Depicting the genetic architecture of pediatric cancers through an integrative gene network approach

www.nature.com/scientificreports OPEN Depicting the genetic architecture of pediatric cancers through an integrative gene network approach Clara Savary1, Artem Kim1, Alexandra Lespagnol2, Virginie Gandemer1,3, Isabelle Pellier4, Charlotte Andrieu5,6, Gilles Pagès7,8, Marie-Dominique Galibert1,2, Yuna Blum 9,10 & Marie de Tayrac1,5,10* The genetic etiology of childhood cancers still remains largely unknown. It is therefore essential to develop novel strategies to unravel the spectrum of pediatric cancer genes. Statistical network modeling techniques have emerged as powerful methodologies for enabling the inference of genedisease relationship and have been performed on adult but not pediatric cancers. We performed a deep multi-layer understanding of pan-cancer transcriptome data selected from the Treehouse Childhood Cancer Initiative through a co-expression network analysis. We identified six modules strongly associated with pediatric tumor histotypes that were functionally linked to developmental processes. Topological analyses highlighted that pediatric cancer predisposition genes and potential therapeutic targets were central regulators of cancer-histotype specific modules. A module was related to multiple pediatric malignancies with functions involved in DNA repair and cell cycle regulation. This canonical oncogenic module gathered most of the childhood cancer predisposition genes and clinically actionable genes. In pediatric acute leukemias, the driver genes were co-expressed in a module related to epigenetic and post-transcriptional processes, suggesting a critical role of these pathways in the progression of hematologic malignancies. This integrative pan-cancer study provides a thorough characterization of pediatric tumor-associated modules and paves the way for investigating novel candidate genes involved in childhood tumorigenesis. Cancer remains the leading cause of death by disease in children of less than fourteen years of age1. Improving the management of pediatric cancer is essential and will benefit from more accurate diagnosis, new personalized treatment and development of specific and less damaging therapies. To face these challenges, it is necessary to unravel the complete genetic repertoire of pediatric malignancies. Recent studies have improved the understanding of the genetics of childhood cancer, but have mainly focused on depicting the germline and somatic mutational landscape of these diseases2–4. Several evidences demonstrated that the biology and genetics of pediatric cancers set them apart from adult tumors4,5. Childhood cancers have a 14-times lower mutation rate compared to adult tumors and mostly arise from mutations in few driver genes. Somatic alterations mostly target a handful of major genes such as CDKN2A, NOTCH1, NRAS, KRAS or TP53, and pathways disrupted by driver alterations are either common to cancer (e.g. cell cycle) or specific to pediatric cancer histotypes4. More than half of the driver genes are restricted to one cancer histotype and 83% of them are not shared between hematologic and solid tumors. This indicates that certain genes and pathways are exclusively dysregulated in a single type of childhood cancer. 1 Univ Rennes, CNRS, IGDR (Institut de génétique et développement de Rennes) - UMR 6290, Rennes, France. Somatic Cancer Genetics Department, Pontchaillou University Hospital, Rennes, France. 3Pediatric Oncology Department, Pontchaillou University Hospital, Rennes, France. 4Pediatric Immuno-Hemato-Oncology Unit, Angers University Hospital, Angers, France. 5Molecular Genetics and Genomics Department, Pontchaillou University Hospital, Rennes, France. 6Chemistry Oncogenesis Stress Signaling (COSS) Laboratory – INSERM U1242, Centre de Lutte Contre le Cancer (CLCC) Eugène Marquis, Rennes, France. 7University Côte d’Azur, IRCAN (Institute for Research on Cancer and Aging of Nice) - CNRS UMR 7284 and INSERM U1081, Centre Antoine Lacassagne, Nice, France. 8Biomedical Department, Centre Scientifique de Monaco, Monaco, Principality of Monaco. 9Programme Cartes d’Identité des Tumeurs (CIT), Ligue Nationale Contre le Cancer, Paris, France. 10These authors contributed equally: Yuna Blum and Marie de Tayrac. *email: 2 Scientific Reports | (2020) 10:1224 | https://doi.org/10.1038/s41598-020-58179-0 1 www.nature.com/scientificreports/ www.nature.com/scientificreports Regarding hereditary predisposition, genome-wide studies reported that pathogenic germline variants were identified in 8–10% of the affected children and adolescents2,6–8. This proportion is likely underestimated considering that only cancer-related genes were analyzed for pathogenicity in these studies. To date, over 100 cancer predisposition genes have been described and most of the associated pathogenic germline variants were loss of function mutations in DNA or double-stranded break repair genes2,3,8. The total spectrum of cancer-predisposition genes involved in childhood tumorigenesis still remains to be uncovered. Tumor initiation and progression result from complex interplay between germline and somatic events that shape the transcriptional landscape of tumors9,10. Integration of transcriptome-based knowledge has emerged as a powerful method for prioritizing genomic alterations in cancers11. Statistical network modeling is essential for interpreting genotype-to-phenotype relationships or discerning transcriptional regulatory programs12–14. Studies reported that mature pediatric tumors mirror the conserved transcriptional programs of embryonic cell populations that have been subject to genomic changes15. A system-level understanding of how the genetic mutations affect transcriptional profile has been provided in adult pan-cancer data16. Such analyses revealed common functional gene clusters that are shared by multiple adult cancer types. In onco-pediatric research, construction of co-expression networks achieved interesting results in identifying predictive molecular biomarkers and in unraveling differential regulatory molecular programs by analyzing matched normal-tumor samples 14,17. The published studies have only focused on deciphering co-expression networks of one particular histotype and, therefore, lack to provide a global view of both common and histotype-specific processes that drive childhood tumorigenesis. This requires a deep exploration of the co-expression network obtained by analyzing pan-cancer childhood data. Here, we carried out computational analyses of the transcriptome data of 820 pediatric cancer samples selected from the Treehouse Childhood Cancer Initiative (TCCI) dataset across six cancer histotypes. We constructed a co-expression network using weighted gene co-expression network analysis (WGCNA) to capture transcriptional relationships between genes in pediatric cancers. We associated the resulting modules with tumor types by examining their transcriptional profiles and by characterizing their biological functions. We determined the (...truncated)