Development, Characterization, and Reversal of Acquired Resistance to the MEK1 Inhibitor Selumetinib (AZD6244) in an In Vivo Model of Childhood Astrocytoma

Astrocytomas are the most common tumors of the central nervous system in children and are subdivided according to histologic subtypes, grades 1–4 (1). The majority of astrocytomas are grade 1 and 2 tumors, yet the biologic behavior of low-grade astrocytomas reflect a heterogeneous spectrum. Radical surgical resection is the standard therapy; however, whereas most cerebellar pilocytic astrocytomas (WHO I) do not frequently involve the diencephalon, those that do are not resectable due to the tumor's involvement in the diencepahlon and surrounding eloquent structures. Thus, adjuvant therapy is warranted, which includes chemotherapy and/or radiation therapy. The 5-year progression-free survival rate for chemotherapy plus radiotherapy has been reported as 68%, which is superior to chemotherapy alone 38% (2). However, significant morbidity is associated with the presence of residual tumor and the current therapy that includes neuroendocrine cognitive deficits, visual deficits, vasculopathy, and secondary tumors (3, 4). Moreover, the metastatic potential and transformation to a high-grade astrocytoma further contributes to the poor prognosis (5, 6). In recent years, there have been considerable advances in defining subsets of pediatric tumors by genotyping and expression profiling (7–9). Whereas histologically, astrocytomas in children and adults are similar, childhood astrocytomas are distinct clinical entities from those in adults and are not associated with many of the critical genetic alterations found in the adult astrocytomas. With the possible exception of TP53 mutations, frequent genetic alterations detected in adult astrocytomas have been identified at lower frequencies in childhood astrocytomas (10–13). For pediatric astrocytomas, low-grade tumors are associated with activation of BRAF through a tandem duplication that results in the KIAA1549–BRAF fusion (14) or through an activating point mutation of BRAF (predominantly V600E). More recent data suggest that the KIAA1549–BRAF fusion is restricted to grade 1 pilocytic astrocytoma (100%), whereas BRAF^V600E occurs more frequently in grade 2–4 gliomas (∼23%; although lower frequencies have been reported; ref. 15) and in 60% of xanthoastrocytomas (15, 16). Thus, activating mutation of BRAF appears to be the most common genetic alteration in intermediate-grade astrocytoma. Homozygous deletion of the CDKN2A locus is frequent (∼70%) in tumors harboring the BRAF^V600E mutation (17). Mutations in PIK3CA are reported to be rare in these tumors (18). Findings for BRAF mutation, similar to other tumors with activated BRAF (e.g., melanoma), suggest that activated BRAF may provide a potential drug target (19).

BRAF is a component of the mitogen-activated protein kinase (MAPK) signaling pathway that induces multiple proliferative or differentiation signals within tumor cells (20, 21). In many adult carcinomas, MAPK activation occurs through activating mutations in RAS or RAF. The frequency of BRAF^V600E mutations range from ≥90% in hairy cell leukemia and 60% to 80% in melanoma to around 10% in colon cancer with other tumors in between (22–28). Cell lines harboring BRAF^V600E may be highly sensitive to MEK inhibition, and these agents may have significant use against melanoma and other tumors with similar mutations (19). As part of the Pediatric Preclinical Testing Program (PPTP), we recently reported the activity of the MEK1/2 inhibitor selumetinib (AZD6244) against 43 in vivo pediatric tumor xenografts and 23 cell lines. At the highest concentration used in vitro (10 μmol/L), selumetinib only inhibited growth by 50% in 5 of the 23 cell lines. Against the in vivo tumor panels, selumetinib induced statistically significant differences in event-free survival (EFS) distribution in 10 of 37 (27%) solid tumor models and 0 of 6 acute lymphoblastic leukemia (ALL) models. However, even at best, growth inhibition was modest with EFS increased more than 2-fold in 3 tumor models (29). There were no objective responses. Pharmacodynamic studies indicated that, at the dose and schedule used, selumetinib completely inhibited ERK1/2 phosphorylation in nonresponding tumors. These data imply that MEK signaling is not required for cell proliferation or survival in most of these pediatric tumor models or that there is rapid compensation and activation of alternative signaling pathways.

In the secondary panel of tumors in the PPTP, we have developed 2 astrocytomas as direct patient/mouse heterografts. Characterization of these models showed BT-35 tumor cells have 2 to 5 copies of wild-type BRAF, whereas BT-40 tumor cells have 5 copies of activated BRAF^V600E (29). Further characterization of BT-40 revealed loss of the CDKN2A locus, and wild-type p53, typical of many grade 2 astrocytomas (17). BT-40 xenografts were highly sensitive to selumetinib, regressing completely during the period of treatment (6 weeks), but recurring in a dose-dependent manner after treatment was finished. In contrast, BT-35 xenografts progressed on selumetinib treatment without any significant drug-induced growth delay.

Resistance to MEK and BRAF inhibitors has been extensively studied in the context of adult cancers (30–35), as has the paradoxical activation of the BRAF pathway by the BRAF inhibitor PLX4032 in the cellular context of wild-type BRAF (36). However, as stated above, the genetic features of childhood astrocytomas may be very different from most adult epithelial-derived cancers. Consequently, development of resistance, or development of therapies that retard or prevent emergence of resistance, may be context-specific. Our studies, presented here, would support a mechanism of resistance that is unstable (hence unlikely through genetic mutation) not previously identified in BRAF mutants (31, 35, 37, 38) and mediated by activation of STAT3 signaling in BT-40 xenografts resistant to selumetinib treatment.

RPMI media, FBS, and Alamar Blue (AB) were purchased from Invitrogen. Anti-CD34 antibody (ab27448) was from Abcam. Terminal deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) detection kit was purchased from GenScript USA Inc. Human Cytokine Array from R&D systems (Cat no ARY005). Primary antibodies to β-actin, glyceraldehyde-3-phosphate dehydrogenase (GAPDH), phosphorylated STAT3 (Y705), STAT3, phosphorylated ERK, and extracellular signal–regulated kinase (ERK) were purchased from Cell Signaling Technology and Ki67 antibody was purchased from DAKO.

Selumetinib was provided by AstraZeneca. Selumetinib was dissolved in 0.5% hydroxypropyl methyl cellulose, 0.1% Polysorbate 80 and administered per os, using a twice daily schedule [excepting weekends, for which a once daily (SID) schedule was used] for a scheduled 6 or 12 weeks at a dose of 75 mg/kg.

LLL12 was synthesized in the Laboratory of Dr. Pui-Kai Li (College of Pharmacy, The Ohio State University, Columbus, OH). The powder was dissolved in sterile dimethyl sulfoxide (DMSO) to make a 5 mg/mL stock solution. Aliquots of the stock solution were stored at −20°C. LLL12 was administered by intraperitoneal (i.p.) injection at 5 mg/kg daily, for 6 weeks.

AM38c1 and DBTRG-05MG glioblastoma cell lines (generously provided by T. Nicolaides, UCSF) were maintained in antibiotic-free RPMI (Invitrogen) in high glucose supplemented medium with 10% FBS and 2 mmol/L glutamine at 37°C with 5% CO₂. All cells were maintained as subconfluent cultures and split 1:3, 24 hours before use. Experiments were carried out within 10 passages of receiving the cells.

Genomic DNA was extracted using a Puregene DNA Isolation Kit (Gentra Systems) and quantified spectrophotometrically. The genomic DNA from BT-40 xenografts and cell lines AM38c1 and DBTRG-05MG were screened for BRAF mutations with primers designed to amplify the exons 1 to 18 using primers described previously (24). Big Dye Terminator Chemistry was used for sequencing.

Total RNA was extracted from three independent tumors per condition (parental, resistant, or revertant) before treatment, or at 72 hours of AZD6244 treatment, using TRIzol (Life Technologies)/chloroform/ethanol precipitation, as per the manufacturer's instructions. Before cDNA synthesis, 5 μg of total RNA per condition was treated with 2.5U of DNase I (Life Technologies) to remove any contaminating genomic DNA. The reverse transcription reaction contained 2.5 μg of pure total RNA, 500 ng of anchored oligo(dT) primer (Life Technologies), as well as 25 ng of random hexamer primers (Life Technologies) to reverse transcribe the control 18S ribosomal RNA. The resulting cDNA was purified using QIAquick PCR cleanup columns (Qiagen) and quantified by spectrophotometry at 260 nm. Each real-time PCR reaction contained 75 ng of the purified cDNA brought up to a final volume of 9 μL. To this, 10 μL of the TaqMan master mix was added, along with 1 μL of the transcript specific or 18S TaqMan probe. All aliquot pipeting was conducted on a Biomek 3000 (Beckman Coulter) to ensure accuracy. Quantitative reverse transcriptase (RT-PCR) amplification was conducted on an ABI 7300 Real Time PCR System, under the following conditions: 2 minutes at 50°C (uracil DNA glycosylase activation), 10 minutes at 95°C, and 40 cycles of 95°C for 15 seconds and 60°C for 1 minute. Relative gene expression was calculated using the ΔΔC_T method, with each C_T value for the probes set relative to the internal 18S control C_T value. The ΔC_T values were then compared between the conditions.

Total RNA was prepared from snap-frozen subcutaneous BT-40 tumor xenografts using the RNeasy Kit (Qiagen). Gene expression analysis was done in the Medical Genomics Core Laboratory, Nationwide Children's Hospital (NCH), using the Affymetrix HG-U133Plus2 GeneChip (54,613 probe sets). RNA quality was confirmed by UV spectrophotometry and by analysis on the Agilent 2100 Bioanalyzer. Processing of RNA samples was done according to the Affymetrix gene expression protocol. Expression signals were calculated using the MAS5 statistical algorithm within the Affymetrix GCOS software (version 1.4). Signal values were scaled using the global normalization method with the 2% trimmed mean set to 500. Detection calls for each transcript (absent, marginal, or present) were determined using the default variables within the GCOS software.

Four BT-40 tumor replicates (4 per condition), each from independent mice, were measured using the Agilent Sureprint G3 version 1 human gene-expression array. The biologic replicates were used to measure differences between conditions given xenograft biologic variability. Gene set enrichment analysis (GSEA; ref. 39) was applied to measure the overall activity of an MEK activity and compensatory signature. To gauge significance of enrichment, we used GSEA v2.08 desktop application (39, 40) and applied default algorithm settings, which included 1,000 gene set permutations, condition difference by signal-to-noise, and a weighted enrichment statistic.

Cell lysis, protein extraction, and immunoblotting were as described previously (41). Immunoreactive bands were visualized by using Super Signal Chemiluminiscence substrate (Pierce) and Biomax MR and XAR film (Eastman Kodak Co.). Fifteen microliters of total sample was resolved on a 4% to 12% SDS-PAGE. Proteins were transferred to a polyvinylidene difluoride (PVDF) membrane, and immunodetection was conducted with specific primary antibodies.

AM38c1 and DBTRG-05MG cells were seeded on 6-well plates at a density of approximately 1 × 10⁵ cells per well in their respective media. Cells were treated with 10 nmol/L to 10 μmol/L AZD2644 or LLL12 (100 nmol/L to 5 μmol/L) to determine IC₅₀ concentrations. For combination studies, cells were exposed to 10 μmol/L selumetinib combined with 2 μmol/L LLL12 one day after seeding. After 2 days, Alamar Blue (AB) was added directly into culture media at a final concentration of 10% and the plates were incubated at 37°C. Optical density was measured spectrophotometrically at 540 and 630 nm within 3 to 4 hours after adding dye. As a negative control, AB was added to medium without cells.

CB17SC-M scid^−/− female mice (Taconic Farms) were used to propagate subcutaneously implanted BT-40 tumors. All mice were maintained under barrier conditions and experiments were carried out using protocols and conditions approved by the institutional animal care and use committee of NCH. Mice were randomized into groups of 10 when tumors were 100 to 200 mm³. Tumor volumes (cm³) were determined as previously described (42).

All animal experiments were carried out in accordance with institutional animal care and use committee of The Research Institute at NCH-approved protocols, designed to minimize the numbers of mice used and to minimize any pain or distress. The named institutional review board or ethics committee specifically approved this study.

Proteome profiler antibody array (R&D Systems; Cat no ARY005) was used according to manufacturer's instructions to detect the relative levels of expression of 36 cytokines, chemokines and acute phase proteins in a single sample in control and treated tumors. After blocking, the membranes 300 μg of protein from the tumor lysate from control and selumetinib-treated BT-40 (drug-sensitive), BT-40/AZD (drug-resistant), and BT-40/REV (drug-revertant) groups were added and incubated for overnight at 4°C. After 16 to 18 hours, the membranes were washed and streptavidin/horseradish peroxidase (HRP) was added or 30 minutes. Immunoreactive signals were visualized by using Super Signal Chemiluminiscence substrate (Pierce) and Biomax MR and XAR film (Eastman Kodak Co.). Array data on developed X-ray film were quantified by scanning the film using Biorad Molecular Image Gel Doc XR+ and data using were analyzed using Image Lab software.

For immunohistochemistry, formalin-fixed tumors were sectioned at 5 μm and then dewaxed and soaked in alcohol. After microwave treatment in antigen unmasking solution (Vector Laboratory) for 10 minutes, endogenous peroxidase activity was inactivated by incubating in 3% hydrogen peroxide (H₂O₂) for 15 minutes, and sections were incubated with primary antibody in PBS at 4°C overnight. After washing with PBS, immunostaining was conducted using the Vectastain Universal Quick Kit and DAB Peroxidase Substrate Kit (Vector Laboratories) according to the manufacturer's instructions. Antiserum was omitted in the negative control. The number of cells staining positive was counted by a blinded observer in 5 random 40× fields and treated versus controls compared (Student t test). Images were obtained with an Olympus AX70 fluorescence microscope and Spot v2.2.2 (Diagnostic Instruments) digital imaging system. Proliferation was measured by Ki67 staining. Apoptosis was detected in deparaffinized tumor sections by TUNEL assay using the In Situ Cell Death Detection Kit (Gene Script USA Inc.) according to manufacturer's instructions.

The VE1 antibody that can differentiate between wild-type and mutant BRAF (43) was used to determine expression of mutant BRAF. Total BRAF (wild-type plus mutant) was detected using rat monoclonal antibody pBR1 (43). The expression levels of mutant BRAF^V600E and total BRAF were assessed in parental BT-40, in BT-40/AZD derivatives (A1 and B4) lines, and BT-40/REV revertant tumor models. FFPE tissues from each individual xenograft tumor were processed as previously described (43). Paraffin blocks or slides were coded and scored “blinded.”

Data are presented as mean ± SE. Differences were analyzed with the Student t test, and significance was set at P < 0.05. Unpaired t tests were used for the continuous variables comparisons, such as expression level differences shown in Fig. 2D. For in vivo testing against pediatric glioma xenograft models, criteria for defining an event (4 times the tumor volume at the start of treatment) were similar to that used by the PPTP (20). Log-rank test was used to compare the time-to-event curves between groups and Holm's method was used to adjust for multiplicity within each xenograft model. SAS 9.3 was used for this analysis (SAS, Inc.).

The BT-40 astrocytoma was derived from a tumor diagnosed as a juvenile pilocytic astrocytoma (JPA; ref. 29). As a xenograft, Ki67 staining showed a proliferating fraction that was high for a JPA with Ki67 index of 16.8 ± 7.2, 16.2 ± 7.1, and 4.26 ± 6.1 on passages 5, 10, and 20, respectively. In addition, BT-40 is heterozygous for mutant BRAF^V600E, suggesting that it is a low-grade astrocytoma rather than a JPA. Our initial screen of selumetinib against the solid tumor and ALL panels within the PPTP showed minimal single-agent activity despite clear evidence of MEK inhibition (42). In contrast, BT-40 astrocytoma xenografts that harbor the BRAF^V600E mutation were highly sensitive to treatment with selumetinib, completely regressing during the period of treatment (6 weeks). However, although treatment of BT-40 xenografts induced complete tumor regression, tumors regrew after discontinuing therapy (Supplementary Fig. S1). Two tumors designated A5 and B4 were selected for further transplant and retreatment. As shown in Fig. 1A, A5 tumors were essentially resistant to selumetinib treatment on passage 2, whereas B4 tumors showed some degree of response but then progressed during selumetinib treatment (Fig. 1B). Tumor line A5/A2 was further transplanted in selumetinib-treated mice. In addition, A5/A2 was transplanted and maintained in untreated mice for 3 passages, then retested for sensitivity to selumetinib. As shown in Fig. 1C, after 3 passages in untreated mice, the A5/A2 tumor was somewhat sensitive to treatment with selumetinib, whereas by the fifth passage, it had regained full sensitivity to selumetinib treatment (Fig. 1D). BT-40 parental, the A5/A2-resistant line (designated BT-40/AZD) and the revertant line from passage 5 in untreated mice (BT-40/REV) were further characterized.

Several studies have suggested that the BRAF/MEK pathway becomes reactivated in cell lines selected for resistance to BRAF and MEK inhibitors (31, 33–35). We examined inhibition of MEK by selumetinib in BT-40 (Fig. 2A) and BT-40/AZD (Fig. 2B) xenografts. Inhibition of MEK signaling resulted in significant hyperphosphorylation of MEK in both sensitive and resistant tumor models, whereas phospho-ERK1/2 was markedly inhibited in both tumor lines by selumetinib treatment. Quantitation of phospho-MEK and phospho-ERK2 is shown for BT-40 xenografts in Fig. 2C. These data are consistent with reduced ERK-mediated negative feedback on RAS or upstream RTKs (44). Phospho-ERK1/2 was also significantly reduced by selumetinib treatment; however, in resistant tumors, there was a statistically significant increase in phosphorylation of ERK1 compared with that in BT-40 xenografts after treatment (Fig. 2D), although there was no substantial difference in the level of inhibition of the MEK/ERK pathway that could account for resistance to selumetinib.

BT-40 and derivative tumor lines are heterozygous for mutant BRAF^V600E (29). We wondered, initially, whether selumetinib-selected resistant sublines expressed predominantly wild-type BRAF and hence became insensitive to MEK inhibition as we found for other tumor models (29). To assess expression of BRAF^V600E protein, tumor sections were processed and stained using VE1 antibody that reacts only with the mutant protein (43). As shown in Supplementary Fig. S2, staining for mutant BRAF was generally homogeneous with moderate intensity in both parental and AZD-resistant xenografts. The intensity of staining was slightly less in the BT-40/REV tumors that had reverted to sensitivity to selumetinib. Genomic profiling showed that mutant and wild-type BRAF alleles were represented in equal proportions in all tumor sublines and also showed no amplification of the BRAF locus in BT-40 or BT-40/AZD xenografts.

Gene expression changes between sensitive and resistant tumors were examined in xenografts carefully staged to be between 200 and 300 mm³ at time of harvest. Four replicates from independent mice for the sensitive (BT-40) and resistant (BT-40/AZD) groups were profiled using the Agilent platform. Variance between replicate tumors in each group was small [coefficient of variation (CV) < 5%]. We determined whether the “MEK-Functional” signature, derived predominantly from cell culture experiments (45), distinguished between parental BT-40 and BT-40/AZD tumors. Enrichment of transcriptional pathway signatures was proposed to predict “addiction” to MEK signaling and response to selumetinib. Rather surprisingly, the “MEK-Functional” expression signature was enriched in BT-40/AZD tumors resistant to selumetinib relative to the parental tumor. As shown in Fig. 2E, we observed that genes driven by MEK activity (45) were significantly enriched (NES = 1.55, ES = 0.53, nominal P = 0.045, FDR q = 0.13, FWER P = 0.061) in BT-40/AZD tumors relative to parental BT-40 tumor. The core enrichment was largely driven by SLCO4A1 (signal-to-noise = 0.51) and DUSP4 (signal-to-noise = 0.50) as well as supported by 6 additional core enriched genes (LGALS3, PROS1, SERPINB1, DUSP6, MAP2K3, and SPRY2) with moderate signal-to-noise (0.11–0.19).

Microarray data indicated that the MEK functional activation signature was significantly enriched, whereas the compensatory resistance signature was not enriched in the BT-40/AZD–resistant tumors. qRT-PCR was carried out using TaqMan probes specific for the genes in the signatures. (Note that a probe for the compensatory gene, STAC, was unavailable.) The purpose of this analysis was 3-fold: to determine the validity of the microarray results, to examine the effects of selumetinib treatment on gene expression, and to incorporate the revertant tumor line gene expression into the data set. From the microarray results, 3 MEK functional activation genes were found to be upregulated in the resistant BT-40 tumors relative to the parental tumors: DUSP4, LGALS3, and SLCO4A1. The ΔΔC_T values calculated from the qRT-PCR results confirmed that in the resistant tumors, the same 3 genes were driving the enrichment of the MEK signature, with values of 138.8-, 70.5-, and 9.2-fold elevation in BT-40/AZD tumors, respectively (Supplementary Table S1). The qRT-PCR results of the compensatory signature differed from the microarray results; in that the microarray showed 2 genes, COL5A1 and LOX, were significantly downregulated, and 2 genes, interleukin-6 (IL-6) and SERPINE1, were significantly upregulated in the resistant tumors. It should be noted, however, that LOX [fold change (FC) = −2.0] and SERPINE1 (FC = 2.5) were barely above the significance cutoff values of ±2.0. From the qRT-PCR results, only IL-6 was significantly upregulated in the resistant tumors, with a ΔΔC_T value of 7.45 (Table 1).

To investigate the effects of MEK1/2 inhibition on the signature gene expression, mice with BT-40- and BT-40/AZD–resistant xenografts were treated with twice daily with selumetinib. Overall, when qRT-PCR was carried out on tumors harvested at 72 hours, gene expressions did not differ greatly when the BT-40- and BT-40/AZD–resistant tumors were compared with their untreated cohorts for either “MEK-Functional” or “compensatory” signatures (Supplementary Tables S1 and S2). It did show, however, that the same 3 genes, DUSP4, LGALS3, and SLCO4A1, that were driving MEK signature enrichment remained upregulated in the resistant tumors under conditions of MEK inhibition. Interestingly, the expression of the compensatory signature gene, IL-6, also remained significantly upregulated in the resistant tumors compared with the BT-40 xenografts (ΔΔC_T = 5.1) after treatment. This increased expression was maintained even though IL-6 was shown to be significantly downregulated in both the parental and AZD-resistant tumors after MEK1/2 inhibition, with ΔΔC_T values of 0.09 and 0.05, respectively (Fig. 2F). qRT-PCR analysis of untreated BT-40 and BT-40/REV (revertant) tumors showed that the MEK functional and compensatory resistance gene signatures differed very little; the exception being that IL-6 gene expression was reduced approximately 10-fold in the revertant tumors compared with the parental BT-40 xenografts (Table 1). Also, as seen in the BT-40 tumors, expression of DUSP4, LGALS3, SLCO4A1, and IL-6 was significantly upregulated in the untreated resistant tumors versus the revertant tumors.

Upon administration of selumetinib, expression of the MEK functional signature genes in the untreated versus treated revertant tumors followed the same pattern as seen in the parental and resistant tumors (Supplementary Table S1). However, unlike the BT-40 drug-sensitive and BT-40/AZD–resistant tumors, the compensatory signature genes in the revertant tumors were greatly affected (Table 1). In the untreated BT-40 or BT-40/AZD tumors versus treated BT-40 and BT-40/AZD tumors, only FZD2 and IL-6 were shown to be differentially expressed due to MEK 1/2 inhibition. In contrast, in the revertant tumors, 9 of 12 genes in the signature were significantly up- or downregulated in response to drug administration. When the selumetinib-treated BT-40 tumors and treated revertant tumors were compared for changes in MEK signature genes, 7 of 16 genes were downregulated, and 7 of 12 compensatory genes were differentially expressed. Comparison of treated resistant versus treated revertant tumor gene signatures displayed the same pattern. Ten of the 16 MEK functional genes were significantly upregulated in the resistant tumors compared with the revertant and 5 of the 12 compensatory resistance genes were differentially expressed.

Both microarray and qRT-PCR analyses showed that IL-6 expression increased in the selumetinib-resistant xenografts and decreased upon reversion to sensitivity. To examine whether targeting MEK/ERK by selumetinib inhibited cytokine production, levels of cytokines in xenografts were evaluated using a Human Cytokine Protein Array II, an assay that can detect 36 different cytokines. Tumor lysates for the antibody array were derived from control and treated BT-40, BT-40/AZD and BT-40/REV tumors. Basal levels of IL-6 were elevated in BT-40/AZD xenografts compared with either parental BT-40 or BT-40/REV tumors (Fig. 3A). In untreated BT-40/REV xenografts, IL-6 was not detected. Relative to control BT-40 xenografts, selumetinib-treated tumors showed a dramatic decrease of IL-6 and IL-8. Similarly, treatment decreased IL-6 levels in BT-40/AZD xenografts. However, although IL-6 decreased during selumetinib treatment in BT-40/AZD tumors, the level in treated tumors was still similar to that in untreated parental BT-40 xenografts. Levels of individual cytokines, and their response to selumetinib, are presented in Fig. 3B. Consistent with increased IL-6 transcripts and protein, there was clear activation of the NF-κB pathway in BT-40/AZD tumors where phospho-Rel-A was detected. In contrast, phosphorylated Rel-A was not detected in BT-40 xenografts and was significantly reduced by selumetinib treatment in BT-40/REV tumors relative to the levels detected in BT-40/AZD tumors (Fig. 3C). Examination of IL-6Rα transcript levels by RT-PCR, using human-specific primers, showed a slight increase in BT-40/AZD over parental BT-40 xenografts, but a 4-fold decrease in receptor transcript levels in BT-40/REV tumors (Supplementary Fig. S3).

The above results suggested that IL-6 potentially could maintain STAT3 signaling leading to resistance. We examined the effect of selumetinib in 2 human astrocytoma lines with BRAF^V600E mutations. By sequencing, AM38c1 cells are heterozygous for the BRAF^V600Emutant whereas DBTRG-05MG has only the mutant gene (data not shown). Neither cell line was sensitive to selumetinib (IC₅₀ > 10 μmol/L; Fig. 4A). Increasing concentrations of selumetinib progressively inhibited phosphorylation of ERK1/2 with 50% inhibition at 5 μmol/L for AM38c1 cells and about 500 nmol/L for DBTRG-05MG cells (Fig. 4B). Of note, however, was a reciprocal increase in STAT3 phosphorylation (Y705) in both cell lines, suggesting that activation of STAT3 signaling could compensate for MEK inhibition (Fig. 4B). Thus, activation of STAT3 signaling could represent a mechanism for both intrinsic and acquired resistance to selumetinib. To test the effect of inhibiting STAT3 activation in the presence of MEK inhibition, we used LLL12 as a potent inhibitor of STAT3 phosphorylation (46). As shown in Fig. 3C, the effects of selumetinib and LLL12 were essentially additive.

Upon binding its receptor, IL-6 activates the JAK/STAT3 signaling pathway (47), resulting in cell survival and increased angiogenesis (48–50). Our data showed that IL-6 levels are elevated in BT-40/AZD xenografts and that levels are maintained in the selumetinib-resistant tumors after 72 hours of drug treatment. We hypothesized that as in the cell lines, the transcription factor, STAT3, may be activated in BT-40/AZD xenografts posttreatment with the MEK1/2 inhibitor.

To test this, protein lysates were extracted from BT-40, BT-40/AZD, and BT-40/REV xenograft tumors, in triplicate, from mice that had been administered selumetinib (75 mg/kg BID). Tumors were harvested before treatment (0 hours) and 24, 72, and 144 hours after the start of treatment. STAT3 and phospho-STAT3 (Y705) were determined by analysis of the respective lysates. Both the BT-40 and BT-40/REV tumors displayed a progressive reduction of phospho-STAT3 over the course of drug treatment, with the highest levels of phosphorylation in the untreated tumors, and the lowest at the 144-hour time point (Fig. 5A). The BT-40/AZD drug–resistant tumors, however, showed a marked increase in phosphorylation of STAT3 after 24 hours of the MEK1/2 inhibitor that was maintained above the levels in untreated tumors at the 144-hour time point. Densitometric analysis of the immunoblots (n = 3 per condition; Fig. 5B) confirmed that after selumetinib treatment at 24 hours, phospho-STAT3 levels were reduced by approximately 37% in BT-40 and BT-40/REV tumors. In contrast, phospho-STAT3 increased by 100% in BT-40/AZD xenografts. After 144 hours of drug administration, the parental and revertant tumors both showed a greater than 90% reduction in the phosphorylation of the STAT3 protein, whereas phospho-STAT3 levels in BT-40/AZD tumors were elevated 50% above control (untreated) levels.

We have shown previously that in an osteosarcoma xenograft model, daily dosing with LLL12 (5 mg/kg) completely abrogates the phospho-STAT3 (Y705) signal (51). As a proof-of-principle experiment, using subcutaneous tumors, we examined LLL12, selumetinib, and the combination against BT-40/AZD xenografts. As shown in Fig. 5C, selumetinib failed to inhibit the growth of BT-40/AZD tumors (P = 0.077), confirming resistance. Similarly, LLL12 had very modest activity as a single agent (P = 0.063). In contrast, combined treatment with selumetinib and LLL12 significantly inhibited growth relative to controls (P < 0.0001). In the combination treatment group, tumors dramatically regressed over the first week of treatment. Interestingly, tumors appeared to regrow but then remained essentially static over the 6-week course of therapy. The result suggests that either MEK or STAT3 signaling may maintain proliferation, but inhibition of STAT3 signaling has profound effects only when MEK signaling is simultaneously inhibited, indicating a synthetic lethal interaction. Examination of tumor sections at termination of the respective control or treatment groups showed that LLL12 and selumetinib as single agents had relatively modest effects on reducing proliferation (Fig. 5D). LLL12 reduced Ki67 staining by 28%, whereas selumetinib reduced this index by 10% in BT-40/AZD xenografts. Apoptosis was increased significantly by both LLL12 and selumetinib but was dramatically elevated (∼90%) in tumors harvested from mice receiving the combination of these agents.

Inhibitors of BRAF^V600E such as vemurafenib and to a lesser extent MEK inhibitors such as selumetinib have impacted treatment of BRAF mutant melanoma. However, intrinsic or acquired resistance to each agent, or the combination of BRAF/MEK inhibitors, limits their use (52). Combination treatment with dabrafenib, a selective BRAF inhibitor, and trametinib, a selective MEK inhibitor, was superior to monotherapy; however, median progression-free survival was less than 1 year (52) with objective response rates of 76% versus 54%, and median progression-free survival increased from 5.8 to 9.4 months. Thus, development of acquired resistance remains a significant problem to targeted therapy in melanoma. Furthermore, responsiveness to inhibitors may be context-specific, as colon carcinoma with BRAF^V600E mutations is unresponsive to BRAF inhibitors (44), possibly a consequence of “ERK rebound” associated with activation of EGF receptor (EGFR) signaling (44, 53). Thus, effects of inhibiting the BRAF/MEK pathway may be context-dependent.

The MEK inhibitor selumetinib has recently entered clinical evaluation by the Pediatric Brain Tumor Consortium for treatment of pediatric low-grade astrocytomas (protocol PBTC-029). The incidence of BRAF^V600E mutations is approximately 23% in low-grade astrocytomas of childhood and about 60% in xanthoastrocytomas. The tandem duplication of BRAF that results in the KIAA1549-=BRAF fusion, considered to activate BRAF, appears to be restricted to grade 1 pilocytic astrocytoma. Thus, there is interest in evaluating both BRAF and MEK inhibitors in these tumors, although there is a lack of preclinical data to support whether these cancers will be responsive, as with melanoma, or unresponsive, as with colon cancer, to these agents. A recent report (54) indicates that in cells expressing KIAA1549–BRAF, the fusion kinase functions as a homodimer that is resistant to vemurafenib and is associated with CRAF-independent paradoxical activation of MAPK signaling. However, there is no reason a priori to anticipate that cells expressing the KIAA1549–BRAF fusion would be resistant to an MEK inhibitor. Previously, we characterized 2 patient-derived JPA xenogafts (29) and showed that BT-35 tumors with wild-type BRAF were unresponsive to selumetinib, whereas BT-40 tumors that are heterozygous for the BRAF^V600E allele were highly responsive to treatment. Neither model has the KIAA1549–BRAF fusion, suggesting that these tumors are low-grade astrocytomas rather than JPAs. Consistent with this, we found that the Ki67 proliferation index was relatively high in early-passage xenografts, compared with published data on JPA clinical samples where labeling index was calculated using bromodeoxyuridine incorporation (0.22%–4.3%; ref. 55) or Ki67 staining (56).

In this study, we sought to identify mechanisms of resistance to a MEK inhibitor selumetinib, with the long-term objective to develop therapeutic strategies to circumvent resistance. By direct transplantation of tumor tissue into SCID mice, we developed a xenograft model of low-grade BRAF^V600E astrocytoma and selected for resistance to selumetinib in vivo. The parental BT-40 xenograft was highly sensitive to selumetinib (29), but resistant sublines were readily derived within 2 passages in treated mice. Of interest, further passage in untreated mice led to revertants that were sensitive to selumetinib treatment. These results do not distinguish between an unstable resistance mechanism or overgrowth of a drug-sensitive population of cells when tumors are passaged in untreated mice. Previously, we reported reversion to rapamycin sensitivity in cloned resistant lines, suggesting that epigenetic resistance may occur to signaling inhibitors (57). The levels of mutant BRAF were similar in all tumor lines, indicating that resistance was not associated downregulation of the mutant protein and expression of wild-type BRAF. In contrast to reported mechanisms for acquired resistance to MEK inhibitors selected in vitro, we did not find amplification of the BRAF locus (35) or a failure of drug to inhibit MEK previously reported as a consequence of activation of MAP3K8 (COT; ref. 31) or mutation in MEK (34). Rather, MEK appeared to be equally inhibited in parental and BT-40/AZD selumetinib-resistant xenografts. Thus, we examined changes in the “MEK-functional” and “compensatory” expression signatures (45) that correlated with sensitivity to selumetinib in parental, resistant, and revertant xenograft lines. Rather surprisingly, in BT-40/AZD–resistant xenografts, we found enrichment in gene expression associated with the MEK-functional pathway, considered to reflect “addiction” to MEK signaling and to predict sensitivity to MEK inhibition. Increased expression of these genes was confirmed by qRT-PCR. Three genes (DUSP, LGALS3, and SLCO4A1) were significantly enriched in BT-40/AZD tumors relative to parental BT-40. The differential was larger when BT-40/AZD was compared with BT-40/REV, where 10 of 16 genes were expressed at significantly higher levels in resistant tumors. In contrast, expression profiling did not show significant changes in genes associated with the “compensatory” profile. However, by qRT-PCR, transcripts for IL-6 were clearly enhanced in BT-40/AZD tumors compared with parental (7.5-fold) or revertant tumors (65-fold) and thus correlated with drug sensitivity. Other studies have implicated elevated IL-6 expression in resistance to MEK inhibitors (45, 58).

Examination of IL-6 in parental, resistant, and revertant tumors showed that IL-6 protein tracked with the qRT-PCR data. IL-6 was elevated about 5-fold in resistant tumors compared with parental tumors, whereas it was not detected in revertant xenografts. Of importance, selumetinib abrogated IL-6 detection in BT-40 tumors, whereas IL-6 was decreased but detectable in BT-40/AZD–treated tumors, being about 75% the level determined in untreated BT-40 xenografts. Although several signaling pathways are implicated in IL-6 transcription, we found that NF-κB signaling was activated in BT-40/AZD tumors and was relatively resistant to selumetinib-induced downregulation that occurred in both parental and revertant xenografts. In addition, levels of IL-6 receptor α were decreased 4-fold in BT-40/REV xenografts, suggesting that in these selumetinib-sensitive revertants, the IL-6/STAT3 pathway was abrogated.

These data suggested that enhanced IL-6 secretion may activate STAT3 signaling and circumvent the effect of MEK inhibition. Previously, it was reported that activation of STAT3 correlated with resistance to selumetinib in non–small cell lung cancer (NSCLC) cell lines (59) and in melanoma cell lines selected for vemurafenib resistance in vitro (60). In NSCLC cell lines, STAT3-regulated microRNA miR-17 mediated resistance to selumetinib by preventing drug-induced BIM and PARP cleavage. We examined 2 additional BRAF^V600E mutant astrocytoma cell lines. Both were intrinsically resistant to selumetinib (IC₅₀ > 10 μmol/L). In both lines, there was a reciprocal activation of STAT3 as MEK was progressively inhibited, suggesting that STAT3 activation could be mediating intrinsic resistance to the MEK inhibitor. Inhibition of ERK activation by AZD6244 leads to derepression of STAT3, as ERK-mediated STAT3 (S727) phosphorylation negatively modulates STAT3 tyrosine phosphorylation at Y705, which is required for dimer formation, nuclear translocation, and DNA-binding activity of this transcriptional regulator (61). Consistent with this, phospho-STAT3 (S727) was lower in BT-40/AZD tumors compared with parental or revertant xenografts (data not shown). However, combination of selumetinib with a STAT3 inhibitor (LLL12) gave only additive effects against the 2 BRAF mutant cell lines in vitro, with no obvious increase in apoptosis.

Similar to the in vitro data, where STAT3 was progressively phosphorylated as MEK was inhibited, we found that only in the BT-40/AZD–resistant model was STAT3 phosphorylation increased during treatment with selumetinib. In contrast, phosho-STAT3 decreased more than 90% in both BT-40 and BT-40/REV tumors treated with drug. These data suggested that STAT3 activation could be compensating in the presence of MEK inhibition to maintain proliferation and survival. We have shown previously that the small-molecule STAT3 inhibitor LLL12 (46) completely abrogated phospho-STAT3 (Y705) when administered at 5 mg/kg to mice bearing OS-1 osteosarcoma xenografts (51). We therefore examined the antitumor activity of selumetinib, LLL12, or the combination against BT-40/AZD xenografts. As anticipated, selumetinib had no significant effect in these resistant tumors. Similarly, LLL12 was ineffective single-agent therapy. In contrast, when selumetinib and LLL12 were combined, there was complete regression of BT-40/AZD tumors, with subsequent slight growth, and tumor stasis over the 6 weeks' treatment. These data are in contrast to that of Dai and colleagues (59) where the JAK/STAT3 inhibitor (JS-124, cucurbitacin I; ref. 62) exerted growth delay as a single agent and was essentially additive when combined with selumetinib. In BT-40/AZD xenografts harvested at the end of treatment, there was a modest decrease in Ki67 staining and increase in TUNEL-positive cells in tumors treated with selumetinib or LLL12 as single agents. However, for tumors treated with the combination, approximately 90% of cells were apoptotic, and the proliferation index was dramatically decreased. Thus, residual tumor had few viable cells, and inflammation/edema may have contributed to the measured tumor volume.

In summary, in one of the only xenograft models of pediatric low-grade astrocytoma, we selected resistance to the MEK inhibitor selumetinib. Resistance was unstable, which may suggest epigenetic rather than a genetic basis for resistance. By comparing parental, resistant, and revertant sublines, our studies pointed to enhanced IL-6/STAT3 signaling as a potential mechanism for resistance. Both in vivo and in vitro inhibition of MEK signaling was associated with increased STAT3 (Y705) phosphorylation and resistance to selumetinib. In vivo, combination of selumetinib with a STAT3 inhibitor reversed acquired resistance to selumetinib. Although based only on a single low-grade astrocytoma model, the data support the contention that compensatory activation of STAT3 signaling may be a mechanism for intrinsic and acquired resistances to MEK inhibitors (58, 59) that is amenable to pharmacologic modulation for treatment of childhood low-grade astrocytoma.

D. Capper has ownership interest (including patents) in Ventana Medical Systems Inc. P.D. Smith is an employee of Astrazeneca. No potential conflicts of interest were disclosed by the other authors.

Conception and design: H.K. Bid, J. Lin, D. Oswald, P.J. Houghton

Development of methodology: H.K. Bid, D. Capper, D. Oswald, P.J. Houghton

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H.K. Bid, A. Kibler, D.A. Phelps, L. Xiao, D. Capper, D. Oswald, M. DeWire, P.J. Houghton

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.K. Bid, D. Oswald, B. Geier, R.T. Kurmasheva, X. Mo, S. Fernandez, P.J. Houghton

Writing, review, and/or revision of the manuscript: H.K. Bid, J. Lin, D. Capper, D. Oswald, M. DeWire, P.D. Smith, R.T. Kurmasheva, X. Mo, S. Fernandez, P.J. Houghton

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Kibler, S. Manap, L. Xiao, D. Oswald, P.J. Houghton

Study supervision: J. Lin, R.T. Kurmasheva, P.J. Houghton

Provision of reagent selumetinib: P.D. Smith

This work was supported by a Pelotonia Ideas Grant from the Ohio State University Comprehensive Cancer Center, NO1-CM42216 and CA165995 from the National Cancer Institute. H.K. Bid is a recipient of a postdoctoral fellowship from Pelotonia.

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.