DIPG Harbors Alterations Targetable by MEK Inhibitors, with Acquired Resistance Mechanisms Overcome by Combinatorial Inhibition

A proportion of diffuse intrinsic pontine gliomas (DIPG) harbor alterations that predict sensitivity to MEK inhibition; however, resistance can quickly occur via MEK1/2 mutation that can be overcome with addition of dasatinib, providing a therapeutic option.


INTRODUCTION
The MAPK pathway plays an important role in signal transduction regulating cell proliferation, differentiation, and cell death (1). Dysregulation of the MAPK signaling pathway is implicated in a wide range of cancers as a result of genetic and epigenetic alterations. In adults, BRAF-mutated tumors include 60% of melanomas, 60% of thyroid cancers, 15% of colorectal cancers, and 5% to 8% of non-small cell lung cancers, with the most prevalent mutation being BRAF V600E (2). In addition, non-V600E mutations have been identified to be oncogenic and can be classified as one of three types based on their effect on BRAF activity (3,4).
Although targeted agents against the MAPK pathway have become an important initial success story in the field of precision oncology, they are frequently associated with the emergence of resistance and treatment failure (22). Therapyinduced resistance can occur from the acquisition of de novo mutations (23,24) or from expansion of rare preexisting resistant cells, also called "persister clones" (25)(26)(27). The use of inhibitors such as those targeting BRAF (vemurafenib, dabrafenib) and MEK (trametinib, selumetinib, cobimetinib) in BRAF V600E -positive melanoma resulted in a moderate success of targeted therapies by showing tumor shrinkage and improving patient survival (28)(29)(30). However, durable responses were limited due to resistance to single agents often mediated by reactivation of MAPK through amplification or splice variants in BRAF, mutations in the upstream oncogene NRAS or the downstream kinase MEK1 (MAP2K1), and PI3K-PTEN-AKT upregulation, among others (24,(31)(32)(33)(34). This led to the combination of BRAF and MEK inhibitors in clinical trials, which showed a significant improvement in overall survival compared with single-agent therapy, and subsequently FDA-approved treatment for advanced BRAF V600Epositive melanoma (35,36). However, acquired resistance was observed in patients under BRAF and MEK inhibitors, with the identification of MEK1 and MEK2 (MAP2K2) mutations as two of the main mechanisms (31,(37)(38)(39)(40).
Despite their clinical availability for treating children with brain tumors harboring targetable alterations (41,42), BRAF and MEK inhibitors have thus far not been explored in DIPG. As part of a coclinical trial of patients with DIPG from the United Kingdom enrolled in BIOMEDE, an adaptive, international, phase II clinical trial in newly diagnosed DIPG (NCT02233049), we have generated patient-derived in vitro and in vivo models subjected to integrated drug screening and molecular profiling. We identify specific genetic dependencies associated with response to trametinib in vitro, explore mechanisms of acquired resistance that emerge to single-agent targeted therapy, and present rational drug combinations that may circumvent this in future clinical trials.

MAPK Alterations Confer In Vitro Sensitivity to Trametinib in DIPG Models from a Coclinical Trial
BIOMEDE (NCT02233049) was a phase II, biopsy-driven clinical trial in patients with DIPG with randomization of stratification between dasatinib, erlotinib, and everolimus. From UK patients, we undertook to generate patient-derived in vitro and in vivo models from biopsy material excess to trial inclusion. From the first 12 cases, exome sequencing was performed on the biopsy specimen itself, and panel sequencing was performed on the in vitro cultures established under stem cell conditions in 2 dimensions (2-D; on a laminin matrix) and 3 dimensions (3-D; as neurospheres), as well as tumors grown as orthotopic xenografts in mice (Fig. 1A). Seven cases presented with H3F3A (now H3-3A) K27M mutations, one had HIST2H3C (now H3C14) K27M, and another presented EZHIP overexpression by RNA sequencing, recently described as phenocopying the H3K27M mutation-driven loss of H3K27me3 (43). Of the three histone H3 wild-type cases, one harbored MYCN/ID2 coamplification. Recognized recurrent DIPG driver mutations were largely preserved between the biopsy and models, with specific exceptions between 2-D and 3-D cultures described below, and some amplicons (PDGFRA, MYCN) discrepant between different samples from the same patient (Fig. 1A). Similarly, all models were highly concordant with their original tumors on the basis of their methylation profiles derived from Illumina 850K EPIC arrays (Fig. 1B). Although all H3F3A K27M and EZHIP cases clustered with other diffuse midline glioma K27M tumors, the HIST2H3C K27M case (ICR-B184) was most similar to ICR-B118 and other MYCN-amplified glioblastoma in the t-statistic-based stochastic neighbor embedding projection of the methylation array data (Fig. 1B). The H3 wild-type case ICR-B128 did not classify strongly with any known subgroup. A summary of the molecular data of the models and the tumors from which they were derived is given in Supplementary Table S1. Seventeen established cultures from 11 patients were subjected to a drug mini-screen against a range of targeted inhibitors (n = 13), selected on the basis of their molecular profiles (Fig. 1C). Exploring these data for genetic dependencies, we observed a striking concordance of mutations in the MAPK pathway (BRAF, NF1, PIK3R1) and sensitivity to the MEK inhibitor trametinib, both in 2-D and 3-D, including the models ICR-B169 in 2-D (Fig. 1D) and ICR-B169, ICR-B181, and ICR-B184 in 3-D (Fig. 1E).

Functional Selection of MAPK Alterations in Model Systems and Translation to the Clinic
ICR-B181 harbored a subclonal [variant allele frequency (VAF) = 18%] hotspot mutation in PIK3R1, N564D ( Fig. 2A). ICR-B184 was found to have a subclonal (VAF = 20%) frameshift mutation in NF1 in the biopsy specimen (E78fs) that was absent from the models, which instead presented the missense mutation I1824S (Fig. 2B). In both cases, the likely MAPK-activating mutations were observed in only the 3-D cultures, not 2-D cultures, which corresponded with a marked increase in sensitivity to trametinib in the neurospheres compared with adherent cells, when validated in a full dose-response analysis [∼420-fold for ICR-B181 (Fig. 2C) and ∼460-fold for ICR-B184 (Fig. 2D) Fig. S1D)].
ICR-B169 (2-D and 3-D) harbored a clonal, heterozygous, noncanonical class II BRAF oncogenic mutation, G469V (Fig. 3A). Following feedback of our molecular and in vitro drug screening data, the patient was clinically treated with trametinib on a compassionate use basis (0.025 mg/kg/d once daily, the pediatric recommended phase II dose previously established in an international phase I/II clinical trial) 2 months after progression at 16 weeks on everolimus and radiotherapy. During 11 weeks on trametinib, the patient improved clinically, was able to be taken off the steroid treatments required over the previous 3 months, and showed radiologically stable disease during this time. However, an MRI at week 12 showed progression, with the appearance of new metastatic lesions within the brainstem and in the lateral ventricles, and the patient died of disease shortly afterward (Fig. 3B). In parallel, mice bearing orthotopic xenografts derived from ICR-B169 cells (3-D) were treated with trametinib, from day 55 postimplantation, Figure 1. In vitro sensitivity to trametinib in patient-derived DIPG models. A, Oncoprint representation of an integrated annotation of single-nucleotide variants, DNA copy number changes, and structural variants for patient-derived models and tumor biopsy specimens. Samples are arranged in columns with genes labeled along rows. Clinicopathologic and molecular annotations are provided as bars according to the included key. B, The t-statistic-based stochastic neighbor embedding (t-SNE) projection of a combined methylation data set comprising the in vitro models (circled) plus a reference set of glioma subtypes (n = 1,766). The first two projections are plotted on the x-and y-axes, with samples represented by dots colored by subtype as labeled on the figure. C, Drug sensitivities in the mini-screens carried out on cells grown under 2-D and 3-D conditions, visualized by heat map of normalized AUC values. Clinicopathologic and molecular annotations are provided as bars according to the included key. D, Dose-response curves for the MEK inhibitor trametinib tested against patient-derived models in vitro grown in 2-D. E, Dose-response curves for the MEK inhibitor trametinib tested against patientderived models in vitro grown in 3-D. Cells harboring MAPK pathway alterations are highlighted in blue. Concentration of compound is plotted on a log scale (x-axis) against cell viability (y-axis). Means plus standard errors are plotted from at least n = 3 experiments.  Everolimus  Erlotinib  Crizotinib  Olaparib  Panobinostat  AZD1775  ONC201  CUDC-907  AZD2014  LY3023414  Regorafenib   ICR-B117-2-D  ICR-B121-2-D  ICR-B128-2-D  ICR-B134-2-D  ICR-B134-3-D  ICR-B169-2-D  ICR-B169-3-D  ICR-B181-2-D  ICR-B184-2-D  ICR-B184-3-D  ICR-B186-3-D  ICR-B194-2-D  ICR-B194-3-D  ICR-B198-2-D  ICR-B181-3-D  ICR-B193-2-D  ICR-B193-3 LGG_MYB LGG_PA_GG_ST LGG_PA_MID LGG_PA_PF ICR-B184 1 mg/kg orally once daily over 53 days (5 days on, 2 days off). No difference in overall survival between trametinib-and vehicle-treated animals was observed (P = 0.7123, log-rank test; Fig. 3C).

MEK1/2 Mutations Drive Resistance to Trametinib in BRAF G469V -Driven DIPG Cells
As single-agent trametinib did not achieve the expected efficacy both in vivo and clinically, we sought to explore the mechanisms by which resistance to MEK inhibition may occur in DIPG and identify strategies to circumvent these. We first attempted to generate subclones of ICR-B169 cells resistant to trametinib by two continuous exposure approaches in vitro, the first by an exponential increase in drug concentration from the GI 50 value over time and the second by a constant GI 80 dose (Fig. 3D). Both experiments produced subclones with increasing resistance to trametinib, with shifts in GI 50 values over 5 to 9 months of 64-to 167-fold with the exponential model ( Supplementary Fig. S2A) and 5-to 56-fold with the constant GI 80 approach ( Supplementary Fig. S2B). To explore what might be underlying the lack of sensitivity in selected subclones, exome sequencing was carried out in the showing the mutant residue for the observed I1824S missense mutation observed in ICR-B184. Generated in ProteinPaint (pecan.stjude.cloud/proteinpaint). C, Dose-response validation curves for trametinib tested against ICR-B181 cells in vitro grown in 3-D (PIK3R1 N564D , blue) and 2-D (PIK3R1 wild-type, gray). D, Dose-response curves for trametinib tested against ICR-B184 cells in vitro grown in 3-D (NF1 I1824S , blue) and 2-D (NF1 wild-type, gray). Concentration of compound is plotted on a log scale (x-axis) against cell viability (y-axis). Mean plus standard error are plotted from at least n = 3 experiments. ****, P < 0.0001, AUC t test. E, Bar plot of quantitative capillary phospho-protein assessment of phospho-ERK1/2 T202/Y204 , plotted as a ratio to total ERK1/2, and normalized to the 2-D (MAPK wild-type) model in each case.    Fig. S2D).
To evaluate if resistance was selected from preexisting tumor subclones or was acquired in vitro in response to trametinib, allele-specific ddPCR assays for the MEK1/2 mutations were conducted in both the original tumor biopsy specimen and the parental cells ( Supplementary Fig. S3A). No mutant droplets were found in either sample for any of the three mutations, with an average number of wild-type droplets of 32,165 and 41,083, respectively. We also found no MEK1/2 mutant droplets in either vehicle-or trametinib-treated cell-derived xenograft (CDX) specimens from our earlier efficacy experiment. Exome sequencing of these latter specimens confirmed the lack of MEK1/2 mutations, but in four of five treated mice, we observed plausible variants in genes that converge on RAF/ MEK signaling, including MYBL1, IRAK1, DUSP26, and others ( Supplementary Fig. S3B), highlighted by STRING proteininteraction analysis ( Supplementary Fig. S3C).
The trametinib-resistant clones were also cross-resistant to the additional MEK inhibitors cobimetinib (P < 0.0001-0.0004 AUC t test; Supplementary Fig. S4A) and binimetinib (P = 0.0021-0.0033 AUC t test; Supplementary Fig. S4B), with GI 50 values not reached at 10 μmol/L drug. When implanted orthotopically in immune-deficient mice, they had a substantially longer latency time to tumor formation, as seen with median survival times of 225 to 435 days, compared with the 185 days of the ICR-B169 parental cells ( Supplementary Fig. S4C).

Mechanisms of MEK1/2 Mutation-Driven Resistance to Trametinib
To explore the mechanism by which mutations in MEK1/2 confer resistance to trametinib in DIPG cells, we carried out gene expression profiling by RNA sequencing (RNAseq), as well as total and phospho-proteome analysis by LC/ MS-MS in our resistant clones. We identified 277, 127, and 241 differentially upregulated genes and 232, 175, and 214 downregulated genes at the transcript level in T1, T3, and T6, respectively, compared with ICR-B169 parental cells (Supplementary Fig. S5A). Of those, a total of 41 genes were shared to be upregulated and 56 genes to be downregulated among the three clones (Supplementary Fig. S5B and S5C). From the total proteome, 447, 378, and 435 proteins were differentially overexpressed and 626, 673, and 729 underexpressed in the clones ( Supplementary Fig. S5D), with 137 and 225, respectively, in common between the three clones (Supplementary Fig. S5E and S5F). Phospho-proteomics identified a total of 212, 37, and 167 phospho-peptides from 124, 22, and 106 proteins with increased phosphorylation and 561, 171, and 395 phospho-peptides from 373, 117, and 221 proteins with decreased phosphorylation in T1, T3, and T6, respectively ( Supplementary Fig. S5G). Of these, a total of 8 proteins with increased phosphorylation and 57 with decreased phosphorylation were shared among the three clones (Supplementary Fig. S5H and S5I).
The intersection of the shared most differentially upregulated (n = 28; Fig. 4A) or downregulated (n = 24; Fig. 4B) genes, proteins, and phospho-sites between any of the three clones resulted in an integrated signature of coordinately represent passage number; left y-axes are specific mutation VAFs; right y-axes plot the concentration of trametinib that cells were exposed to (gray dashed line). H, Emergence of resistance in clones over time (T1, pink; T3, purple; T6, red), plotted as days exposed to inhibitor on the x-axis, with passage numbers labeled. The y-axis is a GI 50 value for trametinib in cells harvested at the given passage. I, Pathway activation in resistant clones (T1, pink; T3, purple; T6, red) assessed by a capillary electrophoresis assay and plotted as a ratio of respective phosphorylated/total protein compared to ICR-B169 parental cells (gray).  Table S3).

Combinatorial Approaches for Overcoming Trametinib Resistance in DIPG
In considering opportunities for how to effectively treat MAPK-driven DIPG cells that have become resistant to MEK inhibitors, we first tested the downstream ERK inhibitor ulixertinib, as reported in other tumor types (44,45). No differences in sensitivity were observed between T1 MEK2 I115N or T3 MEK1 I141S and the ICR-B169 parental cells, whereas T6 MEK1 K57N cells were significantly less sensitive to ulixertinib (P = 0.0024, AUC t test; Fig. 5A). By contrast, treatment with the upstream receptor tyrosine kinase (RTK) inhibitor dasatinib revealed a marked differential sensitivity in all trametinib-resistant clones, with a 68-to 174-fold shift in GI 50 values (P = 0.009-0.0004, AUC t test; Fig. 5B). With this reciprocal potency of trametinib and dasatinib in the ICR-B169 parental cells and subclones, we next investigated whether this was true in all the patient-derived cultures we initially assessed. Cells classed

RNA-seq Proteome Phospho
IC R -B 1 6 9 P a re n ta l IC R -B 1 6 9 P a re n ta l IC R -B 1 6 9 P a re n ta l

RNA-seq
Proteome Phospho B IC R -B 1 6 9 P a re n ta l IC R -B 1 6 9 P a re n ta l IC R -B 1 6 9 P a re n ta l Enrichment score IC R -B 1 6 9 P a re n ta l IC R -B 1 6 9 P a re n ta l as trametinib-resistant in the mini-screen were found to have significantly lower GI 50 values for dasatinib than the trametinib-sensitive cultures (P = 0.0017, unpaired t test; Fig. 5C), while the converse was also true: dasatinib-sensitive cells had significantly higher GI 50 values for trametinib than those classed as dasatinib-resistant (P = 0.0014, unpaired t test; Fig. 5D).
Given the similarities between a range of distinct patientderived cultures (and their relative responses to trametinib) and the resistant clones generated from the BRAF G469Vdriven ICR-B169 cells, we applied GSEA to the RNA-seq data on the panel of untreated in vitro models. There was substantial overlap in the significantly differentially expressed gene signatures observed in the inherently trametinib-resistant     CDC42EP3  OSMR  SNAI2  PRNP  GNG12  LARP6  DPYD  CAV2  SOCS5  EPHB2  ELK3  CALD1  ITGA5  RAI14  REXO2  POPDC3  PCDH7  IFIT3  ANTXR2  TGFBR2  LAYN  CAV1  FSTL1  PARVA  MET  FXYD5  NNMT  COL5A1  ANXA1 Fig. S6C). Most strikingly, there was a highly significant underrepresentation of the VERHAAK_GLIOBLASTOMA_PRONEURAL signature in the inherently trametinib-resistant cultures (nominal P = 0.000231), significantly enriched in dasatinibresistant cells (nominal P = 0.000204; Fig. 5E), along with the reverse pattern for mesenchymal gene sets such as HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSI-TION (trametinib-resistant, nominal P = 0.000172; dasatinibresistant, nominal P = 0.000196; Fig. 5F). As expected, we also observed a significant enrichment in the inherently trametinib-resistant cells, as well as underrepresentation in the dasatinib-resistant primary cultures, of the HUANG_DASATINIB_RESISTANCE_UP gene set (nominal P = 0.000184 and nominal P = 0.000198, respectively; Fig. 5G, Supplementary Table S3). As dasatinib is a multitargeted kinase inhibitor, we explored which of these may underlie the response in the trametinib-resistant cells by phospho-protein array profiling. In T6 (MEK1 K57N ) cells, we observed only a modest decrease in pPDGFRB Y571 ( Supplementary Fig. S6D) and pAKT S473 ( Supplementary  Fig. S6E) in response to dasatinib at the higher doses. By contrast, there were high basal levels of pSRC Y419 (Supplementary Fig. S6F) and pYES1 Y426 (Supplementary Fig. S6G), which were dramatically inhibited at the lowest doses of dasatinib, correlating with the cellular response (Supplementary Table S4). Notably, there was also a specific shift in dasatinib sensitivity in the paired ICR-B181 2-D and ICR-B181 3-D (PIK3R1 N546D ) cells (3-fold, P = 0.0398 AUC t test; Supplementary Fig. S6H), as well as the ICR-B184 2-D and ICR-B184 3-D (NF1 I1824S ) cells (7.7-fold, P = 0.0012 AUC t test; Supplementary Fig. S6I).
Finally, we assessed whether this reciprocal sensitivity to targeted inhibition would lead to an effective combination therapy in MAPK-driven DIPG cells. We used the BLISS independence model to assess the combination of dasatinib and trametinib, which showed profound effects on cell viability in ICR-B169 parental cells, even at low doses of both drugs (Fig. 6A). This was in contrast to the trametinib-resistant clones, which unsurprisingly required high levels of MEK inhibitor to produce an effect (large and trametinib (x-axes) ranging from 0 to 10 μmol/L. A heat map is overlaid to the proportions of viable cells remaining, colored according to the key provided from 1.0 (black, all cells) to 0 (red, no viable cells). B, Excess above BLISS matrices for ICR-B169 parental and trametinib-resistant clones treated with distinct combinations of dasatinib (y-axes) and trametinib (x-axes) ranging from 0 to 10 μmol/L. A heat map is overlaid to the excess score, colored according to the key provided from 0.4 (red, enhanced effects) to 0 (green, no difference). C, BLISS synergy maps for ICR-B169 parental and trametinibresistant clones treated with distinct combinations of dasatinib (y-axes) and trametinib (x-axes) ranging from 0 to 10 μmol/L. The heat map represents the δ score colored according to the key provided from 30 (red, high degree of synergy) to −30 (green, antagonism). δ-score black areas on heat map). This was more formally evaluated by calculating the excess above BLISS scores for each pairwise combination, identifying likely synergistic areas within the combinatorial matrix, highlighted in red on the heat map (Fig. 6B). A summary synergy score was calculated as the average excess response due to drug interactions above expectations, with the resulting value of 19.9 in the ICR-B169 parental cells indicating a high degree of formal synergy at the lowest doses of trametinib, an effect not seen in the resistant clones, in an analogous pattern to the cell viability score above (Fig. 6C). Notably, there was also a highly synergistic interaction of ulixertinib and trametinib in both parental and resistant clones, particularly with the T6 MEK1 K57N cells, which were largely resistant to ulixertinib as a single agent ( Supplementary  Fig. S7A-S7C).
To validate our proposed combinatorial approach, we established mouse ex vivo brain slice cultures, onto which we implanted ICR-B169 parental cells (Fig. 7A). Treatment with combined trametinib and dasatinib resulted in a significantly reduced growth capacity across the brain parenchymal tissue compared with vehicle (P = 0.0003, FDR-corrected t test) or either agent alone (P = 0.0057 vs. dasatinib; P = 0.050 vs. trametinib; FDR-corrected t test; Fig. 7B). Critically, there was a substantial reduction in cell number evident in the tumor cell core in the combination arm not observed with the single agents.
Thus, we present evidence for a subset of patients with DIPG to harbor MAPK pathway-activating mutations, as well as sensitivity to MEK inhibition in vitro with the rapid acquisition of consistent resistance mutations, and identify potentially efficacious combination approaches based on the common mechanisms underlying these events.

DISCUSSION
In the setting of a coclinical trial of prospectively established DIPG patient-derived models, we show the identification of biomarkers of response to trametinib, a selective reversible inhibitor of MEK1/2 that binds to the allosteric pocket of MEK. Specifically, these involve multiple nodes of the MAPK signaling pathway and include PIK3R1 N564D , NF1 I1824S , and BRAF G469V . These results suggest a possible rationale for the use of trametinib in DIPG, such as the ongoing phase II clinical trial called TRAM-01 (NCT03363217), which is exploring the use of trametinib in pediatric gliomas harboring MAPK alterations independently of the tumor entity (46). PIK3R1 N564D is an oncogenic hotspot mutation known to promote cell survival in vitro and oncogenesis in vivo (47) and lies within the regulatory subunit of PI3 kinase, resulting in loss of function, predicted to destabilize protein interaction, which may affect tumor suppressive function (FATHMM pathogenic score of 0.99; refs. 48,49). Such PIK3R1 oncogenic mutations have been shown to activate the MAPK pathway and exhibit sensitivity to MAPK inhibitors (50). NF1 I1824S lies in the neurofibromin chain and the lipid binding region, which may affect the protein folding and the tumor suppressor gene function (48); it has also been reported in a patient with neurofibromatosis type 1 (51) and is absent from population databases. The efficacy of MEK inhibitors has previously been shown in NF1-deficient glioblastoma cell lines (52), in Figure 7. Efficacy of combined dasatinib and trametinib on ex vivo brain slice preparations. A, Coronal slices of normal mouse brain, counterstained with Hoechst 33342 (aqua), are implanted in the pontine region with ICR-B169 parental cells, stained with human nuclear antigen (orange), and treated for 4 days with 1 μmol/L dasatinib, 0.1234 μmol/L trametinib, or both compared with vehicle control. Scale bars, 2 mm. B, Bar plot of quantification of tumor cell infiltration across the brain parenchymal tissue as measured by the calculated area invaded compared with vehicle control. Plotted is the mean of at least six independent slices; error bars represent the SD. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not significant, FDR-corrected t test. Extent of infiltration * addition to clinical benefit in refractory neurofibromatosisassociated glioma harboring the NF1 mutation (53)(54)(55). By stochastic selection, an imbalance of VAF between 3-D and 2-D cultures from the same patient was observed in ICR-B181 (PIK3R1 N564D ) and ICR-B184 (NF1 I1824S ). Trametinib sensitivity was identified only in the mutant cultures, which also had higher basal levels of MAPK pathway activation, which would support the hypothesis that these mutations were responsible for the observed trametinib efficacy in vitro.
BRAF G469V is a class II BRAF mutation within the protein kinase domain, resulting in increased kinase activity and downstream MEK and ERK activation (3). BRAF class II mutations have constitutively activated BRAF dimers independent of RAS activation (3), and BRAF G469V has been shown to confer sensitivity to trametinib in melanoma and lung cancer (4,56). Despite this, in the context of DIPG, targeting this mutation with MEK inhibitor as a single agent in vivo and clinically failed to elicit a significant response. Although there may be issues with the treatment regimen used in our long-latency orthotopic xenografts, and the patient received trametinib only postprogression after several months on everolimus and radiation, there are also concerns over the emergence of resistance to MEK inhibition as previously described in melanoma or colorectal cancers (37)(38)(39)(40). Such a process we showed to be active in DIPG cells through the acquisition in continuous exposure experiments in vitro of MEK1/2 mutations, resulting in pathway reactivation and the irreversible resistance to trametinib. Although not observed in trametinib-insensitive CDX models in vivo, we did observe plausible variants that form part of a BRAF-MEK1/2 circuit, further underlying the importance of reactivation of this element of the signaling pathway in response to drug interventions.
MEK1 and MEK2 exhibit 85% peptide sequence homology (57). MEK1 K57N lies on the helix A domain within the N-terminal negative regulatory region and is associated with high levels of RAF-regulated activation of ERK signaling (31,58). Interestingly, MEK1 K57N has been attributed to cause resistance to BRAF and MEK inhibitors in vitro and in patients with melanoma (45,59). To the best of our knowledge, neither MEK1 I141S nor MEK2 I115N has been previously detected, although both mutations lie within the protein kinase domain affecting the allosteric pocket (45). Particularly, MEK1 I111N , the equivalent of MEK2 I115N , has been demonstrated to confer resistance in vitro to allosteric MEK inhibitors (31,45). In our resistant clones, the presence of the MEK1/2 mutations resulted in shift of 2 to 3 orders of magnitude of trametinib concentration required to inhibit MAPK signaling.
Notably, using integrated RNA-seq and phospho/total proteomics, a drift from a proneural to a mesenchymal phenotype was observed in the MEK1/2-mutant resistant clones, a feature also observed in patient-derived models inherently insensitive to trametinib. This has been broadly reported as a hallmark of metastasis and resistance to multitherapy in cancer (60,61). In particular, glioma-initiating clones displaying drug resistance and radioresistance have been previously linked to a proneural-mesenchymal transition (62). In our models, this was also accompanied by a reciprocal sensitivity of the trametinib-resistant DIPG cells to the multikinase inhibitor dasatinib. In this context, phospho-kinase profiling highlighted Src-family kinases as the likely mediators of this response rather than PDGFR signaling. In other cancers, a proportion of KRAS-mutant cell lines resistant to trametinib was found to have a mesenchymal gene expression signature (63), and dasatinib has been reported to overcome mesenchymal transition-associated resistance to erlotinib in nonsmall cell lung cancers (64). In addition, a study has found that dasatinib sensitizes KRAS-mutant cancers to trametinib both in vivo and in vitro (65). In DIPG, the combined use of dasatinib and trametinib showed a high degree of synergy in vitro and on ex vivo brain slices and may represent a novel combinatorial approach in this disease.
It remains to be explored how best these combinations could be translated clinically in order to prevent or overcome resistance to MEK inhibitors. Critically, we observed consistent results in our primary patient-derived cultures, 2-D/3-D MAPK-altered isogenic cells, and the trametinib-resistant clones for the more clinically relevant brain-penetrant MEK inhibitors cobimetinib and binimetinib. Going forward, assessment of an intermittent multitherapy regimen to control population dynamics and potentially prevent emergence of treatment resistance to begin with is clearly warranted. In this context, mathematical modeling and adaptive drug design will be essential to determine treatment scheduling, exploiting the fitness trade-offs associated with resistance therapy (66).

Primary Patient-Derived Cell Cultures
DIPG biopsy tissue was shipped to our laboratory in Hibernate A transport media (ThermoFisher Scientific, A12475-01) at room temperature or minced with a sterile scalpel blade in DMEM/ F12 (Life Technologies, 11320-074) supplemented with 0.2% BSA (Sigma-Aldrich, A1595) and 10% DMSO (Sigma-Aldrich, D2650), frozen at −80°C and sent on dry ice (Supplementary Table S1). Minced tissue was digested using Liberase DL. Supernatant was removed and the tissue/cells were resuspended in stem cell media and continuously pipetted to ensure dissociation before transfer to culture flasks to grow either attached on a laminin substrate (Merck Millipore, CC095) and/or in suspension as neurospheres in ultra-low attachment flasks (Sigma-Aldrich, CLS3815). The cells were grown at 37°C, 5% CO 2

Nucleic Acid Extraction
DNA and RNA were isolated from the same piece of tissue or cell pellet using ZR-Duet DNA/RNA Miniprep Plus (Zymo Research, D7001). If only DNA was extracted, the DNeasy Blood & Tissue

AACRJournals.org
Kit (Qiagen, 69581) was used, and if only RNA was extracted, the RNeasy Mini Kit protocol (Qiagen, 74104) was used. DNA and RNA quality were measured using a Nanodrop spectrophotometer (Thermo-Scientific). DNA concentration was determined using a Qubit fluorometer (Thermo Fisher Scientific, Q32850 and Q32851). RNA integrity was analyzed and quantified using a 4200-Tapestation (Agilent).

RNA-seq
At least 150 ng RNA was sequenced at Eurofins Genomics. Strand-specific cDNA libraries were made by purification of poly-A containing mRNA molecules followed by mRNA fragmentation and random primed cDNA synthesis (strand-specific). Adapter-ligation and adapter-specific PCR amplification was performed before sequencing on Illumina sequencers (HiSeq or NovaSeq) using 150-bp pairedend reads chemistry according to the manufacturer's instructions. RNA-seq data were aligned with STAR and summarized as gene-level fragments per kilobase per million reads sequenced using BEDTools and HTSeq. Following rlog transformation and normalization, differential expression was assigned with DESeq.2. Fusion transcripts were detected using chimerscan version 0.4.5a filtered to remove common false positives. GSEA was carried out using the R package fastGSEA (fGSEA) based on curated canonical pathways (MsigDB, Broad Institute).

Methylation
A total of 50 to 500 ng DNA was bisulfite modified using the EZ DNA Methylation-Direct Kit (Zymo, D5006) and loaded onto the Illumina Infinium MethylationEPIC BeadChip, and the array intensities were read on the Illumina iScan system at the University College London Genomics Centre according to the manufacturer's instructions. Methylation data from the Illumina Infinium Human-Methylation850 BeadChip were preprocessed using the minfi package in R (v11b4). DNA copy number was recovered from combined intensities using the conumee package. The Heidelberg brain tumor classifier (ref. 67; molecularneuropathology.org) was used to assign a calibrated score to each case, associating it with one of the 91 tumor entities that feature within the current classifier. Clustering of β values from methylation arrays was performed based on correlation distance using a Ward algorithm. DNA copy number was derived from combined log 2 intensity data based on an internal median processed using the R packages minfi and conumee to call copy number in 15,431 bins across the genome.

ddPCR
Custom TaqMan-based quantitative PCR genotyping assays (Applied Biosystem, Thermo Scientific) were designed to specifically detect PIK3R1 N564D and MEK1 K57N , MEK1 I141S , and MEK2 I115N mutations (Supplementary Table S2). ddPCR was performed by using the Bio-Rad automated droplet generator (BioRad, QX200 AutoDG) and the Bio-Rad QX200 droplet reader, based on the manufacturer's guidelines. Results were analyzed using Bio-Rad QuantaSoft Analysis Pro software. At least two single FAM-positive droplets were required to call a sample positive.

Drug Assays
Primary cells were incubated in 96-or 384-well plates for 3 days before adding the compounds with serial dilution at 10 different concentrations, and 8 days later (endpoint, day 11) cell viability was measured using CellTiter-Glo (2.0, or 3-D assays as appropriate; Promega). Relative luminescence units (RLU) for each well were normalized to the median RLU from the DMSO control wells as 100% viability. At least two duplicates per drug condition were performed as well as a minimum of three independent biological replicates. GI 50 values (drug concentration causing 50% inhibition of cell proliferation) were calculated using GraphPad Prism, and the curves show the mean ± SE of the replicates per condition measured. All compounds were purchased from Selleckchem except for CUDC-907 and PTC-209, which were obtained from Apexbio. The compounds were diluted in DMSO to a final concentration of 10 mmol/L. Drug plates were prepared using the acoustic liquid handler Echo 550 (Labcyte). Each plate included six compounds at eight different concentrations as well as a cytotoxic chemotherapeutic (camptothecin) as a positive control, as well as DMSO as a negative control. Drug combinations were assessed by adding one compound in rows and another in columns with serial dilutions resulting in a 6 × 10 dose matrix using different fold dilutions manually prepared. The stand-alone web application SynergyFinder (https://synergyfinder. fimm.fi) was used for interactive analysis and visualization of multidrug combination profiling data following the BLISS independence model (68,69).

Generation of Resistant Clones
ICR-B169 resistant clones were established by culturing the parental line in escalating concentrations of trametinib from 0.05 μmol/L (GI 50 ) to 1 μmol/L in an exponential stepwise manner or by exposing the cells to a constant concentration of trametinib of 0.5 μmol/L (GI 80 value). Both methods were performed using two technical replicates and two independent biological replicates for a total of eight derived cultures plus four replicate control flasks that were treated with the same concentration of DMSO in parallel to the establishment of the resistant cells. Cells were maintained in DMSO for only two passages to keep the baseline clonal population as close as possible to the parental culture. A total of 1 to 1.4 × 10 6 cells were seeded in a T75 flask for the resistance assay, and 2 to 3 days later the media with trametinib were added at the appropriate concentration or DMSO to the control flasks. For the escalating-dose approach, 10 different concentrations were used with an exponential increment, and the same dose was added a total of six times. During the generation of resistance, the stem cell media containing the drug or DMSO were replaced three times a week and the cells split when they reached 90% confluency.

Capillary-Based Protein Quantification
Cells were scraped from the flask and collected in media or cold PBS and then centrifuged at 1,300 rpm for 5 minutes, resuspended in 1 mL cold PBS, and centrifuged at 4,000 rpm for 4 minutes. The supernatant was then removed and the pellet was resuspended in 50 to 100 μL cold lysis buffer (Cell Signaling Technologies, 9803) containing protease inhibitor cocktail mini-tablet (Roche, Diagnostics, 11836153001) and phosphatase inhibitors (Sigma-Aldrich, P044 and P5726). Lysates were sonicated for 10 seconds at 40% amplitude, spun at 14,000 rpm for 10 minutes in a cold microfuge, and quantified using the BCA Protein Assay Kit (Thermo Fisher, 23225). Capillary electrophoresis was conducted using the automated Wes system (ProteinSimple) with the 12-to 230-kDa Separation module (SM-W004) and the anti-rabbit detection module (DM-001) following the manufacturer's instructions, and analyzed with Compass software. The following primary antibodies were used: AKT (

Phospho-Kinase Profiling
Capillary electrophoresis was conducted using the automated Wes system (ProteinSimple) with the 12-to 230-kDa Separation module (SM-W004) and the anti-rabbit detection module (DM-001) following the manufacturer's instructions and analyzed with Compass software. In each run, each well contained an antibody of interest (9101L, pERK1/2 T202/Y204 ; 9102L, total ERK1/2; Cell Signaling) and normalized to loading control antibody (6487S, α-actinin; Cell Signaling) in each well.
Human Phospho-Kinase Arrays (R&D Systems, ARY003C) were placed in 1.5 mL array buffer 1 and agitated gently for 1 hour at room temperature. Array buffer was substituted by dilute lysate (100 μg/mL) and left O.N. at 4°C. Membranes were subsequently washed three times in wash buffer for 10 minutes and then incubated in detection antibody for 2 hours at room temperature. The arrays were rinsed a further three times in wash buffer and then covered by Streptavidin-HRP in 1× Array Buffer (Thermo Fisher) and incubated at room temperature for 30 minutes. They were visualized using the LI-COR Imaging System and analyzed using the Quick Spots Image Analysis Software (R&D Systems). Briefly, the pixel density of each spot on the array was measured, and then the average signal of the pair of duplicate spots representing each pRTK was calculated. Finally, the averaged background signal from each RTK spot was subtracted.
For treatment with trametinib or dasatinib, cells were incubated in complete media with vehicle (DMSO) or increasing concentrations of drug (0.01, 0.1, 1, and 10 μmol/L), and protein was collected at 30 minutes posttreatment. Samples were lysed by using lysis buffer (CST) containing phosphatase inhibitor cocktail (Sigma-Aldrich) and protease inhibitor cocktail (Roche Diagnostics).

Proteomics
Cell pellets were lysed in 5% SDS/100 mmol/L triethylammonium bicarbonate buffer with probe sonication and heating at 95°C for 10 minutes, and protein concentration was measured by the Pierce 660nm Protein Assay. Then, 300 μg protein was reduced with 5 mmol/L Tris (2-carboxy-ethyl)-phosphin-HCl, alkylated by iodoacetamide, and then purified by methanol/chloroform precipitation. Trypsin was added at a 1:30 ratio (trypsin/proteins) for 18 hours of digestion at 37°C. Next, 150 μg of peptides per sample was tandem mass tag (TMT) labeled as instructed by the manufacturer (Thermo Scientific). The TMT-labeled peptide mixture was fractionated by C18 column at pH 10, and fractions were collected and pooled. Phosphopeptide enrichment used the High-Select Fe-NTA Phospho-peptide Enrichment Kit. Samples were analyzed on an Orbitrap Fusion Lumos coupled with an Ultimate 3000 RSLCnano System (Thermo Scientific) using the Synchronous Precursor Selection method with dynamic exclusion enabled (as reported in PMID 28854368). All raw files were processed in Proteome Discoverer 2.3 (phospho-proteome) or 2.4 (full proteome; Thermo Fisher) using the SequestHT search engine. Spectra were searched against reviewed Uniprot Homo sapiens entries (November 2019) and an in-house contaminant database. Search parameters for phospho-proteome were trypsin with up to 2 miscleavage sites; mass tolerances at 20 ppm for Precursor and 0.02 Da for fragment ions; dynamic modifications of deamidated (N, Q), oxidation (M), and phospho (S, T, Y); and static modifications of carbamidomethyl and TMT6plex (peptide N-terminus and K). For full proteome, fragment ion tolerance was 0.5 Da, with dynamic modification of oxidation (M), acetylation (Protein N-terminus), and static modifications of carbamidomethyl and TMT6plex (peptide N-terminus and K). Peptides were validated by Percolator with q value set at 0.05. Phosphorylation site localization probabilities were computed by the ptmRS node. The TMT10plex reporter ion quantifier included 20-ppm integration tolerance on the most confident centroid peak at the MS3 level of unique peptides. Peptides with average reported S/N > 3 were used for protein quantification, and the SPS mass matches threshold was set at 55%. Only master proteins were reported.

Orthotopic Xenograft Studies
All in vivo experiments were approved by the local Animal Welfare and Ethics Review Board at the Institute of Cancer Research and carried out in accordance with the UK Home Office Animals (Scientific Procedures) Act of 1986, the UK National Cancer Research Institute guidelines for the welfare of animals in cancer research, and the ARRIVE (Animal Research: Reporting In Vivo Experiments) guidelines (70). A single cell suspension was prepared from cells from tissue [patient-derived xenograft (PDX), n = 9] immediately prior to implantation in NOD.Cg-Prkdc scid Il2rg tm1WjI /SzJ (NSG) mice (Charles River). Animals were anesthetized with 4% isoflurane and maintained at 2% to 3% isoflurane delivered in oxygen (1 L/ min). Mice were placed on a stereotactic apparatus for orthotopic implantation, with coordinates x = +1.0, z = −0.8, y = −4 mm from the lambda used for delivery to the pons. Then, 5 μL of cell suspension (250,000 cells) was stereotactically implanted per animal, using a 25-gauge SGE standard fixed needle syringe (SGE 005000) at a rate of rate of 1 μL/min for PDX and 2 μL/min for CDX using a digital pump (HA1100, Pico Plus Elite; Harvard Apparatus). For the trametinib efficacy study, mice (29-33 days old) were randomized according to tumor volume determined by MRI into two groups: group 1 included vehicle [10% DMSO, w/v hydroxyproylbetacyclodextrin dissolved in PBS, oral gavage (PO), every day], and group 2 included trametinib (1 mg/kg, PO, every day). Animals were treated for 8 weeks, 5 days on, 2 days off, and treatment started at day 55 postinjection. 1 H MRI was performed using a horizontal bore Bruker Biospec 70/20 equipped with physiologic monitoring equipment (SA Instruments) using a 2-cm × 2-cm mouse brain array coil. Anesthesia was induced using 3% isoflurane delivered in oxygen (0.5 L/min) and maintained at 1% to 2%. Core body temperature was maintained using a thermoregulated water-heated blanket. Following optimization of the magnetic field homogeneity using a localized map shim over the whole brain, a rapid acquisition with relaxation enhancement (RARE) T 2 -weighted sequence (repetition time = 4,500 ms, effective echo time = 36 ms, two averages, RARE factor = 8, in-plane resolution 98 μm × 98 μm, 1-mm-thick contiguous AACRJournals.org slices) was used for localization and assessment of tumors. Mice were weighed twice a week and sacrificed by cervical dislocation upon deterioration of condition, and tissue was taken for further analysis. For immunohistochemistry, sodium citrate (pH 6.0) heatmediated antigen retrieval was performed and staining was carried out using antibodies directed against human nuclear antigen (1:100, Millipore, MAB4383) diluted into 1% Tris buffer solution with 0.05% Tween-20 and incubated 1 hour at room temperature. Novocastra Novolink Polymer Detection Systems Kit (Leica Biosystem, RE-7150) was used for detection. Slides were mounted using a Leica CV Ultra mounting medium (Leica, 070937891).

Coronal Whole-Brain Organotypic Slice
All animal procedures were under the European Communities Council Directive N. 2010/63/EU and the Italian Ministry of Health guidelines (DL 26/2014) and approved by the Italian Ministry of Health and by the local Institutional Animal Care and Use Committee at Istituto Superiore di Sanità (Rome, Italy; protocol n. D9997.N.BYG, 2019). Coronal whole-brain organotypic slices encompassing the pons were prepared from CD1 mice pups (postnatal days 6-7; Charles River) as previously described (71), with some modifications. In brief, mice were decapitated and brains rapidly dissected and placed in ice-cold artificial cerebrospinal fluid containing (in mmol/L): 126 NaCl, 3.5 KCl, 1.2 NaH 2 PO 4 , 1.2 MgCl 2 , 2 CaCl 2 , 25 NaHCO 3 , and 11 glucose (pH 7.3), saturated with 95% O 2 and 5% CO 2 . The brain was then embedded in 3% SeaPlaque agarose (Lonza) in PBS and 300-μm-thick coronal slices were cut on a vibrating microtome (Campden Instruments), constantly cooled, and oxygenated. Each slice was transferred onto a porous membrane (0.45-μm pore size; Millipore), placed on a Millipore culture insert, and inserted into 6-well plates with 1.2 mL of cell culture medium/well, where the inserts were placed. The slices were incubated at 35°C, 5% CO 2 for 7 days before the experiments to allow the inflammatory reaction following the mechanical procedure to subside. Following the first day of culture, the medium was replaced with fresh medium and, from that time, changed every 48 hours. Seven days after slices were sectioned, ICR-B169 parental neurospheres with a diameter of 250 to 300 μm (1 neurosphere/ slice) were implanted on the pontine area, and following 3 days of coculture, preparations were treated for 4 days with 1 μmol/L dasatinib, 0.1234 μmol/L trametinib, or both compared with vehicle control (DMSO). After this time, slices were fixed with 10% buffered formalin for 2 hours at room temperature (RT) and washed twice with PBS for 10 minutes. Slices were then permeabilized with 1% Triton in PBS for 90 minutes and then blocked with 10% goat serum + 1% BSA + 0.1% Triton for 60 minutes. Slices were incubated O.N. with anti-human nuclei antibody (Millipore, 1:300). Slices were washed with PBS twice for 10 minutes and incubated with the secondary antibody (Alexa Fluor-555 goat anti-mouse, 1:500 O.N.; Invitrogen). Hoechst33342 was used as a counterstain (1:10,000 in PBS for 45 minutes at RT; Invitrogen). Images were taken on an Operetta CLS (PerkinElmer) in confocal mode (10×; z-stack 42 μm).

Statistical Analysis
Statistical analysis was carried out using R 3.3.0 (THe R Project for Statistical Computing) and GraphPad Prism 8 (GraphPad Software).
Categorical analyses were carried out by ANOVA, with multiple comparisons subjected to the Dunnett test. In vitro dose-response effects were compared by AUC analysis. Normalized and scaled abundances proteins and phospho-peptides were compared by using a Student t test. For these and gene expression data analysis, multiple testing corrections were made using FDR according to Benjamini and Hochberg. Effects of drug treatment on survival as the primary endpoint and overall survival in the orthotopic in vivo models were assessed using the Mantel Cox log-rank test. Adjusted P values <0.05 were considered significant.

Data Accessibility
All newly generated sequencing data have been deposited in the European Genome-phenome Archive (ebi.ac.uk/ega) with accession number EGAS00001004495 (sequencing) or ArrayExpress (ebi.ac.uk/arrayexpress/) with accession number E-MTAB-9282 (methylation arrays). Proteomic data have been deposited in the Proteomics Identifications Database (PRIDE; ebi.ac.uk/pride) with accession number PXD019701. Curated gene-level mutation, copy number, and expression data are provided as part of the pediatricspecific implementation of the cBioPortal genomic data visualization portal (pedcbioportal.org).