A High-Throughput Screening Platform Identifies Novel Combination Treatments for Malignant Peripheral Nerve Sheath Tumors

Malignant peripheral nerve sheath tumors (MPNST) are aggressive neoplasms that account for 2% to 5% of soft-tissue sarcomas diagnosed annually (1). MPNSTs are encountered in three different clinical settings: associated with Neurofibromatosis type 1 (NF1; 40%–50%), sporadic cases (40%–47%), and associated with sites of previous radiation therapy (10%–13%; ref. 2). In patients with NF1, MPNSTs are the major malignant tumors associated with this autosomal dominant genetic disorder (3) and usually arise from a preexisting plexiform neurofibroma. The lifetime risk is 8% to 15% (4) and it is the leading cause of mortality in patients with NF1. These tumors have a poor prognosis, and multiple centers have reported that the 5-year disease-free survival rates of patients with MPNST are between 30% to 60% (2). Complete surgical excision with wide negative margins is the primary treatment option for local MPNST disease (3, 5). However, the success of surgical excision is limited by tumor infiltration, which results in a high relapse rate, and often is not feasible due to location, size, and metastases (5). Currently, there are no single-agent effective chemotherapeutic treatments for MPNSTs. Clinical trials of typical cytotoxic drugs have shown response rates ranging from 18% to 44%, with some stability as well (6). Doxorubicin and Ifosfamide have traditionally been used as the chemotherapy regimen for MPNSTs. However, a 10-year institutional review showed no correlation between chemotherapy and patient survival (7). The role of adjuvant radiation is not conclusive but is often recommended for high-grade lesions (8). To identify new therapeutic options for MPNST, the research directs towards the understanding of the underlying biology of these tumors and the generation of suitable preclinical models (2). NF1 inactivation is necessary but not sufficient for MPNST development, and mutations in other driver genes have been reported (TP53, CDKN2A, and SUZ12, and EED from Polycomb Repressive Complex 2; ref. 9, 10). The increasing knowledge of altered molecular pathways in MPNST suggested targeted therapies as potential treatments to block these dysregulated signaling pathways (2). Despite the advances, clinical trials, so far, have produced low response rates and have shown that the outcome for unresectable MPNST is poor (11). For this reason, there is a high medical need to identify therapeutic combinations that help overcome resistance and result in durable responses in patients with MPNST. Current technologies allow for rapid screening of multiple cancer cell lines against large libraries of compounds as simple agents (12, 13). These technological advances also allow for the screening of a large number of drug combinations in an unbiased manner, to identify synergistic compound combinations that promote tumor reduction and/or increased survival in vivo (14). Previous reports have used high-throughput screening methods to identify drug sensitivities in NF1 and Neurofibromatosis type 2 (NF2) tumors (15–17). In vivo models are critical to assessing the efficacy and therapeutic index of potential drug treatments suggested from in vitro screening. The current MPNST animal models available include xenograft models produced by the injection of MPNST cells, genetically engineered mouse models (GEMM), and patient-derived orthoxenografts (PDOX) in which a small portion of the patient's tumor is engrafted, without previous culturing near the sciatic nerve of athymic or NSG mice (11).

Here, we screened the National Center for Advancing Translational Sciences (NCATS) oncology-focused collection [Mechanism Interrogation PlatE (MIPE)] of 1,912 mechanistically annotated small molecules on three MPNST cell lines. The library includes approved and preclinical/clinical development drugs for cancer treatment. The aim is to test whether the compounds identified as active as single agents had a synergistic effect when added as pairwise combinations. The most synergistic combinations in a panel of MPNST cell lines were tested in our PDOX mouse models.

The establishment of the MPNST-SP-01, MPNST-NF1–08, and MPNST-NF1–09 cell lines at the Catalan Institute of Oncology is described in a manuscript under preparation. Supplementary Table S1 summarizes clinical and molecular data from the patients and tumors from which the three cell lines were isolated. MPNST-SP-01 was already presented in a previous work from our group (18). All molecular features of the primary tumors were confirmed in the derived cell lines. STS-26T (19), HS-PSS, HS-sch2 (20), S462 (21), and sNF96.2 (22) cell lines were previously established in other laboratories. The Human Foreskin Fibroblast (HFF) cell line was purchased from ATCC. Cells were maintained in DMEM growth medium with 10% FBS (Gibco) and 100 μg/mL penicillin and 100 μg/mL streptomycin (BioWest) under standard incubator conditions. All cell lines are validated as Mycoplasma negative before starting the in vitro validation and were retested every 2 months.

The MIPE 4.0 library of approved and investigational drugs included 1,912 individual small molecules (17). It encompasses a small-molecule modulator of over 400 specific gene targets, cellular pathways, or genotypes. Within well-explored targets, there are multiple, redundant agents incorporated as a means to inform the on-target nature of phenotype-to-mechanism data associations and to explore instances where a phenotype is the result of the specific polypharmacology of an individual drug.

The screening method used in this study in the MPNST-SP-01, S462, and sNF96.2 cell lines was similar to that previously described (12, 23). Briefly, compounds in the MIPE library were transferred to columns 5–48, and controls were added in columns 1–4 of a 1536-well assay plate. Column 1 contained media only, column 2 contained cells with added DMSO, while columns 3 and 4 contained the protease inhibitor Bortezomib or the antibiotic Salinomycin in DMSO (final concentration 10 μmol/L). Compounds were tested once as 11-point dose-response starting at a stock concentration of 10 mmol/L (the highest final compound concentration was 46 μmol/L) and diluted threefold with DMSO, as described previously (24). Relative luminescence units (RLU) for each well were normalized to the median RLUs from the DMSO control wells as 100% viability, and median RLUs from control wells with media only as 0% viability.

The activity of the compounds was determined based on their curve response class (CRC), in which normalized data is fitted to 4-parameter dose-response curves using a custom grid-based algorithm to generate a CRC score for each compound dose-response (25). CRC values of –1.1, –1.2, –2.1, are considered the highest quality hits; CRC values of –1.3, –1.4, –2.3, –2.2, –2.4, and –3 are considered inconclusive hits; and a CRC value of 4 is considered inactive compounds.

Given a selection of compounds, we used the annotated target for these compounds and computed the enrichment for each target class, compared with the background, using Fisher exact test (26). For this test, the background was defined as all the targets annotated in the MIPE collection. The P value from the test was adjusted for multiple hypothesis testing using the Benjamini-Hochberg method (27).

For the combination matrix screen with the selected compounds, protocols were as described before (12, 28). Briefly, compounds were pre-plated using an acoustic dispenser ATS-100 (EDC Biosystems, Fremont, CA). A total of 5 nL of each compound solubilized in DMSO, as well as DMSO control wells, were dispensed with threefold dilutions to generate 6 × 6 dose-response matrices, first, followed by confirmatory retesting using twofold dilutions in 10 × 10 dose-response matrices. 5 μL of a cell suspension (500 cells/well) were added directly to the plates immediately after compounds were pre-plated. Two sets of plates were screened. The first one was for cell proliferation assays, where cells were incubated with the compounds for 48 hours, and proliferation was quantified by adding CellTiter-Glo reagent. RLU was measured with the Viewlux (Perkin Elmer) after 15 minutes incubation at room temperature. The second plate was used for an apoptosis assay, incubating cells with the compounds for 16 hours and adding the Caspase-Glo 3/7 reagent to measure apoptosis.

The IC₅₀ of all compounds used in the 21 combinations selected was calculated for MPNST-SP-01 or S462 cells. First, 10,000 cells/well of MPNST-SP-01 cells or 8,000 cells/well of S462 were seeded in 96-well plates (Corning) and incubated overnight. Then, compounds in DMSO were added to the wells in three replicates. Two series of dilutions were performed at two concentration ranges depending on the potency of the compound. Stocks were prepared at 10 mmol/L, and compounds were 10-fold diluted from 1,000 nmol/L to 0.1 nmol/L or fivefold diluted from 40 μmol/L to 0.064 μmol/L, using DMSO. After 48 hours of incubation with the drugs, cell viability was analyzed using MTT (Sigma-Aldrich). Briefly, 14 μL of MTT (0.5 mg/mL) were added to each well. After 2 hours of incubation at 37°C, media was removed from the wells and 100 μL of glycine buffer solution (25 μL of buffer 0.1 mol/L NaCl and 0.1 mol/L Glycine with 75 μL of DMSO) was added. Plates were shaken for 15 minutes and the absorbance levels were read in the VictorTM X5 2030 Multilabel Reader (PerkinElmer). The IC₅₀ was calculated using GraphPad Prism Version 6 (La Jolla, CA). The percentage of cell viability was calculated by normalizing the values to DMSO control wells as 100% viability. Later, the four chosen synergistic combinations were tested in six additional MPNST cell lines, for which individual IC₅₀ of compounds was also calculated.

For combination assays, compounds were added to wells in three replicates as single treatments or in combination at a fixed 1:1 ratio of the IC₅₀ of both compounds, and threefold dilutions were performed from 10 × IC₅₀ concentration to 0.12 × IC_50, for each compound. After 48 hours of incubation, cell viability was measured by MTT, and combination index (CI) values for the combinations were calculated using CompuSyn Software, based on Chou-Talalay calculations (29). When CI is lower than 0.9 at high values of fraction affected of cells (Fa), we labeled the combination as synergistic. Fa value represents the fraction of cell death by drug treatment, ranging from 0 (no cell death) to 1 (100% cell killing). CI values between 0.9 and 1.1 are indicative of an additive effect and when values are higher than 1.1, the compounds are antagonistic (30).

MPNST-SP-01 and S462 cells were seeded (75,000 cells/mL and 60,000 cells/mL, respectively) in 6-well plates and incubated overnight. Cells were then treated with single compounds or combinations. Combinations with Panobinostat were tested at a fixed 1:1 ratio of the IC₅₀ and MK-1775 with Doxorubicin was combined at a 1:1 ratio of 3 × IC₅₀ to visualize the differences in apoptosis activation between the combination and individual treatments. After 24 hours of incubation, apoptosis activation was evaluated using the Annexin V APC Dead cell kit and SYTOX Green (ThermoScientific) to stain annexin-positive cells, according to the manufacturer's protocol. Samples were analyzed by FACSCanto II (BD Biosciences) flow cytometer. The apoptotic index was calculated as the mean percentage of apoptotic cells in 10⁴ cells measured in each of three independent experiments. Alternatively, apoptosis was determined by the percentage of cells in the sub-G₁ cell-cycle phase using flow cytometric analysis. After treatment, cells were fixed in ice-cold 70% ethanol and left for 24 hours at –20°C. Ethanol-fixed cells were centrifuged at 800 × g, washed, and resuspended in citrate-phosphate buffer (Na₂HPO₄ 0.2 mol/L and citric acid 0.1 mol/L at a ratio 192:8, pH 7.8) for 30 minutes. Cells were centrifuged at 400 × g and incubated in PBS supplemented with 1% FBS, 0.2 mg/mL RNAse (Sigma-Aldrich), and 60 μg/mL propidium iodide (Sigma-Aldrich) for 45 minutes at 37°C. 10,000 to 20,000 cells per sample of three experiments were measured to determine the percentage of cells at the sub-G₁ phase of the cell cycle. The percentage of cells in the G₁, S, and G₂–M phases was also calculated.

We adjusted the dosages and the administration schemes for the different compounds (Supplementary Table S2), using athymic nude 6-week-old male mice, in which no tumor was engrafted. Treatments lasted for 3 weeks and the mice body weight was daily measured to analyze potential toxicity. Carfilzomib and Doxorubicin were obtained from our hospital pharmacy. NVP-BGT226 (S2749), Torin 2 (S2817), and Panobinostat (S1030) were purchased from Selleckchem and MK-1775 (ref. 31; Adavorsetib, HY-10993) from MedChemExpress.

Before drug treatments, an early-passage (P2–P4) MPNST-SP-01 or MPNST-NF1–09 PDOX tumor had to be expanded. To do this, each tumor was implanted in five mice. When tumors reached 1,000 to 1,500 mm³, mice were sacrificed; tumors were harvested and cut into small fragments and engrafted near the sciatic nerve of 60 athymic nude male 6-week-old mice. Once tumors reached 300 to 500 mm³, mice were randomly distributed into treatment groups (n = 5–10/group): (i) single compound 1; (ii) single compound 2; (iii) combination compounds 1+2; and (iv) control receiving solvents of compounds 1+2. The final tumor weight plot from MPNST-NF1–09 PDOX model presented less than 5 points in some groups due to mice left to perform regrowth testing. Dosages and treatment schedules are specified in Supplementary Table S3. The study received IDIBELL IRB (#PR213/13) and IDIBELL Animal Ethics Experimentation Committee (CEEA-IDIBELL; #9111) approval. Combinations in which we did not observe tumor reduction by day 18 were further tested until day 25. Twenty-four hours after the last dose, mice were sacrificed and tumors extracted. Tumors were measured using a caliper twice a week, and volume was calculated using the formula “v = (w² L/2)”, where l is the longest diameter and w the width. Changes in tumor volume were quantified as the “increased tumor volume” since day 0, where each measurement is calculated following the formula “increased tumor volume = tumor volume day X – tumor volume day 0”. The interaction between follow-up time (in days) and treatment was used to assess the effect of each treatment in terms of volume change compared with the control group.

Protein was extracted from control tumors (n = 4) and tumors treated with the combinations (n = 6) using RIPA buffer supplemented with protease inhibitor (cOmplete Tablets, Roche). Briefly, 15 to 20 mg of tissue was homogenized in RIPA using TissueLyser II (Qiagen) and centrifuged at 16,000 × g at 4°C for 10 minutes. Proteins in the supernatant were quantified with the Pierce BCA Protein Assay Kit (ThermoScientific), using BSA as a standard.

For Western blot, equivalent amounts of protein (20 μg) were separated on 12% Acrylamide gels (TGX Stain-Free FastCastAcrylamide Kit, Bio-Rad). Before transference, gels were activated for 45 seconds by UV radiation using ChemiDoc Touch Imaging System (Bio-Rad), so that proteins could fluorescence when exposed to UV. Proteins were then transferred to nitrocellulose membranes from the Trans-Blot Turbo RTA Transfer Kit (Bio-Rad) using an established protocol from the Trans-Blot Turbo Transfer System (Bio-Rad). After transference, total protein from each lane was detected by exposing the membrane to UV. Then, membranes were blocked using 5% BSA (Roche) for 1h. Primary antibodies against Cdk1 (77055S), phospho-Cdk1^Tyr15 (9111T), Caspase-3 (9662S), Cleaved Caspase-3 (9661S), Histone H3 (9715S), Acetyl-Histone H3^Lys18 (9675S), SAPK/JNK (9252T), phospho-SAPK/JNK^{Thr183/Tyr185} (4668T), p70S6 Kinase (9202S), and phospho-p70^Thr389 S6 Kinase (9234S) were purchased from Cell Signaling Technology, diluted 1:1,000 in 5% BSA, and incubated overnight at 4°C. Membranes were probed with secondary antibody Stabilized Peroxidase Conjugated Goat Anti-Rabbit (Thermo Scientific), diluted 1:1,000 in 5% BSA, and incubated at room temperature for 1 hour. Finally, detection of proteins was conducted using SuperSignal West Femto Chemiluminescent substrate kits (Pierce Biotechnology Inc., Rockford, IL). Quantification analyses were carried out by the Image Lab program from Bio-Rad. Each protein of interest was normalized to the total protein of the lane to assess differences due to protein loading. Then, phosphorylated and acetylated proteins were also normalized to their levels of basal protein.

Statistical analyses were performed using Graphpad Prism 6. We calculated differences in terms of cell apoptosis activation between cells treated with combination of compounds and cells treated with single compounds or DMSO-control treated cells (n = 3 in each group), by using the unpaired t test with a significant level of 0.05. We also assessed differences in tumor volume and tumor weight between mice treated in vivo with combinations compared with single treatments and controls (n = 7–10 in each group) using the Mann–Whitney test with a significant level set at 0.05.

A full list of compounds from the MIPE 4.0 library is available upon request.

Data from the 113 10×10 combination screening matrices for the three MPNST cell lines, performed by the NIH-NCATS, are found in the following web links:

• MPNST-SP-01 Apoptosis 16 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6341/table

• MPNST-SP-01 Cell viability 48 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6342/table

• sNF96.2 Apoptosis 16 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6343/table

• sNF96.2 Cell viability 48 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6344/table

• S462 Apoptosis 16 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6345/table

• S462 Cell viability 48 h https://tripod.nih.gov/matrix-client/rest/matrix/blocks/6346/table

We screened the MIPE library using a quantitative high-throughput screen approach against three different MPNST cell lines, one sporadic established in our laboratory (MPNST-SP-01), and two NF1-related: S462 and sNF96.2. The assay used for the screen quantitated cellular ATP levels as a measure of cell viability/proliferation after 48 hours of treatment. From the dose-response screening, several hundred compounds were judged to be active based on CRC scores of –1.1, –1.2, and –2.1, for each cell line (480 active compounds at MPNST-SP-01 cell line, 507 at S462 and 410 at sNF96.2; refs. 24, 32). The CRC score is a measure of both potency (AC₅₀) and maximum response (MAXR), which is the percentage of activity at the maximum tested concentration (46 μmol/L). Differences between cell lines in the hierarchical clustering of the screening responses via MAXR (Fig. 1A) showed that the three MPNST cell lines had distinct pharmacologic vulnerabilities. Another parameter used to analyze the response of the drugs is the Area under the curve (AUC), which includes both the potency and efficacy of the compounds (33). When comparing the median AUC values of different compounds grouped by their mechanism of action (MOA) and target classes, while some mechanisms appear to have similar effects (e.g., tyrosine or serine/threonine kinase inhibitors), for other mechanisms (e.g., proteasome inhibitors or tubulin inhibitors) the response is quite different between cell lines (Fig. 1B). In general, the median AUC for each compound class is higher (less active) in the sporadic cell line compared with the NF1 cell lines. The redundancy in the number of compounds in each target class enabled us to perform target enrichment analysis for the active compounds in each cell line (Fig. 1C). There were only three target classes enriched (–log P value > 1) in all three cell lines (PIK3CA, CDK1, and PSMD1), and three targets enriched in two cell lines: TUBB for MPNST-SP-01 and S462; and histone deacetylase 1 (HDAC1) and MTOR for MPNST-SP-01 and sNF96.2. However, compounds targeting these enriched targets displayed different ranges of responses, as measured by AUC, among the three MPNST cell lines (Fig. 1D). We also performed a pharmacogenomic analysis and used exome sequencing data from the sporadic MPNST-SP-01 primary tumor, and found that the distance between somatic mutations and targets that were pharmacologically enriched was shorter than expected, including ERBB2, EGFR, PIK3CA, and PRKCA (Fig. 1E). The responses of compounds targeting these gene products were, in general, stronger for MPNST-SP-01 compared with the other two cell lines (Fig. 1F). On the basis of the target enrichment data, high-quality CRC, clinical relevance, and pathways of therapeutic interest, 25 compounds were selected to implement a pairwise all-versus-all matrix screen to explore synergistic activity in the MPNST-SP-01 and S462 cell lines and 50 compounds in sNF96.2 cell line.

The selected hits (75 among the three MPNST cell lines) were first screened in an all-versus-all, pairwise, 6 dose × 6 dose-response screening matrix. These experiments resulted in 300 discrete 6 × 6 matrices against MPNST-SP-01 and S462 cell lines, and 1,326 matrices against sNF96.2 cell line, using CellTiter-Glo ATP measurement as readout after 48 hours of treatment. Combinations that displayed synergy at selected concentrations, as quantitated by multiple metrics including the Bliss independence and the excess highest single agent (HSA) models, were advanced into 10 dose × 10 dose-response matrix experiments (12). For the 10×10 matrix screen, two sets of plates were generated, one for 48 hours treatment with CellTiter-Glo measurement, and the second for 16 hours treatment with Caspase-Glo to measure apoptotic cell death. A total of 113 10 × 10 dose-response pairwise combination matrices were generated and screened with all three MPNST cell lines (data in Materials and Methods). Hierarchical clustering analysis using the Delta Bliss Sum Negative (DBsumNeg) values for each matrix, with both the cell viability (Fig. 2A, left) and apoptosis (Fig. 2A, right) readouts, shows that synergies are stronger for the MPNST-SP-01 cell line at 48h, while they are weaker and similar between the S462 and sNF96.2 cell lines.

When selecting combinations with DBSumNeg values lower than –7, we observed different patterns of synergism among the three MPNST cell lines (Fig. 2B). For example, from the eighteen combinations that met this criterion for the MPNST-SP-01 cell line, seven included Panobinostat (HDAC inhibitor) combined with compounds targeting many different targets like Torin 2, NVP-BGT226, Carfilzomib, Fluvastatin, Lenvatinib, Obatoclax, and Doxorubicin. Interestingly, Panobinostat is not the compound driving most of the synergies for the other two MPNST cell lines. Moreover, five combinations including Doxorubicin (DNA damaging agent), and four including NVP-BGT226 (mTOR inhibitor) were observed for this cell line. On the other hand, from the eleven combinations that met the criterion for S462, four included Doxorubicin with NVP-BGT226, Englerin A, Neratinib, and Torin 2; and from 16 combinations for sNF96.2, four combined Sirolimus (mTOR inhibitor) with AT-9283, Cabozantinib, Crizotinib, and CUDC-101 (Supplementary Table S4).

The selection of synergistic combinations for further validation was focused on those combinations showing synergism in at least two of the three MPNST cell lines (MPNST-SP-01, S462, and sNF96.2), based on the DBSumNeg values. Also, visual inspection of the 10 × 10 matrices was taken into account when the DBSumNeg value was not a clear indication of synergism (Supplementary Fig. S1).

Twenty-one combinations were further tested in vitro to obtain a CI for each treatment according to dose-response curves. Synergy was observed in 13 of the 21 combinations, where CI was lower than 0.9 at Fa values higher than 0.8 (ref. 29; Table 1; Supplementary Fig. S2). From these combinations, the most synergistic ones were Panobinostat combined with Torin 2, NVP-BGT226, and Carfilzomib. We also selected the combination of MK-1775 with Doxorubicin because, even though the treatment did not kill cells beyond Fa = 0.85, the response of the combined compounds in terms of cell viability was notably increased compared with the individual drugs. These four combinations exhibited low DBSumNeg (Table 1) and excess HSA values (Fig. 3A) and synergistically decreased cell viability in the MPNST-SP-01, S462, and sNF96.2 cell lines compared with individual treatments (Fig. 3B and C). We then tested the four combinations in a broad panel of MPNST cell lines: three sporadic (STS26T, HS-PSS, and HS-sch2) and two NF1-derived cell lines (MPNST-NF1–08 and MPNST-NF1–09). We observed a synergistic effect in almost all the MPNST cell lines, independent of their NF1 status (Supplementary Fig. S3). In addition, the combinations were also tested in a fibroblastic cell line, HFFs, to evaluate general cell toxicity. Synergism was observed in all the combinations containing Panobinostat, but at higher concentrations than in the MPNST cell lines, indicating that tumor cells are more sensitive to the treatment, but also pointing to potential cytotoxicity. On the contrary, MK-1775 with Doxorubicin did not show synergism in the HFF cell line (Supplementary Fig. S3).

To gain further insight into the nature of these four compound combinations, we investigated their synergistic effect on the activation of apoptosis by measuring in vitro the percentage of annexin-positive cells and the fraction of cells at the sub-diploid (sub-G₁) phase, both as apoptosis readouts. We observed that Panobinostat combinations presented a significantly higher percentage of annexin-positive cells than individual treatments, confirming a potent apoptotic response in both cell lines. The combination MK-1775 with Doxorubicin showed an increased apoptotic response after treatment at high doses of the combined compounds in both cell lines (Fig. 4A). We observed that the combined treatment and the single compounds caused a greater apoptotic response in the S462 cell line, proving that this cell line is more sensitive to the treatments (Fig. 4A). The fraction of cells in the sub-G₁ phase in the four combinations treatment was also significantly higher, confirming the synergy at the apoptotic level (Fig. 4B). The apoptosis results obtained from annexin staining or phase sub-G₁ analyses presented a good correlation in both cell lines (except for MPNST-SP-01 treated with MK-1775 plus Doxorubicin). Moreover, the percentage of cells in the G₁/S/G₂ phases of the cell cycle was calculated to evaluate changes in the cell-cycle progression due to the treatments (Fig. 4C). The most notable change was observed when using MK-1775 (Wee1 kinase inhibitor) as a single treatment, due to its effect on the cell cycle via G₂–M checkpoint inactivation, as we observed a decreasing proportion of cells in the G₁ phase, which results in S-phase accumulation (31). Panobinostat also promotes cell-cycle arrest by blocking cells in the G₂–M phase before cell death, thus observing a lower percentage of G₁ cells and increased levels of G₂ cells in the three combinations using this compound (34).

We further tested the four combinations in vivo in patient-derived orthotopic xenograft (orthoxenograft/PDOX) mouse models from our preclinical platform. Only MK-1775 with Doxorubicin reduced the tumor growth compared with control and individual treatments in two different PDOX-MPNST models: one sporadic PDOX (MPNST-SP-01) and one NF1-related PDOX model (MPNST-NF1–09). The treatment caused the final tumor volume in mice to be seven times lower than the vehicle in the sporadic model after eighteen days of treatment, and almost four times lower in the NF1-PDOX model (Fig. 5A and B). This reduction was validated through tumor weight measures once they were extracted from mice at the end of the treatment. In the sporadic model, tumor weight was almost two times lower than control, and three times lower in the case of the NF1-related PDOX-MPNST. Moreover, treatment minimally reduced mice's body weight: approximately 10% in the NF1-related PDOX and no detectable weight loss in the sporadic model (Fig. 5A and B). In contrast, the three combinations selected with Panobinostat neither reduced the tumor volume nor the final tumor weight in the sporadic PDOX model (MPNST-SP-01) after 25 days of treatment (Supplementary Fig. S4A and S4B) at the tested doses, which were not toxic to the animals (Supplementary Fig. S4C). Higher doses of the compounds were tested, but toxicity was observed (Supplementary Table S2).

We also performed western blot to evaluate the expression of key molecular targets of the compounds in control tumors and tumors treated with the combinations. We observed a reduction of phosphorylation of Cdk1 (P-Cdk1) in some tumors treated with the combination MK-1775 with Doxorubicin due to the inactivation of Wee1 kinase activity by MK-1775 (31). Moreover, this combination caused higher levels of cleaved Caspase-3, which suggests activation of the apoptosis pathway due to DNA damage caused by Doxorubicin (Fig. 5C; ref. 35).

Some tumors treated using Panobinostat with Torin 2 and Panobinostat with NVP-BGT226 presented higher levels of acetylation of Histone 3 than controls, due to Panobinostat working as a HDAC inhibitor (36). Also, phosphorylation of p70S6 kinase was reduced in treated tumors, as both Torin 2 and NVP-BGT226 inhibit mTOR kinase activity (37, 38). On the other hand, tumors treated with Panobinostat and Carfilzomib did not show an increase in Histone 3 acetylation or an increase in the phosphorylation of JNK proteins, as would be expected due to reactive oxygen species being produced by the proteasome disruption by Carfilzomib (Supplementary Fig. S4D; ref. 39). In general, results from Western blot were weak and not statistically significant, likely due to the compounds being washed out in the 24 hours between the last dose and tumor extraction.

Our current study shows the results of a large-scale screening of 1,912 oncology-focused molecules on three MPNST cell lines (S462, sNF96.2, and MPNST-SP-01). This HTS approach has been successfully applied to find synergistic drug combinations for different cancers and has suggested new therapeutic options that worked in vivo (14, 28).

We found that the pharmacologic responses among the MPNST cell lines differed and that synergistic combinations in the sporadic cell line MPNST-SP-01 did not fully correlate with those in the NF1 lines, S462 and sNF96.2, probably due to the complex genomic landscape and genetic variability between different MPNSTs (40). Although most of the enriched targets identified have been previously suggested as possible therapeutic targets for MPNSTs (41–43), clinical trials using single agents have only so far shown response rates in the 18% to 44% range (6). It has been argued that cellular and molecular MPNST heterogeneity could limit the efficacy of single anticancer drugs in eliminating all the malignant cells in a tumor (44). There is much evidence that combination therapy could be superior to monotherapy because of the ability to target multiple pathways, minimizing drug resistance as cancer cells are frequently incapable of adapting to the simultaneous toxic effects. Moreover, if two compounds present synergism, the dose required for efficacy could be reduced (44). We, therefore, looked for synergistic combinations effective against MPNST cell lines. As a result of the single-agent MIPE library screening, some compounds were selected based on potency, efficacy, cellular target, and clinical relevance, and assayed in pairwise combinations to identify potential potent synergies. From the 21 in vitro selected combinations, four were identified as highly synergistic in a panel of MPNST cell lines: MK-1775 with Doxorubicin, Panobinostat with NVP-BGT226, Panobinostat with Torin 2, and Panobinostat with Carfilzomib. The combinations containing Panobinostat (a histone deacetylase inhibitor, HDACi) showed a potent apoptotic response in tumor cells but also exhibited cytotoxicity. Dual PI3K-mTOR inhibitors combined with HDACi have been effective in cell lines and mice models of medulloblastoma, non–small cell lung cancer, and breast cancer, among others (45–47). Moreover, there is also preclinical evidence of the effectiveness of mTOR and HDAC inhibitors for MPNSTs (48), in which the combined inhibition impinges on the thioredoxin pathway, triggering catastrophic oxidative stress and cell death, thus achieving tumor regression in vivo. In our study, none of the two combinations using HDAC and PI3K-mTOR inhibitors (Panobinostat with NVP-BGT226 or Torin 2) resulted in a reduction of tumor volume in our sporadic PDOX-MPNST model (MPNST-SP-01). We also tested Panobinostat in vivo with Carfilzomib (a proteasome inhibitor). This combination is in Phase I/II clinical trials (NCT01496118 and NCT01301807) for patients with relapsed and refractory multiple myeloma (49, 50). Also, it has been validated as synergistic in osteosarcoma cell lines in vitro (51). The combination triggers cell death through the accumulation of unfolded proteins that cannot be degraded by the proteosome or aggresome, in which HDAC-6 plays an important role (52). However, we did not observe the effect of this combination on tumor volume in our MPNST-SP-01 PDOX mouse model.

The observed differences in our in vivo experiments compared with the data in the literature could be due to the polypharmacology of the compounds, as well as to differences between tumor types. Another factor is the drug doses used in the different studies. We initially treated PDOX model MPNST-NF1–09 with 10 mg/kg of Panobinostat daily, as described previously (48), after assessing that combinations with Panobinostat at this dose were not toxic in non-engrafted nude mice (Supplementary Table S2). Contrary to expected, engrafted mice did not tolerate that dosage of Panobinostat. We then reduced the dose of Panobinostat to 5 mg/kg for testing in the MPNST-SP-01 PDOX model, which was well tolerated by the mice but did not produce tumor reduction. This difference in the tolerance may be due to the mouse strain; Malone et al used GEMMs on a C57BL6 background (48), which may tolerate higher doses than our nude mouse strain. Carfilzomib was also used at higher doses in other tumor types in C57BL6 mice (3–5 mg/kg), but such concentrations were lethal in our mouse model (53). We concluded that combinations that induce high levels of apoptosis in vitro, such as the three tested with Panobinostat, may not be good therapeutic approaches for MPNSTs, as toxicity may not allow the administration of the doses needed to effectively inhibit HDAC in nude mice PDOX models, therefore not achieving tumor reduction. It could be interesting to test these combinations in another mouse strain with higher drug tolerance.

The fourth combination validated in our panel of MPNST cell lines is MK-1775, a Wee1 inhibitor, with Doxorubicin, a DNA-damaging agent. Further analyses demonstrated that this combination is not as apoptotic as the combinations using Panobinostat. The combination of Wee1 inhibitors with DNA damaging agents has proved effective in vitro and in vivo for some other cancers. For instance, MK-1775 and Gemcitabine had synergy in vitro against MPNST (among other sarcomas cell lines) in other works (54) and reduced tumor volume in PDOX mouse models of pancreatic cancer, and MK-1775 with Cisplatin reduced tumor growth in gastric cancer (55, 56). This therapeutic approach has been described to be more effective in TP53 mutated tumors, as the G₂–M checkpoint (which is disrupted by MK-1775) is the only one that prevents tumor cells from programmed death due to DNA damage caused by the other compound (31). In vitro, we observed that MK-1775 with Doxorubicin presented higher drug activity in the MPNST cell lines with TP53 mutated (MPNST-SP-01, S462, MPNST-NF1–09, HS-sch2, and STS26T). Synergy was not observed in the wild-type TP53 cell lines MPNST-NF1–08 and HS-PSS and sNF96.2 presented less synergy than the TP53 mutated cell lines. In vivo, MK-1775 with Doxorubicin significantly reduced tumor growth compared with single treatments in two PDOX-MPNST models, one sporadic (MPNST-SP-01) and one NF1-related (MPNST-NF1–09), both TP53 mutated. The effect in both primary tumors matched the effects seen in their isolated cell lines where synergism was validated. Specifically, mice treated with the combination showed a significant tumor growth reduction in the MPNST-SP-01 and MPNST-NF1–09 PDOX models compared with the control group. It has been described that NF1-derived MPNSTs may have a lower response rate to chemotherapy than sporadic tumors, which agrees with our results (6). Overall, this less apoptotic combination proved to be a better therapeutic approach for MPNSTs, as toxicity effects were not observed in mice at the doses used to achieve tumor growth reduction.

In conclusion, the data presented here clearly show the potential of our screening platform for the discovery of potential combination therapies and presented results suggesting the combination of MK-1775 with Doxorubicin as a potential therapy for MPNST treatment, especially for those with TP53 mutated, although further studies on sequencing and dosing preclinically may yield more information that can assist in moving these combinations forward into the clinic.

J. Fernández-Rodríguez reports grants from Children's Boston Hospital, Carlos III National Health Institute, Government of Catalonia, Fundación Proyecto Neurofibromatosis; and grants from Fundació La Marató de TV3 during the conduct of the study. E. Creus-Bachiller reports grants from Children's Boston Hospital, Carlos III National Health Institute, Government of Catalonia, grants from Fundación Proyecto Neurofibromatosis (FPNF); and grants from Fundació La Marató de TV3 during the conduct of the study. K.M. Reilly reports grants from Neurofibromatosis Therapeutics Acceleration Program during the conduct of the study. J.O. Blakeley reports personal fees from TriAct; other support from SpringWorks Therapeutics, AstraZenica; and nonfinancial support from Bristol-Myers Squibb outside the submitted work. No disclosures were reported by the other authors.

J. Fernández-Rodríguez: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. E. Creus-Bachiller: Formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. X. Zhang: Investigation, writing–review and editing. M. Martínez-Iniesta: Investigation, writing–review and editing. S. Ortega-Bertran: Investigation, writing–review and editing. R. Guha: Investigation, writing–review and editing. C.J. Thomas: Methodology, writing–review and editing. M.R. Wallace: Methodology, writing–review and editing. C. Romagosa: Funding acquisition, writing–review and editing. L. Salazar-Huayna: Investigation, writing–review and editing. K.M. Reilly: Methodology, writing–review and editing. J.O. Blakeley: Investigation, writing–review and editing. J. Serra-Musach: Investigation, writing–review and editing. M.A. Pujana: Methodology, writing–review and editing. E. Serra: Methodology, writing–review and editing. A. Villanueva: Conceptualization, supervision, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. M. Ferrer: Conceptualization, supervision, funding acquisition, visualization, methodology, writing–original draft, writing–review and editing. C. Lázaro: Conceptualization, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.

Work developed by J. Fernández-Rodríguez, E. Creus-Bachiller, M. Martínez-Iniesta, S. Ortega-Beltran, A.Villanueva, and C. Lázaro was supported by NFRI grant from Boston Children's Hospital and the Carlos III National Health Institute funded by FEDER funds – a way to build Europe – [PI19/00553, PI16/00563; PI16/01898; PI19/01320 and CIBERONC]; the Government of Catalonia [Pla estratègic de recerca i innovació en salut (2017SGR1282 and 2017SGR496)]; the Fundación Proyecto Neurofibromatosis (FPNF), and Fundació La Marató de TV3. Work developed by X. Zhang, R. Guha, C.J. Thomas, and M.Ferrer at NCATS was funded by the NIH Intramural Research Program and NTAP. L. Salazar-Huayna and C. Romagosa are supported by a grant from the “Programa d'impuls del talent i de l'ocupabilitat del PERIS 2016–2020”.The authors thank CERCA Program/Generalitat de Catalunya for their institutional support. We thank all the patients and families with MPNST tumors, as well as all Spanish NF patients and NF associations for their continuing support, in particular the Spanish Asociación de Afectados de Neurofibromatosis (AANF) and the Associació Catalana de les Neurofibromatosis (ACNefi). We also wish to thank the ICO and IGTP Hereditary Cancer Program teams, as well as the members of the Spanish CSUR of Phakomatoses. We also want to thank Centres Científics i Tecnològics de la UB (CCiTUB) for assistance in the apoptosis and cell-cycle analyses.