PI3K–mTOR Pathway Inhibition Exhibits Efficacy Against High-grade Glioma in Clinically Relevant Mouse Models

Glioblastomas (WHO Grade IV) are the most common primary brain tumors with a median survival of only 15 months, despite intensive treatment including surgery and chemoradiation. Common molecular alterations in glioblastoma include overexpression of EGFR, PDGFR, and cMET, activating mutations in PI3K and EGFR (EGFRvIII) and loss of function of PTEN by deletion or mutations (1, 2). These alterations lead to constitutive activation of PI3K and further downstream effectors such as mTOR, which are crucial for tumor cell growth, proliferation, and survival (3). Preclinical studies suggest that inhibition of this pathway results in either direct inhibition of tumor growth or in sensitizing cells to conventional chemotherapy and radiotherapy (4, 5). Activation of the PI3K–mTOR signaling pathway is a very common event in solid tumors, which triggered the development of numerous small-molecule inhibitors. Although these PI3K inhibitors have been developed mainly for other more common solid tumors, they are now also being tested against glioblastoma (e.g., Clinicaltrials.gov identifiers NCT01339052, NCT00085566, NCT01240460, NCT02430363, NCT01316809) (6). Unfortunately, however, the outcomes of clinical studies with agents that target the PI3K pathway in glioblastoma have been disappointing so far. The prototype mTOR inhibitor rapamycin (sirolimus) and its analogues (so called rapalogs) temsirolimus and everolimus were considered promising targeted agents for glioblastoma, but a number of phase I and II trials applying them either as monotherapy or in combination with an EGFR inhibitor failed to demonstrate meaningful clinical efficacy in recurrent glioblastoma (7, 8). Possible explanations for this lack of efficacy include a lack of mTORC2 inhibition by rapalogs (9), increased upstream signaling of Akt through negative feedback loops following mTOR inhibition (10, 11), and PTEN status (12). Although these events, which focus on PI3K–mTOR signaling itself, may have contributed to the failures, a frequently underappreciated but critical issue is the presence of the blood–brain barrier (BBB) that may have hampered drug delivery to tumor cells in quantities required to elicit a meaningful pharmacologic response. Especially those glioblastoma cells that escaped surgical resection due to their migration away from the tumor core into adjacent regions of the brain will be shielded by a more intact BBB (13, 14).

It is well known that the intact BBB restricts the brain entry of the majority of xenobiotics by its unique structure (13). In particular, the drug efflux transporters expressed at the BBB play a very important role in limiting the brain penetration of a wide variety of compounds including frequently used chemotherapeutics and novel targeted agents. Two well-established drug efflux transporters, ABCB1 (P-glycoprotein, P-gp, MDR1) and ABCG2 (Breast Cancer Resistance Protein, BCRP), are abundantly expressed in the human and murine BBB and restrict most newly developed kinase inhibitors such as erlotinib, lapatinib, and palbociclib (15–17). As a consequence, the usefulness of such agents in glioblastoma treatment might be attenuated by an inadequate brain penetration.

In the current study, we have compared the affinity for ABCB1 and ABCG2 as well as their impacts on brain penetration of the mTOR inhibitors rapamycin and AZD8055, the PI3K/mTOR inhibitor NVP-BEZ235 and the PI3K inhibitor ZSTK474. All these inhibitors are under investigation for treatment of glioblastoma (18–20). We have used the orthotopic U87 brain tumor model to evaluate the antitumor efficacy of the inhibitors and further investigated ZSTK474, being the PI3K inhibitor with the most favorable brain penetration, in a more clinically relevant transgenic glioblastoma model (21).

The PI3K inhibitors rapamycin and ZSTK474 were purchased from LC Laboratories. NVP-BEZ235 was purchased from Selleck Chemicals. AZD8055 was purchased from Active Biochem. These compounds were dissolved in DMSO (Sigma-Aldrich) to yield 10 and 4 nmol/L working solutions and were stored at 4°C. Elacridar was kindly provided by GlaxoSmithKline. Zosuquidar (LY335979) was kindly provided by Eli Lilly. Modified Eagle's medium (MEM), Opti-MEM–reduced serum medium, HBSS (Hank's balanced salt solution), l-glutamine, nonessential amino acids, MEM vitamins, penicillin–streptomycin, FCS, trypsin-EDTA, and other reagents for cell culture were purchased from Invitrogen. All other chemicals were purchased from Merck.

LLC and MDCK cell lines used for in vitro transport experiments (see below) have been generated in our institute. U87MG human glioma cells have been obtained from ATCC and have been transfected with luciferase in house to obtain clone U87lucB5. All cell lines have been cryopreserved in stocks and aliquots are thawed and used for a maximum period of about 45 days (10–15 passages). Mycoplasma testing and testing for absence of mouse pathogens of the stocks is performed by PCR. The human U87lucB5 line was last authenticated by STR analysis on January 2013 (Identicell).

Mice were housed and handled according to institutional guidelines complying with Dutch legislation. All experiments with animals were approved by the animal experiment committee of the institute. The animals used for pharmacokinetics studies were female wild-type (WT), Abcb1a/b^−/−, Abcg2^−/−, and Abcb1a/b^−/−;Abcg2^−/− mice of FVB genetic background, between 9 and 14 weeks of age. p16^Ink4a/p19^Arf;K-Ras^v12;LucR, Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR, p53;p16^Ink4a/p19^Arf;K-Ras^v12;LucR and p53;Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR conditional mice for generation of glioblastoma cell lines and efficacy studies were genotyped as described previously (21). The animals were kept in a temperature-controlled environment with a 12-hour dark/ 12-hour light cycle and received a standard diet (AM-II, Hope Farm B.B.) and acidified water ad libitum.

To determine whether rapamycin, AZD8055, NVP-BEZ235, and ZSTK474 are substrates of murine Abcb1a (Mdr1a), human ABCB1 (MDR1), murine Abcg2, and/or human ABCG2, we analyzed the translocation of these compounds in concentration equilibrium transport assays (CETA) as described previously (22). To this end, we used the parental LLC pig-kidney cell line (LLC-PK1) and sublines transduced with murine Abcb1a (LLC-Mdr1a) or human ABCB1 (LLC-MDR1) and the parental Madine Darby Canine Kidney (MDCK) type II cell line (MDCKII-parental) and murine Abcg2 (MDCKII-Bcrp1) or human ABCG2–transduced sublines (MDCKII-BCRP). Cells were seeded on Transwell microporous polycarbonate membrane filters (3.0-μm pore size, 24-mm diameter; Costar Corning) at a density of 2 × 10⁶ cells per well in 2-mL MEM. When confluency was reached, the transport experiment was started by replacing the drug-free medium with Opti-MEM medium containing 0.5 μmol/L rapamycin, or MEM containing 1 μmol/L AZD8055, 0.5 μmol/L ZSTK474, or 0.5 μmol/L NVP-BEZ235. Zosuquidar (5 μmol/L) or Elacridar (5 μmol/L) was added when appropriate to inhibit either Abcb1 alone or both Abcb1 and Abcg2, respectively. [¹⁴C]-inulin (approximately 1.6 × 10⁶ DPM/mL) was added to check the integrity of the membrane. Samples of 50 μL were taken at 30, 120, 180, and 240 minutes and used for drug analysis.

U87 human glioma cells were seeded in 24-well plates and exposed to various concentrations of NVP-BEZ235 or ZSTK474. Upon reaching confluency in the control wells, all cells were fixed and stained using a glutaraldehyde (6% v/v; Sigma) and crystal violet (0.5% w/v; Sigma) solution. Plates were imaged using a Chemi-Doc MP system (Bio-Rad) and well confluency was measured using the ImageJ plugin ColonyArea as described elsewhere (23). Curves were plotted, fitted with a log(inhibitor) versus response - Variable slope (four parameters) curve and IC₅₀s were determined using GraphPad Prism 5.01 (GraphPad Software, Inc.).

For pharmacokinetic studies, rapamycin was first dissolved in 100% ethanol and further diluted in HBSS to yield a solution of 0.05 mg/mL. AZD8055 was dissolved in DMSO at a concentration of 10 mg/mL. For orally administration, ZSTK474 was mixed with water and sonicated, suspended in a mixture of hydroxypropyl methylcellulose (4%, v/v) and Tween 80 (polysorbate 80; 0.75%, v/v) to yield a drug suspension of 20 mg/mL. For intravenous administration, ZSTK474 was dissolved in DMSO to yield a solution of 10 mg/mL. NVP-BEZ235 was dissolved in DMSO to yield a solution of 5 mg/mL and further mixed with PEG400 to yield a solution of 1 mg/mL.

Rapamycin (1.5 mg/kg, i.p.), AZD8055 (10 mg/kg, i.v.), NVP-BEZ235 (10 mg/kg, orally), and ZSTK474 (10 mg/kg, i.v.; 200 mg/kg orally) were administered to WT and Abcb1a/b;Abcg2^−/− mice and/or Abcb1a/b^−/− and Abcg2^−/− mice. Elacridar, prepared as described previously (24), was given orally at a dose of 100 mg/kg 15 minutes prior to rapamycin. Blood sampling was performed either by collecting tail vein blood or by cardiac puncture at different time points. Brains were dissected after mice were sacrificed and were homogenized in 3 mL 1% (w/v) BSA. Both plasma and brain homogenates were stored at −20°C until analysis.

Rapamycin (MRM 936.6/409.2) was measured using LC-ESI-MS/MS with tacrolimus (826.5/616.2) as internal standard and protein precipitation with methanol for sample pretreatment. Samples of 5 μL were injected onto an Atlantis dC18 column (Waters) coupled with a Polaris 3 C18-A precolumn (Varian). The gradient elution of methanol ranged from 65% to 100% (v/v) in 1% (v/v) formic acid. Spray voltage was set to 3,900 V, capillary temperature to a 400°C and argon collision pressure to 2.0 mTorr. Collision energies were at 54 and 36 V, respectively.

AZD8055 (MRM 466.4/450.3) was measured using LC-ESI-MS/MS with buparlisib (411.3/367.2) as internal standard and ethyl acetate extraction as sample pretreatment. Samples of 50 μL were injected onto an Agilent C18 column (Agilent) coupled with a Phenomenex analytic securityguard C18 precolumn (Phenomenex). The gradient elution of methanol ranged from 65% to 100% (v/v) in 0.1% (v/v) formic acid. Declustering potential was 100 V and collision energy 59 eV.

For quantification of ZSTK474 in plasma and brain samples, a reversed-phase high-performance liquid chromatographic (HPLC) assay with fluorometric detection was used. Following tert-butyl methyl ether extraction, ZSTK474 and its internal standard NVP-BEZ235 were chromatographically separated using XBridge BEH130 C18 column (Waters) by isocratic elution with a mobile phase which consisted of acetonitrile, and 0.1% triethylamine adjusted with hydrochloric acid to pH 9.5 (50:50, v/v). Fluorescence detection was used with excitation and emission wavelengths of 240 and 425 nm, respectively. NVP-BEZ235 concentration in biological samples was measured using an HPLC assay as described previously (25).

The detailed procedures of stereotactic intracranial injection and bioluminescence imaging have been described previously (26). In short, FVB nude mice or p16^Ink4a/p19^Arf;K-Ras^v12;LucR, Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR, p53;p16^Ink4a/p19^Arf;K-Ras^v12;LucR or p53;Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR mice were anaesthetized and placed in a stereotactic frame, 10⁵ U87-luc cells or CMV-Cre lentivirus suspension in 2 μL was injected 2 mm lateral and 1 mm anterior to the bregma, 3 mm below the skull cap. After the initial tumor load was established, tumor development was monitored by bioluminescence using the IVIS 200 Imaging system (Xenogen Corporation). Mice were sacrificed when clear neurologic symptoms occurred or weight loss (≥ 20%) was observed. Brain tissue was fixed in an ethanol–glacial acetic acid mixture containing 4% formaldehyde (EAF), embedded in paraffin, and cut into coronal slices of 4 μm. Sections were H&E stained for verification of tumor growth.

For the U87 xenograft model drug intervention study, animals were stratified on the basis of bioluminescence signal at day 12 after tumor cell injection into three groups receiving daily vehicle control solution (orally 200 μL saline per mouse), NVP-BEZ235 (orally 10 mg/kg), or ZSTK474 (orally 200 mg/kg), respectively, until the day of sacrifice. For the spontaneous glioblastoma model drug intervention study, animals received a dose of 200 mg/kg every day of ZSTK474, starting 3 days after lentiviral injection until the day of sacrifice.

Brain tissue was fixed in 4% formaldehyde, paraffin-embedded, and cut into 4-μm coronal sections that were stained with H&E and for pAKT (D9E), pERK (D13.14.4E), and p4EBP1 (236B4; Cell Signaling Technology).

As previously described, the general linear model repeated-measures procedure was used to determine whether the basolateral-to-apical differences of the drugs levels were significantly increased by the factor of time and at which time point(s) these differences became significant (27). For in vivo pharmacokinetic experiments, pharmacokinetic parameters were calculated using an add-in program for Microsoft Excel PKSolver (28). To determine the differences of brain and plasma concentrations among multiple strains, one-way ANOVA with Bonferroni post hoc test was performed. Differences were considered statistically significant when P < 0.05. For in vivo efficacy studies, survival fractions were calculated using Kaplan–Meier method using GraphPad Prism 5.01 (GraphPad Software). The log-rank test was used to compare survival of groups.

To determine substrate affinity of rapamycin for Abcb1a (Mdr1a) and Abcg2 (Bcrp1), a concentration equilibrium transport assay (CETA) was performed using murine Mdr1a- and Bcrp1-transduced cells and their parental cell lines as described previously (29). Rapamycin was significantly translocated from the basolateral to apical (b-to-a) compartment by LLC-PK1 cells (Fig. 1A). Endogenously expressed porcine Abcb1 was responsible for this translocation, as translocation was abrogated when the specific Abcb1 inhibitor zosuquidar was added. Rapamycin b-to-a translocation in Abcb1a-overexpressing LLC-Mdr1a cells was more pronounced than in LLC-PK1 cells and was also completely inhibited by zosuquidar, showing that rapamycin is a good Abcb1 substrate. In contrast, rapamycin is not a substrate of Abcg2, as no b-to-a translocation was observed in Bcrp1-overexpressing MDCK cells (MDCK-Bcrp1).

Next, we assessed the impact of Abcb1 and Abcg2 on rapamycin pharmacokinetics using WT, Abcb1a/b^−/−, Abcg2^−/−, and Abcb1a/b;Abcg2^−/− mice. Surprisingly, administration of 1.5 mg/kg rapamycin orally resulted in 35- to 52-fold higher AUC_blood values in all different knockout mice compared with WT mice (Fig. 1B). However, no differences in rapamycin levels between WT and knockout mice were found in well-perfused organs such as the kidney or liver, where drug levels are usually well equilibrated with systemic blood levels. In the brain, Abcb1 is clearly a dominant factor because the brain AUC of rapamycin in Abcb1a/b^−/− mice was at least 10-fold higher than in WT mice. The levels in brain samples of WT mice were below the lower limit of quantification (LLQ) of the assay (10 ng/g), whereas the levels in brain samples from Abcg2^−/− mice taken at the earlier time points were above the LLQ (Fig. 1B), This is most likely due to the presence of some remnant blood in the homogenized brain tissue and the much higher blood levels of rapamycin in this strain. No significant difference was found between the Abcb1a/b^−/− and Abcb1a/b;Abcg2^−/− mice, suggesting Abcg2 is not involved in restricting rapamycin from the brain.

To further investigate whether Abcb1- and/or Abcg2-mediated transport is responsible for the differences of plasma and brain levels between WT and knockout mice, the dual Abcb1 and Abcg2 inhibitor elacridar was coadministrated to WT mice 30 minutes prior to intraperitoneal administration of rapamycin. Moreover, the LC/MS-MS assay was adapted to improve the LLQ and allow accurate quantification in brain samples of WT mice. Interestingly, elacridar enhanced the brain concentration of rapamycin in WT mice by about 23-fold, but did not cause any alteration of rapamycin levels in blood and kidney, indicating that Abcb1a/b and Abcg2 are not responsible for the differences observed in blood of knockout mice (Fig. 1C). Together, these results demonstrate that Abcb1 at the BBB profoundly impairs the brain penetration of rapamycin.

It has previously been reported that rapamycin is unstable in plasma and whole blood of human, rabbit, and rat (30). In contrast to plasma, rapamycin levels in the kidney and liver in our study were not different between WT and knockout mice. Therefore, we tested the stability of rapamycin in murine plasma ex vivo and indeed found a much more rapid degradation in plasma of WT compared with ABC-transporter knockout mice. After 6-hour incubation of rapamycin with freshly collected plasma from mice of different knockout strains, we observed that about 85% of the added rapamycin was recovered in plasma of Abcb1a/b^−/−, Abcg2^−/− and Abcb1a/b;Abcg2^−/− mice, but about 10% in plasma of WT mice (Fig. 1D). Recent work with rapamycin's analogue everolimus has demonstrated that carboxyl esterase 1c is more abundantly present in ABC-transporter knockout mice and responsible for enhanced stability in plasma (31). It is very likely that the higher blood level of rapamycin in knockout mice is similarly caused by avid binding of rapamycin to a serum factor, possibly Ces1c, which is more abundantly present in the plasma of knockout mice. Because of this avid binding, only a small fraction of rapamycin is available for degradation and for tissue distribution.

AZD8055 was transported in vitro by human and murine MDR1/Mdr1a and BCRP/Bcrp1 (Fig. 2A). Basolateral-to-apical translocation was observed in CETAs with Abcb1a, ABCB1, Abcg2, and ABCG2-overexpressing cell lines, but not parental control lines. Inhibition with zosuquidar or elacridar diminished translocation in all cases and validated AZD8055 transport by MDR1/Mdr1a and BCRP, although Bcrp1-mediated translocation was not completely inhibited by elacridar, indicative of very efficient transport.

Compared with WT mice, the AZD8055 brain concentrations were 3.7- (P < 0.01) and 7.7-fold (P < 0.0001) higher in Abcb1a/b^−/− and Abcb1a/b;Abcg2^−/− mice, respectively (Fig. 2B), whereas the levels in liver and kidney were not different (not shown). Interestingly, a similar plasma retention effect in ABC-transporter knockout mice was found as observed for rapamycin albeit that the effect size was much more modest. Consequently, the brain–plasma ratios do not correctly reflect AZD8055 brain penetration in WT mice versus the knockout mice. However, the increased ratio in Abcb1a/b;Abcg2^−/− mice in comparison with each of the single knockout strains clearly demonstrates that both Abcb1a/b and Abcg2 restrict the BBB penetration of AZD8055.

Using in vitro CETAs, neither LLC-PK1, LLC-Mdr1a, nor LLC-MDR1 cells displayed significant b-to-a translocation of NVP-BEZ235, suggesting that NVP-BEZ235 is not transported by murine or human Abcb1/ABCB1 (Fig. 3A). In contrast, NVP-BEZ235 was significantly translocated by MDCK-Bcrp1 cells, but not MDCK-parent or MDCK-BCRP cells, and this translocation was inhibited by elacridar, indicating that NVP-BEZ235 is transported by murine Abcg2, but not human ABCG2.

The roles of Abcb1 and Abcg2 in limiting the brain penetration of NVP-BEZ235 were investigated using WT and Abcb1a/b;Abcg2^−/− mice. We administered NVP-BEZ235 orally at a dose of 10 mg/kg. As shown in Fig. 3B, both plasma and brain levels of NVP-BEZ235 were significantly higher in Abcb1a/b;Abcg2^−/− mice than those in WT mice at multiple time points. Moreover, the brain-to-plasma ratio of Abcb1a/b;Abcg2^−/− mice was 2.0- (P < 0.01) and 1.6-fold (P < 0.01) higher than that of WT mice at 1 and 4 hours, respectively. These data suggest that the higher brain level of NVP-BEZ235 was due to the absence of active brain efflux mediated by Abcb1 and Abcg2 and not only a consequence of higher plasma levels in Abcb1a/b;Abcg2^−/− mice. Interestingly, in contrast to the results obtained from the CETAs, pharmacokinetic experiments with single transporter knockouts showed the brain penetration of NVP-BEZ235 to be predominantly impaired by Abcb1a/b (Fig. 3C). The brain concentration and brain-to-plasma ratio in WT mice were 3.6- (P < 0.05) and 1.7-fold (P < 0.01) lower than those of Abcb1a/b^−/− mice and 4.5- (P < 0.01) and 2.0-fold (P < 0.001) lower than those of Abcb1a/b;Abcg2^−/− mice, but similar to Abcg2^−/− mice. Similar to AZD8055, and to a lesser extent to rapamycin, the plasma levels in all knockout strains were higher than in WT mice.

To assess the pharmacodynamic implications of the impaired brain penetration of both NVP-BEZ235 and rapamycin, target engagement was compared between healthy brain tissue of WT and Abcb1a/b;Abcg2^−/− mice. At clinically achievable plasma exposures, both rapamycin and NVP-BEZ235 were not very efficient in reducing the phosphorylation levels of the downstream mTOR target pS6 in WT mice relative to control. In contrast, both compounds did inhibit pS6 phosphorylation in Abcb1a/b;Abcg2^−/− mice, with NVP-BEZ235 also reducing the phospho-AKT^S473 levels (Fig. 3D). In summary, these data show that Abcb1a/b at the BBB can impair rapamycin and NVP-BEZ235 brain penetration, resulting in hampered intracranial target engagement and thus likely restricting therapeutic efficacy.

No significant directional translocation of ZSTK474 was found in vitro, although a minor ZSTK474 concentration difference was observed at 4 hours in the MDCK-Bcrp1 CETA (Fig. 4A). These data indicate that ZSTK-474 is not transported by Abcb1/ABCB1 or Abcg2/ABCG2.

As the ABC-transporter affinities of a drug can be underestimated when using in vitro CETA (see for example NVP-BEZ235), we further investigated the roles of Abcb1 and Abcg2 on brain and plasma pharmacokinetics by oral administration of 200 mg/kg of ZSTK474 to WT and Abcb1a/b;Abcg2^−/− mice. ZSTK474 plasma and brain concentrations as well as brain-to-plasma ratio were similar in WT and Abcb1a/b;Abcg2^−/− mice (Fig. 4B). Moreover, oral administration of ZSTK474 greatly diminished PI3K pathway signaling in brain of both WT and Abcb1a/b;Abcg2^−/− mice, suggesting that sufficient ZSTK474 was delivered to inhibit its target in the brain even in presence of Abcb1 and Abcg2 (Fig. 4C).

We repeated the brain penetration study with ZSTK474 at a lower intravenous dose (10 mg/kg). Again, we did not find any significant difference in plasma, brain concentration, or brain-to-plasma ratio between WT and Abcb1a/b;Abcg2^−/− mice, confirming that neither systemic clearance nor brain penetration are limited by Abcb1 and Abcg2 (Fig. 4D).

The antitumor efficacy of the brain penetrable inhibitors NVP-BEZ235 (IC₅₀ = 17 nmol/L; Fig. 5A) and ZSTK474 (IC₅₀ = 630 nmol/L; Fig 5B) was investigated in the U87 orthotopic xenograft model. Daily administration of NVP-BEZ235 (10 mg/kg orally) and ZSTK474 (200 mg/kg orally) profoundly reduced tumor growth compared with control (Fig. 5C), without diminishing body weight (data not shown). Possibly in line with the slightly lower effect on tumor growth inhibition, NVP-BEZ235–treated mice survival did not significantly differ from control mice, whereas ZSTK474 demonstrated a markedly longer survival than control (median survival 25.0 vs. 17.5 days; P = 0.0019).

The antiproliferative activities of NVP-BEZ235 and ZSTK474 were assessed in vitro using glioblastoma cells derived from spontaneous transgenic high-grade glioma mouse models as described previously (ref. 21;Supplementary Methods). These cells harbor a combination of genetic deletions that are common in human glioblastoma, including p16^Ink4a/p19^Arf, P53, and/or Pten together with an activated Ras–MAPK pathway. Both agents showed a dose-dependent growth-inhibitory activity in line with target inhibition against all glioblastoma cell lines, with NVP-BEZ235 being approximately 10-fold more potent than ZSTK474 (Supplementary Fig. S1). Pten deficiency did not render these glioblastoma cell lines sensitive to PI3K pathway inhibition, as no clear difference in response was found between the different genotypes.

Traditional xenograft models such as the U87 model fail to recapitulate many important features of human glioma, including BBB integrity. We previously established spontaneous high-grade gliomas models using conditional mice of different genetic backgrounds: viz. p16^Ink4a/p19^Arf;K-Ras^v12;LucR, Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR, p53;p16^Ink4a/p19^Arf;K-Ras^v12;LucR and p53;Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR (21). All of these mice spontaneously develop grade III or IV gliomas after intracranial CMV-Cre lentivirus injection that exhibit many histopathologic and biological features of high-grade gliomas. In comparison, our glioblastoma models that are deficient for PTEN proliferated much more rapidly in vivo than PTEN-proficient p16^Ink4a/p19^Arf;K-Ras^v12 tumors (Fig. 6A). As a result, p16^Ink4a/p19^Arf;K-Ras^v12 mice that develop gliomas survived significantly longer than Pten;p16^Ink4a/p19^Arf;K-Ras^v12 mice (median survival 45 vs. 26 days; P < 0.001). Similarly, the deletion of PTEN in mice with P53-null tumors also had an accelerated tumor onset and progression (p53;p16^Ink4a/p19^Arf;K-Ras^v12 and p53;Pten;p16^Ink4a/p19^Arf;K-Ras^v12 gliomas; 28 vs. 20 days, respectively; P < 0.0001). These data suggest that activation of the PI3K pathway by PTEN deletion signaling is important in the transformation of cells into high-grade gliomas. We, therefore decided to treat Pten;p16^Ink4a/p19^Arf;K-Ras^v12 mice daily with 200 mg/kg of ZSTK474, starting 2 days after lentiviral injection. In line with these expectations, ZSTK474 treatment delayed Pten;p16^Ink4a/p19^Arf;K-Ras^v12;LucR tumor onset. However, ZSTK474 was not able to reduce the proliferation rate of these tumors at a more advanced stage (Fig. 6B). Overall, ZSTK474 treatment increased median survival from 26 to 32 days (P = 0.0002). Concordantly, PI3K signaling in these tumors was markedly reduced 2 hours after ZSTK474 treatment, while MAPK signaling was unaffected (Fig. 6C). These results suggest that glioma PI3K signaling can be inhibited by ZSTK474 in vivo, leading to prolonged survival.

This study demonstrates that PI3K inhibitors that are sufficiently brain penetrable can reach levels in glioblastoma that are sufficient for target inhibition and growth delay. Importantly, for ZSTK474, these effects occur at dose levels providing plasma concentrations that are in the range of those achievable in patients (32). However, this work using our spontaneous PTEN-deficient glioblastoma model also shows that inhibition of the PI3K–mTOR pathway alone has very modest activity, most likely due to the fact that, next to the PI3K-mTOR cascade, several other signaling pathways are concomitantly active in glioblastoma.

As manifested by the number of ongoing clinical trials, small-molecule targeted therapies are thought to hold promise for treatment of malignant gliomas. But to be efficacious, these agents need to penetrate the BBB. Besides the more central areas in glioblastoma where the vasculature is leakier, glioblastoma also harbors many tumor cells that have invaded into adjacent normal brain tissue where the BBB is still more or less intact. By using in vitro and in vivo mouse models, we determined the impact of Abcb1 and Abcg2, two well-established drug efflux transporters expressed at BBB, on the brain penetration of rapamycin, AZD8055, NVP-BEZ235, and ZSTK474. Importantly, for all in vivo work with these agents, we have taken the clinically achievable plasma levels into consideration.

Rapamycin is the prototype inhibitor of mTOR and has been tested in clinical trials against glioblastoma without success (33). Because of the Abcb1-mediated brain efflux, the brain concentration of rapamycin (Fig. 1C) remained very low in mice receiving a dose that has previously been used in other preclinical models of cancer (34, 35). In patients, the MTD of daily oral rapamycin is 6 mg/day and results in whole blood levels of about 30 ng/mL (36), thus in the same range as we achieved in WT mice. At this plasma level, however, the brain penetration was insufficient for target inhibition (pS6^S235/S236) in normal brain tissue of WT mice. Similarly, Mendiburu-Elicabe and colleagues already showed that neither target inhibition nor improved survival was achieved in the U87 xenografts even at a higher dose of 3 mg/kg of rapamycin, while further dose escalation caused severe weight loss (37). In patients undergoing resection of their tumor following rapamycin treatment, the levels in tumor tissue were also in the same range as in the WT mice (38). Rapamycin is not a substrate of Abcg2. The higher level of rapamycin measured in the brain homogenates of Abcg2 mice is a consequence of the high plasma level. The brain-to-plasma ratio is 0.04, which is close to the fraction of blood that is present in the brain vasculature and that will end up in the brain homogenate (39). Similarly, as shown before for the rapamycin analogue everolimus, the much higher levels of rapamycin in the plasma of the knockout strains is most likely due to the instability of rapamycin in plasma (31) and binding of rapamycin to carboxyl esterase 1c (Ces 1c), which protects rapamycin from degradation. Higher plasma levels in knockout mouse strains were also observed with AZD8055 and NVP-BEZ235, albeit that the effect size was much smaller. This may be due to a similar higher binding to a plasma protein that is more abundant in knockouts. Because of this higher plasma protein binding, the brain-to-plasma ratio will underestimate the brain retention. For rapamycin, the brain-to-blood ratio in Abcb1a/b and Abcg2;Abcb1a/b knockout mice is about 0.10. The Abcb1/Abcg2 inhibitor elacridar is able to increase the brain level of rapamycin without the increasing the blood level, which results in an increased brain-to-blood ratio from 0.18 to 5.2 in WT versus WT + elacridar animals, respectively. Notably, rapamycin is also strongly retained in brain tissue of Abcb1a/b and Abcg2;Abcb1a/b knockout mice and causes a profound target inhibition (pS6^S235/S236) in these strains (Fig. 3D).

Besides insufficient drug exposure, resistance to rapamycin can also be due to the fact that this compound inhibits the formation of the mTORC1 complex, while still causing feedback activation by phosphorylation of AKT at Ser473 by mTORC2 (10, 11). This feedback loop activation of AKT can be negated by using a dual mTORC1/2 inhibitor, like AZD8055, or a PI3K/mTOR inhibitor, like NVP-BEZ235. AZD8055 appeared to be a very good substrate of both Abcb1 and Abcg2, which is in line with previous work showing that AZD8055 (10 mg/kg twice daily) alone was inactive against intracranial glioma (40). Because the plasma levels in mice receiving 10 mg/kg are also considerably higher than can be achieved in patients receiving the MTD (40, 41), we have not further evaluated this compound in vivo. Interestingly, the in vitro Transwell experiments for NVP-BEZ235 did not show transport by Abcb1 and relatively weak transport by Abcg2. In line with these in vitro findings, the impact of Abcg2 on the brain penetration of NVP-BEZ235 was minimal based on the difference between Abcb1a/b^−/− and Abcb1a/b;Abcg2^−/− mice. However, Abcb1 significantly decreased the brain concentration and caused incomplete target inhibition in the brains of WT mice. Discordance between in vitro Transwell and in vivo brain penetration has been observed before (29). In general, the in vitro transport model is very effective to identify substrates as shown for rapamycin and AZD8055, but a lack of transport does not always accurately predict the impact of Abcb1 at the BBB. Consequently, in vivo studies remain necessary to accurately assess the role of this transporter in drug delivery to the brain. In the case of ZSTK474, the absence of transport by Abcb1 and/or Abcg2 in vitro nicely reflected the in vivo negligible impact at the BBB. The brain penetration of ZSTK474 was similar between WT and Abcb1a/b;Abcg2^−/− mice with a brain-to-plasma ratio of 1.5 and profoundly reduced AKT^S473 phosphorylation even in brains of WT mice. Importantly, this target inhibition occurred at plasma levels that are in the range as those achievable in patients, given that the first-in-human phase I study reported that the plasma level of ZSTK474 in patients who received the MTD (150 mg/kg/day for 21 consecutive days) ranged between 200 and 500 ng/mL (32).

On the basis of these results, we assessed the in vivo efficacy of ZSTK474 against orthotopic U87 glioblastoma and included NVP-BEZ235 as a reference, as previously studies have shown that NVP-BEZ235 is active against ectopic (subcutaneous) and orthotopic U87 when given at a daily oral dose of 25 or 45 mg/kg (19, 42). We also found a modest efficacy of NVP-BEZ235 against intracranial U87, even at a lower dose of 10 mg/kg. The plasma concentration in our WT mice was 200 ng/mL in line with the 1,000 ng/mL (peak) and 15 ng/mL (trough) plasma level at a dose of 50 mg/kg as reported by Maira and colleagues (42). Notably, this appears to be considerably lower than the 1,800 ng/mL found in patients as has been presented in a meeting report (43). No other (full) reports on the plasma pharmacokinetics of NVP-BEZ235 have been published, but apparently patients tolerate higher plasma levels than mice, and the efficacy of NVP-BEZ235 will likely be better at higher doses. On the other hand, intracranial U87 gliomas have a very open (leaky) vasculature, which renders the tumor highly accessible for systemically administered drugs. Although it is being used a lot in preclinical research, it is not the most predictive model for clinical efficacy against glioma. Therefore, we decided to test ZSTK474 in one of our spontaneous transgenic glioma models (21). We have used the Pten;p16^Ink4a/p19^Arf;K-Ras^v12 spontaneous tumors because PTEN-deficient tumors developed much more rapidly than PTEN-proficient tumors in p16^Ink4a/p19^Arf;K-Ras^v12mice. On the basis of this finding, we expected that inhibiting the PI3K pathway in Pten;p16^Ink4a/p19^Arf;K-Ras^v12 mice carrying spontaneous glioblastoma may extend their survival to be comparable as p16^Ink4a/p19^Arf;K-Ras^v12 mice. Indeed, we found that treatment with ZSTK474 significantly improved survival and caused inhibition of AKT^S473 and pS6^S235/S236 in tumors and surrounding brain. However, the response was much more modest than might have been expected on the basis of the notion that PTEN loss is an important factor driving this model. Glioblastoma is characterized by multiple parallel aberrant signaling pathways (1), and therefore, single-target treatment efficacy will be limited. Combinations with other drugs targeting these other pathways (e.g., MAPK, CDK4/6-Rb) may be necessary. Obviously, these drugs will also need to cross the BBB sufficiently to reach tumor cells in pharmacologically relevant levels. In conclusion, its excellent brain penetration renders ZSTK474 the most promising candidate among the PI3K pathway inhibitors included in this study for future multiple pathway–targeting studies.

No potential conflicts of interest were disclosed.

Conception and design: M.C. de Gooijer, J.H. Beijnen, O. van Tellingen

Development of methodology: F. Lin, M.C. de Gooijer, R.W. Sparidans, O. van Tellingen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): F. Lin, M.C. de Gooijer, D. Hanekamp, G. Chandrasekaran, R.W. Sparidans, J.H. Beijnen, O. van Tellingen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): F. Lin, M.C. de Gooijer, D. Hanekamp, L.C.M. Buil, N. Thota, O. van Tellingen

Writing, review, and/or revision of the manuscript: F. Lin, M.C. de Gooijer, J.H. Beijnen, T. Wurdinger, O. van Tellingen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L.C.M. Buil, O. van Tellingen

Study supervision: F. Lin, O. van Tellingen

Other (performed some of the in vitro and in vivo studies): N. Thota

This work has been supported by a grant of the foundation StopHersentumoren.nl (to O. van Tellingen).

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.