GATOR2-dependent mTORC1 activity is a therapeutic vulnerability in FOXO1 fusion–positive rhabdomyosarcoma

Oncogenic FOXO1 gene fusions drive a subset of rhabdomyosarcoma (RMS) with poor survival; to date, these cancer drivers are therapeutically intractable. To identify new therapies for this disease, we undertook an isogenic CRISPR-interference screen to define PAX3-FOXO1–specific genetic dependencies and identified genes in the GATOR2 complex. GATOR2 loss in RMS abrogated aa-induced lysosomal localization of mTORC1 and consequent downstream signaling, slowing G1-S cell cycle transition. In vivo suppression of GATOR2 impaired the growth of tumor xenografts and favored the outgrowth of cells lacking PAX3-FOXO1. Loss of a subset of GATOR2 members can be compensated by direct genetic activation of mTORC1. RAS mutations are also sufficient to decouple mTORC1 activation from GATOR2, and indeed, fusion-negative RMS harboring such mutations exhibit aa-independent mTORC1 activity. A bisteric, mTORC1-selective small molecule induced tumor regressions in fusion-positive patient-derived tumor xenografts. These findings highlight a vulnerability in FOXO1 fusion–positive RMS and provide rationale for the clinical evaluation of bisteric mTORC1 inhibitors, currently in phase I testing, to treat this disease. Isogenic genetic screens can, thus, identify potentially exploitable vulnerabilities in fusion-driven pediatric cancers that otherwise remain mostly undruggable.


Introduction
Rhabdomyosarcoma (RMS) is the most common soft tissue sarcoma of childhood and is comprises 2 genetically defined subtypes: FOXO1 fusion positive (FP) and fusion negative (FN), which are characterized by activating mutations in receptor tyrosine kinase (RTK)/RAS pathways (1,2). Despite genomic, histologic, and clinical factors that distinguish FP from FN disease, patients with either diagnosis are treated with the same backbone of chemoradiotherapy. Only 60% of patients are cured, and relapse is frequently lethal. Patients with FOXO1 FP disease have worse overall and relapse-free survival than do patients with FN disease (3).
Recurrent translocations between chromosomes 2 and 13 in alveolar RMS were mapped to rearrangements of PAX3 over 20 years ago (4). The expressed product of the PAX3-FOXO1 translocation retains the N-terminal DNA binding elements of PAX3 and the C-terminal transactivation domain of FOXO1. A related but less frequent t(1;13) rearrangement results in the PAX7-FOXO1 fusion, associated with similar biological and clinical outcomes as PAX3-FOXO1 (5). PAX3-FOXO1 binds at super-enhancers distal from target genes and enables active chromatin states through recruitment of histone acetylation machinery (1). Targeting FP RMS through either BET/bromodomain inhibitors or targeted degradation of the fusion protein are new and promising directions for therapy. To date, however, these approaches have not shown conclusive benefit in patients, and thus ongoing efforts to identify or refine therapies for FOXO1 FP RMS remain vital.
The mTOR kinase controls the growth of normal and cancerous cells. As part of the mTORC1 complex, mTOR phosphorylates p70S6K to promote ribosome assembly, and 4EBP1 to permit m 7 -GTP Oncogenic FOXO1 gene fusions drive a subset of rhabdomyosarcoma (RMS) with poor survival; to date, these cancer drivers are therapeutically intractable. To identify new therapies for this disease, we undertook an isogenic CRISPR-interference screen to define PAX3-FOXO1-specific genetic dependencies and identified genes in the GATOR2 complex. GATOR2 loss in RMS abrogated aainduced lysosomal localization of mTORC1 and consequent downstream signaling, slowing G1-S cell cycle transition. In vivo suppression of GATOR2 impaired the growth of tumor xenografts and favored the outgrowth of cells lacking PAX3-FOXO1. Loss of a subset of GATOR2 members can be compensated by direct genetic activation of mTORC1. RAS mutations are also sufficient to decouple mTORC1 activation from GATOR2, and indeed, fusion-negative RMS harboring such mutations exhibit aa-independent mTORC1 activity. A bisteric, mTORC1-selective small molecule induced tumor regressions in fusion-positive patient-derived tumor xenografts. These findings highlight a vulnerability in FOXO1 fusion-positive RMS and provide rationale for the clinical evaluation of bisteric mTORC1 inhibitors, currently in phase I testing, to treat this disease. Isogenic genetic screens can, thus, identify potentially exploitable vulnerabilities in fusion-driven pediatric cancers that otherwise remain mostly undruggable.
JCI Insight 2022;7(23):e162207 https://doi.org/10.1172/jci.insight.162207 cap-dependent mRNA translation. The activity of mTORC1 is controlled both by growth-factor signaling that inhibits the TSC1/TSC2 complex, and aa sufficiency signals. In response to aa, including leucine and arginine, the pentameric GATOR2 complex inhibits the GATOR1 complex, which, in turn, prevents mTORC1 localization to its site of activity at the lysosome by accelerating GTP hydrolysis by RAGA and RAGB (6). Thus, GATOR2 is an aa-responsive, positive regulator of mTORC1.
The allosteric mTOR inhibitor temsirolimus delays disease progression but does not improve the survival of patients with relapsed RMS (7). These allosteric inhibitors inhibit phosphorylation of p70S6K, but not 4EBP1, which limits their efficacy in certain cancer models (8,9). Active-site kinase inhibitors suppress phosphorylation of both p70S6K and 4EBP1 but also inhibit mTORC2, resulting in dose-limiting hyperglycemia (10,11) and removal of AKT-dependent feedback inhibition of RTK signaling (12). Newer bisteric inhibitors with combined active-site and allosteric binding suppress p70S6K and 4EBP1 phosphorylation, have been optimized for mTORC1 selectivity, and might improve upon the modest clinical success of rapalogs in RMS (13). However, a cellular mechanistic understanding of how RMS-initiating genetic events (i.e., FOXO1 fusions or mutations in the RAS pathway) create mTOR dependence is lacking, which limits the precise use and further development of these agents in RMS.
We reasoned that genetic screens could uncover targetable dependencies downstream of PAX3-FOXO1. Prior efforts to do so have used shRNA libraries and compared FP and FN cell lines (14). However, because FN RMSs harbor genetic changes besides lack of PAX3-FOXO1, such as large-scale copy number alterations and RAS mutations (15,16), an isogenic system might provide a better means to identify PAX3-FOXO1 dependencies. By engineering and deploying such a system coupled with a more precise CRISPR/ Cas9 genetic screening platform, we report here that PAX3-FOXO1 FP RMS depends on GATOR2 for cellular proliferation, and we identify a clinically translatable therapeutic approach to target this potential vulnerability in patients with RMS.

Results
A CRISPR-interference screen identifies genes required for the growth of FP RMS cells. We developed a system to identify genes that are essential to the growth and survival of PAX3-FOXO1-positive cells, given the poor outcomes for FP RMS. We transduced FP Rh30 cells with either a lentiviral shRNA construct targeting the PAX3-FOXO1 fusion boundary or a nontargeting control shRNA. The former pool was clonally derived through limiting dilution to ensure the purity of cells harboring the PAX3-FOXO1 shRNA. The resultant cells (herein referred to as P3F KD ) or Rh30 cells carrying a nontargeting control shRNA (referred to here as P3F + ) were then transduced with the dCas9-KRAB transcriptional repressor. We confirmed P3F KD cells had decreased levels of the fusion protein, as well as downstream transcriptional targets of PAX3-FOXO1 such as FGFR4, when compared with P3F + controls ( Figure 1A; see complete unedited blots in the supplemental material). P3F KD cells grow in 2D culture but with a doubling time almost twice that of P3F + cells. Furthermore, P3F KD cells showed near-complete loss of colony formation in soft agar assays, demonstrating a loss of oncogenic potential ( Figure 1B).
We then transduced P3F + and P3F KD cells with a pooled library of approximately 20,000 lentiviral sgRNA constructs targeting 2944 genes involved in protein homeostasis, on the basis of our prior work identifying targets in this cellular network as therapeutic opportunities for RMS (17). After growing cells for 10 population doublings, we collected and deep sequenced genomic DNA to identify the abundance of each sgRNA, and we calculated gene-level scores for enrichment or depletion in the P3F + and P3F KD conditions ( Figure 1C and Supplemental Table 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.162207DS1) (18). This analysis was corrected for the observed number of population doublings in each condition, thus accounting for the slower growth of P3F KD cells. We focused on genes whose knockdown depleted P3F + cells but was tolerated in P3F KD cells, because these should be selectively required in PAX3-FOXO1-positive RMS ( Figure 1D).
Of the 2944 genes in the screening library, 36 met the following criteria for future study: fold-depletion of greater than 2 -2 in P3F + cells with an FDR-adjusted Mann-Whitney q value of less than 0.20, and a "neutral" fold-change between 2 2 and 2 -2 in P3F KD cells ( Figure 1E and Supplemental Figure 1). Comparing these results with those of the DepMap project (http://www.depmap.org) (19), more than half of our screen hits (n = 20 of 36) were reported as "common essential"; 5 (USP14, RAB35, PRDX1, FBXO28, and KCTD10) of the 36 screen hits were reported as "strongly selective" but did not meet criteria for lineage enrichment in RMS. We next asked whether our screen hits showed differential effects when knocked down in 6 FP vs. 5 FN RMS cell lines included in the DepMap. Only 2 of the 36 hits (UCHL5 and EIF2B5) had significantly more potent effects in FP RMS, by a Mann-Whitney test. Although possibly reflective of dependencies unique to the Rh30 cell line, this comparison spurred us to test whether the isogenic PAX3-FOXO1 screen identified fusion-specific dependencies that did not emerge from a larger, less focused effort.
We next prioritized genes for further validation by using STRING analysis (20) to identify structurally interacting gene products among screen hits ( Figure 1E). Four clusters of interacting genes were identified, including a 6-gene cluster encompassing ER to Golgi transport, 2 genes involved in ubiquitin processing at the proteasome, 2 genes involved in pre-mRNA splicing, and 2 members of the GATOR2 complex, MIOS  (20). Outlines correspond to selectivity calls from the DepMap database (19). Coloring indicates the P value, by Mann-Whitney test, for stronger CERES gene effect score in FP (n = 6) rather than FN (n = 5) cell lines in the DepMap. and WDR24. GATOR2 activates mTORC1 in response to aa ( Figure 1F; ref. 6), making it of specific interest given clinical trials supporting the use of mTOR inhibitors in RMS (7). To our knowledge, the GATOR2 complex has not previously been implicated in sarcoma cell signaling or growth. The other 3 GATOR2 complex members had varying results in our screen: SEH1L was not included in the library; WDR59 was depleted in both P3F KD and P3F + conditions, but without meeting statistical significance in the latter; and SEC13 was strongly and significantly depleted in both conditions, consistent with its GATOR2-independent role in ER vesicular biogenesis (21) (Supplemental Table 1).
GATOR2 is necessary for proliferation in PAX3-FOXO1-positive RMS cells. To extend these findings, we evaluated the role of each of the 5 members of the GATOR2 complex (MIOS, WDR59, WDR24, SEH1L, and SEC13) in 2 additional FP RMS cell lines: Rh41 and RMS13. To test the effect of PAX3-FOXO1 dosage on GATOR2 dependence, we conducted independent competition assays (22)(23)(24) in either parental (P3F + ) or knockdown (P3F KD ) cells. Cells expressing dCas9-KRAB were transduced with a blue fluorescent proteintagged (BFP-tagged) control sgRNA and GFP-tagged sgRNAs targeting GATOR2 or nontargeting control (  Figure 2E). We conclude that GATOR2 loss more effectively impedes the growth of PAX3-FOXO1-positive cells than those in which the fusion protein, and resultant cellular suppression, have been suppressed.
Tissue culture provides an overabundance of aa and thus might incorrectly measure the effects of the aa-responsive GATOR2 complex on cellular growth. Therefore, we turned to murine xenograft models. Rh30 cells were transduced with either sgRNA targeting WDR59 or a nontargeting control and implanted in the flanks of NSG mice. Tumor formation and growth were monitored over time and confirmed that cells transduced with sgWDR59 had reduced growth compared with cells transduced with a non-targeting control sgRNA (sgCTL; Figure 2, C and D). Thus, GATOR2 promotes both in vivo and in vitro growth of FP RMS. We next used an in vivo competition assay (25) to assess the contribution of PAX3-FOXO1 to GATOR2 dependence. Knockdown of PAX3-FOXO1 was selected against tumor cells in the absence of puromycin selection that could not be maintained in vivo, so we used this experimental design to benchmark the fitness of cells that may have slowly increasing PAX3-FOXO1 expression over time against the parental cell line. First, we confirmed that PAX3-FOXO1 is required for the fitness of Rh30 cells in vivo. A 4-fold excess of GFP tagged P3F KD cells was mixed with BFP-tagged P3F + cells and implanted in the flanks of NSG mice. After 21 days of growth, animals were euthanized and the resultant tumors were analyzed by flow cytometry and immunoblot. Consistent with our in vitro findings and a myoblast-derived in vivo model (26), the 4-fold excess of P3F KD cells was overtaken by P3F + cells during that time ( Figure 2, E and F).
Finally, we asked whether suppression of GATOR2 might favor the slowly proliferating P3F KD cells by depriving P3F + cells of mTORC1 signaling. We repeated the experiment using a 4-fold excess of GFPtagged P3F KD /sgWDR59 cells that competed against BFP-tagged P3F + /sgWDR59 cells. In contrast to the WDR59-sufficient tumors, the majority of cells in the WDR59 knockdown tumors were derived from the P3F KD xenograft ( Figure 2, E and F). These data support the hypothesis that loss of GATOR2 diminishes the growth of PAX3-FOXO1-positive RMS in vivo.
GATOR2 regulates mTORC1 activation and cellular proliferation in FP RMS. The GATOR2 complex inhibits the GATOR1 GTPase activating protein complex, thus enabling RAG-GTP/GDP heterodimers to recruit mTORC1 to the lysosomal membrane in response to aa abundance (6). We examined whether this pathway operates as previously described here in the context of FP RMS and whether loss of PAX3-FOXO1 alters this signaling cascade. Rh30 cells were deprived of leucine, lysine, and arginine for 3 hours and either lysed immediately or after 10 minutes of stimulation with aa-replete medium. As expected, cells transduced . Similarly, knockdown of most GATOR2 complex members prevented the localization of mTOR to the lysosome by widefield deconvolution microscopy, confirming that GATOR2 regulates mTORC1 in RMS through this previously described mechanism ( Figure  3C and Supplemental Figure 3A). SEC13 was an exception to both findings: knockdown of SEC13 did not prevent phosphorylation of p70S6K or localization of mTOR to the lysosome upon aa stimulation. Because SEC13 loss does not perturb GATOR2 function in RMS and was equally depleted in P3F + and P3F KD cells in our screen, we excluded it from further study of the PAX3-FOXO1 and GATOR2 interaction. We found that knockdown of PAX3-FOXO1 did not significantly affect aa stimulation of mTORC1 (P = 0.13 by 2-way ANOVA), nor did knockdown of GATOR2 alter the expression of PAX3-FOXO1. Thus, the protective effect of PAX3-FOXO1 knockdown seen in competition assays is not due changes in aa control of mTORC1 signaling; instead, it may reflect a decreased requirement for mTORC1 activation in P3F KD cells.
We next investigated why GATOR2-deficient cells are lost over time in competition assays. We did not detect significant apoptosis as measured by annexin V and propidium iodide staining (Supplemental Figure  3B). Because mTORC1 is essential for cell cycle progression (27), we interrogated whether GATOR2 loss slowed cell cycle transit in FP RMS. We assessed G1-S progression by blocking cells in G1 with thymidine and releasing into nocodazole. After thymidine block, cells demonstrated slowed transition from G1 into S phase after GATOR2 knockdown ( Figure 3D and Supplemental Figure 3C). Thus, GATOR2 loss, like pharmacological inhibition of mTORC1 (9), results in cytostatic rather than cytotoxic effects through slowed G1-S cell cycle progression.
Activation of mTORC1 is sufficient to compensate for loss of MIOS and WDR59. Although we have demonstrated that MIOS, WDR59, WDR24, and SEH1L are each required for mTORC1 activation, these proteins also have distinct, mTOR-independent roles in cell biology (28)(29)(30). Therefore, we considered whether our findings of increased GATOR2 dependence represent a requirement for mTORC1 signaling or, instead, highlight distinct, mTOR-independent genetic dependencies in FP RMS. We reasoned that if any given GATOR2 member is specifically required to activate mTORC1, then direct activation of mTORC1 should rescue cells from its loss. To test this hypothesis, we activated mTORC1 downstream of GATOR2 ( Figure  4A) through knockdown of DEPDC5, a member of the GATOR1 complex that GATOR2 inhibits.
Dual knockdown of GATOR2 and DEPDC5 rescued p70S6K phosphorylation (Figure 4, B and C) and 4EBP1 release from the m 7 -GTP cap (Supplemental Figure 4). In competition assays, knockdown of DEDPC5 rescued cells from MIOS or WDR59 loss ( Figure 4D) but only partially rescued WDR24 or SEH1L loss (differences assessed by 1-way ANOVA with post hoc Dunnett's test) when compared with BFP-tagged sgCTL cells. Thus, the GATOR2 complex can be segregated into members with entirely mTOR-dependent (MIOS, WDR59) and partially mTOR-independent (WDR24, SEH1L, and as shown above, SEC13) effects on proliferation in FP RMS. Ongoing efforts to elucidate the molecular function of each component as it acts within the GATOR2 complex may help clarify these differences (31)(32)(33)(34).
We next examined mTORC1 regulation in FN RMS as a comparison with the aa dependence exhibited by FP RMS. FN RMS shares histologic features with FP RMS and yet has distinct genetics and superior cure rates. These cells lack FOXO1 fusions; instead, they harbor oncogenic mutations in the RAS-MAP kinase and PI3-kinase pathways (15,16) that might activate mTORC1 in parallel to aa signaling ( Figure 4A) (35,36). FN RMS cells maintain higher levels of p70S6K and RPS6 phosphorylation in the absence of aa (Figure 4, E and F). Correspondingly, these FN cell lines have no decreased fitness upon loss of the mTOR-dependent GATOR2 complex members MIOS and WDR59, but they retain sensitivity to the partially mTOR-independent members WDR24, SEH1L, and SEC13 (Supplemental Figure 5). Exogenous expression of PAX3-FOXO1 was insufficient to enhance dependence on MIOS or WDR59 (Supplemental Figure 5). These data raise the possibility that oncogenic RAS mutation drives the decreased sensitivity of mTORC1 to aa deprivation or GATOR2 loss observed in these FN RMS cell lines.
Oncogenic RAS mutations diminish the role of GATOR2 in mTORC1 activation or growth. To determine whether oncogenic RAS is sufficient to reduce aa thresholds for mTORC1 activity, we expressed either WT or Q61H mutant alleles of NRAS (the most commonly mutated codon and RAS isoform in FN RMS; ref. 2), in FP RMS13 cells ( Figure 5A). NRAS Q61H , but not WT NRAS, bolstered basal levels of p70S6K and RPS6 phosphorylation (but not 4EBP1 release from eIF4E). Like DEPDC5 inactivation, expression of oncogenic NRAS Q61H rescued FP cells from GATOR2 loss in growth assays ( Figure 5B), though with more modest results.
We next asked whether MAP kinase activity is required for the observed mTORC1 signaling in NRAS-mutant RMS in aa-limiting conditions. In FP RMS13 cells, blocking MAP kinase output with the ERK inhibitor ulixertinib had minimal effects on p70S6K phosphorylation or RPS6 phosphorylation in response to aa. In cells harboring an NRAS Q61H mutation, however, ERK inhibition blocked both basal and aa--stimulated phosphorylation of p70S6K and RPS6 ( seen with the MEK inhibitor trametinib, and in BIRCH cells with endogenous HRAS mutation (Supplemental Figure 6). These data suggest that MEK-ERK activity downstream of oncogenic RAS mutations is necessary and sufficient for aa-independent phosphorylation of p70S6K. Recent work has shown that oncogenic RAS can confer RAG-independent activation of mTORC1 in response to select aa (37). However, FN RD cells transduced with a dominant negative RAG construct (38) lost aa-responsive p70S6K phosphorylation  Figure 7). Thus, we conclude that although maximal mTORC1 activity in RAS-mutant RMS requires intact ERK and RAG signaling, hyperactive ERK can sustain mTORC1 in aa-limiting conditions.
Bisteric mTORC1 inhibition induces regressions in xenografts derived from patients with FP RMS. Our genetic findings demonstrate that mTORC1 signaling is central for FP RMS growth and that its activity is under tight homeostatic control by nutrient availability. The allosteric mTOR inhibitor temsirolimus can improve eventfree survival in patients with RMS who have relapsed disease, but temsirolimus is insufficient to cure the disease even when combined with chemotherapy. On the basis of the data presented here, we hypothesized that mTORC1 is, in fact, a critical vulnerability in FP RMS but that the pharmacological means previously used to exploit this vulnerability may have been insufficient to provide maximal benefit. mTORC1-selective inhibitors with distinct modes of action from rapalogs (13) are now in clinical development. Bisteric inhibitors combine allosteric binding with an ATP-competitive inhibitory mechanism (13) and more durably suppress mTORC1 outputs, including phosphorylation of 4EBP1, than do allosteric inhibitors in models of glioma (9). Therefore, we investigated whether RMC-6272, a preclinical tool compound with properties representative of the clinical-stage bisteric mTORC1 inhibitor RMC-5552, currently in phase I testing, more completely suppresses mTORC1 activity and thus more effectively targets mTORC1 dependence in FP RMS.
In vitro, RMC-6272 was significantly more potent than the allosteric inhibitor rapamycin (of which temsirolimus is a prodrug) ( Figure 6A). Although both agents prevented p70S6K phosphorylation and increased phosphorylation of the mTORC2 target AKT at Ser473 (13), RMC-6272 induced dissociation of 4EBP1 from EIF4E, suppressing cap-dependent translation, whereas rapamycin did not ( Figure 6B and Supplemental Figure 8). On the basis of these data, we tested the preclinical activity of RMC-6272 in 2 xenografts derived from patients (PDXs) with FP RMS disease. Flanks of NSG mice were implanted with 1 × 10 6 viable cells in 50% MatriGel, and animals were randomized to either RMC-6272 or vehicle treatment once tumors reached a volume of 100 mm 3 . RMC-6272 was dosed by weekly i.p. injection, either at 6 or 8 mg/kg. At both doses, RMC-6272 demonstrated significant tumor control in these models ( Figure 6, C and D). Of 10 animals treated, we observed 5 complete responses to therapy (67% at 8 mg/kg and 43% at 6 mg/kg), and 7 with a best response of stable disease or better (100% at 8 mg/kg, and 71% at 6 mg/kg). By contrast, although temsirolimus treatment (20 mg/kg i.p. twice a week) (39) led to dephosphorylation of RPS6, it only induced regression in 1 of 8 tumors (-10% volume) and had no effect on tumor growth in 1 of the 2 models tested (Supplemental Figure 9). Treatment with RMC-6272 suppressed both RPS6 phosphorylation and EIF4G binding to EIF4E in treated tumors, confirming effective pharmacodynamic modulation of the target pathway and consistent with in vitro data ( Figure 6E). Three of 5 animals treated at 8 mg/kg had weight loss of greater than 10%, compared with none of the animals treated at 6 mg/kg. Indeed, mice treated at 6 mg/kg did not exhibit any significant weight loss compared with vehicle treatment (Supplemental Figure 10; P = 0.16 by 2-tailed t test), indicating this dose to be the maximum tolerated dose in NSG mice. Together, these data demonstrate that a bisteric mTORC1 inhibitor induces regressions in FP RMS PDX models and highlight 4EBP1 binding to EIF4E as a biomarker of treatment response.

Discussion
FOXO1 fusions are a consistent adverse prognostic factor for patients with RMS (3). Unlike kinase fusions present in other sarcomas involving NTRK3 or ALK, however, FOXO1 fusions lack an enzymatic, readily druggable activity. Novel technologies such as protein degraders may eventually permit direct targeting of nonenzymatic oncoproteins. As a complementary and alternative approach, here we used a functional genomics screen to identify a liability within the aa regulation of protein synthesis. Importantly, our isogenic system identified this vulnerability, alongside others, that could not be specifically linked to the PAX3-FOXO1 oncogene in important, larger-scale efforts. One limitation of our approach is that although our screen suggested that GATOR2 is completely dispensable in PAX3-FOXO1 knockdown cells, more detailed mechanistic studies instead unveiled a dosage response of enhanced sensitivity with higher PAX3-FOXO1 expression. Additionally, different GATOR2 members exhibited differing biochemical, cellular, and genetic interaction phenotypes, perhaps due to distinct contributions to the function of the GATOR2 complex. Although the relationship between PAX3-FOXO1 and GATOR2 is thus less stark and more nuanced than one of synthetic lethality, the differing sensitivity of PAX3-FOXO1 intact and knockdown cells to GATOR2 knockdown is shown here to have a potential therapeutic impact. Another limitation of our study was gene-by-gene validation, which has left the relationship between PAX3-FOXO1 and the other 34 screen hits awaiting additional study. A pooled rescreening strategy in additional FP cell lines may reduce this bottleneck in larger screens. The development and refinement of such isogenic systems may thus further uncover important oncogene dependencies with clinical ramifications.
Although nutrient control of mTORC1 is a conserved regulatory mechanism, we found that 2 RMS oncogenes, PAX3-FOXO1 and NRAS Q61H , differently affected the impact of GATOR2 on mTORC1 output. Reduced PAX3-FOXO1 levels diminish the cell biological effects of GATOR2 loss. It is possible that aa thresholds for mTORC1 activation are higher in PAX3-FOXO1-positive cells relative to knockdown cells, which our aa starvation and stimulation experiments may not reveal. Alternatively, or in addition, it may be that PAX3-FOXO1-positive cells have higher demands on protein synthesis than their knockdown counterparts, opening a potential therapeutic window. A third possibility is that there is a subset  Figure 2B; differences between parental and NRAS Q61H expressing cells assessed by 2-tailed t test. (C) RMS13 or RMS13 cells expressing NRAS Q61H were incubated with DMSO, 1 μM ulixertinib (ERKib), or 100 nM sapanisertib (TORib) for the duration of a 3-hour aa starvation with or without 10-minute stimulation. Immunoblots demonstrate that sapanisertib suppresses p70S6K phosphorylation in either condition but incompletely suppresses RPS6 phosphorylation. By contrast, ulixertinib restores aa control of RPS6 phosphorylation. (D) Quantification of p70S6K and RPS6 phosphorylation from C (n = 3 independent replicates). Significance of basal p70S6K or RPS6 phosphorylation assessed by 1-way ANOVA and Dunnett's multiple comparisons test. CTL, control. *P < 0.05; **P < 0.01. of PAX3-FOXO1 target transcripts that exhibit mTOR-dependent translation, defining an important level of control on fusion oncoprotein output. Unbiased approaches such as ribosome profiling or quantitative nascent peptide mass spectrometry may help distinguish among these possibilities. Broadly, we propose that cells transformed by oncogenes that redefine epigenetic states (e.g., PAX3-FOXO1) rather than those that directly activate signaling cascades (e.g., NRAS Q61H ) retain homeostatic buffers that control protein translation. Thus, PAX3-FOXO1 cannot overcome a strict requirement for aa activation that oncogenic RAS is able to circumvent. How other driver mutations in FN RMS, such as PIK3CA or FGFR4 activation, alter mTORC1 signaling remains to be studied. Driven by genetic data supporting mTORC1 dependence in FP RMS, we tested the bisteric mTORC1 inhibitor RMC-6272 in PDX models of this disease. This agent is sufficient to induce complete remissions in RMS preclinical testing, supporting the central importance of mTORC1 in FP RMS despite the more modest effects of allosteric inhibitors (39,40). We believe the preclinical data presented here provide strong rationale to pursue clinical testing of bisteric inhibitors regardless of the effects seen in ongoing studies with allosteric inhibitors such as temsirolimus. In addition to measuring pharmacokinetics and determining a maximally tolerated dose, response biomarkers to determine optimal biological dosing (41) will be essential to make sense of potentially conflicting data between these 2 classes of inhibitors. Prior studies have confirmed the prognostic impact of multiparameter mTOR pathway activation in RMS, including phosphorylation of both RPS6 and 4EBP1 (42). Our data confirm that dephosphorylation of 4EBP1 is a crucial difference between allosteric and bisteric inhibitors (9,13). Knockdown of 4EBP1 and dual knockdown of 4EBP1 and 4EBP2 enhanced, but did not completely restore, binding of EIF4G to EIF4E in the presence of RMC-6272 (Supplemental Figure 11A). This, in turn, gave rise to only partial RMC-6272 resistance in vitro (Supplemental Figure 11B), suggesting that 4EBP1 and 4EBP2 are not the sole effectors by which mTORC1 inhibition limits the growth of RMS. Studies will be required to define which mTORC1 targets contribute to the antitumor effects of RMC-6272 in RMS and may have a role in predicting mechanisms of drug resistance. Resistance to bisteric mTORC1 inhibitors may also emerge through selection for cells expressing low levels of PAX3-FOXO1, as have been identified in human tumors and murine models (26,43). Combination therapy approaches such as those already tested with temsirolimus may circumvent this possibility.
In conclusion, the data presented here support the development of isogenic systems to probe genetic networks that regulate the survival of oncogene-driven cancer cells. Such screens can identify oncogene-selective vulnerabilities within intact homeostatic mechanisms, for instance, aa control of protein translation, that can be re-examined as new pharmacological tools emerge with which to target them. The general strategy described in this study offers a means of refining precision therapy that could be useful to develop molecular approaches for the treatment of other cancers with driver oncogenes that are challenging to drug by current methods.

Methods
Cell culture and lentiviral transduction. RMS cell lines were obtained from the COG Repository (Rh30, Rh41), purchased from ATCC (RD, RMS13), or were a gift from Marielle E. Yohe (BIRCH, JR1; National Cancer Institute). RD and JR1 cells were grown in DMEM, and all other cell lines were grown in RPMI-1640, supplemented with 10% FBS and 1× penicillin/streptomycin, in a 37°C incubator with 5% CO 2 . Cells were tested quarterly for mycoplasma and tested to confirm identity by short tandem repeat analysis twice a year.
Generation of PAX3-FOXO1 knockdown cells. Rh30 cells were transduced with shP3F-GFP lentiviral particles, then sorted for GFP expression after 72 hours. GFP-positive cells were seeded at a concentration of 1000 cells/15-cm plate, and after 2 weeks, cloning cylinders were used to isolate and expand single-cell clones in 24-well plates. Resultant clones were analyzed by flow cytometry and immunoblot to identify a cell line with high GFP and low PAX3-FOXO1 expression, which was designated P3F KD . These cells were then transduced with lentiviral particles, including the dCas9-KRAB construct, sorted twice for BFP expression, and used in the initial screen.
CRISPR-interference screen. Lentiviral particles were generated from 293T cells transduced with pooled sgRNA libraries as described (18), then used to infect either Rh30-P3F KD cells, or Rh30 cells serially transduced with pLMN-GFP and dCas9-KRAB BFP, and sorted for GFP and BFP positivity (Rh30-P3F + ). Flow cytometry was used to confirm an MOI of approximately 1 at 48 hours after infection. Cells were JCI Insight 2022;7(23):e162207 https://doi.org/10.1172/jci.insight.162207 selected in puromycin for 72 hours, an aliquot was frozen for t 0 analysis, and the remainder were seeded in 500 cm 2 tissue culture plates at equal density (8 × 10 6 cells/plate). Every 3 days, cells were trypsinized, pooled, counted, and then replated at the same density to maintain 1000× coverage of each sgRNA construct. After 10 population doublings, cells were viably frozen.
Deep sequencing and data analysis were performed as described (18). Briefly, genomic DNA was extracted from t 0 and t end cells using a DNEasy Blood & Tissue kit (Qiagen), digested to enrich for lentiviral integration sites, and sgRNA sequences were amplified by PCR for subsequent sequencing on an Illumina HiSeq. Reads were aligned to the sgRNA library, and fold-change from t 0 to t end in the P3F + or P3F KD conditions was calculated. A gene-level score was then calculated as the mean of the top 3 scoring sgRNAs targeting a given transcript.
Flow cytometric competitive fitness and cell-fate assays. After puromycin selection, equal numbers of cells transduced with either GFP-or BFP-control labeled sgRNA constructs were plated in a single well. A sample of this mixed culture was taken at the time of plating, then every 4 days with trypsinization, for flow cytometric analysis. The percentage of GFP + cells was measured until 12 days in culture, normalizing back to the percentage present when mixed cultures were initiated. For apoptotic assays, cells were allowed to recover from puromycin selection for 48 hours prior to staining with annexin V and propidium iodide and analysis per manufacturer's instructions (CF488A Annexin V and PI Apoptosis Kit, Biotium). Analysis was restricted to BFP + cells to exclude residual nontransduced cells. For cell cycle assays, cells were plated in 2 μM thymidine for 18 hours after puromycin selection, then washed and released into fresh medium. After 8 hours, cells were trypsinized, fixed in ice-cold ethanol, and stained and analyzed per manufacturer's instructions (FxCycle PI/RNase staining solution, Thermo Fisher Scientific).
Immunoblots and m 7-GTP pulldown. Cells for immunoblots were lysed in ice cold RIPA buffer with protease and phosphatase inhibitors (Roche). Lysates were quantified using a DC protein assay (Bio-Rad), boiled in 1× Laemmli buffer, and then run on a 4% to 16% TGX gel. Gels were transferred onto nitrocellulose membrane and blotted for 3 to 18 hours in primary Ab, washed 3 times, and then incubated with HRP-conjugated secondary Ab for 1 hour. Blots were washed thrice, then imaged using ECL reagent (Amersham) on a GelDoc (Bio-Rad). Blots are representative of at least 3 independent replicates. Quantitation of phosphorylation was carried out by ImageJ (NIH) and normalized to total protein levels for the p70 band of S6K or RPS6, respectively.
Amino acid starve and stimulation assays. For immunoblots, cells were washed with PBS, then cultured in RPMI-1640 lacking arginine, leucine, or lysine (Thermo Fisher Scientific) for 3 hours (starve). Medium was aspirated and replaced with full RPMI-1640 (stimulation) for 10 minutes, after which cells were washed with ice-cold PBS and lysed. For microscopy experiments, starvation was limited to 2 hours, and stimulation was achieved by careful addition of RPMI-1640 supplemented to 2× arginine, leucine, and lysine, achieving 1× final concentration. Microscopy experiments were carried out in the presence of 10% dialyzed FBS. After 10 minutes, medium was aspirated and cells were immediately fixed.
Microscopy. Cells were plated on chamber slides (BioTek) coated with poly-l-lysine. Cells were fixed for 15 minutes with 4% PFA, washed, permeabilized for 5 minutes with 0.01% Triton-X-100 in PBS, washed, incubated for 1 hour with primary Ab in PBS with 1% BSA, washed, incubated with fluorescent conjugated secondary Abs for 1 hour, washed again, and sealed behind coverslips with ProLong Anti-Fade with Hoechst 33342 (Thermo Fisher Scientific). Slides were imaged on an OMX-SR system using a TIRF 60× objective (Nikon Imaging Center, UCSF Center for Light Microscopy). Colocalization of MTOR and LAMP2 was measured by Pearson's correlation coefficient using DeltaVision SoftWoRx software.
Cell line xenograft experiments. The Rh30 cell line stably expressing dCas9-KRAB was transduced with either guide RNA targeting WDR59, nontargeting control, or bicistronic versions of those vectors including an shRNA targeting PAX3-FOXO1, and selected in puromycin for 7 days. Once WDR59 and/or PAX3-FOXO1 knockdown was confirmed by immunoblot, cells were trypsinized, washed in PBS, and suspended JCI Insight 2022;7(23):e162207 https://doi.org/10.1172/jci.insight.162207 mean tumor volumes or percent GFP composition in 2-variable xenograft experiments were compared using a 2-tailed Student's t test. To quantify phosphorylation of target proteins, band intensities of phosphorylated and total protein were quantified in ImageJ. Phosphorylated levels were normalized to total protein levels. Comparisons were by2-tailed t test for 2-sample comparisons or by 2-way ANOVA to compare the effects of genotype and treatment. Post hoc comparisons to a single control were made with Dunnett's multiple comparisons test, and pairwise comparisons were made using Sidak's multiple comparisons test. Analysis of mTOR localization was made by Pearson's correlation coefficient using SoftWoRx software (DeltaVision). Comparison of PDX growth curves was performed by a linear mixed-effects regression model, with post hoc comparisons to the control by Dunnett's multiple comparisons test. In all experiments, an adjusted P value of less than 0.05 was considered significant.
Study approval. All procedures were carried out in accordance with IACUC policies under an approved protocol.

Author contributions
AJS led the project, conducted experiments, analyzed data, coordinated the contributions of other authors, and wrote the manuscript. TGB, JSW, and WCG provided project oversight and analyzed data, and TGB edited the manuscript. MAH provided experimental guidance in the sgRNA screen and analyzed data that it produced. JM and DVA conducted experiments and analyzed data, with support from KK, PKS, YP, JAG, and DAF. All authors reviewed the final manuscript.