Class I HDAC inhibition reduces DNA damage repair capacity of MYC-amplified medulloblastoma cells

HD-MB03 is a cell line established in our laboratory [14]. The MB cell lines ONS76: CVCL_1624, UW228-2: CVCL_0572, MED8A: CVCL_M137, D425: CVCL_1275 [6], as well as the non-transformed human foreskin fibroblast cell line VH7 [15] were obtained and cultured as previously mentioned. All cell lines were authenticated using Multiplex Cell Line Authentication (MCA) and tested for contamination by Multiplex cell Contamination Testing (McCT) by Multiplexion (Heidelberg, Germany) as described before [9]. Cultured cells were checked weekly with PlasmoTest™- (InvivoGen) and monthly with Venor® GenM Classic—(Minerva Biolabs) Mycoplasma Detection Kits for mycoplasma contamination. MYC amplification status of cell lines were verified previously [6].

HD-MB03 were treated in three biological replicates for 24 h and 48 h with 1 µM entinostat or DMSO solvent control. RNA extraction was done with RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. RNA quantity was assessed with Nanodrop ND-1000. Microarray-based gene expression profile analysis was performed at the Genomics and Proteomics Core Facility at the German Cancer Research Center (DKFZ) including sample preparation (quality control, reverse transcription, labelling and fragmentation) and data preprocessing. Sample hybridization was done on Human Genome U133 Plus 2.0 Affymetrix microarray according to the manufacturer’s instruction. After preprocessing Affymetrix CEL files were GCRMA normalized and log₂-transformed. Differentially expressed probesets (adj. p-value < 0.05) between treatments were identified using the empirical Bayes approach based on moderated t-statistics as implemented in the Bioconductor package limma. For comparisons based on multiple time points, a time point specific effect was accounted for in the linear model by performing a two-way ANOVA. Probesets showing an over-additive effect were identified by testing the interaction between time point and treatment by preforming a two-way ANOVA with cross-effects. Gene set enrichment analysis for KEGG, reactome and GO databases was performed using the camera test. In case a gene was represented by multiple probesets, the probeset with the strongest effect was selected for Gene Set Enrichment Analysis (GSEA). P-values were adjusted for multiple testing using the Benjamini–Hochberg correction in order to control the false discovery rate. All analyses were performed with statistical software R 3.6 [16]. Visualization of functional enrichment in networks was carried out using the Cytoscape plug-in EnrichmentMap [17]. Enriched genesets of the three databases with FDR < 0.1 served as input. Assembled networks and clusters were manually re-organized for biofunctional annotation. Additionally differentially expressed genes were analyzed through the use of Ingenuity Pathway Analysis (IPA) to identify enriched canonical pathways of the IPA database (IPA core-analysis) (QIAGEN Inc., https://​www.​qiagenbioinforma​tics.​com/​products/​ingenuity-pathway-analysis) [18]. Results from Cytoscape EnrichmentMap and IPA were summarized by manual grouping into overarching pathways representing circumscribed biological functions (Suppl. Table 4).

The following compounds were used: entinostat (Biomol GmbH, M4693-15A.25), idasanutlin (TargetMol; T6254), olaparib (BIOZOL; S1060), ribociclib (supplier), selinexor (BioCat; T6106-TM), vinblastinesulfate (Selleckchem; S4505), doxorubicin in 0.9% NaCl (UKHD pharmacy), niraparib (Selleckchem; S2741), pamiparib (Selleckchem; S8592), veliparib (Selleckchem; S1004). The translational drug library for initial drug selection comprised 74 compounds [19].

Cells were seeded in 96-well plates for dose–response investigation and matrix synergism analysis as previously reported [20], drugs were added with the D300e Digital Dispenser (Tecan). Metabolic activity was assessed after 72 h of treatment using the WST-8 based Cell counting Kit-8 (Dojindo, CK04-13) as previously reported [6].

72 h combination treatments of entinostat with idasanutlin, olaparib, ribociclib, selinexor and vinblastine on 96-well cell culture plates included 38 treatment conditions per combination in form of a 5 × 5 checkerboard matrix, applying DMSO normalization and plate randomization [21]. The 5 × 5 concentration matrix was built applying twofold dilution series surrounding each drug’s EC50 as previously reported [21]. After 72 h of treatment metabolic activity was assessed by WST-8 assay. Entinostat dose–response-curve (DRC) shift was analyzed in the presence of a fixed dose of the investigated combination drug. Fixed dose level was set to a concentration between single drug IC20-IC50. DRC left-shift was assessed visually and assigned to increased entinostat potency (i.e. comparable effects were achieved at lower concentrations). CompuSyn software (ComboSyn Inc., Paramus, NJ, USA) was used to calculate combination index (CI) values at three constant ratio combinations including an equipotent ratio-combination (EC 50: EC 50). SynergyFinder 1.0 web application was used to calculate synergism across the whole dose–response surface applying the ZIP synergy model [21].

Cell counts and viability were assessed using Vi-CELL™ Cell Viability Analyzer (Beckman Coulter) as previously described [6]. Treatment with 500 nM staurosporine served as control for maximal achievable effects. Combination index (CI) values were calculated using CompuSyn-Software (ComboSyn Inc., Paramus, NJ, USA).

Western blots were performed as previously described [9]. The following antibodies were used: c-MYC (Abcam; ab32072; 1:5000), PARP1 (CellSignalling; #9532S; 1:5000), Rad51 (GeneTex; GTX70230; 1:5000), γH2A.X (CellSignalling; #9718; 1:2000), beta-actin (Sigma; A5441; 1:10,000). Immunodetection was done by luminescence using Amersham ECL Prime Western Blotting Detection Reagent (GE Healthcare) with Azure Azure 400 imaging system (Azure biosystems).

Equal amounts of cells were fixed by dropwise addition of ice-cold 70% ethanol and incubated for 2 h at 4 °C. For cell cycle analysis cells were washed in 1% BSA – PBS – 0.2% Tween20 and subsequently incubated in PI staining solution (50 µg/ml PI, 5 µg/ml RNAse in 1% BSA—PBS—0.2% Tween20) for 20 min at 37 °C under light protection. Assessment of γH2A.X was done as previously described [22]. Flow cytometric analysis was done on FACS Canto II. Data analysis was performed using FlowJo software. After doublet exclusion using a SSC-A/SSC-W plot, cell cycle phases were assigned. γH2A.X high gates were set to contain ~ 5% of cells in the untreated condition representing intrinsic DNA damage. Mean normalized fluorescence intensity was calculated and normalized as previously reported [22].

Cells were seeded at a density of 30,000 per chamber into a µ-Slide 8 Well (Ibidi) 24 h before treatment. After 24 h treatment cells were washed with PBS and fixation was performed with 4% PFA for 20 min at room temperature. Fixed cells were washed with PBS, permeabilized with 0.2% Triton-X-100 in PBS for 30 min and blocked with 0.05% Triton-X-100—3% BSA—PBS for 1 h at room temperature. After incubation with γH2A.X antibody (1:200) at 4 °C overnight cells were incubated with Alexa-Fluor-488-labeled secondary antibody (ThermoFisher; A-11008; 1:400) for 2 h at room temperature and counterstained with DAPI (4′,6-Diamidin-2-phenylindol) for 10 min. Images were acquired on a Zeiss LSM 710 laser scanning confocal microscope (Oberkochen, Germany) with 40 × times magnification. Exemplary γH2A.X foci quantification was done semi-automatically using ImageJ for > 100 nuclei per condition for two biological duplicates.

Primary mRNA [4] and protein [23] expression data were derived from publicly available datasets. All in vitro experiments were performed in at least three biological replicates. Data across biological replicates were calculated and presented as mean values ± SD. GraphPad Prism version 5.01 (GraphPad Software, La Jolla, CA, USA) for Windows was used for dose–response curve modelling and calculation of half maximal effect concentrations (EC50) applying nonlinear regression curve fitting. Treatment groups were compared using One Way ANOVA and subsequent Bonferroni multiple comparisons testing, multiplicity adjusted P values are shown. P values < 0.05 were considered significant. γH2A.X foci distribution histograms were done with GraphPad Prism version 5.01 (GraphPad Software, La Jolla, CA, USA). Heatmap visualizations were generated using MORPHEUS web application (https://​software.​broadinstitute.​org/​morpheus).

To identify drugs that potentially interact synergistically with entinostat in MYC-amplified MB, first, we aimed to identify targetable pathways that were deregulated in response to entinostat single agent treatment (Fig. 1A). The MYC-amplified MB cell line HD-MB03 was treated with entinostat and changes in gene expression were determined by microarray-based gene expression profiling after 24 h and 48 h (Fig. 1B). Entinostat treatment led to pronounced changes in gene expression, with 9976 differentially expressed genes (DEGs) across both timepoints (Fig. 1B). The majority of all identified DEGs (n = 6108/9976; 61%) showed consistently either an up- or a downregulation. Of these consistently regulated genes 88 upregulated and 137 downregulated genes were computed to show over-additive behavior i.e. a total of 225 (225/6108 = 4%) DEGs displayed a continuously increasing or decreasing expression over time comparing 24 h and 48 h of treatment (Fig. 1B). A total of 119 (119/6108 = 2%) DEGs differed in their direction of regulation (n = 59 genes first up- then downregulated vs. n = 60 genes first down- then upregulated) comparing both timepoints (Fig. 1B).

We hypothesized that DEGs functionally involved in the previously described cytotoxic effect of entinostat treatment in MYC-amplified MB cells [6] are likely to be regulated consistently over time and thus, should be detectable as either up- or down-regulated at both timepoints investigated. In addition, we assumed that deregulated genes showing over-additive change in expression were indicative of a self-sustaining and thus biologically important process. Accordingly, a gene set encompassing all genes differentially regulated at both treatment timepoints (DEG_24/48) and a gene set of genes showing over-additive behavior comparing both timepoints (DEG_OA) were defined as gene sets of interest for the subsequent functional analysis (Fig. 1A). GSEA of DEG_24/48 revealed 1783 significantly (FDR < 0.25) enriched gene sets in the three databases KEGG [24] (n = 141), GO [25] (n = 1360) and REACTOME [26] (n = 282) (Suppl. Table 1). These gene sets clustered into 11 sets of enriched and 11 sets of depleted pathways as evidenced by functional enrichment visualization using cytoscape [17] (Fig. 1C). The respective analysis for DEG_OA resulted in 273 significantly enriched gene sets in KEGG (n = 12), GO (n = 68) and REACTOME (n = 193) databases (Suppl. Table 2), clustering into three enriched and nine depleted pathways (Suppl. Figure 1). To strengthen the informative output of this analysis we applied Ingenuity Pathway Analysis (IPA) [18] as a second algorithm to identify functional patterns of deregulated genes. In DEG_24/48 and DEG_OA IPA analysis resulted in 82 and 5 altered canonical pathways respectively (Suppl. Table 3). Accounting for partial functional overlap of identified pathways by GSEA/functional enrichment and IPA analysis we summarized the results by manually grouping all functionally related altered pathways into matching categories. These categories represented a total of n = 22 overarching biological functions and pathways affected by entinostat treatment in HD-MB03 cells (Suppl. Table 4; Fig. 1A). We reasoned that some of the identified pathways may be responsible for entinostat induced cytotoxicity or provide escape mechanisms to evade the cytotoxic stress of entinostat treatment, all of which could represent emerging vulnerabilities. We hypothesized that additional targeting of these pathways may enhance the efficacy of entinostat treatment. To select drugs for testing this hypothesis we compared the list of affected pathways with the mode of action of all drugs included in a published translational drug library (n = 74 drugs). This library includes drugs which are either clinically approved for oncological diseases or under clinical investigation for the treatment of cancers [19] (Suppl. Table 5). Based on literature research we identified n = 44 drugs from this library with known pharmacological interference with one of n = 9 pathways identified as altered upon entinostat treatment (Fig. 1A). We then filtered for drugs with high potential as novel combination partners for entinostat meeting the following two criteria: (1) suited for in vitro testing (leading to exclusion of drugs targeting the immune system); (2) no published data on combination with HDACi in MB (Fig. 1A). In case of two potential drugs targeting the same pathway the compound with highest blood–brain-barrier (BBB) penetrance according to the BBB score [27] was selected for further analysis. As all three vinca alcaloids vinblastine, vincristine and vinorelbine have the same calculated BBB score of 1.19 vinblastine was selected as a drug class representative with favorable clinical tolerability. The following five drugs met all filtering criteria: idasanutlin, olaparib, ribociclib, selinexor and vinblastine (Suppl. Table 6).

Single agent dose–response curves were determined in metabolic activity assays for all five drugs of interest in three MYC-amplified (HD-MB03, MED8A, D425) and two non-MYC-amplified MB cell lines (ONS76, UW228-2) (Suppl. Figure 2A and B). To be able to interpret the clinical relevance of the observed in vitro drug potency we calculated the ratio of each drug’s in vitro IC50 and the respective maximal clinical plasma concentration (c_max; Suppl. Table 7) (Fig. 2A), as well as each drug’s maximum effect achieved at a concentration below the respective c_max (Fig. 2B). Selinexor and vinblastine showed low IC50:c_max ratios and profound reduction of metabolic activity across all cell lines, indicating a strong overall cytotoxicity (Fig. 2A, B). Only the PARP inhibitor (PARPi) olaparib showed a low IC50:c_max ratio (Fig. 2A) as well as a strong inhibition of metabolic activity at concentrations below c_max (Fig. 2B) exclusively in all three MYC-amplified cell lines, suggesting both a translationally relevant single agent activity and a MYC dependent drug response pattern.

All five compounds were tested for synergistic interaction with entinostat in two MYC-amplified cell lines (HD-MB03 and MED8A) (Fig. 2C, D). We evaluated synergism applying three different methods to increase reliability: I) dose–response-curve (DRC) shift-analysis; II) Chou-Talalay combination index (CI) method; III) response-surface ZIP synergy computation (Fig. 2C, D; left to right). The most robust indication for synergism was observed for the combination of entinostat and olaparib showing synergistic interaction in all three synergy models in HD-MB03, and in two of three models (DRC and ZIP) in MED8A (Fig. 2C, D). Importantly, in both cell lines indications for synergism were observed at clinically achievable concentrations (Suppl. Table S7).

To confirm the sensitivity of MYC-amplified MB cells to single agent olaparib treatment to be a drug class effect the response of MB cells to three additional PARPis (niraparib, pamiparib, veliparib) was evaluated (Fig. 3A, Suppl. Figure 2C). All applied PARPis bind to the catalytic domain of PARP but differ in their PARP trapping ability, off-target kinase affinities, pharmacokinetic profiles and reported BBB penetration [28‐30]. MYC-amplified cell lines showed a trend towards or a significantly higher sensitivity to all three PARPis compared to non-MYC-amplified cell lines (Fig. 3A, Suppl. Figure 2C). Of note, significantly higher sensitivity of MYC-amplified cell lines was observed for veliparib, which has the lowest reported PARP-trapping activity [30] (Fig. 3A). The increased sensitivity of MYC-amplified MB cells to all applied PARPis raised the question whether there is functional interdependence of PARP and MYC protein. Interestingly, the analysis of previously published mass spectrometry data of protein pulldown after immunoprecipitation of endogenous MYC in HD-MB03 cells [9] revealed PARP1 peptides co-localizing with MYC ranking in the top 10^th percentile of all MYC-interacting peptides, suggesting a protein–protein interaction of MYC and PARP1 (Fig. 3B). Synergistic interaction of entinostat and PARPi as a drug class effect was validated with pamiparib, showing synergistic interaction with entinostat in both MYC-amplified cell lines tested (HD-MB03 and MED8A) in 2/3 models (DRC and ZIP) (Fig. 3C). Notably, for both olaparib and pamiparib, the synergistic interaction was observed within the range of clinically achievable plasma concentrations (Suppl. Table S7).

Continuing our investigation with the FDA-approved compound olaparib we aimed to validate the observed synergistic interaction in a different experimental setting. We performed cell count experiments showing a significant, dose-dependent reduction of the number of viable cells upon combination treatment in MED8A and HD-MB03 with computed combination indices below 0.5, indicating strong synergy for all tested conditions (Fig. 3D). Also in the non-MYC-amplified cell line ONS76 the number of viable cells was significantly reduced upon combination treatment with a combination index indicating synergy in one of two tested conditions (Fig. 3D). Conversely, cell viability was significantly reduced only in the MYC-amplified cell lines (Fig. 3E), indicating increased cell death as opposed to a possible cell cycle arrest in non-MYC-amplified cells.

We hypothesized that the observed synergy of entinostat and olaparib is functionally based on a joint inhibiting effect on the DNA damage response (DDR) leading to increased and potentially fatal accumulation of DNA damage. Indeed, we observed a strong overrepresentation of downregulated DDR gene sets (n = 4, see exemplary sets in Fig. 4A) in the top ranked gene sets with an FDR < 0.01 (n = 25) considering all genes with a time dependent over-additive change in expression upon entinostat treatment (DEG_OA) (Suppl. Table 2). Based on this observation we reasoned that entinostat may exert its effect on DDR indirectly by affecting the transcription of genes involved in the DNA damage repair machinery. In line with this observation, we detected a time- (Fig. 4B) and dose-dependent (Fig. 4C) increase in phosphorylated histone variant γH2A.X, an early marker of DNA double strand breaks (DSBs), upon entinostat single agent treatment. Taken together these data point towards a reduced DNA damage repair capacity of HD-MB03 cells upon entinostat treatment.

The role of PARP1 as a crucial initiator of DDR is well described [31]. To verify that olaparib can exert its described effect by inhibiting PARP1 in our model systems we performed target presence analysis. In all three cell lines we confirmed PARP1 protein expression with significantly higher PARP1 expression in the MYC-amplified cell lines (Fig. 4D). PARP1 expression was not significantly altered upon entinostat treatment in HD-MB03 cells on mRNA level (Suppl. Table 8: PARP1 unlisted). On protein level we observed significantly reduced but readily detectable amounts of PARP1 protein after entinostat treatment, (Fig. 4B, Suppl. Figure 3A). We confirmed target presence in primary MB tumors by analyzing publicly available mRNA [4] and protein [23] expression data sets. PARP1 showed strong expression on both mRNA (Fig. 4E top) and protein level (Fig. 4E bottom) in primary MYC-amplified MB tumors of subtype II with a trend towards or significantly higher expression compared to all other subtypes. Additionally, PARP1 expression significantly correlated with MYC expression on mRNA level in Group 3 tumors (Fig. 4F). PARP2 expression levels did not differ significantly between Group 3/4 subgroups I-VIII (Suppl. Figure 3B).

It has been previously shown that the cytotoxic effects of olaparib can be increased by concomitant treatment with DNA-damaging agents such as doxorubicin [32]. To test the impact of a DNA damaging agent in the presence of entinostat and olaparib we quantified the number of DNA DSBs by staining of the phosphorylated histone variant γH2A.X after 24 h of treatment with entinostat, olaparib and doxorubicin in single, double and triple drug treatment. Flow cytometry analysis revealed that single treatments generated comparable amounts of DSBs across all cell lines tested (Fig. 5A). The combination of entinostat and olaparib lead to a significantly increased number of DSBs compared to entinostat single agent treatment only in the MYC-amplified cell lines (Fig. 5A). Similarly, and in line with our hypothesis, addition of entinostat to the combination of olaparib and doxorubicin (OD) significantly enhanced the number of DSB formation only in the MYC-amplified cell lines (Fig. 5A). Cell cycle analysis by PI staining revealed that olaparib treated cells showed highest γH2A.X positivity in S-phase (Suppl. Figure 3C), in line with DSBs caused by PARP-trapping-induced replication fork collapse. Entinostat and doxorubicin single treatment did not show a cell cycle specific distribution of DSBs (Suppl. Figure 3C). Interestingly, upon triple treatment the MYC non-amplified cell line ONS76 showed strongest accumulation of DNA damage in G2/M phase, while the MYC-amplified cell lines showed the lowest number of DNA damage in G2/M phase (Suppl. Figure 3C). Comparing the number of cells without (no γH2A.X foci per cell) and with strong accumulation (≥ 10 γH2A.X foci per cell) of DNA damage revealed a significant increase of cells with ≥ 10 γH2A.X foci per cell upon triple treatment compared to DMSO control in MYC-amplified MED8A cells (Suppl. Figure 3D). Additionally there were significantly less cells without any γH2A.X focus in double and triple treatment conditions compared to DMSO control (Suppl. Figure 3D). In summary, we observed strong induction of DSBs upon triple treatment specifically in MYC-amplified MB cell lines with preferential occurrence during G1/S phases.

We assumed that accumulation of DNA damage, particularly during G1/S phases may lead to increased cell death. Indeed, we observed slightly increased sub-G1 fractions compared to single agent treatment in all cell lines after 72 h of double treatments. Importantly a significant increase of the sub-G1 fraction was detected exclusively in MYC-amplified cell lines upon triple combination treatment (Fig. 5C; Suppl. Figure 4A, B). Conversely, the MYC non-amplified cell line ONS76 showed a G2/M arrest upon treatment (Suppl. Figure 4A, B), which is in line with the preferential occurrence of DSBs during G2/M cell cycle phase. Finally, addition of entinostat (E) to a double treatment of olaparib and doxorubicin (OD) led to a significant decrease in metabolic activity compared to backbone treatment (OD) alone only in MYC-amplified MB cells (Fig. 5D). Notably also in non-malignant non-transformed juvenile fibroblasts the triple treatment (EOD) did not cause significantly lower metabolic activity in comparison to single agent treatment with olaparib and doxorubicin (Fig. 5D).