IDH1R132H is intrinsically tumor-suppressive but functionally attenuated by the glutamate-rich cerebral environment

Recurrent heterozygous mutation of isocitrate dehydrogenase 1 gene (IDH1), predominantly resulting in histidine substitution at arginine 132, was first identified in glioma. The biological significance of IDH1R132H, however, has been controversial, and its prevalent association with glioma remains enigmatic. Although recent studies indicate that IDH1R132H is nonessential to tumor growth or even anti-tumor growth, whether IDH1R132H initiates gliomagenesis remains obscure. In this study, we report that IDH1R132H is intrinsically tumor-suppressive but the activity can be attenuated by glutamate—the cerebral neurotransmitter. We observed that IDH1R132H was highly suppressive of subcutaneous tumor growth driven by platelet-derived growth factor B (PDGFB), but IDH1R132H tumor growth and glioma penetrance were virtually indistinguishable from those of IDH1-wildtype tumors in orthotopic models. In vitro, addition of glutamate compromised IDH1R132H inhibition of neurosphere genesis, indicating glutamate promotion of oncogenic dominance. Furthermore, we observed that IDH1R132H expression was markedly decreased in tumors but became more permissible upon the deletion of tumor-suppressor gene Cdkn2a. To provide direct evidence for the opposing effect of IDH1R132H on PDGFB-driven glioma development, we explored tandem expression of the two molecules from a single transcript to preclude selection against IDH1R132H expression. Our results demonstrate that when juxtaposed with oncogenic PDGFB, IDH1R132H overrides the oncogenic activity and obliterates neurosphere genesis and gliomagenesis even in the glutamate-rich microenvironment. We propose therefore that IDH1R132H is intrinsically suppressive of glioma initiation and growth but such tumor-suppressive activity is compromised by the glutamate-rich cerebral cortex, thereby offering a unifying hypothesis for the perplexing role of IDH1R132H in glioma initiation and growth.


INTRODUCTION
Heterozygous mutations in the isocitrate dehydrogenase 1 (IDH1) gene are found most frequently in glioma, predominantly resulting in the mutant enzyme IDH1 R132H with histidine substitution at arginine 132 [1][2][3]. The biological function of IDH1 R132H , however, remains controversial. The prevailing belief is that IDH1 R132H is oncogenic owing to the gain of neomorphic activity that converts 2-oxoglutarate (aka α-ketoglutarate)-the product of wild-type IDH1-in an NADPH-dependent reduction to an "oncometabolite" D-2-hydroxyglutarate (D2-HG), which in turn inhibits a class of 2-oxoglutarate-dependent dioxygenases involved in epigenetic regulation, collagen

Research Paper
Oncotarget 35101 www.oncotarget.com synthesis, and cell signaling [4]. Supporting evidence for this theory includes 1) the association of IDH1 mutations with glioma evolution, glioma CpG island methylator phenotype, and proneural subtype; 2) the induction of methylator phenotype in normal human astrocyte by IDH1 R132H transduction or D2-HG treatment; and 3) the association of IDH1 mutations with repressive histone methylation marks that contribute to a less differentiated or stem-like state [5]. Despite the circumstantial evidence, the exact mechanism by which IDH1 R132H drives glioma initiation remains ill-defined, and, moreover, evidence from recent studies apparently challenges this belief.
Specifically, despite effective reduction of D2-HG by small-molecule inhibitors specific to mutant IDH1, treated glioma cells, unexpectedly, accelerated proliferation and shortened survival in an animal model [6]. Therapeutic sensitivity is important to improved survival of glioma patients with IDH1 mutations, but mutant IDH1 inhibitors desensitized tumors cells to irradiation and chemotherapy [7]. Apparently, these counterintuitive findings not only argue against therapeutic targeting of IDH1 mutations but also question the presumptive oncogenic activity of IDH1 R132H . Consistently, previous studies showed that IDH1 R132H transduction inhibited rather than stimulated tumor growth [8,9], and gliomas with IDH1 mutations possessed attenuated oncogenic signaling in comparison with those without [8,[10][11][12][13]. These studies have led us to posit that IDH1 mutations are tumor-suppressive on the contrary; the biological consequence of IDH1 mutations in glioma is to ameliorate patient survival, at least in part, by inhibiting oncogenic signaling [13]. This concept is in accordance with the experimental demonstration of antitumor effects of D2-HG, which decreases the stability of MYC/CEBPA transcripts via N 6 -methyladenosine RNA modification and thereby inhibits tumor cell survival and proliferation [14]. We have reported recently that whereas heterozygous but not hemizygous IDH1 R132H suppresses anchorage-independent growth of glioma cells, the surviving cells conversely selects against IDH1 R132H heterozygosity [15]. Our findings not only support the concept of IDH1 R132H being anti-oncogenic but also suggest the strong antagonism between tumor growth and heterozygous IDH1 R132H expression in the experimental setting. This interpretation is consistent with the requirement of a wild-type IDH1 allele for D2-HG production [16,17] and the frequent loss of either wildtype or mutant IDH1 allele in patient-derived xenograft, ex vivo neurosphere culture, and glioma recurrence and progression [11,16,18,19], even though the underlying mechanism of copy number alteration remains unclear.
The concept that IDH1 R132H heterozygosity is antioncogenic and incompatible with tumor growth seems at odds with the fact that greater than 70% of WHO grade II and grade III gliomas and secondary glioblastomas harbor IDH1 mutations [1][2][3]. Moreover, IDH1 R132H -specific inhibitor and mutant IDH1 pan-inhibitor have been shown to be effective in animal studies [20,21]. It is noteworthy, however, that the anti-oncogenic activity of heterozygous IDH1 R132H can be circumvented by genetic and metabolic alterations, including the loss of IDH1 R132H heterozygosity and use of reducing equivalent [15]. Furthermore, deletion or amplification of either mutant or wild-type IDH1 allele decreases D2-HG in glioma recurrence [19]. Moreover, glutamate-a neurotransmitter rich in the cerebral cortex-is sufficient to bypass the inhibitory effect of IDH1 R132H on glioma progenitor proliferation [9]. These findings altogether indicate the delicate nature of heterozygous IDH1 R132H , whose tumor-suppressive activity can be compromised by genetic alterations and tumor microenvironment.

IDH1 R132H transduction suppresses subcutaneous tumor growth
We reported recently that heterozygous IDH1 R132H is functionally anti-oncogenic, as evidenced by the antagonism between IDH1 R132H heterozygosity and anchorageindependent growth; whereas heterozygous IDH1 R132H suppressed neurosphere genesis, the surviving neurosphere selected against the expression of either IDH1 R132H or IDH1 transgene and reduced D2-HG levels [15]. To ascertain the tumor-suppressive activity of IDH1 R132H in vivo, we first established subcutaneous tumor growth of mouse astrocyte NA1 that had been transduced with luc-PDGFB, which expresses luciferase and platelet-derived growth factor B (PDGFB) upon P2A cleavage [15]. PDGFB has been used extensively for gliomagenesis in vivo [9,[22][23][24][25][26]. Accordingly, the transduced NA1 developed robust tumor growth with a volume-based doubling time of 9.4 days in contrast to NA1 transduced with luc*, which expresses the same transcript harboring a stop codon engineered at the P2A (Supplementary Figure 1).
Next, we sought to test whether IDH1 R132H cotransduction inhibits tumor growth using YFP-IDH1 R132H , which expresses nuclear yellow fluorescent protein (YFP) and IDH1 R132H upon P2A cleavage [15]. As expected, YFP-IDH1 R132H significantly inhibited cell proliferation, resulting in 20% increase of the mean doubling time to 28.8 hours compared with 24.0 hours of its control YFP-IDH1 (Supplementary Figure 2A). In keeping with this, YFP-IDH1 R132H cells showed G 2 /M arrest compared with YFP-IDH1 cells (Supplementary Figure 2B). In agreement with its effect on neurosphere genesis [15], YFP-IDH1 R132H markedly inhibited tumor growth, as indicated by bioluminescent imaging and confirmed by tumor weight analysis ( Figure 1A-1C). The mean volumebased doubling time of YFP-IDH1 R132H tumors increased 66% to 15.8 days from 9.5 for YFP-IDH1 tumors. Histological examination confirmed reduced cellularity, nuclear-cytoplasm ratio, and nuclear pleomorphism in www.oncotarget.com YFP-IDH1 R132H tumors compared with YFP-IDH1 tumors ( Figure 1D). Furthermore, immunohistochemistry showed decreased Ki67 as well as PDGFB staining in YFP-IDH1 R132H tumors. It is noteworthy, however, in contrast to the staining of HA-tagged wild-type IDH1 in the control, HA-tagged IDH1 R132H was nearly undetectable in YFP-IDH1 R132H tumors and sparsely stained with an anti-IDH1 R132H antibody (Supplementary Figure 2C). Taken together, these results support the concept that IDH1 R132H is not only tumor-suppressive but is also selected against in the surviving tumors, as reported previously in anchorageindependent growth [15].

Antagonism between IDH1 R132H transgene expression and tumor growth
To provide further evidence for the selection against IDH1 R132H expression in tumor growth, we observed 76% reduction of YFP-IDH1 R132H transcript levels accompanied by 35% reduction of PDGFB transcript levels compared with those in YFP-IDH1 tumors (Figure 2A, 2B), a finding in agreement with the selection against IDH1 R132H transgene in neurosphere culture [15]. No reduction of IDH1 R132H copy number, however, was observed at the genomic DNA level ( Figure 2C), suggesting non-genetic event(s) for IDH1 R132H downregulation.
Next, we employed fluorescent microscopy to visualize the suppression of IDH1 R132H transgene expression by examining tumor cells transduced with mCherry-PDGFB, which expresses the fluorescent mCherry and PDGFB upon P2A cleavage [15]. To that end, we transduced NA1 cells with mCherry-PDGFB and YFP-IDH1 R132H or its control YFP*, which expresses only YFP protein from the same YFP-IDH1 R132H transcript that harbors an engineered stop codon [15]. Of note, we opted for YFP* as a more appropriate control because the upregulation of wild-type IDH1 promotes aggressive growth of malignant glioma [27]. As expected, YFP-IDH1 R132H co-transduction resulted in significant decreases in cell proliferation and subcutaneous tumor growth astrocytes that had been transduced with luc-PDGFB showing significant growth suppression by co-transduction with YFP-IDH1 R132H (YFP-mIDH1) compared with YFP-IDH1 (A). Nonlinear regression curve fit was performed using exponential growth equation and twoway ANOVA for the analysis of statistical significance. RLU, relative luminescent units. **** p < 0.0001. Suppression of tumor growth was supported by bioluminescent imaging (B) and autopsied tumor weight (C). Unpaired t-tests were performed using two-tailed p value. * p < 0.05. (D) Hematoxylin-eosin (H-E) and immunohistochemistry staining revealed less malignant histologic features, decreased Ki67 and PDGFB expression, and weak HA-tagged IDH1 R132H staining in YFP-IDH1 R132H tumor compared with YFP-IDH1 tumor. Scale bar: 50 μm.
Oncotarget 35103 www.oncotarget.com compared with YFP* co-transduction (Supplementary Figure 3). Fluorescent microscopy revealed few cells that were YFP + in the surviving YFP-IDH1 R132H tumor in contrast to widespread YFP + cells in the control ( Figure  2D), which is consistent with the previous finding in neurosphere culture [15]. We conclude therefore that although IDH1 R132H transduction suppresses subcutaneous tumor growth, the surviving tumors, conversely, select against the transgene expression.

IDH1 R132H tumors are histologically indistinguishable from IDH1-wildtype tumors in orthotopic models
Unexpectedly, IDH1 R132H suppression of tumor growth could not be reproduced in orthotopic transplantation models, as shown by bioluminescent imaging and histological examination ( Figure 3A, 3B). Indeed, YFP-IDH1 R132H transduction failed to inhibit PDGFB-driven orthotopic tumor growth, resulting in similar bioluminescent readings in reference to the control. Histological examination showed similar tumor cell proliferation and invasion in both groups of mice ( Figure 3C), an observation consistent with previous reports [25,28,29]. The lack of clear tumor suppression in the orthotopic model indicates a tissue-specific role of the cerebral cortex in the biological effect of IDH1 R132H .

IDH1 R132H expression becomes permissible in glioma with Cdkn2a deletion
Although orthotopic transplantation exhibited similar tumor growth between YFP-IDH1 R132H and YFP* control cells, strong nuclear staining of YFP was seen in YFP*, but not YFP-IDH1 R132H , tumors by immunohistochemistry ( Figure 4A). Additionally, IDH1 R132H staining was scattered in YFP-IDH1 R132H tumor cells (Supplementary Figure 5A). Likewise, in the RCAS/PDGFB glioma model, weak YFP staining was seen in YFP-IDH1 R132H glioma cells in contrast to prevalent nuclear staining in YFP* glioma cells ( Figure  4B; Supplementary Figure 5B). Furthermore, in RCAS/ mCherry-PDGFB-induced gliomas, fluorescent YFP signal was visualized only in the YFP-IDH1, but not YFP-IDH1 R132H , tumors despite a modest decrease of mCherry signal in the latter ( Figure 4C). These results strongly indicate that IDH1 R132H transgene expression is selected against during glial tumor growth irrespective of tumor size and microenvironment, supporting the notion of antagonism between IDH1 R132H expression and tumor growth.
Our findings are apparently at odds with the fact that IDH1 R132H is detectable immunologically in human gliomas and tumor transplantations [10,[36][37][38][39][40]. In fact, IDH1 R132H staining was strong and conspicuous in RCAS/PDGFA gliomas when combined with Cdkn2a knockout and Trp53 knockdown [28]. In light of frequent mutations in various tumor-suppressor genes associated with IDH-mutant glioma [41], we hypothesized that the inactivation of tumor-suppressor gene(s) renders glioma more permissible to IDH1 R132H expression. To test this notion, we compared immunohistochemical IDH1 R132H staining between tumors developed in Cdkn2a-intact and -deleted genetic background. Indeed, IDH1 R132H gliomas derived from Cdkn2a fl/fl mice showed much increased immunohistochemical staining of IDH1 R132H with Cre cotransduction compared with those without ( Figure 4D). Additionally, IDH1 R132H staining was seen in glioma cells forming perineuronal satellitosis ( Figure 4E), as reported previously [36]. Taken together, these results support the selection against IDH1 R132H transgene in PDGFB-driven tumors and the dependence of IDH1 R132H expression on inactivation of tumor-suppressor gene(s).

Expression of IDH1 R132H and PDGFB from the same transcript obliterates gliomagenesis
Although recent studies indicated anti-tumor effects of IDH1 R132H [14,19], whether IDH1 R132H suppresses gliomagenesis remains unclear. To provide evidence that Oncotarget 35105 www.oncotarget.com IDH1 R132H is tumor-suppressive, we engineered a RCAS vector that expresses IDH1 R132H , P2A, and PDGFB from the same transcript. This tandem design of IDH1 R132H -PDGFB not only ensures the expression of the two at 1:1 molar ratio within the same cells but also precludes selection against IDH1 R132H expression especially in the cerebral cortex (Figure 4). Similarly, IDH1-PDGFB was constructed as control.
We confirmed equivalent transgene expression between NA1 cells transduced with IDH1-PDGFB and IDH1 R132H -PDGFB at mRNA and protein levels (Supplementary Figure 6A, 6B). We observed the mean D2-HG levels at 3,583 nmol per mg protein in IDH1 R132H -PDGFB cells ( Figure 5A), a concentration similar to those in human IDH1 R132H glioma cells [15] and fourfold greater than that in IDH1-PDGFB cells. IDH1 R132H -PDGFB cells showed a remarkable decrease in proliferation compared with IDH1-PDGFB cells, resulting in 47% increase of doubling time to 41.9 hours from 28.5 ( Figure 5B). Furthermore, we determined the ability of single cells to form neurosphere; consistent with the inhibitory effect of YFP-IDH1 R132H when co-expressed with luc-PDGFB or mCherry-PDGFB from different transcripts [15], IDH1 R132H -PDGFB cells had a fivefold reduction of Oncotarget 35106 www.oncotarget.com neurosphere genesis from 3.5% in IDH1-PDGFB cells to 0.7, a level equivalent to the parental NA1 ( Figure 5C; Supplementary Figure 6C). These results indicate that IDH1 R132H -PDGFB is a functional platform demonstrating that a single-nucleotide difference in IDH1 is sufficient to confer suppression of anchorage-independent growth by overriding the oncogenic activity of PDGFB.
Previous studies have indicated the importance of glutamate anaplerosis in IDH-mutant glioma metabolism and growth [9,42]. In particular, the addition of glutamate reversed IDH1 R132H -mediated proliferative inhibition of neural progenitor cells co-transduced with PDGFB [9]. Likewise, we previously showed that the addition of reducing equivalent N-acetyl cysteine reversed inhibition of anchorage-independent growth by heterozygous IDH1 R132H [15]. We sought to determine whether IDH1 R132H -PDGFB cells would respond differently to the treatment of glutamate or N-acetyl cysteine in neurosphere genesis. Indeed, treatment with sodium glutamate or N-acetyl cysteine markedly increased size and number of PDGFB-driven neurospheres when YFP-IDH1 R132H was expressed from different transcripts ( Figure 5D, bottom); however, neither treatment had noticeable effect on those transduced with IDH1 R132H -PDGFB (top). Furthermore, results from single-cell analysis confirmed that glutamate treatment had virtually no effect on IDH1 R132H -PDGFB cells in contrast to a 3.5fold increase in the co-transduced cells from 0.69% to 2.43 ( Figure 5E). Therefore, these results not only further support the concept that IDH1 R132H is intrinsically tumorsuppressive but also suggest a complete suppression of glioma development if IDH1 R132H is co-expressed with PDGFB from the same transcript.
Given the overriding role of IDH1 R132H against oncogenic PDGFB in anchorage-independent growth, we predicted that expression of IDH1 R132H with PDGFB from the same transcript would prevent spontaneous glioma initiation and growth even in the glutamine-rich microenvironment. Indeed, none of the mice (13/13) developed glioma with RCAS/IDH1 R132H -PDGFB in contrast to 93% incidence (14/15) in those with RCAS/IDH1-PDGFB ( Figure 6A, 6B). In addition, immunohistochemistry showed widespread Olig2 staining in tumor cells but not in the cortex of IDH1 R132H -PDGFB mice ( Figure 6B). Furthermore, Kaplan-Meier analysis revealed that IDH1 R132H was remarkably beneficial to the survival of IDH1 R132H -PDGFB mice; none of them exhibited clear neurologic signs by the end of the 8-week Oncotarget 35107 www.oncotarget.com period, whereas 73% IDH1-PDGFB mice had to be sacrificed, thereby significantly decreasing the median survival to 43 days ( Figure 6C). Histological examination and Olig2 immunohistochemical staining confirmed the presence or absence of glioma lesions in all of the mice (data not shown). Thus, we conclude that IDH1 R132H is intrinsically suppressive of glioma initiation as well as glioma growth.

DISCUSSION
We present evidence in this study that the outcome of IDH1 R132H transduction in glioma initiation and growth is context dependent even though IDH1 R132H is intrinsically tumor-suppressive. Specifically, we demonstrate that when IDH1 and PDGFB are expressed from the same transcript, a single-nucleotide change of IDH1 at codon 132 determines the fate of gliomagenesis and overall survival of Nes-tva mice. Our results provide direct evidence that IDH1 R132H is not only intrinsically tumor-suppressive but also resistant to functional compromise by the environmental glutamate or reducing power, which would otherwise attenuate the antagonism of IDH1 R132H toward the oncogenic PDGFB when both are expressed from different transcripts. This finding offers an explanation for the distinct effects of IDH1 R132H on tumor growth between the subcutaneous and the glutamate-rich cerebral environment in this study.
Our observation that the addition of glutamate markedly decreased the inhibitory effect of IDH1 R132H on neurosphere growth and genesis from single cells is in agreement with glutamate reversal of IDH1 R132H inhibition of neural progenitor cell proliferation [9] and is consistent with the metabolic dependence of IDH1-mutant glioma on glutamate [43]. Although we did not provide evidence that there is sufficient glutamate in the cerebral cortex to feed glioma in our models, these results nevertheless support the concept that IDH1 R132H -mediated tumor suppression can be compromised by the microenvironmental factors including glutamate and reducing equivalent as escape mechanisms of glioma progression [15]. The study may also account for the prevalence of IDH mutations in glioma and the nonexistence in most other cancer types [2,3,44,45]. Furthermore, our results may provide a clue to the challenging issue of maintaining IDH1 R132H heterozygosity in glioma cell culture [11,18,46].
In accordance with the incompatibility between IDH1 R132H heterozygosity and anchorage-independent growth [15], we observed the strong antagonism between IDH1 R132H expression and PDGFB-driven tumor growth. Interestingly, IDH1 R132H transgene expression was markedly attenuated but more permissible at the expense of Cdkn2a deletion. We anticipate additional Trp53 knockdown would result in even greater IDH1 R132H transgene expression, as shown previously [28]. The biological consequence of tumor-suppressor gene inactivation, however, is the erosion of IDH1 R132H tumorsuppressive activity, as indicated by the complete loss of IDH1 R132H survival benefit in Cdkn2a −/− mice in contrast to Cdkn2a +/+ and Cdkn2a −/+ mice [28]. In light of the association of IDH-mutant gliomas with mutations in tumor-suppressor genes including TP53, ATRX, CIC, NOTCH1, and FUBP1 [41], it stands to reason that IDH1 R132H expression becomes permissible and therefore detectable in these gliomas of various grades [36]. The notion that inactivation of tumor-suppressor gene(s) permits IDH1 R132H existence and expression in glioma may account for the continuous presence of IDH1 R132H in recurrent gliomas [51,52], even though some recurrent gliomas underwent genetic deletion of mutant IDH1 allele and copy number alterations [16,19]. Whether IDH1 R132H detected in recurrent glioma is fully functional remains to be investigated; the finding that IDH1 R132H and D2-HG are nonessential at recurrence nevertheless has raised the question of targeting IDH1 R132H for therapeutic intervention [19]. It is interesting to note that although the IDH1 R132H -specific inhibitor AGI-5198 was shown initially to be effective in inhibiting glioma cell growth in subcutaneous xenograft [20], followup studies found lack of tumor regression in the same tumor model [53]. Despite the high potency in 2HG suppression among available mutant IDH1 inhibitors, their therapeutic efficacies in survival experiments vary from modest to harmful [6,21]. Additionally, studies of gliomagenesis in cell culture models also indicate that 2-HG is nonessential to cell growth [54].
Given the association of PDGFRα with IDH-mutant glioma [39,55,56], the use of PDGFA as an oncogenic driver seems more relevant because PDGFA activates only PDGFRα [57]. Furthermore, since Trp53 deletion is sufficient to induce glioma [29], it will be interesting to test further whether IDH1 R132H obliterates gliomagenesis driven by PDGFA transduction or Trp53 knockdown when expressed from the same transcript. It should be noted that this design, albeit artificial, has enabled us to tease out the intrinsic function of IDH1 R132H , similar to what the genetic engineering of heterozygous IDH1 R132H/+ in HCT116 colon cancer cells, for instance, has done to advance the understanding of glioma biology. Thus far, our tandem design arguably has the advantage of directly determining the antagonism between IDH1 R132H and oncogenic activities in gliomagenesis.
In summary, this study has shown that IDH1 R132H is intrinsically tumor-suppressive, and yet its anti-tumor activity can be compromised by internal factors, such as inactivation of tumor-suppressor gene, and external factors, such as glutamate. The context-dependent effects of IDH1 R132H on tumor initiation and growth may have implications in glioma etiology, model development, and therapeutic targeting.

Cell culture, retroviral infection, and neurosphere culture
Immortalized mouse astrocyte cell line NA1 was prepared and subjected to retroviral infection as described previously [15]. Likewise, resultant cells with fluorescent signals were enriched by flow cytometry, and the IDH1 R132H status was verified by DNA sequencing. Neurosphere growth was compared qualitatively by seeding 10,000 cells in a 48-well plate with 500 μL of neural stem cell medium [Neurobasal media supplemented with B-27, 10 ng/mL bFGF, and 20 ng/mL EGF (Invitrogen)]. An additional 100 μL of neural stem cell medium was added after 4 days. Micrographs were acquired 1 week following seeding. To determine the ability to form neurosphere from single cells, we seeded cells of interest at 1 or 5 cells per well in a 96well plate in triplicate and replenished fresh medium every 2-3 days. Sodium glutamate or N-acetyl cysteine was added at 1 mM and refreshed every 2-3 days. Spheres over 50 μm in diameter were counted after 14 days.

D2-HG analysis
D-2-Hydroxyglutarate Colorimetric Assay Kit (BioVision) was used to measure the intracellular level of D2-HG, as per manufacturer recommendations. Briefly, cell lysates from 1 × 10 5 cells were split into three parts to determine the absorbance of the sample, 5 nmol D2-HG spiked sample, and sample background. The protein concentration of cell lysate was determined using the BCA Protein Assay Kit (Thermo Scientific). D2-HG was determined in triplicate according to the manufacturer's protocol and expressed as nmol/mg protein.

Cell proliferation and cell-cycle analysis
Cells expressing luciferase were seeded in a 96-well plate at 100 cells per well in triplicate. Cell proliferation was determined through luciferase activity every 24 hours for 6 consecutive days with a luciferase assay kit (Biotium) or cell viability kit (Promega) and a microplate reader (Turner BioSystems). Relative luciferase units were normalized with background subtraction. Nonlinear regression curve fit was performed using exponential growth equation, and two-way ANOVA was used for the analysis of statistical significance (GraphPad Prism 7.0). Cell-cycle profiling was performed in quadruplicate by labeling the cells with 4′,6-diamidino-2-phenylindole (DAPI) to a final concentration of 10 μg/mL. Cells were then analyzed by flow cytometry (BD FACSCanto, BD Biosciences) with BD FACSDiva Software (BD Biosciences).

Gene expression
Gene expression analysis at genomic DNA, RNA, and protein levels was performed essentially as described before [15]. Amplicon intensities were quantified using an open-source image analysis platform (Fiji) and normalized to Actb expression. Dilutions of primary antibodies for Western blotting are as follows: 1:1000 anti-PDGFB (Santa Cruz Biotechnology), 1:5000 anti-β-actin (Sigma), and 1:500 anti-HA (Abcam).

Mouse models and bioluminescent imaging
All experiments and procedures involving live mice were approved by the University of Utah Institutional Animal Care and Use Committee. Transplantation of transduced cells into the subcutaneous and intracranial sites and bioluminescent imaging of tumor volume were performed essentially as described previously [58]. Alternatively, subcutaneous tumor growth was measured with an electronic caliper, once a week for 6 weeks. The tumor volume was calculated using the formula (length × width 2 ) / 2.
Briefly, 1-2-day-old newborns were subjected to intracranial injection of DF-1 producer cells expressing genes of interest. The cell number per injection was 3 × 10 4 PDGFB mixed with equal numbers of IDH1 or IDH1 R132H or 3 × 10 4 PDGFB or mCherry-PDGFB mixed with 1 × 10 5 YFP* or YFP-IDH1 R132H . For Cdkn2a deletion, additional 1 × 10 5 DF-1 cells expressing Cre recombinase was included. These mice were terminated by the end of 5 weeks post injection or earlier. The cell number for IDH1-PDGFB or IDH1 R132H -PDGFB was 5 × 10 5 . For survival analysis, these mice were sacrificed by the end of 8 weeks or earlier if any of the following symptoms was found: severe lethargy, pronounced hydrocephalus, and severe cachexia.
Autopsied brains were embedded in paraffin after formalin fixation, sectioned at 3 μm, and stained with hematoxylin and eosin for histological analysis. For fluorescent microscopy, samples were flash-frozen in liquid nitrogen, embedded in OCT compound, and sectioned at 30 μM thickness with a cryostat microtome (Leica CM1950). Sections were mounted with 30% glycerin containing 10 μg/mL DAPI prior to imaging with a Nikon A1R confocal microscope and NIS-elements confocal software (Nikon Instruments). The image was converted with an open-source image analysis platform (Fiji).

Immunohistochemistry
Immunohistochemistry was performed in 3-μm sections of formalin-fixed and paraffin-embedded tissues.