An unsupervised learning approach uncovers divergent mesenchymal-like gene expression programs across human neuroblastoma tumors, preclinical models, and chemotherapy-exposed tumors

Abstract

ABSTRACTNeuroblastoma is a common pediatric cancer, where preclinical studies have suggested chemotherapy resistance is driven by a mesenchymal-like gene expression program. However, the poor clinical outcomes imply we need a better understanding of the relationship between patient tumors and preclinical models. Here, we generated single-cell RNA-seq maps of neuroblastoma cell lines, patient-derived xenograft models (PDX), and a genetically engineered mouse model (GEMM). We developed an unsupervised machine learning approach to compare the gene expression programs found in preclinical models to a large cohort of human neuroblastoma tumors. The dominant adrenergic programs were well preserved in preclinical models, but contrary to previous reports do not unambiguously map to an obvious cell of origin. The mesenchymal-like program was less clearly preserved, and primarily restricted to cancer-associated fibroblasts and Schwann-like cellsin vivo. Surprisingly however, we identified a subtle, weakly expressed, mesenchymal-like program in otherwise adrenergic cancer cells in some high-risk tumors. This program appears distinct from mesenchymal cell lines but was maintained in PDX and a similar program could be chemotherapy-induced in our GEMM after only 24 hours, suggesting an uncharacterized therapy-escape mechanism. Collectively, our findings advance the understanding of neuroblastoma heterogeneity and can inform the development of new treatments.