Diffuse intrinsic pontine gliomas (DIPG) are the most severe pediatric brain tumors. Although accepted as the standard therapeutic, radiotherapy is only efficient transiently and not even in every patient. The goal of the study was to identify the underlying molecular determinants of response to radiotherapy in DIPG.
We assessed in vitro response to ionizing radiations in 13 different DIPG cellular models derived from treatment-naïve stereotactic biopsies reflecting the genotype variability encountered in patients at diagnosis and correlated it to their principal molecular alterations. Clinical and radiologic response to radiotherapy of a large cohort of 73 DIPG was analyzed according to their genotype. Using a kinome-wide synthetic lethality RNAi screen, we further identified target genes that can sensitize DIPG cells to ionizing radiations.
We uncover TP53 mutation as the main driver of increased radioresistance and validated this finding in four isogenic pairs of TP53WT DIPG cells with or without TP53 knockdown. In an integrated clinical, radiological, and molecular study, we show that TP53MUT DIPG patients respond less to irradiation, relapse earlier after radiotherapy, and have a worse prognosis than their TP53WT counterparts. Finally, a kinome-wide synthetic lethality RNAi screen identifies CHK1 as a potential target, whose inhibition increases response to radiation specifically in TP53MUT cells.
Here, we demonstrate that TP53 mutations are driving DIPG radioresistance both in patients and corresponding cellular models. We suggest alternative treatment strategies to mitigate radioresistance with CHK1 inhibitors. These findings will allow to consequently refine radiotherapy schedules in DIPG.
Diffuse intrinsic pontine gliomas (DIPG) are the most severe and common form of malignant brain tumors in children and adolescents. Radiotherapy is the only proven efficient treatment, albeit only bringing a transient response followed by a constant progression within a few months. Our study identifies the inactivation of TP53 as the main driver of resistance to radiation in DIPG cells and TP53 mutation status as a biomarker of poor response to radiotherapy in patients both in terms of magnitude and duration. These results could be used to tailor radiotherapy schedules (including reirradiation) in DIPG patients depending on their TP53 status. Finally, we suggest the potential of mitigating radioresistance with CHK1 inhibitors in this setting.
Brain tumors represent the first cause of death from cancer in children, adolescents, and young adults. High-grade gliomas are the most frequent and aggressive forms. Diffuse intrinsic pontine gliomas (DIPG) are the most severe of these high-grade gliomas and have been recently associated with other diffuse midline gliomas sharing the same driving histone H3-K27M mutation (1). With respect to treatment, DIPGs are not resectable due to their deep-seated location in a sensible area of the brainstem as well as their infiltrative nature, which precludes any surgical attempt. Since the discovery of the effect of radiotherapy (RT) on brainstem tumors (2), this therapeutic modality has remained the only validated treatment for DIPG. This usually consists in three-dimensional conformal photon-based RT to a range of 54 to 59.4 Gy given in 30 to 33 fractions of 1.8 Gy daily (3). Changes in fractionation, cumulative dose as well as the use of various radiosensitizers have failed to improve its efficacy (4). No progress has been made since then despite the concomitant and adjuvant use of various drugs in more than 200 trials (5). Prognosis has not significantly changed, and most children die within the first 2 years after diagnosis (6). Besides, resistance to standard chemotherapy has been well demonstrated in vitro in DIPG cells derived from both diagnostic and autopsy samples (7, 8).
The K27M somatic mutation described in DIPG occurs in H3F3A, encoding the histone variant H3.3, or HIST1H3B/C and more rarely HIST2H3A/C, encoding canonical histones H3.1 and H3.2, respectively (9–11). Clinical work from our group has suggested that response to RT, albeit transient, was not uniform and could be associated with the type of histone H3 harboring the K27M mutation (11, 12), but resistance to ionizing radiation (IR) has not been explained so far. It seems therefore important to explore resistance to RT in DIPG and its determinants in order to invent efficient therapeutic combinations with irradiation.
We thus hypothesized the existence of molecular determinants of the response to radiation, considering histones H3 mutations but also alterations in other genes. Indeed, analysis of DIPG mutational landscape by next-generation sequencing (NGS) has shown additional recurrent genetic changes such as PDGFRA amplification (13), mutations in ACVR1, TP53, and components of the PI3K/mTOR pathway (14–18). Cosegregation of ACVR1 mutation with H3.1-K27M on the one hand and TP53 alteration with H3.3-K27M on the another hand were also reported (15).
First, we used a variety of cellular models of DIPG derived from stereotactic biopsies at diagnosis (i.e., treatment-naïve) to evaluate their radiosensitivity according to their molecular profile. Then, we extended our preclinical findings to patients by correlating their clinicoradiological response to RT with their genomic profiling according to these in vitro findings. Finally, we identified for the first time with a kinome-wide shRNA screen synthetic lethalities with IR in DIPG cells.
Materials and Methods
Cells and culture
Glioma stem-like cells (GSC) derived from stereotactic biopsies at diagnosis (19). Control human neural stem cells hNSC1 were from commercial origin (Applied Stem Cells, ASE-5001), and hNSC2-3 were derived from human embryo of Carnegie stages 18 and 22 obtained following voluntary abortions (20). Tissue collection and use were performed as previously described (21). All cells were maintained at 37°C, 5% CO2, used at less than 20 passages and tested monthly for mycoplasma contamination (22).
Determination of cell survival after irradiation in vitro
Cells were plated at 10,000 to 30,000 cells/cm2 and were irradiated 1 day after at indicated doses (320 kV, 4 mA, and 1.03 Gy/min) with X-ray IR (X-RAD 320 Biological Irradiator, Precision X-Ray Inc.) at 20°C under normoxic atmosphere. Cell growth was monitored daily by videomicroscopy (Incucyte ZOOM, Essen bioscience) over 3 weeks. Half of the medium was renewed every 3 to 4 days. For each dose, the surviving fraction was calculated as described by Buch and colleagues (23) based on the delay required by irradiated cells to reach a specific relative confluence (usually 3) compared with nonirradiated cells taking into account the doubling time of each GSC (Fig. 1A). The confluence threshold was chosen so all considered cells were in exponential growing phase.
Lentiviral shRNA cloning
Two shRNAs targeting TP53, shTP53-1 and shTP53-2, were cloned in a pLKO.1 lentiviral vector (Supplementary Table S1). Oligonucleotides (Eurofins) were annealed and cloned according to the Broad Institute shRNA cloning protocol (GPP Web Portal N0000003755).
As negative controls, two nontargeting shRNAs were used: shCTL-1, a vector containing a non-hairpin insert (Addgene #10879) and shCTL-2, a nonmammalian targeting shRNA (SHC002, Sigma-Aldrich). The puromycin resistance gene downstream the human PGK promoter was replaced by the NLS-tagged fluorescent protein mTagGFP.
Lentiviral production and transduction
Lentiviral particles were produced in HEK293T cells using psPax2 and pMD2.g second-generation packaging plasmids (Addgene #12260, #12259) with jetPRIME Polyplus transfection reagent, and lentiviral titers were determined by fluorescence assay (24). GSCs were transduced at a multiplicity of infection (MOI) of 0.3, and mTagGFP-positive cells were sorted by FACS (FACSAria III, BD Biosciences) after 5 to 7 days. mCitrine-NLS cells were transduced at an MOI of 3.
Protein extraction and immunoblotting
Proteins were extracted as described previously (25), separated on a 4% to 20% polyacrylamide gel (Biorad), and transferred to PVDF membrane (Biorad) with a Trans-Blot Turbo system (Bio-Rad). Membranes were incubated at 4°C overnight with TP53 (#sc-126; Santa Cruz Biotechnology, 1/500) and Cyclophilin A (Bio-Rad, 1/1,000) antibodies in 5% BSA in TBS-Tween20 and with horseradish peroxidase–linked secondary antibody for 1 hour at 20°C (#7076, Cell Signaling Technology, 1/5,000), enhanced by chemiluminescence reagent (#34095, Thermo Scientific) and analyzed with ChemiDoc MP Imaging System (Bio-Rad) with ImageLab 4.1 software.
Clinical samples and data
Clinical samples (n = 79) were retrieved from three retrospective cohorts fully characterized previously (11, 14, 15), and we performed a retrospective analysis of medical and imaging records of these 79 DIPG patients followed at Gustave Roussy between 1996 and 2016. We completed genotyping for H3F3A, HIST1H3B/C, TP53, ACVR1, and PPM1D by targeted sequencing of DNA for samples not evaluated by NGS. Patients without H3-K27M mutation were excluded because they do not represent a homogenous category of tumors according to the World Health Organization 2016 update (1). We focused on a cohort of 73 H3-K27M–mutated patients with corresponding full clinical record: age, signs, symptoms at diagnosis, MRI scans, treatment, response, and evolution over time.
Clinical response to RT was defined into three classes: worse as appearance of new symptoms or significant increase of the severity of an existing symptom, stable as lack of positive clinical response and absence of new symptoms, and improvement as disappearance of two or more main initial clinical symptoms leading to a general clinical improvement lasting more than 2 months concurrent with steroids weaning. Radiological response was evaluated on MRI scans performed within 6 weeks after completion of RT compared with pre-RT MRI scans and defined into four classes: worse with evidenced tumor growth, stable in case of volume changes inferior to 30% without appearance of new lesions, improvement when the tumor volume shrank by more than 30%, and pseudoprogression in case of radiological findings suspicious of tumor progression with a new or enlarging area of contrast agent enhancement which decreased or stabilized without a change in anticancer therapy. When these radiological changes were symptomatic, subsequent clinical improvement was necessary to consider this event as pseudoprogression. Functional imaging criteria were also considered as previously described (26). Assessments were made blindly by two independent reviewers. The time to first progression (TTP1) was defined as the time from RT start to first progression post-RT based on imaging and overall survival (OS) as the time from diagnosis to death.
For clinical data, we described and compared the clinical, radiological response according to H3-K27M status or TP53 mutational status by a Fisher exact test. A multivariable logistic model (Firth's approach) was performed by including age at diagnosis, histone H3 mutational status, and TP53. We used a Kruskal–Wallis test to compare the distribution of TTP1 across histone H3 and TP53 genotypes. OS was estimated by using the Kaplan–Meier method and compared with a log-rank test. Multivariable Cox regression models on TTP1 and OS (with Firth's approach) were performed by including age at diagnosis, histone H3, and TP53 mutational status. For TTP1 and OS, sex was not considered of interest. Statistical analyses were performed using PRISM software v7 (GraphPad) and SAS (version 9.4). All tests were two-sided and considered significant with a P ≤ 0.05.
Targeted DNA sequencing
Targeted sequencing was performed by the Biology and Medical Pathology platform of Gustave Roussy on an Ion-PGM System (Life technologies). Libraries were generated using the Ion AmpliSeq Library kit 2.0 (Life Technologies). The custom-designed ampliseq primer pools covered 100% of PPM1D and 99.8% of TP53 coding sequences, as well as hotspot loci in ACVR1, HIST1H3A-J, H3F3A, H3F3B, H3F3C, and HIST2H3A-D. Data were analyzed with the Torrent Suite Variant Caller software and annotated using hg19 (GRCh37). Variants were visualized using Alamut Visual version 2.9 (Interactive Biosoftware). An average of 7,535X coverage depth was obtained in all samples.
Kinome-wide shRNA screen
Fifteen millions of GSC7-8-9 were transduced with the MISSION human kinase lentiviral shRNA pooled library containing 7,450 shRNAs targeting 642 genes (Sigma-Aldrich) at an MOI of 0.3 in order to get an average 600-fold shRNA representation. Eight days after transduction and antibiotic selection, cells were irradiated at 4 Gy (LD25) and amplified during 64 days before gDNA extraction (QIAmp DNA Blood Maxi Kit, Qiagen). shRNA constructs were PCR amplified following the Broad/GPP protocol, purified (Agencourt AMPure XP, Beckman Coulter), and sequenced (1,000X depth, HiSeq4000, Illumina). Reads were mapped to shRNA library with bowtie v1.1.2. Read count per shRNA was normalized using the following equation:
A multilevel selection criterion was applied for candidate gene identification: (i) shRNAs with more than 50 reads in raw data of untreated cells; (ii) shRNAs with a fold-change decrease equal or superior to 2 in irradiated compared with untreated cells; and (iii) selection of candidate genes with at least three distinct shRNAs decreasing following irradiation.
Evaluation of drug effect in vitro
Cells stably expressing mCitrine-NLS were plated at 20,000 cells/cm2 (n = 6 for each condition). The day after, cells were treated with concentrations of CHK1 inhibitor Prexasertib (LY2606368, Selleckchem) as indicated, with or without IRLD50 5 hour after Prexasertib treatment. Cell growth was monitored by videomicroscopy for 120 hours. To assess viability, medium was supplemented with 80 nmol/L of YOYO-3 Iodide (#Y3606, ThermoFisher Scientific). The fraction of image area occupied by cells in phase, mCitrine-NLS (green fluorescence), or YOYO-3–positive nuclei (red fluorescence) was determined with CellPlayer Analysis software (Essen Bioscience) and expressed as a percentage of confluence.
Cells were plated the day before treatment on coverslips coated with 0.1 mg/mL Matrigel (Corning), fixed at the indicated time of 5 minutes with 1X PBS/4% paraformaldehyde, permeabilized for 5 minutes with 0.5% Triton X-100, and blocked for 30 minutes in 1X PBS/5% normal goat serum (ThermoFisher). Cells were incubated overnight at 20°C with anti-γH2AX (#05-636, clone JBW301, Merck, 1:10,000) and anti-53BP1 (#ab172580, Abcam, 1:1,000) and 45 minutes with secondary antibodies (Alexa fluor, Invitrogen, 1:500) and Hoechst 33342 (#H3570, ThermoFisher, 10 μg/mL). Slides were mounted with Fluoromount-G (SouthernBiotech). Images were acquired with a Leica SP8 confocal microscope. Foci were counted using ImageJ software. For combination treatment, drug was added to the medium, and cells were immediately (<30 minutes) irradiated.
Cell-cycle analysis by flow cytometry
For propidium iodide (PI) staining, cells were harvested and stained according to the manufacturer's instructions (#P1304MP, ThermoScientific). Samples were analyzed on a C6 BD Accuri flow cytometer. Cell-cycle profiles were generated using ModFitLT 5.0.9 software. For dual staining with EdU and PI, cells were exposed during 1 hour to 10 μmol/L of EdU supplied with Click-iT Flow Cytometry Assay Kit (#C10419, ThermoScientific). Samples were acquired using an LSR Fortessa flow cytometer (BD Biosciences), and analysis was perfomed in Kaluza 2.1 software.
For treatments combination, drug was added to the medium, and cells were immediately (<30 minutes) irradiated.
Distribution of DIPG stem cells' radiosensitivity in vitro
In order to decipher the underlying molecular basis of the variable clinical response of DIPG patients to RT, we assessed the cellular consequences of IR in vitro using DIPG cellular models deriving from primary tumors at diagnosis (19). We could not perform in all cell cultures classical clonogenic assays commonly used to evaluate radiosensitivity. Indeed, GSCs presented a high motility when cultured as adherent monolayers in laminin-coated flasks, or presented highly heterogeneous and most often poor survival in semisolid culture as single cells in collagen gel, soft agar, matrigel, or methylcellulose (Supplementary Fig. S1). We thus decided to follow proliferation of GSCs treated or not with IR over 3 weeks by videomicroscopy, in order to mathematically infer the surviving fractions from the growth curves (Fig. 1A). This alternative assay was previously shown to provide comparable information to clonogenic studies in several cell lines (23). Irradiation of H3.1-K27M GSC5 with doses ranging from 0.25 to 4 Gy induced a dose-dependent latency in cell growth kinetics (Fig. 1B; Supplementary Fig. S2). In contrast, H3.3-K27M GSC8 showed very mild alteration of its proliferation for such doses, requiring 6 to 8 Gy to observe a significant decrease of the relative confluence (Fig. 1C). These experiments were repeated 4 times and led to similar results allowing to calculate surviving fractions at each dose for all the cellular models, representative results of two GSCs being represented in Fig. 1D. The Lethal Dose 50% (LD50) was then defined on these survival fraction curves, i.e., 0.70 and 5.95 Gy for GSC5 and GSC8, respectively.
The radioresistance of in vitro avatars correlates with the time to first progression post-RT in patients
We extended our analysis to a total of 13 GSCs with K27M mutations in H3.1/H3.2 canonical histones H3 or in the H3.3 variant, and confirmed the wide spectrum of response to radiation in the cells with LD50 ranging from 0.5 to 7.4 Gy (Fig. 1E). We then compared the cellular response to IR in vitro with the time to first progression post-RT (TTP1) observed in the corresponding patient from which the tumor models were derived. TTP1 also varies importantly ranging from 1.5 to 17 months, and GSCs LD50 in vitro negatively correlated with TTP1. Patients presenting a longer clinical response led to GSC models with lower LD50 to RT, confirming the relevance of our in vitro results of radiosensitivity (Fig. 1E; Spearman correlation coefficient −0.71, P = 0.0086).
In this large dataset, radiosensitivity was not significantly correlated to the type of histone H3 mutated. Eighty-five percent of the models had LD50 below 2 or above 6, with only two GSCs presenting an intermediate radiosensitivity, the H3.2-K27M cellular model and one H3.3-K27M (GSC12). Indeed, although 4 out of the 5 H3.1-mutated cells were associated with LD50 around or below 2 Gy, the last one presented the highest LD50 (7.4 Gy; Fig. 2A). Similarly, H3.3-mutated cells spanned highly variable LD50 suggesting the existence of additional determinant affecting the intrinsic radiosensitivity in DIPG.
Analysis of secondary driver mutations in GSCs identifies TP53 alteration as a molecular determinant of radioresistance in DIPG
In order to identify the molecular origin of the difference of radiosensitivity in these GSCs, we profiled the most common recurrent mutations found in DIPG: AVCR1, TP53, PPM1D, ATRX, or in genes belonging to the PI3K/AKT/MTOR pathway. Our selected panel of GSCs contained the canonical H3.1-K27M and ACVR1 or H3.3-K27M and TP53 double-mutated samples (n = 4 for both; Fig. 2A). It was also enriched for less frequent genotypes with H3.3-K27M and TP53WT or PPM1DMUT GSCs, and even rarer genotypes such as H3.1/2-K27M and TP53MUT (Fig. 2A). Alterations in the PI3K/AKT/MTOR pathway were found in 4 of 5 H3.1-K27M– and 3 of 7 H3.3-K27M–mutated samples and included mutations of PIK3CA, PIK3R1, and AKT3.
DIPG cells appeared to separate into two main groups according to their LD50. All GSCs with an LD50 below 2 Gy were TP53WT, and conversely all TP53MUT belonged to the radioresistant subgroup, with significantly different average LD50 of 1.1 Gy ± 0.5 and 5.5 Gy ± 1.7, respectively (Fig. 2B; Supplementary Table S2; P < 0.0001). Despite some heterogeneity in the LD50, none of the most frequent genetic alterations studied here aside from TP53 mutation seemed to correlate with LD50 values of irradiation.
Three control hNSCs were included in the analysis. Their average LD50, i.e., 0.58 Gy ±0.22, was close to TP53WT GSCs associated with the lowest LD50. In contrast, TP53-mutant cells proved to be more resistant to radiations (P = 0.0015, Fig. 2B). With respect to PPM1D-mutant GSCs, their radiosensitivity was not significantly higher than NSCs and TP53WT GSCs (Fig. 2A).
TP53 knockdown increases in vitro radioresistance of TP53WT DIPG cells
In order to specifically evaluate the influence of the TP53-pathway inactivation on response to radiation, we measured LD50 in isogenic cells with inactivation of TP53. For this, we transduced 2 H3.1-K27M and 2 H3.3-K27M GSCs associated with low LD50 with lentiviral constructs leading to the expression of two independent shRNAs targeting TP53, or negative control shRNAs. As expected, TP53 protein level increased following a 4 Gy irradiation in both nontransduced and cells transduced with control shRNAs (Fig. 3A and B). Cells showed decreased TP53 levels, even greater post-IR, following TP53-shRNAs introduction. Next, the response to IR of these cells was analyzed as previously, and all 4 GSCs tested exhibited a significantly increased LD50 after TP53-KD (Fig. 3C). Interestingly, this increase was overall higher in all shTP53-2–transduced cells in accordance with the more efficient repression of TP53, excepted for GSC6 which displayed important variation among replicates in LD50 evaluation (Fig. 3B and C). Similar results were observed after TP53 inhibition in both TP53WT and PPM1DWT (GSC4-6) and TP53WT and PPM1DMUT models (GSC11). The comparison of LD50 in isogenic TP53WT and TP53-KD cells thus confirmed the pivotal role of TP53 in inducing DIPG radioresistance whichever the histone H3 mutated. As a control, we introduced the shRNAs in hNSC1 and also in two GSCs TP53MUT, and shTP53 increased radioresistance of hNSC1 (Fig. 3D). Conversely, the KD of TP53 sensitized TP53MUT GSCs, thus indicating a contribution of the mutated TP53 isoforms to the radioresistance beyond a loss of function (Fig. 3E).
Patients with a TP53MUT DIPG do not respond to RT
Given the central role of TP53 mutations identified on in vitro radioresistance of DIPG cells, but also the preferential association of TP53 and H3.3-K27M mutations in these tumors (11, 15), we evaluated in detail the clinical and radiological responses to RT, TTP1, and OS according to these markers (Supplementary Table S3).
We first analyzed the clinical response to radiation according to the TP53 status, and identified that TP53WT patients responded better, as opposed to TP53MUT patients with 86% and 44% good clinical responders, respectively (Fig. 4A; P = 0.0030). Thus, we compared the radiological response, i.e., the evolution of the tumor volume after RT followed by MRI. Again, the TP53WT subgroup comprised almost exclusively good radiological responders to RT, with an even stronger difference with TP53MUT patients (92% and 39%, respectively; Fig. 4B; P = 0.0003). Conversely, there was no significant difference in the clinical and radiological response when comparing patients according to the H3 mutational status of their tumor (H3.1-K27M and H3.3-K27M; Fig. 4C and D; P = 0.1543 and P = 0.2018, respectively). Similar results were observed after regrouping the clinical and radiological responses into two classes of response/no response (Supplementary Fig. S3A and S3B for TP53; P = 0.0014 and P ≤ 0.0001 and Supplementary Fig. S3C and S3D for H3 mutational status; P = 0.0958 and P = 0.1429, respectively).
Because we observed a correlation between H3.3-K27M and TP53 mutations in our cohort with more TP53MUT in H3.3 K27M (75.5%) compared with H3.1-K27M (13.6%; P < 0.0001; χ2 test), we decided to analyze concomitantly the contribution of these alterations to the response to RT. Because of the low headcounts and multiple genotypes, we used the clinical and radiological response in two classes (response and no response). A multivariable logistic regression including age at diagnosis, TP53, and H3 mutational status confirmed the previous analyses with an ORMUT/WT = 0.14 [95% confidence interval (CI), 0.03–0.68; P = 0.015] and ORMUT/WT = 1.17 (95% CI, 0.20–6.90; P = 0.864), respectively for TP53 and H3 for predicting a clinical response, and ORMUT/WT = 0.02 (95% CI, 0.001–0.46; P = 0.013) and ORMUT/WT = 8.24 (95% CI, 0.36–190.04; P = 0.188) for TP53 and H3 for predicting a radiological response, respectively (Supplementary Table S4). For the latter, the large variability was explained by the small number of patients.
Altogether, this confirmed that TP53 alterations but not the type of mutated histone H3 are theranostic biomarkers of the response to RT in patients.
Histone H3.3- and TP53-mutated DIPG patients display a worse prognosis
We next thought to study the disease course in patients and analyzed TTP1 after RT. Univariate analysis showed both a significantly earlier relapse for TP53MUT patients compared with TP53WT (median of 4.4 vs. 8.3 months, P = 0.0003; Kruskal–Wallis test) and for H3.3-K27M patients compared with H3.1-K27M (median 4.5 vs. 9.6 months, P = 0.0001; Kruskal–Wallis test). We further found significant differences according to the combination of the TP53 and H3-K27M status: 9.8 months for H3.1-K27M, TP53WT versus 5.8 months for H3.3-K27M, TP53WT versus 4.0 months for H3.3-K27M, TP53MUT (Fig. 4E; P = 0.0369 and P < 0.0001, respectively; Kruskal–Wallis test). The multivariable analysis, including age at diagnosis, TP53, and histone H3 mutational status, showed that only histone H3.3-K27M mutation was marginally associated with TTP1 (HR, 1.83; 95% CI, 0.95–3.53; P = 0.0701; Table 1). We next analyzed OS and showed that H3.3-K27M, TP53MUT patients have the worse prognosis with a median OS of 8.5 months, followed by H3.3-K27M, TP53WT with 12.1 months, and H3.1-K27M, TP53WT with 15.1 months (Fig. 4F, P = 0.0005, global log-rank test). The multivariable analysis including age at diagnosis, TP53, and histone H3 mutational status showed that H3.3-K27M mutation was significantly associated with OS (HR, 2.29; 95% CI, 1.17–4.48; P = 0.0154), but TP53 mutational status was only marginally associated with OS (HR, 1.69; 95% CI, 0.94–3.02; P = 0.0766; Table 1). As opposed to the clinical and radiological responses to RT, H3 mutational status was thus the better predictor for worse prognosis.
|.||TTP1 .||OS .|
|Characteristics .||HR (95% CI) .||P value .||HR (95% CI) .||P value .|
|Age at diagnosis||1.009 (0.908–1.120)||0.8745||0.927 (0.845–1.017)||0.1090|
|H3.3-K27M||1.833 (0.951–3.532)||2.290 (1.171–4.477)|
|TP53 mutated||1.289 (0.719–2.312)||1.690 (0.945–3.020)|
|.||TTP1 .||OS .|
|Characteristics .||HR (95% CI) .||P value .||HR (95% CI) .||P value .|
|Age at diagnosis||1.009 (0.908–1.120)||0.8745||0.927 (0.845–1.017)||0.1090|
|H3.3-K27M||1.833 (0.951–3.532)||2.290 (1.171–4.477)|
|TP53 mutated||1.289 (0.719–2.312)||1.690 (0.945–3.020)|
Kinome-wide RNAi screen identifies targets sensitizing TP53-mutated DIPG to IR
Having identified TP53MUT as the main driver of radioresistance in H3-K27M–mutated DIPG, we thought of identifying genes which extinction sensitize DIPG cells to RT. To this end, we performed a synthetic lethality RNAi screen in 3 H3.3-K27M GSC models with TP53MUT using shRNAs targeting the human kinome to identify those that selectively impair cell survival in combination with IR at LD25 (4 Gy; Fig. 5A).
Deep sequencing assessed the frequency of each shRNA 64 days after transduction, and those associated with a specific decrease in irradiated cells (logFC < –2) were selected. We then focused on the 36 candidate kinases common to the 3 GSCs. Among these, several genes are involved in DNA-damage response such as ATM, CHK1, PLK1, or WEE1. We first evaluated ATM (AZD1390), WEE1 (MK-1775), and CHK1 (Prexasertib, LY2606368) inhibitors on two GSCs. It was possible to determine an IC50 only for Prexasertib, but for ATM and WEE1 inhibitors IC50, these were out of the range of tested concentration (>400 nmol/L; Supplementary Figs. S4 and S5). We chose therefore to focus on Prexasertib as it also showed a good brain penetrance in orthotopic mice xenografts (27).
We evaluated the effect of Prexasertib in vitro on proliferation of H3-K27M ± TP53MUT GSCs and hNSC1, with or without IR at LD50. Without IR, we observed a dose-dependent effect in all cells with IC50 ranging from 11.2 to 20.6 nmol/L (Fig. 5C). We observed a shift toward lower IC50 of Prexasertib in dose–response curves specifically in irradiated TP53MUT GSCs (Fig. 5C). Further, we measured the consequence of the treatment with the cell impermeant dye YOYO-3 on mCitrine-NLS–expressing cells (Supplementary Fig. S6). We observed a higher cytotoxic effect and a significant difference between Prexasertib versus Prexasertib/IR combination specifically in TP53MUT DIPG cells (Fig. 5D; Supplementary Fig. S7). To get more mechanistic insight, we then studied post-IR double-strand breaks (DSB) repair kinetics using γH2AX and 53BP1 stainings. The repair kinetics were mostly identical for the different GSCs analyzed independently of their genotypes (Supplementary Fig. S8A and S8B). After 4 Gy, TP53MUT GSC8 had significantly more unrepaired DSBs at 24 hours compared with nonirradiated cells as opposed to TP53WT GSC5 (Fig. 5E). Accordingly, GSC5 transduced with shTP53–2 showed more DSB after 4 hours than cells with shCTL-2 (Supplementary Fig. S8C and S8D). Cell-cycle analysis at different time points after 2 Gy showed an important increase of cells in G1 and to a lesser extent in G2–M in TP53WT GSCs together with a strong decrease in S phase. In contrast, TP53MUT GSCs did not display a G1 block, but rather accumulated in S and G2 (Supplementary Fig. S9A). In both cases, the effect was observed 6 to 48 hours after IR. Based on these results, we performed a more detailed analysis of the cell cycle at 8 hours after IR by using an EdU pulse in order to precisely evaluate replicating cells (Fig. 5F). In TP53WT GSCs, the G1 block was confirmed with the decrease of EdU-positive cells in early S as well as an accumulation in G2, whereas TP53MUT GSCs did not show a G1 block but accumulated in S and G2 (Fig. 5G; Supplementary Fig. S10B and S10C). The same influence of the TP53 pathway inactivation was observed in GSC5 after introduction of shCTL or shTP53 (Supplementary Fig. S10D). Treatment of cells with Prexasertib alone only induced a slight reduction of the proportion of cells in S phase in some GSCs, irrespective of their TP53 status, possibly mirroring a reduction of this phase of the cell cycle (Fig. 5G; Supplementary Fig. S10). In addition, GSC6 and other TP53WT GSCs appeared insensitive to the addition of Prexasertib to IR and remained blocked in G1 (Fig. 5G; Supplementary Fig. S10B). This contrasted with GSC8 and other TP53MUT cells, that showed after IR/Prexasertib combination a reduced proportion of cells in S phase, but without a G1 block as assessed by the presence of cells in early S, and correspondingly more cells in G1 and G2. Again, the same results were observed in GSC5 with shTP53 versus shCTL (Supplementary Fig. S10D). Finally, we examined γH2AX staining by immunofluorescence and observed a higher level of strong the pan-nuclear–positive staining with the Prexasertib/IR combination 8 hours after treatment in TP53MUT versus TP53WT GSCs (Supplementary Fig. S11). This was in line with the aforementioned data showing higher cytotoxicity in TP53MUT GSCs (Fig. 5D). The mechanism of action of Prexasertib in this setting could therefore be to enforce DNA replication despite the presence of DSB specifically in TP53MUT GSCs due to the absence of TP53 activity at the G1–S and CHK1 activity at the S phase and G2–M DNA damage checkpoints, eventually leading to cell death.
Response to RT in DIPG is not uniform and unequivocally only transient (9). In this work, we demonstrate that it relates to the heterogeneous mutational landscape of the disease and specifically the presence or not of TP53 mutations. Despite their high frequency in DIPG, around 42%, mutations in TP53 are not initiating this disease (15). Patients harboring germline TP53 mutations (Li–Fraumeni Syndrome) rather develop H3-K27 WT high-grade gliomas and rarely in the brainstem (28). The panel of 13 GSCs analyzed was selected to cover most combinations of the principal alterations encountered in DIPG at diagnosis including rare TP53MUT H3.1/2-K27M associations, without reflecting their relative frequencies in the DIPG population. This allowed us to assess in vitro DIPG radioresistance and screen for the contribution of the distinct genomic alterations. As a prerequisite to our explanatory experiments, we correlated LD50 in GSCs to TTP1 in patients and showed therefore that the heterogeneous response to RT observed in DIPG patients seems mostly defined by a cell-autonomous response to radiation that can be evaluated in vitro in a microenvironment-free system. The role of the stroma would deserve further studies in different settings. We demonstrated through a molecular analysis that the main driver of resistance to IR in DIPG cells is the alteration of the TP53 pathway in H3.3- and H3.1-mutated cells as well. In apparent contradiction, we previously found that histone H3.3-K27M mutations were associated with a worse response to RT (11), but we show in the present study that it does not affect radioresistance by itself neither in vitro nor in vivo. The discrepancy between these two studies likely results from a confounding effect as the majority of H3.3-K27M samples are also TP53MUT, whereas H3.1-K27M samples rarely are (29). TP53 dysfunction has been linked to radioresistance in various adult cancers and some pediatric models like neuroblastoma (30). Shown in a few clinical studies like in head and neck cancer (31), this finding has not been ascertained in clinical studies of glioblastoma (32–35). In preclinical models of glioblastoma however, TP53 dysfunction has been associated with radioresistance (36). In medulloblastomas, TP53 mutations have been associated with a worse prognosis after irradiation, and TP53 knockdown was associated with increased radioresistance in vitro (37). In pediatric nonbrainstem high-grade glioma, TP53 mutations have been linked to a worse outcome (38). We show here a strong impact of TP53 mutational status on DIPG response to RT and marginally on survival, identifying a subset of patients with extremely poor outcome.
PPM1D-truncating mutations causing the loss of its regulatory domain were reported to negatively regulate TP53 activity through increased phosphatase activity at Ser15 of TP53 (39, 40). Unexpectedly, the two PPM1DMUT GSCs tested did not show a marked elevated LD50 for irradiation in vitro, albeit their radiosensitivity was slightly lower than most TP53WT GSCs. Moreover, the increase in LD50 following TP53 inhibition was similar in PPM1DMUT, TP53WT and PPM1DWT, TP53WT GSCs. Phosphorylation at Ser15 by PPM1D is stimulating transactivation of TP53-responsive promoters (41), but since the transactivating role of TP53 is not its only mechanism of action, it cannot be expected to phenocopy completely the consequences of TP53 mutations or loss. In addition, the relative radiosensitivity of the two GSCs with PPM1D mutations may be explained by the conservation of the phosphatase activity of the protein because the mutations are at the C-terminus of the protein, i.e., downstream of the phosphatase domain. These two PPM1D mutants could therefore still inhibit CHK1 and even more since they lose only the degradation signal as shown previously (40). Thus, these cells could present an intermediate phenotype with respect to radiosensitivity because of simultaneous TP53 and CHK1 inactivation. Similarly, activating mutations in the PI3K/AKT/MTOR pathway did not overtly affect radioresistance in vitro, whereas mTOR inhibitors were shown to radiosensitize DIPG cells (42). In a recent study, PI3K/AKT/MTOR activation assessed through PTEN loss of expression however appeared as a common feature in DIPG, not restricted to tumors with mutations in genes of the pathway (43). One can therefore consider that the effect of mTOR inhibitors could be independent of genomic alteration of components of PI3K/AKT/MTOR pathway. Anyhow, only a slight difference in term of radiosensitivity was observed between NSCs and H3-K27M/TP53WT tumors suggesting that most of radioresistance in DIPG could be caused by TP53 dysfunction. Interestingly, two TP53MUT GSCs displayed an intermediate LD50 compared with other TP53-mutated and WT GSCs. GSC13 presented no LOH but two mutations (p.W146* and c.919+1G>T splice donor) outside hotspots and associated with neutral or unknown FATHMM pathogenic prediction in the COSMIC database (Supplementary Table S1). For GSC12, the TP53 mutation was more canonical (p.G266R), but cells however harbored a mutation in ATRX, and these were recently reported to increase sensitivity of gliomas cells to DSB-inducing agents by impairing nonhomologous end joining DNA repair (44).
Using univariate analysis, our study confirms that mutations in TP53 identify DIPG patients with the worst prognosis. This is in line with our previous report showing that patients with LOH at TP53 locus identified by CGH array of the tumor had a worse outcome with a median survival of 7 versus 12 months (13), recently confirmed by a meta-analysis of K27M-mutated gliomas (45). However, our multivariate analysis identifies the type of mutated histone H3 (i.e., H3.1 or H3.3) as a stronger predictor of survival in DIPG, despite not influencing the response to RT, whereas mutations in TP53 were only marginally associated with a worse prognosis, likely because of low headcount in the H3.3-K27M, TP53WT subgroup. Thus, the prognosis in DIPG appears to be partly linked to the response to RT, but also to the differences in disease evolution between H3.3- and H3.1-K27M, such as the rate of metastasis development or more generally differences in tumor–stroma interactions as previously shown (11).
The correlation of in vitro radioresistance and clinical response to IR indicated that in vitro GSCs mirror the bulk of tumor cells in vivo. This confirmed the relevance of using stem-like models of DIPG to study the disease in vitro, in accordance with recent results from Filbin and colleagues showing that most DIPG cells in vivo display a stem-like profile (46).
We subsequently screened for IR-induced synthetic lethality in the more aggressive H3-K27M, TP53MUT subgroup of DIPG and identified CHK1 as a vulnerability. Treatment with Prexasertib together with IR shifted the growth inhibition curves to lower doses only in TP53MUT cells (Fig. 5C). Also, combining Prexasertib and irradiation was more cytotoxic than each treatment alone only in TP53MUT GSC (Fig. 5D). In this latter subgroup, radioresistance was not associated with increased DNA repair efficiency, but rather an escape from the G1–S checkpoint. The combination of IR with Prexasertib in this TP53MUT context induced a faster progression from S phase to G2 through abrogation of the S phase and G2–M CHK1-dependent checkpoints without repair of DNA damage that led to an increased cell death. Prexasertib was previously shown to be brain penetrant in mice bearing orthotopic medulloblastomas (27). It reaches a free brain extracellular fluid concentration of 10 ng/mL (corresponding to 22 nmol/L) as measured using microdialysis, thus in the range of the IC50 we measured in our models. As Prexasertib also inhibits other kinases, such as CHK2 and RSK, although only 10 times less effectively (47), we cannot exclude that part of the effect on survival evidenced in our experiments could not be mediated by their inhibition as well, albeit none of these kinases were identified in the shRNA screen. In cellular assays, CHK2 was however the only additional kinase significantly inhibited by Prexasertib (47). This combination appears clinically promising especially in case of TP53MUT which could be used as a theranostic biomarker. In a previous study showing the role of BMI1 inhibition in the response to the radiomimetic drug bleocin in DIPG cell lines, CHK1 downregulation has been postulated as one of the downstream effectors (48). However, no information on the TP53 status of the cell lines was given. Interestingly, a clinical phase I/II study (NCT02808650) opened recently to evaluate Prexasertib efficiency in refractory pediatric tumors including brain tumors, but without combination with RT.
In conclusion, we showed that TP53 mutations explain most of the differences in radiosensitivity between DIPG GSCs and NSCs and can serve as a biomarker to predict radioresistance in DIPG patients. We suggest that these mutations should be considered when designing RT plans. As reirradiation is progressively becoming more frequently used at relapse (49), it seems to be possible to tailor this approach to patients most likely to benefit from it, i.e., those who responded initially with a prolonged progression-free interval of at least 3 to 6 months and who did not have a TP53MUT. The underlying mechanisms of DIPG radioresistance downstream and eventually beyond TP53 dysfunction will require further elucidations. However, CHK1 inhibition in combination with RT may offer new opportunities to overcome this resistance in DIPG with TP53 mutations.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: C. Werbrouck, C.C.S. Evangelista, J. Grill, M.-A. Debily, D. Castel
Development of methodology: C. Werbrouck, C.C.S. Evangelista, E. Barret, M. Mondini, J. Grill, D. Castel
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Werbrouck, C.C.S. Evangelista, M.-J. Lobón-Iglesias, E. Barret, R. Brusini, P. Varlet, K. Beccaria, N. Boddaert, S. Puget, J. Grill, D. Castel
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Werbrouck, C.C.S. Evangelista, M.-J. Lobón-Iglesias, G.L. Teuff, J. Merlevede, R. Brusini, T. Kergrohen, N. Boddaert, J. Grill, M.-A. Debily, D. Castel
Writing, review, and/or revision of the manuscript: C. Werbrouck, C.C.S. Evangelista, M.-J. Lobón-Iglesias, G.L. Teuff, R. Brusini, S. Bolle, N. Boddaert, J. Grill, M.-A. Debily, D. Castel
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.-J. Lobón-Iglesias
Study supervision: J. Grill, D. Castel
D. Castel, J. Grill, and M.-A. Debily acknowledge financial support from Société Française de Lutte contre les Cancers et les leucémies de l'Enfant et de l'Adolescent, Fédérations Enfants Et Santé, INCa (Grant PLBIO-14-253) and charities “The DIPG Collaborative,” “The Cure Starts Now,” and “L'Etoile de Martin.” C. Werbrouck was supported by a fellowship from “Course of Excellence in Oncology – Fondation Philanthropia,” C.C.S. Evangelista by a fellowship from National Council for Scientific and Technological Development (CNPq) Program "Science without borders" in Brazil, and T. Kergrohen by “Imagine for Margo.”
The authors are grateful to the Necker Imagine DNA biobank (BB-033-00065) and tumor bank, to the Necker operating room nurses/assistants for their technical assistance, to L. Lacroix and P. Saulnier from the Translational Research Platform of Gustave Roussy, and to Y. Lecluse and F. De Leeuw from the PFIC core cytometry and microscopy platforms of Gustave Roussy.
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.