Inhibition of mutant IDH1 decreases D-2-HG levels without affecting tumorigenic properties of chondrosarcoma cell lines.

Mutations in isocitrate dehydrogenase 1 (IDH1) and IDH2 are found in a subset of benign and malignant cartilage tumors, gliomas and leukaemias. The mutant enzyme causes the production of D-2-hydroxyglutarate (D-2-HG), affecting CpG island and histone methylation. While mutations in IDH1/2 are early events in benign cartilage tumors, we evaluated whether these mutations play a role in malignant chondrosarcomas. Compared to IDH1/2 wildtype cell lines, chondrosarcoma cell lines harboring an endogenous IDH1 (n=3) or IDH2 mutation (n=2) showed up to a 100-fold increase in intracellular and extracellular D-2-HG levels. Specific inhibition of mutant IDH1 using AGI-5198 decreased levels of D-2-HG in a dose dependent manner. After 72 hours of treatment one out of three mutant IDH1 cell lines showed a moderate decrease in viability, while D-2-HG levels decreased >90%. Likewise, prolonged treatment (up to 20 passages) did not affect proliferation and migration. Furthermore, global gene expression, CpG island methylation as well as histone H3K4, -9, and -27 trimethylation levels remained unchanged. Thus, while IDH1/2 mutations cause enchondroma, malignant progression towards central chondrosarcoma renders chondrosarcoma growth independent of these mutations. Thus, monotherapy based on inhibition of mutant IDH1 appears insufficient for treatment of inoperable or metastasized chondrosarcoma patients.

Chondrosarcoma is the second most common primary malignant bone tumor and represents a heterogeneous group of tumors [14]. So-called dedifferentiation occurs in 10-15% of central chondrosarcomas [15]. Dedifferentiated chondrosarcoma is a highly malignant tumor characterized by a bimorphic histological appearance with distinct and abruptly separated areas of low grade chondrosarcoma juxtaposed to a high grade, non-cartilaginous sarcoma [16]. ~54% of the dedifferentiated chondrosarcomas contain mutations in IDH1 or IDH2 [8,10].
The signaling pathways that regulate endochondral ossification are thought to also play a role in the development of enchondromas and chondrosarcomas [17]. Of these, Hedgehog signaling (Hh) is thought to be most important and constitutively active signaling is found in enchondromas and central chondrosarcomas [18,19]. In gliomas, IDH1/2 mutations are associated with active Hh signaling [20]. Isocitrate dehydrogenase is an enzyme involved in the conversion of isocitrate to α-ketoglutarate. Three isoforms of IDH are known. IDH1 is localized in the cytoplasm while IDH2 and IDH3 act in the mitochondria. Gain of function mutations are exclusively found on the arginine residues R132 in IDH1 and R140 and R172 in IDH2. These mutations in IDH-1 or -2 lead to gain-of-function, by which the mutant enzyme acquires the activity to convert α-ketoglutarate into D-2-hydroxyglutarate (D-2-HG), but not to its enantiomer L-2-hydroxyglutarate (L-2-HG). The newly formed oncometabolite D-2-HG shows structural similarities with α-ketoglutarate, and as a result D-2-HG is able to competitively inhibit α-ketoglutarate dependent enzymes such as the ten-eleven translocation (TET) enzymes [21]. TET enzymes are involved in DNA demethylation [22][23][24][25]. Indeed, increased levels of D-2-HG have been found in cartilage tumors with an IDH1 or IDH2 mutation [8], and DNA hypermethylation was shown in enchondromas with an IDH1/2 mutation [8,9].
Genome-wide CpG methylation sequencing of chondrosarcoma biopsies revealed that IDH1/2 mutations are associated with DNA hypermethylation at CpG islands but not at other genomic regions [26]. In addition, histone demethylases are also α-ketoglutarate dependent [21] and an increase in methylation of the histone H3 lysine residues was shown in knock-in mice with an IDH1 R132H mutation [27]. Trimethylation of H3K4 positively regulates transcription, whereas trimethylation of H3K9 and H3K27 is associated with repression of transcription [28,29]. Furthermore, mutations in IDH1/2 are associated with stabilization of hypoxia inducible factor-1α (HIF1α) through an effect on the prolyl hydroxylases (PHD). Gliomas with an IDH1/2 mutation show upregulation of HIF1α, whereas PHD activity is inhibited in artificial mutant IDH1 cell lines [30].
The high prevalence of IDH1/2 mutations in enchondroma and chondrosarcoma suggest a causal rather than a bystander role. This led us to investigate the function of the IDH1/2 mutation in chondrosarcoma. Chondrosarcoma patients show a poor response to conventional chemotherapy and radiotherapy, and surgery is the mainstay of treatment. Alternative treatment strategies are urgently needed as no treatment options are currently available for patients with inoperable or metastatic disease. To evaluate the functional role of IDH1/2 mutations we used a chondrosarcoma cell line panel including five IDH1/2 wildtype, three endogenous IDH1 mutant and two endogenous IDH2 mutant cell lines, originating from conventional central as well as dedifferentiated chondrosarcomas. Remarkably glioma and leukemia cells harboring mutations in IDH1 or -2 cannot be maintained in culture, however we previously reported that chondrosarcoma cell lines retain these mutations [31] providing us with a model to functionally study the effects of IDH1 or -2 mutations in their naturally occurring context and to evaluate whether inhibition of mutant IDH1/2 could be a potential treatment strategy.

results cell lines with an endogenous mutation in IDH1 or -2 express the mutant allele
Sequencing for mutations in exon 4 of IDH1 and -2 in the cell lines confirmed the previously reported heterozygous mutations in JJ012 (IDH1 R132G), L835 (IDH1 R132C), SW1353 (IDH2 R172S), HT1080 (IDH1 R132C) and the homozygous R172W IDH2 mutation in L2975. Moreover, in all cell lines the mutant allele was expressed, as shown by sequencing of the corresponding cDNA samples (example shown in Figure 1A).

Mutant IDH1 or -2 cell lines have increased levels of d-2-HG intra-and extracellularly
Intracellular levels of D-2-HG as determined by LC-MS/MS were elevated up to a 1000-fold in cell lines with IDH1 or -2 mutations as compared to cell lines wildtype for IDH1/2 ( Figure 1B). Measurements of extracellular D/L-2-HG levels in the culture media revealed that the ratio between D-2-HG and L-2-HG was elevated up to a 100-fold in mutant IDH1/2 cell lines ( Figure 1C). No elevated levels of the enantiomer L-2-HG were found in mutant IDH1 or -2 cell lines. Levels of D-2-HG did not differ between cell lines with a mutation in IDH1 versus those with a mutation in IDH2. www.impactjournals.com/oncotarget cell lines with mutations in IDH1 or -2 have a different methylome as compared to wildtype cell lines Since mutations in IDH1 were shown to cause global hypermethylation in tumors [9,26,32], we assessed the methylome in the chondrosarcoma cell lines. Unsupervised hierarchical clustering for the top 2000 most differentially methylated CpG sites resulted in a cluster containing IDH1/2 wildtype and a cluster containing mutant IDH1/2 cell lines (Supplementary Figure 1A). Unexpectedly, when all 480 000 probes are included, comparison of the mean β-values between mutant IDH1/2 and IDH1/2 wildtype cell lines suggested global hypermethylation of the IDH1/2 wildtype cell lines (Supplementary Figure 1B). When analyzing the different probes separately, only CpG islands probes showed hypermethylation in the chondrosarcoma cell lines with an IDH1/2 mutation ( Figure 2B), whereas the shores (regions flanking the CpG islands) and shelves (regions flanking the shores) show hypermethylation in the IDH1/2 wildtype cell lines (Supplementary Figure 1B). Since CpG islands are known to be located in promoter regions and because IDH1/2 mutations are associated with DNA hypermethylation at CpG islands but not at other genomic regions [26], we focused on the probes that were CpG island specific. Unsupervised hierarchical clustering of CpG islands specific probes resulted in three clusters. The first containing mutant IDH1/2 cell lines only, the second containing IDH1/2 wildtype cell lines and the third cluster contained a mixture of wildtype and mutant IDH1/2 cell lines ( Figure 2A).

Inhibition of mutant IDH1 does not affect viability or migration of mutant IDH1 cell lines
To assess the role of mutant IDH1 in chondrosarcoma cell proliferation, we incubated the cell lines with the IDH1 mutant inhibitor AGI-5198 for 72 hours. Indeed the levels of D-2-HG in culture media decreased in a dose-dependent manner in the mutant IDH1 cell lines JJ012 (R132G), HT1080 (R132C) and L835 (R132C) with IC50s of 0.7 µM, 0.5 µM and 0.35 µM, respectively ( Figure 3A). As expected, D-2-HG levels of the mutant IDH2 SW1353 cell line were not affected ( Figure 3A). Interestingly, even at doses that led to >90%  (A) Unsupervised hierarchical clustering for the top 2000 most differentially methylated CpG sites resulted in a wildtype cluster (blue samples), a mutation specific cluster (red samples) and a cluster with a mixture of wildtype and mutant IDH1/2 cell lines. (b) CpG islands are known to be located in promoter regions. As expected, methylation in the CpG islands was higher in the mutant IDH1/2 cell lines compared to the wildtype counterparts. decrease in D-2-HG (20 µM AGI-5198 for 72 hours), only a minor effect on cell viability using a WST-1 assay was seen (approximately 65% viability left in JJ012 after 72 hours)( Figure 3B). Furthermore, mutant IDH1 inhibition did not influence the colony forming capacity of the mutant IDH1 cell lines HT1080 and JJ012 and the mutant IDH2 cell line SW1353 after 10 days of treatment, whereas L835 was not able to form colonies in both the control and the AGI-5198 conditions ( Figure 3C). Since glioma cells transfected with mutant IDH1 need 20 passages to show an effect on methylation [33], we treated the mutant IDH1 cell lines JJ012, HT1080 and the IDH1/2 wildtype cell line CH2879 for up to 20 passages. Due to its slower growth rate, L835 was treated for up to 10 passages. As expected, treatment of CH2879 with 5mM D-2-HG for 10 passages showed a slight increase of intracellular levels of D-2-HG (10 times compared to control)( Figure 4A). We confirmed that after treatment with 1.5 µM AGI-5198 for 10 and 20 passages intracellular levels of D-2-HG were decreased in mutant IDH1 cell lines L835, HT1080 and JJ012 compared to untreated cells ( Figure 4B-4D). However, the xCelligence assay did not show any effect of AGI-5198 treatment on proliferation or migration ( Figure 4E-4F). Moreover, the ability of IDH1 mutant cells to migrate in 3D extracellular matrix scaffolds was not affected by treatment with 1.5 µM AGI-5198 (Supplementary Figure  2). Indeed, migration was not dependent on the IDH1/2 mutation status, since mutant IDH1 and mutant IDH2 cell lines as well as the IDH1/2 wildtype cell line showed migration in the timeframe of this experiment. long term inhibition of mutant IDH1 does not significantly affect CpG island methylation, histone trimethylation of H3K4, H3K9 and H3K27 or gene expression Since mutations in IDH1 are associated with inhibition of demethylases [21,27], we tested the effect of prolonged treatment with AGI-5198 and D-2-HG on trimethylation levels of H3K4, H3K9 and H3K27. The mutant IDH1 chondrosarcoma cell line L835, treated with 1.5 µM AGI-5198 for 10 passages, and the mutant IDH1 cell lines HT1080 and JJ012, treated for 20 passages, did not show any difference in lysine trimethylation of these specific histone marks ( Figure 5A).
Comparison of the mean overall methylation levels did not show any difference in global methylation levels in treated versus untreated cell lines ( Figure 5B). When focusing on the CpG islands specific probes, a slight decrease in overall CpG island specific methylation was seen ( Figure 5C). When assessing the treated and untreated pairs separately, demethylation of CpG island specific probes was predominantly seen in L835 treated for 10 passages ( Figure 5D).
Studying global gene expression using RNAseq revealed that prolonged inhibition of mutant IDH1 did not dramatically change global gene expression. Unsupervised hierarchical clustering showed that the expression profiles of the treated cell lines are more similar to their untreated counterparts, than to other treated cell lines ( Figure 5E). We identified 2 genes that were differentially expressed (FDR<0.05) after 10 passages of continuous treatment (Table 1), whereas after 20 passages, 10 genes were differentially expressed (FDR<0.05) ( Table 2), without any overlap between 10 and 20 passages of continuous treatment. Interestingly, gene expression in HT1080 is more stable than in L835 or JJ012, since after 10 or 20 passages of continuous treatment, genes with a FDR<0.05 are shown to be influenced exclusively in L835 and JJ012 (Tables 1 and 2). We chose COL1A2 and DIO2, based on their role in differentiation and chondrocyte biology, for further validation using qPCR [34]. Indeed, expression of DIO2 decreased upon inhibition of mutant IDH1 in L835 after 10 passages and in JJ012 after 20 passages of continuous mutant IDH1 inhibition ( Figure  6B). Expression of COL1A2 was very low in all cell lines as compared to osteoblastoma and normal growth plate controls (data not shown).
Analysis of gene expression using qPCR revealed that osteogenic and chondrogenic differentiation related genes (COL1A1, COL2A1, RUNX2, BGLAP, OPN, SPARC) were only very lowly expressed in the cell lines before and after treatment, in comparison to the normal cartilage, growth plate and osteoblastoma controls ( Figure  6A). The target genes of HIF1α (BNIP3, EGLN3, ENO1,     Figure 6C and 6D). Expression levels of genes involved in hedgehog signalling (IHH, SMO, PTCH1, GLI1, GLI2) were variable and no consistent change in expression was seen upon inhibition of mutant IDH1 (data not shown).

dIscussIon
The high prevalence of IDH1 or -2 mutations in enchondroma and chondrosarcoma, suggesting a causal rather than a bystander role in tumorigenesis, led us to investigate the function of these mutations in chondrosarcoma. Since glioma and leukemia cells harboring mutations in IDH1 or -2 cannot be maintained in culture, previous studies in these tumor types used artificial models to study the role of IDH1/2 mutations. We here use a unique panel of chondrosarcoma cell lines harboring endogenous IDH1/2 mutations that were retained during culturing [9,31]. We confirmed that all cell lines with an endogenous mutation in IDH1 or -2 expressed the mutant allele, which was associated with elevated intracellular and extracellular levels of D-2-HG. Our chondrosarcoma cell line panel contained one homozygous mutant IDH2 cell line (L2975) that, contradictory to what has been shown in the anaplastic astrocytoma WHO grade III cell line IMA [35], revealed elevated levels of D-2-HG comparable to heterozygous mutant IDH1/2 cell lines. Thus, our panel provides us with an excellent model to functionally study the consequences of IDH1 or -2 mutations and to evaluate whether inhibition of mutant IDH1 could be a potential treatment strategy.
To inhibit the mutant IDH1 protein we used the commercially available small molecule inhibitor AGI-5198, which is a potent and selective inhibitor of IDH1 with mutations on position R132 [36]. Inhibition of the mutant IDH1 protein indeed decreased the levels of D-2-HG in a dose-dependent manner. In addition to the previously reported IDH1 R132H and R132C mutations [37], we here show that D-2-HG levels of IDH1 R132G mutant cells can also be decreased using AGI-5198. The selectivity of AGI-5198 was confirmed in the mutant IDH2 cell line SW1353, where treatment did not affect D-2-HG levels. Although inhibition of mutant IDH1 decreased levels of D-2-HG up to 90%, only one out of three IDH1 mutant chondrosarcoma cell lines (JJ012) showed a minor effect on viability. In addition, prolonged treatment during 10 or 20 passages also did not affect proliferation or migration in monolayer cultures nor in a 3D collagen scaffold model. Moreover, the colony forming capacity of the mutant IDH1 cell lines after 10 days was not affected by treatment with 10 µM AGI-5198. These results are in contrast to results in glioma since Rohle et. al. showed that AGI-5198 impaired soft-agar colony formation in the transformed glioma cell line TS603. Moreover, an effect on colony formation after knockdown of IDH1 with shRNAs was shown [37]. A possible explanation for this difference might be that the shRNA targets both the wildtype and the mutant allele, which are expressed both as we show in Figure 1A. In addition, the difference may be explained by the difference in tumor types, or by a difference in model systems used (endogenous mutation versus artificially induced mutation). Our results suggest that mutations in IDH1/2, while being essential as an early event in the development of benign enchondroma, are not essential anymore after progression towards high grade chondrosarcoma. It is likely that other, additional genetic events occurred during tumor progression that have taken over driving chondrosarcoma cell proliferation.
Since the elevated levels of the oncometabolite D-2-HG caused by the mutation are reported to affect CpG island hypermethylation and histone methylation thereby affecting gene expression, as well as to interfere with HIF1α signaling, we next investigated the effect of inhibition of mutant IDH1 on these cellular processes in our cell lines. Mutant IDH1/2 causes hypermethylation of the genome via inhibition of α-ketoglutarate dependent enzymes in AML [27], gliomas [38], enchondromas [9] and chondrosarcomas [26]. Most studies focused on CpG island regions. In this study, we used the Illumina HumanMethylation 450k Beadchip array covering CpG islands, shores and shelves. Even though it is well known that prolonged culturing affects global methylation levels in cell lines [39,40], we were able to show a different methylome in IDH1/2 mutant and IDH1/2 wildtype chondrosarcoma cell lines. Interestingly, only CpG islands probes show hypermethylation in the chondrosarcoma cell lines with an IDH1/2 mutation, whereas the shores and shelves show hypermethylation in the IDH1/2 wildtype cell lines. When all 480 000 probes were included, comparison of the mean β-values between mutant IDH1/2 and IDH1/2 wildtype cell lines showed general hypermethylation of the IDH1/2 wildtype cell lines. Thus, even though cell lines have been in culture for a long time, the mutation in IDH1/2 is associated with increased levels of methylation at the CpG islands, and decreased levels of methylation at other areas in the genome. In TS603 glioma xenografts treated with AGI-5198 for two weeks Rohle et. al. showed no statistically significant change in the genome wide distribution of DNA methylation [37]. Likewise, we observed no significant change in gene expression nor in genome wide methylation distribution after continuous inhibition of mutant IDH1 during 10 or 20 passages. However, when focusing only on the CpG islands, methylation slightly decreased, an effect which was predominantly caused by alterations in L835. L835 is a slowly growing chondrosarcoma cell line that we previously generated in our lab [31], which has had a much lower number of passages in culture as compared to the widely used cell lines JJ012 and HT1080. Thus, L835 might resemble the primary tumor more than the cell lines that were cultured for a longer period of time. Indeed, it is therefore of interest that L835 is the only cell line demonstrating a decrease in CpG island specific methylation as well as a decrease in expression of HIF1α downstream targets GLUT1 and VEGF, upon inhibition of mutant IDH1. On the other hand, HT1080, formally derived from a "fibrosarcoma of bone" but in retrospect presumably representing dedifferentiated chondrosarcoma based on its IDH1 mutation, showed no alterations in gene expression, nor in CpG island methylation levels. Not surprisingly, this cell line does not express any of the differentiation markers either. It is tempting to speculate that HT1080 obtained many additional molecular alterations upon dedifferentiation (i.e. fibrosarcomatous change in conventional chondrosarcoma) and prolongation in culture, thereby rendering the IDH1 mutation nonessential.
In addition to downregulation of HIF1α downstream targets in L835 upon inhibition of mutant IDH1, RNAseq analysis revealed downregulation of the deiodinase iodothyronine type 2 (D2) gene DIO2, which was confirmed using qPCR. The gene product of DIO2 catalyzes the conversion of intracellular inactive thyroid hormone (T4) to active hormone (T3) [34]. T3 is involved in growth plate chondrocyte maturation [41,42]. This suggests that a mutation in IDH1/2 would upregulate DIO2, which would increase the levels of T3 thereby promoting chondrocyte maturation.
In summary, both short term as well as long term continuous inhibition of endogenous mutant IDH1 in three different chondrosarcoma cell lines show that although D-2-HG levels are decreased by the treatment, there is no significant effect on proliferation, migration (both in monolayer as well as in 3D), CpG island methylation or histone trimethylation of H3K4, H3K9 and H3K27. At the gene expression level we show inhibition of HIF1α target genes and of DIO2 in L835. The latter finding is novel and suggests a role of DIO2 and thyroid signaling in the development of enchondroma and chondrosarcoma, which would warrant further study. Our results suggest that while mutations in IDH1 or-2 are an early event in tumorigenesis, chondrosarcoma is not dependent on mutant IDH1 anymore. Thus, other events are likely to be involved in chondrosarcoma progression, in which the mutation in IDH1 or -2 no longer seems to be a driver mutation. Even though promising results are being reported for IDH1 inhibitors in early clinical trials in leukemia [43,44], our results indicate that monotherapy using mutant IDH1 inhibition may not be sufficient to use as a treatment option for patients with inoperable or metastatic central chondrosarcomas harboring this mutation.

Analysis of IDH1 and -2 mutation status
Both genomic DNA and cDNA were used for mutation analysis. DNA isolation was performed using the wizard genomic DNA purification kit (Promega, Madison, WI) according to the manufacturer's instructions. RNA was isolated using TRIzol (Invitrogen, cat nr 15596-018), and purified with the RNeasy mini kit (Qiagen, cat nr 74104). cDNA synthesis was performed with AMV-RT enzyme (Roche Diagnostics, Cat. nr. 109118), oligodTs and random hexamer primers were used. IDH1 and -2 mutation status was confirmed on DNA, as described previously [9]. PCR reactions were carried out in a total volume of 25 μl and contained either 100 ng DNA or 5 μl of 25x diluted cDNA (equivalent to 40 ng total RNA). After each PCR run, melting curves were inspected in order to check the formation of a single product. The PCR products were purified using the QIAquick PCR Purification Kit (Qiagen, Hilden, Germany) according to the manufacturer's manual and eluted in a final volume of 40 μl. Bidirectional Sanger sequencing was performed by Macrogen (Amstelveen, The Netherlands) using M13 general primers. The sequence results were evaluated using Mutation Surveyor software (Soft-Genetics).

d/l-2-hydroxyglutarate measurements
Extracellular D/L-2-hydroxyglutarate was measured using an aliquot (1 mL) of culture medium conditioned for 72 hours. Intracellular D/L-2-hydroxyglutarate was measured in cell lysates. The cells were washed with PBS and taken up in Hanks balanced salt solution (HBSS). The cells were lysed by three cycles of freeze-thawing in dry ice. To distinguish between D-2-HG and L-2-HG, samples were measured via LC-MS/MS as described elsewhere [50]. Intracellular D/L-2-HG measurements were normalized to the total protein concentration. For the extracellular D/L-2-HG measurements, the levels of D-2-HG were normalized to the amount of L-2-HG since the L-2-HG levels were constant for all cell lines. Protein concentrations were measured using the Biorad Dc Assay (Bio-Rad Laboratories B.V., Veenendaal, The Netherlands) as described in the manufacturer's manual.

Western blotting
Whole cell lysates were made using the HOT-SDS protocol, lysis buffer contained complete inhibitors and phosphoStop. Proteins were separated by SDS-PAGE and transferred to a polyvinylidene difluoride membrane by a wet blotting system according to manufacturer's manual (Bio-Rad Laboratories B.V., Veenendaal, The Netherlands). After blocking with 5% non-fat dry-milk, the primary antibody, H3K4me3 (Millipore, 07-473); H3K9me3 (Abcam, ab8898); H3K27me3 (Millipore, 07-449) was incubated overnight at 4°C, at dilutions of 1:1000, 1:2000 and 1:2000 respectively. Afterwards the membrane was washed with PBS/0,05% Tween-20, and the appropriate secondary antibody was incubated for 30 minutes at room temperature. After a final washing step, the membrane was developed using the ECL system (Thermo Scientific, #32209) according to the manufacturer's protocol.

Viability assay
Cells were seeded in 96 well plates and allowed to adhere overnight. For SW1353 and JJ012 5000 cells per well were seeded, for HT1080 and L835 10 000 cells and 15 000 cells, respectively. After 72 hours incubation with AGI-5198, a WST-1 assay (REF: 11 644 807 001, Roche) was performed according to the manufacturer's instruction and analyzed with the Victor3V, 1420 Multilabel plate reader (Perking Elmer, NL). Assays were performed in triplicate for three independent experiments. For long term treatment with AGI-5198 and D-2-HG, cells were treated for 10 or 20 passages with 1.5 µM AGI-5198. Medium with compounds or solvent controls were refreshed twice a week.

real time proliferation and migration assays
The RTCA xCelligence system (Acea, http://www. aceabio.com) was used to study either proliferation or migration in real time. Proliferation was monitored for 72 hours in E-plates. For the migration assay, SIM plates were used with the lower compartment filled with RPMI1640 with 20% FCS as a chemoattractant, cells were plated in the upper compartment in RPMI1640 only.

3d migration assay
Cells were pretreated for 72 hours with either mutant IDH1 inhibitor 10 µM AGI-5198 or DMSO, after which cells were trypsinized and collected. The migration assay was performed as described elsewhere [51]. Pictures were taken one hour, one day, two days and three days after injection of cells into the matrix. Pictures were taken using the Motic Moticam 3 CMOS 3.0MP Color Digital Camera and the motic images plus 2.0 ML Software. colony forming assay HT1080, JJ012, L835 and SW1353 cells were trypsinized and 100 and 1000 cells were seeded in a 6-well format. Cells were allowed to form colonies over 10 days and were either treated with DMSO or 10 µM AGI-5198 on day 0. Afterwards, cells were fixed and stained with 6.0% glutaraldehyde and 0.5% crystal violet.

Quantitative real time Pcr
PCR reactions were carried out and relative gene expression levels were normalized for DNA input using the housekeeping genes PPIA, CPSF6 and GPR108, as described previously [52]. We used Real Time-PCR to evaluate the effect of D-2-HG and mutant IDH1 inhibition on Indian hedgehog signaling (IHH, SMO, PTCH1, GLI1, GLI2), HIF1α signaling (GLUT1, BNIP3, EGLN3, ENO1, VEGF), and differentiation (COL1a1, COL2a1, RUNX2, BGLAP, OPN, SPARC). In addition, we evaluated the expression of COL1A2 and DIO2. PCRprimers are shown in Supplementary Table 1. Bio-Rad CFX Manager software was used to analyze data from the PCR experiments. As normal controls for Hedgehog signalling and cartilaginous differentiation we included RNA isolated from normal articular cartilage (n=3) and normal growth plate cartilage (n=3). For genes involved in osteogenic differentiation we included RNA derived from osteoblastoma (n=3). To give an overview of the expression levels of each of the genes and samples, a heatmap was made using the Gene-E tool (http://www. broadinstitute.org/cancer/software/GENE-E/index.html).

Methylation analysis
Genomic DNA was bisulfite-converted using the EZ DNA Methylation Gold Kit (Zymo Research) and used for microarray-based DNA methylation analysis, performed at ServiceXS (ServiceXS B.V., Leiden, The Netherlands) using the HumanMethylation450 BeadChip array (Illumina, Inc., San Diego, CA, U.S.A). This array interrogates over 450,000 CpG sites representing about 99% of the RefSeq genes. The bisulfite-converted DNA was processed and hybridized to the HumanMethylation450 BeadChip (Illumina, Inc.), according to the manufacturer's instructions. Analysis was performed using 'R project'. Samples were normalized using BMIQ [53]. next generation rnA sequencing analysis RNA was isolated using TRIzol (Invitrogen, cat nr 15596-018). High througput sequencing of mRNAs was performed as previously described in detail [54]. This procedure uses libraries obtained using the TruSeq RNA Sample Prep Kit v2 (Illumina) and indexes from the standard TruSeq Kits (12-Set A and 12-Set B) that were incorporated at the adaptor ligation step for multiplex sample loading on the flow cells. Purified and adaptermodified DNA fragments were enriched by 12 cycles of PCR using primers included in the Illumina Sample Prep Kit. The Agilent Bioanalyzer DNA 1000 chip was used for quality control to determine DNA concentrations and size distributions, followed by validation of the quantification using Qubit fluorometry (Invitrogen). Samples were examined as paired-end reads on an Illumina HiSeq 2000 platform using TruSeq SBS sequencing kit version 3, HCS v2.0.12 data collection software and Illumina's RTA version 1.17.21.3 for base-calling. RNA-Seq data were analyzed using a the MAPRSeq v.1.2.1 system and a data pipeline that includes alignment with TopHat 2.0.6 HTSeq software and EdgeR 2.6.2 [55][56][57]. Analyses of the RNA in R using the EdgeR package were performed using the negative binomial distribution. The dispersion factor was estimated in EdgeR using a weighted combination of both a gene-specific dispersion effect and a common dispersion effect calculated from all genes. To determine differentially expressed genes between the treated and untreated samples the threshold for false discovery rate (FDR) was set on 0.05. Hierarchical clustering of samples was performed using the Gene-E tool.

AcKnoWledGMents
The authors would like to thank René Zwartbol for technical assistance. We are grateful to Dr. J.A. Block (Rush University Medical Centre, Chicago, IL, USA), who kindly provided us with the JJ012 cell line, to Prof. A. Llombart Bosch (University of Valencia, Spain) for the cell line CH2879, Dr. Takashi Ariizumi (Niigata University Graduate School of Medical and Dental Sciences, Niigata, Japan) for providing the NDCS-1 cell line and M Namba (Okayama University Medical School, Shikata, Japan) for the OUMS27 cell line. The authors thank Dr. P. French for fruitful discussions. The study was financially supported by the Dutch Cancer Society (KWF Grant number UL 2013-6103).

conFlIcts oF Interest
None of the authors declare to have any conflict of interest with regards to the work presented in the manuscript.