KDM3B inhibitors disrupt the oncogenic activity of PAX3-FOXO1 in fusion-positive rhabdomyosarcoma

Fusion-positive rhabdomyosarcoma (FP-RMS) is an aggressive pediatric sarcoma driven primarily by the PAX3-FOXO1 fusion oncogene, for which therapies targeting PAX3-FOXO1 are lacking. Here, we screen 62,643 compounds using an engineered cell line that monitors PAX3-FOXO1 transcriptional activity identifying a hitherto uncharacterized compound, P3FI-63. RNA-seq, ATAC-seq, and docking analyses implicate histone lysine demethylases (KDMs) as its targets. Enzymatic assays confirm the inhibition of multiple KDMs with the highest selectivity for KDM3B. Structural similarity search of P3FI-63 identifies P3FI-90 with improved solubility and potency. Biophysical binding of P3FI-90 to KDM3B is demonstrated using NMR and SPR. P3FI-90 suppresses the growth of FP-RMS in vitro and in vivo through downregulating PAX3-FOXO1 activity, and combined knockdown of KDM3B and KDM1A phenocopies P3FI-90 effects. Thus, we report KDM inhibitors P3FI-63 and P3FI-90 with the highest specificity for KDM3B. Their potent suppression of PAX3-FOXO1 activity indicates a possible therapeutic approach for FP-RMS and other transcriptionally addicted cancers.

MYOD-SKP2 axis boosts tumorigenesis in fusion negative rhabdomyosarcoma by preventing differentiation through p57Kip2 targeting

Rhabdomyosarcomas (RMS) are pediatric mesenchymal-derived malignancies encompassing PAX3/7-FOXO1 Fusion Positive (FP)-RMS, and Fusion Negative (FN)-RMS with frequent RAS pathway mutations. RMS express the master myogenic transcription factor MYOD that, whilst essential for survival, cannot support differentiation. Here we discover SKP2, an oncogenic E3-ubiquitin ligase, as a critical pro-tumorigenic driver in FN-RMS. We show that SKP2 is overexpressed in RMS through the binding of MYOD to an intronic enhancer. SKP2 in FN-RMS promotes cell cycle progression and prevents differentiation by directly targeting p27Kip1 and p57Kip2, respectively. SKP2 depletion unlocks a partly MYOD-dependent myogenic transcriptional program and strongly affects stemness and tumorigenic features and prevents in vivo tumor growth. These effects are mirrored by the investigational NEDDylation inhibitor MLN4924. Results demonstrate a crucial crosstalk between transcriptional and post-translational mechanisms through the MYOD-SKP2 axis that contributes to tumorigenesis in FN-RMS. Finally, NEDDylation inhibition is identified as a potential therapeutic vulnerability in FN-RMS.

Preclinical development of a chimeric antigen receptor T cell therapy targeting FGFR4 in rhabdomyosarcoma

Pediatric patients with relapsed or refractory rhabdomyosarcoma (RMS) have dismal cure rates, and effective therapy is urgently needed. The oncogenic receptor tyrosine kinase fibroblast growth factor receptor 4 (FGFR4) is highly expressed in RMS and lowly expressed in healthy tissues. Here, we describe a second-generation FGFR4-targeting chimeric antigen receptor (CAR), based on an anti-human FGFR4-specific murine monoclonal antibody 3A11, as an adoptive T cell treatment for RMS. The 3A11 CAR T cells induced robust cytokine production and cytotoxicity against RMS cell lines in vitro. In contrast, a panel of healthy human primary cells failed to activate 3A11 CAR T cells, confirming the selectivity of 3A11 CAR T cells against tumors with high FGFR4 expression. Finally, we demonstrate that 3A11 CAR T cells are persistent in vivo and can effectively eliminate RMS tumors in two metastatic and two orthotopic models. Therefore, our study credentials CAR T cell therapy targeting FGFR4 to treat patients with RMS.

Preclinical Evaluation of the FGFR-Family Inhibitor Futibatinib for Pediatric Rhabdomyosarcoma

Rhabdomyosarcoma (RMS) is the most common pediatric soft tissue sarcoma. Despite decades of clinical trials, the overall survival rate for patients with relapsed and metastatic disease remains below 30%, underscoring the need for novel treatments. FGFR4, a receptor tyrosine kinase that is overexpressed in RMS and mutationally activated in 10% of cases, is a promising target for treatment. Here, we show that futibatinib, an irreversible pan-FGFR inhibitor, inhibits the growth of RMS cell lines in vitro by inhibiting phosphorylation of FGFR4 and its downstream targets. Moreover, we provide evidence that the combination of futibatinib with currently used chemotherapies such as irinotecan and vincristine has a synergistic effect against RMS in vitro. However, in RMS xenograft models, futibatinib monotherapy and combination treatment have limited efficacy in delaying tumor growth and prolonging survival. Moreover, limited efficacy is only observed in a PAX3-FOXO1 fusion-negative (FN) RMS cell line with mutationally activated FGFR4, whereas little or no efficacy is observed in PAX3-FOXO1 fusion-positive (FP) RMS cell lines with FGFR4 overexpression. Alternative treatment modalities such as combining futibatinib with other kinase inhibitors or targeting FGFR4 with CAR T cells or antibody-drug conjugate may be more effective than the approaches tested in this study.

A stem cell epigenome is associated with primary nonresponse to CD19 CAR T cells in pediatric acute lymphoblastic leukemia

CD19 chimeric antigen receptor T-cell therapy (CD19-CAR) has changed the treatment landscape and outcomes for patients with pre-B-cell acute lymphoblastic leukemia (B-ALL). Unfortunately, primary nonresponse (PNR), sustained CD19+ disease, and concurrent expansion of CD19-CAR occur in 20% of the patients and is associated with adverse outcomes. Although some failures may be attributable to CD19 loss, mechanisms of CD19-independent, leukemia-intrinsic resistance to CD19-CAR remain poorly understood. We hypothesize that PNR leukemias are distinct compared with primary sensitive (PS) leukemias and that these differences are present before treatment. We used a multiomic approach to investigate this in 14 patients (7 with PNR and 7 with PS) enrolled in the PLAT-02 trial at Seattle Children's Hospital. Long-read PacBio sequencing helped identify 1 PNR in which 47% of CD19 transcripts had exon 2 skipping, but other samples lacked CD19 transcript abnormalities. Epigenetic profiling discovered DNA hypermethylation at genes targeted by polycomb repressive complex 2 (PRC2) in embryonic stem cells. Similarly, assays of transposase-accessible chromatin-sequencing revealed reduced accessibility at these PRC2 target genes, with a gain in accessibility of regions characteristic of hematopoietic stem cells and multilineage progenitors in PNR. Single-cell RNA sequencing and cytometry by time of flight analyses identified leukemic subpopulations expressing multilineage markers and decreased antigen presentation in PNR. We thus describe the association of a stem cell epigenome with primary resistance to CD19-CAR therapy. Future trials incorporating these biomarkers, with the addition of multispecific CAR T cells targeting against leukemic stem cell or myeloid antigens, and/or combined epigenetic therapy to disrupt this distinct stem cell epigenome may improve outcomes of patients with B-ALL.

Lung endothelial cells regulate pulmonary fibrosis through FOXF1/R-Ras signaling

Pulmonary fibrosis results from dysregulated lung repair and involves multiple cell types. The role of endothelial cells (EC) in lung fibrosis is poorly understood. Using single cell RNA-sequencing we identified endothelial transcription factors involved in lung fibrogenesis, including FOXF1, SMAD6, ETV6 and LEF1. Focusing on FOXF1, we found that FOXF1 is decreased in EC within human idiopathic pulmonary fibrosis (IPF) and mouse bleomycin-injured lungs. Endothelial-specific Foxf1 inhibition in mice increased collagen depositions, promoted lung inflammation, and impaired R-Ras signaling. In vitro, FOXF1-deficient EC increased proliferation, invasion and activation of human lung fibroblasts, and stimulated macrophage migration by secreting IL-6, TNFα, CCL2 and CXCL1. FOXF1 inhibited TNFα and CCL2 through direct transcriptional activation of Rras gene promoter. Transgenic overexpression or endothelial-specific nanoparticle delivery of Foxf1 cDNA decreased pulmonary fibrosis in bleomycin-injured mice. Nanoparticle delivery of FOXF1 cDNA can be considered for future therapies in IPF.

Abstract 1784: FGFR4 and CD276 dual targeting CAR T cells demonstrate synergistic antitumor activity in childhood rhabdomyosarcoma

Background Chimeric antigen receptor T cells (CAR T) are potential therapies for rhabdosarcoma (RMS), the most common soft tissue sarcoma in children, where patients with relapsed or refractory disease have a dismal cure rate, and effective therapies are urgently needed. RMS tumors express high levels of the cell surface receptors FGFR4 and CD276, and both are direct targets of the PAX3-FOXO1 chimeric oncogene. However, CAR T have shown poor performance in solid tumors due to T cell exhaustion, limited persistence, and heterogeneous expression of target antigens. We hypothesize that engineering CAR constructs targeting FGFR4 and CD276 will enhance their activity and dual targeting will overcome the heterogenous target expression to effectively eliminate RMS tumors.

Methods We modified and tested different hinge and transmembrane domains (HTM) or intracellular domain (ICD) of FGFR4 CAR constructs and developed multiple dual targeting Bicistronic CARs (BiCisCARs) against FGFR4 and CD276. We tested these CARs in aggressive mouse models and performed Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) of tumor infiltrating CAR T cells to functionally characterize these constructs.

Results FGFR4 targeting CAR construct with CD8HTM and 4-1BB ICD could eliminate low burden RMS but was less effective against orthotopic tumors. Replacing CD8HTM with CD28HTM improved the efficacy in a moderate burden RMS model but did not show any benefit against a more aggressive RMS559 orthotopic model. Further modification using the CD28 ICD significantly enhanced the activity in a large burden RMS model but was unable to eradicate all tumors. Similarly, CAR T cells targeting CD276 alone, showed significant anti-tumor activity in moderate burden RMS models but could not eliminate all tumors. The BiCisCAR with FGFR4.CD28HTM.CD28ICD and CD276.CD8HTM.4-1BBICD was the most potent construct, eradicating 100% tumors in all tested orthotopic models, and those with heterogenous expression of target antigens. CITE-seq and flow cytometry assays demonstrated that this BiCisCAR showed the most significant tumor infiltration, and persistence, with limited exhaustion. This BiCisCAR exhibited a synergistic effect on cytokine production and anti-tumor activity, compared to single targeting CARs. Moreover, biochemical characterization revealed that the combined use of both CD28 and 4-1BB ICDs in the BiCisCAR resulted in the activation of three TCR downstream signaling pathways including AKT, Erk1/2 and p65.

Conclusions and Future DirectionsThus, we have developed a potent BiCisCAR with dual targeting of FGFR4 and CD276 that shows optimal biochemical activity, persistence, and limited exhaustion, and addresses heterogenous expression of target antigens. This BiCisCAR will be further developed for future clinical trials in patients with high-risk RMS.

Citation Format: Meijie Tian, Jun S. Wei, Adam Cheuk, David Milewski, Zhongmei Zhang, Yong Yuan Kim, Can Liu, Sherif Badr, Michael C. Kelly, Jerry T. Wu, Abdelrahman Rahmy, Hsien-Chao Chou, Xinyu Wen, Javed Khan. FGFR4 and CD276 dual targeting CAR T cells demonstrate synergistic antitumor activity in childhood rhabdomyosarcoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 1784.

Abstract 6736: Suppression of antigen presentation is a hallmark of pediatric rhabdomyosarcoma

Background: Rhabdomyosarcoma (RMS) is the most common soft tissue tumor in children and young adults, accounting for approximately 350-400 new cases annually in the US. These tumors are diagnosed based on their expression of skeletal muscle lineage markers. Clinical trials testing conventional immunotherapies such as cancer vaccines or immune checkpoint blockade (ICB) therapy have found no activity in RMS, although the root causes of failure are largely unexplored. A recent study by our group found that RMS tumors have a lack of immune infiltrate, consistent with previous reports. However, we made an unexpected observation that RMS tumors had a near absence of antigen processing and presentation (APP) gene expression, a key requisite for anti-tumor CD8+ T lymphocytes. In this study, we investigate the root causes of the lack of APP in RMS and evaluate implications for immunotherapy.

Methods: Publicly available DNA sequencing datasets were used to evaluate APP gene mutations in RMS tumor samples. RNAseq from reporter iPSC cell lines demarcating specific stages of skeletal muscle differentiation were used to evaluate APP gene expression during myogenesis. Surface MHC class I expression was evaluated using flow cytometry against a panel of pediatric sarcoma cell lines. RMS PDX and cell line APP protein levels were measured by western blot. Stimulation of APP in RMS cell lines was tested using IFNγ or sub-cytotoxic doses of different epigenetic therapeutics (DNMTi, HDACi, EZH2i) before analyzing surface MHC class I expression by flow cytometry.

Results: RNAseq analysis of the differentiation states across the normal human skeletal muscle lineage revealed a myoblastic cell state which transiently downregulates the APP pathway. RNA expression of HLA-A,B,C genes were lower in RMS than any other pediatric extracranial solid tumor histology studied. RMS cell lines and PDX models also displayed low APP pathway gene expression by western blot and flow cytometry, with FOXO1 fusion positive RMS having a near absence of expression. RMS tumor exome sequencing data from RMS patients showed a lack of enrichment of APP gene mutations. Instead, we found that APP could be induced in nearly all samples tested using IFNγ or epigenetic stimulation with Decitabine, Entinostat, or Tazemetostat.

Conclusions: We uncovered a previously unappreciated lack of antigen presentation in RMS which phenocopies myoblasts in normal skeletal muscle development. The downregulation of MHC class I antigen presentation is not due to APP pathway mutations but rather a partially reversible downregulation mirroring this specific stage of normal skeletal muscle development. Our findings suggest that pharmacological reversal of the APP pathway downregulation may enhance immunotherapies targeting MHC class I antigens in RMS.

Citation Format: David Milewski, Meijie Tian, Yong Kim, Jun Wei, Javed Khan. Suppression of antigen presentation is a hallmark of pediatric rhabdomyosarcoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6736.

Abstract 3538: Endogenous HiBiT-tagging of PAX3-FOXO1 identifies potent suppressors of PAX3-FOXO1 protein levels by high-throughput screening

Background: Oncogenic fusion genes are attractive therapeutic targets due to their tumor-specific expression and driver roles in cancers. PAX3-FOXO1 (P3F) is the dominant oncogenic driver of fusion-positive rhabdomyosarcoma (FP-RMS) with no targeted therapy. We developed methods to directly measure endogenous P3F protein levels amenable to high-throughput drug screens to identify suppressors of P3F.

Methods: HiBiT tag, an 11 amino acid peptide of the small fragment of NanoLuc luciferase, was inserted into the endogenous P3F using CRISPR-Cas9 in FP-RMS cell lines RH4 and SCMC. Western analysis was used for HiBiT tag validation and confirmation of P3F suppression. RNA-seq and ChIP-seq were used to assess transcriptomics and DNA binding of HiBiT-tagged P3F (P3F-HiBiT) respectively. High-throughput drug screen using Nano-Glo luciferase assay was performed using the Mechanism Interrogation PlatE (MIPE 5.0) drug library, which included 2,480 drugs with known mechanisms of action. CellTiter-Glo was used to monitor cell viability. We identified drugs that suppressed P3F by Nano-Glo without acute cytotoxicity by CellTiter-Glo at an early 24-hour timepoint. Mouse xenograft model of FP-RMS was used to investigate in vivo efficacy of top hits.

Results: We validated HiBiT tagging of P3F and not the wild-type FOXO1 by Western analysis. We showed that the HiBiT tag did not change the function of P3F by transducing human fibroblasts with P3F-HiBiT versus unmodified P3F. Gene Set Enrichment Analysis (GSEA) of RNA-seq showed that P3F-HiBiT activated the same downstream target genes as unmodified P3F. ChIP-seq using HiBiT antibody in HiBiT-tagged FP-RMS cell lines RH4 and SCMC matched the genomic locations from ChIP-seq with P3F antibody in parental RH4 and SCMC. Using a cutoff of Area Under the Curve (AUC) of CellTiter-Glo - AUC of Nano-Glo > 90, in both RH4 and SCMC, identified 182 compounds. Filtering for drugs with ≥ 3 hits for the same target identified 14 drug classes that suppressed P3F protein level including HDAC inhibitors (3), mTOR inhibitors (4), CDK inhibitors (8), and BRD4 inhibitors (3). One top hit was the CDK inhibitor TG02 (Zotiraciclib), currently in human trials. TG02 suppressed P3F protein levels by Nano-Glo and Western analysis. We confirmed induction of apoptosis by PARP cleavage in a panel of FP-RMS cell lines. GSEA analysis of RNA-seq after treatment with TG02 showed marked suppression of P3F target gene sets. TG02 also significantly delayed tumor progression of established tumors in a mouse xenograft model of FP-RMS without weight loss.

Conclusion and Future Directions:By HiBiT tagging the fusion oncogene P3F, we identified 182 compounds that suppress P3F levels of which TG02 was a top hit that also showed in vivo efficacy. Drug combination studies are currently underway to identify synergistic suppressors of P3F protein levels that can be translated into clinical trials.

Citation Format: Yong Yean Kim, Robert G. Hawley, Mehal Churiwal, Teresa S. Hawley, Christine N. Evans, Raj Chari, David Milewski, Ranuka Sinniah, Young K. Song, Hsien-Chao Chou, Xinyu Wen, Ying Pang, Jing Wu, Craig J. Thomas, Jun S. Wei, Michele Ceribelli, Javed Khan. Endogenous HiBiT-tagging of PAX3-FOXO1 identifies potent suppressors of PAX3-FOXO1 protein levels by high-throughput screening. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 3538.

Predicting Molecular Subtype and Survival of Rhabdomyosarcoma Patients Using Deep Learning of H&E Images: A Report from the Children's Oncology Group

PURPOSE

Rhabdomyosarcoma (RMS) is an aggressive soft-tissue sarcoma, which primarily occurs in children and young adults. We previously reported specific genomic alterations in RMS, which strongly correlated with survival; however, predicting these mutations or high-risk disease at diagnosis remains a significant challenge. In this study, we utilized convolutional neural networks (CNN) to learn histologic features associated with driver mutations and outcome using hematoxylin and eosin (H&E) images of RMS.

EXPERIMENTAL DESIGN

Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from 321 patients with RMS enrolled in Children's Oncology Group (COG) trials (1998-2017). Patches were extracted and fed into deep learning CNNs to learn features associated with mutations and relative event-free survival risk. The performance of the trained models was evaluated against independent test sample data (n = 136) or holdout test data.

RESULTS

The trained CNN could accurately classify alveolar RMS, a high-risk subtype associated with PAX3/7-FOXO1 fusion genes, with an ROC of 0.85 on an independent test dataset. CNN models trained on mutationally-annotated samples identified tumors with RAS pathway with a ROC of 0.67, and high-risk mutations in MYOD1 or TP53 with a ROC of 0.97 and 0.63, respectively. Remarkably, CNN models were superior in predicting event-free and overall survival compared with current molecular-clinical risk stratification.

CONCLUSIONS

This study demonstrates that high-risk features, including those associated with certain mutations, can be readily identified at diagnosis using deep learning. CNNs are a powerful tool for diagnostic and prognostic prediction of rhabdomyosarcoma, which will be tested in prospective COG clinical trials.

Abstract 552: FGFR4 and CD276 dualtargeting CAR-T cells for treating rhabdomyosarcoma and other solid tumors

Background: Chimeric antigen receptor T-cell therapies (CAR-T) have shown success in treating refractory and relapsed leukemia and lymphoma, while they perform poorly in solid tumors due to heterogenous expression of tumor-associated antigens (TAAs), limited T cell persistence and propensity for exhaustion. The receptor tyrosine kinase FGFR4 and immune checkpoint molecule CD276 are highly and heterogeneously expressed in some solid tumors, including Rhabdomyosarcoma (RMS), a most common soft tissue sarcoma of childhood, and human hepatocellular carcinoma (HCC). However, their expression is usually low in normal human tissues. These features make FGFR4 and CD276 promising therapeutic targets for CAR-T therapy for RMS and HCC. We have developed a FGFR4 targeting CAR construct (3A11-BBz) with a CD8 hinge (H) and a transmembrane domain (TM) infused with a 4-1BB intracellular domain (ICD). 3A11-BBz CAR can efficiently eliminate low RMS disease burden in metastatic models, but less effectively for bulky disease in RMS intramuscular (I.M.) xenograft models. Testing of a CD276 targeting CAR T-cells showed significant shrinking of tumors in RMS I.M. xenograft models.

Methods: To improve the CAR-T cells efficacy, we first modified the H/TM and ICD of 3A11-BBz CAR to CD28 (3A11-CD28z). To overcome tumor heterogeneity, we also created Bicistronic CARs (BiCisCARs) combining the complete FGFR4 and CD276 CAR into a single construct allowing co-expression of both constructs on the same T cells. We then tested the efficacy of these CARs in-vitro and in-vivo using intramuscular FP-RMS xenograft (RH30) or HCC intraperitoneal models.

Results and Conclusions: We found either FGFR4 targeting CARs or dual targeting BiCisCARs, showed similar in-vitro cytotoxicity against RMS cells and HCC cells. However, CARs with CD28 ICD released more IL-2 than those with 4-1BB ICD when co-cultured with target cells. In RMS I.M. xenograft model, 3A11-CD28z CAR-T cells shrank and eliminated the tumor in 5/8 mice whereas 3A11-BBz only suppressed tumor growth. Furthermore, 3A11-CD28z BiCisCAR eradicated tumor cells in 8/8 mice, whereas 3A11-BBz BiCisCAR showed very poor efficacy. Moreover, there are more 3A11-CD28z BiCisCAR T-cells persisting in blood and spleen than the other bicistronic or single CAR-T cells, suggesting this BiCisCAR-T cells have prolonged persistence. Therefore, we have developed a potent BiCisCAR dual targeting both FGFR4 and CD276 that overcomes RMS heterogeneity and effectively eliminates tumors in-vivo, which will be developed as a future therapeutic CAR for clinical trials.

Citation Format: Meijie Tian, Adam Cheuk, David Milewski, Jun S. Wei, Hsien-Chao Chou, Yong Yean Kim, Young K. Song, Brad St. Croix, Mitchell Ho, Javed Khan. FGFR4 and CD276 dualtargeting CAR-T cells for treating rhabdomyosarcoma and other solid tumors [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 552.

Abstract 466: Predicting survival of rhabdomyosarcoma patients based on deep learning of H&E images

Background: Rhabdomyosarcoma (RMS) is an aggressive soft tissue tumor in children and young adults, accounting for 350-400 new cases annually in the US. Diagnosis of RMS is defined by the expression of genes related to skeletal muscle differentiation and can be further subclassified based on histological patterns (embryonal, ERMS; alveolar, ARMS; spindle/sclerosing, SSRMS). Genetic studies have found that the presence of a PAX fusion gene (FP-RMS), which is present in many ARMS tumors, correlates with poor outcome. Subsequent studies have identified additional genetic alterations (ex. TP53 or MYOD1 mutations) which also display distinct histological features and are associated with poor outcome. As a result, there is a growing need to identify these mutations to improve risk stratification. The goal of this study is to develop and test deep learning algorithms from diagnostic H&E images of RMS tumors which can aid in the diagnosis, mutation prediction and risk stratification for RMS patients.

Methods: De-identified RMS patient samples were collected from tissue banking studies from Children’s Oncology Group (n=275), University Hospital Zurich (n=250) and Memorial Sloan Kettering (n=10). H&E stains on whole slides or TMAs were digitally scanned and used for analysis. Clinical information including clinical risk group, event-free survival, and genomic findings were used as available for training and testing. Convolutional neural networks (CNN) using EfficientNet were trained using K-fold cross validation to classify tumor histology, mutation probability, and risk stratification and tested against randomly selected samples or independent datasets when available.

Results: The developed AI algorithm was able to classify tissue as ARMS (FP-RMS), ERMS (FN-RMS), stroma and necrosis with an average weighted intersect-over-union of 0.74 when compared to an expert pathologist annotation. A second deep learning algorithm developed specifically for distinguishing FP-RMS from FN-RMS displayed excellent sensitivity (FP-RMS=0.88) and specificity (FP-RMS=0.86) when tested against an independent RMS TMA dataset. Algorithms were also trained to predict mutations in MYOD1, RAS pathway genes, and TP53 and displayed good performance with ROC values of 0.96, 0.68, and 0.64, respectively. Lastly, we developed an algorithm to provide a Cox proportional hazard prediction based on H&E images. The resulting algorithm was capable of predicting EFS with similar accuracy as current clinical risk group assessment with improved ability to distinguish intermediate and high risk patients.

Conclusions: Deep learning with convolutional neural networks provides pathologist-independent classification of RMS patients from simple H&E images. These AI algorithms can provide probabilities of prognostically relevant genetic alterations and survival which will ultimately contribute to better risk stratification of RMS patients.

Citation Format: David Milewski, Hyun Jung, G. Thomas Brown, Yanling Liu, Jack Collins, Marc Ladanyi, Erin Rudzinski, Javed Khan. Predicting survival of rhabdomyosarcoma patients based on deep learning of H&E images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 466.

Abstract LB213: Potent antitumor activity of a FGFR4 CAR-T in rhabdomyosarcoma

Rhabdomyosarcoma (RMS) is an aggressive soft tissue sarcoma originating from skeletal muscle in children and adolescent young adults. Despite multi-modal aggressive therapies, relapsed, refractory or metastatic rhabdomyosarcoma remains a lethal disease with no significant improvement in outcome over decades of clinical trials. Therefore novel therapies are needed. FGFR4 is a developmentally regulated cell surface receptor tyrosine kinase that is overexpressed in RMS when compared with normal tissues, and mutationally activated in about 7.5% of RMS. Recently we showed that PAX3-FOXO1 establishes a super-enhancer in the FGFR4 genomic locus driving its high expression in fusion positive RMS. CAR T-cell therapy is effective in treating refractory and relapsed B-cell leukemia and lymphoma, with three CARs targeting CD19 approved by the FDA. Multiple CART trials are currently underway for solid tumors. Since FGFR4 is a cell surface protein, we hypothesized that FGFR4 will provide a rational target for immunotherapy in RMS. We confirmed by immunohistochemistry staining, western analysis, and Meso Scale Discovery that FGFR4 protein is highly differentially expressed in RMS samples. We developed a murine anti-FGFR4 antibody, 3A11, by immunizing mouse with FGFR4-IG fusion protein. 3A11 showed high affinity and specificity of binding to FGFR4. We then developed a second-generation CAR using the VL and VH domain of 3A11 antibody and found that the scFvFc retained its specificity and high affinity at nanomolar range. Human T cells transduced with 3A11 CAR construct were found to be highly potent at inducing IFN-γ, TNF-α, IL-2 and cytotoxicity when the FGFR4-CART was co-cultured with RMS cells, but not with RMS cells with FGFR4 knocked out or FGFR4 negative cells. 3A11 CART incubated with human primary cells obtained from liver, kidney, heart, and pancreas, did not elicit a cytokine response, indicating a low potential for “on-target off-tumor” toxicity. In vivo testing also found that 3A11 CART eliminated RMS cells in both murine xenograft metastatic and localized subcutaneous models. Therefore we have developed a CART targeting FGFR4 that shows high potency for treating RMS. A phase 1 FGFR4-CART clinical trial is planned for children and adolescent young adults with relapsed/refractory rhabdomyosarcoma.

Citation Format: Adam Tai Chi Cheuk, Meijie Tian, Nityashree Shivaprasad, Steven Highfill, David Milewski, G Tom Brown, Peter Azorsa, Dina Schneider, Berkley Gryder, Jun S Wei, Young Kwok Song, Hsien-Chao Chou, Jerry Wu, Joon-Yong Chung, Brian Belyea, Corinne Linardic, Stephen M Hewitt, Boro Dropulic, Rimas Orentas, Javed Khan. Potent antitumor activity of a FGFR4 CAR-T in rhabdomyosarcoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr LB213.

Immuno-transcriptomic profiling of extracranial pediatric solid malignancies

We perform an immunogenomics analysis utilizing whole-transcriptome sequencing of 657 pediatric extracranial solid cancer samples representing 14 diagnoses, and additionally utilize transcriptomes of 131 pediatric cancer cell lines and 147 normal tissue samples for comparison. We describe patterns of infiltrating immune cells, T cell receptor (TCR) clonal expansion, and translationally relevant immune checkpoints. We find that tumor-infiltrating lymphocytes and TCR counts vary widely across cancer types and within each diagnosis, and notably are significantly predictive of survival in osteosarcoma patients. We identify potential cancer-specific immunotherapeutic targets for adoptive cell therapies including cell-surface proteins, tumor germline antigens, and lineage-specific transcription factors. Using an orthogonal immunopeptidomics approach, we find several potential immunotherapeutic targets in osteosarcoma and Ewing sarcoma and validated PRAME as a bona fide multi-pediatric cancer target. Importantly, this work provides a critical framework for immune targeting of extracranial solid tumors using parallel immuno-transcriptomic and -peptidomic approaches.

Proteogenomic Analysis Unveils the HLA Class I-Presented Immunopeptidome in Melanoma and EGFR-Mutant Lung Adenocarcinoma

Immune checkpoint inhibitors and adoptive lymphocyte transfer–based therapies have shown great therapeutic potential in cancers with high tumor mutational burden (TMB), such as melanoma, but not in cancers with low TMB, such as mutant epidermal growth factor receptor (EGFR)–driven lung adenocarcinoma. Precision immunotherapy is an unmet need for most cancers, particularly for cancers that respond inadequately to immune checkpoint inhibitors. Here, we employed large-scale MS-based proteogenomic profiling to identify potential immunogenic human leukocyte antigen (HLA) class I-presented peptides in melanoma and EGFR-mutant lung adenocarcinoma. Similar numbers of peptides were identified from both tumor types. Cell line and patient-specific databases (DBs) were constructed using variants identified from whole-exome sequencing. A de novo search algorithm was used to interrogate the HLA class I immunopeptidome MS data. We identified 12 variant peptides and several classes of tumor-associated antigen-derived peptides. We constructed a cancer germ line (CG) antigen DB with 285 antigens. This allowed us to identify 40 class I-presented CG antigen–derived peptides. The class I immunopeptidome comprised more than 1000 post-translationally modified (PTM) peptides representing 58 different PTMs, underscoring the critical role PTMs may play in HLA binding. Finally, leveraging de novo search algorithm and an annotated long noncoding RNA (lncRNA) DB, we developed a novel lncRNA-encoded peptide discovery pipeline to identify 44 lncRNA-derived peptides that are presented by class I. We validated tandem MS spectra of select variant, CG antigen, and lncRNA-derived peptides using synthetic peptides and performed HLA class I-binding assays to demonstrate binding to class I proteins. In summary, we provide direct evidence of HLA class I presentation of a large number of variant and tumor-associated peptides in both low and high TMB cancer. These results can potentially be useful for precision immunotherapies, such as vaccine or adoptive cell therapies in melanoma and EGFR-mutant lung cancers.

•

Proteogenomics identified ∼35,000 class I-presented peptides.

•

CG antigen and PTM peptides identified in melanoma and lung cancer.

•

De novo search identified variant and lncRNA-derived peptides.

•

A new strategy to identify class I-presented lncRNA-derived peptides developed.

Targeting EYA3 in Ewing Sarcoma Retards Tumor Growth and Angiogenesis

EWSR1/FLI1, the most common fusion gene in Ewing sarcoma, upregulates expression of the Eyes Absent 3 (EYA3) transactivator-phosphatase protein. The purpose of this study was to investigate molecular and cellular mechanisms through which EYA3 might promote Ewing sarcoma tumor growth and to determine whether the EYA3 tyrosine phosphatase activity represents a viable therapeutic target. We used genetic and pharmacologic modulation of EYA3 in cell line-based xenografts to examine how loss of EYA3 tyrosine phosphatase activity affects tumor growth and angiogenesis. Molecular mechanisms were evaluated in vivo and in vitro through analyses of tumor tissue and multicellular tumor spheroids. Our results show that both loss of EYA3 in Ewing sarcoma cells and pharmacologic inhibition of the EYA3 tyrosine phosphatase activity inhibit tumor growth and tumor angiogenesis. EYA3 regulates levels of VEGFA in Ewing tumors, as well as promoting DNA damage repair and survival of Ewing sarcoma tumor cells. Target engagement is demonstrated in tumor tissue through elevated levels of the EYA3 substrate H2AX-pY142 upon loss of EYA3 or with Benzarone treatment. The efficacy of EYA3 tyrosine phosphatase inhibition in attenuating tumor growth and angiogenesis is corroborated in an Ewing sarcoma patient-derived tumor xenograft. Together, the results presented here validate EYA3 as a target for the development of novel Ewing sarcoma therapeutic strategies, and set the stage for evaluating the efficacy of combining the antiangiogenic and anti-cell survival effects of EYA3 inhibition with cytotoxic chemotherapy.

Interaction between SNAI2 and MYOD enhances oncogenesis and suppresses differentiation in Fusion Negative Rhabdomyosarcoma

Rhabdomyosarcoma (RMS) is an aggressive pediatric malignancy of the muscle, that includes Fusion Positive (FP)-RMS harboring PAX3/7-FOXO1 and Fusion Negative (FN)-RMS commonly with RAS pathway mutations. RMS express myogenic master transcription factors MYOD and MYOG yet are unable to terminally differentiate. Here, we report that SNAI2 is highly expressed in FN-RMS, is oncogenic, blocks myogenic differentiation, and promotes growth. MYOD activates SNAI2 transcription via super enhancers with striped 3D contact architecture. Genome wide chromatin binding analysis demonstrates that SNAI2 preferentially binds enhancer elements and competes with MYOD at a subset of myogenic enhancers required for terminal differentiation. SNAI2 also suppresses expression of a muscle differentiation program modulated by MYOG, MEF2, and CDKN1A. Further, RAS/MEK-signaling modulates SNAI2 levels and binding to chromatin, suggesting that the differentiation blockade by oncogenic RAS is mediated in part by SNAI2. Thus, an interplay between SNAI2, MYOD, and RAS prevents myogenic differentiation and promotes tumorigenesis.

Abstract 3445: Immuno-transcriptomic profiling identifies actionable genomic alterations in pediatric solid malignancies

Malignancy remains the leading cause of disease-related death in children. To identify potential tumor-driving molecular targets and characterize immunogenomic profiles in pediatric cancers, we performed RNA-seq analysis on a cohort of 788 pediatric solid tumors across 14 different diagnoses in conjunction with additional 147 normal tissues for comparison. Sequencing data were analyzed for expressed mutations, fusion events, and expressional patterns, providing therapeutic targets and rich cancer biology for these childhood cancers. Furthermore, we describe a comprehensive and in-depth immunogenomic landscape of these solid tumors including immune cell infiltrate, neoepitope analysis from expressed mutations and fusions, expressional patterns of clinically relevant immune checkpoint genes, expression of tumor-specific genes as potential pharmacological or immunological targets, and T cell receptor repertoire. Across the cohort, we observed a striking correlation between the expressed neoepitope burden in tumors and enrichment of the effector immune signatures. Intriguingly, canonical fusions (e.g. EWS-FLI1) contribute a disproportionally large number of neoepitopes in these typically low mutational tumors. Histology-specific immunogenomic patterns are also apparent. Several of the pediatric cancers such as alveolar soft part sarcoma and osteosarcoma exhibit rich immune cell infiltration and evidence for activated T cell activities, whereas others such as Wilms tumors and synovial sarcoma generally have a very low T cell infiltration. In addition, we demonstrated a significant positive correlation between tumor-infiltrating CD8+ T cells and overall survival in patients with osteosarcoma, revealing the clinical importance of these tumor-infiltrating immune cells in these childhood cancers. Moreover, an orthogonal evaluation of immunopeptidome in osteosarcoma, a cancer type displaying high immune infiltrates, confirmed our transcriptomic findings on potential targetable tumor-specific genes. Finally, we took an adoptive cell therapy-based approach to target a tumor-specific gene PRAME identified by our transcriptomic and immunopeptidomic studies and showed significant in-vitro cytotoxicity using T cells expressing TCRs specifically targeting PRAME in osteosarcoma U2OS cells. Therefore, we demonstrate that RNA-seq is a powerful tool to identify clinically relevant and histology-specific genomic alterations and translationally relevant immunogenomic patterns for pediatric cancers. This study also represents one of the largest of its type to date and provides a framework for future translational efforts in pediatric cancer.

Citation Format: Jun S. Wei, Andrew S. Brohl, Sivasish Sindiri, David Milewski, Young K. Song, Sushma Nagaraj, Vineela Gangalapudi, Xinyu Wen, Marc Ladanyi, Javed Khan. Immuno-transcriptomic profiling identifies actionable genomic alterations in pediatric solid malignancies [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 3445.

Abstract 6613: Identification of targetable HLA*A2:01 restricted peptides in osteosarcoma

Background: Osteosarcoma is the most common primary tumor of bone with 800-900 new cases diagnosed annually in the United States. The development of effective chemotherapeutic agents in the early 1970's led to a dramatic increase in osteosarcoma survival from less than 10% to over 60%. Unfortunately, little progress has been made in the past several decades despite intensified treatment and the application of targeted therapies with the 5-yr overall survival stalled at ~70%. Several phase I studies have reported occasional clinical responses to immune checkpoint inhibitors in osteosarcoma patients, although the role of immunotherapy in osteosarcoma treatment is largely unexplored. The goal of this work was to perform a deep immuno-genomic and proteomic profile of osteosarcoma tumors as the basis for developing effective immune-based therapy for osteosarcoma.

Methods: Osteosarcoma tumors and cell lines were subjected to whole transcriptomes analysis. Intratumoral T cell receptor sequences were identified using MiXCR and VDJtools from the RNA-seq data. Immune signatures were scored in each tumor sample using single-sample GSEA. MHC class I bound peptides were isolated by immunoprecipitation of MHC class I complex, acid elution of bound peptides, and identification using LC-MS/MS. Healthy donor T cells were transduced with a lentiviral construct targeting PRAME and co-cultured with U2OS cells to evaluate specific PRAME peptide targeting by T cells.

Results: Gene expression profiling of extracranial pediatric solid tumors identified high T cell infiltration in osteosarcoma patient tumors. High CD8+ T cell infiltration were associated with favorable outcome among osteosarcoma patients, suggesting the presence of an underlying endogenous T cell response against osteosarcoma tumors. We identified a set of tumor associated antigens, such as cancer germline antigens (CGA), which are highly expressed in osteosarcomas and display low/absent expression in normal tissues. PRAME, a CGA expressed in multiple solid tumors and hematological malignancies, was identified as one of the most frequently overexpressed CGAs in osteosarcoma. Immunoprecipitation of MHC Class I complexes followed by LC-MS/MS from osteosarcoma cells identified peptides derived from PRAME and other tumor associated antigens that are presented by HLA*A2:01 in osteosarcoma. In a proof-of-concept experiment, we demonstrate effective T cell cytotoxicity against osteosarcoma using a TCR which specifically recognizes an PRAME HLA*A2:01 peptide identified by mass spectrometry.

Conclusions: Osteosarcoma tumors have prognostic high CD8+ T cell infiltration, expression of cancer testis antigens, and presentation of tumor associated peptides on MHC class I complex. These data warrant a broad evaluation of the efficacy of immunotherapy in osteosarcoma.

Citation Format: David Milewski, Sivasish Sindiri, Jun S. Wei, Andrew Brohl, Young Song, Xinyu Wen, Andrew Qi, Udayan Guha, Javed Khan. Identification of targetable HLA*A2:01 restricted peptides in osteosarcoma [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 6613.

FOXM1 nuclear transcription factor translocates into mitochondria and inhibits oxidative phosphorylation

Forkhead box M1 (FOXM1), a nuclear transcription factor that activates cell cycle regulatory genes, is highly expressed in a majority of human cancers. The function of FOXM1 independent of nuclear transcription is unknown. In the present study, we found the FOXM1 protein inside the mitochondria. Using site-directed mutagenesis, we generated FOXM1 mutant proteins that localized to distinct cellular compartments, uncoupling the nuclear and mitochondrial functions of FOXM1. Directing FOXM1 into the mitochondria decreased mitochondrial mass, membrane potential, respiration, and electron transport chain (ETC) activity. In mitochondria, the FOXM1 directly bound to and increased the pentatricopeptide repeat domain 1 (PTCD1) protein, a mitochondrial leucine-specific tRNA binding protein that inhibits leucine-rich ETC complexes. Mitochondrial FOXM1 did not change cellular proliferation. Thus, FOXM1 translocates into mitochondria and inhibits mitochondrial respiration by increasing PTCD1. We identify a new paradigm that FOXM1 regulates mitochondrial homeostasis in a process independent of nuclear transcription.

Loss of FOXM1 in macrophages promotes pulmonary fibrosis by activating p38 MAPK signaling pathway

Idiopathic pulmonary fibrosis (IPF) is a chronic disease with high mortality and is refractory to treatment. Pulmonary macrophages can both promote and repress fibrosis, however molecular mechanisms regulating macrophage functions during fibrosis remain poorly understood. FOXM1 is a transcription factor and is not expressed in quiescent lungs. Herein, we show that FOXM1 is highly expressed in pulmonary macrophages within fibrotic lungs of IPF patients and mouse fibrotic lungs. Macrophage-specific deletion of Foxm1 in mice (myFoxm1-/-) exacerbated pulmonary fibrosis. Inactivation of FOXM1 in vivo and in vitro increased p38 MAPK signaling in macrophages and decreased DUSP1, a negative regulator of p38 MAPK pathway. FOXM1 directly activated Dusp1 promoter. Overexpression of DUSP1 in FOXM1-deficient macrophages prevented activation of p38 MAPK pathway. Adoptive transfer of wild-type monocytes to myFoxm1-/- mice alleviated bleomycin-induced fibrosis. Altogether, contrary to known pro-fibrotic activities in lung epithelium and fibroblasts, FOXM1 has anti-fibrotic function in macrophages by regulating p38 MAPK.

The FOXM1 Inhibitor RCM-1 Decreases Carcinogenesis and Nuclear β-Catenin

The oncogenic transcription factor FOXM1 has been previously shown to play a critical role in carcinogenesis by inducing cellular proliferation in multiple cancer types. A small-molecule compound, Robert Costa Memorial drug-1 (RCM-1), has been recently identified from high-throughput screen as an inhibitor of FOXM1 in vitro and in mouse model of allergen-mediated lung inflammation. In the present study, we examined antitumor activities of RCM-1 using tumor models. Treatment with RCM-1 inhibited tumor cell proliferation as evidenced by increased cell-cycle duration. Confocal imaging of RCM-1-treated tumor cells indicated that delay in cellular proliferation was concordant with inhibition of FOXM1 nuclear localization in these cells. RCM-1 reduced the formation and growth of tumor cell colonies in the colony formation assay. In animal models, RCM-1 treatment inhibited growth of mouse rhabdomyosarcoma Rd76-9, melanoma B16-F10, and human H2122 lung adenocarcinoma. RCM-1 decreased FOXM1 protein in the tumors, reduced tumor cell proliferation, and increased tumor cell apoptosis. RCM-1 decreased protein levels and nuclear localization of β-catenin, and inhibited protein-protein interaction between β-catenin and FOXM1 in cultured tumor cells and in vivo Altogether, our study provides important evidence of antitumor potential of the small-molecule compound RCM-1, suggesting that RCM-1 can be a promising candidate for anticancer therapy.

Targeted Inhibition of the Dual Specificity Phosphatases DUSP1 and DUSP6 Suppress MPNST Growth via JNK

PURPOSE

In neurofibromatosis type 1 (NF1) and in highly aggressive malignant peripheral nerve sheath tumors (MPNSTs), constitutively active RAS-GTP and increased MAPK signaling are important in tumorigenesis. Dual specificity phosphatases (DUSPs) are negative regulators of MAPK signaling that dephosphorylate p38, JNK, and ERK in different settings. Although often acting as tumor suppressors, DUSPs may also act as oncogenes, helping tumor cells adapt to high levels of MAPK signaling. We hypothesized that inhibiting DUSPs might be selectively toxic to cells from NF1-driven tumors.

EXPERIMENTAL DESIGN

We examined DUSP gene and protein expression in neurofibroma and MPNSTs. We used small hairpin RNA (shRNA) to knock down DUSP1 and DUSP6 to evaluate cell growth, downstream MAPK signaling, and mechanisms of action. We evaluated the DUSP inhibitor, (E)-2-benzylidene-3-(cyclohexylamino)-2,3-dihydro-1H-inden-1-one (BCI), in MPNST cell lines and in cell-line and patient-derived MPNST xenografts.

RESULTS

DUSP1 and DUSP6 are expressed in NF1-deleted tumors. Knockdown of DUSP1 and DUSP6, alone or in combination, reduced MPNST cell growth and led to ERK and JNK hyperactivation increasing downstream TP53 and p-ATM. The DUSP inhibitor, BCI, diminished the survival of NF1-deleted Schwann cells and MPNST cell lines through activation of JNK. In vivo, treatment of an established cell-line xenograft or a novel patient-derived xenograft (PDX) of MPNSTs with BCI increased ERK and JNK activation, caused tumor necrosis and fibrosis, and reduced tumor volume in one model.

CONCLUSIONS

Targeting DUSP1 and DUSP6 genetically or with BCI effectively inhibits MPNST cell growth and promotes cell death, in vitro and in xenograft models. The data support further investigation of DUSP inhibition in MPNSTs.

FOXM1 activates AGR2 and causes progression of lung adenomas into invasive mucinous adenocarcinomas

Lung cancer remains one of the most prominent public health challenges, accounting for the highest incidence and mortality among all human cancers. While pulmonary invasive mucinous adenocarcinoma (PIMA) is one of the most aggressive types of non-small cell lung cancer, transcriptional drivers of PIMA remain poorly understood. In the present study, we found that Forkhead box M1 transcription factor (FOXM1) is highly expressed in human PIMAs and associated with increased extracellular mucin deposition and the loss of NKX2.1. To examine consequences of FOXM1 expression in tumor cells in vivo, we employed an inducible, transgenic mouse model to express an activated FOXM1 transcript in urethane-induced benign lung adenomas. FOXM1 accelerated tumor growth, induced progression from benign adenomas to invasive, metastatic adenocarcinomas, and induced SOX2, a marker of poorly differentiated tumor cells. Adenocarcinomas in FOXM1 transgenic mice expressed increased MUC5B and MUC5AC, and reduced NKX2.1, which are characteristics of mucinous adenocarcinomas. Expression of FOXM1 in KrasG12D transgenic mice increased the mucinous phenotype in KrasG12D-driven lung tumors. Anterior Gradient 2 (AGR2), an oncogene critical for intracellular processing and packaging of mucins, was increased in mouse and human PIMAs and was associated with FOXM1. FOXM1 directly bound to and transcriptionally activated human AGR2 gene promoter via the -257/-247 bp region. Finally, using orthotopic xenografts we demonstrated that inhibition of either FOXM1 or AGR2 in human PIMAs inhibited mucinous characteristics, and reduced tumor growth and invasion. Altogether, FOXM1 is necessary and sufficient to induce mucinous phenotypes in lung tumor cells in vivo.