Novel HSP90 Inhibitor NVP-HSP990 Targets Cell-Cycle Regulators to Ablate Olig2-Positive Glioma Tumor–Initiating Cells

Glioblastoma multiforme (GBM), the most common adult glioma, is associated with a dismal prognosis not only because of the high degree of genetic heterogeneity among patients and even within individual tumors but also because of its dynamic genetic instability. The most frequently altered genes are CDKN2A, TP53, EGFR, PTEN, and RB (1). Signal transduction pathways are not linear; they are complex, overlapping, and crosstalking, which may allow alternative pathways to compensate when one is disrupted, potentially leading to resistance to single agents that affect only one target (2). Simultaneously targeting multiple molecules that are deregulated is critical to designing a successful therapeutic strategy for GBM.

HSPs are a highly conserved family of molecular chaperones, which can be upregulated to protect cells from potentially lethal stress. Upregulated HSPs may partially account for glioma cell's ability to survive the otherwise fatal hypoxic environment and tolerate genetic alterations (3). HSP90 is induced in response to cellular stress and stabilizes client proteins involved in cell-cycle control and proliferation/antiapoptotic signaling (4). By binding and chaperoning proteins, HSP90 can buffer the genetic variation at the protein level (5). Many of HSP90′s more than 100 client proteins, including P53, CDK4, ErbB2, PI3K, PTEN, AKT, Raf, c-MET, and EGFR, are reported to be involved in the major aberrant signal transduction pathways identified by The Cancer Genome Atlas (TCGA; refs. 6–9). An advantage of HSP90 inhibitors is their ability to affect multiple oncoproteins simultaneously, including targets considered “undruggable,” and thus they may largely avoid generating resistant phenotypes arising from mutation, activation of alternative signaling pathways, or feedback loops seen with therapeutics targeting a single oncogene or pathway (10).

Tumor-initiating cells are functionally defined through their capacity for sustained self-renewal and tumorigenicity. Glioma tumor–initiating cells (GIC) retain relevant molecular features of GBMs and enable preclinical models for evaluation of both tumor biology and therapeutics (11). GIC maintenance is regulated by an interconnected regulatory circuit consisting of many HSP90 client proteins including AKT and STAT3 (12, 13). Because HSP90 is involved in redundant pathways for maintaining cell viability, its inhibition has the potential to block expression of multiple client proteins involved in tumorigenesis. Therefore, there seems to be a compelling rationale to evaluate an HSP90 inhibitor in GIC models. The aim of the current study therefore was to evaluate the effects and mechanism of an HSP90 inhibitor in GICs both in vitro and in vivo.

The GIC lines were established by isolating neurosphere-forming cells from surgical specimens of human GBM using a method described previously (14, 15) The study was approved by the Institutional Review Board of M. D. Anderson Cancer Center, and informed consent was obtained from all subjects. These GIC lines were cultured as GBM neurospheres in DMEM/F12 medium containing B27 supplement (Invitrogen) and bFGF and EGF (20 ng/mL each). 17-N-allylamino-17-demethoxygeldanamycin (17-AAG) and retinoic acid (RA) were obtained from Sigma. The HSP90 inhibitor NVP-HSP990 was provided by Novartis. NVP-HSP990 was dissolved in dimethyl sulfoxide to a concentration of 10 mmol/L. Normal human astrocyte (NHA) was maintained in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% FBS. Human fetal brain neural stem cell line HFB2050 was kindly provided by Dr. Evan Snyder (Burnham Institute, La Jolla, CA) and cultured as described previously (15).

Cells were seeded in 96-well plates (2 × 10³ cells/well) and incubated at 37°C for 24 hours before serial dilutions of NVP-HSP990 were added. Growth inhibition was determined using the CellTiter-Blue cell viability assay (Promega). Cell viability in the vehicle control was considered as to be 100%. The IC₅₀ value was calculated by using Calcusyn version 2.0 software (Biosoft) as the mean drug concentration required to inhibit cell proliferation by 50% compared with vehicle-treated controls.

GIC tumorspheres were dissociated using Accutase enzyme cocktails (Invitrogen) and plated at a density of 2 × 10⁵ cells per well in 60 mm plates. The cells were treated with NVP-HSP990 for 24 hours, fixed in 70% ethanol, and stored at −20°C. Cells were stained with propidium iodide for cell-cycle analysis using a BD FACSCalibur flow cytometer and CellQuest software (BD Biosciences).

The immunoprecipitation procedure was conducted as described previously (16). Briefly, after NVP-HSP990 (50 nmol/L) or 17-AAG (500 nmol/L) treatment for 2 hours, cells were lysed in immunoprecipitation lysis buffer, and precleared by incubation with poly(G) immunoprecipitation beads for 1 hour at 4°C. The pellet was discarded, and the supernatant (200 μg of lysate protein) was subjected to immunoprecipitation.

Lentiviral vectors encoding Olig2 or short hairpin RNA (shRNA) for Olig2 was purchased from Genecopoie Inc. Sequences of shRNA hairpin are listed in Supplementary Data. The VSV-G pseudotyped lentiviral vectors were produced by transient cotransfection of 3 plasmids into 293T cells and produced lentiviruses stock was titered and stored at −80°C.

Cells were harvested in a lysis solution as previously described (17) and subjected to Western blotting. Membranes were probed with the following primary antibodies: anti-HSP90, anti-HSP70, anti-HSP27, anti-PDGFRα, anti-AKT, anti-CDK4, anti-CDK2, anti-cleaved-caspase-7, anti-cleaved-caspase-8, anti-Bim, anti-phosphop-Rb (Ser807/811), and anti-p27 (all from Cell Signaling), Olig2 antibody was purchased from Chemicon, and β-actin antibody was from Sigma-Aldrich.

Dissociated GICs were plated at 10 cells/μL in 6-well plates and incubated with various concentrations of NVP-HSP990 for 7 days. Formed tumorspheres were dissociated into single cells and counted with hemocytometer using 0.2% Trypan blue exclusion.

Immunofluorescence staining was conducted as described previously (18). Briefly, GICs were grown on chamber slides precoated with poly-lysine–coated coverslips, cells were fixed with 4% paraformaldehyde, permeated with 0.25% Triton X-100, and blocked with 3% normal goat serum. Immunostaining was conducted using the appropriate primary and secondary antibodies, and images were acquired using a confocal laser scanning microscope (Carl Zeiss Microscopy).

Immunohistochemistry was conducted as described previously (19). Briefly, after dewaxing and antigen retrieval, tissue sections were incubated with appropriate primary and secondary antibodies. Sections were stained 3,3-diaminobenzidine and hydrogen peroxide chromogen substrate (DAKO) and counterstained with hematoxylin and mounted. Images were acquired using a Zeiss Microsystems fluorescence microscope linked to a DFC340FX camera (Carl Zeiss Microscopy).

Apoptosis was determined using the terminal deoxynucleotidyltransferase–mediated deoxyuridine triphosphate nick end labeling (TUNEL) method using in situ cell death detection reagent (Roche Applied Science). The percentage of TUNEL- labeled cells in each section was determined at a magnification of 400 by counting 500 cells in a randomly selected field.

After NVP-HSP990 (50 nmol/L) treatment for 24 hours, cells were collected and lysed in a buffer consisting of a 2.5% solution of 2-mercaptoethanol in loading buffer/T-PER (Pierce) plus phosphatase and protease inhibitors. All samples were diluted to a final concentration of 1 mg/mL, and then 30 μL of each sample, arrayed in a series of dilutions, was printed in duplicate on slides. The slides were then subjected to immunostaining with a panel of 207 commercially available antibodies (20). Slides were stained on an automated slide strainer (DAKO) using biotin-linked peroxidase catalyzed signal amplification.

All animal studies were conducted in the veterinary facilities of M.D Anderson Cancer Center (Houston, TX) in accordance with institutional rules. The antitumor efficacy of NVP-HSP990 was examined in intracranial xenografts derived from 2 GIC lines, GSC11 and GSC20. To create the intracranial disease model, we engrafted GICs (5 × 10⁵) into the caudate nucleus of Nude (nu/nu) 6- to 8-week-old mice using a previously described guide-screw system (21) and then randomly divided the mice into 4 groups of 14 mice in each group. Starting on day 4 after the tumor cells implantation, mice were treated by oral gavages with 10 mg/kg NVP-HSP990 in methylcellulose or with an equal volume of methylcellulose alone (vehicle control) on a schedule of once per week for 5 weeks. After NVP-HSP990 treatment for 2 or 4 weeks, 2 animals from each group were euthanized for biologic assessment of tumor response.

Statistical analysis was conducted with using the Student unpaired t test. Results are presented as the mean ± SD of at least 3 independent experiments. Survival analysis was conducted using the log-rank analysis module in SPSS 10.0 (SPSS Inc). Differences were considered significant at P < 0.05 for all the statistical analysis conducted in our study.

Examination of some known HSP90 client proteins, including PDGFRα, CDK4, and AKT, revealed a dose- and time-dependent decrease in GIC line GSC11, which is consistent with the idea that the responsiveness of these cells to NVP-HSP990 is a results of the degradation of HSP90 client proteins (Fig. 1A and B). GIC lines such as GSC13, GSC16, and GSC23 also showed similar changes after NVP-HSP990 treatment (Supplementary Fig. S1A). We also observed increased HSP70 and HSP27 levels following NVP-HSP990 treatment.

We tested the effects of NVP-HSP990 against a panel of 14 GIC lines. NVP-HSP990 inhibited GIC proliferation in all GIC lines, with IC₅₀ values ranging approximately between 10 and 500 nmol/L (Fig. 1C). Furthermore, we observed a differential response in these GIC lines and divided them into responder and nonresponder groups (Fig. 1D). All GIC lines in responder group exhibited high sensitivity to NVP-HSP990 with IC₅₀ values less than 60 mol/L. Although the levels of HSP90 protein expression was similar in responder and nonresponder GICs, the IC₅₀ values in responder GIC lines were significantly lower than those in nonresponder group (Supplementary Fig. S1B and Fig. 1D; P = 0.0002). Treatment of responder GIC with NVP-HSP990 for an extended time from 3 to 5 days did not seem to produce more cytotoxicity on responder GICs (Supplementary Fig. S1C). We also evaluated the effect of NVP-HSP990 on NHAs and human fetal brain neural stem cells (HFB2050). The IC₅₀ values were much higher than responder GICs (NHA, 225.2 ± 28.5 nmol/L; HFB2050, 132.4 ± 20.3 nmol/L).

We selected 2 GIC lines in the responder group (GSC11 and GSC13) and 2 cell lines in the nonresponder group (GSC2 and GSC20) to study NVP-HSP990 effect on cell proliferation and differentiation. Tumorspheres that formed from NVP-HSP990–treated GICs were smaller and fewer than those formed from control untreated GSC11 cells (Fig. 2A). Treatment with NVP-HSP990 markedly decreased the proliferation of responder GICs (Fig. 2B). In addition, NVP-HSP990 treatment markedly impaired the proliferative potential of responder GICs, as shown by the decreased positivity of Ki-67 staining in treated cells (Fig. 2C). Limiting dilution assay also showed that NVP-HSP990 attenuated the self-renewal potential in GSC11 cells (Supplementary Fig. S2A). Thus, our data indicated that inhibition of HSP90 by NVP-HSP990 is associated with attenuation of GIC proliferation at the functional levels.

To further determine the effect of inhibiting HSP90 on GIC differentiation, exposure of GICs to either differentiation condition solution (1% FBS + 1 μmol/L retinoic acid in DMEM/F12) or to NVP-HSP990 (50 nmol/L) for 5 days differentiated GICs into 3 neural lineages, as GFAP, TuJ1, NeuN, and CNPase-stainined cells increased following induction by 1% FBS + 1 μmol/L retinoic acid. Surprisingly, NVP-HSP990 treatment only considerably increased neuronal marker TuJ1 and NeuN expression and did not alter GFAP and CNPase in responder GICs, suggesting that NVP-HSP990 mainly triggered neuronal lineage-restricted differentiation (Fig. 2D). Consistently, the expression of stem cell related genes such as CD133, Nestin, Sox2, and Olig2 decreased following NVP-HSP990 treatment (Supplementary Fig. S2B). However, the nonresponder cells (GSC2 and GSC20) did not show much staining of TuJ1 and NeuN after NVP-HSP990 treatment (Fig. 2E). Nevertheless, RA-differentiated GICs seemed less sensitive to NVP-HSP990 treatment, as the IC₅₀ value of the RA-differentiated GICs were higher than GICs grown in stem cell medium (Supplementary Fig. S2C).

To correlate the drug response with the lineage characteristics of GICs, immunostaining on 13 lineage markers showed that all responder GIC lines exhibited high Olig2 expression, also shown by Western blotting analysis (Fig. 3B), as compared with nonresponder GIC lines (Fig. 3A). Gene expression analysis also confirmed that Olig2 is the top marker associated with response to NVP-HSP990 (Supplementary Fig. S3). Treatment of responder GIC line, GSC11, with HSP990 (50 nmol/L) for different times points showed that Olig2 protein levels markedly decreased after 24 hours of treatment and disappeared after 72 hours (Fig. 3C) accompanied with increased NeuN staining (Fig. 3D).

Olig2 has a well-established function in regulating the proliferation of neural progenitors and glioma cells. In our study, we showed that Olig2-high glioma cells displayed high Ki-67 labeling in GSC11 cell line (Fig. 4A). To further evaluate the role of Olig2 on GIC proliferation, Olig2 knockdown GICs and cell lines ectopically expressing Olig2 in GSC23 were generated. Olig2-high GSC11 and GSC13 cells showed slower cell proliferation rates following Olig2 knockdown (Fig. 4B and C) and GSC23-expressing ectopic Olig2 showed increased cell proliferation (Fig. 4D and E) indicating Olig2 might regulate the proliferation of GICs.

The observation that NVP-HSP990 attenuates the cell proliferation of GICs with high Olig2 prompted us to hypothesize that Olig2 might be a functional molecule associated with response to NVP-HSP990. To confirm this hypothesis, we tested the effect of NVP-HSP990 on various cell models manipulating Olig2 expression. Our results showed that Olig2-knockdown GIC lines (GSC11/shOlig2 and GSC13/shOlig2) were less sensitive to NVP-HSP990 treatment as compared with GSC11/shscr and GSC13/shscr (P < 0.05; Fig. 4F). Consistent with this finding, we showed that Ki-67 index in the Olig2 knockdown GICs is lower than shscr, whereas it decreased slightly, as compared with shscr control following NVP-HSP990 treatment (Fig. 4G). In parallel, overexpression of Olig2 in GSC23-a low Olig2-expressing cell line sensitized GSC23 to NVP-HSP990 treatment, as compared with GFP control (Fig. 4H, P < 0.05). These results suggested that Olig2 might be a functional marker associated with cell proliferation and response to NVP-HSP990.

We analyzed the alteration of various signaling components by hierarchical clustering analysis of the RPPA data from 10 GIC lines (4 responder cell lines and 6 nonresponder cell lines) treated with NVP-HSP990 by comparing patterns of protein expression in the responder group versus the nonresponder group to identify altered signaling events associated with response to NVP-HSP990. We identified a series of cell-cycle regulatory proteins in responder GIC lines that showed changes, specifically an increase in p27, and a decrease in phosphorylated Rb (Ser 807/811), and appreciable induction of proapoptotic proteins, including cleaved caspase-7/8, and Bim. Such changes were not observed in nonresponder GIC lines (Fig. 5A). Western blot analysis validated the RPPA results as shown by reduced expression of phosphorylated Rb and increased p27 levels in GSC11 cells and increased in the apoptosis-related molecules cleaved caspase-7/8 and Bim in GSC11 after NVP-HSP990 treatment (Fig. 5B).

Flow cytometry analysis showed that NVP-HSP90 caused a dramatic depletion of cells in the S-phase in 2 of the responder GIC lines (P < 0.001; Fig. 5C). We did not observe significant change in the proportion of cells in S-phase in nonresponder GIC lines (P > 0.05). TUNEL staining confirmed that the NVP-HSP990 induced apoptosis in both GSC11 and GSC13 cell lines, whereas minimal induction of apoptosis was observed in the nonresponder GIC lines GSC2 and GSC20 (Fig. 5D). The combined treatment of NVP-HSP990 with chemotherapeutic drugs etoposide and topotecan (known apoptosis inducer) showed enhanced TUNEL labeling, implicating a potential combinational strategy to eliminate GICs (Supplementary Fig. S4). These results suggested that the disruption of cell-cycle regulation and the induction of apoptosis might be crucial for NVP-HSP990′s effect on responder GICs.

Because CDK2 and CDK4 were previously reported to be the major kinases for Rb phosphorylation at S807/S811 and to be client proteins for HSP90 (22–24), we tested the possibility that the alteration of cell-cycle–related molecules by NVP-HSP990 might result from the attenuation of CDK2 and CDK4 expression. Coimmunoprecipitation assay showed that HSP90 was associated with both CDK2 and CDK4 in GSC11 and GSC13 cell lysate (Fig. 6A) and inhibition of HSP90 by either NVP-HSP990 (50 nmol/L, 2 hours) or the reference inhibitor 17-AAG (500 nmol/L, 2 hours) disrupted the interactions between HSP90 and CDK2/4 (Fig. 6B).

To further investigate the function of CDK2 and CDK4 in the stem cell maintenance, treatment with flavopiridol (FLP: a selective inhibitor of CDK2 and CDK4) decreased phosphorylated Rb levels and increased p27 expression (Fig. 6C). Similar to NVP-HSP990, FLP treatment showed similar distribution of responder and nonresponder GIC lines in cytotoxicity assay (Fig. 6D), a marked reduction of Ki-67 index (Fig. 6E), and induction of apoptosis in responder GIC lines GSC11 and GSC13 (Fig. 6F). Therefore, NVP-HSP990 might promote the protein degradation of CDK2 and CDK4 and consequently impair GIC proliferation.

We showed that NVP-HSP990 affects both Olig2 and CDK2/4 function in regulating GIC proliferation, and HSP90 does not directly interact with Olig2 protein. Therefore, we hypothesized that HSP90 client protein CDK2/4 might interact with Olig2 to regulate its function. Coimmunoprecipitation assays showed that endogenous Olig2 physiologically interacts with CDK2/4 in GSC11 cells (Fig. 7A) and treatment with FLP disrupted this interaction between CDK2/4 and Olig2 in Olig2-transfected 293T cells suggesting that CDK2/4 activities might be crucial to maintain this interaction between CDK2/4 and Olig2 (Fig. 7B).

To show that interaction of CDK2/4 with Olig2 might regulate Olig2 expression and more importantly Olig2 protein stability, we measured the protein stability of Olig2 protein following CDK2/4 inhibition after protein synthesis was blocked by cycloheximide (CHX). Results show that Olig2 levels were decreased more rapidly in FLP-treated GSC11 cells, as Olig2 protein almost disappeared after FLP treatment for 2 hours (Fig. 7C). Time course treatment with FLP in GSC11 cells showed that Olig2 protein dramatically decreased after 12-hour treatment and completely disappeared after 24 hours. This decrease in Olig2 was also accompanied with downregulated p-Rb and increased p27 (Fig. 7D). Therefore, NVP-HSP990 might affect Olig2 protein expression through attenuation of CDK2/4 activities in GICs.

Therapeutic efficacy of NVP-HSP990 (10 mg/kg) was shown by increase in median survival times to 58 days (95% confidence interval: 52–64 days) in comparison with 46 (95% confidence interval: 45–47 days) days in the vehicle control group for GSC11 (P < 0.01; Fig. 8B, Supplementary Table S1). However, NVP-HSP990 treatment had no effect on the median survival of GSC20 xenografted animals (P = 0.4897). Histopathologic staining revealed that NVP-HSP990 considerably reduced the tumor growth in GSC11 xenografts, as shown by decreased tumor mass in animals treated for 4 weeks with NVP-HSP990 in comparison with vehicle control. However, the tumor mass in GSC20 xenograft showed mild reduction following NVP-HSP990 treatment (Fig. 8A).

The orthotopic tumors after NVP-HSP990 treatment for 4 weeks were analyzed for biologic effects. The levels of HSP90 client proteins CDK2 and CDK4 were lower in NVP-HSP990–treated GSC11 tumors than in untreated tumors, suggesting NVP-HSP990 suppressed HSP90 activities in vivo. In addition, the treated tumors showed decreased Ki-67 positivity and increased cell TUNEL labeling (Fig. 9 and Supplementary Fig. S5) indicating that NVP-HSP990 suppressed proliferation and induced apoptosis in the responder GIC tumors. Furthermore, reduced Olig2 expression was accompanied by increased p27 and TuJ1 positivity after NVP-HSP990 treatment.

We have shown that NVP-HSP990, a HSP90 inhibitor, suppresses the oncogenic properties of a subgroup of high Olig2 expressing GICs with respect to cell proliferation, induces neuronal differentiation, and affects survival in mice. Furthermore, our results indicate that NVP-HSP990 treatment in GICs disrupts cell-cycle control via the CDK2/CDK4 pathway by regulating the protein degradation of these cell-cycle regulators.

NVP-HSP990 is highly potent and selective for HSP90 and represents one of the most potent oral HSP90 inhibitors reported. NVP-HSP990 also displayed significant HSP90-dependent antitumor activity in glioma cell lines by interacting with HSP90 and leading to the destabilization and degradation of several proteins that depend on HSP90 for their stability (25, 26). Furthermore, NVP-HSP990 exhibited promising drug selectivity on responder GICs, as IC₅₀ values of responder GIC lines is much less than those of astrocytes or neural stem cells. NVP-HSP990 significantly impaired GIC proliferation and triggered neuronal differentiation of the GICs, confirming that HSP90 plays a role in maintaining GICs biologic properties.

GICs cultured under neural stem cell conditions can display heterogeneous biologic characteristics and lineage profiles that reflect the heterogeneous features of GBMs (27). Olig2, a bHLH transcription factor, shows both antineural functions and proneural functions in oligodendrocyte progenitors (28, 29, 30). The responder GIC lines highly expressed Olig2, whereas NVP-HSP990 treatment reduced its expression, suggesting NVP-HSP990 might preferentially target GICs that might originate from early neural progenitors such as oligodendrocyte progenitors. More importantly, Olig2 might play a crucial role in promoting the proliferation of neural stem/progenitor cells and malignant glioma. Ligon and colleaguesindicated that Olig2-regulated lineage-restricted pathway controls replication competence in neural stem cells and malignant glioma (31). Recent study also showed that acquisition of Olig2 in the Id1^high GICs population is a critical factor in its tumorigenic potential (32). Consistently, we showed that Olig2 might regulate proliferation of GICs, as Olig2-high GICs exhibit higher Ki-67 index. Knockdown of Olig2 slowed the GIC proliferation, whereas overexpression of Olig2 promotes cell proliferation in Olig2-low GICs. Therefore, Olig2 might be a functional marker in maintaining the proliferative/rapid cycling state of GICs, which might render them susceptible to drugs that disrupt the cell-cycle progression.

Cell-cycle regulators are inserted in complex networks whose alteration might underlie tumorigenesis in the nervous system (33). Accumulating evidence indicates that the regulatory mechanisms in the G₁ phase in neural precursors play an important role in the precursors' decision to proliferate or differentiate and that factors modulating G₁ might be used to influence this decision (34). CDK2 and CDK4 are the master regulators controlling G₁–S progression, which they do by phosphorylating the Rb protein. CDK2 is also critical for the proliferation and self-renewal of neural progenitor cells in the adult subventricular zone (35, 36). In this study, we showed that CDK2 and CDK4 are complexed with HSP90 in GICs; NVP-HSP990 attenuates CDK2 and CDK4 protein levels, consequently decrease cell populations in the S-phase, and impair cell proliferation. Furthermore, our data indeed showed that similar to NVP-HSP990, Olig2-high GIC lines are more sensitive to CDK2/4 inhibitor flavopiridol. Given the potential role of Olig2 as a functional marker in maintaining the rapid cycling state of GICs, disrupting cell-cycle regulation by targeting CDK2 and CDK4 (NVP-HSP990 or flavopiridol) might help to kill GICs with high Olig2 expression.

The mechanism that regulates CDK2/4 and Olig2 function is unknown. Our findings indicate that CDK2/4 can interact with Olig2 in GICs. Nevertheless, CDK2/4 activities are crucial to maintain Olig2 stability, as CDK2/4 inhibition by Flavopiridol disrupts the CDK2/4/Olig2 interaction and destabilize Olig2 protein. The above findings might implicate a mechanism through which HSP90 regulate cell proliferation in Olig2-high GICs. Although the detailed mechanisms still remains to be understood regarding how CDK2/4 interacts with Olig2.

One challenge for molecular inhibitor development is the selection of cancer patients who would benefit from this treatment. TCGA described a robust gene expression-based molecular classification of GBMs, with distinct gene expression profiles in GBM subclasses (37). Establishing the association between drug responses and GBM molecular subclasses may help to identify potential cohorts of patients for targeted therapy. In this study, we identified proneural gene Olig2 as a functional marker to predict the GICs' response to NVP-HSP990. These data suggest that NVP-HSP990 would be effective in targeting as well as eventually minimizing recurrence by inhibiting the proliferation of GICs. Results of this study are therefore anticipated to help identify potential GBM patient cohorts who would gain the most from HSP90 inhibition and could also be potentially valuable in designing future combinational strategy to target the subset of GICs with high Olig2 expression.

W.K.A. Yung has a commercial research grant from Novartis; has a honoraria from speakers' bureau from Novartis, Merck, and Actelion; and is a consultant/advisory board member of Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Fu, D. Koul, W.K.A. Yung

Development of methodology: J. Fu, H. Colman, W.K.A. Yung

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D. Koul, E.P. Sulman, F.F. Lang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Koul, J. Yao, Y. Yuan, W.K.A. Yung

Writing, review, and/or revision of the manuscript: D. Koul, Y. Yuan, H. Colman, W.K.A. Yung

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Wang, W.K.A. Yung

Study supervision: D. Koul, W.K.A. Yung

The authors thank Verlene Henry and Lindsay Holmes for performing the animal studies and Kathryn Carnes (Department of Scientific Publications, The University of Texas M. D. Anderson Cancer Center) for editing the manuscript.

This study was supported by grants from the National Cancer Institute (CA56041 and CA127001 to W.K.A. Yung), Cancer Center Support Grant (CA16672) and a sponsored research grant from Novartis.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.