Abstract It is now well established from single-cell RNA-seq studies that glioblastoma (GBM) tumors are complex, maligned neuro-developmental ecosystems harboring diverse tumor cell types. Neoplastic cells can resemble astrocytes, neural progenitors, oligodendrocyte progenitor cells, mesenchymal cells, and radial glial cells that contribute to tumor growth and homeostasis in specific ways. However, GBM single-cell data sets have failed to produce general models for transitions in and out of specific developmental and proliferative states in tumors. One reason is that human GBM tumors do not neatly resolve into developmental hierarchies. Here we focused on modeling GBM tumor cellular heterogeneity by defining "proliferative compartments" in single-cell transcriptomic data derived from primary tumors and early passage tumorsphere cultures. Previously, we observed that each tumor cell entering S-phase has a unique developmental signature that can be leveraged to define broader partitioned proliferative compartments (PPCs) with distinct developmental gene expression and genomic alteration patterns. Thus, we extracted the S-phase cells from a tumor, defined the proliferative compartments in that tumor using de novo clustering, determined the marker genes for each compartment, and defined the broader PPCs across tumors by de novo clustering based on similarity in marker genes. From a cohort of six tumors we observed eight broader PPCs and found that tumors can contain as many as five PPCs or as few as two. Because tumor growth and recurrence both require cell proliferation, we propose that patient-specific PPCs represent the engines of GBM progression.
Read full abstract