Aurora A Kinase Inhibition Destabilizes PAX3-FOXO1 and MYCN and Synergizes with Navitoclax to Induce Rhabdomyosarcoma Cell Death

Rhabdomyosarcoma (RMS) are the most common pediatric soft-tissue sarcoma. The most aggressive subtype, alveolar RMS, is characterized by the occurrence of balanced reciprocal translocations, resulting in expression of oncogenic fusion proteins (1, 2), whereby the most common fusion is PAX3-FOXO1. Fusion-positive RMS (FP-RMS) has a high propensity to metastasize, and resistance to standard-of-care treatments is common, resulting in 5-year survival rates of only about 30% (3–5). The search for novel targeting strategies is difficult as pediatric tumors generally harbor only few somatic mutations (6), which is especially true for the FP-RMS subtype (7). The lack of other somatic mutations underscores the important role of PAX3-FOXO1, which functions as transcriptional activator affecting multiple oncogenic pathways (8). We previously demonstrated that antisense-mediated loss of PAX3-FOXO1 results in cell death, underscoring the addiction of FP-RMS cells to the fusion protein. However, so far the exact mechanism by which the cells die has not been described (9).

Due to its importance for tumor cell survival, therapeutic targeting of PAX3-FOXO1 has become a paramount goal of RMS research. One approach has been to identify key downstream effectors of the fusion protein as potential therapeutic targets, such as MYCN (10). However, because MYCN and PAX3-FOXO1 are transcription factors lacking enzymatic activity, targeting these proteins directly is challenging. Recently, new strategies have been developed targeting PAX3-FOXO1 indirectly by inhibiting its stabilizer Polo-like kinase 1 (PLK1; ref. 11) or its cofactors, like bromodomain-containing protein 4 (BRD4; ref. 12) or chromodomain-helicase DNA-binding protein 4 (CHD4; ref. 13). Further strategies to target PAX3-FOXO1 and MYCN are reviewed in refs. 14, 15. Although single-target therapies can be efficient in preclinical models and clinical treatment, they are prone to develop therapy resistance. One way to circumvent this problem is to use drug combinations that target different pathways preventing escape.

Recently, we showed that PLK1 stabilizes PAX3-FOXO1 through phosphorylation of serines 503 and 505, protecting the fusion protein from proteasomal degradation (11). Although single-agent inhibition of PLK1 proved very effective in reducing tumor growth in vivo, resistance against this treatment was also observed. One of the regulators of PLK1 is Aurora A kinase (AURKA), which therefore might also functionally contribute to FP-RMS tumorigenesis. Aurora kinases are a family of serine-threonine kinases that are involved in cell-cycle progression, most importantly during mitosis (16). AURKA is highest expressed at the G₂–M transition (17) where it activates PLK1 by phosphorylation at threonine 210 as crucial step for checkpoint recovery (18, 19). Furthermore, AURKA serves important roles in centrosome maturation and mitotic entry (20), as well as for spindle assembly (21). These functions in cell-cycle progression indicate an important role for Aurora kinases in cancer. Indeed, AURKA is upregulated in a variety of tumors and has thus become a focus of inhibitor development (22–25). AURKA has also been shown to phosphorylate AKT and mTOR, indicating a role in promoting chemotherapy resistance (26).

Our goal was thus to find a novel synergistic combination therapy to treat FP-RMS. We demonstrate that FP-RMS cells undergo intrinsic apoptosis in a NOXA-dependent manner after loss of PAX3-FOXO1, which can be enhanced by the BH3-mimetic navitoclax (ABT-263). Moreover, we established a novel functional link between AURKA and stability of both PAX3-FOXO1 and MYCN. Combination of navitoclax and alisertib had synergistic antitumor effects in vitro and in vivo and thus provides the basis for a promising new rationale combination therapeutic approach for FP-RMS.

RMS cell lines RD, Rh4 (Peter Houghton, St. Jude Children's Hospital, Memphis, TN), RhJT (Scott Diede, Fred Hutchinson Cancer Research Center, Seattle, WA), RMS (Janet Shipley, Sarcoma Molecular Pathology, The Institute of Cancer Research, London, UK), KFR (Jindrich Cinatl, Frankfurter Stiftung für krebskranke Kinder, Frankfurt, Germany), Rh30 as well as HEK293T cells (both ATCC LGC Promochem) were cultured in high-glucose DMEM (Sigma-Aldrich), supplemented with 100 U/mL penicillin/streptomycin, 2 mmol/L l-glutamine, and 10% FBS (Life Technologies) at 37°C and 5% CO₂. All RMS cell lines were authenticated upon receipt by short tandem repeat (STR) profiling and used for experimentation from frozen stocks within 10 to 20 further passages. As the human myoblast cell line has not yet been characterized by STR profiling, negative matching with all available cell lines in the database was used for verification. All cells were regularly tested for mycoplasma contamination by a PCR-based assay.

Patient-derived xenograft (PDX) tumors were dissociated as described before (27). In brief, tumor tissue was minced with scalpels and suspended in Hanks' Balanced Salt Solution (Sigma-Aldrich) supplemented with 1 mmol/L MgCl₂, 200 μg/mL Liberase, and 200 U/mL DNase I (both Roche). Tissue was digested for 30 minutes at 37°C and filtered twice through 70 μm cell strainers (BD Biosciences). Dissociated cells were washed with PBS (Sigma-Aldrich) before freezing or resuspending for further culture.

Cells derived from PDX tumors were cultured in Neurobasal medium (Life Technologies) supplemented with 2x B-27 Supplement (Life Technologies), 20 ng/mL EGF, and 20 ng/mL basic FGF (Peprotech) on plates coated with Matrigel (Corning Life Sciences).

Cells were seeded in 384-well plates, and shRNA was induced by 100 ng/mL doxycycline (Sigma-Aldrich). Drugs (Supplementary Table S1) were purchased from Selleckchem. Twenty-four hours after seeding cells, medium was changed to 19 μL culture medium. Drugs were prediluted to 10 μmol/L in culture medium, and 1 μL of each drug was added to the wells for a final concentration of 500 nmol/L. Viability was measured after 48 hours by WST-1 assay.

Cells for immunoblot analysis were seeded in 24-well plates and drugs added at 100 nmol/L after 48 hours of incubation.

NOD/Scid il2rg^−/− (NSG) mice were 8 to 12 weeks old for the experiments. Note that 5 × 10⁶ Rh4 or IC-PPDX-35 cells were injected s.c. into the flanks. After engraftment mice were randomized into four groups (5–6 mice per group) when tumors reached 100 mm³. Tumor growth was assessed by caliper measurements, and the volumes were calculated using the formula V = (4/3) π r3; r = (d1+d2)/4. Mice were sacrificed when the tumor volume reached 1000 mm³. All animal experiments have been approved by the Swiss veterinary authorities (license ZH206/15).

Data analysis was performed with GraphPad Prism 7. Significance was calculated using unpaired two-tailed Student t test or Welch two-tailed test. Two-way ANOVA was used for multiple comparisons. Differences were considered statistically significant with P < 0.05. Drug synergy was calculated using the Bliss independence model in the free SynergyFinder WebApp (28).

For additional methods, see Supplementary Materials and Methods Section.

Because it was previously shown that silencing of PAX3-FOXO1 expression results in FP-RMS cell death (9), we aimed first to further elucidate its precise mechanism.

We generated FP-RMS cell lines expressing an inducible shRNA (shP3F) or control (shsc) to specifically silence PAX3-FOXO1 expression upon doxycycline treatment (Supplementary Fig. S1A). After confirming that the system significantly downregulated the fusion protein both at the mRNA and protein levels in four different cell lines (Supplementary Fig. S1B–S1D), we assessed whether cells would undergo apoptosis. We observed a significant increase in caspase-3/7 activity in Rh4 and Rh30 cells after PAX3-FOXO1 depletion compared with control (Fig. 1A; Supplementary Fig. S1E). This increase was also reflected by the appearance of cleaved products of PARP, caspases 3, 7, and 9 (Fig. 1B). To exclude off-target effects, we overexpressed a nontargetable PAX3-FOXO1 mutant, which indeed could rescue cells from apoptosis despite silencing of the endogenous fusion protein as shown by reduced caspase-3/7 activity (Supplementary Fig. S1F) and decreased levels of cleaved PARP and caspase-3 protein products (Supplementary Fig. S1G).

To further confirm this notion, we also treated cells after fusion protein depletion with increasing concentrations of the pan-caspase inhibitor zvad-FMK, which again could restore viability of shP3F-Rh4 cells (Fig. 1C) and reduced cleaved PARP and caspase-3 protein levels (Supplementary Fig. S1H). To exclude other modes of cell death, we also treated cells with the different cell death inhibitors zvad-FMK (apoptosis), necrostatin-1 (necroptosis), ferrostatin (ferroptosis), as well as E64d (cathepsin inhibitor) and cathepsin GI. However, only zvad-FMK could rescue viability but none of the other inhibitors (Fig. 1D). These data indicate that apoptosis is the major mode of cell death activated in FP-RMS cells upon depletion of PAX3-FOXO1.

Next, we sought to identify the proapoptotic protein(s) responsible for initiating apoptotic cell death. We performed a small-scale CRISPR/Cas9 screen in shP3F-Rh4 cells using the construct depicted in Supplementary Fig. S2A to knock out proapoptotic genes either individually or in combination, and measured cell viability upon PAX3-FOXO1 depletion. Knockdown efficiencies of Bax, Bak, Bad, and Bim as examples are shown in Supplementary Fig. S2B and S2C. In control cells (shsc-Rh4), only depletion of caspase-9, but not caspase-8, and the combination of Bax/Bak were able to significantly reduce caspase-3/7 activity (Supplementary Fig. S2D). This indicates that activation of the extrinsic pathway is less important in FP-RMS cells. Upon depletion of PAX3-FOXO1, NOXA was the only BH3-only protein capable to significantly reduce caspase-3/7 activity (Fig. 1E; Supplementary Fig. S2E), similar to knockout of the pore-forming proteins BAX and BAK. These results were validated in three additional FP-RMS cell lines (Supplementary Fig. S2F), albeit the extent of repression differed among cell lines. Nevertheless, NOXA seems to play an important role in initiating apoptosis after PAX3-FOXO1 depletion, although the contribution of other BH3-only proteins cannot be excluded. Furthermore, cell-cycle analysis after PAX3-FOXO1 silencing (shP3FDox) indicated no significant change in cell-cycle distribution, but appearance of a sub-G₁ peak in all cells except Rh30, indicative of induction of apoptosis (Supplementary Fig. S2G). This is in line with the observation that both NOXA mRNA and protein expression are upregulated upon silencing of the fusion protein in all cell lines tested with the exception of Rh30 cells (Fig. 1F and G; Supplementary Fig. S2H). Taken together, our results demonstrate that intrinsic apoptosis initiated upon silencing of PAX3-FOXO1 depends on upregulation of the BH3-only protein NOXA.

Next, we aimed to find drugs that would enhance apoptosis induced by PAX3-FOXO1 depletion. For this, we set up a compound screen of 208 drugs at a final concentration of 500 nmol/L and treated shP3F-Rh4 cells after doxycycline-mediated shRNA induction (Supplementary Fig. S3A; Supplementary Table S1). Results are depicted as ratio of viability comparing shP3F versus shsc cells. The screen identified 13 candidate drugs that decreased viability by at least an additional 50%, while not affecting control cells (Fig. 2A, left plot; raw data Supplementary Table S2). Classification of the top hits according to their mechanism of action revealed that three of them are BH3-mimetics (Fig. 2A, right plot) with navitoclax (ABT-263) being the most potent drug, also when validating each as single agent (Fig. 2B; Supplementary Fig. S3B–S3M). To directly compare BCL-XL versus BCL-2 targeting, we generated dose–response curves for navitoclax and venetoclax (ABT-199; Fig. 2C and D). Interestingly, navitoclax reduced IC₅₀ to at least a 10 times lower concentration (reduction of 3.0 μmol/L to 0.18 μmol/L) than venetoclax (reduction of 9.6 μmol/L to 5.6 μmol/L). These findings suggest that inhibition of BCL-2 might be less important. To confirm this notion, we treated cells with increasing concentrations of an additional BCL-XL–specific inhibitor, A1331852, and UMI-77, which is MCL-1 specific. Although A1331852 showed comparable effects to navitoclax (Fig. 2E), treatment with UMI-77 did not further increase cell death (Supplementary Fig. S3N). These results strengthen the observation that PAX3-FOXO1 silencing primes FP-RMS cells to apoptosis via balancing NOXA versus BCL-XL levels.

To demonstrate this more directly, we treated shP3F-Rh4-NOXA^−/− cells (Supplementary Fig. S2E) with navitoclax and found indeed reduced sensitivity upon silencing of PAX3-FOXO1 (Fig. 2F). Finally, expression analysis of different datasets in the r² database (R2: Genomics Analysis and Visualization Platform: http://r2.amc.nl) revealed that NOXA expression was elevated at base levels in all RMS datasets compared with healthy skeletal muscle tissue (Supplementary Fig. S3O).

These findings suggest that RMS cells might be already primed towards a proapoptotic state through higher basal NOXA expression, which can be further enhanced by reduction of fusion protein levels and renders them more sensitive towards navitoclax treatment.

Next, we aimed to pharmacologically reduce PAX3-FOXO1 protein levels. To this end, we treated wild-type (WT) Rh4 cells with the same drug library (Supplementary Table S1) and analyzed cell lysates by Western blot (Supplementary Fig. S4A). To exclude general toxic effects, all signals were densitometrically digitalized and normalized to the house keeping protein GAPDH. Initial hits were called when fusion protein levels were lowered below 80%, a criterion fulfilled by 43 compounds (Fig. 3A and B; raw data Supplementary Table S3). When classifying these hits according to their drug targets, we identified five epigenetic regulators, though HDAC inhibitors like entinostat, which was recently described to reduce fusion protein levels (29), showed only mild to no effects (Supplementary Fig. S4B). Furthermore, we classified five proteasome, five AURKA, and three CDK9 inhibitors (Fig. 3C). Strikingly, AURKA inhibitors were already identified in our apoptosis screen as hits (Fig. 2B), and previous work from our laboratory demonstrated a direct interaction between PAX3-FOXO1 and PLK1, which stabilizes the fusion protein (11) and is known to be activated by AURKA (19). Hence, we individually validated AURKA inhibitors at increasing doses and identified alisertib as being the compound active at the lowest doses tested (Fig. 3D; Supplementary Fig. S4C and S4D). Importantly, treatment of PDX-derived primary cells with alisertib also reduced fusion protein levels (Supplementary Fig. S4D). We were then interested to study the mechanism underlying the functional interaction of AURKA and PAX3-FOXO1. We treated Rh4 cells with alisertib, which increased total protein levels of both PLK1 and AURKA, whereas we observed a reduction of PLK1 phosphorylation at threonine 210, as expected (Fig. 3E). Because AURKA has been described to phosphorylate S256 in WT FOXO1 (30), we then investigated phosphorylation of S437 in PAX3-FOXO1 (corresponding to residue S256 in WT FOXO1). Indeed, we also observed a clear reduction at this site upon alisertib treatment (Fig. 3E). As this region of PAX3-FOXO1 has been described to be relevant for protein stability (reviewed in ref. 15), we next studied whether phosphorylation of S437 contributed to fusion protein stability by replacing serine with alanine. Interestingly, and actually even slightly more pronounced than the previously described S503A mutant, the S437A mutant also significantly reduced fusion protein stability as revealed after 8-hour treatment with cycloheximide (Fig. 3F and G).

Lastly, to investigate a potential direct interaction of AURKA and PAX3-FOXO1, we fused PAX3-FOXO1 to the bacterial biotin-ligase BirA (31) and expressed it ectopically in HEK293T cells. After pull down of biotinylated proteins using streptavidin coated beads, PLK1 was identified on Western blots in fusion protein samples but not in the BirA-only control, as shown before (Fig. 3H; ref. 32). Strikingly, also AURKA was identified in this assay as being selectively biotinylated, suggesting a novel direct interaction of this kinase and the fusion protein (Fig. 3H).

Taken together, our results indicate that AURKA inhibition can decrease fusion protein stability through reduced phosphorylation of serine 437, which might contribute to its ability to induce apoptosis in FP-RMS cells.

AURKA is reported to stabilize the MYCN protein that is highly expressed through transcriptional regulation by the fusion protein and genomic amplification (10, 33–35). Previous studies showed that AURKA stabilizes MYCN by interfering with SCF^FbxW7-mediated ubiquitination and that sequential phosphorylation of MYC proteins at S62 and T58 is required for binding of SCF^FbxW7 (36, 37). We therefore investigated whether AURKA inhibition affected MYCN protein levels in FP-RMS cells. We stably transduced three different FP-RMS cell lines with plasmids expressing shRNA against AURKA mRNA. In all cell lines, depletion of AURKA resulted in reduced MYCN protein levels (Fig. 4A). In accordance with published literature (38), MYCN double mutants (T58A/S62A) had more stable MYCN and were rescued from degradation after depletion of AURKA in RMS cells (Fig. 4B). This finding indicates that in FP-RMS cells, MYCN stability is controlled by AURKA through the same mechanisms described for other cell types and cancers (39). Lastly, we wanted to confirm the influence of pharmacologic AURKA inhibition on MYCN protein levels in FP-RMS cells. We treated different FP-RMS cell lines with increasing concentrations of the AURKA inhibitors alisertib, as well as CCT137690 (a kinase inhibitor), and CD532 (an amphosteric inhibitor with similarities to alisertib), respectively. In all cell lines, inhibition of AURKA resulted in loss of MYCN protein levels in a dose-dependent manner at submicromolar doses (Fig. 4C; Supplementary Fig. S5A and S5B).

Taken together, our data show that AURKA stabilizes MYCN in addition to its effects on PAX3-FOXO1, and its inhibition destabilizes MYCN, a target gene of the fusion protein.

Because it is unlikely that single agents will be able to provide significant clinical benefit, we next wanted to identify a combination of synergistically acting drugs. To do this in an unbiased way, we screened our library for compounds that would synergistically reduce cell viability in conjunction with a low concentration of navitoclax (IC₂₀; Supplementary Fig. S6A and S6B; raw data Supplementary Table S4). We assessed cell viability after 48 hours and ranked the results according to synergism. Strikingly, out of the 28 hits identified, 7 were AURKA inhibitors (Fig. 5A). Six of these were also identified in a similar screen carried out in a second FP-RMS cell line (Supplementary Fig. S6C and S6D). Hence, AURKA inhibitors might act in synergy with navitoclax. To test this, we treated FP-RMS cells with increasing concentrations of both alisertib and navitoclax to obtain a combination matrix. Indeed, the combination was able to induce cell death even at lower drug concentrations (Fig. 5B) and was highly synergistic as assessed by SynergyFinder (Fig. 5C; Supplementary Fig. S7A; ref. 28). Importantly, we observed comparable synergistic effects not only in cell lines but also in six PDX-derived primary cells. This synergy was tumor specific as nontumorigenic cells (human foreskin fibroblasts and myoblasts) were not sensitive toward the combination treatment despite expression of various levels of NOXA (Supplementary Fig. S7B). NOXA^−/− cells were also not sensitive to the combination treatment, demonstrating that NOXA expression was required (Supplementary Fig. S8A and S8B).

As AURKA plays an important role for cell-cycle progression during G₂–M phase, we assessed cell-cycle distribution after alisertib or combination treatment. Treating cells with 50 or 100 nmol/L alisertib, we observed an increasing proportion of cells arrested in G₂–M phase (Fig. 5D). Combination with 800 nmol/L navitoclax however strongly reduced the G₂–M peak and increased the sub-G₁ fraction (Fig. 5D; Supplementary Fig. S8C). This suggests that alisertib alone induces cell-cycle arrest whereas in combination with navitoclax pushes cells into apoptosis. These findings are also supported by a synergistic increase in caspase-3/7 activity (Supplementary Fig. S8D). Hence, alisertib acts synergistically with navitoclax in vitro to induce apoptosis in cell lines and primary PDX-derived RMS cells, whereas the drug combination had no major effects on nontumorigenic cells.

Having confirmed a synergistic action of alisertib and navitoclax in vitro, we next aimed to assess the antitumorigenic response of the combination in vivo. We injected NSG mice subcutaneously with either Rh4 or patient-derived IC-PPDX-35 cells. After tumors were palpable, mice were randomized into four groups and treated daily over 3 weeks with either vehicle, navitoclax alone (80 mg/kg), alisertib alone (30 mg/kg), or the combination of both drugs (Supplementary Fig. S9A). Although continuous tumor growth was observed in Rh4 cells in vehicle and navitoclax-only–treated mice, alisertib treatment slightly delayed tumor growth but failed to induce lasting effects (Fig. 6A). In contrast, combination therapy resulted in modest tumor regression and lasting stable disease even after end of the treatment period (Fig. 6A). This was also reflected in the survival of mice, because only animals in the combination group survived (Fig. 6B). In the PDX model, alisertib treatment alone provoked a stronger delay in tumor growth albeit also in this setting, combination treatment showed the most stable growth control (Fig. 6C) and survival was increased (Fig. 6D). In neither experiment, we observed significant weight loss (Supplementary Fig. S9B and S9C). To investigate whether alisertib treatment mimics PAX3-FOXO1 depletion, we also performed an in vivo experiment with the doxycycline-inducible shP3F line in combination with navitoclax. Although silencing of the fusion protein alone already leads to significant growth inhibition, addition of navitoclax treatment has a further, albeit small, combinatorial effect (Supplementary Fig. S9D). We also isolated IC-PPDX-35 tumors after 1 week of treatment, where histological analysis revealed a marked increase in the number of apoptotic cells in combination-treated tumors (Fig. 6E) and increased staining for cleaved caspase-3 (Fig. 6E). Consequently, proliferation was reduced as monitored by Ki67 staining and increased expression of p21 (Supplementary Fig. S9E and S9F).

To underscore the clinical relevance of AURKA inhibition, we finally noticed that AURKA RNA expression is significantly higher in three independent RMS datasets compared with healthy skeletal muscle (Fig. 6F; data from R2 database: http://r2.amc.nl). Further, we demonstrated AURKA protein expression in the majority of primary RMS. Twenty-six of 37 (70%) of FP-RMS and 37 of 44 (84%) of FN-RMS stained positively for AURKA by IHC (Supplementary Fig. S10Aand S10B). In addition, AURKA and the proliferation marker Ki-67 showed a significant correlation in FN-RMS and a trend toward significance in FP-RMS (Supplementary Fig. S10C and S10D). This was consistent with significant correlations between AURKA and MKI67 at the mRNA level (Supplementary Fig. S10E and S10F). No significant correlations between AURKA IHC scores and event-free or overall survival were seen in either the FP or FN groups.

Our aim for this study was to identify a novel synergistic combination treatment strategy based on cell death mechanisms identified in FP-RMS following silencing of PAX3-FOXO1. We showed that cells undergo intrinsic apoptosis in a NOXA-dependent manner upon reduction of PAX3-FOXO1 protein levels. In accordance, we identified BH3-mimetics, specifically navitoclax, to efficiently enhance this mode of cell death. Furthermore, we demonstrated a novel functional interaction between AURKA and the fusion protein and observed strong synergy in vitro and in vivo with the AURKA inhibitor alisertib and navitoclax.

We established a functional link between PAX3-FOXO1 and the BH3-only protein NOXA. However, we were unable to identify exactly how the fusion protein regulates NOXA expression. Previously, it was reported that PAX3-FOXO1 can upregulate basal Noxa levels in mouse myoblasts, which could prime tumor cells to apoptosis (40). Indeed, we also found that NOXA levels in tumor cells are higher than in normal myoblasts. However, in chromatin immunoprecipitation sequencing data, we could not detect any binding of PAX3-FOXO1 to the genomic PMAIP1 locus (41), excluding a direct role of the fusion protein in NOXA regulation. This is supported by the observation that NOXA levels even further increase when PAX3-FOXO1 levels diminish. Because the cell lines used in our experiments are p53 deficient (42), this also excludes a role of p53 in NOXA regulation as previously described (43). However, it is well known that NOXA expression can be induced by a variety of cellular stresses such as hypoxia, DNA damage, genotoxic, ER, or metabolic stress (44). If any of these, which one might be responsible for the observed phenotype remains to be investigated.

Nevertheless, the observation that NOXA upregulation was required for apoptosis of FP-RMS cells led to the identification of navitoclax as sensitizer. This is in line with the previous identification of the BCL-XL inhibitor A-1331852 as sensitizer to chemotherapeutics (45). In addition, BCL-XL has been described as target gene of the fusion protein (46), which also explains the lower sensitivity of FP-RMS cells to venetoclax, targeting BCL-2. However, single-agent activity is likely limited and therefore we searched for drugs capable to reduce fusion protein levels using conventional Western blotting. This led to the identification of multiple AURKA inhibitors that also reduced MYCN protein levels, a functionally very important PAX3-FOXO1 target gene (35).

Previously, it was reported that fusion protein levels could be reduced by entinostat, an HDAC inhibitor (29). Conversely, in our screen for drugs affecting fusion protein levels, epigenetic regulators, like entinostat or SAHA, do not show a striking effect. However, the drug concentrations used in our screen, as well as incubation times (0.1 μmol/L for 48 hours), are substantially lower than what was used by Abraham and colleagues (2 μmol/L for 72 hours; ref. 29). It is possible that these factors are the reasons for epigenetic regulators like entinostat not showing strong effects in our screen (Supplementary Fig. S3B).

Strikingly, we identified a new phosphorylation site in PAX3-FOXO1 (serine 437) that is important for fusion protein stability. This site corresponds to S256 in WT FOXO1, where it regulates nuclear exclusion and subsequent degradation (47), whereas phosphorylation of S437 in PAX3-FOXO1 stabilizes the fusion protein. These seemingly opposing effects on protein stability might suggest that protein turnover of WT versus the fusion protein is regulated by different mechanisms, as recently also suggested for the EWS-FLI1 fusion expressed in Ewing sarcoma (48). A similar observation has already been described for acetylation of K426 and K429 by the histone acetyltransferase KAT2B that also stabilize PAX3-FOXO1 (49). Acetylation of the corresponding sites in WT FOXO1 results in phosphorylation at S256 and subsequent degradation (27). Hence, we might speculate that acetylated K426/K429 might contribute to stability through priming of the fusion protein for S437 phosphorylation.

In the past, clinical trials involving navitoclax or alisertib have been challenging halting the process of clinical development. Navitoclax is known to increase the risk for thrombocytopenia due to its on-target effect (50). The advantage of our synergistic combination approach is that the individual drug doses can be lowered, thereby potentially reducing unwanted effects of each individual drug. Indeed, in our in vivo experiments, we did not observe complications when using both drugs in combination. Furthermore, recent interest in these drugs increased, and both companies are currently recruiting to new clinical trials (www.clinicaltrials.gov).

Taken together, we identified a previously undescribed site in PAX3-FOXO1 that is important for fusion protein stability. With AURKA inhibition, we found a therapeutic option to target this site in the fusion protein while simultaneously also affecting MYCN levels (Supplementary Fig. S11A and S11B). It is likely that both fusion protein and MYCN levels contribute to the inhibitory effects of alisertib because positive feedback mechanisms have been described regulating the expression of these two proteins (10, 12). Furthermore, by characterizing the exact mechanism of cell death associated with loss of fusion gene expression, we were able to identify drugs that enhance this mode of cell death. When used in such a rationale combination, both drugs show a high degree of synergy both in vitro and in vivo, in cell lines and an RMS PDX model. These findings shed more light on a devastating disease and may offer novel therapeutic options for the treatment of patients with alveolar RMS.

No potential conflicts of interest were disclosed.

Conception and design: J. Ommer, M. Wachtel, J. Shipley, B.W. Schäfer

Development of methodology: J. Ommer, J.L. Selfe, M. Wachtel, D. Surdez

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Ommer, J.L. Selfe, M. Wachtel, E.M. O'Brien, D. Laubscher, M. Roemmele, S. Kasper, O. Delattre, D. Surdez, G. Petts, J. Shipley

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Ommer, J.L. Selfe, M. Wachtel, D. Laubscher, O. Delattre, A. Kelsey, J. Shipley, B.W. Schäfer

Writing, review, and/or revision of the manuscript: J. Ommer, J.L. Selfe, M. Wachtel, A. Kelsey, J. Shipley, B.W. Schäfer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D. Surdez, G. Petts

Study supervision: J. Shipley, B.W. Schäfer

The authors wish to acknowledge the financial support from Swiss National Science Foundation (310030_156923 and 31003A_175558), Childhood Cancer Research Foundation Switzerland, and Cancer League Switzerland (KLS-3868-02-2016).

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.