhibitor XL765. 1H-MRS of two cell lines genetically modified to express IDHmut showed that XL765 induced a significant reduction in several intracellular metabolites including 2HG. Importantly, examination of an orthotopic IDHmut tumor model showed that enhanced animal survival following XL765 treatment was associated with a significant in vivo 1H-MRS detectable reduction in 2HG but not with significant inhibition in tumor growth. Further validation is required, but our results indicate that 2HG could serve as a potential noninvasive MRS-detectable metabolic biomarker of IDHmut glioma response to PI3K/mTOR inhibition.

Gliomas are the most common type of brain tumor, representing 80% of all diagnosed malignant central nervous system tumors in the United States 1 . According to the World Health Organization, they are classified into low-grade gliomas (LGG; astrocytoma and oligodendroglioma) and the high-grade glioblastoma (GBM) 2,3 . GBM always presents as Grade IV and is the most aggressive form of glioma with an average survival of 15 months. Astrocytoma and oligodendroglioma tumors typically present as Grades II or III, and in the case of astrocytoma can also upgrade to Grade IV secondary GBM. They are characterized by a slower growth rate and a longer average survival of ~10 years [4][5][6] . Astrocytoma and oligodendroglioma tumors also affect patients at a younger age 1 . Traditionally this classification was based on histopathological analysis. However, recent understanding of the molecular genetic events that drive gliomagenesis 2,7 led in 2016 to the WHO's updated classification of gliomas 3 . In this context, the main characteristic of ~70-90% LGGs and secondary GBM is the presence of a mutation in the cytosolic form of isocitrate dehydrogenase 1 (IDHmut) 8 . In contrast, primary GBMs are now defined as wild-type for isocitrate dehydrogenase 1 (IDH1).
Wild type IDH1 catalyzes the oxidative decarboxylation of isocitrate to α-ketoglutarate (α-KG). Mutation occurs as a single amino acid substitution in the active site of IDH 9,10 , and the IDHmut enzyme catalyzes the reduction of α-KG to 2-hydroxyglutarate (2HG) 11,12 . 2HG has been shown to inhibit the activity of α-KG-dependent dioxygenases such as histone demethylases, prolyl hydroxylases and TET family of 5-methylcytosine hydroxylases 13 . This inhibition leads to alterations in cell signaling and gene expression that drive tumorigenesis 14,15 . As such, IDHmut is considered a 'driver' mutation in the development of LGGs, and 2HG is considered an Results XL765 treatment inhibits 4E-BP1 and S6K phosphorylation downstream of PI3K/mTOR signaling in low-grade gliomas. We investigated immortalized normal human astrocytes (NHA) and U87 cells genetically engineered to express IDHmut (NHAIDHmut and U87IDHmut) as described previously 49 . First, we examined the effect of inhibiting PI3K/mTOR signaling using the dual PI3K/mTOR inhibitor XL765 (see Fig. 1A) on cell proliferation. Each cell line was treated with a dose that resulted in approximately 50% reduction in cell number. NHAIDHmut cells were treated with 32 μM XL765 for 72 h leading to 55.7 ± 16% inhibition while U87IDHmut cells were treated with 12 μM XL765 for 24 h, resulting in 61.0 ± 10.6% inhibition (Fig. 1B). We confirmed inhibition of the PI3K/mTOR pathway following XL765 treatment by examining the phosphorylation status of 4E-BP1 (p4E-BP1) and S6K (pS6K) downstream of PI3K/mTOR signaling (see Fig. 1A). XL765 treatment induced a significant reduction in p4E-BP1 and pS6K levels in both NHAIDHmut and U87IDHmut cells (Fig. 1C-F and Supplementary Figs 1 and 2). Importantly, XL765 treatment did not alter total levels of 4E-BP1 and S6K. Taken together, these results confirmed that XL765 inhibited PI3K/mTOR signaling in our low-grade glioma models.

XL765 treatment alters 1 H MRS-detectable metabolites levels in low-grade gliomas.
We then examined the effect of XL765 on steady-state metabolite levels using 1 H MRS in the NHAIDHmut ( Fig. 2A) and U87IDHmut (Fig. 3A) models. Multivariate analysis was used to compare the XL765-treated group and the control group. First, principle component analysis (PCA) was applied for an unbiased reduction of the complexity of the data. In both cell models PCA was able to reliably distinguish the XL765-treated group from the control group using only the first two principle components t [1] and t [2] (see Fig. 2B: NHAIDHmut, R2X (fraction of variation of X variables explained by model) = 0.407, Q2 (goodness of prediction) = 0.0484; N = 6 for control and N = 7 for XL765; Fig. 3B: U87IDHmut, R2X = 0.515, Q2 = −0.0102, N = 6 for control and N = 7 for XL765). Next, the data was analyzed using a supervised orthogonal partial least squares discriminant analysis (OPLS-DA) (Supplementary Fig. 3; NHAIDHmut R2Y (fraction of variation of Y variables explained by Complete blots can be seen in Supplementary Fig. 1. (D 3C). The correlation coefficient (|r|) of the metabolites responsible for the discrimination between the two groups is illustrated by the color of the points on the S-plot while the relevance to the model (loadings) is indicated by the signal amplitude. Based on this information, we focused on metabolites with |r| higher than 0.8. The cellular concentrations of these metabolites were then subjected to univariate analysis and our findings are summarized in Table 1 (NHAIDHmut) and 2 (U87IDHmut). As shown in the S-plots in Figs 2C and 3C as well as Tables 1 and 2, several metabolites were commonly altered in both cell lines. Importantly, levels of alanine, 2HG, glutamate, glutamine and aspartate were significantly reduced in XL765-treated cells in both of our models, thus identifying potentially translational metabolic biomarkers of IDHmut glioma cell response to PI3K/mTOR inhibition. XL765 inhibits PI3K/mTOR signaling and prolongs animal survival in orthotopic tumor xenografts in vivo. Next, we assessed the impact of XL765 in vivo on U87IDHmut orthotopic tumor xenografts.
Consistent with our findings in our cell models, XL765 reduced levels of p4E-BP1 without affecting total 4E-BP1 levels (Fig. 4A,B and Supplementary Fig. 4). Immunohistochemical analysis pointed to a significant reduction in Ki-67 staining in XL765-treated tumors (Fig. 4C,D). With regard to tumor growth, examination of T2-weighted MR images from control and XL765-treated animals (Fig. 5A) indicated that, although there was a significant reduction in tumor volume at day 8 and day 12, there was no significant difference at the majority of time-points (Fig. 5B). Importantly, XL765 treatment significantly prolonged animal survival as illustrated in the Kaplan-Meier survival plot ( Fig. 5C; N = 11; χ = 8.555 and p = 0.003). Taken together, these results indicated that XL765 inhibited PI3K/mTOR signaling in vivo and pointed to a cytostatic effect of XL765 on tumor proliferation, consistent with previous studies indicating that PI3K/mTOR inhibition leads to tumor stasis 48 .

XL765 treatment reduces 1 H MRS-detectable 2HG levels in vivo.
To assess whether XL765 treatment in vivo induced metabolic alterations similar to those observed in our cell studies, we acquired in vivo 1 H spectra from control and XL765-treated tumors. Figure 5D shows a representative 1 H spectrum from a control animal along with the spectral fit and residual signal from LCModel. We then quantified metabolites for which    Table 1. Quantitative comparison of metabolites quantified from the aqueous phase of control and XL765treated NHAIDHmut cell extract that presented correlation coefficient |r| higher than 0.8 in the S-plot (Fig. 2C). Metabolites in bold mark metabolites that are common in the two cell lines.  Table 2. Quantitative comparison of metabolites quantified from the aqueous phase of control and XL765treated U87IDHmut cell extract that presented correlation coefficient |r| higher than 0.8 in the S-plot (Fig. 3C). Metabolites in bold mark metabolites that are common in the two cell lines. www.nature.com/scientificreports www.nature.com/scientificreports/ Cramer-Rao lower bounds (CRLB) were <40% (Table 3). Consistent with our cell studies, there was a significant drop in 2HG levels from 0.132 ± 0.010 a.u. in controls to 0.103 ± 0.008 a.u. in XL765-treated animal (N = 5 for control group, N = 6 for XL765-treated group, CRLB < 40%), as well as a significant drop in glutamate (from 0.127 ± 0.006 a.u. in controls to 0.101 ± 0.004 a.u. in XL765-treated animal; N = 9 for control group, N = 10 for XL765-treated group, CRLB < 25%). However, in contrast to our cell findings, there was a significant increase in glutamine levels while there was no change in alanine. Aspartate could not be quantified with any reasonable accuracy (CRLB > 40%) in more than 3 animals in XL765-treated tumors.
Since the CRLBs used in the analysis of our in vivo MRS data were relatively high, and because there was some discrepancy between our cell and tumor data, we performed higher resolution ex vivo tumor extract studies to confirm our in vivo observations. 1 H MRS data was recorded from control and XL765-treated tumor tissue extracts (Fig. 6A), and similar to our cell studies, unbiased multivariate analysis was then used to compare the XL765-treated group and the control group. PCA was able to reliably distinguish the XL765-treated   www.nature.com/scientificreports www.nature.com/scientificreports/ group from the control group using the first three principle components t [1], t [2] and t [3] (Fig. 6B: R2X = 0.395, Q2 = 0.0208; N = 10 for control and 7 for XL765). Next, the data was analyzed using a supervised OPLS-DA analysis (Supplementary Fig. 3; R2Y = 0.929, Q2 = 0.241) and an S-plot derived from this classification (Fig. 6C). The only region with a correlation coefficient (|r|) higher than 0.8 was the 2.2-2.3 ppm which corresponds to 2HG. Univariate analysis of the 2HG concentration in the two groups confirmed a significant drop (p = 0.003) from 0.089 ± 0.045 a.u. to 0.027 ± 0.017 a.u. following XL765 treatment. Univariate analysis of the glutamate concentration in the two groups agreed with the multivariate S-plot analysis as there was no significant change (p = 0.32; 0.152 ± 0.032 a.u. and 0.187 ± 0.076 a.u. for control and XL765-treated group respectively). Taken together with the in vivo 1 H MRS data, these results point to 2HG as a robust imaging biomarker of IDHmut glioma response to XL765 treatment.

Production of 2HG from glucose and glutamine is reduced in treated cells. Previous work 24 has
shown that glucose and glutamine are the precursors of 2HG in IDHmut glioma cells. To further confirm our 1 H MRS findings and assess alterations in metabolic fluxes from glucose and glutamine induced by XL765 treatment, we used 13 C MRS to probe the fate of [1-13 C]glucose and [3-13 C]glutamine. Consistent with previous work 24 , our extracts demonstrated that [1-13 C]glucose was incorporated into [4-13 C]2HG (34.5 ppm) in both NHAIDHmut (Fig. 7A) and U87IDHmut (Fig. 7B) models while [3-13 C]glutamine was incorporated into [3-13 C]2HG (32 ppm, Fig. 7C,D). Our results indicate that, following XL765 treatment, flux from both glucose and glutamine towards 2HG was significantly reduced in both NHAIDHmut and U87IDHmut models, thus explaining the drop in 2HG steady state levels observed by 1  www.nature.com/scientificreports www.nature.com/scientificreports/ cell in U87IDHmut control dropped to 0.06 ± 0.04 fmol/cell in U87IDHmut XL765-treated cells; p = 0001; Supplementary Fig. 5, N = 4). Comparison of the steady-state concentration of 2HG determined from the 1 H spectra to the combined 13 C MRS values showed that the data were within experimental error, further validating our findings. Specifically, for NHAIDHmut the 1 H MRS-derived steady state 2HG was 4.71 ± 1.15 fmol/cell for control and dropped to 0.42 ± 0.27 fmol/cell following XL765 treatment, while the combined 13 C values were 5.05 ± 0.36 fmol/cell and dropped to 0.46 ± 0.12 fmol/cell (Fig. 7E,F). For U87IDHmut steady-state 2HG was 1.49 ± 0.30 fmol/cell in control and 0.31 ± 0.09 fmol/cell in XL765-treated, while the combined 13 C values were 1.66 ± 0.20 fmol/cell and 0.34 ± 0.23 fmol/cell respectively (Fig. 7G,H).

Discussion
Treatment of lower grade gliomas increasingly involves chemotherapy and radiotherapy [51][52][53] . In an effort to further improve outcomes for LGG patients, other therapeutic options are being considered. One such approach is to inhibit the PI3K/mTOR pathway 33 . PI3K/mTOR pathway activation has been linked with worse patient survival in LGGs 43 , supporting clinical trials of inhibitors of the pathway as a treatment regimen for those tumors (NCT02023905, NCT01316809). The outcomes of these trials remain to be determined, but studies in other tumor types have shown that PI3K/mTOR signaling inhibitors often lead to tumor stasis rather than shrinkage [46][47][48] . The goal of our study was therefore to identify potential metabolic imaging biomarkers that could help assess drug target engagement and the likely response of LGG to inhibitors of the PI3K/mTOR pathway.
To perform our studies, we chose to focus on monitoring the effect of a dual PI3K and mTOR inhibitor. This choice was driven by research in other tumor types that has shown that inhibition of mTOR only results in a feedback loop, activation of compensatory pathways, and tumor resistance [54][55][56][57] . More effective inhibition of tumor growth was achieved by simultaneously targeting both PI3K and mTOR. Additionally, it has been shown that dual PI3K/mTOR inhibitors, such as XL765, were more effective that individual PI3K or mTOR inhibitors combined 58 . Collectively such studies have led to the use of dual inhibitors in clinical phase I and II trials in a range of tumor types including GBM and, importantly, several such studies have resulted in clinically-relevant disease stability 50,59-64 (NCT03522298, NCT01547546).
We assessed the effect of treatment in two genetically engineered cell models. These models grow reliably and reproducibly as cells and, in the case of the U87IDHmut model, also as tumors, providing us with a robust platform to assess the impact of our inhibitor. Our studies show that whereas response to XL765 in our IDHmut models is associated with a drop in several metabolites, only a drop in 2HG was also reliably detected in the in vivo setting. Future studies will focus on confirming our findings in patient-derived models, but it should be noted www.nature.com/scientificreports www.nature.com/scientificreports/ that few such models are available and they are typically challenging to grow consistently both as cells and in vivo. Nonetheless, the drop in 2HG was observed in both of our models, independent of their genetic background (one is a GBM and the other is a NHA) lending support to the reliability of our findings. Furthermore, our finding that 2HG is reduced following PI3K/mTOR inhibition is in agreement with the report from Hujber et al. 65 who showed reduce 2HG production in IDHmut fibrosarcoma cells following mTOR inhibition.
Our observation that 2HG drops in vivo was independent of tumor size, but was associated with enhanced animal survival, pointing to the value of this biomarker as an indicator of PI3K inhibition and likely response to treatment. The drop in 2HG would serve as a biomarker that is specific to IDHmut LGG tumors. Studies of GBM tumors, including U87 tumors, previously performed by us and others detected changes in lactate production and phosphocholine levels, but changes in 2HG would not be expected in GBM as they do not harbor the IDH1 mutation and do not produce 2HG [66][67][68] .
The use of the genetically-engineered U87IDHmut model for in vivo validation of our findings could be considered a potential limitation of our study. However, it should be noted that there is a only a small number of patient-derived IDHmut glioma models that form orthotopic tumors in vivo (e.g MGG119 and MGG152 and BT142) 38,69 . In our prior studies, we have investigated the patient-derived BT142 model 24,69,70 . However, this model has lost 2HG production due to loss of the wild-type IDH allele, a phenomenon that has been shown to occur in IDH mutant gliomas 71 . In contrast, the U87IDHmut model is an easy to grow and very stable model that has been frequently used, not only by our group, but also in several other studies in the literature [72][73][74][75][76] . Importantly, we have previously demonstrated that 2HG levels in our U87IDHmut tumor tissue extracts (9.8 ± 1.6 μmol/g of tissue) were comparable to 2HG levels reported in patients (5-35 μmol/g) 77 . Furthermore, it should be noted that our metabolic findings in the U87IDHmut model differ significantly from that of a prior study which examined the response of IDH wild-type glioblastomas to XL765 treatment 67 . Based on these observations, we believe that the use of the U87IDHmut glioma model in vivo is justified for the purposes of the current study.
The mechanism by which inhibition of PI3K/mTOR signaling leads to a drop in 2HG remains to be fully elucidated. However, our 13 C MRS studies point to a drop in both the flux of glucose and the flux of glutamine toward that metabolite. These findings are consistent with previous publications that have shown that the PI3K/mTOR pathway controls both glucose and glutamine uptake and metabolism [78][79][80][81] . Inhibition of PI3K/mTOR would thus inhibit glucose and glutamine metabolism providing a possible explanation for our observations. Importantly, when considering translation of our findings, several approaches have now been optimized for 2HG detection at clinical field strengths as recently review by Leather et al. 27 . Each has its strengths and weaknesses [82][83][84][85][86][87] . A study using a standard single-voxel double echo point-resolved spectroscopy (PRESS) sequence with a short echo time (TE = 30 ms) at 3 T combined with a standard fitting algorithm (LCModel) 88 can be readily implemented to detect 2HG but was associated with a 26% false positive detection rate in IDH wild type tumors due to the overlap of 2HG with glutamate and glutamine at 2.25 ppm 85 . An optimized PRESS sequence utilizing a longer TE (TE = 97 ms at 3 T or 78 ms at 7 T) can improve spectral resolution and reduce spectral overlap 82,89 leading to 100% specificity and sensitivity for 2HG detection, but the method can lead to quantification errors due to the longer TE 90 . An alternative approach is the use of a 2D correlation spectroscopy (COSY) sequence at 3T 83,86 . The method counteracts the known confounding spectral overlap observed in 1D MRS spectra and allows for better separation of metabolites, but the clinical applicability of such a technique is limited due to the extended acquisition time, an important factor for incorporation into clinical practice. An alternative approach is the use of three dimensional MRS imaging approach combined with 2HG spectral editing, an approach which has been already applied for treatment response assessment in patients 34 . Finally, a study at 7 T recently utilized an optimized semi-localization by an adiabatic selective refocusing (semi-LASER) sequence 84,87 and a study at 9.4T 91 showed that 2HG can be reliably detected in patients in vivo using a short echo stimulated echo acquisition mode (STEAM) sequence, but such higher field strengths are not available in most hospitals potentially limiting the clinical utility of such sequences. Importantly however, these multiple approaches confirm the ability of MRS to clinically detect 2HG in brain tumor patients.
Further studies are needed in patient derived models. Nonetheless, given the feasibility of detecting 2HG in the clinic, and considering our findings that a drop in 2HG is associated with response to PI3K inhibition, this metabolite could serve as an indicator of drug target modulation in IDHmut glioma, providing an MRS detectable noninvasive imaging biomarker of response to PI3K/mTOR inhibitors that is independent of tumor size.

Methods
Cell culture and treatment. Two previously described genetically engineered IDHmut lines were used in this study: U87IDHmut and NHAIDHmut 24,92-95 . Briefly cells were created by introduction of lentiviral constructs encoding for R132H IDH1 cDNA into U87 GBM cells or immortalized normal human astrocytes (NHA) respectively 49 . Cells were routinely cultured as monolayers in DMEM (Gibco, ThermoFisher Scientific) supplemented with 10% FBS (Gibco, ThermoFisher Scientific), 100 units/mL penicillin and 100 μg/mL streptomycine (Gibco, ThermoFisher Scientific). Both cell lines were authenticated by short tandem repeat profiling within 6 months of any study. NHAIDHmut cells were treated with 32 μM XL765 (Voxtalisib, SAR245409; Selleck, USA) for 72 h and U87IDHmut cells were treated with 12 μM for 24 h. These time points and concentrations were chosen based on the doubling times of the NHAIDHmut (~40 h) and U87IDHmut (~20 h) such that cells were treated for approximately 1.5x their doubling time in order to achieve ~50% inhibition of cell proliferation. DMSO was used as vehicle control (final concentrations of 0.16% for NHAIDHmut cells and 0.06% for U87IDHmut). www.nature.com/scientificreports www.nature.com/scientificreports/ Western blot analysis. Cells and tumor tissues were examined as described previously 94,95 . Phosphorylated initiation factor 4E-binding protein 1 (p4E-BP1) and phosphorylated S6 kinase (pS6K) were used as a downstream readout of PI3K/mTOR signaling and β-actin as loading control. Control and treated cells and tissues were lysed using Cell Lysis Buffer (ThermoFisher Scientific) supplemented with 1 μl/ml protease inhibitor cocktail set III (Calbiochem) and 10 μl/ml phosphatases inhibitor (Sodium Orthovanadate; New England Biolabs, Inc.). Lysates normalized to cell number or to wet tissue weight were then run on 4-20% gels (Bio-Rad) using the SDS-PAGE method and electrotransferred onto nitrocellulose membranes. Membranes were blocked in blocking buffer containing 5% BSA in Tris-Buffered Saline Tween-20 (TBST) and incubated with the primary antibodies anti-p4E-BP1 (Thr37/46, Cell Signaling #2855), anti-4E-BP1 (Cell Signaling #9452), anti-pS6 kinase (Thr389, Cell Signaling #9234), anti-S6 kinase (Cell Signaling #9202) and anti-β-actin (Cell Signaling #4970) overnight at 4 °C. HRP-conjugated secondary antibodies (Cell Signaling #7074) were incubated for 60 min in TBST at room temperature. Immunocomplexes were visualized using ProSignal Pico (Genesee Scientific). Densitometry of the bands was performed using ImageJ software (NIH) to quantify protein expression levels and the data were normalized to the β-actin protein levels. Results are expressed as mean ± standard deviation (n ≥ 3 unless otherwise specified) and the significance of comparisons was determined using unpaired two-tailed Student's t-test with unequal variance, with a p-value ≤ 0.05 considered as statistically significant.
Cell extraction. Extractions were performed using the dual-phase extraction method 24 . Briefly, cells were trypsinized, washed with ice cold 0.85% saline (UCSF Cell Culture Facility) and 10 ml ice cold methanol (Sigma-Aldrich) was added to the cell pellet. The solution was then vortexed and 10 ml of ice-cold chloroform (Acros Organics) added. After another vortexing, 10 ml of ice cold Milli-Q water was added and a final vortexing performed. Phase separation was achieved by centrifugation for 10 min at 3000 rpm at 4 °C, phases separated and solvents removed by lyophylization. The aqueous phase was then reconstituted in 400 μl deuterium oxide (Acros Organics) for MRS studies.
MRS data acquisition. 1 H and proton-decoupled 13 C spectra of the aqueous phase of cell extracts were recorded using a 500 MHz spectrometer (Bruker BioSpin) equipped with a triple resonance cryoprobe. The 1 H spectra were acquired using a 90° flip angle and 3 s repetition time (TR) while the 13 C spectra were acquired using a 30° flip angle and TR of 3 s. For both control and XL765-treated extract samples there were 384 scans for the acquisition of the 1 H spectra. In the case of 13 C spectra, 4000 scans were acquired for controls and 12000 scans for XL765-treated extract samples (due to lower metabolite levels following treatment). In addition, fully relaxed 1 H and 13 C spectra were recorded and served to determine correction factors for saturation and nuclear Overhauser enhancement (NOE; 13 C acquisitions only).
MRS data analysis. Multivariate analysis. 1 H spectra were manually phased, baseline corrected and shifted to align TSP to 0 ppm in MestReNova (version 7.1.1, MestreLab Research). Peaks not associated with the metabolic profile (e.g. residual water (4.5-5.1 ppm), TSP and residual methanol were removed). Icoshift algorithm 96 using Matlab (R2015b, The Mathworks Inc.) was applied to minimize peak misalignments and spectra were normalized to cell number and TSP area. An unsupervised Principle Component Analysis (PCA) was then performed using SIMCA (Version 15.0, Umetrics, Sweden) to determine the intrinsic clustering and distribution of samples between XL765-treated and control cells. In order to give equal weighting to all spectral regions, unit variance scaling was used. The orthogonal partial least squares discriminant analysis (OPLS-DA) supervised classification model was then applied to visualize class separation and the significant changes between groups were identified based on the S-plots. The validation of the PCA and OPLS-DA models against overfitting was determine by the goodness of fit (R2X: fraction of variation of X variables explained by PCA model and R2Y: fraction of variation of Y variables explained by OPLS-DA model) and goodness of prediction (Q2).
Univariate analysis. The concentration of metabolites identified via the multivariate analysis as most significantly altered following treatment (based on |r| > 0.8 in the S-plot) were also quantified using a targeted univariate analysis. Peaks were integrated using MestReNova, integrals corrected for saturation and NOE, and normalized to cell number and to an external sodium 3-(trimethylsilyl)propionate-2,2,3,3-d4 (TSP; Sigma-Aldrich) reference of known concentration. The significance of univariate comparisons was determined using an unpaired two-tailed Student's t-test with unequal variance, with a p-value ≤ 0.05 considered as statistically significant.
In vivo studies. All studies were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the University of California San Francisco Institutional Animal Care and Use Committee (IACUC Protocol No: AN148391). Six-to-seven-week old female athymic nu/nu mice (Charles River) were intracranially injected with approximately 3 × 10 5 U87IDHmut cells as previously 95 . Once tumors reached 2-3 mm in diameter (day zero, D0) the mice were randomized into either a treatment or a control group that were treated, respectively, with 30 mg/kg XL765 twice daily p.o. or with 10 mM HCl (vehicle facilitating drug uptake from the GI track). The XL765 solution was prepared immediately before administration.
Anatomic MR imaging. All in vivo MR studies were performed using a 14.1 T vertical MR system (Agilent Technologies) equipped with a single channel volume 1 H coil. A spin echo multi-slice sequence was used to acquire axial T2-weighted images of the mouse head in order to evaluate tumor volume. The sequence parameters were: echo time (TE) = 20 ms, TR = 1200 ms, field of view = 30 × 30 mm 2 , matrix = 256 × 256, slice thickness = 1 mm and number of averages = 2. Anatomical images were acquired on D0 and then every 3-4 days until animals needed to be sacrificed. Tumor volume was evaluated as the sum of manually contoured tumor areas in each slice multiplied by slice thickness using an in-house IDL based software 67 and normalized to D0 value. (2019) 9:10521 | https://doi.org/10.1038/s41598-019-47021-x www.nature.com/scientificreports www.nature.com/scientificreports/ In Vivo 1 H MRS. At the last time point prior to animal sacrifice 1 H spectra were acquired from an 8 mm 3 voxel placed in the center of the tumor using a point resolved spectroscopy (PRESS) sequence with: TE = 20 ms, TR = 4000 ms, NA = 448, 10000 points and spectral width 10000 Hz combined with water suppression. Additionally, a spectrum of the unsuppressed water signal was acquired with the same parameters as above, but with 4 averages only, for subsequent eddy current correction during LCModel spectral analysis. The spectral region between 0.5 and 4.0 ppm was then analyzed using LCModel 88 and a basis set comprised of alanine (Ala), aspartate (Asp), total choline (tCho), total creatine (tCr), GABA, glutamine (Gln), glutamate (Glu), 2-hydroxyglutarate (2HG), myo-inositol (m-Ins), lactate (Lac), N-acetylaspartate (NAA), N-acetylaspartylglutamate (NAAG), phosphoethanolamine (PE), scyllo-inositol (s-Ins) and taurine (Tau). Each peak was normalized to the total signal calculated as the sum of all the quantified metabolites. The average concentration for each metabolite C was then calculated as follows (per the LCModel manual) so as to give values with lower Cramer-Rao lower bounds (CRLB) a higher weighting than values with a higher CRLB: with an estimated standard deviation where C j the normalized signal of the metabolite for the animal j and CRLB (% ) j its corresponding CRLB expressed in per cent. Average metabolic changes were then calculated for each metabolite of interest using the lowest possible CRLB that led to at least 4 animals per group.
Tumor tissue immunohistochemistry. At the end of the MR studies, tumor-bearing brains were immediately removed and fixed in 10% buffered formalin for 24 h, then dehydrated with ethanol and embedded in wax (Paraplast Plus, McCornick Scientific). For each sample, adjacent tissue sections were stained with hematoxylin and eosin and Ki-67. For Ki-67, the slides were incubated for 60 min with a monoclonal mouse Ki-67 antibody (M7240, Dako, Copenhagen, Denmark), diluted 1:100 at room temperature. Subsequently they were incubated for 60 min with the secondary antibody which was peroxidase-labeled horseradish peroxidase polymer (Dako). The antigen localization was achieved by the 3,3'-diaminobenzidine chromogen (Dako). Nuclei were considered to be Ki-67 positive if any nuclear staining was present, regardless of staining intensity. A Ki-67 labeling index was defined as the average percentage of positive stained cells from 3 regions of each sample.
Analysis of tumor tissue. Extraction. At the end of the MR studies, tumors were excised, snap-frozen in liquid nitrogen and stored at −80 °C until further investigation. For complementary 1 H MRS studies pieces of tumor tissue weighting 9 to 24 mg (wet mass) were homogenized in 500 μl ice cold phosphate buffer (PBS) with 1 μl/ml protease inhibitor cocktail set III (Calbiochem) using beads (TissueLyser LT, QIAGEN). The dual-phase extraction method described above for the cell extracts was then applied and the aqueous phase was reconstituted in 400 μl 100 mM phosphate buffer in deuterium oxide (Acros Organics).
MRS data acquisition and analysis. 1 H spectra of tumor tissue extracts were recorded and analyzed the same way as described above for the cell extracts. The only modification was that the analysis was focused in the in vivo relevant spectral region of 0.5 to 4.2 ppm.

Data Availability
Data generated during the current study are available from the corresponding author upon request.