Mechanisms of Resistance to EGFR Inhibition Reveal Metabolic Vulnerabilities in Human GBM

This article is featured in Highlights of This Issue, p. 1473

Glioblastoma (GBM), a malignant brain cancer, is characterized by diffuse invasion of tumor cells into the adjacent brain parenchyma and by the increased activity of receptor tyrosine kinase (RTK) signaling pathways. The most commonly altered RTK, the epidermal growth factor receptor (EGFR), demonstrates gene amplification in up to 45% of patients (1) and expression of a constitutively active EGFR variant, EGFRvIII, in approximately 19% (2–4). Despite the high frequency of EGFR alterations in GBM and their demonstrated roles in oncogenesis (5, 6), chemotherapeutic modalities targeting EGFR signaling have had disappointing results in clinical trials, and have not increased the median survival for GBM patients, which remains at less than 2 years (7, 8).

Several factors contribute to therapeutic resistance to EGFR inhibition (9). Such factors include: some EGFR inhibitors are suboptimal in their receptor binding pocket occupancy (10); limited brain access of systemically administered drugs, especially to invading tumor cells outside of the enhancing tumor mass; tumor cell utilization of mechanisms that increase drug metabolism and detoxification, including expression of efflux pumps (11, 12); and the molecularly heterogeneous nature of GBM, whose aggressive biology is driven by the summation of multiple signaling inputs (13, 14). Thus, and despite effective on-target inhibition, tumor cells appear able to reestablish downstream PI3K/Akt/mTOR signaling pathway activity via suppression of key negative regulators and the activation of other RTKs (15–17), including MET or platelet-derived growth factor receptor β (18, 19). In general, both acquired and innate resistance mechanisms contribute to therapeutic resistance.

Sustained proliferation, as observed in malignant tumors such as GBM, requires changes in cellular metabolism that support rapid cell membrane turnover and the prevention of oxidative damage (20). EGFR signaling itself can increase the activity of such protective pathways, including de novo lipogenesis (21, 22) and protection against oxidative stress-induced apoptosis (23). In this study, we have investigated relationships between EGFR activity and tumor protective pathways, and how maintaining these protective pathways is associated with resistance to EGFR inhibitor therapy. In so doing we have identified potential points of therapeutic vulnerability that may be relevant to improving GBM treatment outcomes.

EGFRvIII-expressing murine tumorspheres were isolated as previously described (24). GBM43 and GBM6 are primary cell cultures of patient-derived xenografts (PDX), established as previously described (25). All cell lines were analyzed before use and each subsequent year by short tandem repeat analysis to ensure the identity and validity of cells and confirmed to be negative by PCR for Mycoplasma contamination. Tumorspheres were cultured and passaged on low-attachment plates under standard neurosphere conditions as previously described (26). Doubling time was calculated using an online calculator (Roth V. 2006 Doubling Time Computing: http://www.doubling-time.com/compute.php). For Western blotting the following antibodies were used: phospho-EGFR Y1173 (Cell Signaling Technology; 4407), EGFR (Santa Cruz; sc-03), phospho-Akt S473 (Cell Signaling Technology; 4060), Akt (Cell Signaling Technology; 9272), phospho-p44/42 MAPK T202/Y204 (Cell Signaling Technology; 9101), p44/42 MAPK (Cell Signaling Technology; 9107), phospho-Src Family Y416 (Cell Signaling Technology; 2101), Src (Cell Signaling Technology; 2109), ALDH1A1 (Cell Signaling Technology; 12035), and GAPDH (Millipore MAB374).

Erlotinib HCl (Selleckchem S1023), Mitomycin C (Santa Cruz sc-3514), dasatinib (LC Laboratories D3307), lapatinib (LC Laboratories L4804), DEAB (Sigma-Aldrich D86256), Tyrphostin AG1478 (Sigma-Aldrich T4182; ref. 27), and disulfiram (Sigma-Aldrich T1132) were used.

For MTT assay, dissociated cells were plated as 4 × 10⁴ to 8 × 10⁴ cells per well in a 6-well plate with and without appropriate drug treatments. Cells were incubated for 3 days and assayed by MTT (Promega) or DyLight 800 NHS Ester and flow cytometry.

We analyzed paired sets of surgical specimens from patients with GBM before and after treatment with an EGFR inhibitor from three different institutions—the University of California, San Francisco (UCSF), the University of California, Los Angeles, and from a clinical trial cohort (15)—for which tissue samples were acquired before or under treatment, 5 days. Cases were selected based on the availability of paired tumor specimens from before and after EGFR inhibitor therapy and EGFR amplification. Unpaired human GBM from initial presentation were obtained through the Brain Tumor Research Center Biorepository at UCSF. All tumor specimens were obtained according to protocols approved by the respective institutional review boards.

Tumor cells were transplanted orthotopically into the striatum of nude mice (Simonsen) as previously described (24). To assess invasion following EGFR inhibition, tumors were allowed to propagate for 7 days prior to 5 days of treatment with erlotinib (150 mg/kg per day via oral gavage). To assess tumor area in the clones, tumors were allowed to propagate for 2 days prior to 10 days of treatment daily with erlotinib or erlotinib plus disulfiram (100 mg/kg and 150 mg/kg, respectively, via intraperitoneal injection). In both experiments, brains were examined at 12 days after transplant or at humane disease endpoint, if sooner. Control vehicle–treated mice were included in each experiment. An image of each tumor was taken at a similar anterior–posterior position using a Zeiss Observer inverted microscope at 1.25× and 5× magnification. To assess invasion, results are maximum distance invaded, as measured by the longest line between 2 tumor foci, normalized to total tumor area, calculated using ImageJ software. Tumor area was determined by hand tracing total intracerebral tumor area blinded to tumor cell and treatment arm, calculated using ImageJ software. All animal procedures and care were performed according to institutional animal care and use policies (IACUC #AN169893-02).

Two-dimensional invasion assays were performed as previously described (28). Six images of invaded cells per insert were taken at 200× magnification on a Zeiss Axioimager MI microscope. The total number of invaded cells at 16 hours per image were counted using ImageJ software and averaged for each insert. Cells were pretreated for 2.5 days in 1 μmol/L erlotinib and remained exposed to 1 μmol/L of erlotinib during the invasion assay. Data were normalized to control in each experiment. Three-dimensional invasion assays were performed as previously described (28). Images of each sphere were taken at 0, 16, and 24 hours using a Zeiss Observer inverted microscope at 25× magnification. The total area of each sphere was calculated using ImageJ software, and invaded area for each sphere was normalized to its center sphere area. Appropriate drug concentrations were added both to the underlying matrigel coat and to the sphere containing media.

Tumor-bearing mice were euthanized, perfused with 4% paraformaldehyde, and brains were fixed overnight in 4% PFA, rinsed in PBS and stored in 70% ethanol until further processing. Human GBM formalin-fixed paraffin-embedded sections and tissue microarrays were immunostained for ALDH1A1 (Abcam; ab52492). IHC was performed according to standard protocols on the Ventana Medical Systems Benchmark XT. Images were taken using an Olympus BX41 microscope and Olympus DP72 camera. Immunostaining was quantified using ImageJ software, and percentage of DAB positive for each tumor was averaged across 2 images per tumor. Histologic analysis was performed on hematoxylin and eosin–stained sections. For immunofluorescence, 2 × 10³ cells were plated into each well of a Biocoat Poly-D-lysine 8 Well-Culture Slide (Corning 354688) for 3 days, followed by 2 hours with appropriate drug treatments. Cells were stained with 1:50 dilution of 4-hydroxynonenal antibody (Abcam; ab46545). Cells were then stained with 4 μg/mL goat anti-rabbit AlexaFluor 488 IgG (Gibco A11008) and nuclei labeled with DAPI. Images were taken using the Zeiss LSM780 Confocal Microscope. The total number of cells and number of 4HNE-positive cells per image were counted using ImageJ software and averaged across 3 pictures per well.

The ALDEFLUOR assay measures the conversion of the ALDEFLUOR substrate BODIPY aminoacetaldehyde (BAAA) to the negatively charged BODIPY aminoacetate (BAA-; ref. 29). The ALDEFLUOR kit (STEMCELL Technologies 01700) was used as described to measure proportions of ALDH⁺ cells. Cells were ALDEFLUOR stained at a concentration of 5 × 10⁵ cells/mL for 45 minutes at 37°C, followed by DyLight 800 NHS Ester (Life Technologies, cat. #46421) staining for 15 minutes at room temperature as a live/dead marker. DEAB or disulfiram was added to samples in ALDEFLUOR buffer for 10 minutes at room temperature before ALDEFLUOR staining. Flow cytometry was performed on a BD FACSAria using FACSDiva software, followed by analysis with FlowJo. A minimum of 10,000 events were collected for each sample. In instances when cells were sorted, the top 10% (ALDH-High) and bottom 40% (ALDH-Low) were sorted into NBM supplemented with B-27 for use in invasion assays, or into RLT buffer for RNA extraction.

The Image-IT Lipid Peroxidation kit for live-cell analysis (Image-iT Lipid Peroxidation kit, Thermo Fisher Scientific Inc.) was used as described to measure the degree of lipid peroxidation. Results are displayed as a ratio of oxidized BODIPY C11 reagent (∼510 nm) over unoxidized BODIPY C11 reagent (∼591 nm), as per the manufacturer's instructions. Cells were plated at 1 × 10⁴ cells/mL for 3 days, followed by appropriate treatment. Cells were centrifuged and then dissociated with Cell Dissociation Solution, Non-Enzymatic and replated in full media with 10 μmol/L Image-IT Lipid Peroxidation probe for 30 minutes at 37°C, followed by DyLight 800 NHS Ester (Life Technologies, cat. #46421) staining for 15 minutes at room temperature as a live/dead marker. Cumene hydroperoxide (100 μmol/L) induces lipid peroxidation and was added to cells for 2 hours at 37°C prior to staining as a positive control. Flow cytometry was performed on a BD FACSAria using FACSDiva software, followed by analysis with FlowJo. A minimum of 10,000 events were collected for each sample.

Total RNA was collected from cells using an RNeasy Mini Kit (Qiagen; 74106). cDNA was synthesized using SuperScript III First-Strand Synthesis System for RT-PCR with Oligo(dT) primers (Invitrogen, cat. #18080-051). qPCR was performed with FastStart Universal SYBR Green Master (Rox; Roche, cat. # 049138500011) on the Applied Biosystems 7900HT Fast Real-Time PCR System. Ct values were normalized to GAPDH Ct values for each individual sample and then averaged across experiments.

All statistical analyses were performed using GraphPad Prism 5.0 software. One-way ANOVA with Tukey test for multiple comparisons, Mann–Whitney, or the Student t test was used to determine differences between experimental groups as appropriate.

To explore the mechanisms of EGFR inhibitor resistance in GBM, we utilized a murine model for GBM driven by constitutive activation of EGFR (EGFRvIII; Fig. 1A; refs. 24, 26). Tumor cells derived from orthotopic tumors that occur in these mice can be maintained in vitro as tumorspheres that, in turn, generate highly proliferative and invasive tumors upon orthotopic engraftment (Fig. 1B) with robust EGFR activation. We refer to these tumor-propagating cells as cancer stem cells (CSC). To study the emergence of EGFR inhibitor resistance, we cultured cells in the presence of the EGFR inhibitor erlotinib. Within 2 hours of EGFR inhibition, changes in tumorsphere morphology were evident. Ragged, irregular spheres became compact, round spheres with a smooth outer surface upon EGFR inhibition (Fig. 1C and D). Western blot analysis revealed decreased phosphorylation of Y1173 in treated cells, as well as decreased signaling downstream of EGFR, including decreased phosphorylation of AKT (S473) and Src (Y416; Fig. 1E). With respect to biological properties, acute EGFR inhibition increased tumor cell invasion in matrigel boyden chambers (Fig. 1F and G), and increased tumorsphere invasion in matrigel 3D assays (Fig. 1H and I). The proinvasive phenotype conferred by EGFR inhibition was not specific to erlotinib and was observed with 2 additional pharmacologic EGFR inhibitors: AG1478 (27) and lapatinib. As expected, EGFR inhibition also decreased tumor cell proliferation (Supplementary Fig. S1E). Although EGFR inhibition decreased downstream signaling pathway activity (Fig. 1E), inhibition of PI3K/AKT or Src signaling pathway alone was not sufficient to confer the proinvasive phenotype (Supplementary Fig. S1A–S1D).

To determine the in vivo relevance of EGFR inhibitor–stimulated invasiveness, mice harboring orthotopic tumors were treated with erlotinib (150 mg/kg/d) for 5 days, and tumors were examined in the brain of euthanized mice at this time point. Inhibition of EGFR conferred a 1.9-fold increase in the area of the invasive tumor as compared with control-treated tumors (Fig. 1J and K).

A gene family often implicated in chemotherapeutic resistance encode aldehyde dehydrogenases that detoxify reactive aldehyde species. EGFR inhibition in CSCs resulted in a striking 11-fold increase in aldehyde dehydrogenase 1A1 (ALDH1A1) expression relative to control (Fig. 2A). The increase in ALDH1A1 expression conferred by erlotinib was also observed with AG1478 and lapatinib (Supplementary Fig. S1F). EGFR inhibition also resulted in increased expression of several genes implicated in cell adhesion and motility, including integrin beta chain beta 3 (ITGB3), CXCR4, S100A4, and TIMP3 (Fig. 2A; Supplementary Fig. S1G). EGFR inhibition resulted in increased ALDH activity (Fig. 2B and C) by ALDEFLUOR assay (29). Treatment of CSCs with the irreversible ALDH inhibitor disulfiram (DS; Fig. 2D and E) or a structurally distinct ALDH inhibitor N,N-diethylabenzaldehyde (DEAB; Supplementary Fig. S2A–S2B) blocked the proinvasive phenotype stimulated by EGFR inhibition.

To determine whether elevated ALDH activity was associated with invasive cells, we sorted bulk untreated tumor cells based on ALDH activity (Fig. 2F). With EGFR inhibition, sorted ALDH-high cells proved more invasive than ALDH-low cells in 3D matrigel invasion assays (Fig. 2G and H). Strikingly, this difference in invasion was dependent on EGFR inhibition (Fig. 2I and J). Thus, a highly proinvasive phenotype required both high ALDH activity and EGFR inhibition.

EGFR inhibition resulted in a rapid change in tumor invasive behavior dependent on ALDH activity. We hypothesized that heterogeneity in ALDH1A1 expression could be an intrinsic mechanism of resistance to EGFR inhibition. In untreated bulk tumor cell cultures, ALDH1A1 protein was not apparent by Western blotting (Fig. 3A; Supplementary Fig. S2C). Single-cell–derived clonal analysis of bulk tumor, however, revealed a range of ALDH1A1 expression levels across clones (Fig. 3A). Two clones with different ALDH1A1 protein expression, similar levels of EGFRvIII activation, and similar cell growth under control conditions, demonstrated a marked difference in sensitivity to EGFR inhibition (Fig. 3A–D). The ALDH1A1-high clone 2 had a distinct survival advantage over the ALDH1A1-low clone 1 upon EGFR inhibitor treatment (Fig. 3C and D). Moreover, EGFR inhibition conferred a proinvasive phenotype in the ALDH1A1-high clone 2 (Fig. 3F and G). Cell viability of clone 2 was markedly reduced by DS treatment (Fig. 3E).

Oncogenic activation of EGFR alters cellular metabolism and upregulates several prosurvival pathways (30). Inhibition of EGFR can decrease the activity of these pathways and lead to increased oxidative stress (31). Increased oxidative stress can promote lipid membrane peroxidation and the generation of toxic aldehyde species. To investigate changes in lipid membrane peroxidation, in association with EGFR inhibitor treatment, we examined the levels of these metabolites following EGFR inhibition of bulk tumor cells. Inhibition of EGFR nearly doubled the lipid membrane peroxidation levels as demonstrated by a shift in peak fluorescence emission upon oxidation of C11-BODIPY^581/591 (ref. 32; Fig. 4A). In addition, the cellular accumulation of toxic protein adducts containing 4-HNE, a destructive aldehyde generated by lipid peroxidation, was readily observed upon EGFR inhibition (Fig. 4B and C).

Aldehyde dehydrogenases, including ALDH1A1, metabolize and remove harmful reactive aldehydes produced by lipid peroxidation, including 4-HNE (33, 34). To determine whether CSCs with elevated ALDH levels were more resistant to lipid peroxidation, we applied exogenous 4-HNE to clones 1 and 2, with results showing increased viability of ALDHA1-high clone 2 cells, relative to ALDHA1-low clone 1 cells (Fig. 4D). In relation to ALDHA1-low cells, ALDHA1-high clone 2 cells showed decreased levels of lipid peroxidation upon EGFR inhibition (Fig. 4E). High ALDH1A1 levels were also associated with greater protection from DNA damage induced by EGFR inhibition, as demonstrated by decreased phospho-γH2AX levels in these cells (Fig. 4F).

EGFR inhibitor treatment of orthotopic tumors demonstrated that ALDH1A1-high clone 2 cells were more resistant to therapy and developed larger tumors than ALDH1A1-low clone 1 cells. When ALDH1A1 was inhibited by DS, there was no difference in the tumor area in ALDH1A1-high and -low tumors (Fig. 4G). In the absence of any therapy, ALDH1A1-high and -low tumors were indistinguishable.

To investigate potential functions for ALDH activity in human GBM, we analyzed PDXs. ALDH1A1 could be detected in 5 of 9 PDX lines, with variable levels of expression evident among the 5 positive lines (Fig. 5A). GBM6, with amplified and expressed EGFRvIII, expressed low-level ALDH1A1, and showed low ALDH activity (Fig. 5A and B). In ALDH1A1-low GBM6 inhibition of EGFR increased ALDH1A1 expression 2.3-fold (Fig. 5C) and increased tumor cell invasion (Fig. 5D and E). Erlotinib-induced invasion could be blocked by ALDH inhibition by DS. In contrast to GBM6, GBM43 had high endogenous levels of ALDH1A1 protein and high ALDH activity (Fig. 5A and B). Inhibition of ALDH by DS in GBM43 resulted in decreased cell viability relative to ALDH-low GBM6 and nearly abolished tumor cell invasion (Fig. 5F–H).

ALDH1A1 protein is expressed in the majority of human GBM (69%, 35 of 51 GBM at initial diagnosis); however, the fraction of positive tumor cells is relatively low with a mean immunopositivity of 8.0% ± 12.5% among positive tumors (Fig. 6A). ALDH1A1-positive tumor cells were observed in perivascular regions and as individual infiltrating cells (Fig. 6B–E). Although ALDH1A1-positive cells were frequently present at the infiltrative tumor edge, they were also present in central tumor regions. Given the elevated ALDH1A1 expression in our murine tumor cells upon EGFR inhibition, we assessed ALDH1A1 levels in paired human GBM from before and after EGFR inhibitor therapy (Fig. 6F). In tumors with EGFR amplification, there was a significant increase in ALDH1A1-positive tumor cells following EGFR inhibitor therapy (P < 0.05, n = 13 tumor pairs). In contrast, there was no increase in ALDH1A1 in tumors without EGFR amplification (n = 9 tumor pairs).

Alterations in EGFR signaling are common in GBM, yet strategies to target EGFR-associated signaling abnormalities have had limited therapeutic efficacy. Here, we identify a subpopulation of tumor cells with elevated levels of the detoxifying enzyme ALDH that resist EGFR inhibition. We showed that ALDH expression was associated with protection from EGFR inhibitor–mediated lipid peroxidation and a proinvasive phenotype. Pharmacologic inhibition of ALDH selectively targeted the EGFR inhibitor–resistant subpopulation, and in so doing caused decreased viability and invasion of these cells. Our data suggest therapeutic strategies targeting ALDH-high tumor cell populations, which have greater resistance to oxidative stress, may be effective when used in combination with EGFR inhibition.

The molecular biology of GBM includes increased lipid peroxidation and generation of toxic end products, such as 4-HNE, that can lead to DNA damage and cell death (35–37). To offset this antitumor characteristic, GBM use several protective mechanisms. To decrease the production of peroxidized lipids, aggressive tumors often maintain relatively high levels of saturated fatty acids, which are less susceptible to lipid peroxidation. Indeed, oncogenic activation of RTKs, including EGFR, drives de novo lipogenesis and the generation of highly saturated phospholipids (22, 35). The inhibition of EGFR can suppress several of these protective mechanisms, including decreased de novo lipogenesis and increased oxidative phosphorylation (31, 38). Using a murine model for GBM, we demonstrate EGFR inhibition results in increased lipid peroxidation and increased production of toxic 4-HNE adducts.

Aldehyde dehydrogenases, a family of 19 enzymes, detoxify endogenous and exogenous aldehydes in a NAD(P)⁺-dependent manner, and have been implicated in EGFR inhibitor resistance in lung, breast, and gastric cancers (31, 39). Consistent with their role in the detoxification of lipid peroxidation products, we have shown that ALDH-high GBM cells exhibit decreased lipid peroxidation upon EGFR inhibition and increased resistance to 4-HNE–associated cell toxicity.

In murine CSCs, increased aldehyde dehydrogenase (ALDH) activity following EGFR inhibition may be due, in part, to a selective survival of ALDH-high tumor cells (Fig. 6G). Elevated ALDH activity has previously been associated with increased tumor cell invasion, self-renewal, and cancer stem–like properties in several cancers, including GBM (29, 40–43). Our data suggest an association between high ALDH activity and increased invasive phenotype. It is not clear if increased invasive phenotype is due to increased ALDH activity or, rather, elevated ALDH is simply an associated, but not causative, molecular characteristic of invasive GBM cells. Interestingly, tumor metabolism and oxidative phosphorylation may differ in highly invasive versus less invasive regions of GBM (44). Increased ALDH1A1 may be one mechanism to manage high metabolic stress at the invasive edge. In nonneoplastic brain, astrocytes also exhibit metabolic heterogeneity as evidenced by an ALDH1A1-high subpopulation of astrocytes that reside in close proximity to blood vessels (45).

ALDH activity has also been used to define tumor cell populations with increased resistance to chemotherapy, including agents targeting EGFR (31, 39, 46–48). Given the importance of oxidative stress and lipid peroxidation in aggressive cancers, it is not surprising that ALDH-high tumor cell subpopulations may play important roles in resistance to several chemotherapeutics. In general, drug resistance is ascribed to a cell population with increased ALDH activity rather than the function of a single ALDH enzyme. Due to the large number of ALDH enzymes and increased expression of several ALDH enzymes in cancer, it has been difficult to assign causality to a specific ALDH (31). Our study indicates increased aldehyde dehydrogenase activity mediated by ALDH1A1 contributes to increased detoxification of reactive aldehydes downstream of lipid peroxidation and increased resistance to EGFR inhibition.

Both human (GBM43) and murine (CSC clone 2) GBM cells with elevated expression of ALDH1A1 were sensitive to the irreversible ALDH inhibitor DS. Based on data from several studies suggesting its cytotoxic effect on tumor cells (49, 50), DS is currently in clinical trials for the treatment of GBM (NCT03363659, NCT03151772, NCT02678975, and NCT02770378; clinicaltrials.gov). Our data confirm a cytotoxic effect of DS, but additionally suggest that combination therapy with both ALDH and EGFR inhibitors may be especially effective in treating a subset of GBM with abnormal and elevated EGFR activity.

T.F. Cloughesy has ownership interest (including stock, patents, etc.) in Notable Labs; is a consultant/advisory board member for Odonate Therapeutics, Del Mar Pharmaceuticals, VBI Vaccines, Dicephera, VBL, Agios, Merck, Roche, Genocea, Celgene, Puma, Lilly, Pascal Biosciences, BMS, Cortice, Wellcome Trust, Novocure, Novogen, Boston Biomedical, Sunovion, Human Longevity, Insys, ProNai, Bayer Pfizer, Notable Labs, MedQia, Tocagen, Karyopharm, GW Pharma, Kiyatec, AbbVie, and Boehinger Ingelheim; and has provided expert testimony as Member of the Board for the Global Coalition for Adaptive Research 501c3 for U.S. Provisional Application No.: 62/819,322. L.M Liau reports receiving a commercial research grant from Northwest Biotherapeutics and is a consultant/advisory board member for Insightec, Inc. and Arbor Pharmaceuticals. M. Prados is a consultant/advisory board member for Nativis. A. Lai reports receiving a commercial research grant from Genentech/Roche, has received honoraria from the speakers bureau of AbbVie, and is a consultant/advisory board member for Merck. W.A. Weiss has ownership interest (including stock, patents, etc.) in StemSynergy Therapeutics. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A. McKinney, J.J. Phillips

Development of methodology: A. McKinney, O.R. Lindberg, H. Gong, E.F. Simonds, J.J. Phillips

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A. McKinney, O.R. Lindberg, J.R. Engler, K.Y. Chen, T.F. Cloughesy, L.M. Liau, M. Prados, A.W. Bollen, C.D. James, W.H. Yong, A. Lai, M.E. Hegi, J.J. Phillips

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. McKinney, O.R. Lindberg, K.Y. Chen, A. Kumar, H. Gong, M. Prados, J.T.C. Shieh, A. Lai, W.A. Weiss, J.J. Phillips

Writing, review, and/or revision of the manuscript: A. McKinney, K.Y. Chen, E.F. Simonds, T.F. Cloughesy, L.M. Liau, M. Prados, A.W. Bollen, M.S. Berger, J.T.C. Shieh, C.D. James, T.P. Nicolaides, W.H. Yong, A. Lai, M.E. Hegi, W.A. Weiss, J.J. Phillips

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.Y. Chen, K.V. Lu, T.P. Nicolaides, M.E. Hegi, J.J. Phillips

Study supervision: J.J. Phillips

We are grateful and acknowledge the UCSF Brain Tumor SPORE Tissue Core (P50CA097257) for providing histology services and the UCSF Helen Diller Family Comprehensive Cancer Center Laboratory for Cell Analysis Shared Resource Facility (P30CA082103) for microscopy services. This study was funded by NIH/NINDS R01 NS081117 to J.J. Phillips, NIH/NCI U01 CA168878 to J.J. Phillips, NIH/NCI U01 CA229345 to J.J. Phillips, NIH/NCI P50CA221747 to C.D. James, the Oncosuisse (OCS-01680-02-2005 to M.E. Hegi), NIH/NCI R01 CA179071 to A. Lai, and NIH/NCI P50-CA211015 to A. Lai; and resources were provided by the UCSF Brain Tumor SPORE Biorepository NIH/NCI P50CA097257 to J.J. Phillips. We also acknowledge and thank the T.J. Martell Foundation, the Gerson and Barbara Bakar Philanthropic Fund, and the Sence Foundation for their support of this work. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

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.