Concurrent Inhibition of Neurosphere and Monolayer Cells of Pediatric Glioblastoma by Aurora A Inhibitor MLN8237 Predicted Survival Extension in PDOX Models

Glioblastoma multiforme (GBM) is one of the most malignant brain tumors that occurs in adults and children. Unlike adult GBM in which the standard treatment is radiotherapy and adjuvant temozolomide (1), a standard of care does not exist for the treatment of pediatric GBM (pGBM) (2). The five-year survival rate in patients with pGBM remains at 20% (3). Novel therapy for this disease is critically needed.

Recent studies suggest Aurora A serine/threonine kinases as potential oncologic targets (4, 5). Of the three related Aurora kinases (Aurora A, B, and C; ref. 6), Aurora A kinase (AURKA) regulates centrosome maturation, mitotic entry, centrosome separation, spindle formation, and cytokinesis (4). Amplification and/or overexpression of AURKA has been reported in multiple human tumors (7–14). In central nervous system tumors, AURKA overexpression correlates with disease progression and shorter survival times (15–17). Among the newly developed AURKA inhibitors, MLN8237 (alisertib) is an orally available reversible selective inhibitor with strong antitumor activity both in vitro and in vivo (12, 17–22). In addition, MLN8237 passes through the blood–brain barrier (BBB), and thus is an attractive agent to treat CNS malignancies (23). One of the key mechanisms of MLN8237-induced cell death is upregulation of p53 (11, 19). As p53 mutation is far less frequent in pGBM than in adult GBM (24, 25) and MLN8237 exhibited an acceptable safety profile in adult and pediatric phase I/II trials (26–30), the applicability of MLN8237 can potentially be greater and expedited in pGBM tumors.

As the incidence of pGBM is less than adult GBM and the number of available new candidate treatment agents is increasing, it is important to establish strong preclinical rationale to prioritize new agents for a clinical trial, and more importantly, to improve the chances of clinical success. For initial drug screening, it is desirable to develop an in vitro drug testing system that can predict in vivo efficacy in animal models. In addition to traditional monolayer cultures, new three-dimensional (3D) cultures, such as spheroids and organoids (31), have been developed. While neurospheres better represent 3D tumor architecture, microenvironment, and cellular heterogeneity of patient tumor and favor the growth of cancer stem cells (CSC), the lack of paired neurosphere and monolayer cultures derived from the same patient makes it difficult to determine which culture type better predicts in vivo treatment response or whether tumor cells in both cultures need to be targeted. For the subsequent in vivo evaluation of therapeutic efficacy, it is ideal to include model systems derived from tumors at different points of disease presentation. For example, therapies that are effective in treatment-naïve animal models frequently fail in the heavily pretreated patients with refractory tumors who are the subjects of most early-phase clinical trials. While conversely, testing new drugs in comparatively resistant tumor models jeopardizes discounting new therapies, which may prove effective in the context of upfront therapy.

We have optimized a surgical procedure that allows for the safe and rapid implantation of pediatric brain tumor cells into the matching locations in the brains of SCID mice (32–36). Our detailed characterization of these patient-derived orthotopic xenograft (PDOX) mouse models has confirmed their faithful replication of histopathologic features, invasive phenotypes, and major genetic abnormalities of the original patient tumors (32–36). From PDOX tumors of pGBM, we also established three matching pairs of cultured monolayer and neurospheres to facilitate the in vitro and in vivo evaluation of new therapies, such as MLN8237 in pGBMs.

In this report, we evaluated AURKA expression in pGBMs compared with pediatric low-grade gliomas, examined the in vitro antitumor effects of MLN8237 by treating paired monolayer and neurosphere cultures established from three pGBM models derived from untreated, recurrent, and terminal/lethal tumors, performed detailed analyses of in vivo therapeutic efficacy, and determined mechanisms of action of MLN8237 in two pGBM models. Our objectives were to examine (i) whether AURKA is a therapeutic target in pGBM, (ii) whether MLN8237 can effectively target this deadly disease, and (iii) whether effective targeting of both monolayer and neurosphere cells predicts prolonged animal survival time.

Fresh tumor tissue was collected from 11 patients with low-grade gliomas (LGG; WHO grade I/II) and 14 patients with pGBMs (WHO grade IV). Signed informed consent was obtained from the patient or legal guardian prior to sample acquisition in accordance with Institutional Review Board (IRB) policy. All studies were conducted in accordance with the ethical guideline of Declaration of Helsinki. Normal control human cerebellar RNAs from 5 adults as well as total RNAs from 2 fetal brains was procured from a commercial source (Clontech Laboratories, Inc. and Biochain; ref. 37).

Orthotopic free-hand surgical transplantation of tumor cells into mouse cerebrum was performed as described previously (36) following an Institutional Animal Care and Use Committee–approved protocol. PDOX models of intracerebral (IC)-4687GBM, IC-3752GBM (38), and IC-R0315GBM were established by direct injection of surgical or autopsy specimens into mouse cerebra; maintenance of reproducible tumorigenicity was confirmed for 5 in vivo passages. These xenograft tumors replicated major histopathologic features of the original patient tumors (38), and all three models are highly invasive in mouse brains. Patient tumor 4687GBM was obtained at the time of initial tumor resection (therapy-naïve), while patients 3752GBM and R0315GBM were heavily treated prior to sample acquisition (Table 1). The NOD/SCID mice were maintained in a pathogen-free animal facility at Texas Children's Hospital (Baylor College of Medicine, Houston, TX). Mice of both sexes, ages 6–8 weeks, were anesthetized with sodium pentobarbital (50 mg/kg). Tumor cells (1 × 10⁵), isolated from donor xenografts, were suspended in 2 μL of culture medium and injected into the cerebral hemisphere (1 mm to the right of the midline, 1.5 mm anterior to the lambdoid suture, and 3 mm deep) with a 10-μL 26-gauge Hamilton Gastight 1701 syringe needle.

Three pairs of cultured pGBM cells were initiated from tumors harvested from PDOX models and established as both monolayer and neurosphere cultures (32, 39). Monolayer cells were cultured in DMEM supplemented with 10% FBS (Atlanta Biologicals, Inc.), 200 U/mL penicillin/streptomycin. Neurosphere cells, in which putative CSC populations are enriched, were cultured in serum-free cell growth medium consisting of neurobasal media, N2 and B27 supplements (Life Technologies, Grand Island, NY), recombinant human βFGF and EGF (50 ng/mL each; R&D Systems Inc.), and 200 U/mL penicillin/streptomycin (32, 39).

Total RNA was extracted with TRIzol reagent (Invitrogen) from snap frozen patient and xenograft tumor tissue and cultured cell lines. cDNA synthesis from 1 μg of total RNA was performed using High Capacity RNA-to-cDNA Kit (Applied Biosystems) following manufacturer's protocol. Gene-specific quantitative RT-PCR analysis of AURKA was performed with SYBR Select Master Mix (Applied Biosystems) on a StepOnePlus Real-Time PCR system and MicroAmp Fast optical 96-well reaction plate (Applied Biosystems) using synthesized 5 ng cDNA. The following primers were used: AURKA = forward, 5′-CTGCATTTCAGGACCTGTTAAGG-3′; reverse, 5′-AACGCGCTG GGAAGAATTT-3′ (40); and GAPDH = forward, 5′-AAGGTGAA GGTCGGAGTCAA-3′; reverse, 5′-AATGAAGGGGTCATTGATGG-3′. Relative mRNA expression levels against normal pediatric/adult cerebral tissues were determined by the 2^−ΔΔC_t method. All quantitative RT-PCR assays were performed in triplicate.

MLN8237, provided by Millennium Pharmaceuticals, Inc., was dissolved in DMSO for a stock solution of 10 mmol/L for in vitro experiments. To determine the time- and dose-dependent effects of MLN8237, paired pGBM monolayer and neurosphere cultures were seeded into 96-well plates and exposed to 7 doses of MLN8237 (ranging from 1 to 4,000 nmol/L) or vehicle control over a 2-week period. Cell viability was measured at day 1, 4, 7, 10, and 13 using the Cell Counting Kit-8 (CCK8; Dojindo Molecular Technologies; refs. 32, 35).

For in vivo experiments, MLN8237 was dissolved in 10% (2-Hydroxypropyl)-β-cyclodextrin (Sigma-Aldrich)/1% sodium bicarbonate to achieve a final concentration of 5 mg/mL. Two weeks after intracerebral tumor cell implantation, mice were orally administered MLN8237 (30 mg/kg/day; refs. 19, 23, 41, 42) once daily for 12 days. To determine any survival benefits from MLN8237 treatment, the mice were monitored daily until they developed signs of neurologic deficit or became moribund, at which time they were euthanized and their brains were removed for analysis. To test the biologic effects of MLN8237 at the end of treatment, mice were treated with MLN8237 (30 mg/kg/day), a dose frequently used before in animal experiments (19, 23, 41, 42) for 12 days once initial neurologic symptoms caused by GBM growth were observed; these mice were euthanized 1, 24, and 72 hours after the last MLN8237 dose, and their brains were removed for analysis. Plasma was collected from all euthanized mice.

Flow cytometry was performed as we described previously (32, 35, 36, 38). To examine the changes in cell-cycle distribution, tumor cells were washed and incubated at 37°C for 45 minutes with 10 μg/mL Hoechst 33342 (Sigma-Aldrich) and Verapamil (ATP pump inhibitor, 50 μmol/L) in FCM buffer (Dulbecco's PBS, 0.5% BSA, and 2 mmol/L EDTA) for DNA staining following incubation with 0.5 μg/mL PyroninY (Sigma-Aldrich) for another 45 minutes to stain RNA. To exclude mouse cells from analysis, xenograft cells were stained with a cocktail of APC-conjugated antibodies specific to mouse cell surface antigen (CD90.1, CD133, CD140a, and CD24) together with matching APC-conjugated isotype control antibodies (Mouse IgG1 κ, Rat IgG2a κ, and Rat IgG2b κ; BioLegend, Inc.) to set the gate. To analyze the putative CSCs, tumor cells were washed and labeled with APC-conjugated anti-human CD133 (Miltenyi Biotec Inc) and FITC-conjugated anti-human CD15 (Miltenyi Biotec Inc) for 15 minutes on ice in 100 μL of FCM buffer containing FcR blocking reagent (32, 35, 36, 38). Anti-mouse IgG2b-APC and anti-mouse IgM-FITC antibodies (Miltenyi Biotec Inc) were also included as isotype controls. After washing, cells were resuspended in FCM buffer containing 2 μg/mL PI to gate out dead cells and analyzed with a LSR II flow cytometer and Kaluza Analysis Software Version 1.3 (Beckman Coulter, Inc.).

MLN8237 concentrations were determined using an Applied Biosystems API5000 mass spectrometer equipped with a Shimadzu Nexera UFLC System and a LEAP autosampler following an optimized method at Takeda Pharmaceuticals. A reverse-phase gradient method with flow rate of 0.4 mL/minute on a Phenomenex Synergi Polar-RP column was used for analyte separation. The mobile phases used were water and acetonitrile (supplemented with formic acid 0.1% volume-to-volume). MLN8237 was ionized under a positive ion spray mode and detected through the multiple-reaction monitoring of a mass transition pair at a mass-to-charge ratio of 519.1/328.1. Calibration curves of MLN8237 were established using standards, and the peak area ratios of the analyte against an isotopically labeled internal standard were used to quantify samples. Using a plasma volume of 50 μL, linearity was achieved in the MLN8237 concentration range of 1.93–3860 nmol/L. In addition, dilution linearity was also assessed to ensure that study samples above the upper limit of the standard curve could be diluted with blank matrix without affecting the final calculated concentration. This was performed by preparing dilution QCs at 19,300 nmol/L and diluting down to the curve range by 50-fold with blank mouse plasma. The triplicate dilution QCs have met the acceptance criteria of within 15% with respect to accuracy and precision. This approach validated the dilution cap of accurately quantifying samples that are AQL (above quantitation limit) up to 193,00 nmol/L. All data were acquired and processed using Analyst 1.6.2 software (Applied Biosystems).

Proteins were extracted in M-PER Mammalian Protein Extraction Reagent (Life Technologies) including Protease Inhibitor Cocktail, separated with SDS-PAGE in a 4%–12% Bis-Tris gel, transferred to nitrocellulose membrane, and incubated with the primary antibodies: Anti-Aurora A (1:1,000, Cell Signaling Technology, catalog no. 4718), anti-PARP (recognizes both the full and the cleaved fragment, 1:500, Calbiochem, catalog no. AM30), anti-caspase-3 (recognizes both the full and the cleaved fragments, 1:1,000, Cell Signaling Technology, catalog no. 9662), and anti-β-actin (1:5,000, Sigma-Aldrich, catalog no. A2228). Following incubation with IRDye-conjugated secondary antibody (1:10,000, LI-COR Biosciences), proteins were detected and quantitated using an Odyssey Infrared Imaging System (LI-COR Biosciences) application Software version 3.0.

IHC staining was performed using a Vectastain Elite ABC kit Rabbit IgG (Vector Laboratories) or Mouse on Mouse (M.O.M.) Elite Peroxidase Kit (Vector Laboratories) on 5-μm paraffin sections of whole mouse brains as described previously (32, 35, 36, 43). After antigen retrieval with Target Retrieval Solution (Dako North America, Inc.) in a pressure cooker, sections were blocked with horse normal serum or M.O.M. Mouse IgG Blocking Reagent. Primary antibodies utilized were full PARP (1:100; Sigma-Aldrich), cleaved PARP (1:100), full caspase-3 (1:50), cleaved caspase-3 (1:100), AURKA (1:50; Cell Signaling Technology, Inc.), and Ki67 (1:50) (Abcam). After slides were incubated with primary antibodies, appropriate biotinylated secondary antibodies were applied, and the final signal was developed using 3, 3′-diaminobenzidine substrate kit for peroxidase. Negative control was performed by replacing primary antibodies with Dulbecco's PBS. IHC staining was assessed by combining intensity, scored as negative (−), low (+), medium (++), or strongly positive (+++), with extent of immunopositivity (percentage of positive cells). For AURKA, the IHC H-score was calculated as 3 × percentage of strong (+++) positive cells + 2 × medium (++) positive cells + 1 × low (+) positive cells (44).

Values were presented as mean ± SD. The effect on cell proliferation was analyzed with two-way ANOVA, and in vitro cell cycle analyzed with one-way ANOVA followed by a multiple comparison procedure (Holm–Sidak method). Animal survival times were compared through log-rank analysis. In vivo cell cycle and CSC populations were analyzed with pair-wise t test. All statistical analysis was performed by using SigmaStat 3.5 (Systat Software, Inc.). P < 0.05 was considered statistically significant.

To evaluate AURKA mRNA expression in pGBMs, qRT-PCR was performed on patient samples from 11 pediatric LGGs, 14 pGBMs, and 7 normal cerebral tissues. Consistent with previous observations (17), AURKA mRNA was significantly overexpressed in pGBM with a mean tumor/normal ratio of 12.50 ± 7.98 compared with LGG (1.67 ± 0.68) and normal cerebral tissue (1.60 ± 1.71; P < 0.001; Fig. 1A). Subsequent evaluation of 10 PDOX pGBM mouse models and 3 pairs of patient-matched monolayer and neurosphere cultures derived from PDOX models of pGBM confirmed the maintenance of AURKA mRNA overexpression in xenograft tumors (up to 5 in vivo passages; Fig. 1B) and in the cultured pGBM cells (Fig. 1C).

To evaluate AURKA as a potential therapeutic target in pGBM, we examined in vitro effects of MLN8237 using cell cultures derived from 3 PDOX models (Table 1) representing different clinical timepoints of disease progression, that is, previously untreated (IC-4687GBM), recurrent following treatment (IC-3752GBM), and a progressive, terminal stage obtained at autopsy (IC-R0315GBM). Recognizing the role of CSCs in the progression and recurrence of GBM (45), stem cell–enriched neurosphere cell lines were treated together with the matching monolayer cells to MLN8237 (1–4,000 nmol/L).

In CSC-enriched neurospheres, MLN8237 showed significant inhibition of growth at nanomolar concentrations in all three lines in a time- and dose-dependent manner (Fig. 2A; Supplementary Fig. S1). Neurospheres from previously untreated patient tumor (IC-4687GBM) were the most sensitive to treatment with MLN8237, demonstrating significant suppression of cell proliferation after treatment with 15.6 nmol/L of MLN8237 from day 4 (28.2% ± 6.7%) through day 13 (>90%). In the recurrent model IC-3752GBM, the cells were more resistant to treatment as 62.5 nmol/L of MLN8237 led to 45.9% ± 6.2% growth suppression on day 7 and 75 ± 0.6% inhibition on day 13. In the autopsy/terminal model IC-R0315GBM, treatment with 62.5 nmol/L of MLN8237 resulted in >95% suppression of cell proliferation from day 7 to 13 (P < 0.05).

In monolayer cell lines, only the IC-4687GBM cells (previously untreated) were sensitive to MLN8237, but it required a higher dose (250 nmol/L) and longer exposure (day 7) to achieve >90% inhibition. Monolayer cells derived from the recurrent and terminal GBM models were resistant to MLN8237, and <50% inhibition was achieved even with 4,000 nmol/L for 13 days. Worthy of note is that monolayer cells from IC-4687GBM proliferated much faster than the other two models, which may partially explain their early responses to MLN8237. Overall, these data identified the treatment naïve IC-4687GBM as the most responsive tumor model and revealed differential responses between neurosphere and monolayer cells in the recurrent and terminal pGBM models.

To examine the drug's mechanism of action, first we examined apoptosis in vitro through FCM. To examine time–course responses, pGBM cells were treated with MLN8237 at 62.5 nmol/L, a dose active in neurospheres from all three models, for 1–7 days. In neurospheres, significantly elevated apoptosis (% of cells in sub-G₁; P < 0.05) was observed on day 2 (20%) in IC-4687GBM, day 4 (40%) in IC-3752GBM, and day 1 (18%) in IC-R0315GBM. The levels of apoptosis increased over time and peaked on day 7 reaching 65%–75% (Fig. 2C). In monolayer cells, only moderate apoptosis was noted in IC-3752GBM (18% on day 1, 25% on day 7). In agreement with these data, Western hybridization demonstrated similar results with time-dependent cleavage of PARP detected in all neurosphere cultures but minimal effects observed in the monolayer cells, including a minor effect on caspase-3 (Fig. 2D). To examine dose responses, cells were treated with MLN8237 (4, 62.5, and 1,000 nmol/L; Fig. 2C) and analyzed by FCM on day 7 of treatment, a point at which growth inhibition was observed in the CCK8 assays. In neurospheres, a significant increase in the sub-G₁ population was observed on day 7 with 4 nmol/L of MLN8237 in IC-4687GBM (15%), and treatment with 62.5 nmol/L and higher resulted in more significant increases (60%–90%) in apoptosis in all three models. However, in monolayer cells, higher drug doses (e.g., 1,000 nmol/L in IC-4687GBM) were required to induce even lower levels of apoptosis (e.g., 60% in IC-4687GBM; Fig. 2C). Overall, MLN8237 induced significant apoptosis in pGBM cells, particularly in the neurospheres.

Next, we analyzed changes in cell-cycle distribution in viable tumor cells in vitro by excluding the sub-G₁ population (Fig. 2B). In neurospheres, a time-course treatment with 62.5 nmol/L of MLN8237 revealed early (day 1–2) reduction of G₀–G₁ phase cells accompanied by accumulation of G₂–M phase cells in all three models (Supplementary Fig. S2A, right). Prolonged treatment to day 7 resulted in release of G₂–M blockage and an increase in G₀–G₁ phase cells. Analysis of dose responses on day 7 failed to detect dose-dependent G₂–M block (Supplementary Fig. S2A, left). In monolayer cells, time-course analysis detected G₀–G₁ reduction and G₂–M block in IC-3752GBM (recurrent) and IC-R0315GBM (autopsy) cells, but not in IC-4687GBM (treatment naïve; Supplementary Fig. S2A, right). Also, IC-3752GBM was the only model that displayed a dose-dependent G₀–G₁ decrease (from 80% to 38%) and G₂–M blockage (from 10% to 60%) at 1,000 nmol/L (Supplementary Fig. S2A, left). These data indicated that MLN8237 was able to induce short-term (1–2 days) G₂–M blockage in pGBM neurospheres but failed to trigger major responses in monolayer cells from the three models.

One major hurdle in the treatment of CNS tumors is the necessity for drugs to cross the BBB. Before we launch a large-scale in vivo efficacy examination, we examined the CNS penetration capability of MLN8237 in a responsive pGBM by treating IC-4687GBM with MLN8237 for 12 days (30 mg/kg/day; Fig. 3A). Drug levels in xenograft tumors were compared with levels in the neighboring “normal” mouse brains and in the serum (Fig. 3B). In plasma, MLN8237 was detected at a concentration of 6,636 ± 1,687 nmol/L (n = 6) 1 hour following the completion of treatment and decreased to 1,126 ± 546 nmol/L at 24 hours (n = 4) and to undetectable levels (<1 nmol/L) at 72 hours (Fig. 3B). Compared with levels of MLN8237 in the normal mouse brain tissues (397 ± 100 nmol/L; n = 2) at 1 hour posttreatment and (45 ± 4 nmol/L) (n = 2) at 24 hours posttreatment, MLN8237 accumulated in tumor tissue to 615 nmol/L (n = 1) at 1 hour and remained elevated at 90 nmol/L (n = 1) at 24 hours posttreatment (Fig. 3B). Worthy of note is that this tumor is highly invasive. Therefore, the normal tissues harvested from the tumor-bearing mice may contain small amount of invasive tumor cells. The significantly elevated drug concentration in the tumor mass indicated that MLN8237 was able to penetrate the BBB and accumulate in IC-4687GBM. Compared with serum concentrations achieved in adults in a phase I trial (approximately 5,000 nmol/L ± 1,800 nmol/L, 7 days at 100 mg/kg, twice daily), the serum concentration in our animals are slightly higher (27).

We next examined the activity of MLN8237 in vivo and assessed whether in vitro activity predicted the in vivo therapeutic efficacy. Specifically, we examined whether effective targeting of CSC-enriched neurospheres (38, 46–49) was sufficient to predict significant extension of animal survival times or whether simultaneous targeting of both neurospheres and monolayers was required. Mice bearing IC-4687GBM (previously untreated pGBM with monolayer and neurosphere MLN8237 sensitivity) and IC-R0315GBM (terminal pGBM with only neurosphere MLN8237 sensitivity) were orally treated with MLN8237 (30 mg/kg/day for 12 days) 2 weeks after tumor implantation. Animal survival times (median) in IC-4687GBM were significantly prolonged from 53 days in the vehicle group to 78 days in the treatment group (P = 0.022; Fig. 3C). However, in the autopsy model IC-R0315GBM, the median survival time was 41 days in vehicle group and 42 days in the MLN8237-treated group (P = 0.859). While these data suggested that simultaneous suppression of both neurospheres and monolayer cells correlates with improved therapeutic efficacy, in vitro suppression of neurospheres alone was insufficient to extend animal survival times, additional studies are needed in more tumor models to validate the power of paired monolayer and neurosphere cultures in predicting in vivo success of drug treatment.

As our in vitro data showed that MLN8237 induced apoptosis, we validated the role of apoptosis in vivo in mice 1 hour post-drug treatment (acute response) and in the recurrent tumors in the survival group (long-term/delayed effects). In the responsive model IC-4687GBM, MLN8237 (30 mg/kg/day for 12 days by gavage) resulted in significant increases in apoptosis (1 hour after completion of treatment) as evidenced by increased cleaved PARP (from + in ≤25% cells to +++ in nearly 75% of cells), cleaved caspase-3 (from ++ in ≤25% cells to +++ in 50%–75% cells; Fig. 3D; Table 2), and accumulation of cells in the sub-G₁ phase from pretreatment (3.6% ± 0.4%) to 1 hour after completion of treatment (10.4% ± 1.9%; Fig. 3E). As anticipated, in the cells from recurrent tumor (harvested in the survival group), cleaved PARP and cleaved caspase-3 decreased to low levels similar to baseline (+ in ≤25% cells and sub-G₁ to 3.1% ± 1.5%). Parallel to the increased apoptosis, changes in histologic features, such as decreased cell density (Fig. 3F) and increase of cell size (Fig. 3D, H&E staining), were also noted. In the autopsy model IC-R0315GBM, elevated cleavage of PARP and caspase-3 were only found in <25% cells. In all the tumor cells in both models, a high level (+++) expression of full PARP and caspase-3 was maintained, suggesting that surviving cells had functional apoptotic machinery. Cell proliferation (Ki-67) was not significantly altered at any timepoints in either model.

Despite significant prolongation of survival times following MLN8237 treatment in IC-4687GBM mice, all animals in our study ultimately succumbed to their tumors. To understand the mechanisms of therapy resistance, we examined changes of AURKA (molecular target of MLN8237) through IHC and evaluated as H-score, calculated as 3× percentage of strong (+++) positive cells + 2× medium (++) positive cells + 1× low (+) positive cells (ref. 44; Supplementary Fig. S3A). Compared with vehicle-treated IC-4687GBM tumors with H-score 19.8 ± 4.8 for AURKA, it decreased to 8.8 ± 4.4 (P < 0.001) with many areas lacking AURKA expression in recurrent tumors. There was no change in the H-score at the end of treatment tumor (22.4 ± 9.4) compared with vehicle-treated IC-4687GBM. In the resistant model IC-R0315GBM, cells with strong (+++) or medium (++) positivity decreased while low (+) positive cells slightly increased in the recurrent tumor (Supplementary Fig. S3B). These data suggest that pGBM tumor cells with low or no AURKA expression escaped MLN8237-induced cell killing and contributed to progression and recurrence of xenograft tumors in these two pGBM models, although data from additional models are needed to draw a definitive conclusion.

Next, we examined the role of CSCs using FCM analysis of dual-stained CD133⁺ and CD15⁺ cells, the two most commonly utilized cell surface markers for brain tumor stem cells in both neurospheres and xenografts. In vitro, CD133⁺CD15⁻ cells represented the major population of neurosphere cells derived from both IC-4687GBM (36.1%) and IC-R0315GBM (57.2%; Fig. 4A). Treatment with MLN8237 (4–62.5 nmol/L) for 7 days resulted in a minor shift (≤7%) of the mono (CD133⁺CD15⁻ and CD133⁻CD15⁺) and dual (CD133⁺CD15⁺) subpopulations, suggesting that all subpopulations were similarly, if not equally, sensitive to MLN8237 treatment. In vivo in the MLN8237-responsive model (IC-4687GBM), we detected an increase of mono-CD133⁺ (CD133⁺CD15⁻) cells from 9.5% ± 5.9% in vehicle-treated tumors to 28.2 ± 22.2% (P > 0.05) in tumors harvested 1 hour after MLN8237 treatment (30 mg/kg × 12 days), and we observed a further increase to 52.3% ± 12.8% (P < 0.05) in the recurrent tumors (Fig. 4B). These changes were accompanied by a steady decrease of CD15-mono positive (CD133⁻CD15⁺) cells (P = 0.055), whereas changes of dual-positive (CD133⁺CD15⁺) cells were not significant (Fig. 4B). Progressive tumors from the MLN8237-resistant model (IC-R0315GBM) showed similar enrichment of CD133⁺ cells (69.8% ± 8.6% in vehicle and 80.8% ± 0.0% in MLN8237-treated group; P < 0.05). However, the majority of the xenograft CD133⁺ cells in IC-R0315GBM coexpressed CD15 (59.4% ± 7.6% in vehicle and 76.2% ± 0.0% in MLN8237-treated group; Fig. 4B, bottom), which was significantly different from the neurospheres that were predominantly CD133-mono–positive (CD133⁺CD15⁻; Fig. 4A, bottom). In summary, these data suggest that the putative CSC population in vitro in cultured neurosphere cells did not have identical responsiveness toward MLN8237 as the CSCs in xenograft tumors, and CD133⁺ cell were the cause of therapy resistance because nearly all the CD15⁺ cells were dual positive with CD133.

In this report, we identified AURKA as a potential therapeutic target in a subset of pGBM by validating its overexpression in patient tumors, PDOX models, and matching pairs of cultured neurosphere and monolayer cultures of pGBM. We demonstrated strong antitumor activity of the AURKA inhibitor MLN8237 in vitro and in vivo and identified variable responses in pGBM cells derived from different stage of clinical progression (treatment naïve, recurrence, and terminal/autopsy) and between matching pairs of neurospheres and monolayer cultures. While our data suggested that successful inhibition in both neurosphere and monolayer cells correlates with extension of animal survival times, additional studies involving more tumor models are needed to further test the “adherent versus spheroid” hypothesis and to correlate with the frequent molecular changes. Mechanistically, we confirmed apoptosis as the primary cause of MLN8237-induced cell death and suggested the lack of AURKA expression and the expression of CSC marker CD133 as causes of therapy resistance. These findings also support a secondary transplantation of the treated (resistant) tumor cells followed by another round of AURKA inhibitor treatment to confirm the sustained resistance of the AURKA-negative and/or CD133⁺ cells toward AURKA inhibition in pGBMs.

Overexpression of AURKA and preclinical antitumor activity of AURKA inhibitors have been reported extensively in various human cancers. AURKA inhibitor MLN8237 (alisertib) is orally available and has exhibited strong antitumor activity in a wide range of human cancers and has a well-established clinical safety profile (26–30, 50, 51). The advantage of preclinical testing of therapeutic efficacy in PDOX models of pGBM, a disease that desperately needs new therapies, is that in vivo findings can potentially be rapidly translated into clinical trials. In an effort to facilitate the initiation of clinical trials, we address two important issues. First, we examined the treatment effects in samples from different stages of clinical progression and how in vitro models predict in vivo efficacy. Using our unique panel of PDOX models of pGBM, we examined the impact of clinical timepoint of the original patient tumor on the overall response to treatment with MLN8237. Although these models were not derived from the same patients (which is highly desired but very difficult to accomplish), testing new therapies in models representing untreated, recurrent, and terminal/autopsy pGBMs can provide important clues for future selection of patients. Our findings that previously treated recurrent and autopsy pGBM models responded differently to MLN8237 compared with the treatment-naïve tumor model suggests a need for a broader range of tumor models in future preclinical drug testing.

Second, as in vivo evaluation of therapeutic efficacy can be both time consuming and costly, in vitro prescreening assays that can better predict in vivo success are highly desired. Unlike traditional monolayer cells, neurospheres are enriched with CSCs, making them frequently favored over monolayer cells for in vitro preclinical drug testing (22, 52). Using AURKA-overexpressing neurosphere and monolayer pGBM lines, we found that MLN8237 was effective in reducing growth in all three neurosphere cultures enriched with putative glioma stem (CD133⁺ and CD15⁺) cells; these results are similar to preclinical findings in adult GBMs (21, 52). Unexpectedly, monolayer cultures were less responsive to MLN8237 treatment despite expressing similarly high levels of AURKA. Our data demonstrate that tumor cells maintained in these two growth conditions did not always respond identically to the same treatment, and that when MLN8237 effectively suppressed both neurosphere and monolayer cells in vitro, the xenograft models demonstrated the most significant in vivo drug responses. These data support the inclusion and testing of both monolayer and neurosphere cells in future drug development to validate the predictability of this strategy.

Similar to many previous studies on the molecular mechanisms of action of MLN8237 (11, 19, 22, 53–56), we demonstrated that apoptosis and cell-cycle arrest in G₂–M phase mediated the antitumor activity in pGBM both in vitro and in vivo. Despite the significantly prolonged animal survival time in IC-4687GBM, all the animals in the treatment group died of disease. To further understand the mechanisms of therapy resistance, we analyzed the protein expression of AURKA, the target of MLN8237, in the remnant/recurrent xenograft tumors through IHC. Not surprisingly, the number of AURKA-positive cells was dramatically decreased in the remnant/recurrent tumors, suggesting that 12 days of treatment with MLN8237 successfully eliminates AURKA-positive tumor cells, and implying that the AURKA-negative tumor cells that escaped MLN8237 treatment caused tumor recurrence/progression. This finding supports additional studies to establish the role of AURKA levels in determine drug responsiveness, for example, by showing that knockdown of AURKA levels in vitro results in a shift of drug potency. As high expression of full PARP and caspase-3 was detected in the remnant/recurrent tumors, these residual tumor cells may have functional apoptotic machinery that can be targeted in the future development of combination therapy with MLN8237 (22, 57–61).

In addition, our study of the mechanisms of resistance provided new insights on the role of CSCs in MLN8237 resistance. In IC-4687GBM, although CD133 and CD15 (both mono-and dual-positive) cells were killed by MLN8237 in neurospheres in vitro, the fraction of CD133-mono–positive cells increased in the remnant/recurrent xenografts in vivo. In IC-R0315GBM, while CSCs in neurospheres were predominately (>50%) composed of CD133-mono–positive (CD133⁺CD15⁻) cells, the xenografts were dual positive (CD133⁺CD15⁺) cells (59.4% before treatment and 76.2% in recurrent tumor). Such cellular differences, likely caused by changes in microenvironment, may have contributed to the differential responses observed in this terminal-stage pGBM model. This finding suggests that serum-free cell culture may favor the growth of a subpopulation of stem cells, which may not accurately reflect in vivo behavior. Our data shows that some pGBM cells expressing CD133 (alone or with CD15) were resistant to MLN8237 in vivo, and new strategies capable of effectively targeting CSCs are needed to further enhance the in vivo efficacy of MLN8237.

In conclusion, we identified AURKA as a novel therapeutic target for pGBM and demonstrated the strong antitumor activity of MLN8237 in vitro in neurospheres and in vivo in a treatment-naïve PDOX model. Also, this study suggested that simultaneous targeting of neurosphere and monolayer cell lines correlates with prolonged animal survival in vivo. Our data also support additional in vivo studies to confirm the loss of molecular target AURKA and increased expression of CD133 by tumor cells as the mechanisms of MLN8237 resistance in pGBM cells. Our data suggests that combination therapy is required to target AURKA-negative fractions and CD133⁺ cells following treatment with AURKA inhibition to further improve therapeutic efficacy.

No potential conflicts of interest were disclosed.

Conception and design:M. Kogiso, S. Zhao, H. Lindsay, P.A. Baxter, J. Muscal, X.-N. Li

Development of methodology:M. Kogiso, F.K. Braun, P.A. Baxter, X.-N. Li

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.):M. Kogiso, F.K. Braun, L. Zhang, F.Y. Lin, S. Zhao, A.M. Adesina, D. Liao, M.G. Qian, X.-N. Li

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis):M. Kogiso, L. Qi, H. Lindsay, J. Muscal, X.-N. Li

Writing, review, and/or revision of the manuscript:M. Kogiso, S.G. Injac, H. Zhang, F.Y. Lin, S. Zhao, H. Lindsay, J.M. Su, A.M. Adesina, S.L. Berg, J. Muscal, X.-N. Li

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases):M. Kogiso, L. Qi, F.K. Braun, L. Zhang, Y. Du, H. Zhang, F.Y. Lin

Study supervision:X.-N. Li

This project is funded by NIH/NCIRO1 CA185402 (to X.-N. Li), Cancer Prevention and Research Institute of Texas (CPRIT) RP150032 (to X.-N. Li), St Baldrick's Foundation (to J.M. Su), and by the Cytometry and Cell Sorting Core at The Baylor College of Medicine with funding from the NIH (P30 AI036211, P30 CA125123, and S10 RR024574), Texas Children's Hospital Pediatric Pilot Award (to J. Muscal), Kappa Alpha Theta Foundation Research Scholar (J. Muscal), and Alex's Lemonade Stand Center of Excellence (to S.L. Berg).

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.