Corinne Linardic

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 (CNNs) to learn histologic features associated with driver mutations and outcome using H&E images of RMS. PATIENTS AND METHODS: Digital whole slide H&E images were collected from clinically annotated diagnostic tumor samples from n=321 RMS patients 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 to current molecular-clinical risk stratification. CONCLUSION: 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.

Rhabdomyosarcoma (RMS), a cancer characterized by features of skeletal muscle, is the most common soft-tissue sarcoma of childhood. With 5-year survival rates among high-risk groups at < 30%, new therapeutics are desperately needed. Previously, using a myoblast-based model of fusion-negative RMS (FN-RMS), we found that expression of the Hippo pathway effector transcriptional coactivator YAP1 (YAP1) permitted senescence bypass and subsequent transformation to malignant cells, mimicking FN-RMS. We also found that YAP1 engages in a positive feedback loop with Notch signaling to promote FN-RMS tumorigenesis. However, we could not identify an immediate downstream impact of this Hippo-Notch relationship. Here, we identify a HES1-YAP1-CDKN1C functional interaction, and show that knockdown of the Notch effector HES1 (Hes family BHLH transcription factor 1) impairs growth of multiple FN-RMS cell lines, with knockdown resulting in decreased YAP1 and increased CDKN1C expression. In silico mining of published proteomic and transcriptomic profiles of human RMS patient-derived xenografts revealed the same pattern of HES1-YAP1-CDKN1C expression. Treatment of FN-RMS cells in vitro with the recently described HES1 small-molecule inhibitor, JI130, limited FN-RMS cell growth. Inhibition of HES1 in vivo via conditional expression of a HES1-directed shRNA or JI130 dosing impaired FN-RMS tumor xenograft growth. Lastly, targeted transcriptomic profiling of FN-RMS xenografts in the context of HES1 suppression identified associations between HES1 and RAS-MAPK signaling. In summary, these in vitro and in vivo preclinical studies support the further investigation of HES1 as a therapeutic target in FN-RMS.

PURPOSE: Rhabdomyosarcoma is the most common soft tissue sarcoma of childhood. Despite aggressive therapy, the 5-year survival rate for patients with metastatic or recurrent disease remains poor, and beyond PAX-FOXO1 fusion status, no genomic markers are available for risk stratification. We present an international consortium study designed to determine the incidence of driver mutations and their association with clinical outcome. PATIENTS AND METHODS: Tumor samples collected from patients enrolled on Children's Oncology Group trials (1998-2017) and UK patients enrolled on malignant mesenchymal tumor and RMS2005 (1995-2016) trials were subjected to custom-capture sequencing. Mutations, indels, gene deletions, and amplifications were identified, and survival analysis was performed. RESULTS: DNA from 641 patients was suitable for analyses. A median of one mutation was found per tumor. In FOXO1 fusion-negative cases, mutation of any RAS pathway member was found in > 50% of cases, and 21% had no putative driver mutation identified. BCOR (15%), NF1 (15%), and TP53 (13%) mutations were found at a higher incidence than previously reported and TP53 mutations were associated with worse outcomes in both fusion-negative and FOXO1 fusion-positive cases. Interestingly, mutations in RAS isoforms predominated in infants < 1 year (64% of cases). Mutation of MYOD1 was associated with histologic patterns beyond those previously described, older age, head and neck primary site, and a dismal survival. Finally, we provide a searchable companion database (ClinOmics), containing all genomic variants, and clinical annotation including survival data. CONCLUSION: This is the largest genomic characterization of clinically annotated rhabdomyosarcoma tumors to date and provides prognostic genetic features that refine risk stratification and will be incorporated into prospective trials.

Rhabdomyosarcomas (RMSs) are the most common soft tissue sarcomas in children/adolescents less than 18 years of age with an annual incidence of 1-2/million. Inter/intra-tumour heterogeneity raise challenges in clinical, pathological and biological research studies. Risk stratification in European and North American clinical trials previously relied on clinico-pathological features, but now, incorporates PAX3/7-FOXO1-fusion gene status in the place of alveolar histology. International working groups propose a coordinated approach through the INternational Soft Tissue SaRcoma ConsorTium to evaluate the specific genetic abnormalities and generate and integrate molecular and clinical data related to patients with RMS across different trial settings. We review relevant data and present a consensus view on what molecular features should be assessed. In particular, we recommend the assessment of the MYOD1-LR122R mutation for risk escalation, as it has been associated with poor outcomes in spindle/sclerosing RMS and rare RMS with classic embryonal histopathology. The prospective analyses of rare fusion genes beyond PAX3/7-FOXO1 will generate new data linked to outcomes and assessment of TP53 mutations and CDK4 amplification may confirm their prognostic value. Pathogenic/likely pathogenic germline variants in TP53 and other cancer predisposition genes should also be assessed. DNA/RNA profiling of tumours at diagnosis/relapse and serial analyses of plasma samples is recommended where possible to validate potential molecular biomarkers, identify new biomarkers and assess how liquid biopsy analyses can have the greatest benefit. Together with the development of new molecularly-derived therapeutic strategies that we review, a synchronised international approach is expected to enhance progress towards improved treatment assignment, management and outcomes for patients with RMS.

The Children's Oncology Group (COG) uses Clinical Group (CG) and modified Tumor Node Metastasis (TNM) stage to classify rhabdomyosarcoma (RMS). CG is based on surgicopathologic findings and is determined after the completion of initial surgical procedure(s) but prior to chemotherapy and/or radiation therapy. The modified TNM stage is based on clinical and radiographic findings and is assigned prior to any treatment. These systems have evolved over several decades. We review the history, evolution, and rationale behind the current CG and modified TNM classification systems used by COG for RMS. Data from the seven most recently completed and reported frontline COG trials (D9602, D9802, D9803, ARST0331, ARST0431, ARST0531, ARST08P1) were analyzed, and confirm that CG and modified TNM stage remain relevant and useful for predicting prognosis in RMS. We propose updates based on recent data and discuss factors warranting future study to further optimize these classification systems.